<?xml version="1.0" encoding="utf-8"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:media="http://search.yahoo.com/mrss/"><channel><title>Customers</title><link>https://cloud.google.com/blog/topics/customers/</link><description>Customers</description><atom:link href="https://cloudblog.withgoogle.com/blog/topics/customers/rss/" rel="self"></atom:link><language>en</language><lastBuildDate>Thu, 12 Nov 2026 17:06:54 +0000</lastBuildDate><image><url>https://cloud.google.com/blog/topics/customers/static/blog/images/google.a51985becaa6.png</url><title>Customers</title><link>https://cloud.google.com/blog/topics/customers/</link></image><item><title>How lemonflow built real-time agents for EV charging using Gemma and TPUs</title><link>https://cloud.google.com/blog/topics/startups/building-real-time-agents-for-ev-charging-using-gemma-and-tpus/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;Welcome back to The Blueprint, a regular feature where we highlight how Google Cloud customers are tackling unique and common challenges across industries using the latest AI and cloud technologies. We hope to inspire others looking to innovate in their work.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;As the electric vehicle revolution gains steam in Europe, every charging session is a moment of truth, for the driver and for the transition itself. At &lt;/span&gt;&lt;a href="https://lemonflow.ai" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;lemonflow&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, we build AI agents purpose-built for EV charging: support agents for drivers and monitoring agents for charging networks. Our support agents answer drivers around the clock in more than 90 languages, and today they help an EV driver every 90 seconds. &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;The Challenge&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Before we founded lemonflow, we worked closely with leading mobility players across Europe and saw first-hand how often charging goes wrong: Between 15% and 25% of EV charging attempts in Europe end in failure. Across Europe’s rapidly expanding EV market, where battery-electric car sales grew &lt;/span&gt;&lt;a href="https://www.electrive.com/2026/01/27/eu-wide-new-electric-car-registrations-rise-by-30" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;30% in 2025&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and &lt;/span&gt;&lt;a href="https://zevalliance.org/wp-content/uploads/2025/02/Toward-health-competition_Feb-2025.pdf" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;thousands of charge point operators&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; run public chargers, drivers who encountered stalled charging points were often left stranded or stuck in slow call center queues.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We set out to change that hands-on: to bring the cutting edge of AI to EV charging, and to build the world’s best conversational AI for distributed hardware, an automated platform that resolves real-time charging issues via phone, chat, email, and WhatsApp. However, handling live voice support created several technical requirements:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Strict latency thresholds:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Frustrated drivers want immediate assistance. Our testing confirmed that serving latencies above 800 milliseconds risk conversational barge-ins, causing drivers to interrupt the agent and break the interaction flow.&lt;/span&gt;&lt;/li&gt;
&lt;li role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Managed API bottlenecks:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Initially relying on managed LLM APIs, we faced latency deviations during peak hours (spiking from 1.0–1.2 seconds to over 1.7 seconds) and experienced HTTP 429 rate-limit errors when global usage peaked.&lt;/span&gt;&lt;/li&gt;
&lt;li role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Compute-intensive conversations:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Every agent works from one comprehensive prompt, paired with dynamic tool calls that read and act on live charger data. This design creates intense bursts of compute demand during calls, making throughput on shared, managed capacity difficult to guarantee under peak loads.&lt;/span&gt;&lt;/li&gt;
&lt;li role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Highly fragmented ecosystem:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; We needed a standardized conversational engine capable of bridging around 30 distinct Charge Point Management Systems (CPMS), the backends operators use to run their chargers, each with its own APIs, data models and quirks.&lt;/span&gt;&lt;/li&gt;
&lt;li role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Hundreds of tailored agents: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;No two operators run support the same way. Each has its own tariffs, payment methods, escalation paths and hours for human backup, and most run several agents, for example one per country, brand or hotline, with behavior that can change by time of day. Each agent’s prompt is unique and must be served at full speed and improved without breaking any of the others.&lt;/span&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To deliver ultra-low-latency voice interactions and guaranteed serving capacity, we needed direct control over our model hosting environment.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;The Solution&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We wanted every driver to get best-in-class support that feels effortless, almost magical: no queue, no menu, just a short conversation that ends with the car charging. Picture a driver on a long cross-border trip, pulling into a motorway service station at 11 p.m. She plugs in, but the charger won’t start: the previous session never closed properly. She calls the number on the charger and is answered in her own language. Within seconds, our agent spots the stuck session in the charger’s logs, restarts the charger remotely and stays on the line until power is flowing.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To get there, we had to solve our latency bottlenecks and secure dedicated compute capacity. We migrated from managed APIs to self-hosting the open &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Gemma 4&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; model on &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Google Cloud TPU v6e&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; hardware in the EU.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;a href="https://cloud.google.com/blog/topics/startups/why-your-startup-needs-open-models-alongside-frontier-apis?e=0"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Like many startups using Gemma&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, the compatibility of these open models with Gemini allowed us to transition our existing prompt infrastructure with minimal modification.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Moving to self-hosted TPUs unlocked the more complete architectural control our engineering team required to establish a deterministic core layer that handles EV-specific guardrails, guards against prompt injections, and standardizes previously fragmented backend protocols into our internal format before ever reaching the LLM. Control had a second payoff: because the model and serving stack change only when we change them, every test result stays meaningful. That stability laid the foundation for Agent Coach, our background coaching engine for AI agents, described below.  &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We then developed an intelligent routing gateway to prioritize workloads according to operational urgency, serving live voice calls first, followed by web chats, and deferring automated re-simulations to low-traffic periods.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Custom optimizations to the open-source vLLM serving engine and memory tiering between the TPU and host memory let a single TPU slice keep many more operator agents ready for instant answers, without recalculating their context.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We also implemented automated fallback routing to self-hosted GPUs and Gemini Enterprise endpoints. Thanks to TPU reliability, Gemma answers more than 99.9% of all live conversation turns, and these fallbacks serve primarily as deployment buffers during system updates. Primary or fallback, every request is served from secure EU compute.&lt;/span&gt;&lt;/p&gt;
&lt;h2&gt;&lt;span style="vertical-align: baseline;"&gt;The Architecture&lt;/span&gt;&lt;/h2&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/28788___GFS___Diagram_Redesign___Design_fo.max-1000x1000.png"
        
          alt="image2"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;The Outcome&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;For the e-mobility companies we work with, and for their drivers, the result is support that simply works: our agents typically answer every call, handle more than 80% of conversations end to end without a human handover, and earn 95% driver satisfaction. Behind the scenes, self-hosting Gemma on Google Cloud TPUs transformed our performance:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Rapid deployment cycles:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Prompt update and test cycle times dropped from 2–3 hours to about 10 minutes, a more than tenfold faster improvement loop.&lt;/span&gt;&lt;/li&gt;
&lt;li role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Sub-second serving latency:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; We lowered our P50 latency to around 350 milliseconds and our P90 latency to around 600 milliseconds. That is roughly three times faster than our managed-API baseline and fast enough to sharply reduce customer barge-ins.&lt;/span&gt;&lt;/li&gt;
&lt;li role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Zero throttling and flat latency:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Dedicated TPU infrastructure eliminated HTTP 429 rate-limit errors, and latency stays flat through peak afternoon call traffic.&lt;/span&gt;&lt;/li&gt;
&lt;li role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Testing without limits:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; On dedicated capacity, every TPU cycle that live conversations don’t need becomes testing capacity, so our team can experiment and validate at will.&lt;/span&gt;&lt;/li&gt;
&lt;li role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Built for scale: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Today, our agents support more than 40 charging networks in 17 countries, covering more than 250,000 charge points, around the clock. And this is just the beginning: the same platform is ready to take on many more networks and markets.&lt;/span&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With additional compute capacity, we introduced &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Coach&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, a background engine that provides agents with relevant context during live calls and turns real conversations into permanent test cases, so every call makes the next one better.  We also added nightly automated regression testing across thousands of fixed conversational scenarios.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Looking ahead, we are preparing to move to Ironwood, Google’s seventh-generation TPU, as soon as it is available in our EU region. With six times the high-bandwidth memory per chip, Ironwood will multiply the number of operator agents a single TPU can keep ready in memory for instant answers.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Ultimately, our goal is that no driver is ever left stranded at a charger. They get an instant, expert answer in their own language, the charger is rebooted remotely if necessary, and our AI monitoring agents catch the next fault before another driver ever notices. Open models on dedicated Cloud TPUs give us the speed and control to make that the standard across Europe.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;The Details&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Our platform runs two kinds of AI agents: support agents that talk to drivers in real time, served by Gemma on Cloud TPUs for minimal latency, and monitoring agents that analyze conversations and charger logs in the background on Gemini, so operators can fix faults before the next driver is affected. To keep its core platform consistent across many kinds of e-mobility companies, from charge point operators (CPOs) and e-mobility service providers (eMSPs) to charger manufacturers, lemonflow built a translation layer that converts varied CPMS interfaces into a standardized format.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;When a driver calls the support line posted on a charger, the call reaches our Conversational Engine through the operator’s own hotline. The engine then pulls the charger’s live status and Open Charge Point Protocol (OCPP) logs, the messages chargers exchange with their backend, from the operator’s CPMS in real time. Because charging stations often lack digital displays, drivers are frequently unaware of underlying issues like pending mobile 3D-Secure payment authorizations, session timeouts, charging cards not enabled for roaming, power reduced by local load management, or a car that ended the session at its own charge limit. Our engine feeds structured log summaries directly into Gemma on TPU to inform interactions with drivers.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Direct model control allows us to stream token output selectively. Partial text tokens stream instantly to our Text-to-Speech (TTS) engine, while structured tool-call tokens stream asynchronously to backend systems to execute real-world actions, such as remotely restarting chargers, unlocking stuck cables, stopping sessions that failed to end, sending drivers details via SMS or email, or notifying service technicians.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Deterministic security controls screen inputs before they reach the TPU-hosted LLM, blocking malicious prompts. This architecture is designed with European critical-infrastructure requirements in mind, and it still keeps conversations snappy, with a typical response starting in around 350 milliseconds.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;In the end, all of this should stay invisible to the driver: they just get a fast answer and a charger that works. We’re proud to work with Google Cloud to make that the norm and help EV adoption keep growing.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Wed, 11 Nov 2026 17:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/topics/startups/building-real-time-agents-for-ev-charging-using-gemma-and-tpus/</guid><category>Manufacturing</category><category>Customers</category><category>Startups</category><media:content height="540" url="https://storage.googleapis.com/gweb-cloudblog-publish/images/Lemonflow-Blueprint-Header.max-600x600.png" width="540"></media:content><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>How lemonflow built real-time agents for EV charging using Gemma and TPUs</title><description></description><image>https://storage.googleapis.com/gweb-cloudblog-publish/images/Lemonflow-Blueprint-Header.max-600x600.png</image><site_name>Google</site_name><url>https://cloud.google.com/blog/topics/startups/building-real-time-agents-for-ev-charging-using-gemma-and-tpus/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Dr. Lukas Fieber</name><title>Co-Founder &amp; CTO, lemonflow</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Konstantin Huneke</name><title>Co-Founder &amp; CEO, lemonflow</title><department></department><company></company></author></item><item><title>The Blueprint: Orchestrating 5,000+ AI assets for a World Cup TV spot</title><link>https://cloud.google.com/blog/products/media-entertainment/how-mito-orchestrated-5000-plus-ai-assets-for-a-world-cup-tv-spot/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;Welcome back to The Blueprint, a regular feature where we highlight how Google Cloud customers are tackling unique and common challenges across industries using the latest AI and cloud technologies. We hope to inspire others looking to innovate in their work.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;AI is rapidly transforming video production at virtually every step of the process, from ideation and storyboarding to filming, editing, and release. At &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Mito&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, we’re striving to democratize AI-powered video creation, enabling everyone from solo artists to large enterprises to get the most out of the best creative technologies available.&lt;/span&gt;&lt;/p&gt;
&lt;h2&gt;&lt;strong style="vertical-align: baseline;"&gt;The Challenge&lt;/strong&gt;&lt;/h2&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;When we founded Mito, our goal was to eliminate the complex technical friction that stops creators from telling long-form, narrative stories with generative AI. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;However, transitioning from a single five-second clip to a cohesive campaign — such as a 120-second national TV spot for Movistar Plus+, a part of Telefónica, to broadcast across Spain during the World Cup this summer. This effort revealed massive infrastructure and workflow bottlenecks:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Asset &amp;amp; workflow overhead:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Managing local media libraries across individual workstations was unscalable. For Movistar alone, our production required generating a massive "iceberg" of over 5000 assets (1000 of those videos) generated in around 65 iteration rounds.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Zero real-time collaboration:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Traditional feedback loops required manual downloads, Google Drive uploads, and fragmented communication across chat apps and email.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Rendering latency:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Standard AI video rendering takes four to five minutes per short clip. Running these processes synchronously froze our user interface and stalled creative velocity.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Model disparity &amp;amp; consistency:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Every generative model requires distinct prompting logic. Maintaining character persistence, specific lighting, and camera angles across dozens of shots — while preserving the likeness of real public figures like chef Ferran Adrià or journalist Juanma Castaño — was nearly impossible without central orchestration.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Capacity constraints:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Operating on public pay-as-you-go model APIs left us vulnerable to rate limits and regional throughput throttling during peak traffic.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;h2&gt;&lt;strong style="vertical-align: baseline;"&gt;The Solution&lt;/strong&gt;&lt;/h2&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To deliver an infinite, real-time creative canvas for filmmakers, we designed a serverless, asynchronous architecture built on Google Cloud.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We run our API, AI Director, and background workers as Java services on Cloud Run, which scales with request volume.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;In order to address message routing challenges, we decoupled job execution using Google Cloud &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Pub/Sub&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; as our central event bus. Long-running video generations are offloaded to background Cloud Run workers, allowing creators to keep working uninterrupted on the UI canvas.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To minimize latency, media assets are stored in Google Cloud Storage and delivered via Cloud CDN. Expiring key signatures keep pre-release client assets strictly private. Keeping our storage alongside the models we call through Vertex AI reduces network transfer delays.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;An isolated database architecture helps with the heavy business intelligence and analytics workloads while mitigating potential negative impacts on production performance through Cloud SQL and PostgreSQL. We rely heavily on foundational models like Omni, Veo and Nano Banana, which we access through Vertex AI alongside our custom system prompts.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Finally, our structured canvas frontend enforces a "Rule of 6" (limiting prompt variables across World, Camera, Light, Texture, Actions, and Character) and prioritizes still-frame generation before video rendering to maximize cost efficiency.&lt;/span&gt;&lt;/p&gt;
&lt;h2&gt;&lt;strong style="vertical-align: baseline;"&gt;The Architecture&lt;/strong&gt;&lt;/h2&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/image2_VoY9Mls.max-1000x1000.jpg"
        
          alt="image2"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h2&gt;&lt;strong style="vertical-align: baseline;"&gt;The Outcome&lt;/strong&gt;&lt;/h2&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;By pairing our creative orchestration platform with Google Cloud’s infrastructure, we successfully delivered a landmark national broadcast campaign that the production team describes as 95% AI-generated:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Prime-time scale:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; The 120-second Movistar campaign for the Spanish National Team  aired across every major television channel in Spain, with 1,441 assets live on a single shared Canvas near launch.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;AI Director copilot:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Our conversational agent translates natural language instructions into precise model parameters while maintaining context, lighting, and visual style across every scene.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Seamless collaboration:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Distributed teams and brand clients can work on the same canvas in real time. Because new generations are saved as separate assets instead of overwriting earlier ones, stakeholders can duplicate projects to explore parallel creative ideas.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Continuous dogfooding:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Through our internal studio, Mito Films, we put our tools to the test on high-stakes client projects — whether that's global television spots or keynotes for top robotics startups — ensuring direct feedback flows back into engineering.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Future-proof foundation:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Our architecture is primed for rendering times to collapse from minutes to seconds, positioning Mito to handle upcoming world models, deeper memory states, and personalized content delivery.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;h2&gt;&lt;strong style="vertical-align: baseline;"&gt;The Details&lt;/strong&gt;&lt;/h2&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Building an enterprise-grade platform for AI storytelling required solving several backend and creative challenges. This includes our creation of a virtual queuing system that mitigates third-party API rate limiting during usage spikes for smooth request concurrency, with bounded retries and a clear outcome for every job.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Because creative output cannot be fully validated with standard unit tests, we built an automated evaluation harness powered by Gemini. Acting as an independent judge, it evaluates AI Director outputs against original user intents across an automated test suite, which we run alongside standard tests for high-impact Director changes.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We knew early on that we would need multi-model intelligence routing. As such, we built the AI Director to dynamically coordinate tasks across multiple specialized models — including Sonnet, Gemini, and ElevenLabs for audio — treating the spatial canvas like a live code repository to infer adjacent contextual cues.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We also needed to embrace an “imperfect aesthetic” to arrive at a more authentic outcome. In our Movistar spot, the story follows Tommy, a 17-year-old student making an AI video project. At the climax, the AI misunderstands a prompt, making coach Luis de la Fuente say "In the same milkshake" while holding a drink. Tommy edits the prompt live to "In the same heartbeat," and the video recalibrates instantly. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;By incorporating surreal visual elements — like stadium stands built into Galician cliffs or a neighborhood bar covered in pitch grass — we showcased how generative AI can elevate human storytelling rather than obscure it.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Fri, 06 Nov 2026 17:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/media-entertainment/how-mito-orchestrated-5000-plus-ai-assets-for-a-world-cup-tv-spot/</guid><category>AI &amp; Machine Learning</category><category>Customers</category><category>Media &amp; Entertainment</category><media:content height="540" url="https://storage.googleapis.com/gweb-cloudblog-publish/images/Mito-hero.max-600x600.png" width="540"></media:content><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>The Blueprint: Orchestrating 5,000+ AI assets for a World Cup TV spot</title><description></description><image>https://storage.googleapis.com/gweb-cloudblog-publish/images/Mito-hero.max-600x600.png</image><site_name>Google</site_name><url>https://cloud.google.com/blog/products/media-entertainment/how-mito-orchestrated-5000-plus-ai-assets-for-a-world-cup-tv-spot/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Iñaki Berenguer</name><title>CEO, Mito</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Keyvan Akbary</name><title>Head of Engineering, Mito</title><department></department><company></company></author></item><item><title>How Nammu21 automates syndicated loans on Google Cloud Universal Ledger</title><link>https://cloud.google.com/blog/topics/financial-services/how-nammu21-automates-syndicated-loans-on-google-cloud-universal-ledger/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The syndicated loan and private credit markets, both critical components of corporate finance, have long been constrained by manual processes, leading to multi-week settlements, valuation opacity, and trade friction. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To solve these long-standing challenges, fintech innovator &lt;/span&gt;&lt;a href="https://www.nammu21.com/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Nammu21&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; partnered with Google Cloud on a new proof-of-concept. Using its AI-powered platform to structure legal agreements and instantiate them as smart contracts on the &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/universal-ledger"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google Cloud Universal Ledger (GCUL)&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, Nammu21 automates the entire loan lifecycle, paving the way for improved settlement times and greater market efficiency.&lt;/span&gt;&lt;/p&gt;
&lt;h2&gt;&lt;span style="vertical-align: baseline;"&gt;Digitizing the loan market&lt;/span&gt;&lt;/h2&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;While loan markets provide billions in capital to businesses worldwide, their underlying infrastructure has not kept pace with the speed of digital finance. This landscape requires continuous coordination across facility agents, borrowers, and diverse lender syndicates, while also managing highly customized credit agreements with dynamic tranches and complex covenants. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;In particular, relying on fragmented, manual interpretation of legal text often results in significant operational drag and data silos. According to the &lt;/span&gt;&lt;a href="https://blog.lma.eu.com/challenging-the-timeline-issue-no-4/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Loan Market Association&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, secondary market settlement times for par/near par loans in Q4 2024 still averaged 42 days, with a median of 27 days. These notoriously slow timelines tie up capital, increase counterparty risk, and create a heavy reconciliation burden for participants.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Nammu21 transforms bespoke credit documents into programmable assets at scale by rendering each facility's governing terms as a Digital Credit Instrument (DCI). The DCI maps every term — including conditions, consents, eligibility criteria, and operational mechanics that determine what is permitted under a facility — directly back to its originating legal provision. These elements are then expressed as executable behaviours against a loan's live state. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To instantiate these instruments, the proof-of-concept utilizes the secure, high-performance execution environment of GCUL, an institutional-grade, permissioned distributed ledger that helps automate complex financial workflows to deliver near-instant transaction finality and predictable costs without the volatility or security concerns of public blockchains.&lt;/span&gt;&lt;/p&gt;
&lt;h2&gt;&lt;span style="vertical-align: baseline;"&gt;A purpose-built stack for digital credit&lt;/span&gt;&lt;/h2&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Combined with Nammu21’s data intelligence, GCUL provides the foundational infrastructure for converting legal document logic into live, programmable contracts. Designed to handle the privacy, scale, and multi-party coordination required by institutional finance, GCUL provides a permissioned, high-performance distributed ledger that delivers deterministic, instant transaction finality for high-value financial settlements.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Programmable contracts on GCUL are written in &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;gculpy&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt;, GCUL's domain-optimized programming language. As a strict, statically-typed subset of Python, &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;gculpy&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; is designed to be safe, readable, and auditable, making it ideal for processing complex financial logic.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;For regulated financial institutions, &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;gculpy&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; offers distinct operational advantages over other ledger programming models. The language simplifies development by removing the need for complex memory management or workarounds to meet highly restrictive call stack limits, allowing engineers to focus on the financial logic itself. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;GCUL also provides robust, built-in permissioning, requiring explicit user authorization before a contract can store data on their behalf. This solves the "unsolicited asset" problem — common on public networks — where unwanted tokens can be sent to an organization's wallet, creating compliance and accounting issues. With GCUL, institutions can maintain full control over which assets they interact with to support their high-value, highly regulated use cases.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Nammu21 serves a dual role in this architecture. First, the platform renders the DCI programmatically from the original legal agreement without manual document reading to establish the terms. Next, it derives &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;gculpy&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; contracts that define the structure of the obligation and govern the loan on GCUL, using the DCI to supply the terms that determine what the loan documents permit, require and block.&lt;/span&gt;&lt;/p&gt;
&lt;h2&gt;&lt;span style="vertical-align: baseline;"&gt;Technical architecture: The loan lifecycle on GCUL&lt;/span&gt;&lt;/h2&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Nammu21’s loan implementation on GCUL demonstrates the value of creating a single, immutable source of truth. The proof-of-concept modeled a $1.945 billion Term Loan B — one segment of a broader multi-facility transaction — where the agent bank, borrower, transferor, and transferee all worked off a permissioned view of the same on-ledger rendering.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The lifecycle begins when an agent bank selects a specific credit agreement transformed by Nammu21. The platform instantly provides the structured terms from the document, which are used to instantiate a unique &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;gculpy&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; contract on GCUL, establishing a verifiable, on-ledger digital twin of that specific loan facility.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/1-facility-tokenization.max-1000x1000.png"
        
          alt="1-facility-tokenization"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Next, the agent bank initiates the onboarding of the borrower and lenders. This multilateral action requires that each participant review the request and then provide explicit consent by approving the onboarding invitation from their own interface. Once all parties are onboarded, the agent bank records an immutable hash of the loan's key terms on-ledger, attesting that the on-ledger data matches the canonical off-ledger records.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Subsequent lifecycle events, such as secondary market transfers or repayments, are also managed as multi-step, automated workflows governed by a consent-driven model. For example, a transfer of a loan position requires new lender, borrower, and agent bank consent  before it can be executed atomically on the ledger. All payment flows, from interest payments to loan close-outs, settle directly on GCUL using tokenized commercial bank money, eliminating the need for external payment rails.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/2-assignment-transfer-legend.max-1000x1000.png"
        
