Google Cloud Next

Meet the latest Database Center, now with Gemini-powered fleet intelligence

Mon, 11 May 2026 19:00:00 +0000

Managing a modern database fleet is both a scale and cognitive problem. As database estates grow, the effort required to monitor, troubleshoot, and optimize them often outpaces teams’ capacity, who find themselves fighting database issues in isolation, buried under a mountain of fragmented signals.

We designed Database Center as a single pane of glass that provides fleet-wide visibility across all Google Cloud managed database services. At Google Cloud Next ’26, we announced an AI-native manageability interface for Database Center powered by Gemini that is designed to reason across your entire Data Cloud. In this new era of database operations, Gemini acts as an expert teammate, replacing manual scripts and error-prone workflows with AI-driven observability. In this blog, we showcase several key innovations in Database Center that you can take advantage of today.

Gemini-powered enhancements

Fleet-level intelligence with Gemini-powered analysis: Go from reactive firefighting to proactive, fleet-wide AI analysis. Gemini correlates performance shifts across your estate, highlighting patterns and providing actionable insights with an option for detailed investigations for diagnosis and remediation. These features are in preview with select customers.

Generative views (coming soon): Database Center now offers a hyper-personalized, dynamic interface driven by natural language, moving beyond standard dashboards to surface only the most relevant insights. Users will also be able to iteratively update these views.

Integrating into developer workflows: Database Center APIs are now public and integrated with the Model Context Protocol (MCP), bringing fleet management directly to tools like VS Code and Gemini CLI, as well as enabling custom third-party dashboards.

Gemini-powered chat: A new conversational interface uses natural language to interact with the entire database estate, across services like Cloud SQL, Spanner, or Bigtable, for fleet-wide exploratory questions and contextual troubleshooting, including triggering Investigations for root-cause analysis and remediation.

Gemini-backed recommendation validation (coming soon): Trust is vital for AI recommendations. New Gemini-based recommendation validations are available for specific performance optimizations. Users can trigger a Testing agent to simulate impacts on latency, IOPS, or throughput before applying changes like new indexes or machine upgrades, enabling confident, automated optimization.

Other platform enhancements

There’s more to Database Center than just Gemini. Here are the other enhancements we’ve made to the platform.

BigQuery inventory and data affiliation: BigQuery reservations and datasets inventory are now integrated into Database Center. Reservations and datasets inventory provide a single unified view across Google Cloud Databases and BigQuery. Data affiliation, meanwhile,maps data flows between transactional databases and BigQuery, helping surface hidden dependencies for faster troubleshooting.

Fleet-wide slow query analysis: Instead of hunting through logs and multiple pages, Database Center now centralizes and explains slow queries across the entire fleet, helping you prioritize your remediation efforts with AI-assisted troubleshooting. You can sort query patterns across the organization based on CPU execution time, number of instances, average rows returns, etc., and investigate the most impacted queries first.

Observability for top resources: Instantly identify the top 10 resources by critical metrics (CPU, IOPS, latency, etc.) to jumpstart investigations before they impact users.

Intelligent maintenance policies: For Cloud SQL and AlloyDB, Database Center now provides a unified, intelligent view of fleet maintenance status and compliance across all resources. You can also receive maintenance window suggestions based on your unique usage patterns, preventing downtime during peak business hours.

Reporting: Configure Database Center to generate natural language summaries of fleet health and inventory, delivered directly to your inbox so stakeholders stay informed without ever needing to log into a console.

What are Database Center customers saying

“Database Center gives our teams a comprehensive view of our Google Cloud database fleet and enables proactive risk management across security, performance, and capacity. Some of our product teams are already integrating it into their daily standups to improve monitoring and response. With Database Center’s APIs and MCP tools, we will be able to embed real-time fleet health directly into application team workflows — combining Google’s signals with our internal context like team ownership and application mapping to make insights truly actionable. This reduces context switching, accelerates recovery times, and strengthens proactive ownership across our engineering teams. We’re excited to see how the product continues to evolve.” - Bogdan Capatina, Technical Expert in Database Technologies, Ford Motor Company

Get started with Database Center

The future of database management isn't just unified — it's intelligent. With these new Database Center features and capabilities, you can:

Reduce operational overhead: Eliminate management silos and the need for expensive third-party observability software.
Enjoy faster mean time to resolution (MTTR): Resolve cross-domain issues in minutes rather than hours through Gemini-led correlation.
Scale with confidence: End-to-end lineage monitoring and automated health checks minimize blind spots, so that your AI and apps are powered by fresh, reliable data.

You can access Database Center from Google Cloud managed database services console for Cloud SQL, AlloyDB, Spanner, Bigtable, Firestore and Memorystore. Database Center is enabled by default for users with the necessary IAM permissions at the desired hierarchy. It is available at no cost, although certain premium features, including Gemini-backed fleet performance/ inventory insights, cost recommenders and natural language chat require Gemini Cloud Assist. Advanced security and compliance monitoring requires a Google Security Command Central (SCC) subscription.

Get started with Database Center today:

New Bigtable in-memory tier for sub-millisecond read latency

Thu, 07 May 2026 16:00:00 +0000

In the high-stakes world of digital infrastructure, speed isn't just a metric — it’s currency. At Google Cloud Next ‘26 we announced the Bigtable in-memory tier, a breakthrough for our fully managed cloud database service that delivers:

Sub-millisecond read latency for time-sensitive data
~10x higher point read throughput per dollar, dramatically reducing TCO
Hotspot resistance, supporting up to 120,000 queries per second on a single row without breaking a sweat.

For more information see Bigtable performance documentation. Now, let's look at the impact Bigtable in-memory tier can have on your workload performance and operational processes.

The cache-miss nightmare: A familiar story

Imagine it’s 2:00 AM. Your promotional campaign just went viral, and traffic is spiking. Your database architecture, meanwhile, is a house of cards: a primary database struggling to keep up and a separate caching layer acting as a shield.

Suddenly, you're hit with a hot key problem: everyone is trying to access the same viral content. Your cache node saturates. You’re forced to upgrade to larger nodes or add read replicas. You and your team are exhausted. Not only are you managing two different systems, maintaining a complex cache-aside logic (and praying the data in the cache stays in sync with the database), but you also need to respond to the actual incident. To do so, you overprovision CPU to handle the peak, and add more RAM so that everything fits in memory, as well as to avoid cache-aside complexity. Now you’re paying premium prices for warm data that doesn't actually need to be in memory. And while your hypothetical throughput-per-dollar looks great on paper, 90% of your resources sit idle most of the time.

Enter Bigtable’s in-memory tier

The Bigtable in-memory tier ends this cycle. By bringing data tiering across RAM, SSD, and HDD into a single, unified service with a hybrid storage architecture, we've removed the middleman.

The result? You get the raw throughput and speed of a cache with the durability and scale that Bigtable was designed for. When that viral spike hits, Bigtable automatically moves hot rows into memory to handle the load. No CPU spikes, no performance degradation. If the traffic grows, so does your Bigtable cluster by giving you more in-memory read capacity. You no longer need to overpay for idle RAM or cache nodes; Bigtable intelligently manages your data, keeping only the hot data in memory and ensuring data consistency between in-memory tier and SSD storage.

The TCO benefits are tangible, but maybe the most important part is the peace of mind that comes with it — and that’s priceless.

A peek behind the curtain

Almost every database server uses memory to give CPU fast access to latency-sensitive, frequently accessed data such as indexes and Bloom filters. You might be wondering, what makes this announcement different?

The secret lies in Remote Direct Memory Access (RDMA), a high-performance networking technology that allows computers to transfer data directly from one machine's memory to another without involving either the system's operating system or CPU. Our architecture uses RDMA to provide a high-speed, direct path to server memory, and as a result, throughput and latency of in-memory tier isn’t bound by server CPU, translating to impressive benefits. Much like Data Boost enables direct disk access for heavy workloads such as ML training, RDMA provides high-speed, direct memory access for real-time processing.

Imagine you’re running a popular social media site where 98% of users have fewer than 250 followers, while your most popular users have over 100 million. 60% of users post less than once a week, and the top 10% of users generate 80% of the content. And while a typical post receives 500 impressions, popular ones receive tens of millions.

To efficiently address this use case you will want data tiering that will likely look like this:

Memory: Content from profiles of users with large followings
SSD: Recent content, active user profiles
HDD: Older content, inactive user profiles

Luckily this is very easy to accomplish in Bigtable. Simply enable in-memory for your cluster and use a memory-enabled application profile when issuing your database requests to automatically manage the hot data lifecycle. You can also set an age-based policy to tier cold data to infrequent access. With this setup, when a piece of content is read, it gets promoted to the memory tier from persistent storage and stays there until it is evicted to make room for more recently read items. It is hands-free; even if a post from 5 years ago makes a viral comeback all of a sudden, you don’t have to worry about it.

But let’s say you want more fine-grained control of what you cache. You have a list of popular content creators and want to limit memory usage to only that small subset of their posts. Simply route the traffic for those users via the memory-enabled app profile, and for the rest of the content use an app profile that isn’t memory-enabled.

The cache-miss nightmare, revisited

Let’s rewind and replay our cache-miss scenario, but with Bigtable’s in-memory tier enabled. It’s 2:00 AM Sunday morning. Your promotional campaign just went viral, and traffic is spiking, you need to serve an additional 80K reads per second for the next hour. You don’t get paged. You wake up at 11 AM to the sound of birds chirping and enjoy a peaceful breakfast. It’s a beautiful day. The only sign that traffic spiked between 2-3 AM is that your bill shows an extra $0.40 charge.

Power laws govern distribution of requests for applications in a wide range of industries so scenarios like this are not limited to social media. For example, stock exchanges trade several thousands of securities but the top 30 most active stocks typically represent more than 40% of the total daily trading volume. At the same time, the most recent data points (last trade, ask/bid price) are requested frequently with an expectation of low latency responses, while historical data is accessed much less frequently, and has a rather forgiving latency budget. Let’s break down this example into Bigtable data tiers:

Memory: Most recent price of securities for most sought out stocks
SSD: Recent history, aggregated metrics (hourly, daily, monthly etc.)
HDD: Older data, raw events like individual trades

The list of possible use cases for this capability is long. Automated trading systems access latest prices from memory, while retail investors build their candlestick charts from data on SSD, and quants access historical data on HDD using Data Boost to backtest their models. All in one database, without interfering with each other. You can replace financial time series with telemetry data, sensor networks, digital twins and the story wouldn’t be much different.

Nor does using Bigtable’s in-memory tier interfere with other enterprise features like high availability, scaling, auditing, governance, and access controls, which typically introduce significant overhead. Achieving sub-millisecond latency despite these enterprise requirements is extremely impressive. By optimizing our clients and network, we’ve also successfully reduced p50 SSD latencies to below 2 milliseconds.

Get started with Bigtable Enterprise Plus

The Bigtable in-memory tier is available exclusively as part of the new Bigtable Enterprise Plus edition, which offers many additional features and is designed for organizations that demand the highest levels of performance, and management efficiency.

Elevate your stack to Bigtable Enterprise Plus and in-memory capabilities today so you can stop managing infrastructure and start building the future!

Learn more

Learn more about Bigtable Enterprise Plus edition and its capabilities beyond the in-memory tier. Try it out by heading over to Google Cloud console and creating new clusters upgrading existing ones.
If you’re new to Bigtable, you can now experience Google’s pioneering NoSQL database with the new Bigtable Free Trial. Get a dedicated Enterprise Edition node, 500GB storage, and a guided tour of Bigtable.
For more detailed information on getting started, technical specifications, and regional availability, visit the official Bigtable product page.

260 things we announced at Google Cloud Next '26 – a recap

Fri, 24 Apr 2026 19:00:00 +0000

Google Cloud Next ‘26 took place this week in Las Vegas, and the energy was incredible as we welcomed over 32,000 leaders, developers, and partners to explore the Agentic Era with us.

Across three keynotes, 25 spotlights, 700+ breakout sessions, 490+ sponsorships, our Showcase & Expo tech playground, and innumerable impromptu networking and collaboration moments, you could really feel the excitement as organizations move from pilot to production at a scale the world has never seen.

During our opening keynote and throughout the week, we shared our blueprint for the agentic enterprise and all the latest innovations for businesses to meet this moment. To name just a few:

- Gemini Enterprise Agent Platform, our new, comprehensive platform to build, scale, govern, and optimize agents
- Gemini Enterprise app brings the best of Google AI to every employee, for every workflow and includes new features that let every employee build with AI
- Google Cloud’s AI Infrastructure offers the industry’s widest variety of compute options and new innovations at every layer of the AI Hypercomputer, including our new 8th generation TPUs
- Agentic Defense is delivered via a cybersecurity platform that combines Google’s Threat Intelligence and Security Operations with Wiz’s Cloud and AI Security Platform to detect, prevent, and respond to threats
- Agentic Data Cloud introduces a new, AI-native architecture that allows your data to be utilized at the speed and scale required by agentic AI
- Agentic Taskforce redefines engagement and productivity with new innovations across our Gemini Enterprise for Customer Experience and Workspace products
- And so much more

That was just the tip of the iceberg from a jam-packed week of programming. So, we’ve pulled together a comprehensive summary of all 260 announcements we made across our products, partners, and customers at Next ‘26 below.

Product news

AI models, platforms and agents

As you move toward an autonomous enterprise, where agents act with the same independence and reliability as members of your team, you need a new agentic foundation that can sustain your trust. Here’s how we do it:

1. Gemini Enterprise Agent Platform: Build, scale, govern, and optimize agents. As the evolution of Vertex AI, it offers the model selection, model building, and agent building capabilities that customers love, plus new features for agent integration, DevOps, orchestration, and security.

Within Agent Platform, we announced many new features that help you move from managing individual AI tasks to delegating business outcomes with total confidence:

2. Agent Development Kit (ADK): Unlocks more powerful reasoning by organizing agents into a network of sub-agents. This new, graph-based framework allows you to define clear, reliable logic for how agents work together to solve complex problems.

3. Agent Studio: Go from building simple prompts to deploying complex agents in Agent Studio. Once you're ready for deep customization, export your logic directly into ADK to continue development in a full-code environment.

4. Agent Runtime: Delivers sub-second cold starts and allows you to provision new agents in seconds.

5. Agent Sandbox: Now available for everyone, provides a hardened environment to safely execute model-generated code and perform computer-use tasks like browser-based automation without risk to your host systems. Read more here.

6. Agent Memory Bank: Personalizes interactions by dynamically generating and curating long-term memories from conversations. Using new Memory Profiles, agents can recall high-accuracy details with low latency, so as not to lose context. Read more here.

7. Agent Sessions: Stores and manages history. With Custom Session IDs, you can use your own unique identifiers to track sessions and map them directly to your internal database and CRM records. Read more here.

8. Agent Identity: Assigns every agent a verifiable identity in the form of a unique cryptographic ID. This creates a clear, auditable trail for every action an agent takes, mapped back to defined authorization policies. Read more here.

9. Agent Registry: Maintains a central library of approved tools. It indexes every internal agent, tool, and skill, simplifying discovery and ensuring only governed, approved assets are available to your users. Read more here.

10. Agent Gateway: Manages your agent fleet from one control point, providing secure, unified connectivity between agents and tools across any environment, while enforcing consistent security policies and Model Armor protections to safeguard against prompt injection and data leakage. Read more here.

11. Agent Anomaly Detection: Detects suspicious behavior in real time and uses statistical models and an LLM-as-a-judge framework to flag unusual reasoning. It works alongside Agent Threat Detection to provide visibility into malicious activity, such as reverse shells or connections to known bad IP addresses.

12. Agent Security dashboard: Uncovers vulnerabilities automatically. Powered by Security Command Center, it unifies threat detection and risk analysis, allowing teams to map relationships between agents and models, automate asset discovery, and scan for vulnerabilities in the underlying operating system and language packages. Read more here.

13. Agent Simulation: Simulates realistic conversations and tests agents against human-like synthetic user interactions and virtualized tools in a controlled environment. Agents are automatically scored based on task success and safety across multi-step conversations. Read more here.

14. Agent Evaluation: Tracks live performance to continuously score agents against live traffic using multi-turn autoraters that can evaluate the logic of an entire conversation, not just a single response. With turnkey dashboards and Agent Observability, you can visually trace complex reasoning to debug issues as they happen. Read more here.

15. Agent Optimizer: Automatically clusters real-world failures and suggests refined system instructions to improve accuracy. Read more here.

