Customers

Startups are building the next big thing with Google Cloud AI

Wed, 22 Apr 2026 12:00:00 +0000

The future is taking shape in Las Vegas this week, where the world’s leading startups are showcasing their groundbreaking AI work at Google Cloud Next. Whether they need the top AI models, infrastructure purpose-built for the most demanding workloads, intelligent multi-model databases, or security and sovereignty solutions to grow with confidence — or all of the above — startups of all sizes and in every industry are choosing Google Cloud.

To take one example of thousands, Lovable, the popular software creation platform, utilizes Google Cloud infrastructure and builders are now creating over 200,000 new projects on Lovable every day. OpenEvidence, a medical AI platform widely used by U.S. physicians, uses Google's AI tools to provide evidence-based answers at the point of care; the company recently announced major partnerships with Wiley and Mount Sinai, expanding access to trusted medical research and to its AI assistant.

This week at Next we’re announcing even more startups who are starting or expanding their work with Google Cloud. These include leading names in areas like vibe coding, model training and inference; healthcare and life sciences; image generation; financial services; gaming and much more. It also includes an exciting new group of startups who have recently begun accessing GPU-based AI compute through Google Cloud’s AI Hypercomputer, thanks to our deep technical partnership with NVIDIA.

Google Cloud has always offered startups the entire AI stack in one unified place, giving founders maximum velocity. With new exciting launches at Next ‘26, the options and opportunities are only growing.

We’re helping you build complex agents faster with Gemini Enterprise Agent Platform, optimize compute needs with two differentiated TPU offerings for inference and training, reduce data friction with our next-gen cross-cloud Lakehouse, and better govern everything with the help of Wiz AI agents and Google-scale threat intelligence, to name just a few advances. We’ve also unveiled the ability to sell your AI solutions directly to enterprise users through marketplace integrations in Gemini Enterprise.

Supporting startup innovation across categories

If there was any question of just how dramatically AI is transforming every facet of work and life, look no further than the innovations being brought to market by our startup customers.

Coding and developer tools

Emergent AI uses Google Kubernetes Engine (GKE) as its core infrastructure to orchestrate thousands of on-demand, isolated development sandboxes for vibe coding. By leveraging GKE Autopilot for automated scaling and Gemini models for code generation, the platform transforms natural language prompts into live, production-ready applications with instant previews.
Lovable is expanding its work with Google Cloud and launching a new coding agent, now available to purchase through Google Cloud Marketplace and Gemini Enterprise.
Parallel builds high accuracy web search and research APIs, purpose-built for AI agents. Leveraging Google Cloud offerings including Gemini, BigQuery, BigTable, and Spanner has enabled them to build, launch and massively scale their platform since launching in August 2025.
Temporal partnered with Google Cloud to build a unique Application Development Kit (ADK) integration. This enables organizations to deploy agentic systems for mission-critical use cases that require strong execution guarantees, uptime, auditability, and operational resilience.

Customer engagement

Satisfi Labs is an agentic AI platform designed to transform customer engagement for destinations and experiences, serving 800 clients across sports, entertainment, and tourism. They are built on the Google Cloud ecosystem and leverage technologies including Gemini, BigQuery, AlloyDB, and Datastream to enable intelligent interactions, scalable data processing, and real-time insights.
Vapi, a platform for developers creating conversational voice AI, leverages Gemini as its intelligence core, allowing developers to maximize their compute-per-dollar while delivering a premium voice experience with incredible scale.
Vurvey Labs leverages Gemini Enterprise Agent Platform to build agent populations and simulated worlds that predict human behavior and future needs.
ZenBusiness utilizes Gemini Enterprise Agent Platform to simplify business ownership, including the launch of 'Velo,' an AI assistant for formation, compliance, and website building.

Document intelligence

Reducto leverages Gemini Enterprise Agent Platform and AI Studio alongside its core models to automate high-stakes workflows and power production AI pipelines at scale.

Finops

Mavvrik is a FinOps platform using Gemini and BigQuery to power their unified cost management platform for FinOps, IT, and finance leaders to gain visibility, automate allocation, and govern spending across public clouds, private clouds, AI workloads, and SaaS applications.

Generative media

ComfyUI is streamlining complex node-based gen media production by integrating Nano Banana Pro while improving the consistency of their customer experience using Provisioned Throughput.
Gamma used by over 70 million people to create presentations, documents, websites, and social posts, uses Nano Banana 2 to instantly generate localized visuals and transform ideas into ready-to-share content in seconds.

Industry AI

Chorus, spun out of Alphabet’s X innovation arm, is building AI orchestration tools to transform how physical assets are made, moved, and managed across industries. Leveraging Gemini Enterprise Agent Platform, advanced machine learning, and proprietary algorithms, Chorus provides granular asset-level visibility, temperature intelligence, and dynamic delivery estimates across mission-critical supply chains.
ExaCare AI, an AI platform for post-acute care operations, is using Gemini Enterprise Agent Platform and Gemini for inference and to process clinical and operational data across admissions and reimbursement workflows.
Industrility, an AI-powered aftersales platform for machine manufacturers, built a maintenance intelligence engine on Google Gemini’s advanced language models, extracting critical maintenance tasks and reducing a two-month manual project to three hours.
Insilica uses Gemini Enterprise Agent Platform to process millions of toxicology regulatory documents and thousands of databases into a unified, agentic risk assessment platform.
Proximal Health utilizes DocAI to automate the insurance claims adjudication process.
Stord, which offers fulfillment and software for omnichannel retail, leverages Gemini Enterprise Agent Platform to build ML models that predict estimated delivery dates and optimize demand planning across millions of orders, all grounded in real data behind over $10B in annual Gross Merchandise Value.
Stylitics uses Gemini Enterprise Agent Platform to power more than 1,000 AI programs for world-class retailers, including AI image generation, outfitting and styling, and AI analytics.
Wand uses Gemini Enterprise Agent Platform to power Game Guide, a contextual companion built into their single-player PC gaming platform, that helps players get unstuck, uncover hidden content, and stay immersed without ever leaving the game window.

Productivity

Notion is at the forefront of integrating artificial intelligence into productivity tools, leveraging Gemini models to power its text and image generation capabilities.

Sustainability

Watershed, a leading enterprise climate platform, is leveraging Google Cloud to transform sustainability data into actionable climate intelligence, accelerating global decarbonization and enabling organizations to move from measurement to action at an unprecedented scale.

Supporting model training and inference with NVIDIA AI infrastructure

Today, when the world’s most innovative startups need to access NVIDIA’s cutting-edge GPU-powered compute, many of them rely on Google Cloud. Our longstanding partnership with NVIDIA allows these startups to train and serve AI models, support fine-tuning, and scale new agentic workloads.

This week at Google Cloud Next, we’re highlighting the successes these customers are having and announcing exciting new partnerships with more startups choosing to utilize NVIDIA AI infrastructure and Google Cloud’s AI stack.

These new and expanding customers include:

Thinking Machines Labs will utilize the latest NVIDIA Blackwell chips through Google Cloud, including being among the first Google Cloud customers to utilize GB300s. In early testing, Thinking Machines has seen training and serving speed increase 2x on GBs compared to prior generation chips.
GMI Cloud leverages Google Cloud’s GPU infrastructure to run its inference-optimized platform, powering GMI Studio workflows and large-scale model serving across text, image, and video. This infrastructure supports high utilization driven by growing demand for real-time, production-grade AI applications.
Inferact is developing vLLM, a highly performant and easy to use inference engine for open source LLMs on data center hardwares, utilizing a cluster of NVIDIA GB200’s on Google Cloud, as well as services like Managed Lustre and Cloud Filestore.
Neuraldefend is partnering with Google Cloud and NVIDIA to utilize H100 and L4 GPUs for model training and production inference for their real-time deepfake detection engine and digital trust solutions.
Nucleus AI is utilizing a cluster of NVIDIA H100 GPUs through Google Cloud for model training.
Reflection has expanded its usage of G4 VMs through Google Cloud, as well as services like Google Cloud Storage to support training and inference for its open-weight models.
Reve uses A3 Mega (NVIDIA H100 GPUs) and A3 Ultra (NVIDIA H200 GPUs) instances on Google Cloud to handle their intensive multimodal foundation model training and inference. They also use GKE to manage their high performance training needs.

Startups most commonly access NVIDIA GPUs through Google Cloud G4 VMs, powered by NVIDIA’s latest Blackwell infrastructure. We’re also excited to offer NVIDIA’s Vera Rubin architecture to startups later this year. Our deep partnership with NVIDIA means that the Google Cloud AI Hypercomputer is highly optimized for GPU customers and can deliver ultra-low latency, high-throughput, and cost-effective performance across multiple use cases.

Startups are choosing to utilize GPUs through Google Cloud in part because of this high level of optimization and because they can easily access highly differentiated technologies in our AI stack like Gemini Enterprise, Agent Builder, BigQuery, Google Kubernetes Engine, and much more.

We also continue to partner closely with NVIDIA to support early-stage startups with AI compute and cloud services through NVIDIA’s Inception program and our Google for Startups Cloud Program. This includes exciting startups like:

Aible, which is building long‑running enterprise AI agents. They are accessing GPUs on Google Cloud to help them run Nemotron 3 and NemoClaw agents directly on BigQuery, orchestrate multi‑agent blueprints like the Intelligent Warehouse, and deploy serverless RTX‑accelerated inference on Cloud Run.
Baseten provides a managed inference platform that optimizes model serving for large language and vision models. They are accessing GPUs on Google Cloud to help them deploy NVIDIA‑accelerated inference with improved tokens‑per‑second, cost‑per‑1K requests, and GPU utilization for customers running production gen AI workloads.
CodeRabbit builds AI-driven workflows for code review, supported by Nemotron models via Google Enterprise Agent Platform, delivering low-latency PR summarization, embedding-powered context enrichment, and scalable concurrent review pipelines.
Factory AI develops autonomous code-editing agents that can traverse complex code graphs, resolve dependencies, and generate safe patches against real production repositories. Nemotron models in Gemini Enterprise Agent Platform allow Factory AI to power Nemotron Super-driven "droids" that perform semantic code search, tool-augmented reasoning such as grep and static analysis, and multi-file refactoring plans at scale.
Mantis AI is creating a video‑native intelligence layer for media and entertainment companies across Latin America. Through NVIDIA’s Inception program and its collaboration with Google Cloud, Mantis AI operates high‑throughput inference pipelines, rigorously benchmarks fine‑tuned model variants, and experiments with NVIDIA NIM, Blueprints, and vector search services to deliver production‑ready stacks that help broadcasters and streamers move from analyzing video to orchestrating it at scale.
Photoroom specializes in diffusion‑ and transformer‑based vision models that generate and edit product imagery. Photoroom processes over 1,000 images per minute. They use TensorRT for a 4x speedup and xformers to make their Stable Diffusion pipelines 100% faster. They’ve consumed over 500,000 GPU hours on Google Cloud without issue.

Sell agents directly in Gemini Enterprise

For AI startups, navigating enterprise procurement is often a bottleneck to growth. By integrating the Google Cloud Marketplace into Gemini Enterprise, we're providing a global commercialization engine to accelerate procurement.

Today at Next, we’re announcing an exciting group of more than 70 startups who are bringing their agents to our Marketplace.

And, to help even more startups build agents and bring them to market, we are launching a new $750 million dollar fund to support agent development and marketing for our partners. You can read more about this new funding here.

Presenting the Gemini Startup Forum

In June, we’re hosting our Global Gemini Startup Forum, a two-day, in-depth gathering of founders from seed to Series A startups. At the event, attendees will engage with Google leaders and engineers, build and scale products, and share feedback with teams across our AI portfolio. Take a deeper look at the program and register your interest for upcoming editions.

Building on that momentum, we’re also excited to share that applications are open for our next program: the Gemini Startup Forum: Cybersecurity, taking place in London later this year. If your startup is utilizing AI to build the next generation of security solutions, we invite you to apply to join us!

Join the Google for Startups AI Agents Challenge

Startups, it's your chance to get hands-on right away. We’re excited to launch the Google for Startups AI Agents Challenge at Next 2026.

Open to anyone (not just Next attendees), this global competition runs for six weeks and gives teams $500 in credits and the tools — like the new Gemini Enterprise Agent Platform — to build autonomous systems and compete for a share of a $90,000 prize pool.

The challenge, which opens for submissions today, and runs until June 5, 2026, offers three tracks tailored to where you are in your development journey. You can build a net-new agent from scratch, optimize an existing prototype for production reliability, or refactor a business-ready agent for potential enterprise distribution on Google Cloud Marketplace and the Gemini Enterprise app.

Submissions will be evaluated based on the following weighted criteria: Technical Implementation (30%), Business Case (30%), Innovation and Creativity (20%), and Demo and Presentation (20%). Learn more here and sign-up now.

Read the new Future of AI: Perspectives on generative media for Startups report

Our Future of AI: Perspectives on generative media for Startups report compiles exclusive interviews with eight startup founders, investors, and Googlers to capture their candid advice, strategic priorities, and predictions for the industry's future.

Understanding these firsthand founder perspectives will be instrumental in helping you navigate the next wave of AI-driven content creation and build the future of generative media on Google Cloud — and maybe find yourself featured on a list like this at a future Next.

Small and midsize businesses jumpstart their AI transformations with Gemini Enterprise

Wed, 22 Apr 2026 12:00:00 +0000

Small businesses are the backbone of the global economy. With 400 million SMBs worldwide and 36 million in the U.S. alone, they provide 50% of global employment. Now, with Google Cloud AI, they’re scaling faster, operating more efficiently, and delivering better results for their customers.

For years SMBs have been working with cutting edge products like Google Workspace, Google Ads, Search, YouTube, Maps, Google Wallet, and more to grow their business. Now, more SMBs are taking the first big steps in their AI journeys with our leading models and our agentic platform, Gemini Enterprise.

By making AI accessible to every employee, Gemini Enterprise serves as a launchpad for SMBs ready to develop an AI strategy and build new solutions to work smarter than ever before.Businesses are using AI to automate manual tasks and build custom solutions that extend their reach and help them better support their own customers.

We are proud to support these early milestones, and at Next we’re excited to showcase dozens of new customer partnerships with organizations taking their first steps toward an AI-driven future. Read on to explore their success stories and find our guide to the best SMB resources at Google Cloud Next ’26.

How Gemini Enterprise helps SMBs solve complex challenges

Across the globe, small and medium businesses are using AI to help grow their business. Here’s just a few of the many exciting examples of how SMBs are already benefiting from the secure, scalable, enterprise-grade tools in Gemini Enterprise:

Apex Leaders, a US-based business services company, is using Gemini Enterprise to power an internal search engine, providing easy access to internal data sources and automated information summarization and content drafting for consultant teams.
ARCHINERGY, a US based architecture firm, utilized Gemini Enterprise to create an AI agent that answers questions about architectural design and internal documentation, generating faster insights in just a fraction of the time compared to traditional search methods.
Constructora Las Galias, a Colombia-based construction firm, is using Gemini Enterprise to power a centralized internal knowledge platform, providing streamlined sales and legal workflows and immediate employee access to critical business information.
Eficode, which operates as a software development and DevOps firm within Finland, is using Gemini Enterprise to create a multi-platform support engine that quickly indexes complex internal wikis, so employees can provide customers with faster technical support
Grant Thornton Chile, a leading audit and consulting firm, is using Gemini Enterprise to enhance its agile management by powering secure financial data validation, accelerating HR and sales questions, and enabling unified search across disparate document types.
Grupo Ruiz, a leading Spanish company in sustainable mobility, uses Gemini Enterprise to improve team productivity and process efficiency, achieving an optimized workflow and greater data visibility
Huge, an American business services agency, utilized Gemini Enterprise to create AI agents that automate market research and contract analysis, generating new business intake in just minutes compared to several days.
Incrementa, a Chilean marketing company, is using Gemini Enterprise to automate client reporting and data analysis, helping it scale personalized service and boost campaign reach by 40%.
Koufu, one of Singapore's largest food and beverage companies, which operates and manages a large network of food courts and coffee shops, is using Gemini Enterprise to automate sales reporting and BigQuery data analysis. This reduces spreadsheet bottlenecks so shop managers can focus on front-of-house service quality.
La Maison du Whisky (whisky.fr), a specialized spirits and beverages retailer in France and internationally, chose Gemini Enterprise to create a 'Digital Sommelier' that transforms technical product data into high-end storytelling and marketing copy in seconds
SAI360, a US-based software company, is using Gemini Enterprise to power the design, production, and localization of ethics training, achieving 99% faster course creation and a 95% reduction in training media production costs.
Tauá Resorts group is using Gemini Enterprise to create an intelligent agent that generates real-time insights and dialogue suggestions. This technology allows the sales team to answer complex questions instantly, leveraging data from previous events to craft the perfect proposal for every client.
Tirol, a leading dairy producer in Brazil, is using Gemini Enterprise to create an interactive knowledge base, providing democratized data access and enhanced visibility across its entire value chain.
Wimberly Allison Tong & Goo leverages Gemini Enterprise to streamline digital planning, landscape, architecture and interior design workflows and visual queries, enabling the International Hospitality Design firm to accelerate the transitions from idea generation to design concepts and high-fidelity visualizations.