          alt="2-assignment-transfer-legend"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Transacting with familiar, regulated money on a programmable ledger provides an architecture purpose-built for institutional finance, grounded in three core pillars:&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;1. &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Compliance via gated execution: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Nammu21’s &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;gculpy&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; contracts instantiated with governing terms, validate every action against the terms transformed from the agreement before execution, mitigating potential covenant breaches or invalid transfers upfront while eliminating the risk of silent failures. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;2. &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Role-based privacy:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; GCUL enforces strict, role-based access, ensuring that trading strategies, positions, and sensitive loan data remain confidential and are visible only to the necessary counterparties in a transaction.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;3. &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Settlement finality: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Every on-ledger action completes near-instantly and anchors on canonical data derived from the original credit agreement, replacing provisional, reconciliation-heavy models with real-time, trusted verification.&lt;/span&gt;&lt;/p&gt;
&lt;h2&gt;&lt;span style="vertical-align: baseline;"&gt;The business impact of moving on-ledger&lt;/span&gt;&lt;/h2&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This proof-of-concept illustrates a structural shift in how credit markets can operate. When each facility's governing terms are rendered as a DCI and executed on GCUL, the operational mechanics of lending markets change fundamentally, rather than incrementally. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Values can be reconciled bilaterally, but governance logic cannot – and the market’s current friction stems directly from each party's independent interpretation of the rules. Resolving this issue on-ledger is what enables settlement times to compress, back-office overhead to shrink, and counterparty risk to be mitigated by design rather than by manual checking. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This same architecture can be applied wherever credit terms are bespoke and manually interpreted. By freeing up capital that would otherwise be trapped in lengthy settlement cycles, new liquidity is unlocked and a more dynamic and efficient secondary loan market can emerge.&lt;/span&gt;&lt;/p&gt;
&lt;hr/&gt;
&lt;p&gt;&lt;sup&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;The team would also like to thank Someera Khokhar, Jordan Shelley, and Saskia A. Aalders at Nammu21 and Googlers Ishan Singh, Kathryn Stavish, Tom Chudy, and Stathis Onasoglou for their contributions.&lt;/span&gt;&lt;/sup&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Tue, 03 Nov 2026 18:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/topics/financial-services/how-nammu21-automates-syndicated-loans-on-google-cloud-universal-ledger/</guid><category>AI &amp; Machine Learning</category><category>Customers</category><category>Web3</category><category>Financial Services</category><media:content height="540" url="https://storage.googleapis.com/gweb-cloudblog-publish/images/nammu21-automating-sydicated-loans-header.max-600x600.png" width="540"></media:content><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>How Nammu21 automates syndicated loans on Google Cloud Universal Ledger</title><description></description><image>https://storage.googleapis.com/gweb-cloudblog-publish/images/nammu21-automating-sydicated-loans-header.max-600x600.png</image><site_name>Google</site_name><url>https://cloud.google.com/blog/topics/financial-services/how-nammu21-automates-syndicated-loans-on-google-cloud-universal-ledger/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Chang Yang Jiao</name><title>Head of Blockchain, Nammu21</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Moritz Platt</name><title>Capital Markets Technology Manager</title><department></department><company></company></author></item><item><title>Easy to integrate, compliant by design: Bringing the Swiss Franc to Google Cloud Universal Ledger</title><link>https://cloud.google.com/blog/topics/financial-services/easy-to-integrate-compliant-by-design-bringing-the-swiss-franc-to-google-cloud-universal-ledger/</link><description>&lt;div class="block-paragraph"&gt;&lt;p data-block-key="j77gm"&gt;Every day, countless financial transactions are interrupted by slow, expensive, and inflexible conventional payment systems. As a result, public commercial stablecoins like Tether, USDC, and PayPal USD have become increasingly popular by making it easy to move money efficiently with instant, 24/7 blockchain settlement.&lt;/p&gt;&lt;p data-block-key="4jij8"&gt;Yet major regulated banks have largely kept such transactions at arm’s length; for these financial institutions, reserve-backed public stablecoins present numerous risks, including strict capital and liquidity requirements and unmanageable regulatory and operational uncertainty.&lt;/p&gt;&lt;p data-block-key="dn9g0"&gt;Indeed, banks don't want to adopt existing stablecoins. Instead, they need the speed and efficiency of tokenized currency built on the trusted economic foundations they’ve nurtured over years or even decades in business. That’s why many are now turning to &lt;a href="https://cloud.google.com/application/web3/universal-ledger"&gt;Google Cloud Universal Ledger&lt;/a&gt; (GCUL). Rather than channeling public stablecoins, GCUL provides the secure infrastructure banks need to issue and settle tokenized deposits. That brings real-time, 24/7 programmable payments into commercial bank money while satisfying regulators and without disrupting core banking operations.&lt;/p&gt;&lt;p data-block-key="7mq32"&gt;For many banks, this is a compelling vision, but adopting new payment technology and integrating it with core banking systems can seem to bank CTOs like a daunting undertaking. That’s particularly true as they face mounting demands for the engineering resources that keep their businesses running while also striving to stay compliant with new regulations. But with a blockchain purpose-built for financial services like GCUL, this is anything but a herculean task.&lt;/p&gt;&lt;p data-block-key="8dht"&gt;Looking to expand its offerings for an increasingly global clientele, Switzerland’s &lt;a href="https://www.crypto-finance.com/" target="_blank"&gt;Crypto Finance Group&lt;/a&gt; wanted to show it could deploy payment innovations like this in just weeks. That was its inspiration for launching a &lt;a href="https://www.crypto-finance.com/crypto-finance-amina-and-incore-bank-complete-second-phase-payment-infrastructure-pilot-on-google-cloud-universal-ledger-for-instant-account-to-account-transfers/" target="_blank"&gt;digital payments infrastructure pilot&lt;/a&gt; that enabled instant account-to-account transfers between customers of two Swiss banks on the GCUL network. The pilot, which was designed to operate within one of the world's most demanding financial security frameworks, provided a blueprint for how any commercial bank can build on Google Cloud technology to fundamentally modernize its payment operations.&lt;/p&gt;&lt;h3 data-block-key="5mhmo"&gt;&lt;b&gt;Pain points in today's payments landscape&lt;/b&gt;&lt;/h3&gt;&lt;p data-block-key="dpb5l"&gt;In recent years, there’s been a marked proliferation of tokenized payment methods. But stablecoins have transformed this previously niche technology, by effectively resolving some of traditional finance’s most enduring pain points, such as weekend transfer blackouts, direct transfers without correspondent banks, and fast settlement rails. Stablecoins also limit the frustrations caused by opaque cross-border transactions, where fragmented compliance checks can leave legitimate payments in extended and unexplained limbo and without a path to clear or timely resolution.&lt;/p&gt;&lt;p data-block-key="8ol4p"&gt;Still, commercial banks have been reluctant to offer popular reserve-backed stablecoins from permissionless blockchains to their customers despite their retail popularity. That’s because they come with numerous challenges to smooth and secure operations: an uncertain regulatory status, compliance issues around anonymous accounts, &lt;a href="https://ethereum.org/gas/" target="_blank"&gt;gas fees&lt;/a&gt;, and the scalability limitations of layer 1 blockchain technology. But perhaps banks’ biggest concerns about stablecoins is that they create a problematic economic disconnect — giving customers access to reserve-backed stablecoins reduces reserves and, as &lt;a href="https://www.mckinsey.com/industries/financial-services/our-insights/beyond-stablecoins-the-emerging-architecture-of-on-chain-money" target="_blank"&gt;a recent McKinsey study&lt;/a&gt; found, banks’ ability to lend.&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/swiss-franc-stablecoin-gcul-1.max-1000x1000.png"
        
          alt="swiss-franc-stablecoin-gcul-1"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph"&gt;&lt;p data-block-key="j77gm"&gt;As a result, many banks are trying to figure out how to offer their customers the benefits of stablecoins — high speed, low cost, 24/7 availability, and programmability — without introducing problems often associated with existing reserve-backed stablecoins, such as regulatory issues, poor throughput performance, negative influence on bank reserves, and potentially no yield participation.&lt;/p&gt;&lt;h3 data-block-key="dj2t5"&gt;&lt;b&gt;Solution proposal&lt;/b&gt;&lt;/h3&gt;&lt;p data-block-key="8jeai"&gt;Realizing this vision means modernized payments architectures must seamlessly and instantly enable individuals and institutions to transfer bank-held assets to accounts at any target bank. Orchestrating this shared ecosystem requires central entities to establish the governing rulebook, enforce issuance limits, and manage network-wide settlement parameters. During its pilot, Crypto Finance assumed this role, which within GCUL is known as the “currency operator.”&lt;/p&gt;&lt;p data-block-key="2ca3k"&gt;In this position, Crypto Finance can configure the network for either instant or deferred settlement. This choice is analogous to traditional banking: instant settlement acts like real-time gross settlement, where transactions settle individually and immediately and require high liquid reserves. In contrast, deferred settlement operates like netting, where payments are accumulated and offset and only periodically settle the net difference. For the pilot, Crypto Finance chose to adopt deferred settlement, in which GCUL manifests capital efficiency. In this way, clients enjoy instant transfers while banks conserve liquidity through periodic settlement.&lt;/p&gt;&lt;h3 data-block-key="5p3hh"&gt;&lt;b&gt;Concept validation&lt;/b&gt;&lt;/h3&gt;&lt;p data-block-key="6pqch"&gt;To demonstrate instant, bi-directional account-to-account transfers on GCUL, Crypto Finance orchestrated a pilot with two digital token issuers, a Swiss digital asset bank, and a Swiss B2B transaction bank. Crypto Finance established the rulebook for interactions between the participant banks, such as interbank settlement, integration with traditional SIC payment rails, and AML/compliance protocols. Although some backend steps still needed manual intervention, Crypto Finance’s pilot proved that fully automated, cross-institution integration is feasible and established a generalized architecture that can scale to virtually any asset class.&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/swiss-franc-stablecoin-gcul-2.max-1000x1000.png"
        
          alt="swiss-franc-stablecoin-gcul-2"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph"&gt;&lt;h3 data-block-key="j77gm"&gt;&lt;b&gt;Robustness and performance&lt;/b&gt;&lt;/h3&gt;&lt;p data-block-key="9mke5"&gt;A temporary pilot is one thing; moving into live client transactions carrying legal liability brings a whole new set of operational challenges. Live payment rails are considered critical infrastructure that demands security and availability. GCUL addresses these needs through redundancy between geographically distributed nodes operating on an enterprise-grade consensus model, ensuring reliability. Deploying client workloads on Google Cloud provides that enterprise-grade foundation and allows financial institutions to pair the highly resilient GCUL network with secure-by-design infrastructure, encryption at rest and in transit, and robust access controls that satisfy many of banks’ common regulatory, privacy, and data sovereignty mandates. While some banks may consider integrating their systems with fully public and permissionless blockchains, this approach introduces severe regulatory hurdles, anonymity risks, unpredictable gas fees, and low throughput. These are all roadblocks that in the past would have significantly delayed the pilot or rendered it a non-starter.&lt;/p&gt;&lt;h3 data-block-key="4li96"&gt;&lt;b&gt;Ease of integration&lt;/b&gt;&lt;/h3&gt;&lt;p data-block-key="4jrd8"&gt;The pilot's implementation, with systems integration completed in under three months, was compartmentalized across dedicated Google Cloud projects per participant. The currency operator, Crypto Finance, connected specialized components to a dedicated GCUL API. The first was a role admin service that controls account managers and token managers on the network. Next, it implemented a settlement registry to record when periodic off-ledger settlement between participant banks has occurred once a predefined threshold is reached. Finally, a cloud key management service (KMS) stored the currency operator keys to sign transactions.&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/swiss-franc-stablecoin-gcul-3.max-1000x1000.png"
        
          alt="swiss-franc-stablecoin-gcul-3"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph"&gt;&lt;p data-block-key="j77gm"&gt;At the same time, participant banks connected their core banking systems to their own GCUL API instances. But these participant banks don’t need a large Google Cloud footprint. Instead, they can connect to GCUL nodes via Remote Procedure Call from their existing on-premise deployments and manage private key material through solutions other than KMS. Regardless of the setup, transactions are seamlessly routed to the shared global GCUL nodes.&lt;/p&gt;&lt;p data-block-key="5g9o"&gt;Whether GCUL is exposed to a bank’s client or used merely as a bank backend is entirely at the integrating institution’s discretion. The banks participating in the pilot chose the latter, utilizing the ledger to upgrade their backend infrastructure. With GCUL under the hood, they improved intra-day transaction performance and provided customers with the user experience they were already accustomed to.&lt;/p&gt;&lt;h3 data-block-key="2rknj"&gt;&lt;b&gt;Regulatory compliance and governance by design&lt;/b&gt;&lt;/h3&gt;&lt;p data-block-key="5l6gl"&gt;To join the Swiss Franc currency network on GCUL, participants must have regulatory licenses that let them perform third-party transfers and issue commercial bank money. Crucially, features like token and account management are not rigid network-wide limitations. Instead, they rely entirely on the currency operator and its specific rulebook. This standardizes fees, limits, settlement times, and technical configurations.&lt;/p&gt;&lt;p data-block-key="9tt8c"&gt;Managing counterparty risk is central to this governance. Rather than relying on unscalable bilateral agreements or complex clearinghouse default waterfalls, GCUL mitigates risk through these currency operator-enforced issuance limits.&lt;/p&gt;&lt;p data-block-key="d3opm"&gt;It’s important to distinguish GCUL’s broader feature set from the concrete pilot setup. In the pilot, both participating banks operated with conservative initial limits based on their credit ratings in order to safely prove the architecture. But beyond the pilot, the network frees up capital by periodically netting outstanding obligations, which frees banks from having to fund and settle every transaction.&lt;/p&gt;&lt;p data-block-key="b15si"&gt;Looking ahead, risk management can be further innovated. One clever future enhancement would be to allow participants to dynamically increase their issuance limits by committing digital assets as collateral in segregated accounts, using solutions like Crypto Finance’s &lt;a href="https://www.crypto-finance.com/anchornote/" target="_blank"&gt;AnchorNote&lt;/a&gt; pledging solution. These assets could then be used to instantly settle outstanding street-side obligations, virtually eliminating counterparty risk without tying up traditional liquidity reserves.&lt;/p&gt;&lt;h3 data-block-key="19sr7"&gt;&lt;b&gt;The future&lt;/b&gt;&lt;/h3&gt;&lt;p data-block-key="20ig6"&gt;While Crypto Finance’s pilot was designed to focus on domestic commercial bank money transfers in Switzerland, GCUL's programmable architecture and high-performance smart contract language, GCULpy, are paving the way for deeper innovation. As institutions like Crypto Finance scale their use of GCUL, the network can expand to support real-time cross-currency settlement, automated foreign exchange, and dynamic collateral management involving tokenized digital assets. And looking further ahead, this infrastructure has laid the groundwork for next-generation financial services that range from point-of-sale payments to autonomous, AI-driven agentic payments.&lt;/p&gt;&lt;p data-block-key="6ll23"&gt;The pilot proved that deploying high-performance, blockchain-based payment rails doesn’t mean compromising on strict regulatory compliance or core banking stability. By combining the high-throughput of GCUL with a secure, API-driven integration architecture, commercial banks can offer stablecoin-like speed to their customers while seamlessly preserving their capital efficiency. Ultimately, this compliant-by-design approach targets the operational bottlenecks of traditional finance while mitigating the inherent risks and complexities of public blockchains.&lt;/p&gt;&lt;p data-block-key="5t8hr"&gt;To learn how your institution can build on this blueprint to fundamentally upgrade your operational payments infrastructure, contact Google Cloud Sales to explore the Google Cloud Universal Ledger.&lt;/p&gt;&lt;p data-block-key="ab5su"&gt;&lt;i&gt;The team would also like to thank Phil Dettwiler, Andras Miklos, Ignazio Pollina, and Ozan Kizildogan at Crypto Finance and Googlers Chris Esch, Ishan Singh, Olena Clayton, and Stathis Onasoglou for their contributions.&lt;/i&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Mon, 26 Oct 2026 20:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/topics/financial-services/easy-to-integrate-compliant-by-design-bringing-the-swiss-franc-to-google-cloud-universal-ledger/</guid><category>AI &amp; Machine Learning</category><category>Customers</category><category>Financial Services</category><media:content height="540" url="https://storage.googleapis.com/gweb-cloudblog-publish/images/swiss-franc-stablecoin-gcul-2_3lwzLOd.max-600x600.png" width="540"></media:content><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>Easy to integrate, compliant by design: Bringing the Swiss Franc to Google Cloud Universal Ledger</title><description></description><image>https://storage.googleapis.com/gweb-cloudblog-publish/images/swiss-franc-stablecoin-gcul-2_3lwzLOd.max-600x600.png</image><site_name>Google</site_name><url>https://cloud.google.com/blog/topics/financial-services/easy-to-integrate-compliant-by-design-bringing-the-swiss-franc-to-google-cloud-universal-ledger/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Nathaniel Zollinger</name><title>CTO, Crypto Finance Group</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Moritz Platt</name><title>Capital Markets Technology Manager</title><department></department><company></company></author></item><item><title>ZenteiQ scales batch inference 4x with GKE Inference Gateway powered by llm-d</title><link>https://cloud.google.com/blog/products/containers-kubernetes/zenteiq-scales-physical-ai-inference-4x-with-gke-and-llm-d/</link><description>&lt;div class="block-paragraph"&gt;&lt;p data-block-key="3d0iw"&gt;&lt;i&gt;Today we hear from&lt;/i&gt; &lt;a href="https://zenteiq.ai/" target="_blank"&gt;&lt;i&gt;ZenteiQ&lt;/i&gt;&lt;/a&gt;&lt;i&gt;, an artificial intelligence startup spun out of the Indian Institute of Science (IISc) and selected under the&lt;/i&gt; &lt;a href="https://indiaai.gov.in/" target="_blank"&gt;&lt;i&gt;IndiaAI Mission&lt;/i&gt;&lt;/a&gt;&lt;i&gt; to build sovereign scientific foundation models from the ground up. To pre-train models that understand physical laws, governing equations, and numerical solvers, ZenteiQ designed and built a large-scale batch synthetic data-generation pipeline that runs across multiple model families and feeds verified datasets into training runs.&lt;/i&gt; &lt;a href="https://cloud.google.com/kubernetes-engine"&gt;&lt;i&gt;Google Kubernetes Engine (GKE)&lt;/i&gt;&lt;/a&gt;&lt;i&gt; and&lt;/i&gt; &lt;a href="https://cloud.google.com/kubernetes-engine/docs/concepts/about-gke-inference-gateway"&gt;&lt;i&gt;GKE Inference Gateway&lt;/i&gt;&lt;/a&gt;&lt;i&gt; gave ZenteiQ the inference-routing layer within its pipeline to scale aggregate inference throughput 4.07x and grow daily output from 1.47 billion to 5.97 billion generated tokens for every 1,000 accelerators.&lt;/i&gt;&lt;b&gt; - Fisayo Feyisetan, Product Manager, Google Cloud&lt;/b&gt;&lt;/p&gt;&lt;p data-block-key="acnlq"&gt;Unlike general-purpose language models trained primarily on web text and code, scientific foundation models must understand complex physical systems, mathematical governing equations, and numerical solvers across domains such as thermodynamics, computational fluid dynamics, and electromagnetics. At ZenteiQ, we are building sovereign scientific foundation models from scratch that interpret simulation outputs, evaluate equation formulations, and work through step-by-step mathematical derivations that obey physical laws.&lt;/p&gt;&lt;p data-block-key="62ged"&gt;Pre-training models on large datasets of governing equations takes hundreds of billions of reasoning tokens. Alongside curated scientific literature and simulation datasets, we also rely on a batch synthetic data-generation pipeline that we independently designed and implemented to generate, evaluate, and filter candidate derivations and solver validation problems before feeding them into our &lt;a href="https://github.com/AI-Hypercomputer/maxtext" target="_blank"&gt;MaxText&lt;/a&gt; pre-training jobs.&lt;/p&gt;&lt;p data-block-key="bqp5k"&gt;Because we operate on a fixed compute allocation, every hour our accelerators spend generating and verifying training data is an hour taken away from pre-training. When data generation falls behind, multi-host accelerator clusters reserved for pre-training sit idle, waiting on datasets.&lt;/p&gt;&lt;h3 data-block-key="3ud5h"&gt;&lt;b&gt;When data generation slows down pre-training&lt;/b&gt;&lt;/h3&gt;&lt;p data-block-key="14k3c"&gt;To maximize data generation speed, our pipeline separates prompt preparation on worker pods from model serving on vLLM across multi-node accelerator pools. Each batch moves through four stages: expanding scientific seed concepts into prompt batches, generating candidate reasoning traces with generator models, scoring responses using evaluator models, and running guardrail checks with a separate model family before packaging verified datasets for pre-training.&lt;/p&gt;&lt;p data-block-key="69j1"&gt;We originally ran these model pools on Kubernetes with standard ClusterIP round-robin load balancing across inference endpoints. As we scaled our pipeline to feed our 9B- and 35B-parameter model schedules, two routing bottlenecks began slowing down dataset creation.&lt;/p&gt;&lt;ul&gt;&lt;li data-block-key="f2nig"&gt;&lt;b&gt;Uneven engine queues:&lt;/b&gt; Our upstream worker pods expand seed concepts into prompts of widely varying complexity and response length. Because standard round-robin load balancing had no visibility into vLLM engine queues, it kept sending new requests to serving pods that were already busy processing long reasoning traces, while neighboring nodes had capacity to spare.&lt;/li&gt;&lt;li data-block-key="5l1kp"&gt;&lt;b&gt;Redundant prefix computation:&lt;/b&gt; Prompts within a generation batch share roughly 70% to 80% of their instructional context and task specifications. Round-robin routing scattered requests with identical instructions across different serving instances, forcing each pod to recompute the same shared prompt prefix from scratch instead of reusing warm caches in accelerator memory.&lt;/li&gt;&lt;/ul&gt;&lt;p data-block-key="utgm"&gt;Together, uneven queues and redundant prefill work caused our generation throughput to swing by as much as half within a single run, stretching out dataset creation and delaying the start of scheduled pre-training runs. We needed routing that understood live vLLM queue depths and prompt caches, without pulling our researchers off model development to build and maintain a custom inference scheduler.&lt;/p&gt;&lt;h3 data-block-key="btfc1"&gt;&lt;b&gt;Building on GKE Inference Gateway powered by llm-d&lt;/b&gt;&lt;/h3&gt;&lt;p data-block-key="393mm"&gt;We already run our data preparation and MaxText pre-training jobs on GKE, which lets our team manage data generation and model training on the same Kubernetes platform. Following a recommendation from Google Cloud to address our routing bottlenecks without changing our underlying scientific methodology or pipeline design, we deployed &lt;a href="https://cloud.google.com/kubernetes-engine/docs/concepts/about-gke-inference-gateway"&gt;GKE Inference Gateway&lt;/a&gt;, powered by the open-source &lt;a href="https://llm-d.ai/" target="_blank"&gt;llm-d&lt;/a&gt; distributed inference framework, in front of our vLLM accelerator pools.&lt;/p&gt;&lt;p data-block-key="7vsot"&gt;In our pipeline, worker pods pull scientific seed documents and domain taxonomies from Cloud Storage and expand them into batches of multi-step prompts. Those batches flow through a single GKE Inference Gateway deployment that manages traffic across all of our model families and serving nodes.&lt;/p&gt;&lt;p data-block-key="4b7b3"&gt;Rather than maintaining isolated routing for each model tier, we route generation requests, evaluation queries, and guardrail checks across their respective model pools through the same gateway. Once candidate responses pass evaluation and guardrail checks, worker pods deduplicate and package the verified datasets for direct consumption by downstream MaxText pre-training jobs on GKE.&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/image1_YjgtWZi.max-1000x1000.png"
        
          alt="image1"&gt;
        
        &lt;/a&gt;
      