16. Long-running agents: Execute complex business processes that can work autonomously in secure cloud sandboxes, giving agents the ability to orchestrate business logic, write code to build custom tools, and complete multi-step work — without needing constant prompting. Read more here.

Then there’s Gemini Enterprise, the end-to-end system for the agentic era, which includes the Gemini Enterprise app and the following new capabilities:

17. Agent Designer: Uses the same no-code agent designer experience of Agent Platform and lets employees build sophisticated schedule- or trigger-based agents using any enterprise connector, and providing a virtual flowchart of your agent.

18. Inbox in Gemini Enterprise: Provides a central location to monitor, guide, and help manage all of your agent activity, including long-running agents. Notifications are categorized into actionable groups like "Needs your input," "Errors," and "Completed.”

19. Projects: Create a dedicated space where the agent’s memory is confined to the files and conversations your team adds, and connect it to data sources like Google Drive, NotebookLM, and Google Group Chats.

20. Skills: Create simple shortcuts using an “@” mention for repetitive tasks like applying brand guidelines, formatting a report, or accessing specific data.

21. Canvas: An interactive editor directly within Gemini Enterprise that lets teams easily create and edit Docs and Slides, and even export to Microsoft 365 files, within the same experience.

22. Enhanced Deep Research agent: Executes complex, multi-step research plans that autonomously synthesize information from the open web and internal company data.

23. Dynamic and interactive Agent UI experiences: Move beyond simple text chat with new support for the Agent-to-UI protocol (A2UI), enabling custom agents to dynamically generate rich, native UI components directly within the Gemini Enterprise app.

24. High-code agents: Allow your engineering teams and IT leaders to build sophisticated agents with full code and publish them directly into the Gemini Enterprise app for the business to use.

25. Data Insights agent for unified analysis: Uncovers deep patterns across your entire data landscape.

26. Integrations for Bring Your Own Model Context Protocol (BYO-MCP): Allows admins to securely connect the Gemini Enterprise app directly to your own custom or third-party business tools, internal workflows, and custom-built applications.

27. Cross-platform unified intelligence: Delivers unified intelligence across a fragmented data landscape where data lives in Google Workspace, Microsoft 365, or other business applications.

You can try the Gemini Enterprise App today by signing up here.

Our incredible AI ecosystem is also growing with Agent Gallery, where customers can browse a highly vetted set of agents built by our partners.

28. Agent Gallery: Provides access to new third-party agents from partners including Accenture, Adobe, Atlassian, Deloitte, Lovable, Oracle, Palo Alto Networks, Replit, S&P Global, Salesforce, ServiceNow, Workday, and more.

AI and compute infrastructure

We announced the eighth generation Tensor Processing Units, or TPUs, which consist of:

29. TPU 8t: The training powerhouse, specifically designed for high-throughput AI workloads, delivering nearly 3x higher compute performance than the previous generation to shrink training timelines for massive models. Learn all about TPU 8t here.

30. TPU 8i: A breakthrough system for inference and reinforcement learning (RL) workloads, engineered to deliver the ultra-low latency and up to 80% better performance-per-dollar advantage for agentic workflows and Mixture of Experts (MoE) models compared to the previous generation. Go deep on TPU 8i here.

31. Native PyTorch support for TPU, or TorchTPU: With TorchTPU, you can run models on TPUs as they are, with full support for native PyTorch features like Eager Mode. Now in preview with select customers.

Whether you need a fully disconnected deployment, or want an integrated, Google-managed environment on your own hardware, Google Distributed Cloud (GDC) is the foundation of our sovereign neocloud architecture. Here’s the latest:

32. NVIDIA Blackwell GPUs: NVIDIA Blackwell (NVIDIA HGX B200) and NVIDIA Blackwell Ultra platforms (NVIDIA HGX B300) GPUs accelerate AI performance, leveraging fifth-gen NVIDIA NVLink to deliver data-center scale bandwidth directly to your environment.

33. New machine families: In addition to the N2 and N3 machine families for general-purpose workloads, GDC now supports the new A4 machine family for inference, and the memory-optimized M2 and M3 machine families for workloads that require higher memory-to-vCPU ratios.

34. Enhanced storage: GDC now supports 6PB object storage per zone — 6x the previous storage capacity. In addition, it now offers 30 IOPS/GB per zone, a 10x performance boost.

We also announced several new instances and features across our Compute Engine family:

35. A5X with NVIDIA Vera Rubin platform: A5X will be powered by NVIDIA Vera Rubin NVL72 when it becomes available later this year, including concepts from the open-source Falcon networking protocol. Learn more here.

36. Google Axion N4A: Google’s custom Arm-based Axion CPU achieves up to 2x better price-performance than comparable current-generation x86-based VMs for cost-sensitive workloads, now GA. Read more here.

37. Google Axion N4A supports GKE Agent Sandbox. GKE Agent Sandbox lets agents safely execute untrusted code and tool calls, and running on Axion N4A instances provides up to 30% better price-performance than the next leading hyperscale cloud provider. Now GA. Read more here.

38. Google Axion C4A.metal: Our first Axion bare metal instance is designed to power Android development, automotive simulation, CI/CD pipelines, security workloads, and custom hypervisors. Now in preview. Read more here.

39. C4 instances support for Intel Xeon 6 (Granite Rapids): Achieve high-performance for AI workloads with Intel AMX and native FP16 support. C4 VMs are available with Intel Xeon 6 processors across all shapes. Read more here.

40. Flexible CUDs for more VM families: Flex CUDs are now available for memory-optimized (M1-M4) and HPC-optimized (H3, H4D) VM families, as well as Cloud Run. Learn more here.

41. C4N VMs: For high-volume network applications, C4N with Titanium adapters provides nearly 400 Gbps of VM-to-VM bandwidth, and C4N with Hyperdisk Extreme provides 25 GiB/s of block storage throughput and nearly 1M IOPS. Now in preview; sign-up here.

42. M4N instances: Designed for memory-intensive databases, M4N with Hyperdisk Extreme offers 26.57 GiB of RAM per vCPU. Sign-up for the preview here.

43. Z4D instances: Optimize I/O-intensive workloads such as SQL, NoSQL, and vector databases with up to 84 TiB of high-performance local SSD directly on the node. Z4D virtual machines and bare metal instances will be in preview soon.

44. Hyperdisk Balanced improvements: Hyperdisk Balanced High Availability now offers a 4x performance improvement for high-availability databases like SQL Server or PostgreSQL by dynamically routing full-disk performance to the active VM. Learn more here.

45. Hyperdisk ML performance improvements: With 2 TiB/s of aggregate throughput (up from 1.2 TiB/s), Hyperdisk ML helps eliminate AI storage bottlenecks, with more than 200x higher throughput per disk than competitive offerings. Learn more here.

46. Hyperdisk Exapools: For large-scale training, Hyperdisk Exapools offer the highest aggregate block storage performance and capacity per AI cluster of any hyperscaler. Now GA, learn more here.

47. Z4M: For distributed parallel file systems and large-scale AI/ML workloads, Z4M provides up to 168 TiB of local SSD, up to 400 Gbps of network bandwidth, and support for RDMA. Z4M VMs and bare metal instances are expected in preview in Q3 2026.

If you want help deploying agents in your environment, you can partner with our experts for a diagnostic of your data and infrastructure by signing up here.

Agentic Data Cloud

As AI shifts from reactive intelligence to proactive action, legacy data architectures are breaking down under the agentic load. Our Agentic Data Cloud closes the gap between thinking and doing, merging analytical history with transactional power in a real-time loop, so your agents can securely act on your behalf. The Agentic Data Cloud comprises many products and platforms, starting with:

48. Knowledge Catalog: A universal context engine that maps and infers business meaning across your data estate using aggregation, enrichment, and search to help agents execute tasks accurately. Read all about Knowledge Catalog here.

49. Deep Research Agent powered by the Knowledge Catalog: Synthesizes business data, internal docs, and web research to autonomously answer your hardest, most complex questions.

50. Zero-copy federation: Provides direct access to data and context across applications, operating systems, and AI platforms — including Palantir, Salesforce Data360, SAP, ServiceNow, and Workday. Read more here.

51. LookML Agent: Autonomously reads strategy documents, spreadsheets, and reports to generate business-ready semantics, ensuring AI and analysts use the exact same enterprise definitions.

52. BigQuery measures: Redefines data consistency by embedding programmatic business logic directly into the SQL engine, ensuring calculations are reusable, accurate, and governed.

53. Data Products: Packages data assets with built-in intent, SLAs, and governance constraints. These self-contained intelligence blocks ground agents reliably in production.

54. Reusable data quality rules: Scales consistent, high-quality data governance by combining reusable, expert-defined test logic across multiple assets to ensure trusted data.

55. Structured approval workflows: Validates new business terms and data products with structured workflows, ensuring changes are verified by domain experts before reaching AI.

56. Column-level lineage: Tracks exactly where your agent's truth originated. We’ve expanded end-to-end lineage tracking to include Looker, plus column-level lineage for BigQuery and Spark.

An AI-native, cross-cloud lakehouse is another foundation element of the Agentic Data Cloud. Here’s what’s new in ours:

57. Cross-cloud Lakehouse (formerly BigLake): Breaks down data silos with a borderless lakehouse. Powered by Iceberg REST Catalog, agents seamlessly access data across AWS, Azure, and a vast partner ecosystem.

58. Lakehouse catalog federation: Enables easy discovery, analysis, and zero-copy sharing of your data seamlessly across AWS Glue, Databricks, Snowflake, and SAP.

59. Lightning Engine for Apache Spark: Accelerates your agentic data science workloads with an engine that delivers up to 2x the price-performance over proprietary market alternatives. Read more here .

60. Lakehouse federation for AlloyDB: Enables live joins between your transactional data and massive historical analytical insights without moving data or building complex ETL pipelines.

61. Managed Iceberg storage & REST Catalog: Benefits from automatic management and multi-table transactions, plus read/write interoperability across BigQuery, Spark, and OSS engines.

62. Cross-cloud caching: This intelligent cache stores cross-cloud data on the first read to slash massive egress fees and accelerate follow-on queries for AWS and Azure data.

For practitioners, we announced several products with enhanced automation:

63. Google Cloud Data Agent Kit: A portable suite of skills, MCP tools, and plugins that turns environments like VS Code and Gemini CLI into native data workspaces that autonomously orchestrate outcomes. Read more here.

64. Data Science Agent: Simply state your goals in plain English to automatically execute complex plans for loading, cleaning, and visualizing data in BigQuery notebooks. Read more here.

65. Data Engineering Agent: Autonomously builds complex Dataform pipelines from scratch while enforcing governance rules and testing to keep bad data out of your production systems. Read more here.

66. Database onboarding and observability agents: Lets agents guide you through database provisioning, or proactively monitor health and execute automated troubleshooting across your database fleet. Read more here.

67. Conversational Analytics: Empowers teams to query complex datasets using natural language and publish custom analytical agents directly into Gemini Enterprise. Read more here.

We also announced new products and features across our database portfolio:

68. Spanner Omni: This edition brings Spanner’s globally consistent, multi-model database beyond Google Cloud. Run it anywhere — across multiple clouds, on-premises, or on your laptop. Read more here.

69. Spanner Columnar Engine: Accelerates analytical queries up to 200x on live operational data using vectorized execution, without impacting your production transactional workloads.

70. Managed remote MCP servers for databases: Securely manages the infrastructure to connect AI models directly to your operational data, eliminating the burden of hosting MCP servers.

71. Vibe coding integrations with AI Studio: Agent-led workflow that generates live applications from simple text prompts, easily connecting to trusted database services.

72. Tools for Data Agents: Modular building blocks that provide ready-to-use functions, like high-accuracy text-to-SQL, giving custom agents secure and direct database access.

73. Bigtable in-memory: A new hybrid storage architecture natively integrates frequently accessed data in-memory to deliver sub-millisecond read latency for real-time apps.

74. Memorystore for Valkey 9.0: Provides a fully managed migration path from self-managed Redis, complete with enterprise-grade security features like Access Control Lists (ACLs).

75. AlloyDB AI-powered search at scale: Scales enterprise search to 10 billion vectors using Google’s ScaNN index, offering up to 6x faster queries than standard PostgreSQL.

76. New AI functions for AlloyDB: Uses optimized modes for ai.if, ai.analyze_sentiment, and ai.summarize to embed Gemini’s powerful reasoning directly into database queries.

77. Firestore full-text and geospatial search: Google’s search technology is now natively integrated into Firestore, delivering highly relevant keyword and geospatial queries.

78. Move-to-managed migrations: Move your PostgreSQL databases to Cloud SQL or AlloyDB easily and securely with Database Migration Service, now featuring automated networking, support for consolidating multiple sources in parallel into a single destination, migrating tables without primary keys, and many more capabilities.

79. Database Center with BigQuery: Monitors your entire data estate — from transactional databases to analytical warehouses — from a single management plane powered by Gemini fleet analytics.

BigQuery, now more than ever, is the linchpin of modern, agentic data analytics, and includes many new capabilities:

80. Reverse ETL for BigQuery: A one-click solution to sync heavy analytical insights from your lakehouse back to AlloyDB or Spanner for low-latency, high-QPS serving to end users. Read more here.

81. BigQuery Graph: Maps entities, relationships, and business logic directly within your data platform to anchor AI agents in a governed reality for complex reasoning tasks.

82. BigQuery fluid scaling: Executes highly variable, agentic workloads with per-second billing. This premier autoscaling model instantly adjusts resources.

83. ObjectRef: Processes unstructured data alongside structured data using SQL and Python. Build rich, multimodal context on your Knowledge Catalog without moving data.

84. AI.PARSE_DOCUMENT: Part of BigQuery AI, simplifies complex document processing with a single SQL function that automates OCR, layout parsing, and chunking using Gemini’s layout parser.

85. TabularFM model: Part of BigQuery AI, brings high-quality, zero-shot regression and classification natively to BigQuery without requiring extensive feature selection or model management.

86. Optimized mode for SQL-first AI: Part of BigQuery AI, trains task-specific models on the fly for functions like AI.CLASSIFY and AI.IF, delivering up to a 230x reduction in token consumption and cost.

87. BigQuery-native Gemma embeddings: Part of BigQuery AI, generates high-quality vector embeddings at scale on standard CPUs directly within your data warehouse to simplify enterprise AI preparation.

88. Autonomous embedding generation: Part of BigQuery AI, fully manages pipelines for unstructured data, automatically keeping vector indexes in sync asynchronously as new data is ingested at BigQuery scale.

89. BigQuery hybrid search: Part of BigQuery AI, unifies semantic and full-text search into a single powerful function to deliver superior retrieval precision for RAG and complex exploration tasks.

90. Python UDF: Part of BigQuery AI, enrich, transform, or clean data using fully managed Python scalar functions that autoscale serverlessly to millions of rows directly in BigQuery.

91. Connected Sheets with TimesFM: Part of BigQuery AI, brings the scale of BigQuery to Google Sheets, featuring state-of-the-art zero-shot forecasting with the TimesFM model and robust anomaly detection.

92. Geospatial analytics datasets: Part of BigQuery AI, combines native Google Earth AI and roads management data directly with your enterprise data for deep, real-world agentic analysis without wrangling.

93. Proactive agentic workflows: With BigQuery, agents can go beyond basic Q&A to detect metric shifts, perform root-cause analysis, and deliver scheduled research briefings directly to your inbox.

94. BigQuery Agent Analytics plugins: Logs detailed interactions from frameworks like ADK and LangGraph directly to BigQuery to audit, troubleshoot, and optimize long-term agent performance.

95. BigQuery Studio productivity tools: Brings a contextually aware assistant, SQL and Visualization Cells, Files Explorer, and Git workflows directly into BigQuery.

96. Advanced runtime, small query, and history-based optimizations: These BigQuery optimizations have accelerated performance by 35% year-over-year while cutting processing costs.

97. New BigQuery workload management features: Granular cost controls arrive with reservation groups, dynamic assignments, and project-level slot controls, simplified by rules-based management.

98. Colab Data Apps: Transforms your notebook analyses into shareable, fully managed interactive Python applications that business teams can access directly from Data Studio.

With Looker Business Intelligence, you get not just agentic AI, but agentic BI. Here’s the latest:

99. Data Studio (formerly Looker Studio): Available in Free and Pro versions, this data visualization engine now also provides a unified interface to browse data sources, apps, and agents. Read more here.

100. Looker Dashboard Agents: Transforms static views into interactive hubs. Ask natural language questions within dashboards for instant summaries and context-aware, governed answers.