Practical Resources for Lean Teams at Next ‘26

At Google Cloud Next 2026, we are showcasing how SMBs can enhance their impact by integrating AI into their daily workflows. For attendees focused on maximizing impact with focused resources, we have curated a dedicated track to help you navigate the AI era.

When browsing the Session Library, use the Small IT Team filter to find deep dives into:

Cost Optimization: How to deploy AI effectively and manage resources on a budget that’s right for your goals.
Low-Code AI: Training on how to accelerate your business goals by building custom apps and agents.
Adaptive Security: Proactive defense strategies designed to provide enterprise-grade protection regardless of your security team's size.

Get Started Today

If you can’t make every session, we’ve made it easy to keep the momentum going:

The AI Starter Kit: A streamlined, four-step guide to deploying Cloud AI across your organization.
Monthly AI Newsletter: Sign up to receive the latest industry highlights and Cloud AI updates all year long.

Check out the full SMB guide here and start planning your itinerary. See you there!

Oracle AI Database@Google Cloud: The foundation for the Agentic Enterprise

Wed, 22 Apr 2026 12:00:00 +0000

For decades, many of the world’s most critical enterprise datasets have relied on the performance of Oracle databases. Today, we are bringing that reliability even closer to the cutting edge. By enabling customers to run Oracle AI Database services natively within Google Cloud, we’ve bridged the gap between foundational data and the modern AI stack.

With the latest wave of upcoming launches for Oracle AI Database@Google Cloud, we aren't just making it easier to migrate; we are building a direct pipeline from your Oracle systems of record to the insight layer of Google Cloud. By bringing mission-critical data easily and securely to Gemini models and BigQuery, customers can transform static records into autonomous, agentic workflows.

New capabilities announced at Next ‘26

Here is a breakdown of the key new features designed to strengthen your Oracle-to-agentic- AI strategy:

New regions launched: We have significantly expanded the availability of Oracle AI Database@Google Cloud, across 15 regions (and 20 sites) globally. The recent rollout included key global hubs such as Milan, Iowa, São Paulo, Tokyo, Sydney, and Mumbai, among others. With additional regions like Mexico and Turin coming soon, this expansion ensures higher availability and lower latency for your mission-critical workloads across the globe for our Google Cloud customers.
Enhanced AI capabilities: This is the foundation for agentic AI. We are introducing the preview of Managed MCP Server for Oracle workloads, which allows agents like Gemini to interact directly and seamlessly with your Oracle infrastructure. Building on this, the new Oracle AI Database Agent, available in the Google Cloud AI Agent Marketplace, lets you talk to your Oracle data directly from Gemini Enterprise — no custom chatbot or NL-to-SQL solution required.
Database Center integration (Generally Available): To move at the speed of AI, your infrastructure must be healthy and visible. Database Center now supports Oracle AI Database@Google Cloud, providing a "single pane of glass" for your entire data estate. Whether you are running Exadata or Autonomous Database, you can now monitor your inventory and streamline operations through a unified experience within the Google Cloud console.
Knowledge Catalog integration (Preview): Data discovery is the first step toward intelligence. By extending the Knowledge Catalog to Oracle AI Database@Google Cloud, we are breaking down the walls between your Oracle systems and the Google Data Cloud. This allows for a unified governance and metadata layer, making it easier for customers to find, trust, and use Oracle data and provide context to AI agents.
OCI GoldenGate Service integration (Preview): Real-time data is the lifeblood of AI. This integration enables low-impact, continuous data movement, allowing you to streamline migrations from on-premises environments to Oracle AI Database@Google Cloud. In addition, it provides a live link to BigQuery, enabling operational data analytics that reflect the "here and now" of your business.
VPC Service Controls: Oracle AI Database@Google Cloud administrators can use VPC Service Controls to restrict access to the admin API and create databases within a service perimeter. VPC Service Controls protect businesses from unauthorized access outside the security perimeter, even if credentials have been compromised.

The agentic future

The goal of these integrations is simple: To make your data active. When your Oracle data resides natively in Google Cloud, Gemini doesn't just “talk about” your data — it can work with it. Whether it's an AI agent forecasting supply chain shifts in BigQuery based on live Oracle ERP feeds, or a customer service bot with real-time access to legacy account history, the data vault is more open, accessible, and valuable than ever (while remaining just as secure).

Hear directly from our customer, Banco Actinver, regarding the transformative impact of relocating their Oracle data to Google Cloud.

Get started

You can access Oracle AI Database@Google Cloud through the Google Cloud Marketplace using your existing Google Cloud account and billing.

For more information, visit:

Building the Agentic Enterprise with Google Cloud partners and a $750M innovation fund

Wed, 22 Apr 2026 12:00:00 +0000

We are now seeing the Agentic Enterprise become reality for customers, and this week at Next ‘26 we are announcing exciting, new innovations to help customers accelerate agentic AI even further.

Our partners play a critical role in enabling the Agentic Enterprise, and today we are also announcing new resources, technologies, and deep technical partnerships to ensure we offer customers the industry’s most capable partner ecosystem for the agentic era, including:

A $750 million partner fund for agentic development applicable across global consulting firms, software partners, and our channel partners.
New ways for customers to deploy partner agents in Gemini Enterprise.
Deeper and more technical partnerships with global consulting firms to support customers, including with new teams of Google forward deployed engineers.
Integrating Gemini models more deeply into enterprise platforms from Palantir, Salesforce, SAP, ServiceNow, and more.
More AI-powered features in Google Cloud Partner Network to help our partners deliver high quality services.

Investing to accelerate AI agent development

We are committed to offering customers the most AI-capable partner ecosystem in the industry. To empower our partners to drive real transformation in the agentic AI era, we are launching a $750 million innovation fund to accelerate agent development and deployment globally, applicable to every business process, function, and industry.

This funding will support a wide range of activities including:

Hands-on support for software companies to build AI agents into their products with the Gemini Enterprise Agent Platform and bring them to market through our Agent Marketplace and through the new Agent Gallery in Gemini Enterprise.
Expert Google forward deployed engineers (FDEs) who will partner with major systems integrators to help their customers solve deep technical challenges and deploy Google AI more rapidly.
Deployment and usage incentives to help services partners thrive in the agentic era.
Training, technical development initiatives, and workshops to help partners build and deploy agents for customers using Gemini Enterprise Agent Platform.

Surfacing partner-built agents in Gemini Enterprise

At Next, we’re announcing Gemini Enterprise Agent Platform, a comprehensive platform to build, scale, govern, and optimize agents. It includes Agent Gallery, where customers can browse a highly vetted set of agents built by our partners.

Today, Agent Gallery provides access to a wide range of third-party agents. These agents have been built on top of our secure, enterprise-grade infrastructure, meaning customers can deploy them within their businesses with the highest levels of governance and confidence. Today, this includes agents built by Accenture, Adobe, Atlassian, Deloitte, Lovable, Oracle, Palo Alto Networks, Replit, S&P Global, Salesforce, ServiceNow, Workday, and more.

Empowering global consulting partners to drive AI transformations

Today, Google Cloud’s global consulting and systems integrator partners offer customers more than 330,000 experts trained on implementing Google AI. At Next, we are expanding our partnerships with every major systems integrator, including:

Accenture is helping enterprises drive AI-powered reinvention and business value faster and at scale with the launch of a first-of-its-kind Gemini Enterprise Acceleration Program. The program brings elite engineering and forward deployed engineers from Google Cloud and Accenture directly to customers.
BCG is expanding its partnership with Google Cloud to accelerate Gemini Enterprise transformation, helping organizations deliver at-scale agentic adoption.
Capgemini is establishing a Google Cloud AI Enterprise Hub to accelerate enterprise-scale adoption of Gemini Enterprise.
Cognizant is launching a dedicated Gemini Enterprise practice group to accelerate enterprise adoption of Gemini Enterprise.
Deloitte is forming a dedicated Google Cloud Agentic Transformation practice focused on Gemini Enterprise and will roll out Gemini Enterprise to more than 100,000 of its own teams.
HCLTech is launching a Gemini Enterprise Business Unit to accelerate the development and adoption of industry-specific solutions built on Gemini Enterprise.
Infosys is leveraging Gemini Enterprise within its Infosys Topaz platform and is equipping more than 100,000 Infosys developers across Infosys’ global delivery teams with Gemini Enterprise.
KPMG is deploying Gemini Enterprise with a life sciences company and launching a new Financial Close Companion agent built with Workday and Google Cloud.
Kyndryl is deepening its Google Cloud partnership with expanded Google Distributed Cloud services for sovereign, AI-ready applications.
McKinsey is launching the McKinsey Google Transformation Group to accelerate enterprise AI outcomes with Gemini Enterprise, combining its strategic expertise with Google's AI stack to help organizations scale agentic transformation
PwC is launching a dedicated Google Cloud AI Center of Excellence to help organizations scale AI adoption, pairing industry expertise with Gemini Enterprise to deploy AI agents that reason, act, and automate processes at scale.
TCS is launching new agentic AI offerings and a dedicated Gemini Enterprise practice, featuring more than 3,000 industry-focused AI agents and an expanded global network of Gemini Experience Centres to accelerate AI-native, autonomous enterprise operations.

Additionally, we will begin to embed teams of Google Cloud engineers with a subset of global partners, including Accenture, Capgemini, Cognizant, Deloitte, , HCLTech, PwC, and TCS in order to help their customers more rapidly prototype and deploy AI agents within their businesses. AI-native services partners, including Altimetrik, Artefact, Covasant, deepsense.ai, Distyl.ai, Northslope, Quantium, Tribe.ai, and Tryolabs will launch Gemini Enterprise practices, receiving credits for sandbox development, technical upskilling, and referral opportunities.

We are also rolling out a new program offering early model access for a select group of partners, including Accenture, Bain & Company, BCG, Deloitte, and McKinsey, who will be able to preview and begin building with pre-release versions of upcoming Google DeepMind models.

Bringing Gemini to More Customers through Popular SaaS Platforms

Already, many of the world’s leading agentic SaaS and AI platform companies integrate Gemini into their products. At Next, we’re expanding these integrations even further, including:

Atlassian is bringing Gemini 3 Flash to Rovo and integrating multimodal capabilities into Remix in Confluence, helping teams instantly transform text-based documentation into high-fidelity diagrams and charts for faster stakeholder decision-making.
Box is launching new Box Agents powered by Gemini 3 Flash and Gemini Enterprise, helping enterprises transform static files into actionable intelligence by natively integrating AI orchestration into their secure content management workflows.
DocuSign is using Gemini to power new features that summarize complex agreements, identify key clauses, and help users understand the implications of their contracts.
Oracle is launching the Oracle AI Database Agent for Gemini Enterprise. This new agent enables end users to ask business questions of their Oracle data in natural language in Gemini Enterprise, without needing to write SQL or understand the underlying data model.
Palantir is adding Gemini and BigQuery integrations for commercial customers, enabling customers to connect best-in-class models to their most critical AI workflows and operations.
Salesforce is adding native Gemini support to its Atlas Reasoning Engine. This enables Agentforce to “see” across text, image, and video formats, drawing from years of customer history to accurately solve complex problems. This means faster, smarter resolutions, building upon the success thousands of customers are already seeing from Gemini within Agentforce to build prompts.
SAP is integrating Gemini Enterprise into its Engagement Cloud to deliver AI-powered customer service and sales insights alongside creative tools for image and text generation.
ServiceNow is integrating its AI agents with Gemini Enterprise, bringing autonomous operations to the world’s largest enterprises

Building a partner channel for the agentic era

Our new partner program, Google Cloud Partner Network, is designed to help partners thrive in the agentic era. Last year, we used AI to unlock deep insights across our partner tools; now, we are building the agentic workflows that turn those insights into autonomous growth. Key updates include:

The Partner Agent: Integrated into the Partner Network Hub, this agentic tool acts as a central orchestrator for the partner experience. Beyond answering questions, it actively guides partners on next steps, summarizes complex assets, and provides real-time coaching for registrations and statements of work.
The Agentic Earnings Hub: Here, partners can find new capabilities to auto-draft statements of work and monitor consumption milestones to auto-generate claim requests. When paired with the Earnings Potential Modeler, these tools provide contextual recommendations to map every available incentive down to the individual client level.
Partner Finder: We are also extending this intelligence to customers, turning discovery into a conversational experience where natural language prompts pinpoint the ideal partners for the most hyper-specific workloads.

Finally, we’re honored to highlight the winners of Google Cloud’s 2026 Partner Awards, which celebrate the transformative impact and incredible value our partners have delivered for customers over the past year. Our ecosystem continues to evolve to meet the needs of businesses across every industry, and we are constantly impressed by their ability to solve complex, global challenges using our technology. To discover more about these exceptional achievements, please read our full list of partner award winners.

I can’t wait to meet with thousands of you this week to build the future of the Agentic Enterprise!

How WPP accelerates humanoid robot training 10x with G4 VMs

Thu, 16 Apr 2026 16:00:00 +0000

Editor’s note: Today we hear from Perry Nightingale, SVP of Creative AI at WPP about the workflow that cuts training time for humanoid robots from days to minutes — plus access to the open-source code to do it yourself.

Robots are pushing the boundaries of what content creators and directors can capture. These technologies have become critical in the film industry because they open up new possibilities for controlled camera moves in locations where traditional methods would be unsafe or infeasible.

We’ve found that the programming for these robots is arguably as technical and complicated as the shoots they’re being tested on. To achieve our goals, we needed a hardware stack that was equally advanced as the robots we’d be programming. In this post, we explore how WPP used the new G4 VM instance powered by NVIDIA RTX PRO 6000 Blackwell on Google Cloud, which is a great fit for the unique challenges of training physical AI.

While robotics is always a complex task, especially in the environments we were working in, thanks to the unique software stack we were using, we were able to reduce training cycles from 24 hours down to less than one.

Below, we detail the specific reinforcement learning (RL) workflow, the challenges of the "sim-to-real" gap, and the infrastructure that makes it possible.

The lessons learned here apply far beyond entertainment. Our process for mastering complex, natural movement on a film set can be replicated across industries to overcome the massive computational complexity of training robots.

Where we started: redefining the agency model

WPP is one of the world’s largest marketing organizations, handling $70 billion of media for enterprise clients. For us, building AI into our production workflows meant fundamentally redefining the agency model, both in terms of processes, relationships, and tools. Notably, we launched WPP Open last year, our proprietary AI operating platform, where we’re able to take the best of Gemini’s multimodal intelligence, along with other models, and incorporate them directly into every creative step.

The results were immediate. For one of our clients, Verizon, we built an AI-infused promo pipeline that delivered 15 videos in 70% less time, with 50% to 70% efficiency gains across the production cycle. WPP Open has proven so effective for our teams, we’ve begun offering it to our clients, so they can tackle projects in new ways and we can collaborate faster and better.