        &lt;figcaption class="article-image__caption "&gt;&lt;p data-block-key="9v0bp"&gt;GKE Inference Gateway routes batch generation, evaluation, and guardrail requests across multi-node accelerator pools running vLLM before verified datasets are fed into MaxText pre-training.&lt;/p&gt;&lt;/figcaption&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph"&gt;&lt;h3 data-block-key="1vzmh"&gt;&lt;b&gt;From round-robin to load- and cache-aware routing&lt;/b&gt;&lt;/h3&gt;&lt;p data-block-key="82k62"&gt;With GKE Inference Gateway and llm-d, routing decisions use live serving signals from our vLLM instances rather than static round-robin rotation. GKE Inference Gateway continuously monitors real-time queue depth and key-value (KV) cache utilization across our vLLM endpoints. When a serving node is busy decoding a long equation derivation, the gateway routes incoming requests to nodes with open capacity, so heavy prompts no longer pile up behind busy workers while neighboring nodes sit idle.&lt;/p&gt;&lt;p data-block-key="7gl5a"&gt;At the same time, the gateway routes requests with matching scientific instructions to the serving node where that prefix’s KV cache is already warm in accelerator memory. Because 70% to 80% of each prompt is shared instructional context while the generated reasoning trajectories vary from request to request, combining prefix-cache locality with live queue balancing lets our pods skip repeated prefill work without overloading any single replica.&lt;/p&gt;&lt;h3 data-block-key="adli2"&gt;&lt;b&gt;Higher throughput, predictable training runs&lt;/b&gt;&lt;/h3&gt;&lt;p data-block-key="2sss5"&gt;Adopting GKE Inference Gateway powered by llm-d transformed the speed and predictability of our data generation pipeline. Across our production clusters, intelligent routing delivered:&lt;/p&gt;&lt;ul&gt;&lt;li data-block-key="5kv3n"&gt;&lt;b&gt;4x higher fleet throughput:&lt;/b&gt; Aggregate throughput jumped from 16,964 to 69,114 tokens per second per 1,000 accelerators — growing daily output from &lt;b&gt;1.47 billion to nearly 6 billion generated tokens&lt;/b&gt;.&lt;/li&gt;&lt;li data-block-key="809dp"&gt;&lt;b&gt;3.5x to 5.5x gains across individual model pools:&lt;/b&gt; Real-time queue balancing kept serving nodes evenly loaded throughout each run, preventing the mid-run slowdowns caused by long reasoning derivations.&lt;/li&gt;&lt;li data-block-key="5cdue"&gt;&lt;b&gt;14-percentage-point higher prefix-cache hit rate:&lt;/b&gt; Routing prompts with matching scientific instructions to warm caches avoided redundant prefill work before each request branched into its own derivation.&lt;/li&gt;&lt;/ul&gt;&lt;p data-block-key="a3t17"&gt;For our research and infrastructure teams, the most important outcome was schedule predictability. Quadrupling daily output meant verified datasets were ready on time, so our multi-host pre-training runs started on schedule instead of waiting on data generation.&lt;/p&gt;&lt;h3 data-block-key="4m801"&gt;&lt;b&gt;What’s next for physical AI at ZenteiQ&lt;/b&gt;&lt;/h3&gt;&lt;p data-block-key="cglin"&gt;As we scale our scientific foundation models from 9B to 35B parameters and beyond, we plan to use the same GKE Inference Gateway and llm-d routing setup for post-training reinforcement learning (RL) and automated numerical solver evaluation loops. Because each turn in those multi-turn rollouts builds on the context of the turns before it, they depend even more heavily on live queue balancing and warm prefix caches.&lt;/p&gt;&lt;p data-block-key="75tc5"&gt;To try these capabilities on your own workloads, explore the &lt;a href="https://cloud.google.com/kubernetes-engine/docs/concepts/about-gke-inference-gateway"&gt;GKE Inference Gateway documentation&lt;/a&gt; and check out the open-source &lt;a href="https://llm-d.ai/" target="_blank"&gt;llm-d project&lt;/a&gt;. You can also read how the Gemini platform team &lt;a href="https://cloud.google.com/blog/products/containers-kubernetes/how-gke-inference-gateway-improved-latency-for-vertex-ai"&gt;cut latency by 35% with GKE Inference Gateway&lt;/a&gt;.&lt;/p&gt;&lt;/div&gt;</description><pubDate>Wed, 21 Oct 2026 16:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/containers-kubernetes/zenteiq-scales-physical-ai-inference-4x-with-gke-and-llm-d/</guid><category>GKE</category><category>AI infrastructure</category><category>Customers</category><category>llm-d</category><category>Containers &amp; Kubernetes</category><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>ZenteiQ scales batch inference 4x with GKE Inference Gateway powered by llm-d</title><description></description><site_name>Google</site_name><url>https://cloud.google.com/blog/products/containers-kubernetes/zenteiq-scales-physical-ai-inference-4x-with-gke-and-llm-d/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Prof. Sashikumaar Ganesan</name><title>Founder, ZenteiQ</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Fisayo Feyisetan</name><title>Product Manager, Google Kubernetes Engine</title><department></department><company></company></author></item><item><title>Building full virtual vehicles on Google Cloud: From AAOS to the smallest MCU</title><link>https://cloud.google.com/blog/topics/manufacturing/building-full-virtual-vehicles-on-google-cloud-from-aaos-to-the-smallest-mcu/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Imagining traditional gearheads getting their hands on software-defined vehicles is an amusing thought experiment. With their socket wrenches and air compressors at the ready, they might be confused to encounter core functions more likely to run on 1s and 0s than oil and belts. But across the auto industry these days, more and more vehicles feature prominent systems built using cloud-native development, continuous integration/continuous delivery, fast digital iteration, and plenty of AI.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;But there’s still a big gap to be found in many automakers’ and Tier 1 suppliers’ cloud-based development processes. On the one hand, these companies are commonly using software to develop vehicles’ digital entertainment and information screens, or other onboard high-performance computers. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;On the other, the dozens of microchips and control modules that make a car a car these days — such as gateways, body controllers, zone control units, sensor and actuator nodes — are still frequently hardwired to physical chips when they’re constructed, making them very difficult to update, reprogram, or expand after the vehicle drives off the factory floor. Teams have to wait for benches, samples, and rigs, and integration can easily slip to the end of the development process, where defects are most expensive to fix.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This inherent inefficiency has become a leading bottleneck. After all, manufacturers can’t really claim vehicles are software-defined if most of their electronics only function when connected to physical hardware and if developers are leaning heavily on actual hardware.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With Horizon, our agentic innovation platform for developing software-defined vehicles, our vision is to help manufacturers avoid these inefficiencies by working with partners like &lt;/span&gt;&lt;a href="https://www.remotivelabs.com/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;RemotiveLabs&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;The cockpit already runs in the cloud&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The Android team has invested a significant amount of time and effort over the last few years on enabling seamless development of the many digital systems in vehicles’ cockpits. &lt;/span&gt;&lt;a href="https://source.android.com/docs/automotive" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Android Automotive OS&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; (AAOS) runs inside Google’s official virtual testing environment for Android, &lt;/span&gt;&lt;a href="https://source.android.com/docs/devices/cuttlefish" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Cuttlefish&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, which itself runs comfortably on&lt;/span&gt; &lt;a href="https://cloud.google.com/compute/docs/instances/arm-on-compute"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Arm-based Google Cloud instances&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. With this approach, developers can build, boot, and test virtual versions of their infotainment stacks on the same Arm architecture that ships in vehicles.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;And with the recent launch of &lt;/span&gt;&lt;a href="https://source.android.com/docs/whatsnew/android-17-release#sdv" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Android 17 and AAOS SDV&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, this capability is also extended to further domains, whenever AAOS SDV is used. Deploying it with Cuttlefish, for example, proves the model: representative software, on representative architecture, in the cloud, on demand. The question that still needs to be answered is how to extend that model to the rest of the vehicle, especially when other operating systems are involved.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;The harness around the virtual vehicle&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;That’s where &lt;/span&gt;&lt;a href="https://www.remotivelabs.com/products/remotivetopology" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;RemotiveTopology&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; comes into play. As the harness around the virtual vehicle, it’s the layer that connects AAOS in Cuttlefish to a full, simulated electrical and electronic (E/E) architecture: the rest of the vehicle’s onboard computers, its data networks, and the live signals that make the system function like a real vehicle rather than a collection of standalone apps.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;RemotiveTopology automatically utilizes manufacturers’ existing network design files to build a working, virtual version of their cars’ networks. This is vehicle-as-code: the E/E architecture expressed as an executable specification, versioned and provisioned like infrastructure-as-code. In this way, an entire vehicle can be spun up on demand. Cockpit software talks to the rest of the vehicle exactly as it would in a real vehicle. As a result, developers get full and continuous signal and cross-domain visibility, so they can see real-time system status. And because it is built on standard Linux tooling, the same topology runs on a laptop and in a contiguous integration pipeline on Google Cloud.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Integrate across domains early, not late&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The payoff for using this approach is that all onboard systems can be developed and integrated with software before any of them are connected to a vehicle’s hardware. That’s beneficial because the most difficult (and latest) integration in a vehicle’s development is bringing all its systems together. For example, advanced driver assistance systems (ADAS) produce object lists, trajectories, and mode requests that the cockpit has to render and react to, while the infotainment system and information cluster have to reflect driving state while body and chassis nodes sit in the loop.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With the full topology live on Google Cloud, the infotainment system and ADAS can operate simultaneously within a virtual vehicle, exchanging real signals over the simulated network. This means it’s possible to exercise handoffs like an ADAS event surfacing in the cluster or a driver-mode change rippling across domains from day one in continuous integration long before the benches and samples line up. In this way, developers can be confident they’ll discover multi-domain integration long before the end of the process so that they can test continuously from the start.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;The full vehicle frontier&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;At the heart of this work is a shared desire to bring scalable software development to even the smallest and most common nodes in a vehicle, its microchips and control modules. That’s because these are the parts that have traditionally been wired to hardware, and as a result are the biggest blocker to fully software-defined vehicles.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The ultimate goal is that the same software that runs in the cloud can run on vehicles’ actual control modules with no re-tooling or re-validation needed to move between them. That parity is natural with Arm-based Google Cloud infrastructure because the cloud and the car share an instruction set. Google's &lt;/span&gt;&lt;a href="https://cloud.google.com/products/axion"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Axion&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; makes it straightforward:&lt;/span&gt; &lt;a href="https://cloud.google.com/blog/products/compute/axion-based-n4a-vms-now-in-preview"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;C4A metal, a bare-metal Arm instance&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, is provided explicitly for automotive systems development and Android software development, giving direct hardware access without unnecessary software layers and delivering the consistent performance that vehicle systems require.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Taken together, this holistic system allows teams, for the first time, to develop and integration-test an entire vehicle’s systems in the cloud and then apply it to the actual vehicle without having to rewrite it.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Why now, and where to start&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The payoff is not abstract. Treating the vehicle as code and moving integration into the cloud compresses the slowest, most expensive phase of a program. Early customer engagements with RemotiveLabs have measured more than 10x faster integration cycles. And the development loop is now changing faster than the hardware loop: AI-assisted and agentic workflows can rapidly generate, test, and refine vehicle software, but only virtual systems provide the required scale. Every element that can’t be virtualized can slow the entire development process down.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;RemotiveTopology provides this architecture as vehicle-as-code and makes it accessible for developers, testers, and autonomous AI agents alike. &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;Best of all, there is no new third-party tooling to adopt just to get started. Instead, RemotiveLabs publishes an&lt;/span&gt; &lt;a href="https://github.com/remotivelabs/remotivelabs-topology-examples" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;open-source simplified vehicle on GitHub&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; that anyone can run in minutes on standard Linux tooling and platform definitions already in place. Full existing toolchains plug in as a manufacturer or Tier1 supplier scales, and RemotiveTopology can be easily dropped into an Arm-based Google Cloud environment.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;For those building software-defined vehicles whose virtual development stops at the cockpit, it’s time to start with an open-source example, bring unique platform definitions, and run an entire vehicle in the cloud. &lt;/span&gt;&lt;a href="https://github.com/GoogleCloudPlatform/horizon-sdv/tree/main/gitops/modules/remotive-topology" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;RemotiveTopology is available in Horizon&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; for easy usage in combination with other Horizon modules.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Mon, 19 Oct 2026 16:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/topics/manufacturing/building-full-virtual-vehicles-on-google-cloud-from-aaos-to-the-smallest-mcu/</guid><category>AI &amp; Machine Learning</category><category>Customers</category><category>Manufacturing</category><media:content height="540" url="https://storage.googleapis.com/gweb-cloudblog-publish/images/virtual-vehicles-on-Google-Cloud-header.max-600x600.png" width="540"></media:content><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>Building full virtual vehicles on Google Cloud: From AAOS to the smallest MCU</title><description></description><image>https://storage.googleapis.com/gweb-cloudblog-publish/images/virtual-vehicles-on-Google-Cloud-header.max-600x600.png</image><site_name>Google</site_name><url>https://cloud.google.com/blog/topics/manufacturing/building-full-virtual-vehicles-on-google-cloud-from-aaos-to-the-smallest-mcu/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Florian Haubner</name><title>Industry Architect Lead Automotive</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Per Sigurdson</name><title>CEO &amp; co-founder, RemotiveLabs</title><department></department><company></company></author></item><item><title>Architecting autonomy: A design pattern for high-compute, governed agents in biotech</title><link>https://cloud.google.com/blog/topics/healthcare-life-sciences/architecting-scalable-agentic-biotech-workloads/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;In production systems today, most AI agents operate as lightweight dispatchers. They parse a prompt, make an API call, retrieve a record, and hand back control. Their execution is fast, stateless and cheap. Biotechnology does not work this way. &lt;/span&gt;&lt;/p&gt;
&lt;p style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;Folding a protein target, screening a 50,000 compound ligand library, or evaluating structure-activity relationships (SAR) are not quick API lookups. They are sustained, stateful, GPU-bound simulations that run for minutes or even hours across specialized silicon. When you ask a general-purpose large language model (LLM) endpoint to coordinate a complex drug discovery workflow, the standard paradigm breaks: Timeouts occur, context windows overflow with atomic coordinates, and proprietary chemical Intellectual Property (IP) risks exposure to external networks. &lt;/span&gt;&lt;/p&gt;
&lt;p style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;Building enterprise grade research and development (R&amp;amp;D) agents requires moving past the monolithic assistant model. In this blog post, we will discuss how to use a three-layer architecture: separating cognitive orchestration (the brain) from computational execution (the muscle), and anchoring both inside an enterprise governance plane to build scalable agents for drug discovery use cases in biotech. &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-video"&gt;



&lt;div class="article-module article-video "&gt;
  &lt;figure&gt;
    &lt;a class="h-c-video h-c-video--marquee"
      href="https://youtube.com/watch?v=TZTDqtSpzv8"
      data-glue-modal-trigger="uni-modal-TZTDqtSpzv8-"
      data-glue-modal-disabled-on-mobile="true"&gt;

      
        &lt;img src="//img.youtube.com/vi/TZTDqtSpzv8/maxresdefault.jpg"
             alt="AI Agent Infrastructure Decoded"/&gt;
      
      &lt;svg role="img" class="h-c-video__play h-c-icon h-c-icon--color-white"&gt;
        &lt;use xlink:href="#mi-youtube-icon"&gt;&lt;/use&gt;
      &lt;/svg&gt;
    &lt;/a&gt;

    
  &lt;/figure&gt;
&lt;/div&gt;

&lt;div class="h-c-modal--video"
     data-glue-modal="uni-modal-TZTDqtSpzv8-"
     data-glue-modal-close-label="Close Dialog"&gt;
   &lt;a class="glue-yt-video"
      data-glue-yt-video-autoplay="true"
      data-glue-yt-video-height="99%"
      data-glue-yt-video-vid="TZTDqtSpzv8"
      data-glue-yt-video-width="100%"
      href="https://youtube.com/watch?v=TZTDqtSpzv8"
      ng-cloak&gt;
   &lt;/a&gt;
&lt;/div&gt;

&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3 style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;A three-layer pattern&lt;/span&gt;&lt;/h3&gt;
&lt;p style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;The reference architecture below shows how to architect, optimize, and govern high-compute autonomous agents in biotechnology, specifically Drug Discovery. &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/Figure_1_TukBnfH.max-1000x1000.jpg"
        
          alt="Figure 1"&gt;
        
        &lt;/a&gt;
      