101. Agentic Workflows in Looker: Moves beyond reactive queries to proactive results. Agents monitor critical metrics for irregularities and identify hidden correlations so you can instantly address business shifts.

102. Embedded conversational experiences in Looker: Leverages Looker's API and SDK to integrate Gemini-powered, multi-turn conversational experiences into custom data applications and internal workflows.

103. Looker self-service explores: Blends personal files with enterprise data using an updated AI-Explore experience that automatically generates LookML, giving you the agility of a spreadsheet with enterprise governance.

104. Looker AI assistants: Simplifies BI with Visualization, Expression, and Insight Assistants. Use natural language to create charts, generate custom filters, and automatically summarize key trends.

105. Open BI and MCP: Empowers external AI agents to autonomously query and analyze data. We offer both open-source MCP via the MCP Toolbox and an all-new Google-managed MCP server that’s native to Looker for ease of management.

106. Continuous Integration: In Looker, automates SQL validation and content testing within your development workflow to identify potentially breaking changes before they reach production, for reliable model updates.

Application development, management and runtime platforms

Cloud Run, Google’s fully managed application platform, now supports:

107. Full-stack apps in Google AI Studio: Build and deploy apps vibe-coded in AI Studio, including a Firestore database and user authentication, to Cloud Run, now GA.

108. Integration with Gemini Enterprise Agent Platform: Through its integration with Cloud Run, Agent Platform helps you transition agents from experimental environments into fully managed, production-grade systems. Now in preview with select customers.

109. Fully managed remote MCP server: Enable easy deployments from developers and agents, now GA.

110. NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs: Run high-end inference on Cloud Run — up to 70B+ parameter models — without having to manage any underlying GPU infrastructure, and scaling to zero when the resource is not in use.

111. Billing caps: Define your maximum spend per month, after which point, Cloud Run resources are de-activated. Coming soon.

112. Cloud Run instances: Host long-running background jobs, in preview to select customers.

113. Cloud Run sandboxes: Spin up an ephemeral sandbox that’s strictly isolated from your agent code using a built-in sandbox tool, coming soon.

114. SSH: Gain secure shell access (SSH) directly into a running Cloud Run container, now in preview with select customers.

115. Cloud Run service bindings: Service-to-service communication for scalable microservices architectures, coming soon.

116. Ephemeral disk: Now in preview, ephemeral disk storage is created when an instance starts and deleted when it stops.

And for mobile and web developers, we added new capabilities to Firebase, our premier platform for client developers:

117. Firebase SQL Connect: Firebase Data Connect is now Firebase SQL Connect, letting you connect your mobile and web client apps to the power of Cloud SQL for PostgreSQL, now with support for realtime syncing, offline cache, and native SQL— all available via a no-cost trial.

118. Firebase integration with Application Design Center: The new Firebase integration with Application Design Center (ADC) offers a unified model for deploying and managing the resources used by your mobile and web client apps, such as Firestore, App Check, Authentication, and Firebase AI Logic, in the same way you manage the rest of your Google Cloud infrastructure.

119. Firestore Enterprise edition: Introducing native full-text search, geospatial queries, relational-style JOIN using subqueries, and a robust data manipulation language (DML) as part of the new query engine in Firestore Enterprise.

120. Firebase AI Logic: Build generative AI features directly into your mobile and web client apps with no server-side setup. New features, like auto function calling and JSON schema mapping, help reduce technical overhead and give you more control to optimize your resources and costs. Read more here.

121. Firebase Phone Number Verification: Helps onboard new users to your app by retrieving the phone number assigned to the SIM in a device over cellular or WiFi. Now supported by 10+ carriers across six regions.

Then, there’s Google Kubernetes Engine, or GKE, which has emerged as the platform of choice for complex, demanding AI and agentic applications. Here’s what’s new in GKE:

122. GKE Agent Sandbox: A GKE add-on based on the open-source Agent Sandbox controller project that manages isolated, stateful, and single-replica workloads on GKE, optimized for AI agent runtimes. Learn more here.

123. GKE hypercluster: A single, conformant GKE control plane to manage millions of accelerators across Google Cloud regions, in private GA.

124. Predictive latency boost in GKE Inference Gateway: Reduce time-to-first-token latency by up 70% by replacing heuristic guesswork with real-time capacity-aware routing — no manual tuning required. In preview. Learn more here.

125. Automatic KV Cache storage tiering across RAM, Local SSD, and Google Cloud Storage/Lustre, solving long-context memory bottlenecks. Learn more here.

126. RL Scheduler: Solve for the "straggler effect" and inter-batch tail latency, maximizing throughput via intelligent routing. More here.

127. RL Sandbox: Kernel-level isolation for tool-calling and reward evaluation with millisecond-scale provisioning.

128. RL Observability and Reliability dashboards: The deep visibility you need to troubleshoot and optimize the entire RL loop instantly, out of the box. See more here.

129. Scaling on custom metrics: Support for intent-based autoscaling on triggers besides CPU and memory. Learn more here.

130. Accelerated node and pod startup: GKE nodes now start up to 4x faster, while pod startup times have been slashed by up to 80%. Learn more here.

131. Rapid model loading: Leveraging the run:AI Model Streamer and Rapid Cache in Google Cloud Storage, models now load 5x faster, removing a traditional storage bottleneck. Learn more here.

Developers and operators, get ready for the next generation of Gemini Cloud Assist, our AI-assisted cloud operations platform. Be on the lookout for:

132. Redesigned Application Design Center: Using natural language and the power of Gemini, reduce the time from architecture design to deployment of new or existing multi-resource environments.

133. Support for gcloud, kubectl, and Terraform: Automate infrastructure operations while using proactive multi-turn agents to troubleshoot and resolve incidents.

134. Proactive cost anomalies detection: A new FinOps agent analyzes spending spikes and generates granular cost reports on demand.

135. MCP servers: Get help wherever you work, including in your integrated development environment (IDE), CLI, or third-party tools with MCP servers for Gemini Cloud Assist’s design, operations, troubleshooting and optimization capabilities. Read more here.

Whatever your preferred Google Cloud management interface, you’ll find a variety of new features:

136. App-centric visibility for Unified Maintenance: Gives a consolidated view of maintenance events across your entire application landscape, moving from resource-level to business-level visibility. Learn more here.

137. App Lifecycle Manager: Now in preview, App Lifecycle Manager lets you design, deploy and operate Apps at scale across tenants with automated Day 2 ops and visibility, while feature flags decouple your feature releases from code deployments. Learn more here.

138. FinOps Explainability agent: Autonomously investigates the drivers of your AI-related cloud costs. This is in addition to new FinOps tooling which provides commercial auditability. Get started here.

139. Spend Caps: In private preview, FinOps and DevOps managers can set budgets and enforce cost boundaries at the project level for Google AI Studio, Gemini Enterprise Agent Platform, Cloud Run, Cloud Run Functions, and Maps. These caps alert and ultimately pause API traffic once your set budget is reached. Sign up for the preview here.

Customers

The real impact of these new capabilities is in the value they bring to our customers. We updated our running list of AI customer stories with more than 300 new entries, bringing the total to 1,302!

We also talked about several new customer use cases:

140. Capcom worked with Google Cloud to build agents that improve the gameplay experience, freeing developers for creative work. These specialized AI agents — including visual inspection, predictive and institutional knowledge agents — support the highly complex process of playtesting.

141. Citi Wealth is launching Citi Sky, an always-on AI-powered agent for customers to access market insights, act on opportunities, and engage with their advisors. Citi Sky is conversational and multilingual for a more intuitive, responsive and personalized wealth experience.

142. Citadel Securities built a scalable, cloud-based research environment that uses TPUs to run AI workloads up to 4x faster with 30% lower costs. Work that used to take days is now executed in minutes.

143. Deutsche Telekom announced MINDR, a multi-agent system built on Gemini models that enables autonomous, self-healing network operations across multiple domains to resolve network issues before they impact customers. MINDR reduces event management times by more than 95%.

144. GE Appliances is deploying more than 800 AI agents with Gemini Enterprise across its manufacturing, logistics, and supply chain operations.

145. Highmark Health scaled employee impact with Sidekick, an AI assistant that has delivered $27.9 million in value in 2025 alone by automating research protocols and providing "AI search proxies" for internal teams.

146. Home Depot created Magic Apron, a sophisticated digital agent built on Gemini Enterprise for Customer Experience that brings the brand’s "Orange Apron" expertise to every customer, and a new AI-powered phone agent that allows customers to skip complicated phone menus — identifying needs in under 10 seconds.

147. Macy’s created “Ask Macy’s,” a new digital shopping concierge and a leap forward in multimodal, agentic AI shopping powered by Gemini Enterprise for Customer Experience. Far beyond a chat bot, this multimodal agent is capable of handling text and images, and it also has a unique virtual try-on feature.

148. Mars selected Gemini Enterprise as the primary AI operating system for its global workforce. The company is providing its associates with a suite of agentic capabilities, including sophisticated AI assistants designed to handle complex, multistep tasks that can help them work more efficiently and creatively.

149. Merck will deploy a new agentic platform across their research and development, manufacturing, commercial and corporate operations. Core to the work will be digitizing data and boosting productivity for Merck’s 75,000 employees worldwide.

150. Papa John’s is the first user of the Food Ordering agent in Gemini Enterprise for Customer Experience. The agent features a unified voice-and-text AI ordering system, , while integration with the Omnichannel Gateway helps the agent remember customer preferences and past orders and get food to the door faster.

151. Tata Steel is rapidly scaling autonomous capabilities across its vast global organization, deploying a fleet of more than 300 specialized AI agents in just nine months to drive efficiency and precision across its global operations. These include a low-code platform for easy agent deployment and a digital assistant for informed decision-making.

152. Unilever is designing and deploying agents at scale — safely, securely, and globally — to transform how they serve billions of customers. Using Agent Platform, Unilever has developed a multi-agentic solution to help procurement teams make quicker and smarter buying decisions.

153. Virgin Voyages’ new Rovey agent is a personal concierge for the high seas, powered by Gemini Enterprise and Google Distributed Cloud. Rovey makes sure the crew has the latest, most on-brand information 24/7, and even on ships with limited connectivity at sea, Google Distributed Cloud keeps everything connected.

154. Vodafone Business is launching a managed detection and response service enabled by Google Security Operations, and the Vodafone Business AI Concierge with Google Gemini. The detection and response service will further protect Vodafone’s customers from increasingly frequent and sophisticated cyber threats.

Powering many of these organization’s new agents is Gemini Enterprise for Customer Experience, which comes with new agents and capabilities:

155. Shopping agent and Food Ordering agent: Bring new conversational sales and ordering capabilities direct to businesses and third-party chat and digital interfaces.

156. Omnichannel Gateway: Helps agents maintain context across web, mobile, and voice, so a company’s agents can offer more personalized service.

157. Agent Assist: For the most complex customer service situations, coaches employees to deliver fast and more accurate answers to customer questions with organizational data readily available through generative AI grounding.

Networking

We also announced a variety of enhancements to the broader Cross-Cloud Network family:

158. Virgo Network: A scale-out fabric for modern AI workloads. Built on high-radix switches that allow more ports per switch, it delivers massive bisection bandwidth and deterministic low latency for distributed training and serving workloads. Check out the deep dive.

159. Agent Gateway partnerships: Allows customers to integrate their trusted vendors to help secure agentic communications. Includes partner solutions from Broadcom, Check Point, Cisco, CrowdStrike, Exabeam, F5, Netskope, Okta, Palo Alto Networks, Ping Identity, Saviynt, Silverfort, Thales, and Zscaler. In preview soon.

160. Ambient networking: This new integrated data plane for GKE and Cloud Run provides service discovery, zero-trust access, and traffic management without the need for complex and resource-heavy sidecar proxies. Now in preview.

161. Service bindings: Enabled by ambient networking, service bindings automatically establish service-to-service connectivity between agentic applications and services.

162. Network Services Monitoring: Also enabled by ambient networking, the preview of Network Services Monitoring bridges application and network observability gaps for faster root-cause analysis and simplified troubleshooting.

163. Multi-region support in GKE Inference Gateway: Scale inference services across regions, for cross-regional failover, optimized utilization, and reduced global latency. In preview.

164. GKE Inference Gateway disaggregated serving: Choose between vLLM and SGLang for model serving by leveraging llm-d’s SGLang support. Now GA.

165. Accelerator network profiles: Set up accelerator-equipped GKE node pools with DRANET, and improve the bandwidth for distributed AI/ML workloads by up to 60%. Now GA.

166. Cloud Interconnect 400 Gbps circuits: Use up to 3.2 Tbps in a single connection. Now GA.

167. Partner Cross-Cloud Interconnect: GA for AWS; in preview soon for CoreWeave and Lumen.

168. Ultra Low Latency Solution for financial exchanges: In partnership with CME Group, this fully managed cloud environment will allow CME Group and its clients to migrate core trading systems to Google Cloud. Now in preview.

169. Cloud Network Insights: Network performance monitoring (NPM) and digital experience monitoring (DEM) enabled by Broadcom AppNeta to reduce the mean time to detect and mitigate network-related agent, application, and API issues. Now in preview.

170. Private Service Connect published services: Enable secure private global access to your managed services through 40+ Google and third-party published services. Read more here.

171. Private Service Connect endpoint-based security: Use granular authorization policies for producer-to-consumer service communications, now in preview.

172. Cloud Number Registry: Now in preview, this agentic IPAM solution finds free IP ranges, tracks utilization, and allocates resources. It also integrates with Infoblox Universal DDI for Cross-Cloud Network IPAM discovery and enforcement.

173. Hybrid Subnets: Migrate legacy workloads from on-premises to a VPC without needing to change hard-coded IP addresses. GA.

174. Cloud NAT: Connect IPv6-only workloads to private IPv4 destinations using the combined power of DNS64 and private NAT64. In preview soon.

175. Granular networking charge visibility: Cost Explorer and the new App Optimize API now provide attribution of associated data transfer costs to the originating resources for Google Cloud products, in preview soon.

176. Gemini Cloud Assist network security agent: Streamlines network security operations with policy generation, recommendations, and impact analysis. In preview soon.

177. Gemini Cloud Assist network agent: Optimizes workload placement for performance and reliability, Private Service Connect troubleshooting, and also provides advanced cost estimation for observability services. In preview soon.

178. Network observability MCP tools and agent skills: Partners can build agents to leverage connectivity tests, query VPC Flow Logs in natural language. In preview.

As part of Cross-Cloud Network, new capabilities in Global Front End include:

179. Global Front End Enterprise for simplified consumption: Combining capabilities from global Cloud Load Balancing, Google Cloud Armor, Cloud CDN, and Service Extensions with up to 15% lower TCO. In preview soon. Learn more here.

180. Post quantum cryptography (PQC): Helps secure your workloads with industry-standard algorithms that provide a layered defense against both classical and quantum adversaries. Learn more here.

181. Google tag gateway: This allows advertisers to serve tags from their own domain, which can significantly improve the accuracy and resilience of measurement signals. GA soon.

182. Image optimization in Cloud CDN: Deliver content that best fits your end users’ screens and save on bandwidth costs (in preview soon).

183. Cloud CDN GKE Gateway support: Enable and manage caching services using GKE APIs, now GA.

Cloud WAN is a fully managed, reliable global backbone to connect your enterprise, and includes the following new capabilities:

184. Expanded geographic reach: Network Connectivity Center’s site-to-site data transfer is now available in over 25 countries. Read more here.

185. NCC Gateway: Use third-party secure service edge (SSE) integrations from Palo Alto Networks (GA soon) and Symantec (preview).

186. Expanded Verified Peering Provider program: Get highly reliable internet connectivity to Google through 175+ providers worldwide. Read more here.

187. Last mile connectivity: Provision site-to-cloud private connectivity in minutes with preferred partners from the Google Cloud console. In preview soon.

We added new capabilities in our AI-powered network security offerings to protect against evolving threats.

188. Cloud Armor managed rules: Available in preview, and powered by Thales Imperva, detects Layer 7 application attacks and zero-day vulnerabilities.

189. Cloud Armor adaptive protection for Network Load Balancers and VMs: Advanced machine learning to L3/L4 traffic, to detect and mitigate volumetric DDoS attacks. In preview soon.

190. Cloud Armor visual rule builder: Makes custom rule creation easier with a simplified user experience, in preview soon. Read more here.