WPP Open has also challenged and inspired us to look for more ambitious applications of AI. With the latest advancements in Google Cloud’s AI Infrastructure, we saw an opportunity to tackle more complex problems at the cutting edge of creative development.

Why teach a robot to dance?

Our robotics work started with teaching a machine to dance — and not just because we knew a dancing robot would make for a compelling demo. Dance, along with martial arts, is generally accepted as the cutting edge of complex human motion. Mastering these complex movements is a critical step toward achieving natural robotic motion.

For our benchmarking project, we trained our robot to perform a dance sequence captured in a previous project with Universal Music Group.

The workflow

To achieve this complex motion, we needed a workflow that could iterate fast.

We captured human motion with the OptiTrack mocap system and retargeted it to an official OpenUSD digital twin of the robot. This is a complex engineering challenge: a human has over 200 degrees of freedom compared to just 29 on a robot. So our team needed to remap the human skeletal data to the far more constrained physical structure of our robot, creating a sophisticated 3D model.

MuJoCo, an open source physics engine by Google DeepMind, was a critical piece of simulation software that validated our accuracy continuously, in real-time.

The workflow then moves to reinforcement learning. In our previous environment using single on-premises GPUs, training took ten hours. This time, we used G4 VMs, together with the NVIDIA Isaac Sim image on Google Cloud Marketplace. Google Cloud’s unique innovations, such as… with G4 enabled us to utilize a P2P topology that moves data directly between GPUs without the bottleneck of central processing.

Using Google Cloud’s AI Hypercomputer, we saw speed increases in excess of 10x, taking training times down to less than one hour.

Over the training run, the digital twins of the robots are rewarded for getting closer to the intended motion sequence while under the simulated effects of physical gravity, momentum, friction, and small simulated “pushes” that might be expected forces in the real world.

As the simulation training begins, the robots fall almost instantly (top). After about 3 billion simulations, the robots have learned the complex dance sequence (bottom).

Bridging the "sim-to-real" gap

The "sim-to-real" gap is one of the toughest challenges in robotics. A policy that works perfectly in the above workflow often fails in the physical world due to unmodeled physics or sensor noise. A foot might land differently due to small changes in carpet friction or a gap in the floor.

We ran billions of simulations to develop the reinforcement learning model, which was then condensed into an ONNX policy and deployed to the robots. These policies take real-time observations from the robot — IMU data, joint positions, etc. — and return the necessary movements.

Through high-volume simulation, the humanoid learned how to respond to these small changes and determine what move to make next to keep the sequence on track. Again, we used MuJoCo for critical real-time validation, ensuring that the robot's ability to adapt in simulation translated to safety and stability in the real world.

What’s Next

To accompany this project, Unitree has released their in-house reinforcement learning code as a sample project on GitHub. Alongside the NVIDIA Isaac Sim image on Google Cloud Marketplace, this means almost instant access to advanced robotic motion research.

Check it out for yourself:

Explore: Unitree Robotics RL on GitHub
Deploy: NVIDIA Isaac Sim on Google Cloud Marketplace
Learn more: G4 VM Instance Documentation

Building the agentic future: A spotlight on Google Cloud’s media & entertainment partner ecosystem

Thu, 16 Apr 2026 13:15:00 +0000

As we gather in Las Vegas for NAB Show 2026, the industry conversation has shifted. We are no longer asking if AI works; we’re now focused on how it scales. The era of AI experimentation is over — production-grade execution is here.

At Google Cloud, we believe no studio or broadcaster should have to build this future in isolation. Our mission is to provide the agentic platform and AI and cloud tools that allow our partners to innovate at the speed of ideas — from the tools used in the edit suite, to the technology that delivers video to millions of viewers worldwide.

Enhancing production: From manual tasks to intelligent assistants

Modern creative workflows are often slowed down by manual technical tasks. Google Cloud is working with ecosystem leaders to integrate advanced AI capabilities directly into the core of production software, so creators can focus on their artistry, not tedious tasks. With AI acting as a proactive assistant within the creative suite, production teams can significantly reduce the time between a raw idea and a finished frame.

Avid: With the launch of Content Core on Google Cloud, Avid is delivering a truly cloud-native studio. And by integrating multimodal AI search into Media Composer, editors can find the exact frame they need using natural language, turning hours of logging into seconds of discovery.
Backlight: Backlight makes complex media workflows simple for teams of all sizes, from production through monetization. Built on Google Cloud with the Video Intelligence API, Backlight's Iconik platform automatically adds searchable metadata upon upload. Customers see up to 50% shorter production cycles and save up to 60% on storage by deeply understanding their media libraries, reducing duplications, and optimizing asset placement.
Brahma.ai: Brahma AI, an enterprise AI content platform, is powering high-fidelity digital likenesses across retail, entertainment, and healthcare, making them interactive and intelligence-driven within a secure and governed framework.

Unlocking content value: From static archives to active assets

Data is only as valuable as the insights you can extract from it. Our partners, listed on the Google Cloud Marketplace, are using generative media models to transform massive, static archives into searchable, revenue-generating engines. By making every frame discoverable, we’re helping media companies turn decades of history into immediate opportunities.

Ateme: Ateme helps simplify global distribution with its new generative AI-powered subtitling solution, which can significantly reduce the manual labor of localizing different media types.
Perfect Memory: Perfect Memory helps customers turn traditional storage into a context-aware knowledge engine. The platform understands the relationships between athletes, historical events, and emotional nuances — transforming massive media archives into an intelligent library that lets creative teams instantly surface the perfect content for any story.
VionLabs: Working with companies like Cineverse, Plex, and Crunchyroll, Vionlabs uses AI to analyze and index content libraries — making video assets more accessible and enabling metadata generation. By understanding the specific mood and aesthetic of each scene, Vionlabs helps streaming platforms move beyond basic genre tags toward more nuanced, sentiment-driven content discovery and marketing.

Scaling global reach: From simple streams to audience growth

To succeed today, media companies must provide a smooth viewing experience and easy payment options. Our ecosystem provides the tools to grow a company’s reach and maximize the value of every subscriber through reliable, high-quality delivery.

Bending Spoons: By leveraging the global scale of Google Cloud, Bending Spoons’ properties such as Brightcove and Vimeo are delivering professional-grade tools for large enterprises, SMBs, the next generation of creators, and more. Its platforms ensure that high-quality video production and distribution are accessible to everyone, from global brands to independent storytellers.
Bitmovin: Bitmovin enables streaming services to scale efficiently while delivering a premium experience across the widest range of devices. By combining real-time observability with AI-driven insights, media teams can proactively optimize engagement and monetization. Furthermore, Bitmovin’s advanced encoding ensures superior visual quality at lower bitrates, supporting everything from high-demand Video on Demand (VOD) to massive, 24/7 live events.
Evergent: Evergent automates complex billing and monetization workflows for AI-powered revenue management. Media and telecommunications companies can use Evergent’s tools to maximize subscription growth and improve long-term customer retention through personalized and agile payment offers.
Harmonic: Harmonic is helping major broadcasters like Grupo Globo modernize their operations. By integrating new digital broadcast capabilities into their cloud-based streaming solutions, Harmonic provides leaders with a faster, more efficient path to manage video processing and delivery at a massive scale.

Ensuring reliability: From infrastructure to a foundation of trust

High-quality content requires a high-performance foundation. We are partnering with infrastructure leaders to ensure that even the most complex global broadcasts remain stable, secure, and responsive.

Zixi provides the broadcast-grade transport and workflow automation needed to move professional video across any network. By offering centralized control and complete visibility into the delivery process, Zixi ensures that leaders like Fubo can manage high-stakes, broadcast-quality live events without the risk of a signal drop.

Visit the ecosystem in action

The strength of our ecosystem is its integration across all aspects of the media and entertainment landscape. From the cameras, to the cloud, to the viewers' screens, these partners represent the future of a more creative, efficient, and agentic media industry.

Visit the Google Cloud Booth (West Hall, #W2731) at NAB Show from April 19-22 to see many of these partners in action through live demonstrations and theater sessions.

aside_block: <ListValue: [StructValue([('title', '2026 AI Agent Trends in Media and Entertainment'), ('body', <wagtail.rich_text.RichText object at 0x7f8d011650d0>), ('btn_text', 'Read it now.'), ('href', 'https://cloud.google.com/resources/content/ai-agent-trends-media-entertainment-2026'), ('image', <GAEImage: Confirmation email_500x450>)])]>

Cool stuff Google Cloud customers built, April edition: BMW big on SLMs, MLB’s Scout Insights AI, personalized resort experiences

Wed, 15 Apr 2026 16:00:00 +0000

AI and cloud technology are reshaping every corner of every industry around the world. Without our customers, who are building the future on our platform, there would be no Google Cloud. In this regular round-up, we dive into some of the exciting projects redefining businesses, shaping industries, and creating new categories.

For our latest edition, we learn why BMW Group is experimenting with small language models (SLMs); catch AI-powered commentary from Major League Baseball; hit the slopes with Vail Resort’s AI concierge; build an intelligent grid with CTC Global; witness how ID.me created secure global scale; and see how Manhattan Associates supply chain tools now handle 1 billion daily API calls.

Be sure to check back next month to see how more industry leaders and exciting startups are putting Google Cloud technologies to use. And if you haven’t already, please peruse our list of 1,001 real-world gen AI use cases from our customers.

BMW tests the big potential of small models

Who: As one of the world’s leading providers of premium cars and motorcycles, BMW Group is always at the forefront of automotive technology. This ethos pushed the company to test what type of AI language models are ideally suited to driving situations, where access to cloud-based LLMs isn’t always possible.

What they did: BMW Group wanted to explore the potential of small language models (SLMs), which could run within the limited hardware on a vehicle. Finding the right trade-off between size and capability requires careful optimization, and the sheer volume of viable combinations renders manual searches for the optimal configuration an incredibly tedious, if not impossible, undertaking. To overcome this challenge, BMW and Google built automated, reproducible workflows through executable pipelines using Vertex AI.

Why it matters: The path from a general-purpose LLM to a specialized SLM isn’t straightforward. Every choice — from type of quantization to characteristics and contents of the fine-tuning domain-specific dataset — affects the quality and efficiency of the final model. This creates an exponential range of configurations, each with different trade-offs. It’s a great example of using AI to scale an optimization problem for other AI.

Learn from us: “With automated pipelines, we can rapidly adapt models to our domain and rigorously test and evaluate them against domain-specific benchmarks. This allows us to iterate and optimize models in hours rather than days.” – Dr. Céline Laurent-Winter, vice president, Connected Vehicle Platforms at BMW Group

MLB Scout Insights: AI-powered color commentary

Who: Major League Baseball is famous for its colorful announcers. Now, MLB is bringing more baseball color straight to your pocket, and Gemini is helping give it a voice.

What they did: Each season, millions of baseball fans use the MLB app and tap over to the Gameday feature for live, up-to-the-pitch action across more than a dozen games. Starting this season, the league launched MLB Scout Insights in Gameday, which uses Gemini models to quickly scan decades of game and player data, cross-references it with situational game scenarios, and then delivers game-relevant context during key matchups.

Why it matters: Given the sport’s storied history, 162-game regular season, and global reach, baseball fans are among the most sophisticated and passionate out there. To keep them engaged with Gameday and the MLB app, the league wanted to deliver insights that truly felt meaningful and interesting. Building the tool meant answering a rather squishy question: What makes an insight actually insightful, not just an accurate fact, and how can an AI learn that distinction? The answer came from some clever “surprisal” analysis.

Learn from us: "With Scout Insights, every fan can feel like the smartest person in the stands, at the water cooler, or on the couch. It’s about deepening connections to the game, and sharing that passion with others. That’s the magic of sports, and we’re making more of it possible with the magic of AI." – Josh Frost, senior vice president of product & Matt Graser, director of engineering, Major League Baseball

Vail Resorts makes personalized AI assistance easy

Who: Vail Resorts operates some of the most iconic and beloved mountain destinations in the world, including Whistler Blackcomb, Park City Mountain, Stowe, and Crested Butte.

What they did: Vail Resorts launched My Epic Assistant during the 2024-2025 snow season, and expanded it this year to add even more AI-powered chat features powered by Google’s powerful Gemini models. The result is an agentic guide to the slopes that can help skiers and snowboarders decide on the right season pass, share the latest snow report, check on lesson preparations, or suggest a good stop for cocoa.

Why it matters: Vail Resorts wanted more than a chatbot; they wanted a digital concierge that understands the nuance between a powder day at Whistler and a family trip to Beaver Creek. As the company implemented and refined personalization, improved search, summary capabilities, and conversational flow within My Epic Assistant, the app has delivered a 45% reduction in escalation to human agents since launch.

Learn from us: "Utilizing tooling from Google Cloud, we could lean into agentic design patterns that gave us a way to unlock natural, personalized conversations. These boosted customer satisfaction, while reducing the need for manual intent design. These tools also let us combine flexibility and control to enable the assistant to respond fluidly but always within the boundaries of our brand, policies, and product strategy.”— The Vail Resorts technical team

CTC Global turns the smart grid into an intelligent one

Who: CTC Global is a leading manufacturer of advanced transmission conductors and power lines. While many nodes in the grid contain IoT sensors, it recognized a literal gap in the transmission lines themselves.

What they did: CTC’s new GridVista platform threads fiber-optic cable into its high-strength carbon fiber composite core, and connects these to monitoring technology built with AI and monitoring technology from Google Cloud and Tapestry. With GridVista, CTC can turn every inch of transmission into a smart sensor.

Why it matters: GridVista gives CTC grid operators an accurate and reliable view of what’s happening across the entire line — based on actual, real-time data from the entire length of the conductor, not point estimates from a static model of line conditions or the occasional clamped-on sensor. This means they can significantly improve safety, manage costs, increase the line’s capacity to transmit power, and enhance reliability with more precise insights about events that might trigger an outage.

Learn from us: “This awareness allows for a grid that can truly sense its own health in real time and provide unprecedented awareness of conditions on the entire line. Whether that’s real time storm impacts, ice load, wind load, branches on the wire, or temperatures on or under the line. The GridVista system truly represents next generation capabilities. ” — J.D. Sitton, CEO, CTC Global

ID.me reduces risk while scaling past 160 million users

Who: ID.me is transforming digital identity security for the modern era, offering a single login that lets you easily prove you’re you across a wide range of platforms and wallets.

What they did: ID.me currently serves more than 160 million users, including as many as 40,000 at any time, so they can prove their identity online as easily as flashing their driver’s license in person. Over the last two years, ID.me migrated more than 50 terabytes of data across 15 database instances to Google Cloud with minimal downtime. They also introduced a two-tier architecture with Cloud SQL supporting its smaller and more standard services, while AlloyDB runs heavier workflows that form the backbone of the ID.me platform.

Why it matters: AlloyDB AI has allowed ID.me to scale its systems to handle 10X-20X of what was possible before — and at a lower price to boot. That responsiveness and reliability led the U.S. federal government to recognize ID.me for its role in preventing large-scale fraud within national systems.

Learn from us: "We’ve been able to scale both our infrastructure and trust. With a platform that’s faster, smarter, and built to handle portable identity at massive scale, we’re one step closer to our goal: a secure, digital way to prove who you are, wherever you need it, that works everywhere you need it." — Kevin Liu, Cloud Platform Architect, ID.me

Manhattan Associates powers more than a billion daily API calls

Who: Manhattan Associates is a global leader in supply chain and omnichannel commerce solutions, offering tools and platforms that reach more than 2 billion people across 20 billion consumer touchpoints.

What they did: Manhattan Associates modernized its Manhattan Active SaaS platform by migrating from legacy Oracle and DB2 systems to Google Cloud databases. Each capability of Manhattan Active now runs as an independent, containerized service orchestrated by Google Kubernetes Engine (GKE). Data flows through Pub/Sub into BigQuery for real-time analytics, while Cloud Logging and Cloud Monitoring deliver observability at scale.