        &lt;figcaption class="article-image__caption "&gt;&lt;p data-block-key="4nr3v"&gt;Figure 1 — The three-layer pattern: a cognitive orchestrator (brain), a scalable GPU execution layer (muscle), and an enterprise governance plane.&lt;/p&gt;&lt;/figcaption&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h4&gt;&lt;span style="vertical-align: baseline;"&gt;The execution layer: scaling the muscle - high-compute execution&lt;/span&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt; &lt;/span&gt;&lt;/h4&gt;
&lt;p style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;The execution layer is responsible for the heavy scientific computations. The first architectural decision is determining where workloads should run:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation" style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;Strategic, proprietary pipelines require dedicated GPU infrastructure within your security perimeter&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation" style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;Routine or standardized workloads can safely be delegated to managed cloud endpoints&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;In this architecture, protein folding is driven by &lt;/span&gt;&lt;a href="https://catalog.ngc.nvidia.com/orgs/nvidia/teams/nim/collections/bionemo-esmfold" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;ESMFold&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and generative design by &lt;/span&gt;&lt;a href="https://docs.nvidia.com/nim/bionemo/genmol/latest/overview.html" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;GenMol&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. Both run as &lt;/span&gt;&lt;a href="https://www.nvidia.com/en-us/ai-data-science/products/nim-microservices/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;NVIDIA Inference Microservices&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; or NIMs deployed on &lt;/span&gt;&lt;a href="https://cloud.google.com/kubernetes-engine?hl=en"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google Kubernetes Engine&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; (GKE) using G4 VMs backed by &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/compute/introducing-g4-vm-with-nvidia-rtx-pro-6000"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;NVIDIA RTX PRO 6000 Blackwell&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; GPUs.&lt;/span&gt;&lt;/p&gt;
&lt;p style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;A critical hurdle in containerized GPU workloads is the cold-start penalty, the latency delay required to pull multi-gigabyte model weights from an image registry and stage them into GPU memory (VRAM).&lt;/span&gt;&lt;/p&gt;
&lt;p style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;To ensure low latency responsiveness, we build the following as a recommendation into the reference architecture:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation" style="text-align: justify;"&gt;&lt;strong style="vertical-align: baseline;"&gt;Maintain warm pools:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Pin the GKE node pool to a minimum of one active node.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation" style="text-align: justify;"&gt;&lt;strong style="vertical-align: baseline;"&gt;Preload weights:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Model weights are loaded directly into VRAM at pod initialization, ensuring containers are immediately ready for inference.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation" style="text-align: justify;"&gt;&lt;strong style="vertical-align: baseline;"&gt;Resilient scheduling:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; GKE manages pod lifecycles and restarts failed instances without tearing down the underlying VM infrastructure. &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;This approach trades custom, manual CUDA kernel tuning for NVIDIA's pre-compiled, containerized runtimes, dramatically reducing infrastructure engineering overhead while ensuring deterministic throughput.&lt;/span&gt;&lt;/p&gt;
&lt;p style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;In biotech, chemical structures represent millions of dollars in proprietary IP. A docking request transmits SMILE strings or 3D coordinate matrices that must never leave company custody.&lt;/span&gt;&lt;/p&gt;
&lt;p style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;Containerized NIMs eliminate the friction between rapid prototyping and rigorous data isolation. During early exploration, researchers can query NVIDIA hosted endpoints. When transitioning to proprietary lead candidates, the exact same container is deployed to GKE inside a private &lt;/span&gt;&lt;a href="https://cloud.google.com/vpc?hl=en"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;VPC&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. The agent’s orchestration logic remains identical; only the target URI updates from an external proxy to an internal VPC endpoint. No structural data leaves the enterprise perimeter.&lt;/span&gt;&lt;/p&gt;
&lt;h4&gt;&lt;span style="vertical-align: baseline;"&gt;The cognitive layer: Multi-agent reasoning&lt;/span&gt;&lt;/h4&gt;
&lt;p style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;With compute execution isolated, the cognitive layer manages agentic workflow sequencing. Built on &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/gemini-enterprise-agent-platform/build/runtime"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Agent Runtime&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, within the Gemini platform, this reference architecture shows a system that deploys a hierarchical multi-agent structure, one lead orchestrator, and two specialized domain sub-agents (Lead Optimization and Safety Screening) . The multi-agent design can be modified to suit the use case at hand. &lt;/span&gt;&lt;/p&gt;
&lt;p style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;Based on this design, when an end-user submits a high-level task such as &lt;/span&gt;&lt;strong style="font-style: italic; vertical-align: baseline;"&gt;“ Analyze Candidate 4b against Target X and produce a SAR report”&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, the system applies four core orchestration patterns discussed below; &lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li role="presentation" style="text-align: justify;"&gt;&lt;strong style="vertical-align: baseline;"&gt;In-process native agent handoffs: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;The lead orchestrator delegates tasks to specialist sub-agents using Agent Runtime’s native SDK rather than network-serialized agent-to-agent HTTP protocols. This eliminates intermediate network hops and keeps internal handoffs instantaneous. Only the outermost boundary communicates outward to user-facing surfaces like the &lt;/span&gt;&lt;a href="https://cloud.google.com/gemini-enterprise?utm_source=google&amp;amp;utm_medium=cpc&amp;amp;utm_campaign=1713762-Gemini_Enterprise-DR-NA-US-en-Google-BKWS-EXA-GEnterprise&amp;amp;utm_content=c-Hybrid+%7C+BKWS+-+MIX+%7C+Txt_Gemini+Enterprise-189528400785&amp;amp;utm_term=gemini+enterprise&amp;amp;gclsrc=aw.ds&amp;amp;gad_source=1&amp;amp;gad_campaignid=23370621055&amp;amp;gclid=CjwKCAjwl97RBhBWEiwAa9rbXeetAjZcIRGcaAR57VWVWk7ik5DJ9ESqHiDav-2ervoiyzzRZ1ZWyRoCj6sQAvD_BwE&amp;amp;hl=en"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Gemini&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; agent.&lt;/span&gt;&lt;/li&gt;
&lt;li role="presentation" style="text-align: justify;"&gt;&lt;strong style="vertical-align: baseline;"&gt;Passing bulky payloads by reference: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;LLMs degrade quickly when forced to ingest massive arrays of 3D spatial coordinates. In this architecture, raw structural data (PDB files, docking poses, trajectory logs) is written directly to &lt;/span&gt;&lt;a href="https://cloud.google.com/storage?utm_source=google&amp;amp;utm_medium=cpc&amp;amp;utm_campaign=Cloud-SS-DR-GCP-1713658-GCP-DR-NA-US-en-Google-SKWS-BRO-storage&amp;amp;utm_content=c-Hybrid+%7C+SKWS+-+BRO+%7C+Txt-Infrastructure-Storage-Cloud+Storage-4642545472&amp;amp;utm_term=cloud+storage&amp;amp;gclsrc=aw.ds&amp;amp;gad_source=1&amp;amp;gad_campaignid=23767162264&amp;amp;gclid=CjwKCAjwl97RBhBWEiwAa9rbXYcqs8qBJehBeWaezdgKd4am7QbV7uo3__CJIZqoa8lqgP8ErM0LkBoCJ3cQAvD_BwE&amp;amp;hl=en"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud Storage&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. Agents pass lightweight, signed object URIs and high-level structural summaries, keeping context windows compact and inference costs down.&lt;/span&gt;&lt;/li&gt;
&lt;li role="presentation" style="text-align: justify;"&gt;&lt;strong style="vertical-align: baseline;"&gt;Decoupling workflow logic from biochemical reasoning: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;General-purpose LLMs excel at task planning and synthesis, but are prone to hallucinating biochemical nuances. To control this behavior, this architecture maintains strong separation of concerns, namely, workflow and tasks decomposition are handled by Gemini. Domain chemistry and pharmacological reasoning are handled by specialized &lt;/span&gt;&lt;a href="https://www.nvidia.com/en-us/ai-data-science/foundation-models/nemotron/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Nemotron&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; models hosted on dedicated GPU nodes. &lt;/span&gt;&lt;/li&gt;
&lt;li role="presentation" style="text-align: justify;"&gt;&lt;strong style="vertical-align: baseline;"&gt;Context-driven evaluation over rigid cutoffs: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Traditional computational screening enforces arbitrary numerical cutoffs (e.g., discard if binding affinity ,-7.5 kcal/mol) However, therapeutic context fundamentally alters risk tolerance. Preventive cardiovascular therapy taken daily by healthy patients requires extremely strict safety margins, or a novel oncology candidate targeting an intractable cancer may accept higher toxicity if efficacy is demonstrated.&lt;/span&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;p style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;By passing disease context directly into the specialized reasoning models, drug candidates are evaluated against dynamic, target-specific profiles rather than static thresholds, preventing the premature culling of viable drug leads.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Performance best practices for high-compute pipelines&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Optimizing an agentic system that interfaces with heavy GPU compute requires challenging common engineering instincts:&lt;br/&gt;&lt;br/&gt;&lt;/span&gt;&lt;/p&gt;
&lt;div align="left"&gt;
&lt;div style="color: #5f6368; overflow-x: auto; overflow-y: hidden; width: 100%;"&gt;
&lt;div style="color: #5f6368; overflow-x: auto; overflow-y: hidden; width: 100%;"&gt;
&lt;div style="color: #5f6368; overflow-x: auto; overflow-y: hidden; width: 100%;"&gt;
&lt;div style="color: #5f6368; overflow-x: auto; overflow-y: hidden; width: 100%;"&gt;
&lt;div style="color: #5f6368; overflow-x: auto; overflow-y: hidden; width: 100%;"&gt;
&lt;div style="color: #5f6368; overflow-x: auto; overflow-y: hidden; width: 100%;"&gt;
&lt;div style="color: #5f6368; overflow-x: auto; overflow-y: hidden; width: 100%;"&gt;
&lt;div style="color: #5f6368; overflow-x: auto; overflow-y: hidden; width: 100%;"&gt;
&lt;div style="color: #5f6368; overflow-x: auto; overflow-y: hidden; width: 100%;"&gt;
&lt;div style="color: #5f6368; overflow-x: auto; overflow-y: hidden; width: 100%;"&gt;
&lt;div style="color: #5f6368; overflow-x: auto; overflow-y: hidden; width: 100%;"&gt;
&lt;div style="color: #5f6368; overflow-x: auto; overflow-y: hidden; width: 100%;"&gt;
&lt;div style="color: #5f6368; overflow-x: auto; overflow-y: hidden; width: 100%;"&gt;
&lt;div style="color: #5f6368; overflow-x: auto; overflow-y: hidden; width: 100%;"&gt;&lt;table&gt;&lt;colgroup&gt;&lt;col/&gt;&lt;col/&gt;&lt;col/&gt;&lt;/colgroup&gt;
&lt;tbody&gt;
&lt;tr&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p style="text-align: center;"&gt;&lt;span style="vertical-align: baseline;"&gt;Common Assumption&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p style="text-align: center;"&gt;&lt;span style="vertical-align: baseline;"&gt;Architectural Reality&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p style="text-align: center;"&gt;&lt;span style="vertical-align: baseline;"&gt;Recommended Best Practice&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;"Swap in a faster driver model to cut latency."&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The primary orchestrator's token generation is rarely the bottleneck in multi-step R&amp;amp;D pipelines.&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Optimize I/O, not token speed: Focus on reducing payload sizes and intermediate network hops rather than compromising orchestrator reasoning depth.&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;"Parallelize tool calls across workers."&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;External scientific services and containerized models often serialize requests internally on dedicated hardware.&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Respect execution topology: Align client-side concurrency with backend queue mechanics to prevent pipeline lockouts and connection timeouts.&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;"Pass full simulation states between tools."&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Large JSON bodies cause schema validation failures, prompt truncation, and token bloat.&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Pass heavy payloads by reference: Store raw coordinates in object storage and pass lightweight URIs between agent boundaries.&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;"Return outputs only when the pipeline finishes."&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Long multi-minute runs leave scientists blind to whether jobs are proceeding or stalled.&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Stream intermediate stage completions: Emit progress updates at each completed milestone (e.g., &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;folding complete&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;, &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;docking running&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;) for real-time visibility.&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;tr&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;"Chain multiple granular micro-tools."&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Deep tool-chains increase the probability of hallucinated arguments and schema mismatches.&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;td style="vertical-align: top; border: 1px solid #000000; padding: 16px;"&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Collapse compute pipelines: Wrap end-to-end simulation pipelines behind a single, clean tool interface that takes clear parameters and returns a consolidated object.&lt;/span&gt;&lt;/p&gt;
&lt;/td&gt;
&lt;/tr&gt;
&lt;/tbody&gt;
&lt;/table&gt;&lt;/div&gt;
&lt;/div&gt;
&lt;/div&gt;
&lt;/div&gt;
&lt;/div&gt;
&lt;/div&gt;
&lt;/div&gt;
&lt;/div&gt;
&lt;/div&gt;
&lt;/div&gt;
&lt;/div&gt;
&lt;/div&gt;
&lt;/div&gt;
&lt;/div&gt;
&lt;/div&gt;
&lt;h4&gt;&lt;span style="vertical-align: baseline;"&gt;The enterprise governance and asymmetric autonomy&lt;/span&gt;&lt;/h4&gt;
&lt;p style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;Deploying high-compute agents in enterprise environments requires robust data sovereignty, audit trails, and process control. The &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/ai-machine-learning/introducing-gemini-enterprise-agent-platform"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Gemini platform&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; establishes these safeguards natively with the following components: &lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation" style="text-align: justify;"&gt;&lt;a href="https://docs.cloud.google.com/iam/docs/agent-identity-overview"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Agent Identity&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;: Every autonomous agent operates with a cryptographically verifiable identity. Thus, for a biotech use case, when a candidate molecule is advanced or rejected, the event is permanently logged to that specific agent principal. &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation" style="text-align: justify;"&gt;&lt;a href="https://docs.cloud.google.com/gemini-enterprise-agent-platform/govern/gateways/agent-gateway-overview"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Agent Gateway&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;: This serves the single ingress/egress point for every interaction within the perimeter. Centralized IAM and regional residency rules help ensure sensitive molecular data remains within designated cloud regions. &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation" style="text-align: justify;"&gt;&lt;a href="https://docs.cloud.google.com/gemini-enterprise-agent-platform/govern/agent-registry"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Agent Registry&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;: This feature maintains a curated catalog of approved agents, validated tool schemas, and authorized Model Context Protocol (MCP) servers, preventing rogue tools from executing in production. &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation" style="text-align: justify;"&gt;&lt;a href="https://cloud.google.com/security/products/model-armor?hl=en"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Model Armor&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;: This functionality inspects incoming prompts and outgoing payloads in real time to prevent prompt injection and data exfiltration. &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;The payoff is clear. Identity, RBAC, auditing, residency, and data protection are native properties of the Gemini platform; this allows enterprise customers to focus on science rather than having to build a compliance stack for every project. &lt;/span&gt;&lt;/p&gt;
&lt;p style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;In high-stakes research and development, full autonomy introduces unacceptable liability, while requiring approval for trivial steps can create frictions. The solution addressing this to use asymmetric gating in the following ways;&lt;/span&gt;&lt;/p&gt;
&lt;p style="text-align: justify;"&gt;&lt;strong style="vertical-align: baseline;"&gt;Automate reversible operations:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; For example, allow the agent to resolve chemical naming synonyms, pull structural PDBs, and run preliminary Lipinski screens autonomously without interrupting the researcher.&lt;/span&gt;&lt;/p&gt;
&lt;p style="text-align: justify;"&gt;&lt;strong style="vertical-align: baseline;"&gt;Gate irreversible operations:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; For instance, initiating multi-GPU docking run, advancing a lead compound to wet-lab synthesis, or archiving a therapeutic program must require explicit human sign-off.&lt;/span&gt;&lt;/p&gt;
&lt;h3 style="text-align: justify;"&gt;&lt;strong style="vertical-align: baseline;"&gt;The pattern, and the path forward&lt;/strong&gt;&lt;/h3&gt;
&lt;p style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;High-compute autonomous agents succeed when the boundaries between cognitive planning, numerical computation, and system governance are clean in the following ways:&lt;/span&gt;&lt;/p&gt;
&lt;ol&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation" style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;Keep the orchestrator focused on decomposition and synthesis, leaving intensive biophysical simulation to specialized, containerized microservices and models.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation" style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;Eliminate context bloat by passing complex chemical data by reference through cloud storage.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation" style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;Anchor the workflow in enterprise governance, pairing cryptographic agent identity with asymmetric human oversight.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ol&gt;
&lt;p style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;As generative and structural biology models expand toward whole-cell modeling and complex dynamic simulations, this decoupled architecture ensures research systems scale efficiently while keeping proprietary IP secure and auditable.&lt;/span&gt;&lt;/p&gt;
&lt;p style="text-align: justify;"&gt;&lt;span style="vertical-align: baseline;"&gt;Reach out to your Google sales representative if you have any requirements to build a use case like this and we will be happy to talk and support you in your agent development journey on Google Cloud.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Fri, 16 Oct 2026 16:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/topics/healthcare-life-sciences/architecting-scalable-agentic-biotech-workloads/</guid><category>AI &amp; Machine Learning</category><category>AI infrastructure</category><category>HPC</category><category>Customers</category><category>Healthcare &amp; Life Sciences</category><media:content height="540" url="https://storage.googleapis.com/gweb-cloudblog-publish/images/Blog_Headers_-_Cloudstyle_3.0.max-600x600.png" width="540"></media:content><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>Architecting autonomy: A design pattern for high-compute, governed agents in biotech</title><description></description><image>https://storage.googleapis.com/gweb-cloudblog-publish/images/Blog_Headers_-_Cloudstyle_3.0.max-600x600.png</image><site_name>Google</site_name><url>https://cloud.google.com/blog/topics/healthcare-life-sciences/architecting-scalable-agentic-biotech-workloads/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Schneider Larbi</name><title>Sr. Manager, Global Partner Technical Architecture</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Chelsie Czop</name><title>Sr. Product Manager, AI Infrastructure</title><department></department><company></company></author></item><item><title>What the Gemini agent means for startups</title><link>https://cloud.google.com/blog/topics/startups/what-the-gemini-agent-means-for-startups/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Small teams, short runway, and constant pressure to ship faster — it’s the reality we see every day among the founders we work with at Google for Startups. When you are building a company from the ground up, every hour spent on operational overhead or infrastructure plumbing is an hour taken away from your core product.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Last week at Gemini at Work 2026, &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/ai-machine-learning/welcome-to-gemini-at-work-2026"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;we introduced the Gemini agent&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;: a single, universal agent designed to work both as your personal assistant and as a dedicated member of your team. And this week, we made our new frontier model,&lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt; &lt;/strong&gt;&lt;a href="https://cloud.google.com/blog/products/ai-machine-learning/gemini-4-argon-available-on-google-cloud?e=48754805"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Gemini 4 Argon&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, widely available — trained to maintain context across long, multi-step tasks.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Here is a look at how the Gemini agent works, five practical takeaways for founders, and how startups are already using it to punch above their weight.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;What is the Gemini agent?&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Instead of prompting separate chatbots for writing, coding, and data analysis, the Gemini agent is a single agent that takes high-level objectives and returns finished work. Three principles define it:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Unified:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; A single agent can now handle knowledge work, data analysis, creative asset generation, and writing and running code. For complex jobs, it dynamically spins up temporary, job-specific sub-agents and coordinates their work in parallel or in sequence.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Omnipresent access:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; It works across the surfaces you use every day — on the web, command line, desktop, and mobile, as well as natively inside Gmail, Docs, Sheets, Slides, Drive, Chat, and Calendar.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Persistent:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; It runs in the cloud with continuous memory across four dimensions (session, semantic, procedural, and episodic). It onboards itself like a new hire — learning your team, tools, and business context so you don't have to re-explain your startup with every prompt, and keeping multi-day tasks moving even after you close your laptop.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;For founders wearing five hats at once, Gemini works like a digital chief of staff that's already briefed on your calendar, projects and documents. Ask it to research market trends, build a financial model in Sheets and turn both into a pitch deck, without repeating what the project or goals are at each step. It also works ahead of you. When an investor emails asking for your latest product update as a deck, Gemini spots the request and offers to handle it; one click and it comes back with completed work, ready for sharing or review.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Five takeaways for startups&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;1. Coworker agents and skills help you scale your output&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Beyond personal assistance, you can create persistent &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;coworker agents&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; that work with your entire team. Each agent comes with its own Workspace account, email address, Drive storage, and directory presence. You can define a specific role — such as a launch coordinator, QA triager, or financial analyst — and your teammates can add agents to a Chat space or &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;@mention&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; it in a Google Doc comment to delegate work.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;You can also equip agents with reusable &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;skills&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; — or let Gemini write skills for itself as it learns your processes. For example, with operational reporting skills connected to BigQuery and the Knowledge Catalog, anyone on your team can ask a plain-language question to build and save a live report. Once saved, those reports run on demand without incurring additional token costs. Gemini also offers pre-built specializations, starting with legal and financial services.&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;What it means for startups:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Coworker agents can immediately increase the capacity of your teams, and skills can standardize repeatable processes like investor and operational reporting without burning tokens on every run.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;2. Open and third-party model choice&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Gemini is the agent, and the model underneath is your choice. As startup architectures mature, sending every task to a single frontier model quickly eats into gross margins. With the Gemini agent, you can dynamically route workloads across Gemini models, third-party models like Anthropic’s Claude, and lightweight open models like&lt;/span&gt; &lt;a href="https://cloud.google.com/blog/topics/startups/why-your-startup-needs-open-models-alongside-frontier-apis"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Gemma&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;What it means for startups:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; You get flexibility without lock-in. You can optimize your cost per task — using compact models for high-volume, deterministic jobs and frontier models for complex reasoning — and swap models as your product evolves without rebuilding your workflows.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;3. Connected to your services and tools &lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;An agent is only as useful as the systems it can reach. In addition to Google Workspace, the Gemini agent connects directly to your engineering, data, and business stack, including partner solutions:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Development and collaboration:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Git, Slack, Jira and more &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Data and local files:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; BigQuery, Postgres, Snowflake, Databricks, and more alongside files on your desktop&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Custom tools via MCP:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Any Model Context Protocol (MCP) server inside or outside your network so your engineers can build and share internal tools.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;What it means for startups:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Instead of spending engineering cycles maintaining custom connectors, your team can plug the agent straight into your existing codebase, databases, and MCP tools so it can act on live data wherever it lives.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;4. Built-in security and governance&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;For B2B startups, the enterprise security audit is often where deals stall. Under the hood, every agent is assigned a cryptographic &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Identity&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; for clear audit trails, runs inside an isolated sandbox, and protected with built-in prompt injection defenses. External traffic is routed through &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Agent Gateway&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, allowing you to enforce governance policies. And managing coworker agents is as straightforward as managing users today, since they all have a Workspace account.&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;What it means for startups:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; When an enterprise buyer’s CISO asks how your AI workflows handle data privacy, prompt injection, and access control, you have an enterprise-grade architecture ready — helping you clear procurement hurdles.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;5. Protect your runway with real-time cost controls&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Autonomous agents that run sub-agents for hours or days can quickly run up a bill if a loop goes unchecked. You can now set hard, real-time spend caps directly in the Cloud Billing Console; if a cap is reached, the project's agent automatically pauses so a bug never turns into a billing surprise. Underneath, Gemini runs on &lt;/span&gt;&lt;a href="https://cloud.google.com/ai-infrastructure?e=48754805"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google’s specialized AI infrastructure,&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; including our inference-optimized TPU 8i, which delivers 80% better performance-per-dollar than the prior generation.&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;What it means for startups:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; You and your developers can experiment with long-running, parallel agentic loops with a built-in financial safety net that protects your cash reserves and builds confidence with your backers.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Founders already building&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Startups and AI-native innovators are already putting Gemini to work. Immersive storytelling platform&lt;/span&gt;&lt;a href="https://cloud.google.com/transform/whenere-dreamlands-gemini-jane-austen-fortnite-future-of-fan-fiction"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt; &lt;/span&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Whenere&lt;/strong&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt; used Dreamlands&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, an agentic 3D worldbuilding platform built on Gemini platform, to bring &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;Pride and Prejudice&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; to Fortnite — with Gemini generating textures for each new object in about 7 seconds for under 10 cents; one artist created nearly 250 custom assets in a single day and completed a historically accurate estate in under a week. They join application-building platform &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Lovable&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; and fellow startups and innovators including &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt; OpenEvidence&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Recursion&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, and &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Replit&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/ai-machine-learning/welcome-to-gemini-at-work-2026?e=48754805"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;and more&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Build with Google for Startups&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The agentic era gives lean teams the leverage to punch far above their weight — whether you’re deploying coworker agents to cover internal operations or building agentic workflows directly into your product.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;While these new Gemini agent capabilities are not publicly available yet, startups can&lt;/span&gt;&lt;a href="https://cloud.google.com/contact"&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;contact our sales team&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to learn more and discuss how to get ready. In the meantime, you can explore the&lt;/span&gt;&lt;a href="https://startup.google.com/" rel="noopener" target="_blank"&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Google for Startups program&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to access credits, hands-on technical guidance, mentorship, global founder community programs and more. &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Thu, 15 Oct 2026 16:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/topics/startups/what-the-gemini-agent-means-for-startups/</guid><category>AI &amp; Machine Learning</category><category>Customers</category><category>Startups</category><media:content height="540" url="https://storage.googleapis.com/gweb-cloudblog-publish/images/startups-gemini-agent-header2.max-600x600.png" width="540"></media:content><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>What the Gemini agent means for startups</title><description></description><image>https://storage.googleapis.com/gweb-cloudblog-publish/images/startups-gemini-agent-header2.max-600x600.png</image><site_name>Google</site_name><url>https://cloud.google.com/blog/topics/startups/what-the-gemini-agent-means-for-startups/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Darren Mowry</name><title>VP, Global Startups &amp; Investor Ecosystem</title><department></department><company></company></author></item><item><title>How Visko streams real-time world models on Google Cloud TPUs</title><link>https://cloud.google.com/blog/products/media-entertainment/how-visko-streams-real-time-world-models-on-google-cloud-tpus/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;Editor's note: Generative AI startups are moving rapidly beyond static clip generation to build interactive, real-time worlds. In this technical deep dive, we explore how AI startup &lt;/span&gt;&lt;strong style="font-style: italic; vertical-align: baseline;"&gt;Visko&lt;/strong&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt; collaborated with Google Cloud TPU engineers to deploy Orbis — its continuously streaming Live World Model—on Cloud TPUs, going from initial kickoff to production handover in just 30 days.&lt;/span&gt;&lt;/p&gt;
&lt;hr/&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;What makes a live video world model so challenging to serve in production? The bottleneck isn't average throughput — it's beating a relentless real-time clock. Unlike offline video generators where a user submits a prompt and waits in a queue, a live world model streams continuously. Every new chunk of video and synchronized audio must finish rendering before the viewer finishes watching the previous one, or the world freezes.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Today, we're excited to share how &lt;/span&gt;&lt;a href="https://visko.ai" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Visko&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, an AI startup pioneering interactive world models, teamed up with Google Cloud to solve this exact challenge. Working side by side, Visko and Google Cloud engineers deployed Visko's end-to-end live inference pipeline — combining video generation, super-resolution upscaling, and synchronized audio—onto a single &lt;/span&gt;&lt;a href="https://cloud.google.com/tpu?e=48754805"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud TPU slice&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. Together, the teams brought end-to-end generation down, rendering at 1.07x playback speed.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Here is an inside look at how we engineered the pipeline to hit that milestone in a single month, the architectural optimizations that moved the needle, and practical lessons for other startups building ultra-low-latency generative AI on Google Cloud.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Redefining interactive media with Live World Models&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;While most generative video platforms focus on rendering fixed, pre-baked clips, Visko is building what it calls &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Live Models&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: video world models that generate continuously and respond on the fly.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Visko's flagship model, &lt;/span&gt;&lt;a href="https://www.visko.ai/models" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Orbis&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, streams live video starting from a single text or image prompt. As the stream plays, viewers can enter new steering prompts in real time, dynamically evolving the scene without restarting or breaking continuity. For instance, in one test session, the livestream opens on a cat napping beside a crackling fireplace; when the viewer steers the scene to introduce a bowl of food, the bowl materializes seamlessly, and the cat stretches, stands up, and meows.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Since launching its public demo on September 1, 2026, Orbis has unlocked new possibilities across interactive entertainment, live commerce, AI companions, immersive education, and robotics simulation. To scale these interactive sessions reliably, Visko partnered with Google Cloud TPU engineers in August 2026 to port the Orbis live pipeline to Google's Tensor Processing Unit (TPU). Within 30 days, the full multimodal pipeline was running in real time on TPUs.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Visko’s ambition is to power a new generation of interactive worlds that people can step into and shape in real time. Streaming continuous video and audio under a strict clock is a fundamentally different system challenge than offline clip generation. Working side by side with Google Cloud's TPU engineers, we brought the full Orbis pipeline to TPUs in just 30 days — giving us the low-latency compute and memory bandwidth required to bring Live Models to creators and developers at scale.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;"Visko is bringing an exciting new class of real-time, steerable workloads to the Cloud TPU ecosystem. By co-engineering their live multimodal pipeline on a singleTPU slice using JAX, torchax, and custom Pallas kernels, we proved how quickly a lean startup team can go from PyTorch prototype to render at 1.07x playback speed on Google Cloud's AI Hypercomputer."&lt;br/&gt;&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;– &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Dave Elliott, Head of Developer Advocacy for AI, Google Cloud&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Live video is a hard-deadline problem&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;When you serve an offline clip generator, latency is forgiving — the user waits once while the backend processes frames. A live world model has no waiting room.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Every single chunk must be denoised, decoded from latents, upscaled to 1664x936 (at 18 fps), and paired with generated audio faster than realtime playback speed, for as long as the session runs.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Two structural constraints make live world models far more demanding than standard batch inference:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Batch size of one:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Each live stream represents an individual viewer's interactive session. You cannot mask latency by grouping dozens of concurrent requests into a massive batch — the standard playbook for keeping AI accelerators saturated.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Strict sequential dependency:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Every chunk depends directly on the one before it. Because the video model maintains a rolling memory of recent video latents so that chunk N+1 smoothly continues the action from chunk N, you cannot pre-render chunks out of order or parallelize across future timestamps.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;For startup infrastructure engineers coming from LLM serving, it helps to think of live world modeling not as a tokens-per-second throughput benchmark, but as an &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;unforgiving per-step latency SLO&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Architecting the multimodal pipeline on a single TPU slice&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To keep infrastructure lean and cost-effective as Visko scales, the team designed the live pipeline to chain three distinct generative models on a single &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/tpu/docs/multislice-introduction"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud TPU slice&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; running in BF16 precision.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;When a viewer submits an initial prompt or a live steering update, all three generative models — the core video diffusion model, the super-resolution upscaler, and the audio generator — share the same chips, taking turns on the hardware for each chunk.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/figure1_.max-1000x1000.png"
        
          alt="figure1_"&gt;
        
        &lt;/a&gt;
      
        &lt;figcaption class="article-image__caption "&gt;&lt;p data-block-key="1g87t"&gt;Figure 1: Visko's end-to-end Orbis live pipeline chains video diffusion, latent decoding, super-resolution upscaling, and pure-JAX audio generation sequentially on a single Cloud TPU slice, completing each chunk faster than the playback speed.&lt;/p&gt;&lt;/figcaption&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To combine rapid startup iteration with bare-metal accelerator speed, the software stack is built on three open-source pillars:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://github.com/jax-ml/jax" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;JAX&lt;/strong&gt;&lt;/a&gt;&lt;strong style="vertical-align: baseline;"&gt;:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Google's high-performance Python array library for accelerators, compiled into optimized machine code by the &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;XLA&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; compiler.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://github.com/google/torchax" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;torchax&lt;/strong&gt;&lt;/a&gt;&lt;strong style="vertical-align: baseline;"&gt;:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; An interoperability layer that runs PyTorch models directly on Cloud TPUs via JAX, enabling startup teams to bring existing PyTorch components over without an upfront, ground-up rewrite.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://jax.readthedocs.io/en/latest/pallas/index.html" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Pallas&lt;/strong&gt;&lt;/a&gt;&lt;strong style="vertical-align: baseline;"&gt;:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; The custom kernel programming language for TPUs in JAX. &lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Real-time performance breakdown: Hitting the clock &lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;By profiling and optimizing each stage on a single slice in BF16, the engineering team achieved a total per-chunk latency of under the target while — maintaining roughly &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;7% headroom &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;to absorb variance during live streaming:&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/figure_2_orbis_latency_breakdown.max-1000x1000.png"
        
          alt="figure_2_orbis_latency_breakdown"&gt;
        
        &lt;/a&gt;
      
        &lt;figcaption class="article-image__caption "&gt;&lt;p data-block-key="1g87t"&gt;Figure 2: Chart of per-chunk latency budget on one TPU slice.&lt;/p&gt;&lt;/figcaption&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Because video denoising accounts for nearly two-thirds of the total compute budget, that stage became the primary focus of our kernel and parallelism optimizations.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Inside the stack: Three optimizations that unlocked real-time streaming&lt;/span&gt;&lt;/h3&gt;
&lt;h4&gt;&lt;span style="vertical-align: baseline;"&gt;1. Ulysses context parallelism across all TPU chips&lt;/span&gt;&lt;/h4&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;A video chunk expands into a massive sequence of spatiotemporal tokens inside the video transformer. To distribute that workload across the TPU chips at batch size 1, the team implemented &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;Ulysses context parallelism&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Instead of splitting model weights, Ulysses splits the token sequence across chips. Immediately before the attention layer, a high-speed all-to-all interconnect exchange regroups the tensors so that each chip holds the &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;entire&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; video sequence, but only for one-quarter of the attention heads. After each chip computes attention over the full sequence for its assigned heads, a second all-to-all exchange restores the sequence-split layout for the feed-forward layers.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/figure3_ulysses_context.max-1000x1000.png"
        
          alt="figure3_ulysses_context"&gt;
        
        &lt;/a&gt;
      