191. Cloud NGFW Internal Application and proxy Network Load Balancer support: Helps enforce consistent, service-centric security for abstracted services like GKE, Cloud Run, and Private Service Connect traffic. In preview.

192. Cloud NGFW project-level policies: Creates and manages Cloud NGFW endpoints, security profiles, and security profile groups at the project level. In preview soon.

193. Cloud NGFW Advanced Malware Sandbox: In preview later this year, helps defend against highly evasive zero-day threats. This capability is powered by Palo Alto Networks Advanced Wildfire, trained on data from more than 70,000 Palo Alto Networks customers to stop 99% of known and unknown malware.

We added new functionality to Network Connectivity Center (NCC), our hub for managing global connectivity:

194. Support for static routes: By using an internal load balancer as the next hop, you can integrate Secure Web Proxy and third-party network security virtual appliances. Now GA. Learn more here.

195. Support for privately used public IP (PUPI): Allows the exchange of PUPI IPv4 addresses with VPC spokes and producer VPC spokes, now GA. Learn more here.

Partners

A huge part of Google Cloud’s success is built on our partner ecosystem, which helps conceive, build, and implement solutions for our joint customers. To strengthen this partner partnership, we made a number of important announcements, most of which are detailed here.

196. $750 million innovation fund: Investing in partners with new models, tools, engineering support, and incentives to accelerate agent development and deployment globally for every business process, function, and industry.

197. Sell partner agents in Gemini Enterprise: The new Agent Marketplace with the Agent Gallery of Gemini Enterprise allows partners, startups, and more to sell their agents directly to other enterprise users.

198. 70-plus partner-built agents in Agent Marketplace: We’re launching the Agent Marketplace with more than 70 pre-built agents. Discover and manage specialized AI from partners like Accenture, Adobe, Atlassian, Deloitte, and more. Read more here.

199. Partnering on forward deployed engineers: Google Cloud will make its engineering talent available for the customers of select partners, like Accenture, Deloitte, and McKinsey, who are also establishing advanced engineering practices.

200. Deeper model integration: Palantir joins many of the leading SaaS and AI platforms already featuring Gemini access, while Box, DocuSign, Oracle, Salesforce, and SAP have added Gemini features in more of their agents and tools.

201. AI-powered features in Google Cloud Partner Network: A new partner agent and agentic features in the Earnings Hub and Partner Finder help partners amplify their impact for joint customers.

202. New Unified Data Foundation with SAP: Integrates SAP and non-SAP data into a bi-directional source of truth, helping power AI workflows and deliver data-driven value. Zero-copy data sharing opens up access to outside platforms securely.

203. More SAP news: We also unveiled new Gemini features in SAP Engagement Cloud, solution accelerators in Cortex Framework, and support for SAP Sovereign applications. Read all our SAP announcements here.

204. Oracle AI Database@Google Cloud expansion: Our recently launched Oracle data connectors are expanding to 20 regions, offer new AI capabilities, feature Oracle GoldenGate support for near real-time data replication to BigQuery and Knowledge Catalog integration, and more. Read more here.

205. Partners of the year: Celebrating hundreds of partners across specialties, industries, and geographies. See them all here.

Security

From security operations, to fraud detection, to cryptography and key management, there was a ton of security news at Next ‘26.

206. Threat hunting agent in Google Security Operations: Helps teams proactively hunt for novel attack patterns and stealthy adversary behaviors that bypass traditional defenses.

207. Detection Engineering agent Google Security Operations: Identifies coverage gaps and creates new detections for threat scenarios, reducing toil and transforming detection creation from a manual craft into an automated science.

208. Third-Party Context agent Google Security Operations: Enriches your workflows with contextual data from third-party content.

209. Google Security Operations: A robust cohort of new partner integrations for Google Security Operations has been designed to deliver high-fidelity security workflows right out of the box. Our latest participating Google Cloud Security integration ecosystem partners include Darktrace, Gigamon, and SAP. Read more here.

210. Model Armor: Now integrated with Agent Gateway, Agent Runtime, and Langchain available in preview, and Firebase, generally available, to help developers add inline enforcement and sanitization of agent traffic and interactions without the need to change code. These integrations expand Model Armor's protection against runtime risks such as prompt injections, tool poisoning, and sensitive data leakage across Google Cloud services and our AI portfolio. Learn more here.

211. Google Cloud Fraud Defense: Now generally available, we’re evolving reCAPTCHA into a comprehensive platform designed to secure the agentic web by discerning the legitimacy and authorization of bots, humans, and agents. Using the same scale and signals that protect Google’s own ecosystem, Fraud Defense will soon offer in preview agent-specific capabilities for human users and AI agents that can help secure the digital commerce journey, from account creation and login to payment and checkout. Learn more here.

212. Chrome Enterprise: Now in preview, AI-aware extension threat detections can surface advanced extension telemetry that helps security teams detect and respond to anomalous AI agent activity. Learn more here.

213. Chrome Enterprise: Generally available soon, new shadow AI reporting can help you gain visibility into the shadow AI landscape by flagging employee use of unsanctioned web-based AI and SaaS applications. Learn more here.

214. Identity and Access Management: To help achieve least privilege quickly and simply, we’ve streamlined our predefined roles catalog with easy-to-use administrator, editor, and viewer roles, such as the IAM role picker and the ability to re-authenticate sensitive actions.

215. Confidential Computing support for G4 VMs: In partnership with NVIDIA, Confidential Computing support for G4 VMs is available in preview globally, featuring NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs on Compute Engine Confidential G4 VMs, to help strengthen confidentiality and integrity for a wide spectrum of sensitive AI workloads. Learn more here.

216. C4 Confidential VMs: In partnership with Intel, we’re also introducing the preview of C4 Confidential VMs, bringing Intel TDX to 6th Gen Xeon processors to help protect diverse AI and analytics workloads while providing industry-leading compute density and performance.

217. Cloud Key Management Services (KMS): Available in preview, the new Confidential External Key Manager (cEKM) can give you the flexibility to host and protect external keys in any region and maintain verifiable control in a confidential environment.

218. Post-quantum cryptography (PQC): Available in preview, KMS Quantum Safe Key Imports can help you bring your own keys with quantum-safe algorithms.

219. Secret Manager: Generally available, the native integration of our Secret Manager with Agent Development Kit can help prevent password leaks and mitigate prompt injection risks.

Google has completed its acquisition of Wiz. By bringing two industry leaders together, we are building a comprehensive, AI-ready cybersecurity platform designed to protect your organization across all your cloud environments.

Wiz is constantly expanding and deepening your ability to adopt AI quickly and securely, with the following new capabilities:

220. Support for agent studios: Now supports Databricks as well as new agent studios like AWS Agentcore, Gemini Enterprise Agent Platform, Microsoft Azure Copilot Studio, and Salesforce Agentforce to help customers gain visibility however their teams choose to build. Read more here.

221. Security detections updates from Wiz Defend: We updated how we integrate security detections from Wiz Defend with Google Security Operations and Mandiant Threat Defense to help analysts more easily configure automatic threat information forwarding.

222. Vibe coding integration: Generally available in May, Wiz can help secure vibe-coded applications. This new integration runs Wiz security scanning directly inside the Lovable platform so vulnerabilities, secrets, and misconfigurations caught by Wiz are surfaced in Lovable's built-in security view.

223. Inline AI security hooks: Integrates Wiz directly into integrated development environments (IDEs) and agent workflows to evaluate prompts and scan AI-generated output instantly, injecting security guardrails before code is committed.

224. Wiz Skills: Help agent-based remediation by equipping coding agents and AI-native IDEs with full code-to-cloud context and validated attack surface findings from the Wiz Security Graph. These capabilities enable teams to trigger automated, agent-driven remediation workflows locally from the developer's individual IDE and globally at the repository and pull request level in your version control system.

225. AI-Bill of Materials: To help eliminate shadow AI, Wiz’s dynamic AI-Bill of Materials automatically inventories all AI frameworks, models, and IDE extensions across your environment. This helps provide visibility into what is writing code across your stack, allowing you to track sanctioned corporate tools like Gemini Code Assist and GitHub Copilot while simultaneously uncovering unapproved shadow AI plugins.

Storage

Enhancements to Google Cloud’s storage offerings span high-performance storage infrastructure, smart storage, new management capabilities, and expanded ecosystem offerings.

226. Rapid Bucket: As part of the new Cloud Storage Rapid family of high-performance object storage, Rapid Bucket delivers more than 15 TB/s of bandwidth, 20 million requests per second, and sub-millisecond latency in a single zonal bucket. Now GA; read more here.

227. Rapid Cache: Formerly Anywhere Cache, Rapid Cache accelerates bandwidth for bursty workloads, delivering an aggregate read throughput of 2.5 TB/s for existing buckets, with no code changes. Learn more here.

228. Rapid Cache ingest-on-write: Get up to 2.2x faster checkpoint restores, allowing training clusters to recover faster from interruptions. Learn more.

229. Increased throughput for Google Cloud Managed Lustre: With up to 10 TB/s of throughput, Managed Lustre writes and restores checkpoints 2.6x faster than other Google Cloud storage solutions. Learn more.

230. Managed Lustre Dynamic tier: Low-latency performance for intense AI workloads like training and checkpointing, offered at $0.06/GB-month.

231. Managed Lustre support for new TPUDirect and RDMA: Minimize latency by allowing data to bypass the host, moving directly to the accelerators. Learn more here.

232. Smart storage object context: Now GA, this metadata substrate adds structured, mutable, IAM-governed context to every object. Learn more.

233. Smart storage automated annotations: Unlock unstructured data with automated metadata annotation.

234. Cloud Storage MCP server: Read, write, and analyze Cloud Storage data using the standard MCP protocol. Learn more.

235. Storage Intelligence zero-configuration dashboards: Surface cost anomalies and integrate Security Command Center’s Data Security Posture Management (DSPM) data governance feature, to detect security vulnerabilities across Cloud Storage. Now in preview.

236. Storage Intelligence object events and bucket activity tables: Available in Insights Datasets, use these new tables to, for example, optimize bucket placement based on egress patterns, or quickly troubleshoot 429 errors.

237. Enhanced batch operations in Storage Intelligence: New change ACL and storage class operations make it simpler to act on billions of objects, and adds support for multi-bucket operations.

238. Google Cloud NetApp Volumes: With new Flex Unified, provision both block (iSCSI, NVMe/TCP) and file (NFS/SMB) on the same storage pool. New ONTAP-mode lets you bring your existing automation (Terraform, Ansible) and ONTAP APIs directly to NetApp Volumes.

239. Filestore for GKE: Developers building AI workloads on GKE can start small, with shares as small as 100 GiB, and scale capacity and IOPS independently. Tighter integration with Colossus provides more scale and enterprise capabilities.

240. Google Cloud Backup and DR agentic AI capabilities: Let these agents autonomously audit your backup estate and remediate coverage gaps, with new GA integrations for AlloyDB and Filestore.

Startups

Startups are building on Google Cloud for our models, purpose-built infrastructure, and unique security solutions. You can read more about some of the exciting new ones, and new startup programs, here.

241. Notable startups using Google Cloud: We announced dozens of new startups using our services or expanding with us, including Lovable, Notion, Proximal Health, Gamma, Mavvrik, Insilica, and Parrallel.

242. Access to NVIDIA hardware: A number of leading startups use Google Cloud for training and inference workloads on NVIDIA chips, including Thinking Machines, Interfact, Nucleus AI, Reflection, and Neuraldefend.

243. Global Gemini Startup Forum: This June, we’re hosting a two-day, in-depth gathering of founders from seed to Series A startups; attendees will engage with Google leaders and engineers, build and scale products, and share feedback.

244. Google for Startups AI Agents Challenge: Open to anyone, this global competition runs for six weeks and gives teams $500 in credits and tools — like the new Agent Platform — to build autonomous systems and compete for a $90,000 prize pool.

245. New gen media report: Our Future of AI: Perspectives on generative media for Startups report compiles exclusive interviews with eight startup founders, investors, and Googlers.

Workspace

Google Workspace with Gemini is where work happens for over 3 billion users and more than 13 million customers — from solopreneurs to small businesses to the Fortune 500. We introduced new innovations that transform how people work with AI, including:

246. Workspace Intelligence: Unified, real-time understanding to power agentic work, by understanding complex semantic relationships within Workspace apps (such as Docs, Slides, or Gmail), active projects, collaborators, and your organization's domain knowledge. See it here.

247. Ask Gemini in Chat: Acts as a unified command line for all of your work; simply state your goal and Gemini can complete tasks, generate documents or slides, or find files using just a description.

248. New Gemini capabilities in Sheets, Docs, and Slides: Build or edit entire spreadsheets using natural language, collaborate with Gemini in Docs to create infographics grounded in your data or edit based on comment feedback, and create full, editable slide decks in one shot.

249. AI Inbox and AI Overviews in Gmail: Get a clear, streamlined look at what’s most important in your inbox and use AI Overviews in Gmail search to get answers without digging.

250. New Gemini capabilities in Drive: Use AI Overviews and Ask Gemini to find information fast and get detailed insights about all of your files, and use Drive Projects to organize your files and emails, ensuring your colleagues and Gemini have the full context.

251. Third-party data in Sheets: Import third-party data from platforms like HubSpot and Salesforce to Gemini in Sheets, and create interactive visualizations — such as dashboards and kanban boards — directly within the new Sheets canvas.

252. Workspace skills: Organizations can now deploy agentic automation across workflows using skills. Collaboratively build and share skills in Workspace Studio, then invoke them wherever you use Gemini in Workspace.

253. Custom Avatars in Vids: Add branding elements to custom avatars — everything from a company logo on a shirt to a fully branded backdrop. And you can soon add avatars when you convert a presentation into a video.

254. Help from Gemini in every meeting: We’re expanding Take Notes for Me in Google Meet to capture automated meeting summaries and action items for any meeting, whether it’s in-person, or hosted on another provider like Zoom or Teams.

255. Workspace capabilities in the Gemini Enterprise app: Coming to private preview, customers will soon also be able to schedule Google Calendar meetings and seamlessly create, view, and edit Docs and Slides directly from canvas mode in the Gemini Enterprise app.

256. Auto browse with Gemini in Chrome: Workspace customers in the US can now turn on the new Gemini auto browse capabilities in Chrome Enterprise. Learn more here.

257. Workspace MCP Server: The new Workspace MCP Server, available in preview, lets developers bring advanced Workspace capabilities such as synthesizing Drive documents and drafting Gmail responses into their AI applications and agents. Learn more here.

258. Simplified agent governance: New agent governance controls, such as AI control center, agent management, and Workspace Studio controls, help you monitor, control, and audit agent access to your data in Workspace to reduce indirect prompt injection, oversharing, and data loss risks.

259. New sovereign controls and client-side encryption: To meet regulatory requirements, you can lock data processing and storage to the US and EU, with more countries supported in the future, such as Germany and India. And for your most sensitive data, client-side encryption lets you deny access to any agent and any entity, including Google.

260. Improved migration and interoperability features: Migrating your entire organization from Microsoft 365 to Workspace with a new cloud service that lets you easily move emails, files, and conversations. Improved interoperability features allow collaboration with customers and partners who use Office apps.

A big thank you to all of our customers, partners, and attendees who joined us on the ground or tuned in virtually. We’ll see you next year for Cloud Next ‘27 on April 13-15 in Las Vegas!

Day 2 at Google Cloud Next: A marathon developer keynote

Thu, 23 Apr 2026 13:00:00 +0000

At Google Cloud, every day is Developer Day, but none so much as day 2 of Google Cloud Next, when we hold the developer keynote. This year’s topic? An in-depth look at Gemini Enterprise Agent Platform. This year’s theme? Planning a marathon for 10,000 participants through the Las Vegas Strip.

OK, let’s run with it.

Gemini Enterprise Agent Platform: A warm up

As the evolution of Vertex AI, Agent Platform “allows you to build autonomous agents that proactively help users — and complete tasks independently,” said Brad Calder, President, GCP and SRE. The platform does so with a whole suite of tools and capabilities to build, scale, govern, and optimize your agents.

Brad then passed the baton to keynote emcees Richard Seroter, Chief Evangelist, and Emma Twersky, Developer Relations Engineer, for an in-depth run-through of the agentic marathon simulator.

The system uses three main agents: A planner to determine routes; an evaluator that assesses routes based on business and community requirements; and a simulator that takes the route, adding actors and randomized behaviors to test the impact on the city,

This, Emma said, turns out to be “a great example of how agents can help us plan, simulate, and think about solving a really big challenge.”