Why it matters: With its new microservices-first design, Manhattan gained the agility to evolve faster and the confidence that mission-critical operations would remain resilient across regions. With Cloud SQL and BigQuery, the company now processes more than a billion daily API calls with average response times of less than 150 milliseconds. This evolution supports hundreds of thousands of monthly active users across tens of thousands of stores and distribution centers. The new platform also created the foundation for Manhattan’s Agentic AI suite, which includes prebuilt agents — like the Intelligent Store Manager and Labor Optimizer — that coordinate real-time decisions across store and distribution center operations.

Learn from us: "Operationally, the platform has become more elastic and efficient. The system automatically handles hundreds of thousands of scaling events per day, ensuring performance remains consistent during peak surges without expensive overprovisioning." — Narayana Reddy Kothapu, Senior Director, Manhattan Associates & Rajkumar Ramani, Technical Director, Manhattan Associates

How SAP Concur automates expense reporting with agentic AI

Fri, 10 Apr 2026 16:00:00 +0000

For decades, expense automation relied on a simple premise: If the machine can read the text, it can do the work. But anyone who has ever tried to scan a crumpled, smudged, or sun-bleached receipt from their pocket knows that reading isn't enough. When key data is missing, such as a city name or a clear date, the machine halts and the burden falls back onto the user for manual entry.

To close this gap, where traditional Optical Character Recognition (OCR) fails, SAP Concur’s engineering team set out to break new ground. While much of the industry was still focused on the design of conversational interfaces, SAP Concur foresaw a bigger shift. They recognized early on that the next leap in efficiency wouldn't come from better scanning, but from intelligent reasoning.

The result is an agentic AI upgrade for ExpenseIt, moving automation beyond simply reading text to solving messy logic puzzles, significantly reducing the need for manual intervention. Now, travelers can simply snap photos of their receipts as they receive them, upload digital scans, or forward receipts as emails, and ExpenseIt instantly transforms them into accurate expense entries with no date entry or itemization required.

Bringing this next-generation system called for a partner who could push the boundaries of innovation while matching the ambition to execute at startup speeds. SAP Concur fused its visionary roadmap with Google Cloud’s full-stack AI power, partnering with the only provider that co-designs every layer, from custom silicon and data platforms to world-class models and agents. Together, the teams engineered a true breakthrough in cost management — an AI agent that not only captures the receipt but intuitively understands the business traveler’s reality.

Speed, scale, and ingenuity

Standard expense automation is great at seeing what is on receipts but can’t see what is not there. SAP Concur saw the emergence of AI agents as an opportunity to create systems that could reason, decide, and act.

Suppose you upload a lunch receipt from “The Main St. Café,” which doesn’t include the address. In the past, this missing information would completely derail the automation and require you to manually enter this data to continue.

Agentic capabilities enable analyzing contextual clues, such as a vendor’s name, expense types, and trip itinerary data, to fill in the gaps. SAP Concur wanted to create an AI agent that could think like a human assistant: "I see 'Main St. Café.' I also see this transaction coincides with a business trip, where the user has a flight to Dallas and a hotel in Greenville, Texas. Therefore, this vendor is probably the restaurant located near the hotel in Paris, Texas — not Paris, France."

To solve this challenge, the teams approached the problem with a dynamic, startup-style mindset. Instead of a lengthy development cycle, the collaboration was defined by rapid prototyping and bold problem-solving.

Utilizing Google’s Gemini models, they built the Receipt Analysis Agent, underpinned by a cognitive architecture.

Here’s how it works:

Ingestion: The user snaps a photo in the SAP Concur mobile app, uploads a digital scan, or forwards a digital receipt as an email.
Deterministic core: SAP’s foundational technology, refined over decades of processing global expenses, applies finely tuned logic to lift the visible text on receipts with high precision.
Intelligent rRouting layer: If the scanned receipt data is clear, there’s no need to trigger additional actions. If the data is ambiguous (e.g., "Missing location"), the routing logic dynamically directs the task to the Receipt Analysis Agent.
Contextual reasoning: Built with Gemini models, the AI agent doesn’t just guess — it uses tools and grounding to infer missing information. ExpenseIt feeds the partial receipt data to the agent, alongside grounding data like the user’s travel itinerary and business calendar.
ReAct (Reason and Act framework): The Receipt Analysis Agent connects the dots, validating the vendor against the location history, and then completes the expense entry.

ExpenseIt with agentic AI (Receipt Analysis Agent)

Based on the example above, ExpenseIt identifies the receipt image as missing the location, and the intelligent routing layer triggers the Receipt Analysis Agent. Using Gemini, the agent will then identify what’s missing, analyze surrounding contextual clues and user-specific data, and make decisions based on information like travel bookings and calendar events.

Key design patterns for successful AI agents

The Receipt Analysis Agent was designed based on the core principles from Agentic Design Patterns, a hands-on guide written by senior Google engineer Antonio Gulli. This critical guidance helped SAP Concur successfully transform ExpenseIt into a system that can reason on data both inside and outside of receipts to accurately create expense entries.

First, the teams implemented the Routing Pattern to avoid running every receipt through the AI agent, helping to optimize for both cost and intelligence. A routing architecture classifies incoming tasks: Receipts with a high OCR confidence score are routed to the standard deterministic path, while those with low scores (e.g., “Missing location) are dynamically routed to the Receipt Analysis Agent.

Next, the Reflection Pattern is applied to solve issues like the Paris Paradox, ensuring the agent doesn’t just generate an answer like a basic chatbot. This pattern involves an internal generator-critic loop, where the model generates a hypothesis (“I think this is Paris, France”) and then acts a critic, checking it against established facts (“The itinerary says Dallas, Texas. This hypothesis is likely false.”).

Finally, the agent follows the Tool Use Pattern, providing explicit API access to grounding sources like trip itineraries from Concur Travel. This approach allows the agent to fetch the truth rather than hallucinating it, turning the system from a text generator to a factual researcher.

Architecting for ambiguity: Google Cloud’s ecosystem advantage

This project highlights a pivotal shift in intelligent system design. By combining a deterministic core with an agentic reasoning layer, SAP Concur demonstrated that AI’s highest value often isn't in processing the data we have, but in reasoning to find the data we are missing. A defining moment in this engineering journey was the shift in how the model was utilized. The teams moved beyond treating Gemini as a generative interface and instead deployed it as a logic engine.

Why did SAP Concur choose to build this future with Google Cloud? Because an agent is only as good as its understanding of the world — and no one understands the digital world like Google.

While this current release relies on the reasoning power of Gemini, the partnership opens the door to a future of multimodal, full-stack intelligence that’s unique in the market, including:

Real-world grounding: Imagine an agent that cross-references a receipt with Google Maps data to ensure the business actually exists at that location.
Frictionless flow: Future integrations could use Google Wallet to match transaction timestamps instantly, or Gmail to surface hotel folio receipts automatically.
Edge intelligence: With mobile advancements like Gemini Nano and the service system Android AICore, sensitive processing could eventually happen right on devices, giving users speed and privacy without the data ever leaving their phone.

SAP Concur has the deep domain expertise that powers the world’s financial transactions. Google Cloud brings the full AI stack from the custom-designed chips (TPUs) optimized for training, to the mobile OS in the user’s pocket.

Ready to build your next-generation agent?

You don't need to reinvent the wheel to build a reasoning engine like ExpenseIt. The architectural patterns discussed here — Routing, Reflection, and Tool Use — are codified directly in the Google Agent Development Kit (ADK). The ADK provides the frameworks and best practices to help you move from "prompt engineering" to "system engineering," serving as a blueprint for building agents that are reliable, scalable, and ready for the enterprise.

Behind the Analysis with Google Cloud and Team USA: Architecting AI infrastructure for U.S. Winter Olympians

Fri, 10 Apr 2026 16:00:00 +0000

In freeskiing and snowboarding, traditional video replay shows you what happened during a complex aerial maneuver, but it fails to explain the physics of how it was possible. At the speed of the sport, it's incredibly difficult to translate high-speed motion into actionable data—joint angles, rotational velocities, body compression. This requires tracking and analyzing a full three-dimensional model of the athlete, frame by frame, in real-time.

In collaboration with Google DeepMind, we built a system to provide this analysis to U.S. Olympians ahead of the Olympic Winter Games. Our AI pose estimation model transforms a single 2D video into a complete 3D biomechanical analysis, plotting 63 joints in a localized coordinate system. For athletes and coaches, it provides a revolutionary competitive edge. For broader use cases, it turns human movement into objective data.

The challenge: extreme conditions break standard vision

Generating a 63-joint 3D skeleton from 2D video is a massive computational workload. Generating it without lab-grade sensors and in unpredictable outdoor environments, pushes computer vision to its limits. Snowboarders and skiers move at extreme velocities. They wear bulky gear. When they tuck for a grab or spin, limbs disappear from view. Standard pose estimation models lose tracking the moment this occlusion occurs.

Our solution relies on a proprietary model of human motion. Instead of treating each frame in isolation, it uses learned priors to infer the position of hidden joints based on the body's overall trajectory. This temporal reasoning maintains a stable digital skeleton even through rapid, inverted rotations.

The infrastructure: TPUs and Vertex AI

Solving occlusion is only half the battle. Delivering these insights quickly—seconds after a U.S. Olympian lands —requires heavy-duty infrastructure. We built a high-performance inference engine on Google Cloud to handle the intense MLOps demands of the competition.

The hardware foundation: TPUs

At the core of the pipeline are Google’s Tensor Processing Units (TPUs), tasked with the heaviest matrix math. An encoder first compresses the video into a latent representation, and a video transformer model predicts the 3D joint positions.

To eliminate the standard cloud "cold start" delay, we statically provisioned dedicated TPU slices for the duration of Team USA's competition at the Olympic Winter Games. This kept the models perpetually loaded in High-Bandwidth Memory (HBM). When a video arrives, it hits a "warm" TPU, guaranteeing near-instantaneous, predictable inference without the resource contention of a multi-tenant environment.

Orchestration at scale: Vertex AI

Deploying to a single lab server is easy; orchestrating live action at the Olympic Games is not. Vertex AI provided the unified control plane to manage volume, complexity, and latency:

Horizontal scaling with batch prediction: Using the Vertex AI Batch Prediction API, incoming video is instantly directed to a distributed network of workers. This decouples model loading from inference, allowing the system to scale horizontally and process multiple athletes simultaneously without choking.
Volume and elasticity: Video analysis of U.S. Olympians is what we describe as ‘bursty’ - computational needs spike for the short duration of the athlete runs. . Vertex AI dynamically provisions resources to absorb these data spikes, rather than keeping resources always-on.
Security and exclusivity: To protect proprietary Team USA data, we established a Private Endpoint within a Virtual Private Cloud (VPC). Authorized traffic travels via dedicated network pathways, isolating the engine from the public internet to reduce the attack surface and minimize latency.

Beyond the snow

A system capable of reliable pose estimation under extreme winter conditions—high speeds, constant occlusion, and a requirement for speed—is a system that generalizes. We believe the underlying AI architecture, and the ability to provide generalized intelligence from structured data feeds can enable a number of use cases beyond winter athletics.

Imagine a conversational AI physical therapy coach that analyzes and helps with movement form. Or, robot assistance for a factory worker that is triggered by cues noticed in their posture. These are all potential use cases where specialized sensor AI, paired with powerful reasoning models, can provide helpful insights and actions.

aside_block: <ListValue: [StructValue([('title', '2026 AI Agent Trends in Media and Entertainment'), ('body', <wagtail.rich_text.RichText object at 0x7f8cfdfaedf0>), ('btn_text', 'Read it now.'), ('href', 'https://cloud.google.com/resources/content/ai-agent-trends-media-entertainment-2026'), ('image', <GAEImage: Confirmation email_500x450>)])]>

Under one roof: Rightmove reinvents property search with unified data

Tue, 07 Apr 2026 16:00:00 +0000

At Rightmove, we want to make home moving easier for everyone, from house hunters and homeowners to estate agents and brokers. Behind every search, listing, and connection on our platform lies a complex network of users, partners, and properties — and we’ve built our data and AI strategy to serve all three.

To deliver on this mission, we migrated from siloed, on-premises databases to Google Cloud. This move wasn’t just about technology. It was about unlocking smarter, faster, more personalized experiences for our users and partners, and helping them find the right match for each property more efficiently.

Our strategy is guided by four core data and AI value areas:

Delighting consumers with personalized search and discovery
Empowering partners, such as estate agents, with smarter tools and insights
Monetizing data through innovations such as property price prediction
Driving operational efficiency across our platform

Today, we're building this future with a unified analytics and AI stack — BigQuery, Vertex AI, and Looker — that we call “the data hive.” Already, around 300 team members (a third of our workforce) are tapping into its capabilities to turn data into action and insights into impact.

Making it easier to find a home with personalized, dynamic suggestions

When someone’s looking for a new home, they often have a wish list: a garden, a modern kitchen, maybe a home office. We’re using Vertex AI to make that search feel more intuitive and tailored than ever. By extracting metadata from property descriptions and images, we automatically create listing features and keywords, even ones that weren’t manually tagged before, to provide more accurate search results.

We’re also exploring ways to streamline communication. Recently released is an AI-powered feature that uses Gemini to help estate agents respond to inquiries faster. With context-aware, automatically generated replies, agents can keep conversations moving, and potential buyers and sellers can get answers faster, even during the busiest periods.

Helping partners work smarter with AI-powered recommendations

Our partners — which include estate agents,new home developers, mortgage lenders, and other industry professionals — rely on Rightmove to connect with the right audience at the right time. With Vertex AI and Gemini models working behind the scenes, we’re helping them do that more efficiently and effectively.

Take lead generation, for example. We’ve built a vendor scoring engine that analyzes user search patterns and on-site behavior to predict the likelihood that someone is a homeowner. This insight helps partners focus their time and marketing efforts on high-conversion leads, while offering more relevant products — such as mortgage options — to the right people at the right moment.

Next, we’re excited to use generative AI to build agentic, conversational user interfaces, enabling anyone across our network to interact with data or find insights using natural language. Whether it's a business user running a query, or a partner navigating market trends, we’re working toward a more natural, accessible way to engage with data.

Turning data into insight and insight into value

Another exciting way we’re unlocking the value of our data is through an Automated Valuation Model (AVM). This AI-powered tool predicts the sale and rental price of every property in the UK, every month, by analyzing a wide range of signals including market trends, supply and demand, and the condition of individual homes.

Traditionally, valuations were fairly static, based on fixed data points that didn’t reflect recent improvements or shifting market conditions. Vertex AI makes them dynamic. Whether it’s a newly renovated kitchen or a shift in local market conditions, we can factor in real-time changes to properties on our website, delivering more accurate, up-to-date valuations for both homeowners and estate agents.

These monthly valuations are invaluable to our partners. Mortgage lenders, and estate agents use them as trusted pricing guides to understand local markets and assess risk, especially when managing large property portfolios or backbooks.

Behind the scenes, the hive gives us access to both structured and unstructured data, including more than 25 years of property images that were previously siloed. Now stored securely in Cloud Storage, this rich visual data is fueling advanced use cases, including these enhanced valuation models and deeper market analysis.

Driving operational efficiency with a smarter, unified data platform

With our migration to the cloud, we’ve embraced a “hub and spoke” model to ensure both consistency and flexibility in how data and AI are used across the business. The “hubs” are our central teams — experts in BigQuery, Looker, and Vertex AI — who set best practices and help scale innovation. The “spokes” are our vertical business units, such as the New Homes department, that tap into the hive platform to run their own business intelligence and AI use cases, tailored to their specific needs.

By consolidating multiple legacy business intelligence tools into a single platform with Looker, we’ve simplified our tech stack and created operational gains. For example, the New Homes team has cut down meeting prep with developers from hours to minutes, thanks to easily accessible, self-serve Looker dashboards.

And we’re constantly discovering new ways to create value from our new platform. For example, as Google Cloud rolls out new features such as BigQuery’s Series FM function for low-code forecasting, our teams are quickly adopting them to move from descriptive to predictive analytics. Forecasting leads, time spent on site, or whatever KPI a business unit has, was previously unthinkable, with siloed data and manual processes for developing models and ingesting data somewhere else for analysis. In our new platform, we can quickly trial this kind of forecasting in a spoke using just 10 lines of code and our BigQuery-Looker integration. It only took weeks for many of our business units to start using Series FM for forecasting.