        &lt;figcaption class="article-image__caption "&gt;&lt;p data-block-key="1g87t"&gt;Figure 3: Ulysses context parallelism across TPU chips at batch size 1. High-speed all-to-all Inter-Chip Interconnect (ICI) exchanges transition tensors between a sequence-split layout for feed-forward layers and a head-split layout for Pallas splash attention.&lt;/p&gt;&lt;/figcaption&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h4&gt;&lt;span style="vertical-align: baseline;"&gt;2. Custom Pallas splash attention from MaxDiffusion&lt;/span&gt;&lt;/h4&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To accelerate the attention computation itself, the team integrated a custom Pallas &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;splash attention&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; kernel from &lt;/span&gt;&lt;a href="https://github.com/AI-Hypercomputer/maxdiffusion" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;MaxDiffusion&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, Google Cloud's open-source reference library for diffusion models on Cloud TPU.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Splash attention brings FlashAttention-style tiling to TPU architecture: rather than materializing the full quadratic attention matrix in High Bandwidth Memory (HBM), it computes attention in compact tiles that remain inside ultra-fast on-chip VMEM. Startup teams can drop the exact kernel used in this port straight into their own pipelines via &lt;/span&gt;&lt;a href="https://github.com/AI-Hypercomputer/maxdiffusion/blob/main/src/maxdiffusion/kernels/custom_splash_attention.py" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;custom_splash_attention.py&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;h4&gt;&lt;span style="vertical-align: baseline;"&gt;3. A 5.8x audio speedup by moving from torchax to pure JAX&lt;/span&gt;&lt;/h4&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We first brought all three Orbis models onto Cloud TPU with torchax, which runs PyTorch code on TPU through JAX and XLA without a rewrite. For the video and upscaling models, torchax paired with Pallas attention kernels met our latency target directly, and both stay in torchax in the final pipeline.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Only the audio model needed a rewrite. Moving the audio stage from torchax to native JAX made it 5.8x faster and brought the full pipeline under the playback deadline.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Field notes for AI builders: Cloud TPU lessons&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Takeaways from tuning live multimodal inference on Cloud TPU:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li&gt;&lt;strong style="vertical-align: baseline;"&gt;Ulysses context parallelism for single stream video&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: At batch size 1, splitting the token sequence across chips with all-to-all interconnect exchanges was faster than both tensor parallelism and ring attention.&lt;/span&gt;&lt;/li&gt;
&lt;li&gt;&lt;strong style="vertical-align: baseline;"&gt;Native BF16 for continuous streams&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: FP8 quantization cuts memory bandwidth pressure, but small quantization errors build up across chunks in a continuous stream. Native BF16 kept visual quality intact within the latency budget.&lt;/span&gt;&lt;/li&gt;
&lt;li&gt;&lt;strong style="vertical-align: baseline;"&gt;Selective JAX ports&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: torchax kept the video and upscaling models in PyTorch, so the team only rewrote the audio bottleneck in native JAX.&lt;/span&gt;&lt;/li&gt;
&lt;li&gt;&lt;strong style="vertical-align: baseline;"&gt;XLA compilation caching:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; The first boot spends a few minutes compiling XLA graphs. Caching compiled artifacts brings worker restarts under a minute.&lt;/span&gt;&lt;/li&gt;
&lt;/ul&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Building real-time generative models on Cloud TPU? Start here&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;If your startup is pushing the boundaries of live video, world models, or ultra-low-latency multimodal inference, you can build with the same toolkit that powered Visko's port:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Accelerate diffusion attention out of the box:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Grab the open-source splash attention kernel used in this collaboration from &lt;/span&gt;&lt;a href="https://github.com/AI-Hypercomputer/maxdiffusion" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;MaxDiffusion&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, and explore the &lt;/span&gt;&lt;a href="https://jax.readthedocs.io/en/latest/pallas/index.html" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Pallas documentation&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to write custom TPU kernels.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Migrate PyTorch fast, then optimize selectively:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Bring your existing PyTorch models onto TPU quickly with &lt;/span&gt;&lt;a href="https://github.com/google/torchax" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;torchax&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, pinpoint bottlenecks with the &lt;/span&gt;&lt;a href="https://cloud.google.com/tpu/docs/cloud-tpu-tools"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud TPU Profiler&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, and port latency-critical stages to native &lt;/span&gt;&lt;a href="https://github.com/jax-ml/jax" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;JAX&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Spin up TPUs on demand:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Use &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/tpu/docs/request-using-flex-start"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;TPU Flex-start VMs&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to provision TPU capacity for short development and benchmarking runs without a long-term reservation.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Scale with Google for Startups:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Early-stage AI startups looking for cloud credits, technical mentorship, and access to Google Cloud's AI Hypercomputer stack can apply to the &lt;/span&gt;&lt;a href="https://cloud.google.com/startup"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Google for Startups Cloud Program&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;hr/&gt;
&lt;p&gt;&lt;sup&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;Special thanks to the Visko.ai team for the close technical partnership, and to the Google Cloud TPU engineers who lead the project including: Yixin Shi, James Huang, and Elisa Tsai.&lt;/span&gt;&lt;/sup&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Thu, 15 Oct 2026 16:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/media-entertainment/how-visko-streams-real-time-world-models-on-google-cloud-tpus/</guid><category>AI &amp; Machine Learning</category><category>Customers</category><category>Media &amp; Entertainment</category><media:content height="540" url="https://storage.googleapis.com/gweb-cloudblog-publish/images/Visko-header.max-600x600.png" width="540"></media:content><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>How Visko streams real-time world models on Google Cloud TPUs</title><description></description><image>https://storage.googleapis.com/gweb-cloudblog-publish/images/Visko-header.max-600x600.png</image><site_name>Google</site_name><url>https://cloud.google.com/blog/products/media-entertainment/how-visko-streams-real-time-world-models-on-google-cloud-tpus/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Will Yin</name><title>CEO, Visko</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Ivan Nardini</name><title>AI Developer Relations Engineer</title><department></department><company></company></author></item><item><title>Innovation in Ireland: How Irish brands scale with Gemini Enterprise</title><link>https://cloud.google.com/blog/topics/customers/ireland-innovation-companies-startups-governments-scale-with-gemini/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;In recent decades, Ireland has grown into a vibrant hub for global technology. As modernization accelerates, Irish organizations — spanning government agencies, established enterprise brands, and growing startups in AI and other areas — are all advancing their technical foundations. Today, leaders across the country are injecting agentic AI into every layer of their business and operational workflows. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We are seeing organizations move from early prototypes to production-ready deployments using &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/ai-machine-learning/welcome-to-gemini-at-work-2026"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Gemini Enterprise&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. This progress supports Ireland's competitive position within the European market. According to a recent Implement Consulting Group report assessing &lt;/span&gt;&lt;a href="https://services.google.com/fh/files/blogs/google_ireland_economic_impact_25_final.pdf" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google’s local contribution&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, public and private leaders estimate between 40 billion to 45 billion euros of economic potential in global AI markets.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This ambition aligns with broader government initiatives. Since assuming the presidency of the Council of the European Union for the second half of 2026, Ireland has steered higher-level technology and policy discussions, outlining a &lt;/span&gt;&lt;a href="https://www.gov.ie/en/department-of-the-taoiseach/press-releases/government-publishes-new-digital-ai-strategy-90-actions-to-strengthen-irelands-position-as-a-digital-leader-and-ai-hub/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;National Digital and AI Strategy&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to improve productivity and innovation across sectors.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To support this strategy, we are focused on making these capabilities more accessible. Our local teams are collaborating with industry leaders to implement Gemini Enterprise which offers capabilities that range from conversational AI agents and natural language data analysis, to optimizing operations with agentic ecosystems.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The economic impact is already taking shape. In 2025 alone, Google operations — including our AI and cloud services — added an estimated 10 billion euros to the Irish gross domestic product. And this momentum is shared: Organizations across Ireland are using Google Cloud to modernize infrastructure and put AI to work.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Here are several ways Irish organizations are modernizing with Google Cloud and Gemini Enterprise:&lt;/span&gt;&lt;/p&gt;
&lt;h4&gt;&lt;span style="vertical-align: baseline;"&gt;Ryanair supports millions of passengers through AI-backed operations&lt;/span&gt;&lt;/h4&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Ryanair, Europe’s largest airline, &lt;/span&gt;&lt;a href="https://www.googlecloudpresscorner.com/2026-08-12-Ryanair-and-Google-Cloud-Announce-Five-Year-Data-and-AI-Partnership" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;is modernizing&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; its collaborative operations and infrastructure by deploying Gemini Enterprise and Google Workspace for some 35,000 employees. Flight crews, operations planners, and corporate staff will benefit from streamlined decision-making, optimized crew logistics, and enhanced corporate productivity. At the same time, a dual-cloud strategy strengthens infrastructure resilience as the airline targets growth to 300 million passengers by 2034. &lt;/span&gt;&lt;/p&gt;
&lt;h4&gt;&lt;span style="vertical-align: baseline;"&gt;Smyths Toys rethinks the customer experience with agentic support&lt;/span&gt;&lt;/h4&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;As one the world’s largest specialty toy retailers, Smyths Toys is &lt;/span&gt;&lt;a href="https://cloud.google.com/customers/smyths-toys-ge?e=48754805&amp;amp;hl=en"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;reshaping its customer support&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; with Google Cloud and Gemini Enterprise. By introducing an AI agent named Codie, the retailer has resolved more than 60% of its web inquiries and reduced holiday email backlogs. Customer care teams that previously handled up to 1,000 peak-season emails a day can now focus on the roughly 50 cases that need more complex judgment.&lt;/span&gt;&lt;/p&gt;
&lt;h4&gt;&lt;span style="vertical-align: baseline;"&gt;The Irish Revenue Commissioners: Advancing public sector data&lt;/span&gt;&lt;/h4&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Public sector agencies are also making strides in Ireland. The &lt;/span&gt;&lt;a href="https://www.revenue.ie/en/home.aspx" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Irish Revenue Commissioners&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; — responsible for taxes, customs, and keeping Ireland’s economy running smoothly — is &lt;/span&gt;&lt;a href="https://blog.google/intl/en-ie/products/revenue-ireland/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;working with Google Cloud&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to migrate from legacy on-premise systems to a unified AI-lakehouse architecture designed to break down data silos and establish a single source of truth. Strict data sovereignty controls safeguard taxpayer information while Gemini Enterprise powers agentic data analysis, helping Revenue deliver secure, responsive public services.&lt;/span&gt;&lt;/p&gt;
&lt;h4&gt;&lt;span style="vertical-align: baseline;"&gt;Virgin Media Ireland: Accelerating workflows&lt;/span&gt;&lt;/h4&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Supported by &lt;/span&gt;&lt;a href="https://www.libertyglobal.com/liberty-global-google-cloud/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Liberty Global's partnership&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; with Google Cloud, Virgin Media Ireland is moving AI from exploration to production. Through modern infrastructure and automation, the team now deploys workloads three times faster, is in the process of rolling out multiple AI agents, and is launching new customer services designed to drive business growth.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Irish startups scaling with enterprise tools&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;While established enterprises are modernizing legacy systems, Irish startups are building on modern technology from day one. Backed by a strong local tech ecosystem, these agile teams are creating services that compete on an international stage. &lt;/span&gt;&lt;/p&gt;
&lt;h4&gt;&lt;span style="vertical-align: baseline;"&gt;Hexis: Scaling personalized sports nutrition&lt;/span&gt;&lt;/h4&gt;
&lt;p&gt;&lt;a href="https://hexis.live/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Hexis&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; builds technology solutions that help athletes refine their performance and recovery. By storing and processing that real-time biometric and behavioral data in Google Cloud, Hexis uses Gemini Enterprise to power natural language conversations that explain fueling adjustments to users throughout the day. Today, the platform supports 40% of the&lt;/span&gt;&lt;a href="https://www.letour.fr/en/landing-page" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt; Tour de France peloton &lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;and half of &lt;/span&gt;&lt;a href="https://www.premierleague.com/en" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Premier League&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; clubs during the 2026–2027 season.&lt;/span&gt;&lt;/p&gt;
&lt;h4&gt;&lt;span style="vertical-align: baseline;"&gt;Kitman Labs: Unifying performance analytics&lt;/span&gt;&lt;/h4&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Founded in Dublin in 2012, &lt;/span&gt;&lt;a href="https://www.kitmanlabs.com/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Kitman Labs&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; provides sports science and performance intelligence to more than 4,000 sporting organizations across 26 countries. By combining its intelligence platform with advanced analytics and Gemini Enterprise, Kitman Labs helps teams make data-informed decisions that support athlete health, lower injury risk, and extend player career longevity. &lt;/span&gt;&lt;/p&gt;
&lt;h4&gt;&lt;span style="vertical-align: baseline;"&gt;Spryt: Automating healthcare coordination&lt;/span&gt;&lt;/h4&gt;
&lt;p&gt;&lt;a href="https://spryt.com/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Spryt&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; streamlines healthcare coordination across the U.K. and U.S., &lt;/span&gt;&lt;a href="https://cloud.google.com/customers/spryt?hl=en&amp;amp;e=48754805"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;using AI built on Google Cloud and Gemini Enterprise&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to optimize patients’ experiences, reduce missed appointments, and cut administrative burden. &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;The company collaborates with the NHS, operates in more than 70 US clinics, and has analyzed 224 million appointment records as part of a multiyear initiative with Mayo Clinic.&lt;/span&gt;&lt;/p&gt;
&lt;h4&gt;&lt;span style="vertical-align: baseline;"&gt;IMPT: Automating carbon offsetting&lt;/span&gt;&lt;/h4&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;By connecting AI, blockchain, and Google Cloud, &lt;/span&gt;&lt;a href="https://impt.io/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;IMPT&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; embeds verified carbon offsetting into everyday hotel bookings and online purchases. Following a 15-week migration to Google Cloud, IMPT supports 8 million listings across 195 countries. With Gemini Enterprise, the company’s lean team handles about 400 daily customer chats with a six-second median response time, allowing staff to focus on growing its business rather than managing systems.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Transform ideas from planning to production&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;As Ireland continues to expand  as a global technology hub, local organizations are demonstrating how to translate AI potential into practical results. &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;Explore our customer stories to see how other organizations are building with Google Cloud, or reach out to our team in Ireland to accelerate your journey.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Thu, 08 Oct 2026 12:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/topics/customers/ireland-innovation-companies-startups-governments-scale-with-gemini/</guid><category>AI &amp; Machine Learning</category><category>Application Modernization</category><category>Retail</category><category>Telecommunications</category><category>Startups</category><category>Customers</category><media:content height="540" url="https://storage.googleapis.com/gweb-cloudblog-publish/images/ireland-innovation-companies-startups-govern.max-600x600.jpg" width="540"></media:content><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>Innovation in Ireland: How Irish brands scale with Gemini Enterprise</title><description></description><image>https://storage.googleapis.com/gweb-cloudblog-publish/images/ireland-innovation-companies-startups-govern.max-600x600.jpg</image><site_name>Google</site_name><url>https://cloud.google.com/blog/topics/customers/ireland-innovation-companies-startups-governments-scale-with-gemini/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Sandeep Khanna</name><title>Managing Director, Google Cloud U.K. &amp; Ireland</title><department></department><company></company></author></item><item><title>AI21 achieves an 83% reduction in time-to-start for AI workloads with AI Hypercomputer</title><link>https://cloud.google.com/blog/products/containers-kubernetes/ai21-trains-its-models-on-ai-hypercomputer/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;&lt;strong&gt;Editor’s note&lt;/strong&gt;: AI21 Labs is a leading global AI lab with a long track record of building foundation models, most notably the Jamba family, and today focuses on specialized LLMs and agent optimization technology. By adopting Google Cloud AI Hypercomputer, AI21 cut high-priority job wait times from 72 hours to 12 and manual scheduling interventions from 20 per week to zero. &lt;/span&gt;&lt;/p&gt;
&lt;hr/&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;At &lt;/span&gt;&lt;a href="https://www.ai21.com/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;AI21&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, we build foundation models and agent optimization products that help enterprises run agents at frontier quality, efficiently. Our language models, including the &lt;/span&gt;&lt;a href="https://www.ai21.com/jamba/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Jamba&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; family, and our agent optimization product suite run demanding production workloads, including our own. We chose &lt;/span&gt;&lt;a href="https://cloud.google.com/ai-infrastructure"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google Cloud AI Hypercomputer&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to support them at scale.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To keep our model training runs highly utilized, we needed a performant, scalable environment codesigned across infrastructure, orchestration, and consumption models. Our model training runs on one of our shared &lt;/span&gt;&lt;a href="https://cloud.google.com/kubernetes-engine"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google Kubernetes Engine&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; (GKE) clusters, pooling thousands of Google Cloud A3 (powered by NVIDIA H100 Tensor Core GPUs) and A3 Ultra (powered by NVIDIA H200 Tensor Core GPUs) instances, so any team can draw on the full capacity of the fleet rather than being boxed into its own slice. The cluster also trains models and agent-optimization workloads beyond the Jamba family. That approach keeps utilization high, and it makes scheduling hard. &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;The gridlock of high-utilization clusters&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Prior to leveraging GKE for orchestration, we used to negotiate capacity by hand in Slack. If you needed capacity for a training run, you posted in #gpu-resources and hoped for the best.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;That worked fine when the cluster had headroom. It stopped working once utilization pinned near 100%, which is where you want a reserved compute fleet to sit.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Over time, every request became a negotiation. Team leads spent their time refereeing compute disputes. Our high-priority jobs — the large, multi-node training runs that need half or more of the cluster at once and serve as the critical path for model projects — could sit blocked for up to 72 hours waiting for enough contiguous capacity to open up. &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/original_images/1_5g4QmVa.gif"
        
          alt="1"&gt;
        
        &lt;/a&gt;
      
        &lt;figcaption class="article-image__caption "&gt;&lt;p data-block-key="cwh8v"&gt;Figure 1: A typical day on #gpu-resources — manual requests, ad-hoc coordination, and researchers waiting on replies.&lt;/p&gt;&lt;/figcaption&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Scarcity created two distinct problems, and it took us a while to see them as separate. The first was contention: determining who gets compute access next, which we resolved through negotiation. The second was fragmentation: capacity that was technically free but scattered in pieces too small for a large job to use, a bin-packing problem no amount of negotiation could fix. &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/image2_vk4hCkH.max-1000x1000.png"
        
          alt="2"&gt;
        
        &lt;/a&gt;
      