First off the blocks: Building the agent

Mofi Rahman, Developer Relations Engineer, came onstage to demo how the Agent Development Kit (ADK), Google Cloud remote Model Context Protocol (MCP) servers, and Agent Runtime provide the planner agent with the Instructions, Skills, and Tools it needs to improve the initial agent. By the end of the demo, the simulator had generated a route.

“The simulated route looks beautiful,” said Mofi. “Looks like the runners are going to get an amazing view of the entire Las Vegas Strip.”

An agent to evaluate the agent

Next, Ivan Nardini, Developer Relations Engineer, and Casey West, Architecture Advocate, showed us how to evaluate the agent, and build a UI for it. “We want to show you how to move from fragile, unpredictable agentic loops, to a rigorously evaluated network of experts that literally build their own UI,” Casey said.

They did so by deploying a separate model to judge the route, checking both deterministic (e.g., route length) and non-deterministic (e.g., community impact) criteria. For UI development, they showed off the Agent-to-User Interface, or A2UI, an open-source standard developed by Google that created an interface in a single shot. They also used the Agent-to-Agent Protocol, or A2A, and Agent Platform’s Agent Registry, to connect and see which agents are deployed.

“Think of Agent Registry as the DNS of your internet of agents,” Casey said.

Agents that never forget

Richard then mused about how to build agents that get better with time, that “take the learnings from the simulation and optimize for the next run.” Because the answer shouldn’t be to “cram raw text in every request we send back to our agents.”

To capture this learned knowledge, Agent Platform offers Agent Platform Sessions and Memory Bank, plus the ability to turn to tools like Spark or a database to retrieve more information, resulting in an even stronger simulator.

When agents go off course

Thus far, everything had gone swimmingly, but then Richard accidentally “broke” the simulator agent. That provided a perfect opportunity for Megan O’Keefe, Senior Staff Developer Advocate, to show off Agent Interoperability and Gemini Cloud Assist, and how to use them to debug agents at scale.

“With these autonomous agents, the production challenge isn’t just scaling the infrastructure, it’s managing the reasoning, the tool calls — all the places in the system where something can go wrong!” Megan said.

Megan used Agent Runtime trace view to see where the problem was, and using natural language, launched a Cloud Assist Investigation to explore logs and events, which pointed to a specific line of code as the offender. Megan then opened up her Antigravity IDE (powered by Gemini 3, and connected via MCP) to find the problem (an insufficiently run “event compaction” run) and to suggest a fix (add a token_threshold parameter to the event compaction config). She approved the fix and committed it to source, triggering a redeployment to Agent Platform. Problem solved!

Scaling the agents

To this point, all of the presenters had been showing off agent services running as Cloud Run services. Bobby Allen, Group Product Manager, then showed how to convert the apps to Google Kubernetes Engine (GKE), which provides greater control, as well as to use a customized Gemma 4 model, all by vibe coding in the Antigravity editor, which is connected to Cloud Assist. Along the way, Bobby also migrated the agents from GCSFuse to a high-performance Lustre file system.

Closely related to scaling is sharing — making agents available for the world to use and build on. Ines Envid, Senior Director, Product Management and Jason Davenport, Area Technical Lead, showed how to build no-code agents from the Gemini Enterprise app, and how to integrate them with other, “high-code” agents.

Shifting down

Last but not least, it was time to talk about security and governance. “Agents give users and other agents new ways to intentionally — or unintentionally — expose data and behavior in ways that we may not want,” mused Emma.

The standard response to that is to “shift left” — move testing, quality, and performance evaluation earlier in the development process — but for developers, that usually means more work, Richard said. “It’s not sustainable for developers to be responsible for all the layers of the stack,” he said. Instead, “we need to shift down.”

To help, there’s Agent Identity and Agent Gateway, demoed by Ankur Kotwal, head of Cloud Developer Relations. Ankur showed how Agent Gateway uses IAM policies to ensure agent actions are only accessible by approved sources, and how Agent Identity provides each agent with a unique and immutable credential. Then, Agent Policies can be configured to provide guardrails for the agents.

Yinon Costica, Co-Founder and VP of Product at Wiz, then went a step further and showed how Wiz can scan your agent code and infrastructure, and Wiz Green Agent can suggest root cause remediations.

“It’s a full architecture for security to easily understand what you built without you having to actually explain it,” Yinon said. Better yet, he also showed using this functionality from Anthropic’s Claude Code with Opus. “With Wiz, we want to enable your choice of tools and models to fix and prevent real risks,” he said.

The finish line

With this, the developer keynote came to an end. But for Google Cloud developers, it’s just the beginning, as the entire solution is available as source code in GitHub, and all the demos are available as Codelabs. Because when it comes to agentic development, these resources will really help you hit the ground running.

Day 1 at Google Cloud Next ‘26 recap

Wed, 22 Apr 2026 23:00:00 +0000

Last year at Google Cloud Next ‘25, we asked you to imagine a new future for AI. At Next ‘26, the question before you is how do you move AI into production across your entire enterprise?

According to Google Cloud CEO Thomas Kurian, the answer is straightforward: You need a unified stack, with “chips that are designed for models, models that are grounded in your data, agents and applications that are built with those models,” and the whole thing “secured by the infrastructure,” Thomas said in his keynote. (This is the same unified stack that Google uses for Search, YouTube, Chrome, and Android. As Alphabet CEO Sundar Pichai said in his opening remarks, “a big focus of ours is to always be customer zero for our own technologies.”)

As AI matures, we’ve laid out a blueprint on how to succeed. Read on for a whirlwind tour of what we announced from the keynote stage.

Gemini Enterprise: The end-to-end system for the agentic era

Throughout this unified stack is Gemini Enterprise — “the connective tissue between your data, your people, and your goals,” Thomas said, providing a combination of intelligence and automation across multiple layers. Here’s what’s new.

1. Gemini Enterprise Agent Platform

Gemini Enterprise Agent Platform is where you go to build, scale, govern, and optimize agents. As the evolution of Vertex AI, it’s built on top of our leading infrastructure, and deeply integrated with our data and security capabilities — the foundation of the Agentic Enterprise. Here’s a sampling of Agent Platform’s new features and capabilities:

Build: Choose the right environment for the job — from the low-code, visual interface of the new Agent Studio, to the code-first logic of the upgraded Agent Development Kit (ADK). We’ve simplified the entire lifecycle with AI-native coding capabilities to help you ship production-grade agents faster.
Scale: Clear the path to production with the re-engineered Agent Runtime. This supports long-running agents that maintain state for days at a time and are backed by Memory Bank for persistent, long-term context.
Govern: Establish centralized control with Agent Identity, Agent Registry, and Agent Gateway. These capabilities help ensure every agent — whether built on Agent Platform or sourced from our partner ecosystem — has a trackable identity and operates within enterprise-grade guardrails.
Optimize: Guarantee quality with Agent Simulation, Agent Evaluation, and Agent Observability. These tools provide full execution traces and a real-time lens into agent reasoning to help ensure your agents always hit their goals.

To dive deep into Agent Platform, read more in our announcement blog here.

2. Gemini Enterprise app

The Gemini Enterprise app is “the primary environment where your business actually operates,” Thomas explained. The app is where many workers, especially non-technical ones, can ask questions of enterprise agents, create generative media, engage with prebuilt agents, and even create their own with conversational interfaces — all with governance, compliance, and security built in. Here’s a sample of what’s new in this foundational interface:

Gemini Enterprise Projects give your agents permanent memory.
Deep Think solves your most complex business challenges without context pollution.
Microsoft 365 interoperability makes it easy to export docs you create with Canvas into Microsoft Office formats.

To illustrate the power of Gemini Enterprise, Shaun White, three-time Olympic gold medalist, entrepreneur, and snowboarding legend, joined us on stage.

“Back when I was training, our tools were camcorders and guesswork. You’d land a trick and watch it back. And you’d be thinking, ‘How can I make that trick better?’” he said.

Alongside Jason Davenport, Google Cloud Tech Lead, they showed a model that Google Cloud built in collaboration with Google DeepMind that tracked Shaun in space from a two-dimensional video, helping him understand what he was doing right and wrong.

“Learning the trick on the mountain is one thing, but actually understanding the physics of a trick is a whole other thing,” Shaun said.

Read more on the Gemini Enterprise app here.

3. AI Hypercomputer

The same technology foundation that athletes like Shaun White use to understand their performance is being used by enterprises to transform their businesses.

Amin Vadhat, SVP and chief technologist, AI and Infrastructure, took to the stage to announce enhancements to AI Hypercomputer, the integrated supercomputing underneath every AI workload on Google Cloud.

First, there’s the eighth-generation Tensor Processing Unit, or TPU — “a thing of beauty,” Amin said. And because “the demands of training and serving have completely diverged,” this TPU family actually consists of two chips:

TPU 8t, optimized for training, uses new Inter-Chip Interconnect (ICI) technology to scale up to 9,600 TPUs and 2 petabytes of shared, high-bandwidth memory in a single superpod. It achieves three times the processing power of Ironwood and delivers up to double the performance per watt.
TPU 8i, optimized for inference, uses the new Boardfly topology to directly connect 1,152 TPUs in a single pod. It features three times more on-chip SRAM compared to previous versions and a specialized Collectives Acceleration Engine offloads resource-heavy tasks. Taken together, TPU 8i delivers 80% better performance per dollar for inference than the prior generation, helping millions of concurrent agents to run cost-effectively.

AI Hypercomputer also supports Arm-based Google Cloud Axion processors, such as the N4A, now generally available, and will be one of the first platforms to offer NVIDIA’s Vera Rubin NVL72 platform when it is released.

Other parts of the network need to keep up with the demands of the agentic era. For example, the new Virgo Network doubles connectivity to scale training beyond AI Hypercomputer superpods, and Google Cloud Managed Lustre now supports an industry-leading 10 terabytes per second of throughput.

Learn more about all of our AI infrastructure innovations here.

4. Agentic Data Cloud

The AI era hinges on data. Lots of data. That data comes with a catch: It needs to be grounded in context. That’s because “reasoning without context is just a guess,” explained Karthik Narain, chief product and business officer.

To that end, we’re totally rethinking our data platform, and giving it a new name: the Agentic Data Cloud.

Here’s a sampling of what you’ll find in the Agentic Data Cloud:

Knowledge Catalog constructs a unified, dynamic context graph of your entire business enabling you to ground agents in all of your business data and semantics. With Smart Storage and the Object Context API, files in Google Cloud Storage are instantly tagged and enriched with metadata before an agent touches them. Knowledge Catalog is also integrated with Gemini’s Deep Research Agent.
Data Agent Kit delivers a Gemini-powered data science authoring experience across your IDEs, Notebooks, and agentic terminals.
Lightning Engine for Apache Spark is a real-time, serverless engine that is up to 4.5 times faster than open-source alternatives and offers up to double price-performance over the leading competitor for large datasets.
Finally, Cross-Cloud Lakehouse, based on Apache Iceberg, lets you query data in Amazon Web Services or Azure without having to copy it.

Learn all about all the innovations in the Agentic Data Cloud here.

5. Agentic Defense

AI makes — and demands — that everything go faster, and security operations are no exception. Increasingly, “human analysts can’t keep up with AI-driven attacks,” said Francis deSouza, COO, and president, Security Products.

“Security must become an autonomous force, responding faster than the threat itself,” he said.

To help, we introduced three new agents in Google Security Operations to help you defend at the speed of AI.

Threat Hunting agent: Helps teams proactively hunt for novel attack patterns and stealthy adversary behaviors that bypass traditional defenses. Now in preview.
Detection Engineering agent: Identifies coverage gaps and creates new detections for threat scenarios. Now in preview.
Third-Party Context agent: Enriches workflows with contextual data from third-party content. Coming soon to preview.

You can also build your own security agents with remote Google Cloud MCP server support for Google Security Operations, now generally available, and access it from the Google Security Operations chat interface, now in preview.

Then there’s Wiz, now a part of Google Cloud, whose AI-Application Protection Platform (AI-APP), Wiz Security Agents, and Wiz Workflow help you identify and respond to risks and threats at machine speed. New in the Wiz family today are:

Secure vibe-coded applications: A new integration runs Wiz security scanning directly inside the Lovable platform so vulnerabilities, secrets, and misconfigurations caught by Wiz surface in Lovable's built-in security view. Generally available in May.
Secure AI-generated code: Wiz can now remove risks from AI-generated code with inline AI security hooks integrated directly into IDEs and agent workflows, injecting security guardrails before code is committed.
Agent-based remediation: Wiz Skills can equip coding agents and AI-native IDEs with full code-to-cloud context and validated attack surface findings from the Wiz Security Graph.
AI-Bill of Materials (AI-BOM): Work towards eliminating shadow IT by automatically inventorying all AI frameworks, models, and IDE extensions across your environment.
Google Cloud Fraud Defense: The evolution of reCAPTCHA, this platform is designed to discern the legitimacy and authorization of bots, humans, and agents. Now generally available.

Read more about these security innovations here.

6. Workspace Intelligence

For a look at AI for end-users, we heard from Yulie Kwon Kim, VP, Product, Google Workspace, who shared new ways that AI is manifesting in our collaboration and productivity suite. Workspace Intelligence is a unifying semantic layer that breaks down information and context silos for you and your agents. It understands your work, your priorities and the people you work with to help you get more done.

“Think of it as a unified intelligence layer that lives inside every Workspace app. It connects the dots and lets AI do the heavy lifting," Yulie said.

Here’s what’s new:

Ask Gemini in Google Chat allows you to instantly synthesize information, surface insights, and query projects from across Workspace directly from your Google Chat window. It provides proactive daily briefings to help you prioritize, and also lets you take immediate action — such as scheduling a meeting on your calendar or creating a Google Doc to develop a pre-meeting brief — turning your conversations into momentum without having to switch tabs.

Reimagined content creation in Docs, Sheets, and Slides uses Workspace Intelligence to synthesize information from across Workspace and the web and creates professionally formatted drafts that match your voice, style, and brand.
AI Inbox and AI Overviews in Gmail creates a personal, proactive inbox assistant with Gemini.
Google Drive Projects instantly organizes your team's files and emails to manage workflows, generate content, and deliver specific answers based on rich project context. In addition to newly added AI Overviews and Ask Gemini, Projects is another way we’re transforming Drive from a storage tool into an active collaborator to provide insights about your data.
Workspace agent in Gemini Enterprise executes complex, multi-step tasks across Google Workspace apps without having to leave Gemini Enterprise.

For more, check out the full Workspace Intelligence blog.

7. Agentic Commerce

For enterprises, AI agents are reshaping how consumers engage with companies and their products.

“In the agentic era, an agent isn’t just a tool; it’s a strategic extension of your business, built to expand your reach, deepen engagement, and personalize service at scale,” said Carrie Tharp, Google Cloud Vice President of Go To Market Strategic Industries.

Gemini Enterprise for Customer Experience offers a suite of tools to enhance the entire customer journey, from the first moment of discovery through on-going interactions that remember the customer like the best shopkeeper would.

Shopping agent and Food Ordering agent bring new conversational sales and ordering capabilities direct to businesses and third-party chat and digital interfaces.
Omnichannel Gateway helps agents maintain context across web, mobile, and voice, so a company’s agents can offer more personalized service.
Agent Assist helps during complex customer service situations, coaching employees to deliver fast and more accurate answers to customer questions by having organizational data readily available through gen AI grounding.

With Omnichannel Gateway in particular, about bridging the physical, digital, and agentic shopping experience, so consumers always have a familiar, brand-aware experience.

“If a customer moves from text chat to a phone call, the agent seamlessly remembers exactly where they left off,” said Carrie. Now that’s progress!

For more customer stories, check out all 1,302 of the latest gen AI use cases from businesses around the globe.

Innovate all the things

From new products, to new solutions, to new ways of working, there are so many other ways that we’re helping organizations take their AI from pilot to production.

Over the following week, we’ll share even more news, helpful how-to guides, and go deeper on today’s announcements. Stay tuned!

Enabling the agentic enterprise: business and industry agents arrive in Gemini Enterprise

Wed, 22 Apr 2026 21:00:00 +0000

We are officially moving past the era of one-size-fits-all AI. Enterprises today require highly specialized, role-specific tools to drive real productivity — but they cannot afford to sacrifice security, oversight, or control to get there.