Helping users at every stage of homeownership with AI

As we look to the future, our AI strategy is expanding beyond helping people find a home. Rightmove is evolving into a smart, supportive companion for every stage of the home journey: find, afford, transact, move, and live.

That means using data and AI to continue helping users to find properties but also to make better-informed decisions throughout the entire lifecycle of homeownership after that. We’re already rolling out new capabilities, such as smarter mortgage in principle matching and insights into the total cost of ownership, including broadband, energy, and utility costs, so buyers can move with confidence.

One exciting step in this direction is using Vertex AI to power the models and data behind our Track a Property feature — a way for home-owners to regularly check the value of their home. This upgrade means the valuation models are built and trained faster, will improve their accuracy over the long term with added model engineering and tuning, as well as taking advantage of better cloud architecture to host them.

This is just the beginning. As Google AI continues to evolve, so does our platform, becoming a home and living assistant that supports not just the move, but the life that follows it.

How a leading consumer insight brand uses Dataproc to hyper-personalise faster

Mon, 06 Apr 2026 16:00:00 +0000

At RVU, we have a clear and vital mission: empower people, transform industries.

For our market-leading home management and switching brands — Confused.com, Uswitch, Tempcover, Money.co.uk, and Mojo Mortgages — transparency and accurate information are everything. Today’s consumer expects more than a simple comparison table; they want personalized recommendations tailored to their unique circumstances.

Delivering on that promise — building a true personalization engine that powers all our brands — requires a data foundation capable of processing massive, complex datasets for sophisticated ML models. Today, our platform powers hundreds of automated personalization campaigns, optimized with billions of data points from across all our brands. We tackled this challenge using the power of Google Cloud and its two solutions for Apache Spark: Dataproc and Google Cloud Serverless for Apache Spark. Together, we’re making our mission a reality.

The high-speed engine for feature engineering

Our relationship with Google Cloud isn’t new. In fact, we've used BigQuery as our unified data platform for over a decade. Coming from a performance marketing background, we’ve always dealt with a lot of data, but we recognized early on that we’re not a digital infrastructure company. Instead, our focus must always be on where the value is. Managed solutions like BigQuery that eliminate infrastructure and capacity headaches were a natural fit from the start.

The key challenge was stitching together a meaningful and coherent picture of customer behavior across our brands — turning countless fragmented interactions into something that genuinely reflects how a user behaves, clicks, and makes decisions. Instead of relying on isolated events and aggregate views, we’ve had to build a platform capable of connecting these signals into a narrative that works for our machine learning models.

Using Dataproc to support this was a gamechanger. The biggest impact has been its role as our core high-speed Spark processing engine, primarily for feature engineering for our ML model development. Feature engineering, which is the crucial process of shaping all that raw customer data for our data science models, is a real value-driver for us. It’s where we have a marked competitive edge.

The result has been a significant leap in our innovation velocity. With Serverless for Apache Spark, we now have the ability to shape our customer data for feature engineering in just a matter of days. Previously, this would have taken weeks. We’ve also dramatically reduced our time-to-market, which also used to take weeks. Now, a new contractor can join the team and deliver a model, including exploratory data analysis and all feature engineering, in only a week and a half. That’s incredibly fast.

Delivering personalized experiences

By improving our speed of innovation, we’re better positioned to deliver a personalized user experience to our customers and partners.

Our hyper-personalization journey accelerated once we moved to Spark. We can now run heavyweight data processing jobs that crunch vast amounts of behavioral and contextual data, allowing us to build models that generate genuinely meaningful predictions.

These models help us understand not just what to say to a customer, but when and how to say it — selecting the right moment and the right channel to deliver personalized insight that genuinely resonates.

Building a future vision

Google’s Data Cloud directly aligns with our culture of prioritizing value, and its impact on our business is profound. I call it the network effect, where everything seamlessly connects within the same ecosystem: Our data resides in BigQuery, our ability to validate, enrich, and transform that data is tied to Dataproc and Serverless for Apache Spark, and our capacity to deploy the ML models spans the network. It’s all co-located and integrated, powering the real-time accuracy of our consumer brands and giving us a competitive advantage.

For our engineers, the big win is the lack of infrastructure they have to deal with. They can press a button that processes all the data in 10 minutes, rather than having to set up a network of clusters and servers and make them talk to each other. It’s incredibly efficient and frees up time for more valuable work like building and iterating data products.

Dataproc has upped our speed, scale, and agility. It also gives us the tools to innovate with AI as we build the future of hyper-personalization. Today, we’re proud to say RVU’s cutting-edge tech and data are helping millions of UK consumers make smarter, more informed decisions, and truly transforming industries.

Inspired by RVU's success? Whether you need persistent clusters with Dataproc or the agility of Serverless Spark, Google Cloud has a managed solution to help you focus on value, not infrastructure. Discover the right Spark on Google Cloud for your use case.

How Honeylove boosts product quality and service efficiency with BigQuery

Thu, 02 Apr 2026 16:00:00 +0000

Building the perfect bra takes thousands of data points.

That’s why Honeylove isn’t just another intimates brand. We’re a technology company that happens to make exceptional bras, tops, shapewear, and bodysuits. Technology shapes everything we do, from how we iterate garments based on customer feedback to how we optimize sizing across those thousands of data points.

When Honeylove was born in 2018, though, our data wasn’t consolidated. We were looking at analytics in Shopify, checking email campaign performance in one platform, and reviewing ad metrics in another. We weren’t connecting the dots as effectively as we could have.

Then we fell in love with BigQuery. In this post, we’ll cover how Honeylove uses BigQuery and Gemini to unify our data, automate key business insights, and leverage AI to boost product quality and service efficiency — as well as how other organizations looking to make the most of their data can follow our approach intimately.

Transforming insights with BigQuery and Gemini

The first step was getting all our data in one place. BigQuery gave us exactly what we needed: a performant, economical, unified data platform that integrates seamlessly with the tools our team already uses within the Google ecosystem, such as Google Ads and Google Sheets . This helped eliminate manual data silos and enabled us to quickly adopt AI and ML capabilities across the business.

The real transformation came when we started leveraging BigQuery ML functions for contribution analysis. We built models to analyze the key drivers behind some of our most critical metrics: conversion rate, customer satisfaction scores, website performance, and return rates.

What’s really powerful for us is that we can feed these contribution analysis results directly into Gemini to produce accessible reports and summaries. Before implementing this approach, 10 to 15 people would spend an hour before key meetings manually reviewing dashboards, trying to drill into the data and find meaningful insights. We’ve saved hundreds of hours per year just by automating this process with Gemini.

But the impact of BigQuery and Gemini goes beyond time savings. These tools help us find patterns and insights we would’ve missed entirely. Even if you have the best marketing analysts looking over dashboards, they just wouldn’t be able to slice it in the same way these reports allow us to do.

We’ve also been able to transform forecasting inventory and demand planning, another area where manual processes previously dominated. By deploying and training BigQuery ML’s ARIMA univariate forecasting models, we’ve used high-accuracy SKU-level demand forecasts that automatically adjust for seasonality and recent changes.

These automated forecasts consistently come within 5% of what we calculate manually — a huge improvement over third-party vendors that were sometimes off by 20% to 30%. Having that additional checkpoint gives us more confidence when making critical inventory decisions.

Unlocking value and creative with multimodal embeddings

Customer service tickets can be a treasure trove of valuable feedback and information for ecommerce brands. But only if you can extract insights from them, and with Google Cloud, we can.

We leverage Gemini embedding models and BigQuery vector search to transform the unstructured text of our tickets into actionable data. We generate vector embeddings for tickets already in our data warehouse using simple SQL commands, and then use those vectors for semantic searching through retrieval-augmented generation (RAG).

This allows us to ask precise, natural-language questions, such as “What do customers love about our bras?” or “What changes would you like to see to our bodysuits?” In response, Gemini instantly identifies similar use cases, enabling us to move beyond keyword matching and quickly find the root causes of any issues, which are often nuanced. This proactively guides product improvements and enhances service efficiency. We’re saving about 30 seconds per ticket, which might not sound dramatic until you multiply it across thousands of interactions.

We’re also experimenting with multimodal embeddings for video asset search across our ad and influencer content library. It’s been fun to test queries like “find me videos with dogs” or “find me a video with a red dress” and watch it actually work. The next step is to use those embeddings to compare new creative assets with existing ones and predict performance based on our historical data.

Growth creative has traditionally been driven by gut feelings rather than numerical analysis, but we hope to change that by using our huge library of existing ad creative to inform what we test and create in the future.

Building for the future with Google Cloud

Today, Google Cloud and BigQuery are a central pillar of our company. They allow us to spend less time on manual tasks and more time on high-value work that solves real-world problems, making us very efficient as a small team.

Working with the Google Cloud team is invaluable. They’ve been a true partner, and they continue to support our roadmap. We’re leaning further into BigQuery ML functionality, moving more of our data science work into automated, always-available models rather than offline analyses.

We’re also developing internal knowledge bots using the Vertex AI RAG Engine, connected directly to our internal documents hosted on Google Drive, to provide instant answers to internal policy and process questions. Additionally, we’re experimenting with Conversational Analytics API to provide a “BI in a box” experience so our teams can ask plain-text questions and get metrics and charts without needing an analyst.

As a technology-first company, this transformation continues to have a profound impact on what we do at Honeylove. It accelerated innovation in product quality, improved operational efficiency, and ensured that our customers receive a more intelligent and consistent service experience.

Real-world success with Spanner’s fully interoperable multi-model database

Tue, 31 Mar 2026 17:00:00 +0000

In the first post on the power of multi-model databases to lay the foundations for gen AI, we highlighted how Google Cloud Spanner helps organizations overcome some of the challenges presented by traditional approaches to database architecture and management. In this post, we dive deeper on the specific examples, across four common use cases.

We are seeing customers increasingly choose Spanner's multi-model capabilities to address three key strategic goals:

A foundation of scale and reliability: Many specialized databases for graph, vector, or search, are built on traditional, single-machine architectures. As a result, they face fundamental challenges with scalability, availability, and consistency. We see customers migrate off these specialized systems because they have - or are about to - hit a wall. All Spanner’s data models are built on its tried-and-true platform offering 99.999% availability, automatic scaling, and limitless horizontal scale, and they can easily extend to new capabilities. For example, adding a vector embedding column to the existing graph schema.
Consolidating database sprawl and eliminating ETL: Managing, securing, and patching multiple disparate databases, each with its own data model, query language and backup policy can be an operational nightmare for users. Extract, transform, load (ETL) pipelines required to sync data, are especially frustrating as they often create inconsistency and delays. Spanner eliminates this complexity by offering multiple data models in a single unified database, eliminating extra data copies, inconsistency and management overhead. Moreover, Spanner’s interoperable multi-model capabilities allow a developer to write one SQL query that joins relational tables, traverses a graph relationship, and filters on a vector or text search function.
Future-proofing for evolving application needs: While many customers start with a simple application, they know it will need to get smarter and more complex over time. In Spanner, adding a graph-based recommendation or AI-powered vector search can be an afterthought. A developer can simply turn on graph or search capabilities on their operational data, with a simple data definition language (DDL) command. With Spanner, there is no painful migration, no complex re-architecturing and no growth ceiling. Instead, customers can build on a reliable relational database while seamlessly adding new, advanced data models as their application evolves.

Here’s how customers across industries are already leveraging Spanner's evolving multi-model capabilities to solve their toughest data challenges and achieve early success:

1. Fraud detection

Fraudsters often exploit complex, non-obvious patterns across multiple transactions and accounts. Traditional relational databases struggle to detect these intricate relationships in real-time. Spanner combines relational queries with graph analytics to enable real-time pattern recognition. This allows businesses to efficiently identify suspicious clusters or unusual connections that might indicate fraudulent behavior, significantly reducing financial losses and enhancing security.

DANA: Anti-money laundering for fast growing customer base

DANA, an Indonesia-based e-wallet app, offering payments and digital financial services including lending, insurance and investments, has adopted Spanner Graph to support its critical anti-money-laundering, or AML, efforts.

Challenge: With a massive and still rapidly growing user base, DANA struggled to scale and meet query performance SLAs using existing relational databases to detect money laundering patterns in transactions. Moving to do the analytics in graph databases was obvious, but many graph database providers in the market simply could not handle the scale.
Solution: Spanner was selected after an elaborate RFP process due to its high availability, virtually unlimited scale, and external consistency. The ability to use full-text search (FTS) and vector search directly within the Graph model were key differentiators.

Palo Alto Networks: Access graph for SaaS identity

Palo Alto Networks, one of the leading cybersecurity firms, leverages Spanner to provide insights into organizational identity posture, surfacing misconfigurations and over-privileged accounts, dormant accounts, unrotated credentials, over-privileged accounts, and accounts missing in the Identity Provider (IDP).

Challenge: The team needed to build a world-class agent security product for the AI era that could innovate quickly while ensuring highly scalability without creating data silos.
Solution: They built an "Access Graph" on Spanner to connect user identities, access permissions, and the associated user activities within the SaaS applications. Spanner allows them to achieve massive scale with a single schema for both graph and non-graph use cases seamlessly.

Verisoul.ai: Real-time fake user detection

Verisoul offers a unified AI-powered platform to detect and prevent fake users, ensuring accounts are real, unique, and trustworthy.

Challenge: Verisoul previously built and maintained 10 different independent services across Postgres, Cassandra, and Neo4j to handle a variety of types of data, such as network intelligence, device intelligence, behavioral and sensor data, email and multi accounting. This complexity made it difficult to provide zero-latency detection to counter the speed, scale and sophistication of modern-day fraud attacks,
Solution: By consolidating onto Spanner, Verisoul now can monitor hundreds of customers with millions of accounts in real time, capturing every login, page view, click, and mouse move. Spanner provided an all-in-one database for Graph, vector search, and seamless integration with BigQuery, allowing them to eliminate maintenance overhead while delivering unlimited throughput all with a simple architecture.

2. Recommendation engines

Personalized recommendations are at the heart of online consumer businesses. Building an effective recommendation engine requires analyzing vast amounts of user behavior data, product and service attributes, and historical interactions. Spanner’s interoperable queries allow you to combine user profiles (relational), interaction history (search), and product similarity (graph) to generate highly relevant recommendations in real time, driving better user engagement and improving conversion rates.

Target: Combining Vector and Graph Search for gift recommendations

Target sought to elevate the holiday shopping experience with a generative AI-powered Gift Finder for highly personalized gift recommendations.

Challenge: The application was run on a specialized search database, providing limited gift recommendations. To enhance and personalize the experience, Target needed a sophisticated upgrade.
Solution: Target selected Spanner Graph for its versatile hybrid query model. The solution blends graph traversals with vector search with their proprietary embeddings delivering intuitive, real-time product suggestions — all delivered just in time for the 2025 Black-Friday-Cyber-Monday shopping rush.

True Digital Group: Consolidating AI search

True Digital Group, Thailand's leading telecom-tech company, offers customers a wide array of high-quality digital services, encompassing both streaming and print media, along with customer loyalty tracking. Their AI-driven intelligent search feature ensures accurate content retrieval based on keywords and user intent.

Challenge: A fragmented stack with multiple databases resulted in outdated data, inconsistent tokenization, multiple query languages, and poor search quality, causing users to avoid the search feature.
Solution: True Digital consolidated all search functionality onto Spanner. By combining keyword and intent-based search results using SQL, they significantly improved search relevancy and accuracy, leading to increased customer engagement and satisfaction.

3. Hybrid search

Information retrieval is the critical bridge that grounds AI models in factual, up-to-date data and enables agentic workflow. Often, users must locate a specific needle in a haystack — searching through a massive corpus of legal documents, financial reports, or research papers. Interoperable multi-model Spanner empowers customers with hybrid search capabilities, ensuring AI models retrieve the most relevant context at any scale with pinpoint accuracy.