        &lt;figcaption class="article-image__caption "&gt;&lt;p data-block-key="cwh8v"&gt;Figure 2: GPU fragmentation. 8 free GPUs scattered as 1+1+4+2 across four nodes can’t fit an 8-GPU workload.&lt;/p&gt;&lt;/figcaption&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Sometimes we had plenty of capacity free on paper, but it was scattered across different machines in chunks too small for a larger job to actually land. Without all-or-nothing admission, the cluster could reach a deadlock, with machines holding resources without doing useful work until someone stepped in manually. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;It was clear the status quo wasn’t working and we needed something better.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Choosing the right scheduler&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We looked at a few open-source batch schedulers, including Apache YuniKorn, Volcano, and &lt;/span&gt;&lt;a href="https://kueue.sigs.k8s.io/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Kueue&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. YuniKorn didn’t cover all our use cases. And while Volcano had more features, integrating it with our environment would have required replacing core Kubernetes scheduler components. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Kueue won on simplicity and integration. It worked with standard Kubernetes, didn’t require replacing core components, and didn’t force us to rewrite our job specs.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Pairing Kueue with AI Hypercomputer’s flexible, open operations through GKE also contributed to our success. We rely on GKE because it gives us the right level of control for compute-intensive AI work — close access to GPU hardware and drivers, without the overhead of managing raw instances ourselves. Kueue’s native integration with GKE, including with Google Cloud capacity types like&lt;/span&gt; &lt;a href="https://cloud.google.com/spot-vms"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Spot VMs&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and&lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/compute/introducing-dynamic-workload-scheduler"&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Dynamic Workload Scheduler&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, meant that once our reserved capacity filled up, the same scheduling logic could reach out to elastic capacity automatically instead of leaving jobs stuck.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Closing the loop with open source&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Adopting Kueue turned into something bigger than a simple process change. Because we partner with Google Cloud, we have a direct line to a Technical Account Manager, who saw an opportunity to make AI21 a design partner for the Kueue team. This way, we wouldn’t just be a user, but a source of real production requirements that could help shape where the tool went next. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;One of the first things to come out of that partnership was a requirements document we shared with the Kueue team, describing behavior we needed that didn’t exist yet: fair admission ordering across teams for multi-node jobs, without the preemption that usually comes bundled with fairness. The Kueue team built it.&lt;/span&gt; &lt;a href="https://kueue.sigs.k8s.io/docs/concepts/admission_fair_sharing/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Admission Fair Sharing&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; (AFS) reorders the admission queue to favor teams that have historically used less capacity without disrupting jobs already running. Around the same time, we also turned on&lt;/span&gt; &lt;a href="https://kueue.sigs.k8s.io/docs/concepts/topology_aware_scheduling/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Topology Aware Scheduling&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. This existing feature made Kueue aware of our physical cluster layout, so it can refuse to admit jobs that won’t fit on a single node rather than placing them with nowhere to run. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;For us, that’s what the best-case open-source feedback loop looks like: Real requirements surfaced through production use, fed directly back into Kueue.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Same fleet, less friction&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The Slack &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;#gpu-resources channel is archived now. Every workload, whether a debug pod, a multi-node training run, or an inference deployment, gets queued, prioritized, and scheduled automatically. The results were immediate. Manual interventions dropped from about 20 a week to zero. High-priority jobs that used to wait up to 72 hours now wait 12. Fragmentation across the cluster fell from 15% to 8%, and the “zombie job” problem — workloads partially admitted with nowhere to run — is gone.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;None of this changed our total cost. We run our reserved fleet at close to 100% utilization on purpose, so raw spend was never the variable we were optimizing. What changed is where our people’s time goes. Team leads aren’t refereeing compute disputes anymore, and researchers aren’t waiting on replies in Slack. That frees everyone to run more experiments and iterate faster.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Fri, 02 Oct 2026 19:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/containers-kubernetes/ai21-trains-its-models-on-ai-hypercomputer/</guid><category>AI Hypercomputer</category><category>GKE</category><category>AI infrastructure</category><category>Customers</category><category>Containers &amp; Kubernetes</category><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>AI21 achieves an 83% reduction in time-to-start for AI workloads with AI Hypercomputer</title><description></description><site_name>Google</site_name><url>https://cloud.google.com/blog/products/containers-kubernetes/ai21-trains-its-models-on-ai-hypercomputer/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Barak Peleg</name><title>VP, Technology and Architecture, AI21</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Asaf Ben-Tovim</name><title>DevOps Engineer, AI21</title><department></department><company></company></author></item><item><title>Scribd, Inc. classifies more than 400 million documents with Gemini batch inference on Gemini Enterprise</title><link>https://cloud.google.com/blog/topics/customers/scribd-inc-classifies-millions-of-documents-on-gemini-enterprise/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;a href="https://www.scribd.com/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Scribd, Inc&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. is home to one of the world's largest collections of human-created content. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Scribd’s  products leverage one of the world's largest collections of human-created content and intelligent tools to help people move from information access to real understanding and application.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This past year, Scribd used Gemini's native PDF understanding and Gemini Enterprise batch prediction to run trust and safety classification across its entire user-generated content corpus of more than 400 million documents, spanning over 12 billion pages, in a matter of months.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Here were the results: &lt;/strong&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Classified 400M+ user-uploaded documents (12B+ pages of text and images) across Scribd and Slideshare&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Completed the corpus-wide backfill in a matter of months, with Google Cloud scaling batch throughput to meet the timeline&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Native PDF input meant more than 99% of the corpus was processed as-is, with no OCR, rendering, or screenshotting pipeline to build&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Gemini Enterprise’s batch prediction at a 50% discount to interactive pricing made LLM classification viable at corpus scale&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Trust and safety at the scale of an entire corpus&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Scribd, Inc. is the parent company to four distinct products: Scribd, Slideshare, Everand, and Fable. Across Scribd and Slideshare, hundreds of millions of user-uploaded PDFs, presentations, and documents help people find information, build understanding, and finish projects. With that scale comes responsibility. We aim to balance access with protecting our communities. We leverage a mix of human and automated methods to review and best ensure the content on our platforms complies with our community rules. As the corpus continues to grow and technology evolves, this challenge requires even more resources.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Understanding a document requires reading its text and its images together, in context. Classification has to work across all possible use cases, all possible languages, all possible contexts. There is no single solution that can translate cleanly across all of it. And each policy area traditionally demanded its own specialized detection model, which meant either years of in-house engineering effort or specialized vendor solutions that don't fit the economics of a 400-million-document backfill. The team evaluated several off-the-shelf moderation tools and open models, but none delivered the quality they needed at their scale.&lt;/span&gt;&lt;/p&gt;
&lt;p style="padding-left: 40px;"&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;“This is a genuinely hard problem that we have been working on for a long time. Every category of content behaves differently, and historically each one required its own specialized solution. Gemini collapsed all of that into one model, one prompt, and one pipeline.”&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; – &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;Sachin Sebastian, Senior Engineering Manager, Scribd, Inc.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Why Gemini: PDFs are a first-class input&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The turning point was realizing that Gemini treats Scribd's corpus the way it actually exists: as PDFs. Gemini accepts PDF input natively and reads each page as both text and image, so a single multimodal model could evaluate everything from dense text documents to image-heavy presentations, with no OCR pipeline, page rendering, or screenshot infrastructure in between. Because Gemini processes each PDF page at a fixed, predictable token count, costs scale linearly and stay low even across 12 billion pages.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;After benchmarking model families and versions, the team selected Gemini 2.5 Flash Lite as the classification workhorse, with Gemini 2.5 Pro serving as an LLM judge in a full second consistency pass over the corpus to validate output quality. In the team's evaluations, Gemini's multimodal understanding caught visual policy signals that text-only moderation endpoints routinely missed.&lt;/span&gt;&lt;/p&gt;
&lt;p style="padding-left: 40px;"&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;“Gemini's peculiar advantage is that it meets our content in its native format. It reads the text, layout, and images of a PDF directly. More than 99% of our corpus went in exactly as it lives on our site without any pre-processing”&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; – &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;Sachin Sebastian, Senior Engineering Manager, Scribd, Inc.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Batch prediction, simple enough to bet the corpus on&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The execution model was deliberately simple. Documents were staged in Cloud Storage, submitted to Gemini Enterprise batch prediction, and the results flowed back into the team's data platform for downstream analysis. There was no serving infrastructure to operate, no rate-limiting logic to write, and no GPU capacity to manage.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Batch pricing, at 50% below interactive rates, is what made the economics work at corpus scale. The team later layered on Gemini Enterprise’s implicit prefix caching, restructuring prompts so the static policy text hit the cache, which pushed efficiency further with no loss in classification quality.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;A partnership measured in throughput&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Processing 400 million documents is ultimately a throughput problem, and this is where the partnership with Google Cloud mattered most. Scribd's team connected directly with Google Cloud engineering and product to plan the backfill, advise on region strategy, and make sure the right capacity was in place ahead of launch.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;As the backfill ramped up, Google Cloud worked closely with the team to scale throughput to the demands of the project. The effect was dramatic: batch jobs began completing far faster than projected, and for much of the run Gemini Enterprise was not the bottleneck. Scribd's own upstream pipeline was.&lt;/span&gt;&lt;/p&gt;
&lt;p style="padding-left: 40px;"&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;“Google Cloud didn't just answer support tickets. They partnered with us on the backfill, and there were stretches where Gemini Enterprise finished work faster than our own systems could produce it. That is a good problem to have.”&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; – &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;Sachin Sebastian, Senior Engineering Manager, Scribd, Inc.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;What's next&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The backfill is now the foundation of an ongoing program: newly uploaded content flows through the same Gemini classification pipeline, keeping the corpus continuously evaluated rather than periodically cleaned. And because the pattern of PDFs in Cloud Storage, Gemini batch prediction, and results in the lakehouse proved so operationally simple, the team is applying it to a growing set of content-understanding workloads across its platforms.&lt;/span&gt;&lt;/p&gt;
&lt;p style="padding-left: 40px;"&gt;&lt;span style="vertical-align: baseline;"&gt;“This project changed how we think about our roadmap. Work we had classified as multi-year, multi-team efforts is now a prompt, a batch pipeline, and a few weeks of runtime.” – &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;Sachin Sebastian, Senior Engineering Manager, Scribd, Inc.&lt;/span&gt;&lt;/p&gt;
&lt;hr/&gt;
&lt;p&gt;&lt;em&gt;&lt;sup&gt;&lt;span style="vertical-align: baseline;"&gt;This work was a collaboration between Google Cloud and Scribd. We'd like to thank everyone involved for their support throughout this project:&lt;/span&gt;&lt;/sup&gt;&lt;/em&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li role="presentation"&gt;&lt;em&gt;&lt;sup&gt;&lt;strong style="vertical-align: baseline;"&gt;Scribd Engineering:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Anish Kumar, Jeanie Lam, James Watkins, Hima Alladi&lt;/span&gt;&lt;/sup&gt;&lt;/em&gt;&lt;/li&gt;
&lt;li role="presentation"&gt;&lt;em&gt;&lt;sup&gt;&lt;strong style="vertical-align: baseline;"&gt;Scribd Applied Research:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Rafael Pedrosa Lacerda de Melo, Kara Killough, Eric Chang&lt;/span&gt;&lt;/sup&gt;&lt;/em&gt;&lt;/li&gt;
&lt;li role="presentation"&gt;&lt;em&gt;&lt;sup&gt;&lt;strong style="vertical-align: baseline;"&gt;Scribd Product:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Seyoon Kim, Nicole Pauls&lt;/span&gt;&lt;/sup&gt;&lt;/em&gt;&lt;/li&gt;
&lt;li role="presentation"&gt;&lt;em&gt;&lt;sup&gt;&lt;strong style="vertical-align: baseline;"&gt;Google Cloud AI Batch Inference team: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;James Liu, Digvijay Singh, Wei-chung Wang, Yan Wang, Kun Shi &lt;/span&gt;&lt;/sup&gt;&lt;/em&gt;&lt;/li&gt;
&lt;li role="presentation"&gt;&lt;em&gt;&lt;sup&gt;&lt;strong style="vertical-align: baseline;"&gt;Google Cloud Customer Engineer:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Jennifer Liang&lt;/span&gt;&lt;/sup&gt;&lt;/em&gt;&lt;/li&gt;
&lt;/ul&gt;&lt;/div&gt;</description><pubDate>Thu, 24 Sep 2026 16:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/topics/customers/scribd-inc-classifies-millions-of-documents-on-gemini-enterprise/</guid><category>Customers</category><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>Scribd, Inc. classifies more than 400 million documents with Gemini batch inference on Gemini Enterprise</title><description></description><site_name>Google</site_name><url>https://cloud.google.com/blog/topics/customers/scribd-inc-classifies-millions-of-documents-on-gemini-enterprise/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Travis Martin</name><title>Account Manager, Google Cloud Platform</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Sachin Sebastian</name><title>Sr. Engineering Manager, Scribd</title><department></department><company></company></author></item><item><title>How growing Latin American midsize businesses are building in the AI era</title><link>https://cloud.google.com/blog/topics/customers/how-midsize-latam-companies-build-with-ai/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Latin America’s small and medium-sized businesses are the heartbeat of the region's economy — accounting for more than &lt;/span&gt;&lt;a href="https://www.undp.org/latin-america/blog/yes-there-hope-msmes-region-and-beyond" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;60% of total employment&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; in the region, according to United Nations estimates. And just like their enterprise peers, everywhere you look, ambitious teams are moving fast to embrace AI. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Many have already transitioned from experimenting with generative tools and agentic workflows to using them every day to work smarter, save time, and deliver exceptional customer experiences. These growing businesses are particularly focused on maximizing the benefit they get from their investments in AI, whether that’s using a fast, low-cost model to summarize daily emails or deploying an advanced model for complex data analysis, teams can match the right AI capability to their exact task and budget. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;It’s this range of options, and a familiarity with the broader suite of Google business, media, and advertising tools that has led many SMBs to choose Google Cloud, and &lt;/span&gt;&lt;a href="https://cloud.google.com/gemini-enterprise"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Gemini Enterprise&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; in particular, as their AI platform of choice. By doing so, they’re able to build custom AI agents, streamline daily tasks and paperwork, and offer customers instant support with the speed and reach needed to compete on a global scale. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With our unique front row seat, we’ve seen the benefit SMBs are getting from leveraging Gemini Enterprise, not only for generative AI, but as a catalyst for adopting other essential cloud tools like &lt;/span&gt;&lt;a href="https://cloud.google.com/kubernetes-engine"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google Kubernetes Engine&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and &lt;/span&gt;&lt;a href="https://cloud.google.com/bigquery"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;BigQuery&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; for complete end-to-end modernization. The number of Latin American-based small and medium businesses using Google Cloud AI tools has grown 8x year-over-year and the number of Brazil based small and medium businesses using Google Cloud AI tools has grown 9x year-over-year.&lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt; &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;This rapid adoption spans our Gemini models, Gemini Enterprise, and core &lt;/span&gt;&lt;a href="https://cloud.google.com/infrastructure"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud infrastructure&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, and are helping businesses to:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Roll out better customer support systems to help escalate and resolve customer support calls more quickly.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Automate repetitive actions in areas like payroll and accounting.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Help more employees understand and leverage data at work — even those not trained as data analysts.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Rapidly create and implement new designs for marketing collateral.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Help more people build their own AI agents to help them in their everyday jobs.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;As we head into today’s &lt;/span&gt;&lt;a href="https://www.wiz.io/events/gcp-summit-brasil-2026" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google Cloud Summit in Brazil&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, we were proud to showcase nearly 20 of our newest Latin American SMB customers using Google AI to reduce busywork, serve their customers faster, and grow their businesses.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Announcing new Latin American customers putting Google AI to work&lt;/strong&gt;&lt;/h3&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://adgoat.io/" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;AdGoat&lt;/strong&gt;&lt;/a&gt;&lt;strong style="vertical-align: baseline;"&gt;, &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;an Argentina-based adtech company processing more than 10 billion annual ad requests across more than 100 global websites. It uses &lt;/span&gt;&lt;a href="https://cloud.google.com/run"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud Run&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, the &lt;/span&gt;&lt;a href="https://ai.google.dev/gemini-api/docs" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Gemini API&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, and Gemini Enterprise to automate content analysis, ad bidding, and audience targeting to help e-commerce brands drive higher campaign returns.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://angelus.ind.br/en/" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Angelus&lt;/strong&gt;&lt;/a&gt;&lt;strong style="vertical-align: baseline;"&gt;, &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;a Brazilian dental and healthcare manufacturing company, uses Gemini Enterprise to streamline project management across its research and development department. This enables its teams to automatically pull technical project data into pre-approved templates aligned with the company’s brand identity and regulatory requirements.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://bunkerdb.com/en" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;BunkerDB&lt;/strong&gt;&lt;/a&gt;&lt;strong style="vertical-align: baseline;"&gt;, &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;a marketing science company operating across Latin America, uses Gemini Enterprise, Cloud Run, and &lt;/span&gt;&lt;a href="https://cloud.google.com/storage"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud Storage&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to power an AI platform that organizes marketing assets, checks brand compliance, generates or adapts multimodal content, and predicts ad performance before launch. All of this helps it reduce creative turnaround times from weeks to hours and cut cost per lead by up to 25%.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://caffeinearmy.com/?utm_source=google&amp;amp;utm_medium=paid&amp;amp;utm_campaign=20275358873&amp;amp;utm_content=158789549988&amp;amp;utm_term=caffeine+army&amp;amp;gadid=662356321422&amp;amp;tw_source=google&amp;amp;tw_adid=662356321422&amp;amp;tw_campaign=20275358873&amp;amp;gad_source=1&amp;amp;gad_campaignid=20275358873&amp;amp;gbraid=0AAAAAo2729bPtRTinWcSsz7rH5K6skQRG&amp;amp;gclid=CjwKCAjwn67VBhBnEiwAXUIN1ScGTlY2qUbv_w2LDvbf-Gaz5p0VbmhcB1AUgyVMODM8Ww5an4PGSBoChw4QAvD_BwE" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Caffeine Army&lt;/strong&gt;&lt;/a&gt;&lt;strong style="vertical-align: baseline;"&gt;, &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;a Brazilian wellness and high-performance company that connects people with solutions in nutrition, sports, and well-being, deployed BigQuery and Gemini Enterprise on Google Cloud to unify customer purchase insights, enabling faster creative campaign turnarounds and boosting team productivity across the organization.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://convertperforma.com.br/" target="_blank"&gt;&lt;strong style="vertical-align: baseline;"&gt;Convert&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, a Brazilian marketing and analytics provider, uses &lt;/span&gt;&lt;a href="https://cloud.google.com/looker"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Looker&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, BigQuery, and Cloud Run to power five specialized AI agents that answer complex business questions in natural language, speeding up report deliveries by 65% and reducing operational costs by 32%.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://www.growthdigital.biz/" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Growth Digital&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, a Google Ad sales rep operating across 13 Latin American countries, used BigQuery and Gemini Enterprise to build over 113 AI agents, enabling teams to build proposals 5x faster, cut campaign reporting time by 80%, and reduce financial error rates to under 0.01%.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="http://grupotusmaquinas.com" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;GrupoTusMaquinas.com&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, an equipment management platform based in Chile, deployed &lt;/span&gt;&lt;a href="https://cloud.google.com/products/ai"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google Cloud AI&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; tools and Gemini models to create digital tracking profiles for trucks and machinery, allowing businesses to query fleet status in plain language and manage vehicles regardless of brand or location.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://www.healthatom.com/" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;HealthAtom&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, a healthcare technology company, uses the Gemini API, &lt;/span&gt;&lt;a href="https://cloud.google.com/products/firestore"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Firestore&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, and &lt;/span&gt;&lt;a href="https://cloud.google.com/functions"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud Functions&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to power AI assistants across its clinical platforms, automating appointment scheduling and medical record reviews while supporting 80 million annual patient interactions.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="http://klog.co" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;KLog.co&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, a Chilean logistics technology company digitizing freight forwarding across Latin America, uses Gemini Enterprise, BigQuery, and &lt;/span&gt;&lt;a href="https://workspace.google.com/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google Workspace&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to automate cargo tracking and shipping paperwork, cutting manual data entry errors by over 90% and increasing document processing capacity tenfold.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://neooh.com.br/" rel="noopener" target="_blank"&gt;&lt;strong style="vertical-align: baseline;"&gt;NEEOH&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, a leading Brazilian out-of-home advertising communication platform, uses Gemini Enterprise to standardize secure AI usage across its organization, enabling teams to generate campaign copy and build pitch proposals faster while keeping corporate client data secure.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://luxiaagro.com/" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Luxia Agro&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, an Argentinian foreign trade supplier of crop protection products, uses &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/gemini/3-5-flash"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Gemini 3.5 Flash&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and Gemini Enterprise to automatically pull key details from complicated shipping emails and update their central business systems. This allows it to automate 80% of foreign trade operations and cut manual processing errors in half.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://macal.cl/venta" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Macal&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, a Chilean auction company, uses the Gemini Enterprise, Cloud Run, BigQuery, and &lt;/span&gt;&lt;a href="https://cloud.google.com/security/products/security-command-center"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Security Command Center&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to automatically verify property records and modernize its technology systems, cutting software development times from weeks to days and lowering infrastructure costs by up to 30%.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://www.ninecon.com.br/en/home/" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Ninecon&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, a Brazilian tech consulting firm, deployed Gemini Enterprise to integrate AI directly into employee workflows, allowing managers to track usage patterns and optimize project turnaround times with real-time insights.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://www.novagne.com.br/" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Nova Gestões&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, a customer service and operations provider in Brazil, uses &lt;/span&gt;&lt;a href="https://cloud.google.com/speech-to-text"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud Speech-to-Text&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and Gemini Enterprise to translate and analyze 100% of customer calls in real time, reducing post-call manual data entry and boosting team productivity by 30%.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://www.romi.com/en/" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Romi&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, a Brazilian industrial machinery manufacturer, uses the Gemini API and Gemini Enterprise to power an interactive chat assistant directly on CNC machine HMI (human machine iInterface), giving factory operators instant answers grounded in official manuals and generating QR codes for step-by-step instructional videos.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://www.eldorado.com.uy/" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Supermercados El Dorado&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, a leading supermarket chain in Uruguay, leverages &lt;/span&gt;&lt;a href="https://cloud.google.com/products/compute"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google Compute Engine&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and Gemini Enterprise to modernize legacy testing infrastructure and connect custom AI agents within their daily workflows, boosting team productivity across departments.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://tryvia.com.br/" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Tryvia&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, a Brazilian IT and business solutions provider, uses Google Cloud, Looker, and Gemini Enterprise to move off legacy physical servers, giving teams real-time reporting dashboards and AI tools that speed up software development and daily tasks.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://www.vinci-concessions.com/en/infrastructure/via-cristais" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;Via Cristais&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, a major highway operator in Brazil, leverages &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/contact-center/ccai-platform/docs"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google Contact Center as a Service&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to speed up emergency routing for highway accidents, reducing caller wait times, improving driver satisfaction, and mitigating the impact of call center staff turnover.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;a href="https://www.wespeak.pro/en/" rel="noopener" target="_blank"&gt;&lt;strong style="text-decoration: underline; vertical-align: baseline;"&gt;WeSpeak&lt;/strong&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, an AI conversational platform for the hospitality industry in Latin America, uses Cloud Run, &lt;/span&gt;&lt;a href="https://deepmind.google/models/gemini/pro/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Gemini Pro&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, and Gemini Flash to automate end-to-end guest interactions across messaging channels like WhatsApp and Instagram. This has helped it achieve an 85% resolution rate and a 2x increase in overall sales volume for hotel clients.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Helping your team build AI skills&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To help growing teams get the absolute most out of AI, we’ve created easy, no-cost learning programs that anyone can use:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Programs for small and medium businesses:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Explore beginner-friendly training paths or join specialized programs to learn how to build custom AI assistants for your day-to-day work.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Google skills for organizations:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Access thousands of free, on-demand AI courses and hands-on practice labs designed by experts at Google Cloud and Google DeepMind.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Get certified:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Help your staff gain industry-recognized AI certificates through guided courses, expert mentoring, and skill badges.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;By offering easy-to-use tools and free training — from everyday office apps in Workspace to advanced AI on Google Cloud — Google is here to help Latin American businesses thrive today and in the future.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Thu, 24 Sep 2026 14:30:00 +0000</pubDate><guid>https://cloud.google.com/blog/topics/customers/how-midsize-latam-companies-build-with-ai/</guid><category>AI &amp; Machine Learning</category><category>Google Cloud</category><category>Customers</category><media:content height="540" url="https://storage.googleapis.com/gweb-cloudblog-publish/images/latam-midsize-biz-hero.max-600x600.png" width="540"></media:content><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>How growing Latin American midsize businesses are building in the AI era</title><description></description><image>https://storage.googleapis.com/gweb-cloudblog-publish/images/latam-midsize-biz-hero.max-600x600.png</image><site_name>Google</site_name><url>https://cloud.google.com/blog/topics/customers/how-midsize-latam-companies-build-with-ai/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Andre Alves</name><title>Director, SMB Sales LATAM</title><department></department><company></company></author></item><item><title>The future of orchestration: Pine59’s journey to Airflow 3 on Google Cloud</title><link>https://cloud.google.com/blog/topics/supply-chain-logistics/the-future-of-orchestration-pine59s-journey-to-airflow-3-on-google-cloud/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Operating large data pipelines requires an orchestration layer that scales smoothly as workloads expand. When your pipelines process millions of complex data points every day to feed predictive models, staying up-to-date with your technology stack is a strategic necessity.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;a href="https://www.pine59.com/" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Pine59&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; provides location intelligence data through data pipelines that produce analytical metrics on cadences ranging from hourly to quarterly. One of the company’s most data-intensive metrics, Daily Foot Traffic, computes data for as many as 14 million distinct locations in a single job. To handle this massive volume, Pine59’s system runs entirely on Google Cloud, with the heavy lifting in &lt;/span&gt;&lt;a href="https://cloud.google.com/bigquery"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;BigQuery&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and all of it orchestrated by &lt;/span&gt;&lt;a href="https://cloud.google.com/products/managed-service-for-apache-airflow"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Managed Service for Apache Airflow&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; (formerly Cloud Composer) running Apache Airflow 3.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;As the company’s volume of data and number of machine learning workloads scaled up, Pine59 decided to modernize its monorepo, which contains hundreds of directed acyclic graphs (DAGs). Here is a look at how that transition improved Pine59’s MLOps capabilities, developer workflow, and pipeline speed.&lt;/span&gt;&lt;/p&gt;
&lt;h2&gt;&lt;span style="vertical-align: baseline;"&gt;Proactive modernization for growth&lt;/span&gt;&lt;/h2&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Pine59 has long relied on a shared monorepo with code and tooling spanning multiple projects to run its metric production pipelines. As it considered its infrastructure’s future, the company wanted to help its data pipelines run faster and more reliably.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;That’s why it decided to stress-test production workloads against the newly available Managed Airflow (Gen 3) architecture running Airflow 3. The initial results were unambiguous: the Gen 3 environment delivered immediate and significant processing speed, task scheduling, and overall stability improvements. Recognizing the clear potential for performance gains, Pine59 initiated a full transition to the new environment.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/Vertical_Version_yyTMhsG.max-1000x1000.jpg"
        
          alt="1 - Pine59 Google Cloud Architecture Vertical Version"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h2&gt;&lt;span style="vertical-align: baseline;"&gt;Orchestrating advanced MLOps&lt;/span&gt;&lt;/h2&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Pine59’s pipelines don’t just move data; they drive complex ML models, so a core aspect of its migration was optimizing the orchestration of its ML inference workloads.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Previously, Pine59 had used standard Kubernetes operators for these tasks. By moving to Managed Airflow (Gen 3), which features a highly optimized and abstracted infrastructure layer, the company’s engineering team refined its MLOps architecture. They did so by setting up a dedicated &lt;/span&gt;&lt;a href="https://cloud.google.com/kubernetes-engine"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google Kubernetes Engine&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; (GKE) cluster that was specifically optimized for model inference and integrated it into the Pine59 pipelines.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This clear separation of orchestration and heavy ML execution compute allows data processing and model inference to run efficiently, showcasing Managed Airflow as a resilient, scalable backbone for enterprise MLOps.&lt;/span&gt;&lt;/p&gt;
&lt;h2&gt;&lt;span style="vertical-align: baseline;"&gt;Supporting developers with custom extensibility&lt;/span&gt;&lt;/h2&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Beyond infrastructure improvements, Pine59 was also able to immediately capitalize on Airflow 3’s delivery of a vastly improved developer workflow and user interface. Indeed, managing hundreds of interconnected DAGs requires excellent observability, and Pine59 found Airflow 3’s plugin authoring system remarkably easy to use.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To improve internal developer velocity, the company quickly built a number of custom plugins that it integrated directly into its new Airflow UI:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;BigQuery Auto-linkify:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; A tool that automatically detects internal BigQuery table references within the Airflow Logs and XCom tabs, dynamically generating direct links to BigQuery Studio for faster debugging (available as a &lt;/span&gt;&lt;a href="https://gist.github.com/jan-hajny-unacast/74e1e504e3e3c8765323bd019a87fb30" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;public GitHub gist&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;)&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;DAG Run Configuration Search:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; A custom search form added directly to the DAG overview page. It allows Pine59 engineers to query specific key-value pairs within DAG run payloads (configs) and instantly surface matching runs. This in turn drastically reduces troubleshooting time.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;In addition, the team also deployed a compatibility shim layer within its monorepo. This “compat” module dynamically abstracts logic between Airflow versions, streamlining operator migration across versions.&lt;/span&gt;&lt;/p&gt;
&lt;h2&gt;&lt;span style="vertical-align: baseline;"&gt;Faster, more reliable pipelines&lt;/span&gt;&lt;/h2&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;For Pine59, migrating to Managed Airflow (Gen 3) with Airflow 3 has yielded clear, quantifiable results.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The most important improvement was the speed of its DAG runs. In the company’s previous setup, tasks often got stuck in a queued state during peak processing surges. With Gen 3, queue latency has dropped dramatically, allowing tasks to start running almost immediately.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Consider the comparison below of total aggregated “queued” &amp;amp; “running” time of more than 300 runs of the same DAG between Managed Airflow (Gen2) with Airflow 2.11 vs. Managed Airflow (Gen3) with Airflow 3.1 below. As we can readily see, the difference in queued time is significant.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/image2_fCfKA2m.max-1000x1000.png"
        
          alt="image2"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Coupled with internal DAG optimizations made during the transition, the performance gains are also highly tangible. For example, the Daily Foot Traffic pipeline previously took nearly 38 minutes to complete. With the new instance, the same workload now takes less than 26 minutes —nearly 32% less processing time.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Today, Pine59 processes all its production workloads on its new Managed Airflow (Gen 3) instance. By moving to this next generation orchestration, the company improved its MLOps capabilities, equipped its developers with better tools, and built a faster, more resilient foundation for future workloads.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;If your engineering team spends more time managing infrastructure than delivering value, consider a similar transition and discover how it can help you move from maintaining servers to building the future of your data and AI pipelines today.&lt;/span&gt;&lt;/p&gt;
&lt;hr/&gt;
&lt;p&gt;&lt;sup&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;Special thanks to the following contributor to this post: Alexandre Crespo-Perez&lt;/span&gt;&lt;/sup&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Thu, 17 Sep 2026 17:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/topics/supply-chain-logistics/the-future-of-orchestration-pine59s-journey-to-airflow-3-on-google-cloud/</guid><category>Data Analytics</category><category>Infrastructure Modernization</category><category>Customers</category><category>Supply Chain &amp; Logistics</category><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>The future of orchestration: Pine59’s journey to Airflow 3 on Google Cloud</title><description></description><site_name>Google</site_name><url>https://cloud.google.com/blog/topics/supply-chain-logistics/the-future-of-orchestration-pine59s-journey-to-airflow-3-on-google-cloud/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Piotr Wieczorek</name><title>Lead Senior Product Manager, Google</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Jan Hajný</name><title>Senior Data Engineer, Pine59</title><department></department><company></company></author></item><item><title>How a solo founder runs a five-continent tender platform on AlloyDB and MCP</title><link>https://cloud.google.com/blog/products/databases/solo-founder-runs-a-global-tender-platform-on-alloydb-and-mcp/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;strong style="vertical-align: baseline;"&gt;Editor's note:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;Lucius AI, a tender-intelligence startup covering markets across five continents, runs its entire data platform on AlloyDB for PostgreSQL with a single operator. By migrating semantic search to a ScaNN index and managing database operations through Model Context Protocol (MCP), query latency dropped by 47x while automating day-to-day administrative tasks via MCP.&lt;/span&gt;&lt;/p&gt;
&lt;hr/&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Executive summary&lt;/span&gt;&lt;/h3&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Lucius AI runs a global tender platform spanning more than 210,000 tenders across the UK, EU, India, and Australia, requiring minimal operational overhead for a solo founder.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Lucius AI deployed AlloyDB for PostgreSQL to consolidate its relational catalog, audit logs, and vector embeddings into a single managed database engine.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Migrating semantic search to a ScaNN index lowered query latency from 1.14 seconds to 24 milliseconds — a 47x speedup on a representative production query.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Connecting an AI agent to AlloyDB using the Model Context Protocol (MCP) helps Lucius AI automate query analysis, data freshness checks, and incident forensics under strict least-privilege permissions.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;h2&gt;&lt;span style="vertical-align: baseline;"&gt;Making tender intelligence work as a company of one&lt;/span&gt;&lt;/h2&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Lucius AI helps businesses bidding on public contracts evaluate opportunities across global markets. The platform ingests public procurement notices from the UK, the EU, the US and Canada, Australia and New Zealand, India and Singapore, alongside World Bank donor-funded notices across Africa and Asia. Lucius AI analyzes tender documents using Gemini to generate compliance matrices, bid recommendations, and draft responses citing original source pages. For small and mid-sized suppliers, this replaces days of manual document reviews and costly external consulting.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Running a platform of this scope requires extensive operational coordination:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Nightly ingestion from thirteen public procurement sources&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;A catalog of more than 210,000 tenders, including tens of thousands open for active bidding&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Two production regions on Cloud Run: Europe, and an Australian deployment on its own AlloyDB cluster with customer-managed encryption keys (CMEK) for defense-adjacent customers&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Ongoing analytics, performance tuning, data validation, and incident response&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Managing these responsibilities without dedicated data engineering or database administration teams requires offloading operational maintenance. Lucius AI addressed this challenge on two fronts: using AlloyDB for PostgreSQL as the core system of record, and connecting an AI agent through the Model Context Protocol (MCP) to safely execute database operations.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/1_Zpv001B.max-1000x1000.png"
        