What's new: Today, we’re bringing partner-built agents from our Agent Marketplace directly into the Agent Gallery inside the Gemini Enterprise app. This creates a powerful, centrally governed hub for discovering and managing specialized AI from partners like Accenture, Adobe, Atlassian, Deloitte, Lovable, Oracle, Palo Alto Networks, Replit, S&P Global, Salesforce, ServiceNow, Workday, and more.

Why this matters for your business: With millions of paid seats across thousands of companies, the Gemini Enterprise app allows organizations to automate complex, end-to-end workflows in a single, unified environment. Five key benefits include:

Move beyond agentic overclaims: The market is flooded with passive chatbots and rigid workflows. Our featured Marketplace partners deliver true autonomous agents capable of executing complex, end-to-end tasks. They achieve this by combining deep context and memory, intelligent orchestration, and safe tool use.
High-caliber agents: Every featured agent earns the "Google Cloud Ready - Gemini Enterprise" designation only after passing a strict four-step evaluation for basic functionality, output accuracy, autonomous execution, and enterprise standards.
Built-in safeguards: Gemini Enterprise Agent Platform assigns every agent a cryptographically secure identity for a clear audit trail, while our Agent Gateway and Model Armor actively screen traffic to guarantee your data is never used for model training.
Strong IT governance: To balance workforce productivity with strict organizational oversight, we've built Agent Gallery around a secure, two-step governance model. Employees can browse the gallery and submit procurement or access requests for the agents that fit their needs, while IT remains firmly in the driver's seat. Administrators review all incoming requests to approve or deny deployments, retaining granular control over exactly who can interact with each agent across the organization.
A unified ecosystem breaking down silos: Typically, purchasing third-party AI means bolting on disconnected platforms. Our approach eliminates that friction. These partner agents are built to natively run within Gemini Enterprise Agent Platform. This means third-party agents and internal custom builds live side-by-side in a single, cohesive ecosystem, allowing your organization to scale its AI capabilities infinitely without ever fracturing your infrastructure.

Go-to-market opportunity for Partners: Today, we announced a $750 million partner fund for agentic development. In addition, partners can tap into a $460 billion plus backlog of committed enterprise spend and reach millions of Gemini Enterprise users directly in their daily workflows. We are turning everyday user discovery into a procurement flow for IT, standardizing contracts and accelerating purchasing cycles by up to 50%. With Google sales reps incentivized to work with you, and Marketplace vendors closing deals 112% larger, there has never been a better time to build with Google Cloud.

Where you can get started:

For customers: Discover these agents directly via the Agent Gallery in the Gemini Enterprise app. Start your 30-day trial today.

For partners: Take advantage of our rapid agent deployment framework, which provides a clear path to onboard, commercialize, and make agents discoverable to Gemini Enterprise users. Apply for our AI Agents Program now.

In addition, we’re launching the Google for Startups AI Agents Challenge today, including a track dedicated to startups with agents that can be primed for use in Gemini Enterprise.

Meet the agents your organization can start using today

A sampling of featured partner-built agents from Agent Marketplace accessible via the Agent Gallery in the Gemini Enterprise app

Acalvio: ShadowPlex Preemptive Cyberdefense Agent deploys honeytokens in Cloud IAM and in cloud workloads to disrupt agentic AI attacks at the reconnaissance phase. This helps secure cloud workloads from malicious or unsafe agent actions in real time.

Accenture: The Supply Chain Inventory Intelligent Advisor agent is an agentic AI-powered decision support system designed to optimize inventory planning and tracking efficiently. Using real-time Data and AI, it delivers actionable insights to balance service levels and minimize excess inventory.

Adobe: Adobe Marketing Agent for Gemini Enterprise connects natural language queries to Adobe's Customer Experience agentic capabilities. It enables instant insights on campaign performance, target audiences and journey monitoring powered by Adobe, helping teams work smarter without leaving their workflow.

Alteryx: Alteryx AI Insights Agent delivers analyst-curated insights inside Gemini Enterprise. Built on Alteryx One, analysts define trusted datasets while business users simply ask questions to receive governed, repeatable answers. Requires an Alteryx One subscription.

Ambiguous AI: Ambiguous Recruiting Coworker manages candidate communication, multi-round scheduling, and pipeline tracking from application to offer. It works natively across Gmail, Google Calendar, Google Sheets, Google Drive, and more, while looping you in for key decisions.

Amdocs: The Telco Customer Experience Agent autonomously resolves complex customer service issues across digital and contact center channels. Combining Amdocs Cognitive Core with Gemini Enterprise for CX, it reasons, orchestrates, and executes end‑to‑end telco processes, improving resolution speed, cost efficiency, and customer satisfaction.

AODocs: AODocs AI Agent provides instant, trustworthy answers grounded exclusively in the latest validated versions of controlled documents. It supports diverse use cases including standard operating procedures (SOPs), engineering specs, legal policies, and risk management procedures.

Articul8 AI: Articul8 Table Understanding Agent converts complex documents into analytics-ready table data. It reconstructs hierarchical layouts and relationships missed by traditional OCR for accurate analysis. Articul8 Diagram Understanding Agent turns complex diagrams into structured data, interpreting symbols and spatial relationships for automated analysis and seamless workflow integration.

Ascendo: Ascendo AI Agent helps critical infrastructure teams automate workflows, improve diagnostics, and reduce field escalations. By ingesting logs, telemetry, and service manuals, it recommends optimal actions to support faster, accurate decision-making in complex service environments.

Atlassian: Rovo is an AI teammate seamlessly integrated across Jira, Confluence, and your tool stack. Built on Atlassian’s Teamwork Graph, it surfaces knowledge, automates tasks, and takes intelligent actions to help teams pull context, move faster and stay aligned.

AutoCIO: Portfolio managers and investment teams use the AutoCIO Financial Forecasting Agent to design, optimize, and manage portfolios. It automates end-to-end workflows, delivers predictive signals across 50,000+ securities, and enables real-time, hyper-customized strategies via natural language prompts.

Avalara: Avi Agent is an AI-powered gateway for tax and compliance automation. Built on Avalara’s Agentic platform, it smartly coordinates between external systems and compliance agents to instantly and securely calculate taxes, file returns, and manage invoicing.

Backstory: Backstory Revenue Answers Agent tells revenue leaders what's happening on any account, identifies deal risks, and suggests actions. It achieves this by synthesizing CRM data, engagement activity, stakeholder relationships, and public company news into actionable insights.

Botcopy: TrueQ is an AI analyst for government contact centers, enabling supervisors to understand and improve conversational AI performance in the Gemini Enterprise app. It identifies conversation patterns, escalation triggers, and knowledge gaps to improve automation, strengthen governance, and reduce support costs.

Carto: Carto Site Selection AI Agent helps enterprises analyze and visualize prospective commercial locations using spatial analytics. Users ask conversational questions about foot traffic, demographics, or competitors, and results automatically fuel a comprehensive location analysis.

Corvic AI: Corvic AI Agent connects to your enterprise data and delivers production-ready decision intelligence — no pipelines and no data engineering required. Use cases include turning competitive signals and market trend data into actionable intelligence, root cause analysis and customer support automation.

Cotality: The Cotality Payoff Analysis Agent helps mortgage lenders improve customer retention and reduce loan runoff. By analyzing loan originations and payoffs, it delivers actionable intelligence to understand recapture performance, borrower behavior, and competitive insights.

Crescendo Lab Ltd.: DAAC AI Agent transforms complex data into actionable growth strategies for marketing and sales. Using natural language, it delivers instant conversational insights and automated dashboards, effectively eliminating manual reporting and guesswork.

Deloitte: Tariff Management Agentic Suite automates customs regulatory workflows via sub-agents that ingest documentation, extract attributes, and reconcile tariffs against filings. It flags discrepancies to ensure full audit coverage, eliminate manual effort, and recover unnoticed costs.

Devoteam: Demand Sensing Agent helps leaders improve forecast accuracy and optimize inventory. It analyzes sales and real-time market signals to provide immediate demand insights, enabling proactive adjustments. Request for Proposal Agent helps sales teams draft proposals, leveraging centralized knowledge to rapidly generate tailored and compliant proposal drafts.

Dun & Bradstreet: The Dun & Bradstreet Business Verification Agent enables trusted business verification in the Gemini Enterprise app. Powered by Dun & Bradstreet’s trusted data, it resolves businesses to a D-U-N-S® Number and validates identity attributes, enabling front-line teams to move faster and focus on exceptions.

Dynatrace: Dynatrace for Gemini Enterprise integrates AI agents with Dynatrace observability data via Gemini Enterprise Agent Platform and A2A protocol. This enables real-time insights into performance, infrastructure, and user experience, seamlessly enhancing AI workflows while ensuring enterprise-grade security and compliance.

Enigma: Enigma KYB Agent gives compliance teams instant, AI-driven business verification backed by ground-truth business identity from 100M+ registrations and authoritative government records, plus sanctions watchlists and real-time transaction signals.

Eon: Eon AI Agent empowers cloud infrastructure teams to streamline data protection operations. It connects Google Cloud accounts, triggers resource discovery, monitors protected resources, tracks backup jobs, and inspects restore status — all through natural conversation, replacing manual console workflows.

EPAM: Creative Conductor Marketing Agent turns creative briefs into polished promotional videos. It coordinates specialized sub-agents for visuals, voiceover, and music, assembling assets into a final MP4 to deliver fast, scalable, high-quality marketing content.

Exa AI: Exa Agent gives workers a live web view, turning it into a searchable database of code, people, and companies. It searches billions of pages to accelerate tasks like developer tooling, recruiting, and market mapping.

Foxtrot Communications: Forge - AI Agent helps teams eliminate manual ETL pipeline development. Connect any API, and Forge autonomously delivers clean, normalized, analytics-ready datasets, reducing time-to-insight to minutes and freeing engineers to focus on strategic initiatives.

Genpact (UK) Ltd: Genpact Finance One – Revenue Lens and PnL Agents are a suite of agents built on Google ADK and the A2A protocol, enabling finance users to gain actionable insights from revenue and profit & loss data through natural language conversations in Gemini Enterprise.

Genspark: Genspark AI Agent turns multi-step work into finished outputs with one prompt by combining search, web interaction, content generation, and media creation. It helps users execute research, create documents, and automate workflows through a single agent built for real work.

HCLTech: HCLTech ITOps Agent is built on Google Cloud and the HCLTech AI Force platform and enables autonomous IT operations within ServiceNow. Using natural language, it triages incidents, surfaces relevant SOPs, and drives guided remediation to improve overall efficiency, consistency, and operational performance.

Invideo: Invideo empowers marketing teams to scale video production while maintaining consistency. By leveraging brand guidelines and assets, it delivers production-ready ads and short films with built-in quality controls and output in 50+ languages.

Iron Mountain, Inc.: The Iron Mountain InSight® DXP AI Agent bridges Gemini Enterprise and your managed content for unified search and AI-powered Q&A. Its intended purpose is to accelerate results and reduce hours of manual document review to seconds of conversational inquiry.

Lilt: LILT Assist empowers global content managers to transition from manual tasks to autonomous orchestration, managing end-to-end multilingual production and brand governance. Assist scales global content volumes with AI while ensuring enterprise-grade consistency and compliance.

Lovable: Lovable empowers anyone to build real apps and websites by chatting with AI. Teams across product, design, marketing, sales and operations can turn ideas into real, working software.

LTM: LTM Video Intelligence Agent is based on Google Cloud AI technology and converts recorded videos into BDD test cases in a standard format for faster time-to-test and improved productivity.

LumApps: The LumApps AI Agent helps organizations retrieve company knowledge and announcements, trigger workflows, and complete tasks across enterprise systems.

Manhattan: Manhattan Active Agents are AI digital teammates for warehouse management, transportation management, order management, and platform operations. These agents help monitor work, resolve exceptions, guide users, automate tasks, and recommend actions to reduce manual effort; improving speed and accuracy, increasing utilization, and delivering better operational resilience.

Menlo Security: The Menlo Security Orchestrator for Gemini Enterprise reduces mean time to repair (MTTR) from hours to milliseconds. Command global policy and perform deep threat reporting through natural conversation in the Gemini Enterprise app.

monday.com: monday.com agent in Gemini Enterprise turns planning into execution. Teams can quickly build boards, summarize updates, and trigger automations without leaving their workspace—keeping everyone aligned, tracking progress, and driving measurable results.

Nativeorange: LexAI Agent streamlines property and casualty (P&C) underwriting and operations. It ingests submissions, extracts data, and automates risk evaluation, delivering actionable insights on classification and offering next steps to improve speed, accuracy, and decision consistency.

Neo4j: Neo4j Agent translates natural language into Cypher, enabling applications to read from and write to the Neo4j database. Integrated with Gemini Enterprise, it can be used to build agents grounded with GraphRAG and connected enterprise knowledge.

Obin AI: Obin Financial Agent helps private markets and commercial lending teams run financial analysis from Gemini Enterprise. Users submit model spreadsheets and deal-related documents with a scenario description in plain-language. The agent then extracts relevant data and computes financial metrics across modeling assumptions.

Onix: The Onix Risk & Compliance Intelligence Agent automates compliance reviews and risk assessments across regulated industries. It reduces review times from weeks to minutes, proactively identifies regulatory risks, and creates detailed audit trails with citations.

OpenText: OpenText™ Content Aviator™ is a generative AI assistant that enhances content search and productivity. Built on Gemini Enterprise Agent Platform, it allows users to perform natural language searches and efficiently summarize enterprise documents.

Oracle: Oracle AI Database Agent lets you query Oracle data directly from Gemini Enterprise. Available via Google Cloud Marketplace, it delivers fast, enterprise-grade answers without requiring custom chatbots or NL-to-SQL maintenance.

Orion by Gravity: Orion by Gravity is an AI analyst that connects to Looker, BigQuery, and more to proactively hunt for insights. It empowers data teams to identify anomalies and recommend actions that uncover 'unknown-unknowns' in analytics-rich industries.

Palo Alto Networks: Prisma AIRS Model Security agent scans over 35 model formats to eliminate malicious code hidden deep within model layers. This proactive defense goes beyond standard scanning to ensure neural integrity, detecting poisoned weights and backdoors to prevent unauthorized data exfiltration and model hijacking.

Pendo: The Pendo Agent connects your product intelligence to Gemini Enterprise. Query usage trends across SaaS and AI tools, surface adoption gaps, and understand the impact of your AI investments without leaving the Gemini Enterprise app.

Persistent Systems: Ethical Content Auditor Agent helps editors at publishing houses audit political and social content for potential biases and risks. Editorial Assistant Agent enables efficient manuscript evaluation by reading and understanding long manuscripts, analyzing market trends and providing data-backed summaries and recommendations.

Pluto7: Pluto7’s Planning in a Box - Pi Agent orchestrates 50+ sub-agents for supply chain and manufacturing teams. This enables self-healing, autonomous supply chains that optimize demand and production in real-time.

Product Genius: Product Genius brings AI-powered ecommerce intelligence directly into Gemini Enterprise. Share your website URL to get instant insights, actionable policies, and a conversational analyst that helps you optimize your store—no code install required.

Quantum Metric: Felix Agentic for Gemini Enterprise delivers instant insight, contextualized in business objectives, to digital leaders and analytics teams. Leveraging Gemini models, Felix continuously analyzes your digital experience to explain what changed, why it matters, and where it’s costing you.

Red Hat, LLC: Red Hat Lightspeed Agent for Google Cloud helps SREs and admins manage Red Hat Enterprise Linux infrastructure on Google Cloud via natural language. Built on A2A and Gemini, it securely connects with Red Hat Lightspeed services to streamline enterprise IT operations on Google Cloud.

Replit: Replit Agent enables business users to build enterprise-grade applications using natural language. With no coding knowledge required, users turn complex requirements—like a BigQuery-backed customer health dashboard—into functional custom tools in minutes.

S&P Global: The S&P Global Data Retrieval Agent by Kensho enables financial professionals to access cited data from S&P Global's proprietary datasets. It allows analysts and portfolio managers to analyze earnings calls, conduct market research, and instantly retrieve specific financial metrics.

Salesforce: Agentforce Sales for Gemini Enterprise can engage leads, create meeting briefs, surface deal risks and guidance, and manage pipeline and CRM updates — securely and in real-time — without leaving Gemini Enterprise. With the agent handling the grind of manual tasks, sellers can focus on relationship building and strategic selling to win.

Sana from Workday: Sana Self-Service Agent instantly finds and summarizes information from Workday and other knowledge sources, providing personalized answers and assistance with tasks. With over 300 skills across areas like pay, time, and absence, the agent handles everyday HR and finance tasks globally, reducing support tickets and allowing teams to focus on high-value work.