Rogo: Financial workflow automation

Rogo connects proprietary internal data with external financial sources like filings, PitchBook, LSEG, FactSet, and S&P Capital IQ to help finance professionals automate their workflows, from building pitch decks to drafting investment memos.

Challenge: Rogo needs to ingest and connect data from dozens of sources at once, across both structured and unstructured formats. Finding the right backend to support that wasn't straightforward.
Solution: Rogo chose Google Cloud Spanner for its high performance, scalability, and easy management. It lets them store and query both relational and document-based data in one place, which has made it easier to audit and maintain as the platform has grown.

Inspira: Streaming legal intelligence

Inspira, a leading legal tech company, provides AI-driven solutions tailored for legal research and general workforce optimization. Their platform serves law firms, corporations, and government entities, managing a massive repository of 75 million legal documents and 440 million vectors.

Challenge: Before migrating to Spanner, Inspira struggled with a complex, fragmented architecture, relying on a polyglot system consisting of Elasticsearch, BigQuery and Cloud SQL. This led to complicated data synchronization, and complex “two-stage” query filtering to combine keyword and vector searches. The team also needs a path to scale beyond 1 billion vectors without sacrificing latency and high read/write throughput.
Solution: To resolve these inefficiencies, Inspira consolidated their entire stack intoSpanner, drastically simplifying a 4.5 TB data pipeline into a unified, high-performance single-source of truth. Leveraging Spanner’s native support for both FTS and vector search, Inspira enabled single-stage filtering for hybrid queries and achieved high-precision snippets for LLM-based legal analysis with RAG workflow.

4. Autonomous network operations

Autonomous network operations (ANO) represents the transition from reactive maintenance to predictive, self-healing networks. By creating a comprehensive digital twin of the network topology and overlaying it with real-time operational data, telecommunications providers can automate root cause analysis, predict anomalies, and resolve network incidents without human intervention.

MasOrange: The digital twin

A temporal digital twin is at the heart of MasOrange’s ANO efforts, replicating its country-wide wireless network topology, alongside operations support systems (OSS), and business support systems (BSS) data.

Challenge: MasOrange needed a graph database that was highly available, infinitely scalable with zero RPO/RTO to serve as the foundation of its ANO stack. They required vector, and FTS capabilities without the operational overhead of managing multiple disparate solutions.
Solution: MasOrange chose Spanner for its ability to meet strict scalability and availability requirements while offering fully interoperable Graph, vector and FTS capability. Today MasOrange’s digital twin is live on Spanner, powering end-to-end anomaly detection and root cause analysis.

Looking Ahead

With the scale insurance, high reliability, global consistency, and versatility in handling different data models interoperably, Spanner is a future-proof database for your agentic workload.

We envision a future where the database becomes a simple implementation detail, allowing you to focus purely on accelerating developer productivity, improving operational efficiency and delivering your business goals. Visit our Spanner page to learn more and get started today.

How ID.me Scaled to 160M Users While Reducing Operational Risk

Mon, 30 Mar 2026 16:00:00 +0000

Editor’s note: ID.me is transforming digital identity security, proving that establishing your identity can be easy. What's more, their platform has scaled to support 160 million members and can support up to 40,000 users per minute.To support services like tax filing that require massive scale and power real-time AI, the team migrated 50 terabytes of data from their legacy platform to Google Cloud, adopting a modern architecture on AlloyDB, Cloud SQL, and Vertex AI. This architecture resulted in faster development, more accurate fraud detection, and a 40% reduction in their data teams’ overall work completion time.

Phone, driver’s license, credit card – you probably don’t leave home without some form of ID. It proves who you are, and it works almost anywhere. But online, you’re made to prove your identity again and again, and create new logins for every new service or tool you use. At ID.me, we think people should be able to verify their identity once, securely, and bring that same credential everywhere they go online.

Our goal is to create the digital wallet for identity: a trusted sign-in that works across the public and private sectors. Today, we serve over 160 million members. As identity grows ever more essential to how we live and work, we’re scaling to make it as easy to prove who you are online as it is to flash your driver’s license in person.

But demand doesn’t wait

In the last few years, ID.me has gone from serving 50 million members to more than 160 million. We no longer track usage by the day, but instead monitor it moment-to-moment. At this stage, our platform is designed to support up to 40,000 members per minute. That kind of demand changes the equation. For ID.me members, access is everything. It’s not just about uptime; it’s about ensuring that when a member needs to prove who they are—whether for government benefits, healthcare, or exclusive offers—we verify them securely and instantly. As usage grew, we realized our infrastructure couldn’t scale the way we needed it to. We were reaching the limits of our architecture. So we made the decision to rebuild the ID.me data foundation on Google Cloud for our next phase of growth.

A new database for us and our 160 million closest friends

The first order of business was to select a database that could truly provide the massive scalability and reliability we needed. AlloyDB for PostgreSQL immediately stood out. It directly addressed the scaling bottlenecks and operational complexities we faced with our prior setup, allowing us to confidently handle our peak demands. This shift didn't just solve our technical hurdles; it dramatically improved our developer experience. Our teams now spend far less time on provisioning, maintenance, and patching, accelerating our development cycle from weeks to just days.

Over the last two years, we migrated more than 50 terabytes of data across 15 database instances to Google Cloud – with minimal downtime. We also introduced a two-tier architecture where Cloud SQL supports our smaller and more standard services, while AlloyDB runs heavier workforces that form the backbone of the ID.me platform. That way, we can move fast without sacrificing stability. It also frees up our teams to focus on the work that actually drives innovation, like making sure AI works for, not against, our identity efforts.

aside_block: <ListValue: [StructValue([('title', 'Build smarter with Google Cloud databases'), ('body', <wagtail.rich_text.RichText object at 0x7f8cdf33d7f0>), ('btn_text', ''), ('href', ''), ('image', None)])]>

Training AI to fight . . . AI?

Of course, everyone is racing to use AI. At ID.me, it’s just as important to defend against its misuse. The threat landscape is evolving, especially as generative models get better at impersonating individuals and even creating synthetic identities. And since we’re in the business of verifying that people are who they say they are, that threat lands squarely on our doorstep.

One of the great things about AlloyDB is its ability to create multiple read pools. For us, those read pools have become data clean rooms that we can quickly share out with our data engineers and data scientists. Fraud analysts can go in, find what’s wrong, and either remediate or prevent it in real time.

Overall, AlloyDB AI has allowed us to scale our systems 10-20X of what we were able to handle – and with a decrease in price to boot. The impact of this is huge. ID.me has been recognized by the U.S. federal government for its role in preventing large-scale fraud within national systems. A crucial factor in this success was AlloyDB's built-in high availability and easy-to-scale read pools, which enabled the Internal Revenue Service (IRS)—the U.S. national taxing authority— to seamlessly process over 120,000 transactions per second during the last peak tax season without a blip. This effectively doubled their previous self-hosted PostgreSQL performance.

We’ve been playing with a lot of new capabilities, but the ones we’re most excited about are Gemini Enterprise and AlloyDB AI natural language as they represent a fundamental shift in how we build and interact with AI.

One login. Every system. Zero friction.

Our data teams are big fans of Google Cloud; it's made their work substantially easier. Since migrating, they can make changes much faster, leading to a 40% reduction in their overall work completion time. And across ID.me engineering teams, the developer experience has improved dramatically. Our teams can ship full product features in days instead of weeks, spending more time solving meaningful problems for our members.

We’ve been able to scale both our infrastructure and trust. With a platform that’s faster, smarter, and built to handle portable identity at massive scale, we’re one step closer to our goal: a secure, digital way to prove who you are, wherever you need it, that works everywhere you need it. You may never get asked security questions about your childhood pet again.

Learn more:

Learn more about the power of AlloyDB and AlloyDB AI.
Start your AlloyDB free trial today.
Learn how customers like Bayer and Character.ai are leveraging AlloyDB to transform their business.

Easy as a green run: How Vail Resorts built an AI assistant to automate personalized recommendations

Fri, 27 Mar 2026 16:00:00 +0000

For skiers and snowboarders, every moment on the mountain is about maximizing the fun — chasing fresh lines, perfecting a new trick, or exploring new terrain. Whether they're exploring a familiar favorite or visiting a new mountain for the first time, riders want the right info at their fingertips so they can move with confidence, find hidden gems, and feel like they belong from the moment they arrive.

It’s a big part of the reason Vail Resorts launched My Epic Assistant during the 2024-2025 snow season. Vail operates some of the most iconic and beloved mountain destinations in the world (Whistler Blackcomb, Park City Mountain, Stowe, and Crested Butte, to name a few). The company wanted every guest to feel fully supported by their new app — able to get quick, helpful answers and discover everything their mountain has to offer, all in a way that keeps them immersed in the experience.

Just think, would you rather be looking for an info booth, or getting help from the app while riding the lift to your next run?

My Epic Assistant is an AI-powered assistant that takes the expertise of Vail Resort’s IT, hospitality, and operational teams and feeds it into Google’s powerful Gemini models. The result is an agentic guide to the slopes that can help decide on the right season pass, share the latest snow report, check on lesson preparations, or suggest a good stop for cocoa. Vail Resorts wanted more than a chatbot; they wanted a digital concierge that understands the nuance of a powder day at Whistler versus a family trip to Beaver Creek.

To take these offerings to the next level, the teams at Vail Resorts and Google Cloud specialist 66degrees updated My Epic Assistant with new features for the 2025-2026 season, including season pass recommendations and improved personalization. This feature now intelligently guides guests to the best pass for them — understanding guest questions and handling complex requests.

It’s not just savvy Vail guests who can benefit from AI-powered tools like these, either.

In this blog post, we’ll detail how we built the My Epic Assistant, orchestrating a multi-agent system that can expertly slalom around the unique challenges of this rugged corner of the hospitality industry. It’s an approach that organizations in other customer-centric sectors can follow to help build agentic system for their own unique situations.

A new season, and better results

As we’ve implemented and refined personalization, improved search, summary capabilities, and conversational flow within My Epic Assistant, the app has achieved a 45% reduction in escalation to human agents since its initial launch.

By automating routine logistics and expanding personalization, Vail Resorts' Guest Experience Technology team has been able to ensure that technology never replaces the personal touch, it only enhances it. The overall experience for guests seeking support is now faster, more reliable, and available in more ways and times.

According to the team, "Utilizing tooling from Google Cloud, we could lean into agentic design patterns that gave us a way to unlock natural, personalized conversations. These boosted customer satisfaction, while reducing the need for manual intent design. These tools also let us combine flexibility and control to enable the assistant to respond fluidly but always within the boundaries of our brand, policies, and product strategy.”

The steps to building My Epic Assistant

First, My Epic Assistant uses a subtopic classification agent to understand the guest's initial request — whether they want to compare passes, get a recommendation, or just find general information. A date object is used to determine the time of year, as passes are only available for purchase for a limited time. If passes are no longer on sale, it will automatically route to information on lift tickets. If both passes and lift tickets are available, it asks a few clarifying questions to help customers decide which option will provide the best value for their trip.Next, the assistant hands off to the data collection agent, to determine what additional data is needed to provide a recommendation. For authenticated users, Vail Resorts provides their existing data like their home resort, average visits, age, upcoming visit resort, and peak day preferences called from a webhook. For a new non-authenticated user, it asks clarifying, easy to answer questions to fill in the gaps. There’s no one-way, static form filling here!

The generative AI’s conversational capabilities really shine during this step. My Epic Assistant is designed to chat freely back and forth, allowing guests to ask clarifying questions at any point and then return to the recommendation process without losing context. Via the instructions in the playbook, the model continually assesses if all parameters have been filled for the output in order to move on to the next step, and still allows guests to ask other questions during this process.

For instance, it’s possible that when asked if they want peak-day access, a guest might follow up with “What are peak-restricted dates this year?” My Epic Assistant will respond by invoking the website datastores that codify Vail Resorts’ extensive knowledge. My Epic Assistant then provides an answer and returns to the previous turn in conversation without losing context.

Finally, the recommendation agent takes over. Using the collected user data, it queries a structured database of all pass options to find the perfect match. The system then generates a user-friendly response explaining why an identified pass is a good fit and provides content cards with direct purchase links, simplifying the final step for the guest.

The core functionality of the pass recommendation agent is a datastore tool of a pass matrix that contains a structured file of all existing passes, their features, and their restrictions. With a broad number of options available, extensive testing was necessary to ensure the tool returned valid and appropriate options once it received all of the input parameters, and 66degrees relied on Vail Resorts’ deep domain knowledge to validate every possible outcome.

Then, prompt engineering ensures that the ultimate response presents the recommendations with detailed explanations as to why those specific passes were the best options for them. Content cards are also invoked with purchase links for those recommendations, without having to exit the playbook. For some occasions, a generative response was not appropriate, such as with specific details on the newly launched Epic Friends feature. Static responses are called via a code snippet and a conditional action, again housed within the playbook itself, simplifying the overall architecture.

Personalization made easy as a bunny slope

Thanks to the new combination of smart technology and user-focused design in the My Epic Assistant, Vail Resorts moves beyond simple customer Q&A to a truly conversational yet fully automated experience for many customer requests. It’s about providing the same concierge services resort-goers expect from an operator like Vail Resorts, yet achieving them at scale that’s impossible without AI and cloud technologies.

With pass recommendations from My Epic Assistant, Vail Resorts is gearing up every guest to receive custom recommendations for their best season yet.

Want to turn your technical black-diamond challenges into easy green groomers? Contact the experts at 66degrees and Google Cloud to discover what’s possible with the latest AI and cloud technologies.

How FM Logistic tackled the traveling salesman problem at warehouse scale with AlphaEvolve

Thu, 26 Mar 2026 16:00:00 +0000

The traveling salesman problem asks a deceptively simple question: What's the shortest route that visits every point exactly once? It's one of the hardest problems in computer science, and mathematicians have been working on it for nearly a century. It's also what FM Logistic's warehouse operators face every day in Poland.

The facility spans eight football fields. It holds over 17,700 picking locations. And across every shift, up to several dozen operators on ride-on electric trucks crisscross the floor collecting cartons, each one navigating dozens of storage locations per tour. Every unnecessary step adds up: in time, in wear on the fleet, and in delayed fulfillment.

FM Logistic, a global logistics provider operating in over 14 countries, had already optimized their routing once. Their existing model used a fast, cost-prioritized allocation logic built for real-time responsiveness. It worked well, but it made decisions step by step, which limited how well it could coordinate routes across the full warehouse. With dozens of operators working the same floor across shifts, even a small routing improvement would compound quickly.

So they turned to AlphaEvolve on Google Cloud.

Teaching an AI to write better algorithms

AlphaEvolve is an evolutionary coding agent that generates and refines algorithms autonomously using Gemini models. Rather than calculating a schedule from fixed rules, it works as a coding partner: writing new code, scoring it, and iterating until it finds a better solution than the one it started with.

The team didn't start from scratch. They gave AlphaEvolve a "seed" program: their existing algorithm, which made routing decisions one step at a time based on what looked best in the moment. This gave the agent a working baseline that already solved the problem, just not optimally. From there, AlphaEvolve used Gemini to generate variations of this code, introducing mutations and new logic to see if it could beat the human-designed original.

Measuring what good looks like

For AlphaEvolve to improve, it needs a way to measure how well each algorithm performs. FM Logistic designed a custom evaluation function using a representative dataset of 60 tours (over one hour of workforce data), letting the agent test thousands of generated algorithms against real-world conditions.

The evaluation scored every new piece of code on a primary goal: minimize the average travel distance per pick, while avoiding operational failures. The team built in specific penalties to steer the model away from unworkable solutions — things like exceeding forklift capacity, missing pending orders, assigning the same box twice, violating FIFO priority for older orders, or exceeding the computation time required for real-time operations.

The results

The new routing logic delivered immediate, measurable gains over the previous best baseline:

10.4% improvement in routing efficiency over the previous best solution.
15,000+ fewer kilometers of warehouse travel per year at full operational scale.

That efficiency gives FM Logistic room to handle larger order volumes with the same team and equipment, without adding headcount or expanding their fleet.