          alt="1"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Consolidating systems into AlloyDB&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Rather than deploying separate relational databases, vector databases, and log stores, Lucius AI houses all core data in AlloyDB for PostgreSQL. The relational tender catalog, document metadata, audit logs, and vector embeddings reside in the same database engine. Storing vector embeddings alongside relational rows avoids managing separate vector stores, establishes a unified backup schedule, and centralizes identity management.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Authentication relies strictly on Cloud IAM. Services connect using dedicated Google Cloud service accounts mapped to database roles scoped to specific access requirements, without storing database passwords in application environments. Database reliability is managed natively by AlloyDB through automated backups and point-in-time recovery, avoiding custom disaster recovery procedures.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;In production, this consolidated architecture supports:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;More than 210,000 tenders in the catalog&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;, with embeddings stored directly alongside them&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Rebuilding the semantic index embedded &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;115,820 records in 10.6 minutes&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; with the Gemini embedding model, for around three dollars in API spend; AlloyDB auto embeddings now keep those vectors current.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Retrieval reranking executed directly inside the database using the &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;ai.rank&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; function — with &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;mean latency of 77-milliseconds&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; - returning the most relevant results for search queries without requiring a standalone reranking microservice&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Accelerating semantic search by 47x&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Semantic search across the tender catalog initially relied on unindexed vector comparisons, where a representative query took 1.14 seconds. Migrating this workload to a ScaNN index in AlloyDB reduced query latency to &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;24 milliseconds — a 47x improvement&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The index recommendation originated from the AI agent during an automated performance audit, where it benchmarked the query plan before preparing the index migration.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/2_SEuqQ6L.max-1000x1000.png"
        
          alt="2"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Automating database operations with MCP&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To delegate routine administrative tasks, Lucius AI configured the open-source MCP Toolbox for Databases using the prebuilt &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;alloydb-postgres&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; server.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Operational delegation requires strict access controls. The agent connects using a dedicated PostgreSQL role granted SELECT across the schema and UPDATE on a single operational table. Destructive commands (&lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;DROP&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt;, &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;DELETE&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt;, &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;TRUNCATE&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt;) are omitted, restricting agent actions to authorized operational boundaries.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Under this configuration, the AI agent performs regular database operations across four key areas:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;On-demand analytics&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Compiles retention cohorts, activation funnels, and catalog coverage by country via ad hoc SQL queries, removing the need to build and maintain manual dashboards or complex analytical pipelines.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Performance optimization&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Performs query-plan inspections and index analysis, such as identifying the ScaNN indexing strategy.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Incident forensics&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: In response to an external security probe, the agent parsed audit logs to reconstruct the request timeline in minutes, verifying that tenant isolation remained intact.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Automated data-quality checks&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Evaluates ingestion watermarks and freshness across all thirteen procurement sources every morning.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/original_images/3_MXwqVC6.gif"
        
          alt="3"&gt;
        
        &lt;/a&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;For teams adopting this architecture, establishing a progressive permission structure provides clear guardrails: start with read-only access, expand permissions as requirements dictate, and keep destructive operations restricted to human administrators.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Looking ahead&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Lucius AI is planning three technical initiatives to further reduce operational overhead:&lt;/span&gt;&lt;/p&gt;
&lt;ol&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Automated vector embeddings in AlloyDB AI&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: After validating &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;ai.initialize_embeddings&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; across the full catalog, a weekly maintenance job uses &lt;/span&gt;&lt;code style="vertical-align: baseline;"&gt;ai.refresh_embeddings&lt;/code&gt;&lt;span style="vertical-align: baseline;"&gt; to update vectors.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Columnar engine acceleration&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Having enabled AlloyDB’s columnar engine with auto-columnarization, the database identified and stored 40 frequently queried columns across four tables in memory within a day, accelerating reporting queries without a separate analytical store.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Managed Remote MCP Server&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Transitioning from self-hosted Toolbox processes to Google Cloud's fully managed Remote MCP Server for AlloyDB will offload MCP server hosting and maintenance.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ol&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;By anchoring core data in AlloyDB and managing routine operations through MCP, Lucius AI demonstrates how a single engineer can build and operate a resilient, multi-region procurement platform.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To explore Lucius AI, visit &lt;/span&gt;&lt;a href="https://ailucius.com" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;ailucius.com&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;. To evaluate AlloyDB for PostgreSQL, deploy an &lt;/span&gt;&lt;a href="https://cloud.google.com/alloydb"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;AlloyDB cluster&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to test performance against your own workloads.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Thu, 17 Sep 2026 16:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/databases/solo-founder-runs-a-global-tender-platform-on-alloydb-and-mcp/</guid><category>Customers</category><category>Startups</category><category>Databases</category><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>How a solo founder runs a five-continent tender platform on AlloyDB and MCP</title><description></description><site_name>Google</site_name><url>https://cloud.google.com/blog/products/databases/solo-founder-runs-a-global-tender-platform-on-alloydb-and-mcp/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Davor Jerković</name><title>Founder, Lucius AI</title><department></department><company></company></author></item><item><title>For SeaVerse, GKE Agent Sandbox reduces infrastructure costs by 60%</title><link>https://cloud.google.com/blog/products/containers-kubernetes/seaverse-chooses-gke-agent-sandbox/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;strong style="font-style: italic; vertical-align: baseline;"&gt;Editor’s note:&lt;/strong&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt; Today we hear from &lt;/span&gt;&lt;a href="https://seaverse.ai/" rel="noopener" target="_blank"&gt;&lt;span style="font-style: italic; text-decoration: underline; vertical-align: baseline;"&gt;SeaVerse&lt;/span&gt;&lt;/a&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;, &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;a gaming startup from &lt;/span&gt;&lt;a href="https://www.seaart.ai" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;SeaArt&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; that is building a platform for playable AI experiences&lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;, where users can open lightweight games, character chats, and interactive apps, or create their own experiences from a prompt. To support that creative loop, SeaVerse needed infrastructure that could run dynamic, multi-tenant sandbox workloads with strong isolation, low latency, better observability, and more flexible costs. &lt;/span&gt;&lt;a href="https://cloud.google.com/kubernetes-engine"&gt;&lt;span style="font-style: italic; text-decoration: underline; vertical-align: baseline;"&gt;Google Kubernetes Engine (GKE)&lt;/span&gt;&lt;/a&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt; and &lt;/span&gt;&lt;a href="https://cloud.google.com/blog/products/containers-kubernetes/bringing-you-agent-sandbox-on-gke-and-agent-substrate"&gt;&lt;span style="font-style: italic; text-decoration: underline; vertical-align: baseline;"&gt;GKE Agent Sandbox&lt;/span&gt;&lt;/a&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt; gave SeaVerse the managed foundation from which to execute these AI workloads, helping the team reduce their infrastructure costs by up to 60%, while giving creators a faster path from idea to playable experiences.&lt;/span&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt; Read on to learn more.&lt;/span&gt;&lt;/p&gt;
&lt;hr/&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;What if AI were a playground? Welcome to SeaVerse, a creation-first platform for playable AI experiences. Here, an AI creation can be as peaceful as drawing a path for a snake to follow, or as chaotic as a music-backed stickman simulation. Some people come to play lightweight games. Others come to chat with AI characters, try interactive apps, create visual patterns, share what they made, or remix an idea into something new.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We built SeaVerse around a simple promise: Every experience should feel immediate and easy to share. A creator should be able to describe an idea in plain language, refine the result, and publish it in moments, without a traditional coding workflow.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Delivering that simplicity requires serious infrastructure. Every creation that users make moves through the same chain: generate, run, preview, debug, publish, remix. If any part of that chain is slow, unstable, or poorly isolated, users feel it immediately. That’s why we turned to GKE and GKE Agent Sandbox. &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;The infrastructure challenge of instant interaction&lt;/strong&gt;&lt;strong style="vertical-align: baseline;"&gt; &lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;What looks effortless to a user is anything but on our end. Every creation on SeaVerse runs as a distinct workload and is expected to behave reliably from the first interaction.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Because each workload runs in its own environment, we needed clear security boundaries between users, creations, and sandboxes. But overly strict isolation could slow the very creative loop we were trying to protect, and when something went wrong, diagnosing it was costly. Our engineers had to trace problems across multiple parts of the execution chain with little visibility into what was happening inside the environment.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We explored existing sandbox approaches, but needed deeper kernel-level isolation and native observability at scale to support fast diagnosis across multi-tenant environments. Something had to change.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Building on GKE and GKE Agent Sandbox&lt;/strong&gt;&lt;strong style="vertical-align: baseline;"&gt; &lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We chose &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;GKE&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; because we needed a reliable, secure way to operate Kubernetes without turning our engineering team into a cluster maintenance team. GKE brought together the proven ecosystem and operational tooling we needed, freeing us to focus on building the platform rather than managing the infrastructure beneath it.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;As a Kubernetes primitive designed for agent code execution and computer use, &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;GKE Agent Sandbox&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; addressed our requirement for strong isolation, enforcing strong security boundaries without slowing down the creation experience. By utilizing GKE Agent Sandbox with Kata Containers+Cloudhypervisor (microVM), we’ve achieved the perfect balance of multi-cloud flexibility and robust security, option to switch isolation runtime between microVM and gVisor, running our AI sandboxes safely. GKE empowers us to scale toward our long-term vision of supporting over a million sandboxes. Built on gVisor, it provides kernel-level isolation for dynamic sandbox workloads while preserving the Kubernetes orchestration model, so that they can be managed through the same scheduling, monitoring, and operations as the rest of the cluster. &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With SeaVerse, users can generate interactive experiences from a single prompt. After an experience is generated, GKE Agent Sandbox supports the run, test, integration, and verification steps needed to make it ready to preview, refine, and publish. At general availability, it supports allocating up to 300 sandboxes per second, per cluster, with 90% of allocations completing in 200 milliseconds. Together, GKE and GKE Agent Sandbox gave us a reliable foundation for AI-generated interactive workloads that helped keep our team focused on the product experience.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;From black box to glass box&lt;/strong&gt;&lt;strong style="vertical-align: baseline;"&gt; &lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Before GKE Agent Sandbox, a failed sandbox workload could feel like flying blind. We could often see that something had gone wrong, but didn’t have enough runtime status, metrics, or failure signals to understand why.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Now, Google Cloud’s native logging and monitoring reach directly into those sandboxed environments, giving us a clearer view of workload behavior, faster issue resolution, and a stronger foundation for managing multi-tenant workloads.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;That visibility matters to developers, but it also matters to the platform’s users: A creator never sees the logs, the cluster, or the orchestration layer. They see whether an experience opens quickly, whether it responds when they draw, click, chat, or share, and whether they can keep building without friction. &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Flexibility that translates to savings&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;GKE Agent Sandbox also changed how we think about cost. Previously, running secure sandboxed environments meant stronger dependencies on specific server types, which limited how precisely we could match resources to each workload. With GKE Agent Sandbox, we can run secure, isolated workloads on appropriately sized cloud VMs. This gives us greater flexibility in resource allocation and helped us cut our infrastructure costs by up to 60%.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;That same flexibility extended to storage. Not all SeaVerse creations are built in a single session. Some evolve over time as creators return to refine them, build on earlier ideas, or invite others to remix what they’ve made. Our previous architecture didn’t support the persistent file-system capabilities those more complex use cases demanded, but that gap is gone now. We can attach persistent storage where workloads require it while maintaining the isolation boundaries that multi-tenant AI experiences need. For creators, that means experiences that are fast to open and easier to refine, revisit, and build on over time.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;The next remix&lt;/strong&gt;&lt;strong style="vertical-align: baseline;"&gt; &lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Supporting creations that can evolve and deepen is central to what we’re building. It’s still early in what playable AI can become. As the platform grows, we need to keep strengthening what matters most: stability, observability, elastic scaling, and cost efficiency, all in service of a creator experience that stays fast, reliable, and expressive.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;We’re also exploring additional Google Cloud tools to support smarter analytics and creation assistance. Gemini and agent models could help operators and creators better understand how experiences perform. &lt;/span&gt;&lt;a href="https://cloud.google.com/bigquery"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;BigQuery&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; AI and ML capabilities can support use cases such as churn prediction, LTV and ROI prediction, and user segmentation. Multimodal tools such as Imagen and Veo on &lt;/span&gt;&lt;a href="https://cloud.google.com/products/gemini-enterprise-agent-platform"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Gemini Enterprise Agent Platform&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; open up new possibilities for material analysis, creative generation, and AI interactive content production.&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Our goal is to make AI experiences feel immediate, expressive, and connected. With &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;GKE&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; and &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;GKE Agent Sandbox&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;, we have a stronger foundation for the next generation of playable AI.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Wed, 16 Sep 2026 16:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/containers-kubernetes/seaverse-chooses-gke-agent-sandbox/</guid><category>GKE</category><category>AI infrastructure</category><category>Customers</category><category>Containers &amp; Kubernetes</category><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>For SeaVerse, GKE Agent Sandbox reduces infrastructure costs by 60%</title><description></description><site_name>Google</site_name><url>https://cloud.google.com/blog/products/containers-kubernetes/seaverse-chooses-gke-agent-sandbox/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Zongyun Hu</name><title>COO, SeaVerse</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Tinsley Shi</name><title>Product Manager</title><department></department><company></company></author></item><item><title>How Orange built FinOps accountability, and why agents are next</title><link>https://cloud.google.com/blog/topics/telecommunications/how-orange-uses-agents-to-make-finops-everyones-responsibility/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;At &lt;/span&gt;&lt;a href="https://cloud.google.com/customers/orange"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Orange&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, the leading France-based multinational telecom provider, there are days when engineering teams set aside their delivery backlogs and spend the day cleaning up cloud spend together. There's a leaderboard. There are goodies on the line. Experienced practitioners guide the newcomers, so people learn the work while doing it. By the end of the day, sponsors can see the results.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Orange calls these FinOps Clean Days. Together with gamified hackathons, they've earned the company's 100-plus person FinOps community a Net Promoter Score within the organization that’s above 70.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Those numbers point at something the wider industry is wrestling with. Recent State of FinOps reports identify getting engineers to take action as one of the top challenges organizations face. Moving from awareness to action means finding ways to build FinOps accountability, and to get teams to genuinely care.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;That makes FinOps a business change problem. And business change problems have known solutions. We spoke with Camille Marini, the FinOps lead at Orange, to get a deeper understanding of how the company overcame these hurdles to accelerate AI adoption and ROI, and how your organization might follow the same course.&lt;/span&gt;&lt;/p&gt;
&lt;h2&gt;&lt;span style="vertical-align: baseline;"&gt;Why the Clean Days work&lt;/span&gt;&lt;/h2&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Orange has held two principles since it set up its FinOps team. First, Cloud FinOps is a shared responsibility, with every stakeholder in a project involved in their own way. And the only path to that shared responsibility runs through communication and a deliberate change effort. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;“We insisted on the concept of shared responsibility across the organization for our FinOps practices,” Marini told us. “It’s very similar to how we approach cloud security. We needed to make teams understand that every single stakeholder in a project is involved in FinOps, each in their own way, if we are going to achieve responsible and impactful AI spending and usage.”&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Those principles led Orange to create a FinOps Community of Practice, with support from Google Cloud Consulting. The team ran it on standardized communication channels so the methodology reached well beyond the central group, and kept the meetings actionable, sharing optimizations and billing updates so every session provided value.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The Clean Days came from a clear-eyed reading of how agile teams actually operate. In agile environments with deployment running constantly, optimization work rarely wins against the sprint. Delivery priorities, backlogs, and daily operations take the available time first. So Orange created protected time, made it collaborative, and made it fun.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;McKinsey's four building blocks of change explain why this approach lands. Any large organizational change, the framework holds, requires action across four areas:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Conviction and understanding: "I know what is expected of me and I agree with it."&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Formal mechanisms: "The structures, processes, and systems reinforce the change."&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Role modeling: "I see my leaders and colleagues behaving differently."&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Talent and skills: "I have the skills and opportunities to behave in a new way."&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Map Orange's practice onto those blocks and the pattern is visible. Gamification and rewards give engineers colleagues to emulate: The leaderboard makes different behavior visible, and sponsors see the quick wins for themselves. Experienced practitioners guiding novices builds talent and skills through the community itself. The regular sessions, sharing optimizations and billing updates, build the conviction that comes from knowing where the money goes.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/1_m5giQlH.max-1000x1000.jpg"
        
          alt="image2"&gt;
        
        &lt;/a&gt;
      
        &lt;figcaption class="article-image__caption "&gt;&lt;p data-block-key="j87xi"&gt;FinOps activities mapped to the four building blocks of change, with the points where AI agents can reinforce them.&lt;/p&gt;&lt;/figcaption&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h2&gt;&lt;span style="vertical-align: baseline;"&gt;What happens beyond 100 people&lt;/span&gt;&lt;/h2&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;A community of 100 engaged people is an achievement. But in an organization with thousands of engineers, no central FinOps team can reach everyone directly. The question for leaders is how to extend what a community like Orange's creates — the awareness, the shared ownership, the habit of acting — to people the FinOps team will never meet.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This is where AI agents extend the capabilities of a FinOps team with two core benefits. They take on complex, time-intensive activities that previously needed a human, and they reduce friction around FinOps for individuals across the business.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Getting teams to adopt them takes a strategy aimed at your own organization's pain points, which often come from high cognitive load, unclear accountability, or competing priorities. Start by finding where engagement drops off in your FinOps lifecycle:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;An awareness gap: If teams are unsure of their spend impact, an insight agent can push real-time cost data into their daily tools.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;A bandwidth gap: If engineers are too busy with backlogs, a remediation agent can identify quick wins and present them as ready-to-merge code changes.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;A complexity gap: If reporting feels like a manual chore, an orchestration agent can gather the data and simplify the process.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;h2&gt;&lt;span style="vertical-align: baseline;"&gt;Start with trust, then add autonomy&lt;/span&gt;&lt;/h2&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The sensible path runs in sequence. Establish the community practice, the way Orange did. Then introduce read-only agents that inform and suggest. Only once those are established across the community should you build agents that execute changes. Direct action carries operational risk, so manage it carefully. It's also where significant wins often sit.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;How you build depends on who's building. For teams that want to deploy quickly with minimal code, the &lt;/span&gt;&lt;a href="https://cloud.google.com/gemini-enterprise"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Gemini Enterprise App&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; provides a no-code environment for creating agents. For developers who need granular control, the &lt;/span&gt;&lt;a href="https://cloud.google.com/products/gemini-enterprise-agent-platform"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Gemini Enterprise Agent Platform&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; (formerly Vertex AI) offers advanced tools for launching and governing agents built with frameworks like the Agent Development Kit (ADK).&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Cloud FinOps is moving beyond centralized reporting toward action that happens where the work does. The organizations getting there start with the culture, then use agents to carry it further than any one team could reach. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Orange's numbers came out of the community work. Building that foundation is the part worth copying first. When you're ready to extend it, &lt;/span&gt;&lt;a href="https://cloud.google.com/consulting" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Google Cloud Consulting&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; can help you shape the community practice, and the Gemini Enterprise App is a low-lift way to put your first read-only agent in front of your teams.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Wed, 16 Sep 2026 16:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/topics/telecommunications/how-orange-uses-agents-to-make-finops-everyones-responsibility/</guid><category>AI &amp; Machine Learning</category><category>Customers</category><category>Google Cloud Consulting</category><category>Telecommunications</category><media:content height="540" url="https://storage.googleapis.com/gweb-cloudblog-publish/images/orange-finops-shared-responsibility.max-600x600.png" width="540"></media:content><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>How Orange built FinOps accountability, and why agents are next</title><description></description><image>https://storage.googleapis.com/gweb-cloudblog-publish/images/orange-finops-shared-responsibility.max-600x600.png</image><site_name>Google</site_name><url>https://cloud.google.com/blog/topics/telecommunications/how-orange-uses-agents-to-make-finops-everyones-responsibility/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Samuel Moss</name><title>AI Transformation and FinOps Consultant, Google</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Celine Devie</name><title>AI Transformation Consultant, Google</title><department></department><company></company></author></item><item><title>How KDDI built Buffmee, a faster, reliable consumer RAG app</title><link>https://cloud.google.com/blog/topics/customers/how-kddi-optimized-rag-performance-with-agent-development-kit/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;When building consumer-facing generative AI applications,  balancing high generation quality with fast response times across diverse media types, can be challenging. KDDI, a major telecommunications carrier in Japan, tackled this challenge head-on when they developed Buffmee, their consumer Retrieval-Augmented Generation (RAG) app.  &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span&gt;&lt;span style="vertical-align: baseline;"&gt;Buffmee is an interactive AI service built on the concept of 'AI that helps you grow.' By grounding responses in over 100 sources — including books, magazines, and web media — it helps users search for information, summarize key points, and explore personalized learning and hobby interests. By citing sources, Buffmee alleviates concerns about information reliability, allowing users to safely deepen their knowledge. &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; To achieve this, KDDI collaborated closely with their development partner KDDI iret, Google Cloud Consulting and our specialized AI engineers.&lt;/span&gt;&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;As part of their app launch, the engineer team needed to ground a massive variety of proprietary content, including books and magazines. However, they struggled with latency issues that prevented them from meeting their target response times, and they needed a reliable way to ensure hallucination-free results. &lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/image1_9ZYjQgt.max-1000x1000.png"
        
          alt="image1"&gt;
        
        &lt;/a&gt;
      
        &lt;figcaption class="article-image__caption "&gt;&lt;p data-block-key="xjvun"&gt;Buffmee App Description and Images&lt;/p&gt;&lt;/figcaption&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To meet these performance targets, organizations need a systematic approach to AI evaluation and real-time bottleneck identification. That is why we are sharing the automated evaluation framework and performance optimization techniques that helped KDDI successfully launch their application. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The results were inspiring: &lt;/span&gt;&lt;strong style="vertical-align: baseline;"&gt;KDDI reduced total application response latency by 38%, successfully hitting their target response performance. They also achieved a nearly 18% improvement in TTFT.&lt;/strong&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;"Our vision hinged on a platform where content, once ingested, would instantly function as a working RAG system. Google's careful, hands-on guidance made that a reality — we're sincerely grateful for their support." — Shunya Onoda, AI Product Department, KDDI.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With these performance and accuracy improvements, Buffmee now empowers users to safely explore their favorite media through interactive Q&amp;amp;A and deep-dive analysis, delivering a highly personalized experience while maintaining strict trust and compliance for content providers.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Let’s deep dive into how they achieved these results. &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Establish automated evaluation for diverse content&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Traditional manual testing requires immense effort and cannot scale to accommodate a large content library. To solve this, the development team designed a systematic AI evaluation process using Gemini Enterprise Agent Platform Evaluation Service.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;By implementing automated evaluation frameworks like LLM-as-a-Judge and the Rule of Hundreds, the team replaced labor-intensive manual testing with a data-driven process. They ingested their extensive document corpus, constructed hundreds of automated evaluation tests, and built a comprehensive benchmark dataset to measure the reliability of answers for each use case. As a result, the team improved their groundedness scores by 25%, helping deliver highly accurate and reliable outputs.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/image2_R5gWUyX.max-1000x1000.png"
        
          alt="image2"&gt;
        
        &lt;/a&gt;
      