Saviynt: Identity Security for AI enables continuous discovery and inventory of agents, identifying their functions, access, and risks. It treats agents as first-class identities with defined purposes and accountable owners. This blend of posture visibility and enforced accountability converts opaque automation into governable, enterprise-ready identities.

ServiceNow: Now Assist for IT Operations Management operates inside the ServiceNow AI Platform, harnessing the power of GenAI to elevate operational intelligence and automation across IT environments. It introduces a suite of AI-driven capabilities designed to optimize every stage of alert and incident management.

Sheetgo: Sheetgo is an AI-ready data execution platform that transforms spreadsheet-driven operations into governed, reliable pipelines for BigQuery and Gemini Enterprise.

Skyflow: The Skyflow Runtime Data Security Agent helps teams securely deploy agents on Gemini Enterprise Agent Platform. It provides expert guidance and on-demand data protection with fine-grained access controls, enabling enterprises to meet compliance without risking privacy.

Supermetrics: The Supermetrics Marketing Intelligence Agent delivers live, cross-channel performance insights directly in Gemini Enterprise. Connecting to 175+ sources like Google Ads and GA4, it provides fast, hallucination-free analysis using simple natural language prompts.

Synthpop Inc.: The Synthpop Patient Journey Orchestration Agent turns unstructured referrals into validated, payer-ready orders in real time by executing intake and patient-facing workflows. These flows help payers save time as part of Synthpop’s coordinated, production-grade administrative backbone.

Tech Mahindra: Frontline agents use Order Assist to resolve complex orders and service cases faster. By delivering real-time, contextual intelligence through conversation, it reduces handling time, improves first-contact resolution, and enables consistent, data-driven responses.

Teradata: Data Analyst AI Agent enables natural-language analytics on enterprise data in Google Cloud. It translates business questions into governed SQL and Python workflows on Teradata, delivering faster, secure insights without requiring data movement.

Typeface: Typeface Agent acts as a seamless bridge to creative campaigns. It enables instant access to Typeface teams, brand assets, and campaigns via natural language from Gemini Enterprise. Users can trigger high-quality, brand-aligned content creation workflows, streamlining the journey from ideation to production.

UKG: Agentic People Assist — HR Service Delivery elevates the agentic capabilities of the UKG Workforce Operating Platform by empowering employees and managers with knowledge and connected agents to turn intent into action. It orchestrates workflows across time, pay, and HR to deliver real-time, personalized, and measurable outcomes.

WRITER: WRITER Agent is autonomous AI for Fortune 500 enterprises. Built for high-stakes enterprise workflows, it safely plans and executes autonomous work across data and tools based on company context, from brand standards to repeatable playbooks.

XM Cyber: XM Cyber’s PostureAI agent serves as an autonomous posture specialist, providing continuous assessment of Google Workspace applications against security best practices. It is part of a broader Continuous Exposure Management offering that allows organizations to safely adopt AI and secure their critical business processes.

Together, we are creating the world's largest ecosystem of agents — one that will redefine every industry, supercharge every function, and accelerate every business process. I cannot wait to see what we all achieve together.

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!

What’s new in BigQuery: Powering the Agentic Era

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

Succeeding in the agentic era requires a transformation in your data strategy: moving from human-scale to agent-first workloads, evolving from reactive intelligence to proactive action, and shifting from raw data to semantic knowledge that agents can use to reason accurately.

For over a decade, BigQuery’s continuous innovations have helped tens of thousands of organizations build a scalable data and AI foundation and navigate several industry and technological transformations. BigQuery has evolved into an autonomous data-to-AI platform, growing over 30x in data processed with Gemini, 25x in AI functions processing unstructured data, and 20x in agent-building tools with Model Context Protocol (MCP).

Customers like Definity are building data platforms to enhance their customers’ experience, improve back-office operations, and boost data team productivity.

“We stood up our data platform in Google Cloud and ingested all critical insurance data in 10 months, which is about half of the time that people see in the industry. The technology that BigQuery provides, processing large amounts of data very quickly, is giving our practitioners and engineers tools that are advanced and a platform that has AI and ML built in. We have doubled the number of users [in a very short period of time].” — Tatjana Lalkovic, Chief Technology Officer, Definity

Today, we are announcing new BigQuery capabilities in lakehouse, built-in AI processing and reasoning, and agentic experiences, all anchored by our commitment to industry-leading price-performance and enterprise readiness.

Open, cross-cloud lakehouse

Enterprise data is often scattered across applications, multiple cloud environments and on-premises. While early lakehouse solutions reduced data duplication, the agentic era demands a foundation that is natively multimodal, cross-cloud, and AI-ready. Our approach blends Apache Iceberg’s interoperability and Google’s differentiated infrastructure with new capabilities, including:

Managed Iceberg tables in Lakehouse (GA, formerly BigLake) enables the openness of Iceberg with advanced BigQuery capabilities, including automatic table management, Iceberg partitioning, multi-table transactions, change data capture, enhanced vectorization, and history-based optimizations.
Iceberg REST catalog enables read/write interoperability (preview) on Iceberg tables between BigQuery, Spark, and other OSS and third-party engines, so you don't have to make complex engine trade-offs.
Cross-cloud Lakehouse (preview) brings BigQuery AI and analytics to other clouds, starting with AWS and Azure. Using open standards like the Iceberg REST Catalog, high-bandwidth networking via Cross-Cloud Interconnect, and transparent caching, BigQuery achieves performance and total cost of ownership comparable to native warehouses, enabling true cross-cloud for enterprises.
Catalog federation (preview) enables easy discovery, analysis, and zero-copy sharing of data across AWS Glue, Databricks, SAP, Salesforce, Snowflake, and Confluent Tableflow (coming later this year).
Real-time data replication that closes the loop between raw data and operational action by allowing you to replicate data from Spanner, AlloyDB, and Cloud SQL instantly into BigQuery tables (GA) and Iceberg (preview).

Analyze data across BigQuery and Iceberg table on AWS using SQL or natural language

Refer to this blog to learn more about our latest lakehouse innovations.

Graph-based reasoning for enterprise agents

For AI to solve truly complex, multi-hop operational problems, business logic must extend all the way down to the data platform layer. Defining logic at this layer ensures that definitions remain consistent and governed from ingestion to consumption. BigQuery Graph (preview) provides the foundation to activate this context, allowing data practitioners to map entities, relationships, and business logic directly within the data platform. This anchors AI agents in a governed reality, enabling them to solve sophisticated challenges at scale with consistent accuracy.

Today, we are announcing new graph capabilities to enhance AI reasoning:

Native support for measures in BigQuery Graph (preview) enables you to unify analytical metrics and relationships into a single, governed entity. It transforms data into a "business map" for multi-hop structural reasoning, allowing agents to move beyond simple searches to trace the ripple effects of business events.
Graph support in BigQuery Conversational Analytics (preview) allows conversational analytics agents to navigate a deterministic business map instead of raw tables to provide answers with higher accuracy. Graphs enable dual reasoning: agents can instantly calculate precise KPIs using measures while simultaneously traversing complex relationships to uncover the "why" behind the numbers.
BigQuery Graph and Looker integration (preview) allows you to reuse measures defined in BigQuery across your data stack by exposing graphs as Looker views. You can also define BigQuery Graphs using Looker with source control and validation. This interoperability ensures that metrics, such as Churn Rate, remain identical across both dashboards and AI agents.
Visual modeling experience in BigQuery Studio (preview) provides an intuitive interface to easily build and manage the entities, relationships, and business logic that power your agentic context.

Analyze data in natural language using Conversational Analytics Agents with BigQuery Graph

Native AI processing to unlock structured and unstructured data

Your data is no longer confined to rows and columns; agents demand a platform that can work across structured and unstructured data at scale without requiring data copies or movement. BigQuery AI makes it easy to perform predictive machine learning tasks (like forecasting, anomaly detection, and recommendations) and generative AI tasks (like entity extraction from images and text, content generation, and data enrichment), with built-in access to over 170 foundational models. New capabilities in BigQuery AI include:

AI.PARSE_DOCUMENT (preview) simplifies complex document processing workflows with a single SQL function that automates Optical Character Recognition (OCR), layout parsing, and chunking.
TabularFM model (preview) brings high-quality regression and classification to BigQuery without the need for extensive feature selection, tuning, training or model management.
ObjectRef (GA) allows you to process unstructured data alongside structured data using SQL and Python. This establishes the foundation for building rich, multimodal context directly on your Knowledge Catalog.
Optimized mode (preview) for SQL-first, AI co-processing-managed functions like AI.CLASSIFY and AI.IF trains task-specific models on the fly, delivering a 230x reduction in tokens consumed compared to row-by-row gen AI processing.
BigQuery-native Gemma embeddings (preview) allow you to generate high-quality embeddings at scale on standard CPUs.
Autonomous embedding generation (GA) fully manages the pipeline for unstructured data, automatically, keeping vector indexes in sync as new data gets ingested.
BigQuery hybrid search (preview) unifies retrieval by integrating semantic and full-text search into a single function, delivering superior precision for RAG and complex exploration.
Python UDF (GA) allows you to enrich, transform or clean data with fully managed Python scalar functions. You can bring your own code or libraries and the functions will autoscale to millions of rows with serverless, scale-out execution. This will be rolled out in phases over the next few weeks.
Connected Sheets brings the scale of BigQuery to the familiar Google Sheets interface, now supporting forecasting with TimesFM model (GA) and anomaly detection (preview).
Geospatial analytics datasets (preview) can now be accessed directly from BigQuery, allowing you to combine geospatial data — such as infrastructure assets and road management insights — with your enterprise data for deeper analysis, without complex manual data wrangling.

Access, process and activate unstructured data in BigQuery

Agentic experiences

AI enables highly skilled data teams to focus on high-impact strategic initiatives by offloading tedious and time-consuming tasks. We are pioneering this space by integrating Google’s foundational research, capable models, and specialized tools directly into BigQuery, providing automation and assistance at every stage of the data lifecycle. New agentic experiences in BigQuery include:

Conversational Analytics in BigQuery (GA) allows teams to query complex datasets using natural language. It provides a secure and transparent path to insights while supporting advanced features like predictive analytics and reasoning across structured and unstructured data. In addition to BigQuery Studio, BigQuery Conversational Analytics agents can be published to and consumed from Data Studio (Preview), Gemini Enterprise (Preview), or via an API for custom applications (GA).

Gain insights from your data using natural language with Conversational Analytics in BigQuery

Proactive agentic workflows (preview) go beyond simple questions to detect metric shifts, perform root-cause analysis to explain why a change occurred, and deliver scheduled research briefings directly to your inbox.

Proactive agentic workflows in BigQuery

BigQuery Agent Analytics offers plugins for ADK (GA) and LangGraph (Preview) frameworks to record agent activity to BigQuery and analyze it for troubleshooting, optimization, and agent evaluation.
BigQuery Studio gets new productivity tools, including a contextually aware assistant (preview) for resource discovery and troubleshooting, SQL Cells (GA) for blending SQL and DataFrames, and Visualization Cells (GA) to create visuals directly within notebooks. The Files Explorer (GA) allows developers to organize, share, and manage code assets in folders. Additionally, we are introducing Git integration and workflows (preview) in BigQuery Notebooks, providing complete SCM coverage across GitHub, GitLab, Bitbucket, and Azure DevOps for data science workflows.
Data Science Agent (GA) in BigQuery notebooks allows you to simply state goals in plain English to automatically execute plans for loading, cleaning, and visualizing data using BigQuery ML, DataFrames, or Spark.
Data Engineering Agent (GA) - develop, manage, migrate, and troubleshoot data pipelines in BigQuery. It uses your custom context through instructions and integration with the Knowledge Catalog and is available in BigQuery Pipelines and Dataform.
Colab Data Apps (preview) bridge the gap between analysis and action by transforming notebook analyses into shareable, fully managed interactive Python applications that business teams can access from Data Studio.
BigQuery remote MCP server (GA) and BigQuery ADK toolset (GA) reduce the need for manual database connectors for agents.
Google Cloud Data Agent Kit (Preview) provides a portable suite of skills, Model Context Protocol (MCP) tools, environment-specific extensions, and native plugins. By meeting you where you build, like VS Code, Gemini CLI, Codex, and Claude Code, the Data Agent Kit turns your IDE, notebook, or terminal into a native data environment. This enables your environment to orchestrate a massive range of data workflows, automatically selecting the right frameworks (like BigQuery, dbt, Apache Spark, or Apache Airflow) and generating production-ready code.

Unparalleled performance and scale

Modern analytical workloads are increasingly unpredictable and distributed. We continue to invest in enhancing BigQuery's core engine to ensure that as data grows in your environments, your costs and operational overhead do not.

Fluid scaling (GA) enables you to execute highly variable workloads with a premier autoscaling model that does not require a cost-and-performance trade-off. Fluid scaling in BigQuery enables true per-second billing, offering up to 34% cost savings.
Advanced runtime, small query, and history-based optimizations (GA) accelerate native and Iceberg workloads without code or schema changes. BigQuery has improved query speed by 35% year over year while reducing query processing costs by 40% year over year.
New workload management features — including reservation groups (GA), flexible dynamic assignments (preview), and project-level slot and concurrency controls (preview) — provide granular cost attribution and price-performance control, all simplified by declarative, rules-based workload management (preview).
Enhanced observability (GA) and intersection routing (preview) provide flexible disaster recovery capabilities for mission-critical workloads. Agent-powered observability (preview) for turnkey troubleshooting further simplifies operations, while the agent-ready security center (preview) provides a unified, fine-grained access control experience.

BigQuery fluid scaling for unpredictable workloads

In the agentic era, BigQuery isn’t just where your data lives — it’s where your data thinks, reasons, and acts. The future is here, and it’s more open, autonomous, and capable than ever.

Get started on your data and AI journey by taking advantage of our BigQuery data migration offer. We can’t wait to hear how you’re innovating with data.

What’s new with the Cross-Cloud Network at Next ‘26

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

While generative AI sparked a revolution, the true paradigm shift is the rapid evolution from standalone AI models to multi-agent autonomous systems. In this new era, the network transcends basic connectivity to become the critical integration layer for your agentic enterprise.

As AI agents and services surge, your core applications remain as vital as ever. To thrive in this rapidly evolving landscape, you need a planet-scale network to connect, protect, govern, deliver, and secure all your users, data, agents, AI services, and core applications across clouds and on-premises.

Google Cloud's Cross-Cloud Network provides this unified foundation, and is now used by 65% of the Fortune 100 and handles up to 27 exabytes of data per month. At Google Cloud Next, we are introducing networking innovations to accelerate your AI infrastructure, strengthen security, and simplify operations.

Optimized networking infrastructure for AI

As we move toward an agentic world, the network must support massive-scale inference paired with reinforcement learning. At Google, we’ve spent years refining this cycle to power our own global AI services. Today, we’re announcing AI infrastructure network innovations that bring this same architecture directly to your workloads, across agents, inference, training, and beyond.

Networking for agents

The Gemini Enterprise Agent Platform is a comprehensive enterprise environment designed to build, scale, govern, and optimize the next generation of autonomous agents. Key innovations being announced in preview include:

Agent Gateway: Air-traffic control for agentic traffic

Agent Gateway understands MCP and A2A agentic protocols and provides an open, extensible, scalable way to enforce centralized governance policies to securely connect agents, models, and tools across runtimes.

Ambient networking: A seismic shift in service-to-service connectivity

Ambient networking, a new integrated data plane for Google Kubernetes Engine (GKE) and Cloud Run, provides service discovery, zero-trust access, and traffic management without the need for complex and resource-heavy sidecar proxies. It reduces operational overhead and enables up to a 10x reduction in GKE resource usage for layer 4 (L4) mesh capabilities

Ambient networking underpins two new capabilities:

Service bindings automatically establish service-to-service connectivity, allowing developers to move faster to build and scale their agentic applications and services.
Network Services Monitoring bridges application and network observability gaps resulting in faster root-cause analysis and simplified troubleshooting.

Rich partner integrations and customizations

With the help of Service Extensions, we are developing solutions for identity, governance, and AI security for agent-to-anywhere traffic. Coming soon in preview to Agent Gateway are:

Identity and governance administration: Offering delegated authorization to Cloud IAM and partner services from Okta, Ping, Saviynt, and Silverfort to enforce real-time, contextual governance policies based on application and business context.
Runtime security: As a universal enforcement point by integrating with Google Cloud’s Model Armor and partner solutions from Broadcom, Check Point, Cisco, CrowdStrike, Exabeam, F5, Netskope, Palo Alto Networks, Thales, and Zscaler. Together, these can help to secure agentic communications against emerging AI attack vectors.