"Through our partnership with Google Cloud and the implementation of AlphaEvolve and Gemini, we further optimized our routing approach for fast-moving operations," said Rodolphe Bey, Group CIO at FM Logistic. "The 10.4% improvement was achieved on top of an already highly optimized baseline, where further gains are typically hard to come by. This translates directly to faster fulfillment, improved working conditions for our teams, and reduced wear on our fleet."

What the winning algorithm actually does

By running a series of experiments, each generating hundreds of candidate programs, AlphaEvolve developed a new algorithm that outperformed the previous best human-engineered one. The result is a set of clear, human-readable rules that warehouse teams can review and adjust as needs change.

The three core improvements:

Density-based starting points (Anchor selection): The previous system chose a starting mission based on the single location where the most missions overlapped. The new algorithm looks more broadly, identifying clusters of items that are close together and using those dense areas as "starting anchors" for building routes. Every tour begins with a highly efficient core.
Two-step filtering with distance simulation: To maintain real-time speed, the algorithm uses a two-stage process. First, a quick filter eliminates orders that do not fit the route's logic. Second, a precise distance simulation runs only on the best remaining candidates to find the most efficient path, without slowing down warehouse operations.
Flexible route building: If the algorithm can’t fill a truck efficiently around a specific starting point, it doesn’t force a bad route. It returns those orders to the main pool so they can be picked up by a better-fitting route later, improving efficiency across the entire warehouse.

What’s next

The Poland pilot (which is now running in production) demonstrated what evolutionary AI can do for complex routing at warehouse scale. FM Logistic is now exploring extensions — applying the algorithm to other high-volume e-commerce facilities, researching how AlphaEvolve might help optimize road transport for less-than-truckload shipments, and investigating AI-driven product placement inside warehouses to further cut travel distances.

^{This project was a collaboration between the FM logistic team including: Mateusz Klimowicz, Jarosław Urbański, Florent Martin and Alberto Brogio and the AI for Science team at Google Cloud including (but not limited to): Kartik Sanu, Laurynas Tamulevičius, Nicolas Stroppa, Chris Page, Gary Ng, John Semerdjian, Skandar Hannachi, Vishal Agarwal and Anant Nawalgaria as well asMariusz Czopiński from the Google account team as well and partners at Google DeepMind}

Manhattan Associates powers over a billion daily API calls with Google Cloud databases

Thu, 26 Mar 2026 16:00:00 +0000

Editor’s note: Manhattan Associates, a global leader in supply chain and omnichannel commerce solutions, modernized its Manhattan Active SaaS platform by moving from legacy Oracle and DB2 systems to Google Cloud databases. With Cloud SQL and BigQuery, the company now processes over a billion API calls per day with average response times under 150 milliseconds, supporting hundreds of thousands of monthly active users across tens of thousands of stores and distribution centers.

From monolithic roots to cloud resilience

Our Manhattan Active SaaS platform supports global supply chains, requiring constant uptime and performance. Our legacy Oracle and DB2 infrastructure created operational drag through manual scaling, complex licensing, and high maintenance overhead. We needed a new database foundation that provided contractual SLAs for availability, automated resilience, and a predictable cost model.

We chose Google Cloud databases because they give us the right balance of flexibility, scalability, and operational simplicity needed to run Manhattan Active at global scale. With managed databases like Google Cloud SQL, we traded manual upkeep for built-in high availability, scalability, and cross-region disaster recovery.

Each capability of Manhattan Active now runs as an independent, containerized service orchestrated by Google Kubernetes Engine (GKE). Data streams through Pub/Sub into BigQuery for real-time analytics, while Cloud Logging and Cloud Monitoring deliver observability at scale. This microservices-first design, powered by Google Cloud’s managed services, gave us the agility to evolve faster and the confidence that mission-critical operations would remain resilient across regions.

aside_block: <ListValue: [StructValue([('title', 'Build smarter with Google Cloud databases.'), ('body', <wagtail.rich_text.RichText object at 0x7f8d0006ae20>), ('btn_text', ''), ('href', ''), ('image', None)])]>

Building resilience and speed into every transaction

With a new foundation on Google Cloud, we could rethink how our platform delivers value at scale. The Manhattan Active architecture works alongside managed databases to turn supply chain complexity into responsive, resilient systems.

The benefits show up across the platform. Cloud SQL powers the core of Manhattan Active, quickly and reliably running millions of supply chain transactions per day. Real-time analytics flow into BigQuery, giving retailers sharper forecasting and faster anomaly detection. Automated failover and cross-region replicas safeguard business continuity, so critical services stay available even when disruptions hit.

A modernized foundation: From database to intelligence

The move from legacy Oracle and DB2 systems to Google Cloud databases solved more than just a performance issue; it gave us a resilient foundation for what came next. That reliability and scale let Manhattan Associates bring generative AI directly into the supply chain.

Our Agentic AI suite includes prebuilt agents — like the Intelligent Store Manager and Labor Optimizer — that coordinate real-time decisions across store and distribution center operations. The Manhattan Agent Foundry also lets customers build custom AI agents using a low-code environment. That same foundation powers internal efficiency too, with use cases like real-time log analysis, developer code assistance, and scenario simulations.

Unprecedented speed, scale, and operational efficiency in practice

For retailers, the impact of this platform modernization is immediate: tangible speed and reliability.

With the help of Cloud SQL and BigQuery, Manhattan Active now supports an astounding number of API calls per day, with an average response time under 150 milliseconds. This speed supports hundreds of thousands of monthly active users across tens of thousands of stores and distribution centers, enabling real-time decision-making where it matters most.

Operationally, the platform has become more elastic and efficient. The system automatically handles hundreds of thousands of scaling events per day, ensuring performance remains consistent during peak surges without expensive overprovisioning. Database observability tools like query insights give engineers clear visibility, so we spend less time on database patching and reactive troubleshooting and more on feature development and performance tuning.

For Manhattan Associates, resilience is now a built-in capability. And for retailers depending on our software, that translates into supply chains that are smarter, faster, and ready for whatever comes next.

Learn more

Discover how Cloud SQL can transform your business! Start a free trial today!
Download this IDC report to learn how migrating to Cloud SQL can lower costs, boost agility, and speed up deployments.
Learn how Ford and Yahoo gained high performance and cut costs by modernizing with Cloud SQL.

Small models, high quality: Inside BMW Group’s experiments evaluating domain-specific language models

Wed, 04 Mar 2026 17:00:00 +0000

A car you can talk to has been a longstanding dream, whether as the basis for television shows or more recent smartphone integrations.

One way of achieving better, more natural voice commands is by incorporating AI foundation models into vehicle systems, which offer more intelligence than traditional voice commands. AI foundation models can connect everyday questions with vehicle functions in a seamless dialogue. These models allow drivers to focus on the road ahead and enjoy every aspect of the journey while making interactions more intuitive.

While large language models (LLMs) offer powerful capabilities, they present one considerable drawback, at least in automotive settings: their reliance on consistent network access makes LLMs impractical for in-vehicle use due to potential lag and interruption.

To deliver reliable, next-level intelligence, BMW Group and Google Cloud successfully completed a proof of concept to build an efficient, reproducible solution to automate the workflows for fine-tuning, optimizing, evaluating, and deploying language models for specific domains, with special focus on small-language models, or SLMs. In this blog, we want to show results, findings and provide source code to encourage wider adoption.

“Finding the optimal trade-off for small-language models is a challenging, iterative process,” Dr. Céline Laurent-Winter, vice president, Connected Vehicle Platforms at BMW Group, said. “Automating the workflow for training, testing, and deploying domain-specific SLM allows a big push for our development efficiency. With automated pipelines, we can rapidly adapt models to our domain and rigorously test and evaluate them against domain-specific benchmarks. This allows us to iterate and optimize models in hours rather than days, in an automated, reproducible workflow.”

Small language models: small concept, big potential

Generative AI offers automakers powerful new capabilities, enabling complex voice commands. Before, it would have been almost impossible for a voice command system to understand a request like: “Find me a restaurant with vegetarian offerings along my route that is open now and has a customer rating higher than four stars.” With its language understanding and reasoning capabilities, gen AI can puzzle out such a request.

Integrating this intelligence, however, presents a challenge: Cloud-based LLMs are powerful but rely on a stable network to avoid frustrating lag. Conversely, onboard LLM are constrained by a vehicle’s limited computing hardware.

Small language models may offer an ideal balance — but finding the right trade-off between size and capability requires careful optimization.

These purpose-built, right-sized generative AI models can be run directly on edge devices, including vehicles. A common approach is having the SLMs handle the most frequently used features locally and only routing more complex requests to a cloud-based LLM. An SLM must be small enough to run on the target device, yet capable enough to be useful — especially when tailored to the specific automotive context via fine-tuning

Challenges of integrating foundation models into vehicles

Compared to the cloud, vehicle infotainment systems have limited storage and computing power. A 5 Series sedan or X3 SUV might look big, but there’s still limited space given all the performance, technology, and luxury that must fit between their four wheels.

Therefore, integrating a large language model, such as Gemma 3 27B which can consume over 40 GB of memory at 16-bit precision, is difficult. While smaller versions exist (e.g., Gemma 3 270M), they still tend to have a broad, generalized focus albeit with potential reduced accuracy compared to bigger models.

Hence, model compression (to reduce size) and tuning (to ensure high accuracy) become necessary for specialized use cases like ours. The goal then is finding the best tradeoffs between model size, inference time, and accuracy for the most frequent tasks.

Converting LLMs to SLMs

Turning large, resource-intensive LLMs into efficient SLMs requires well-known compression and quality enhancement techniques. Here’s a (reduced) overview of common techniques we’ve explored:

Compression techniques:

The primary goal is to reduce the model's compute and memory complexity. This can be done via:

Quantization: Reducing the model's memory footprint by converting high-precision parameters (e.g., 32-bit floats) to lower-precision formats (e.g., 8-bit integers or 4-bit floats). This leads, however, to a potential, but often minor, reduction in accuracy.
Pruning: Systematically identifying and removing the least important parameters or connections within the neural network, streamlining the SLM while retaining core capabilities.
Knowledge distillation: A compact "student" model is trained to replicate the performance of a larger "teacher" LLM, transferring complex knowledge into a much smaller, more efficient architecture.

Post-compression quality enhancement

We further engaged methods that can help recover or improve performance lost during compression.

Fine-tuning (and LoRA): Adapts the compressed model to a specific domain using targeted datasets. Standard approaches are Parameter-efficient fine-tuning (PEFT), such as Low-Rank Adaptation (LoRA). LoRA freezes the original weights and injects smaller, trainable matrices, dramatically reducing computational and storage costs while matching the performance of full fine-tuning.
Reinforcement Learning (RL): Methods like Proximal Policy Optimization (PPO), Direct Policy Optimization (DPO), and Group relative policy optimization (GRPO) are used for alignment with human preferences. RL iteratively improves model outputs by rewarding desired behaviors, guiding the model to generate more useful and accurate responses.

Evaluating performance for automotive tasks

Once a model has been compressed and enhanced, a crucial final step is to rigorously evaluate its performance. This covers system performance (e.g., latency, resource utilization on target hardware) and the qualitative assessment of the model's generated responses. For assessing quality, established methods are:

Point-wise evaluation: These methods assess the quality of a single generated response by comparing it against a pre-defined "ground truth" or reference answer. Examples include ROUGE and BLEU metrics.
Pair-wise evaluation: This approach determines which of two different model outputs is better, often aligning more closely with subjective human preferences for conversational quality. This can be executed with an Auto-rater (or LLM-as-a-judge) or direct Human Feedback.

Developing a robust testing strategy combining these evaluation methods is essential for validating the success of the compression and fine-tuning efforts.

The challenge of finding the optimal configuration

The path from a general-purpose LLM to a specialized SLM is not straightforward. Every choice — from type of quantization to characteristics and contents of the fine-tuning domain-specific dataset — directly affects the quality and efficiency of the final model. This creates an exponential range of possible configurations each with its own trade-offs.

This intricate landscape is further complicated by practical constraints: Not every compression or enhancement technique is applicable to every language model, and some methods are incompatible. For example, API-only models like Google Gemini permit fine-tuning only through a fixed set of methods.

The sheer volume of viable combinations renders a manual search for the optimal configuration an incredibly tedious, if not impossible, undertaking. To overcome this challenge, we built automated, reproducible workflows through executable pipelines.

Solution: An automated workflow for SLM optimization

Our solution is an automated workflow that orchestrates compression, adaptation, and evaluation steps needed to produce optimized SLMs. This is achieved by designing a flexible pipeline where each step is a modular, parameterized component. This workflow-based approach allows us to systematically explore the vast configuration space and pinpoint the best-performing models for in-vehicle deployment.

The process is structured as a workflow that can be executed automatically on a powerful workflow engine, such as Vertex AI Pipelines. In this workflow, we can define the sequence of operations (e.g., quantization, followed by LoRA fine-tuning and DPO) as a chain of interconnected components. Through pipeline parameters, we can search the entire configuration space, test different base models, compression techniques, tuning methods, and evaluation datasets.

This automated search allows for the comprehensive exploration of possibilities that would be unfeasible to test manually. The final artifacts from each pipeline execution are fully traceable and ready for deployment. This includes the versioned SLM itself, exact configuration parameters that produced the model, datasets used for evaluation, and a detailed report of its performance metrics, ensuring complete reproducibility.

Implementation: An automated workflow with Vertex AI Pipelines

Our solution is built on Google Cloud's Vertex AI platform, using configurable, executable pipeline templates. This offers a structured and automated way to find optimal SLMs in the vast possible search space. Figure 1 illustrates this workflow, its steps and their interactions with various data and model stores.

Figure 1: High-level overview of the automated pipeline's steps and its interaction with data and model stores.

Step 1: Versioning and configuration

Every Vertex AI workflow begins in Vertex AI Experiments. This initial step ensures the entire process is version controlled. The chosen LLM and datasets as well as the pipeline's configuration parameters are all logged as a single, versioned entity, ensuring complete traceability and reproducibility for every experiment.

Step 2: Optimization and compression

This stage puts the compression and enhancement techniques we discussed earlier into practice. Crucially, the pipeline is designed to manage the complex compatibility matrix between models, methods, and parameters. A pipeline template can, for example, enforce that only certain fine-tuning methods are applied to specific model architectures they are known to support, thereby automating the management of these constraints.

Our implementation provides reusable and standardized components for various fine-tuning (e.g., LoRA) and reinforcement learning methods (e.g., DPO, GRPO, and PPO). For compression, we adopt post-training quantization methods mapping models to lower-bit data types (e.g., bfloat16, 4-bit floats, or 8-bit integers) tailored to the target hardware's specifications.

Step 3: Conversion and deployment testing

Once an SLM is optimized, the pipeline deploys it to an environment. This allows testing if the model deployment succeeds on hardware representative of the target environment. This step provides a crucial, early validation point for the model's technical viability under realistic conditions.

An example would be running SLM on Android devices directly and natively (i.e. without emulation layers) on compute instances in the cloud. This allows testing how the model works on the target environment.

Step 4: Evaluation

A comprehensive evaluation is conducted to measure the SLM's true performance. This goes beyond simple accuracy, encompassing hardware-specific metrics like memory usage and inference latency as measured on the cloud-based device emulators. We also assess response quality using a combination of evaluation methods.

This can include point-wise metrics like ROUGE and BLEU, as well as more advanced pair-wise methods like auto-raters. The pipeline is designed to use custom test datasets reflecting a wide range of in-car tasks, such as multi-turn response generation or query rewriting with conversational context. This robust evaluation framework is also forward-looking, with the capability to assess multimodal SLMs such as Google Gemini and Gemma.

Step 5: Visualization and analysis

The Vertex AI Experiments allows storing generated metrics, comparing different experiment runs side-by-side, and creating visualizations using integrated tools like TensorBoard and Looker, making it easy to identify the most promising SLM candidates.

Figure 2: The automated pipeline as viewed in the Vertex AI Pipeline interface.