        &lt;figcaption class="article-image__caption "&gt;&lt;p data-block-key="xjvun"&gt;KDDI's automated evaluation loop: AI generates questions and scores answers, while humans calibrate thresholds and analyze edge-case failures.&lt;/p&gt;&lt;/figcaption&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Identify bottlenecks and optimize performance with an agentic loop&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To improve response speeds, the team implemented BigQuery Agent Analytics and the Agent Development Kit (ADK) log analysis agent. By analyzing actual production logs, they visualized how skill division and prompt bloat—especially with highly complex, multi-page system prompts — impacted the Time To First Token (TTFT).&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The team optimized the system prompt, including the inline integration of skills, and reviewed the sub-agent routing. This allowed them to identify and resolve deep-stack bottlenecks in real time without sacrificing response accuracy.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Four core principles for reliable evaluation &lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;To achieve these results, the team implemented four core technical practices:&lt;/span&gt;&lt;/p&gt;
&lt;ol&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Transitioning to binary evaluation: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;By selectively moving away from ambiguous 1–5 ratings to a binary "pass (1) / fail (0)" system for critical metrics, the team minimized variance and noise, helping improve automation accuracy.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Strategic content sampling:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Rather than attempting to evaluate every single document, the team classified their entire corpus along a two-dimensional grid: File Format (Web articles, EPUBs, PDFs, structured data) and Media Composition (Text-heavy, image-heavy, or mixed). By selecting representative samples from each cell of this difficulty grid, they reduced the evaluation workload by 75% while maintaining comprehensive test coverage.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Thresholds grounded in product judgment:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Instead of relying solely on default tool parameters, the product owner reviewed randomly sampled answers alongside their automated scores to calibrate and establish what "good enough to ship" actually meant for the user experience.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Modular splitting of massive prompts into ADK Skills:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Because massive system prompts exceeding 800 lines can cause LLM attention drift and latency degradation, the team split prompts by function into Agent Development Kit (ADK) Skills, dynamically loading only the required logic to optimize response times.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ol&gt;
&lt;h3&gt;&lt;span style="vertical-align: baseline;"&gt;Get started&lt;/span&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Building scalable, reliable generative AI applications requires both automated evaluation and deep performance analytics. To apply these techniques to your own applications:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Measure quality systematically with the &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/evaluation-overview?hl=ja"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Gen AI evaluation service&lt;/span&gt;&lt;/a&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Structure your agents with &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/gemini-enterprise-agent-platform/build/adk"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Agent Development Kit&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and apply progressive disclosure deliberately&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Ground your agents with &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/generative-ai-app-builder/docs"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Agent Search&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and inspect your retrieval queries&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;&lt;/div&gt;</description><pubDate>Tue, 08 Sep 2026 16:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/topics/customers/how-kddi-optimized-rag-performance-with-agent-development-kit/</guid><category>AI &amp; Machine Learning</category><category>Telecommunications</category><category>Customers</category><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>How KDDI built Buffmee, a faster, reliable consumer RAG app</title><description></description><site_name>Google</site_name><url>https://cloud.google.com/blog/topics/customers/how-kddi-optimized-rag-performance-with-agent-development-kit/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Junichi Kashino</name><title>Platform Business Strategy Department, KDDI</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Miki Katsuragi</name><title>AI Consultant, Google Cloud Japan</title><department></department><company></company></author></item><item><title>How Yahoo optimizes resources with flexible VMs in Managed Service for Apache Spark</title><link>https://cloud.google.com/blog/products/data-analytics/how-yahoo-optimizes-apache-spark-with-flexible-vms/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;As a global media and technology company connecting hundreds of millions of users to finance, sports, and entertainment platforms, Yahoo operates a massive data infrastructure where analytics workloads must run continuously at high speed. In deadline-driven data environments, relying on fixed virtual machine (VM) configurations creates a brittle system; if a specific machine shape faces a regional capacity constraint, cluster provisioning in &lt;/span&gt;&lt;a href="https://cloud.google.com/products/managed-service-for-apache-spark"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Managed Service for Apache Spark&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; (formerly Dataproc) can experience delays and stall critical data pipelines.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Yahoo utilizes &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/managed-spark/docs/concepts/configuring-clusters/flexible-vms"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;flexible VMs&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; in &lt;/span&gt;&lt;a href="https://cloud.google.com/products/managed-service-for-apache-spark"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Managed Service for Apache Spark&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; clusters to automatically absorb these resource fluctuations by defining a ranked list of acceptable VM shapes. This allows the system to dynamically search regional zones and maintain pipeline execution without manual intervention. To search for capacity across a region, teams must also enable &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/managed-spark/docs/concepts/configuring-clusters/flexible-vms"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Auto-Zone placement&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This optimization builds on Yahoo's broader data modernization journey, which involved &lt;/span&gt;&lt;a href="https://www.youtube.com/watch?v=_7Oz1V1-ZiE" rel="noopener" target="_blank"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;migrating on-premises Hadoop and big data estates&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; directly to Google Cloud. By transitioning those legacy workloads, the team established a cloud foundation capable of running high-scale batch and streaming analytics with dynamic resource flexibility.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-video"&gt;



&lt;div class="article-module article-video "&gt;
  &lt;figure&gt;
    &lt;a class="h-c-video h-c-video--marquee"
      href="https://youtube.com/watch?v=_7Oz1V1-ZiE"
      data-glue-modal-trigger="uni-modal-_7Oz1V1-ZiE-"
      data-glue-modal-disabled-on-mobile="true"&gt;

      
        

        &lt;div class="article-video__aspect-image"
          style="background-image: url(https://storage.googleapis.com/gweb-cloudblog-publish/images/maxresdefault_iMaqL8o.max-1000x1000.jpg);"&gt;
          &lt;span class="h-u-visually-hidden"&gt;Hadoop pioneer to cloud innovator: Yahoo’s data lake modernization journey&lt;/span&gt;
        &lt;/div&gt;
      
      &lt;svg role="img" class="h-c-video__play h-c-icon h-c-icon--color-white"&gt;
        &lt;use xlink:href="#mi-youtube-icon"&gt;&lt;/use&gt;
      &lt;/svg&gt;
    &lt;/a&gt;

    
  &lt;/figure&gt;
&lt;/div&gt;

&lt;div class="h-c-modal--video"
     data-glue-modal="uni-modal-_7Oz1V1-ZiE-"
     data-glue-modal-close-label="Close Dialog"&gt;
   &lt;a class="glue-yt-video"
      data-glue-yt-video-autoplay="true"
      data-glue-yt-video-height="99%"
      data-glue-yt-video-vid="_7Oz1V1-ZiE"
      data-glue-yt-video-width="100%"
      href="https://youtube.com/watch?v=_7Oz1V1-ZiE"
      ng-cloak&gt;
   &lt;/a&gt;
&lt;/div&gt;

&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This post provides a technical blueprint for configuring flexible VM instance rankings in &lt;/span&gt;&lt;a href="https://cloud.google.com/products/managed-service-for-apache-spark"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Managed Service for Apache Spark&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to automatically manage capacity constraints and maintain pipeline execution.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Operational trade-offs of static configurations&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Configuring clusters with a single, fixed machine type in a specific zone introduces constraints when regional zonal capacity fluctuations occur, potentially impacting cluster provisioning. Rather than manage these capacity variations through custom retry logic or manual intervention, using flexible configurations allows your infrastructure to automatically adapt. By accepting multiple VM shapes and searching across zones in the selected region, flexible configurations help streamline provisioning to better support high-scale analytics workloads.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Rules for configuring flexible clusters&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Deploying flexible configurations requires aligning several connected design choices:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Enable auto-zone placement:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; You must pass a region(&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;--region=${REGION}&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;) or an empty zone string (&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;--zone=""&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;) so Managed Spark can search for available capacity across the entire region.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Maintain core and memory symmetry:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; If your Managed Spark cluster uses &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/managed-spark/docs/concepts/configuring-clusters/autoscaling"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;autoscaling&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, all machine types in your flexible list must share a similar core count and memory size, even if they come from different VM families. A uniform CPU-to-memory ratio across primary and secondary workers prevents performance degradation, as the smallest ratio determines your effective container sizing.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Align component properties:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Managed Spark calculates system properties based on VM cores and memory. When mixing machine shapes, you may need explicit property overrides to keep YARN and Spark resource allocations aligned with your expected worker behavior.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Two ways flexible VMs support massive workloads&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;For large-scale data environments, flexible configurations support operations in two ways:&lt;/span&gt;&lt;/p&gt;
&lt;ol&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Higher cluster creation success:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Instead of failing when a preferred VM type is out of stock, Managed Spark selects from a ranked list to keep provisioning moving.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Better regional resource use:&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt; Auto-zone placement searches the entire region to find capacity, which reduces provisioning friction during high-demand periods.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ol&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;gcloud example&lt;/strong&gt;&lt;/h3&gt;&lt;/div&gt;
&lt;div class="block-code"&gt;&lt;dl&gt;
    &lt;dt&gt;code_block&lt;/dt&gt;
    &lt;dd&gt;&amp;lt;ListValue: [StructValue([(&amp;#x27;code&amp;#x27;, &amp;#x27;gcloud dataproc clusters create analytics-cluster \\\r\n  --region=us-central1 \\\r\n  --zone=&amp;quot;&amp;quot; \\\r\n  --num-workers=10 \\\r\n  --master-instance-selection=\&amp;#x27;{&amp;quot;machineTypes&amp;quot;:[&amp;quot;e2-standard-8&amp;quot;],&amp;quot;rank&amp;quot;:0}\&amp;#x27; \\\r\n  --master-instance-selection=\&amp;#x27;{&amp;quot;machineTypes&amp;quot;:[&amp;quot;n2-standard-8&amp;quot;],&amp;quot;rank&amp;quot;:1}\&amp;#x27; \\\r\n  --worker-instance-selection=\&amp;#x27;{&amp;quot;machineTypes&amp;quot;:[&amp;quot;e2-standard-8&amp;quot;],&amp;quot;rank&amp;quot;:0}\&amp;#x27; \\\r\n  --worker-instance-selection=\&amp;#x27;{&amp;quot;machineTypes&amp;quot;:[&amp;quot;n2-standard-8&amp;quot;],&amp;quot;rank&amp;quot;:1}&amp;#x27;), (&amp;#x27;language&amp;#x27;, &amp;#x27;&amp;#x27;), (&amp;#x27;caption&amp;#x27;, &amp;lt;wagtail.rich_text.RichText object at 0x7fa19fab9950&amp;gt;)])]&amp;gt;&lt;/dd&gt;
&lt;/dl&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;API example&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;You can also build this capacity policy into your automated pipelines or &lt;/span&gt;&lt;a href="https://cloud.google.com/products/managed-service-for-apache-airflow"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Managed Service for Apache Airflow&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; DAGS using the &lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt;instanceFlexibilityPolicy&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; field in the ‘Dataproc’ API:&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-code"&gt;&lt;dl&gt;
    &lt;dt&gt;code_block&lt;/dt&gt;
    &lt;dd&gt;&amp;lt;ListValue: [StructValue([(&amp;#x27;code&amp;#x27;, &amp;#x27;{\r\n  &amp;quot;projectId&amp;quot;: &amp;quot;PROJECT_ID&amp;quot;,\r\n  &amp;quot;clusterName&amp;quot;: &amp;quot;analytics-cluster&amp;quot;,\r\n  &amp;quot;config&amp;quot;: {\r\n    &amp;quot;gceClusterConfig&amp;quot;: {\r\n      &amp;quot;zoneUri&amp;quot;: &amp;quot;&amp;quot;\r\n    },\r\n    &amp;quot;secondaryWorkerConfig&amp;quot;: {\r\n      &amp;quot;numInstances&amp;quot;: 8,\r\n      &amp;quot;instanceFlexibilityPolicy&amp;quot;: {\r\n        &amp;quot;instanceSelectionList&amp;quot;: [\r\n          {\r\n            &amp;quot;machineTypes&amp;quot;: [&amp;quot;n2-standard-8&amp;quot;],\r\n            &amp;quot;rank&amp;quot;: 0\r\n          },\r\n          {\r\n            &amp;quot;machineTypes&amp;quot;: [&amp;quot;e2-standard-8&amp;quot;, &amp;quot;t2d-standard-8&amp;quot;],\r\n            &amp;quot;rank&amp;quot;: 1\r\n          }\r\n        ]\r\n      }\r\n    }\r\n  }\r\n}&amp;#x27;), (&amp;#x27;language&amp;#x27;, &amp;#x27;&amp;#x27;), (&amp;#x27;caption&amp;#x27;, &amp;lt;wagtail.rich_text.RichText object at 0x7fa19fa4c3d0&amp;gt;)])]&amp;gt;&lt;/dd&gt;
&lt;/dl&gt;&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;This API policy achieves the same goal: it establishes your preferred shape, documents valid fallbacks, and lets Managed Spark resolve resource constraints without breaking your automation scripts.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Establishing an infrastructure policy&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Managing data at this scale requires standardizing a clear resource policy rather than relying on a single rigid machine type. Your configuration standards should outline:&lt;/span&gt;&lt;/p&gt;
&lt;ul&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Preferred and fallback VM families for secondary workers.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Default auto-zone placement to enable flexible provisioning.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Identical core and memory configurations when using autoscaling.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Uniform CPU-to-memory ratios across all worker groups to maintain predictable container sizing.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Explicit YARN or Spark property overrides to guarantee consistent runtime behavior across different machine lines.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: disc; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;span style="vertical-align: baseline;"&gt;Shuffle-safe patterns for Spark workloads running on Spot or highly elastic capacity.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ul&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;By adopting flexible configurations, you turn infrastructure scarcity into a predictable fallback plan, keeping your critical data pipelines up and running.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Yahoo impact and results&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;By implementing flexible VMs in Managed Service for Apache Spark, Yahoo successfully reduced cluster provisioning failures by 85% which were caused by regional capacity stockouts. This flexible configuration allows their data infrastructure to automatically handle capacity constraints and successfully provision resources without requiring manual intervention. As a result, Yahoo ensures continuous workload execution and prevents downstream processing delays across their massive data pipelines.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;&lt;span style="font-style: italic; vertical-align: baseline;"&gt;"Managing high-scale data analytics at Yahoo requires resilient, automated infrastructure. Moving to flexible VMs in Managed Service for Apache Spark has transformed our approach; instead of stalling when a specific machine shape faces capacity constraints, our clusters now automatically pivot to our ranked fallback options. This has helped us reduce provisioning failures by 85%, providing the reliability we need to keep our global media platforms running smoothly."&lt;/span&gt;&lt;span style="vertical-align: baseline;"&gt; - Akshay Jain, Senior Software Developer Engineer, Yahoo! &lt;/span&gt;&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Strategic benefits of flexible infrastructure&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Adopting a flexible compute stack transforms your environment into a dynamic pool of resources that adapts to your operational needs. By moving away from rigid, single-machine type configurations, you ensure that your workloads reliably access the compute they need, regardless of supply fluctuations. This shift not only maximizes workload obtainability and reliability but also facilitates seamless hardware modernization by allowing you to prioritize newer VM generations while maintaining older types as reliable fallback options.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Build your resilient data pipeline&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Transitioning to a fluid compute strategy ensures your critical analytics remain operational despite regional resource shifts. Here is how you can begin optimizing your infrastructure today:&lt;/span&gt;&lt;/p&gt;
&lt;ol&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Audit your workloads: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Identify applications tightly coupled to specific VM families or zones and map out viable alternative hardware shapes.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Standardize resource policies: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Explore the &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/managed-spark/docs/concepts/configuring-clusters/flexible-vms"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;documentation for Managed Spark flexible VMs&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to establish your preferred and fallback VM families.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Align financial strategy: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;Utilize Flexible Committed Use Discounts (Flex CUDs) to maintain cost predictability when workloads dynamically pivot to alternative machine types.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Claim your credits: &lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;New customers may be eligible for &lt;/span&gt;&lt;a href="https://cloud.google.com/free"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;$300 in credits&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; to try Managed Service for Apache Spark and other Google Cloud products at no cost.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ol&gt;&lt;/div&gt;</description><pubDate>Fri, 04 Sep 2026 16:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/products/data-analytics/how-yahoo-optimizes-apache-spark-with-flexible-vms/</guid><category>Streaming</category><category>Customers</category><category>Data Analytics</category><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>How Yahoo optimizes resources with flexible VMs in Managed Service for Apache Spark</title><description></description><site_name>Google</site_name><url>https://cloud.google.com/blog/products/data-analytics/how-yahoo-optimizes-apache-spark-with-flexible-vms/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Akshay Jain</name><title>Senior Software Engineer, Yahoo</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Surjit Singh</name><title>Data &amp; AI Engineer, Google Cloud</title><department></department><company></company></author></item><item><title>How BlackLine simplifies perimeter policy intelligence with VPC Service Controls</title><link>https://cloud.google.com/blog/topics/customers/how-blackline-prevents-data-exfiltration-with-vpc-service-controls/</link><description>&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Establishing network-level perimeters with VPC Service Controls (VPC-SC) is a critical step that can help you protect your cloud environment against data exfiltration, compromised accounts, and insider threats.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Today, Google Cloud is excited to share new policy intelligence capabilities in VPC-SC that can help drive even greater operational simplicity. With our latest release of the &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/vpc-service-controls/docs/violation-analyzer"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;VPC-SC violation analyzer&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/vpc-service-controls/docs/violation-dashboard"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;violation dashboard&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt;, we have simplified policy management and troubleshooting, to make managing and optimizing your security perimeter more efficient and straightforward than ever. &lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;How BlackLine streamlines incident response&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;BlackLine, a leader in financial operations management, adopted the VPC-SC policy intelligence solution to maintain strict security perimeters. Chosen by over half of Fortune 500 companies, BlackLine uses Google Cloud's full suite of managed services and built-in security capabilities to protect sensitive customer financial data.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;VPC Service Controls are the foundation of BlackLine's preventative compliance and security controls in our Google Cloud environment, helping us to mitigate data exfiltration risks and ensure clear separation between our higher and lower environments by establishing strong security perimeters.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Managing these complex perimeters is a continuous process. VPC Service Controls violation analyzer helps BlackLine cloud infrastructure administrators adapt to changing API connection requirements of the business by adjusting security perimeters through approved access levels, ingress policies, and egress policies. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With only the troubleshooting token or unique ID from any VPC-SC violation error message, we can produce a detailed report identifying the principals and target resources involved in a failed API request, and explaining why and how that API request violated BlackLine's service perimeters. We don’t need to write a Cloud Logging SQL query to extract the data.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;The clear access context and actionable insights in the violation details report are an invaluable starting point as we collaborate to resolve violations, significantly reducing our mean-time-to-resolution (MTTR) for service perimeter issues, and helping BlackLine maintain our focus on our customers and continue to innovate on their behalf.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Streamlining the perimeter operations lifecycle&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Our new policy intelligence tools — the VPC-SC &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/vpc-service-controls/docs/violation-analyzer"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Violation analyzer&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/vpc-service-controls/docs/violation-dashboard"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Violation dashboard&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; — simplify real-time monitoring and active incident response. These tools provide clear, actionable insights in the Google Cloud Console, offering greater speed and automation to help you confidently enforce least-privilege perimeters, and quickly resolve access denials.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Violation Dashboard aggregates and visualizes all service perimeter violations across your entire Google Cloud organization in a single pane of glass, helping your team identify trends, spot spikes in access denials, and shareable filters on violations by specific perimeters, projects, or identities.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Violation Analyzer streamlines investigating violations, eliminating the need to query &lt;/span&gt;&lt;a href="https://cloud.google.com/logging"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;Cloud Logging&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and manually piece together the details. When you click a troubleshooting token from the dashboard (or input a unique denial ID), the analyzer maps out the identity, source, target, and VPC-SC rule triggered, creating a report telling you why that specific request was blocked. This helps your team more quickly take action to determine whether to modify existing policy rules or create a new one, and resolve incidents more quickly.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Together, the new VPC Service Controls policy intelligence tools go beyond automated log analysis to provide unified visibility of violations and actionable insights to investigate them, making your perimeter deployment and management simpler and lower-risk.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/images/1_HT1HJeh.max-1000x1000.png"
        
          alt="1"&gt;
        
        &lt;/a&gt;
      
        &lt;figcaption class="article-image__caption "&gt;&lt;p data-block-key="ndthf"&gt;Streamlining the VPC Service Controls lifecycle, from deployment to policy refinement.&lt;/p&gt;&lt;/figcaption&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;With the new VPC-SC troubleshooting tools you can more easily:&lt;/span&gt;&lt;/p&gt;
&lt;ol&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Test new perimeters (deployment)&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Use the violation dashboard to visualize the impact of a service perimeter during your initial dry run phase, helping to verify that enforcement is accurate and predictable before it affects production traffic. Filter violations to track and resolve with prebuilt contextual filters for principals, service perimeters, enforcement type, and more.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Track perimeter denials (monitor)&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: The violation dashboard offers a unified view of your perimeter health, allowing your security operations team to monitor status in real time, including dynamic agentic access denials.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Triage an event (investigate)&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Violation analyzer provides the identity, source, target, and operations for any violation. It cross-references identity and access management (IAM) permissions, resource ancestry, and context evaluation to identify which rule was triggered, reducing manual effort.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;li aria-level="1" style="list-style-type: decimal; vertical-align: baseline;"&gt;
&lt;p role="presentation"&gt;&lt;strong style="vertical-align: baseline;"&gt;Fix the rule (refine policy)&lt;/strong&gt;&lt;span style="vertical-align: baseline;"&gt;: Instead of searching through configuration files, violation analyzer maps violations directly to the relevant line in your VPC-SC policy, allowing you to make updates more quickly and with less manual overhead.&lt;/span&gt;&lt;/p&gt;
&lt;/li&gt;
&lt;/ol&gt;&lt;/div&gt;
&lt;div class="block-image_full_width"&gt;






  
    &lt;div class="article-module h-c-page"&gt;
      &lt;div class="h-c-grid"&gt;
  

    &lt;figure class="article-image--large
      
      
        h-c-grid__col
        h-c-grid__col--6 h-c-grid__col--offset-3
        
        
      "
      &gt;

      
      
        
        &lt;img
            src="https://storage.googleapis.com/gweb-cloudblog-publish/original_images/output_hq.gif"
        
          alt="output_hq"&gt;
        
        &lt;/a&gt;
      
        &lt;figcaption class="article-image__caption "&gt;&lt;p data-block-key="ndthf"&gt;The VPC Service Controls violation dashboard produces detailed reports to jump-start perimeter access investigations that are simplified using the violation analyzer.&lt;/p&gt;&lt;/figcaption&gt;
      
    &lt;/figure&gt;

  
      &lt;/div&gt;
    &lt;/div&gt;
  




&lt;/div&gt;
&lt;div class="block-paragraph_advanced"&gt;&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Core VPC-SC operations: Simple perimeter enforcement&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Our new troubleshooting capabilities build on VPC Service Controls’ foundational simplicity for designing, enforcing, and managing strong perimeters. &lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;By using dry run mode, your teams can build precise, contextual ingress and egress rules based on observed traffic — without disrupting vital business workflows. Once you validate these access patterns, moving to full enforcement becomes a more confident, data-driven process. To keep perimeter maintenance more efficient and straightforward, scoped policies allow you to delegate management directly to project-level administrators, empowering the teams closest to the workload.&lt;/span&gt;&lt;/p&gt;
&lt;h3&gt;&lt;strong style="vertical-align: baseline;"&gt;Getting started&lt;/strong&gt;&lt;/h3&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Simplify data security with VPC Service Controls. With the new Violation Analyzer and Violation dashboard, you can spend less time investigating incidents and more time safely scaling your cloud initiatives. Your data is your most valuable asset — protect it with a perimeter that’s as simple to manage as it is effective in enforcing controls.&lt;/span&gt;&lt;/p&gt;
&lt;p&gt;&lt;span style="vertical-align: baseline;"&gt;Learn more and get started with the VPC-SC &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/vpc-service-controls/docs/violation-analyzer"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;violation analyzer&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; and &lt;/span&gt;&lt;a href="https://docs.cloud.google.com/vpc-service-controls/docs/violation-dashboard"&gt;&lt;span style="text-decoration: underline; vertical-align: baseline;"&gt;violation dashboard&lt;/span&gt;&lt;/a&gt;&lt;span style="vertical-align: baseline;"&gt; in our documentation.&lt;/span&gt;&lt;/p&gt;&lt;/div&gt;</description><pubDate>Tue, 01 Sep 2026 16:00:00 +0000</pubDate><guid>https://cloud.google.com/blog/topics/customers/how-blackline-prevents-data-exfiltration-with-vpc-service-controls/</guid><category>Security &amp; Identity</category><category>Customers</category><og xmlns:og="http://ogp.me/ns#"><type>article</type><title>How BlackLine simplifies perimeter policy intelligence with VPC Service Controls</title><description></description><site_name>Google</site_name><url>https://cloud.google.com/blog/topics/customers/how-blackline-prevents-data-exfiltration-with-vpc-service-controls/</url></og><author xmlns:author="http://www.w3.org/2005/Atom"><name>Pratik Bhangale</name><title>Product Manager, Google Cloud</title><department></department><company></company></author><author xmlns:author="http://www.w3.org/2005/Atom"><name>Jimmy Huang</name><title>Staff Cloud Engineer, BlackLine</title><department></department><company></company></author></item></channel></rss>