This entire automated workflow, from versioning to evaluation, creates a powerful feedback loop. It enables continuous integration and refinement, allowing teams to rapidly iterate and adapt their SLMs to evolving requirements and discover optimal configurations that would be almost impossible to find via manual efforts.

Conclusion and looking ahead

In this blog, we detailed how the automated workflow built on Google Cloud's Vertex AI successfully streamlines SLM development. This enables rigorous evaluation which model architectures or types (like Gemini, Gemma, and Llama) offer the best trade-off for our domain regarding performance, accuracy and size.

Crucially, we are linking our approach with the BMW Group’s "Head unit in the cloud", running the Android Open Source Project (AOSP) based infotainment system natively on cloud compute instances. This allows to test SLMs, including multimodal functions, in a virtual, scalable environment without the need for limited embedded devices.

The BMW Group's commitment to delivering cutting-edge in-vehicle experiences via artificial intelligence aligns seamlessly with Google Cloud's expertise in AI and machine learning. As we look ahead, we anticipate a continued partnership that will push the boundaries of what's possible in automotive AI.

We are publishing the solution of our PoC in the form of a SLM pipeline on GitHub. Feel free to adapt it to your needs and build your own optimized SLM!

_{This blog was written by Dr. Michael Menzel, Google Inc., and Dr. Jens Kohl, BMW Group, and is based on work done in a PoC which involved Dr. Arian Bär, David Katz, Dr. Felix Willnecker, Dr. Jens Kohl, Karsten Knebel, Dr. Manuel Luitz, Paul Weber, Raphael Perri, Thomas Riedl (all BMW Group) as well as Florian Haubner, Marcel Gotza, Dr. Michael Menzel, Raul Escalante (all Google Inc.).}

Cool stuff Google Cloud customers built, Feb. edition: Telco data reinvention; Golden State’s “G.O.A.T.T.”; John Lewis explores DORA

Thu, 26 Feb 2026 17:00:00 +0000

AI and cloud technology are reshaping every corner of every industry around the world. Without our customers, there would be no Google Cloud, as they are the ones building the future on our platform. In this regular round-up, we dive into some of the exciting projects redefining businesses, shaping industries, and creating new categories.

For our latest edition, we explore a new data approach for Vodafone and Fastweb; evaluating John Lewis Partnership’s developer platforms; the Golden State Warrior’s AI playbook; healthy, stable networks at Hackensack Meridian Health; and Ab Initio brings better context to data for AI.

Be sure to check back next year to see how more industry leaders and exciting startups are putting Google Cloud technologies to use. And if you haven’t already, please peruse our list of 1,001 real-world gen AI use cases from our customers.

Fastweb + Vodafone reimagined data workflows

Who: Following the acquisition of Vodafone Italy by Swisscom in 2025, these leading European telecom providers wanted to rethink how they serve customers and deliver timely, personalized experiences across mobile, broadband, and digital channels.

What they did: Both companies had already begun modernizing customer data workflows with BigQuery, but combining ecosystems exposed certain limits of the existing setup. In order to give every channel real-time access to accurate customer data, they implemented Spanner as a service and governance layer, delivering low-latency reads, horizontal scalability, high availability, and a fully managed environment with zero ops overhead. The team is also using Gemini to generate clear documentation directly from the code, which saves hours of manual work.

Why it matters: Using Spanner Graph allowed the organization to map lineage in a way that reflects how its platform actually works: which tables drive specific jobs, how transformations cascade, and where dependencies sit. Call centers now see more complete, up-to-date customer information, digital channels can rely on consistent data without custom integrations, and partners can access what they need with low latency through Apigee.

Learn from us: “Rebuilding our Customer 360 platform with Google Cloud services has already changed how Fastweb + Vodafone works. Workflow monitoring is simpler, pipelines are leaner, and real-time serving is now the norm. ” – Vincenzo Forciniti, IT AI Adoption & Platform Engineering Lead, Fastweb + Vodafone

John Lewis measures the value of its developer platform

Who: The John Lewis Partnership is a major UK retailer operating John Lewis department stores and Waitrose supermarkets. To power their digital transformation, they built the John Lewis Digital Platform (JLDP) to support dozens of product teams building high-quality software for johnlewis.com.

What they did: Moving beyond simple usage metrics, John Lewis developed a sophisticated, multi-stage approach to measuring the real value of their platform. They transitioned from initial speed-based metrics (like "Onboarding Lead Time") to a comprehensive model using DORA metrics and subjective engineer feedback via the DX platform. This included a custom "Technical Health" feature that uses small, automated jobs to monitor more than 35 health measures — such as Kubernetes best practices, security, and operational readiness — providing teams with real-time "traffic light" indicators of their service health.

Why it matters: By focusing on value rather than just activity, John Lewis ensured the platform was actually reducing friction for developers rather than just being a mandatory tool. Their automated Technical Health checks allow product teams to manage technical debt and security vulnerabilities proactively. This approach has decoupled centralized operations teams from individual services, leading to faster incident resolution (MTTR), fewer outages, and significant cost savings.

Learn from us: "Measurement is a journey, not a destination. Start by measuring something meaningful to your stakeholders, but be prepared to adapt as your platform evolves. The things that mattered when you were proving out the platform's viability are unlikely to be what are important several years later when your features are mature." – Alex Moss, Principal Platform Engineer, John Lewis Partnership

Hackensack Meridian Health de-risks network migration using VPC Flow Logs

Who: Hackensack Meridian Health is a leading not-for-profit healthcare organization and the largest hospital system in New Jersey. System reliability is a cornerstone value for HMH as they manage a vast network of hospitals, urgent care centers, and physician practices.

What they did: Preparing for a large-scale migration to a new Google Cloud network design, Hackensack Meridian Health used VPC Flow Logs and Flow Analyzer to eliminate the "black box" of hybrid traffic. By enabling logs on their Cloud Interconnect VLAN attachments, they captured granular telemetry — including source/destination IPs, ports, and protocols. They then exported this data to create a visual "who-is-talking-to-what" map. This allowed them to identify critical traffic patterns between on-premises data centers and specific Google Cloud regions, VPCs, and applications.

Why it matters: In a healthcare environment, even minor network disruptions can have major consequences. By mapping traffic proactively, Hudson Meridian Health pinpointed exactly which moments in the cutover carried the highest risk. This preparation allowed them to detect a migration issue in just three minutes and resolve it within five — a process that previously could have taken hours. Beyond migration, this level of visibility enables the organization to better manage capacity planning, cost attribution, and security compliance across their hybrid infrastructure.

Learn from us: "Getting a clear picture of our interconnect traffic always felt like a black box. Enabling VPC Flow Logs and feeding it into Flow Analyzer finally gave us the map we needed. Identifying those critical traffic flows before we changed any routes was key to de-risking the entire migration." — Randall Brokaw, Cloud Engineering Manager, Hackensack Meridian Health

The Golden State Warriors’ AI-powered back office

Who: The Golden State Warriors are one of the NBA’s most successful modern franchises. Behind their on-court wins are a specialized operations team who run what might be called organization’s "G.O.A.T.T." (Greatest of All-Time Technologies), a data and AI platform that helps drive game-time insights, trading decisions, and fan experience enhancements.

What they did: The Warriors transitioned from a "gut-feeling" culture to an "analytics-first" strategy by building an internal "digital brain" on Google Cloud. Using BigQuery and Gemini, the team now automates complex workflows that previously took hours, such as generating pre-game scouting reports. They use machine learning to run thousands of trade simulations that prioritize "team fit" over raw individual stats and employ computer vision to track the "shot quality" of every attempt in the NBA. On the business side, they built a content recommendation engine using the Discovery API to deliver personalized digital experiences to their global fan base.

Why it matters: This AI-driven approach narrows the decision tree for leadership, allowing them to focus human expertise on the most viable options. By automating the “science” of data processing, coaches and scouts have more time for the "art" of face-to-face training, planning, and player development. This integration has not only influenced on-court strategy — like the three-point revolution — but has also improved business efficiency, with employees now proactively bringing AI-driven ideas to the IT team rather than waiting for top-down mandates.

Learn from us: "You can never reach a point where either humans or machines are making all the decisions. The sweet spot is finding that middle ground where intuition and data converge on the same conclusion. Data helps us narrow our decision tree before we even start evaluating specific options." — Nick Manning, Senior Director of Consumer Products & Emerging Technology, Golden State Warriors

Ab Initio unlocks enterprise data for the agentic AI era

Who: Ab Initio is an enterprise software company specializing in high-volume data integration and governance. Their platform is trusted by large-scale organizations to manage complex data lifecycles across hybrid and multi-cloud environments.

What they did: To solve the challenge of grounding AI agents in accurate data, Ab Initio partnered with Google Cloud to integrate its data fabric with BigQuery, Dataplex Universal Catalog, and Gemini. They launched a suite of more than 500 metadata and data connectors that bridge the gap between legacy systems (like mainframes, COBOL, and SAS) and modern cloud environments. This integration provides field-level, end-to-end lineage, allowing Gemini to access well-documented, "AI-ready" data regardless of where it resides.

Why it matters: AI agents are only as effective as the data they can access. By using Ab Initio as a "neutral hub," enterprises can federate data from on-premises and multi-cloud sources into a single unified layer without moving the data itself. This provides the rich semantic context and lineage needed for Gemini to perform grounded, explainable reasoning. For businesses, this means faster transition from experimental AI to production-ready agentic workflows that are auditable, compliant, and capable of making complex, automated decisions.

Learn from us: "Agentic AI requires trusted, AI-ready data and metadata. Understanding the origin, quality, and meaning of information matters as much as the data itself. Gemini serves as a key component of the agentic layer, using this context to make decisions that are explainable and auditable." — Scott Studer, Head of Development, Ab Initio & Chai Pydimukkala, Data Governance, Sharing & Integration Product Lead, Google Cloud

Decommission your legacy Apache Cassandra stack and build for the future with Spanner

Thu, 19 Feb 2026 17:00:00 +0000

An increasing number of customers are migrating to Spanner from legacy NoSQL environments like Apache Cassandra. The strategic drivers are evident: a markedly lower total cost of ownership (TCO), elastic scalability, and near-zero maintenance overhead.

With the general availability of the native endpoint enabling Cassandra Query Language (CQL) APIs on Spanner, your existing Cassandra applications can now leverage Spanner’s enterprise foundation, featuring strong consistency, virtually limitless scale, and 99.999% availability — all while utilizing familiar CQL.

Better yet, migrating your application to Spanner with the CQL interface typically requires only a one-line code change, as your existing CQL statements remain valid. Combined with our integrated, high performance bulk and live migration tools, your transition from Cassandra to Spanner is simple.

Beyond NoSQL: Strategic solutions for Cassandra Users

While the CQL API facilitates the move, Spanner addresses the fundamental data integrity and operational constraints inherent in traditional Cassandra architectures:

Global ACID transactions: Minimize concerns regarding eventual consistency. Achieve comprehensive global ACID transactions to help ensure data integrity at any scale.
Powerful indexing: Strongly consistent secondary indexes support complex query patterns with built-in optimization and no integrity risks.
Rich SQL: Utilize a sophisticated SQL interface that supports joins and aggregations.
High reliability: Benefit from 99.99% availability in regional setups and 99.999% in multi-regional configurations.
Compliance and latency: Simplify data residency compliance with geo-partitioning, delivering low-latency local reads and writes to a global user base.
Built-in observability: Access a suite of performance metrics and charts directly in the Google Cloud console at no additional cost.

The native CQL endpoint provides a clear pathway to decouple your existing Cassandra applications and modernize them using the full power of Spanner. Let’s look at the next steps after migrating your data and applications from Cassandra to Spanner.

Tweak Spanner for your specific workload

Following your migration, here’s how to optimize your Spanner environment for your workload.

1. Optimize costs and operational efficiency

Workload Characteristics	Recommended Solution	Primary Benefit
Write-intensive traffic	Throughput-optimized writes	Up to a 6x increase in write throughput via request bundling (with minimal latency impact).
Variable or fluctuating traffic	Autoscaler	Automatically aligns capacity with demand, eliminating over-provisioning costs.
Steady-state, baseline capacity	Committed use discounts (CUDs)	Secure up to 40% savings on steady-state operational costs.
Storage-intensive workloads	Tiered storage (HDD)	Utilize cost-effective HDD storage for a significant reduction in long-term storage expenses.

2. Achieve low latencies
We continuously enhance Spanner's performance to support mission-critical, high-concurrency workloads.

Single-digit millisecond performance: Spanner consistently delivers under 5ms latency for both read and write operations.
Repeatable read isolation: This feature utilizes optimistic concurrency to reduce latency and transaction aborts in read-heavy, low-contention scenarios.
Read leases: Enable strongly consistent reads in multi-region instances without cross-region coordination, maximizing node efficiency and performance.

3. Prepare for peak traffic surges
For planned events like marketing launches or massive data ingestions, you can proactively manage capacity:

Manual split APIs: While Spanner handles data partitioning automatically, the pre-splitting capability allows you to exactly define how your database distributes data ahead of peak loads. This helps ensure immediate utilization of new capacity for stable performance.

4. Isolate operational and analytical pipelines
Prevent resource contention by isolating BI and ETL processes from core operations:

Dedicated resources: Leverage read-only replicas and directed reads to achieve workload isolation.
Advanced analytics: Spanner provides high-performance operational analytics through its columnar engine, and integrates with BigQuery via Data Boost, reverse ETL, and continuous queries.

Deconstruct your Cassandra ecosystem

Migrating from Apache Cassandra to Spanner is a strategic opportunity to decouple your architecture from a complex web of sidecar utilities. While the Cassandra-compatible API serves as the entry point, the true value lies in collapsing the operational "Cassandra tax" into a unified, managed multi-model ecosystem.

Here’s a quick guide to drastically lowering your TCO while significantly boosting performance.

1. Transition easily via connectors

Minimize initial friction by leveraging the Spanner Cassandra Adapter and connectors to port management and application layers with near-zero code modifications:

Orchestration: Maintain your existing Airflow DAGs by redirecting to the Spanner proxy.
Application frameworks: Transition from Spring Data Cassandra to Spring Data Spanner, preserving established repository patterns while unlocking a superior transaction model.
Data processing: Continue leveraging Apache Spark by connecting to Spanner’s native CQL endpoint or using the specialized Spanner Connector for Spark.

2. Retire sidecars via native integrations

Cassandra demands a "babysitting" layer that Spanner automates. Decommission legacy maintenance tools and unlock Spanner’s advanced capabilities:

Anti-entropy: Decommission Cassandra Reaper. Spanner’s Paxos-based replication manages consistency natively, removing the need for manual repair cycles.
Data protection: Replace Medusa with Spanner-native backups. Move away from fragile SSTable snapshots to reliable, Point-in-Time Recovery (PITR).
Observability: Swap complex JMX exporters for Cloud Monitoring. Focus on high-value metrics like Query Insights and Lock Statistics rather than maintenance debt.
Integrated search: Replace external Elasticsearch sidecars and complex ETL pipelines with Spanner full-text search to eliminate indices-synchronization issues.
Modern streaming: Swap legacy CDC for Spanner change streams, providing native integration with Dataflow and Kafka.
Graph analytics: Port JanusGraph to Spanner Graph to run openCypher queries directly on operational data without complex ETL processes.
Query federation: Replace Trino/Presto with BigQuery Federation via Data Boost. Join real-time transactional data with massive data lakes without impacting production I/O.

Build the future with Spanner’s multi-model advantage

Spanner is the definitive always-on database that integrates relational, key-value, graph, search, and vector search capabilities into a single, interoperable platform. By transitioning to Spanner, you eliminate the overhead of managing fragmented databases and unlock the ability to develop innovative applications on a unified data foundation. Build new applications or modernize your legacy applications to take full advantage of Spanner.

Get started today

Ready to refocus on building rather than managing? Initiate your migration today and experience the power of Spanner with your existing Cassandra Query Language.

Codelab: Gain practical experience with the native CQL endpoint.
Free trial: Explore Spanner for 90 days or start for as little as $65/month for a production-ready instance that scales without disruption.
Migration guide: Access deep technical documentation and comprehensive migration resources.