These innovations are built on an open foundation including Envoy and Kubernetes, providing strong, integrated governance in multicloud environments using standard Kubernetes Gateway APIs.

Networking for inference

At Google we run inference at scale with optimized use of distributed GPU and TPU resources, automatic failover between regions for high availability, and optimized global request routing for fast end-user performance. GKE Inference Gateway delivers these capabilities to our cloud customers including the following new innovations:

Multi-region support allows scaling inference services across regions, enabling cross-regional failover, optimized utilization, and reduced global latency (preview).
Predictive latency boost improves utilization with intelligent request routing based on predefined performance targets (preview).
Disaggregated serving leverages llm-d’s SGLang support, offering the flexibility to choose between vLLM and SGLang for model serving (GA).

Gemini Enterprise Agent Platform reduced Time to First Token (TTFT) latency by over 35% for Qwen3-Coder by using GKE Inference Gateway.

“Before GKE Inference Gateway, managing our inference stack with Ray Serve created a complex, dual-orchestration layer that was a significant burden on our small operations team. Moving to the Inference Gateway and native Kubernetes deployments was the 'North Star' architecture we needed to simplify management and achieve robust production stability with a GKE-native batteries-included solution.” - Mikhail Lubinets, Lead HPC Engineer, Technology Innovation Institute

Networking for training

At Google, we build and run the largest AI models in the world — and we built a network to support that. Some of the new enhancements are:

Massive scale with Virgo Network

This new non-blocking data center fabric removes latency barriers:

Virgo can link up-to 134,000 chips with 47 Petabits/sec of non-blocking bi-sectional bandwidth to deliver 1.7K Exaflops of compute.
With enhancements in Pathways and JAX, you can further connect these Virgo fabrics to scale to over 1 million TPU chips in a single training cluster.
We are also making Virgo Network available on NVIDIA Vera Rubin NVL72, supporting up to 960,000 GPUs.

For more on Virgo Network, check out this blog.

Accelerator network profiles

It’s easier than ever to handle the complex networking prerequisites for accelerator-equipped GKE node pools with DRANET, which improves bandwidth for distributed AI/ML workloads by up to 60% (GA).

AI-native Cloud Interconnect

SLA-backed, and optimized for efficiency, Cloud Interconnect supports petabit-scale data transfers and is available with a fixed price option. Cloud Interconnect now supports:

400 Gbps circuits with up to 3.2 Tbps in a single connection (GA)
Partner Cross-Cloud Interconnect for AWS (GA), CoreWeave (in preview soon), and Lumen (in preview soon)

Cross-Cloud Network for AI and core applications

The Cross-Cloud Network helps ensure you can securely connect users, data, locations, applications, services, and infrastructure anywhere in the world, at planetary scale. We designed our global multi-shard network to scale horizontally to meet the demands of the AI era and enable us to accommodate our 10x WAN traffic growth from 2020 to 2025.

These are some of the improvements we’re making to the Cross-Cloud Network:

Ultra Low Latency Solution for financial exchanges

In partnership with CME Group, we are bringing the world's leading derivatives marketplace to Google Cloud. To support CME Group’s performance requirements, we developed an ultra low latency (ULL) networking and compute solution. This fully managed cloud environment will allow CME Group and its clients to migrate its core trading systems to Google Cloud.

Now in preview, the solution is designed to meet the unique and exacting requirements of running financial exchanges in the cloud. It includes several new technologies:

Deterministic high-performance compute powered by ULL networking, with bare metal and VM form factors, delivers a comprehensive portfolio for your trading compute needs.
Scalable multicast data distribution with hardware-based ultra-low latency enables reliable one-to-many market data sharing.
Nanosecond-level clock sync enabled by Firefly, a novel clock synchronization system. Firefly achieves sub-10ns NIC-to-NIC synchronization to support high-frequency trading.
Advanced network observability with 64-bit nanosecond timestamps, support for multiple traffic-mirroring destinations and multicast traffic, and support for auditing and regulatory requirements.
Low-latency inference allowing exchange participants to connect their AI-driven services to the exchange’s infrastructure.

“The Google Cloud Ultra Low Latency Solution provides the level of performance necessary for CME Group futures and options markets to run in the cloud, expanding access to clients worldwide.” - Sunil Cutinho, CIO, CME Group

Cross-cloud observability for networks, applications, and agents

Whether you’re running core applications or new AI agents, you need visibility into your network infrastructure. Cloud Network Insights, now in preview, offers network performance monitoring (NPM) and digital experience monitoring (DEM) to dramatically reduce the mean time to detect and mitigate network-related agent, application, and API issues.

Cloud Network Insights is enabled by technologies from Broadcom’s AppNeta and powered by AI-enabling natural language queries through Gemini Cloud Assist.

"In an environment as complex and high-scale as Sabre’s, total visibility isn't just a luxury — it's a requirement for operational resilience. Cloud Network Insights will enable us to further shift our posture from reactive troubleshooting to proactive optimization. By providing granular, real-time telemetry across our global cloud footprint, it helps eliminate the traditional 'black box' of the network, allowing our teams to resolve bottlenecks before they impact the traveler experience." - Alfredo Rodriguez, VP Cloud Platform Infrastructure, Sabre Corporation

“Cloud Network Insights closes the 'visibility gap' between the private corporate network and the public cloud, empowering our joint customers to pinpoint performance bottlenecks in seconds rather than hours.” - Alan Davidson, CIO, Broadcom

Cross-Cloud Network for distributed applications

Multicloud and hybrid networks require secure, reliable, and high-performance connectivity. New enhancements for our foundational networking services and tools include:

Private Service Connect

Private Service Connect traffic volume grew 4x in 2025 and it now supports 40+ Google and third-party published services, enabling secure private global access to your managed services.
Private Service Connect endpoint-based security allows for granular authorization policies for producer-to-consumer service communications (preview).
Gemini Cloud Assist for Private Service Connect provides for automated troubleshooting (preview).

Cloud-native IP address management (IPAM)

Cloud Number Registry is an IPAM solution powered by agentic technologies. Network admins can easily find free IP ranges, track utilization, and allocate resources (preview). It also integrates with Infoblox Universal DDI for Cross-Cloud Network IPAM discovery and enforcement.
Hybrid Subnets allow you to migrate legacy workloads from on-premises to a VPC without needing to change hard-coded IP addresses (GA).
Cloud NAT allows you to connect your IPv6-only workloads to private IPv4 destinations using the combined power of DNS64 and private NAT64 (in preview soon).

Network Connectivity Center (NCC)

Partner Cross-Cloud Interconnect for AWS is available as a connectivity type in NCC (preview).
Support for static routes using an internal load balancer as the next hop allows the integration of Secure Web Proxy and third-party network security virtual appliances (GA).
Support for privately used public IP (PUPI) allows the exchange of PUPI IPv4 addresses with VPC spokes and producer VPC spokes (GA).

Granular networking charge visibility

Cost Explorer and the new App Optimize API now provide attribution of associated Data Transfer costs to the originating resources for Google Cloud products (in preview soon).

Cross-Cloud Network for internet-facing services

As part of Cross-Cloud Network, the Global Front End simplifies how you deliver, scale, and protect web, API, and AI workloads. New capabilities include:

Global Front End Enterprise delivers simplified consumption by combining capabilities from global Cloud Load Balancing, Google Cloud Armor, Cloud CDN, and Service Extensions with up to 15% lower TCO (in preview soon).
Post quantum cryptography (PQC) helps secure your workloads with industry-standard algorithms that provide a layered defense against both classical and quantum adversaries.
Google tag gateway, enabling advertisers to serve tags from their own domain, which can significantly improve the accuracy and resilience of measurement signals (GA soon).

In addition, Cloud CDN, an important part of the Global Front End, now offers:

Built-in image optimization to help you deliver content that best fits your end users’ screens and saves on bandwidth costs (in preview soon).
GKE Gateway support so you can enable and manage caching services using GKE APIs (GA).

Cross-Cloud Network’s Cloud WAN for global enterprises

Cloud WAN is a fully managed, reliable global backbone to connect your enterprise. New capabilities include:

Expanded geographic reach: Our network spans more than 10 million kilometers of terrestrial and subsea fiber, and Network Connectivity Center’s site-to-site data transfer is now available in over 25 countries.
NCC Gateway enables third-party secure service edge (SSE) integrations from Palo Alto Networks (GA soon) and Symantec (preview).
The Verified Peering Provider program, which offers highly reliable internet connectivity to Google, now has dramatically expanded availability through 175+ providers worldwide.
Last mile connectivity: Provision site-to-cloud private connectivity in minutes with preferred partners from the Google Cloud console (in preview soon).

“Cloud WAN enables Dun & Bradstreet to evolve our global network via composable, cloud-native constructs. Leveraging NCC, we’ve built a resilient, high-performance platform that simplifies operations and optimizes costs. This foundation supports continued modernization and AI-driven workloads. We expect to extend this architecture as new patterns emerge, maintaining our blueprints-first approach.” - Josh Barry, VP, Network Engineering, Dun & Bradstreet

AI-powered security against evolving threats

The threat landscape is evolving faster than ever, with AI-driven attacks. Staying ahead requires the latest defenses. Cross-Cloud Network relies on Cloud NGFW and Cloud Armor for advanced security capabilities. Here’s the latest on those offerings.

Cloud NGFW

Advanced malware sandbox uses AI models trained on data from 70k+ customers to stop 99% of known and unknown malware, including evasive zero-days. Advanced malware sandbox is powered by Palo Alto Networks Advanced Wildfire (in preview soon).
Internal Application and proxy Network Load Balancer support helps to enforce consistent, service-centric security for abstracted services like GKE, Cloud Run, and Private Service Connect traffic (preview).
Project-level policies allow for creating and managing Cloud NGFW endpoints, security profiles, and security profile groups at the project level (in preview soon).

Cloud Armor

Managed rules, built-in rulesets across 15 threat categories, deliver automated threat protection against a broad set of attacks and zero-day CVEs. This is powered by Thales Imperva based on visibility to 1.5 trillion web requests each month (in preview soon).
Google Cloud Fraud Defense integration helps to discern the legitimacy and authorization of bots, humans, and agents. Fraud Defense is the evolution of reCAPTCHA, which protects over 14 million domains from fraud and abuse.
Adaptive protection for Network Load Balancers & VMs brings advanced machine learning to L3/L4 traffic, to detect and mitigate volumetric DDoS attacks (in preview soon).
A simplified user experience with a visual rule builder makes custom rule creation easier (in preview soon).

AI-powered network operations

Finally, new AI-powered technologies in Gemini Cloud Assist can help automate manual tasks, ease troubleshooting, predict reliability issues, improve security, and help optimize your network to reduce toil and improve reliability with new specialist agents. These include:

A network security agent that streamlines network security operations by assisting with policy generation, recommendations, and impact analysis (in preview soon).
A network agent that optimizes workload placement for performance and reliability, and also provides advanced cost estimation for observability services (in preview soon).

Additionally, to enable customers and partners to build their own agents, we are releasing Network observability MCP tools and agent skills. This will allow their agents to leverage connectivity tests, and allows for natural language querying of VPC Flow Logs (both in preview).

The network that scales with you

We built our Cross-Cloud Network on the same global infrastructure that powers Google’s largest AI and internet services. This provides you with a blazing-fast, planet-scale foundation that is both secure by design and open by principle, allowing you to integrate your trusted partners across any environment.

As we move into the agentic era, our flexible, future-proof solutions ensure you can quickly adopt the latest AI technologies while maintaining the reliability of your core applications.

Whatever comes next, we’ve built the network to help you lead it. Attend our networking sessions at Next ’26 to learn more, or learn more about the Cross-Cloud Network!

What’s new in GKE at Next ‘26

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

This week at Google Cloud Next ‘26, we are sharing the evolution of Google Kubernetes Engine (GKE), delivering leading performance, efficiency, security, and scale for your most demanding and complex workloads, and the next generation of AI and agentic applications.

Why it matters: Kubernetes has rapidly become the operating system for the AI era, with GKE now powering AI workloads for all of our top 50 customers on the platform, including the largest frontier model builders. We are witnessing a massive acceleration in enterprise AI. In just a few months, the number of multi-agent AI workflows has surged by 327%. At the same time, 66% of organizations rely on Kubernetes to power generative AI apps and agents.

This new era of autonomous agents operating at massive scale requires a foundational change in how we manage infrastructure — a change that is more demanding than the shift from stateless to stateful applications.

What’s new:

GKE Agent Sandbox: Secure, highly scalable, low-latency agent infrastructure
GKE hypercluster: A single, conformant GKE control plane to manage millions of accelerators across Google Cloud regions
Improved inference performance: Foundational enhancements to GKE Inference Gateway and KV Cache management
Reinforcement learning (RL) enhancers: Native capabilities to relieve bottlenecks that throttle accelerator utilization
Scaling on custom metrics: Support for intent-based autoscaling on triggers besides CPU and memory

Read on for details about these GKE announcements.

GKE Agent Sandbox: Accelerating the agentic era

As AI evolves from simple conversational chatbots to entire ecosystems of proactive, autonomous agents, the underlying infrastructure must adapt to handle hundreds or thousands of agents collaborating with workers to plan, evaluate, and execute complex tasks. At scale, infrastructure performance, responsiveness, and rigorous security are essential.

We are excited to announce GKE Agent Sandbox, the industry’s most scalable and low-latency agent infrastructure. Built with gVisor kernel isolation — the same technology securing Gemini — Agent Sandbox allows you to safely execute untrusted code, tools, and entire agents without sacrificing performance. GKE provides leading speed and efficiency for fully isolated agents with 300 sandboxes per second at sub-second latency and up to 30% better price-performance when running on Axion compared to other hyperscale clouds.

Lovable empowers anyone to build apps and websites — with builders creating 200,000+ new projects daily. Lovable runs these AI-generated applications in GKE Agent Sandboxes because of the fast startup, fast scaling and secure isolation.

GKE's cutting-edge sandboxing capabilities allow us to reliably scale to hundreds of secure sandboxes per second, ensuring we can seamlessly empower builders, even during massive, unpredictable demand." - Fabian Hedin, Co-founder, Lovable

GKE hypercluster redefines the scalability ceiling

As foundational AI models grow exponentially and accelerators remain in high demand, organizations resort to fracturing Kubernetes compute infrastructure into hundreds of disconnected clusters, which can create a massive operational burden. To help, we’re announcing the private GA of GKE hypercluster, which allows a single, Kubernetes conformant GKE control plane to manage a million chips distributed across 256,000 nodes — spanning multiple Google Cloud regions. With the GKE hypercluster, widely distributed infrastructure becomes a single, unified capacity reserve that spans geographical locations.

To scale globally without compromising security, GKE hypercluster relies on Google’s Titanium Intelligence Enclave, a software-hardened security engine that delivers private AI compute. This "no-admin-access" model provides hardware-attested, pod-level isolation, so that proprietary model weights and prompts remain cryptographically sealed from platform administrators and infrastructure layers.

Supercharging state-of-the-art inference

Achieving frontier inference requires months of complex performance tuning. To reduce this heavy lifting, GKE now slashes your "time to SOTA" across TPUs and GPUs to mere minutes. We do this with new capabilities:

ML-driven Predictive Latency Boost in GKE Inference Gateway, which can reduce time-to-first-token latency by up 70% by replacing heuristic guesswork with real-time capacity-aware routing — no manual tuning required.
Automatic KV Cache storage tiering across RAM, Local SSD, and GCS/Lustre solves long-context memory bottlenecks. Offloading KV Cache to RAM yielded a more than 40% TTFT reduction and a 50% throughput gain for a 10K system prompt length. Offloading KV Cache to Local SSD yielded an almost 70% throughput improvement for a 50K system prompt length. Learn more about these benchmarks in the llm-d Offloading Prefix Cache to Shared Storage guide.

Built as part of a layered composable suite, these new GKE capabilities leverage llm-d, now an official CNCF Sandbox project. To give you maximum flexibility, we’ve partnered closely with NVIDIA to seamlessly integrate Dynamo for scaling massive Mixture-of-Experts (MoE) models. Whichever tools you choose, GKE provides the highly-optimized, flexible infrastructure you need to safely run any frontier AI workload — including the advanced agentic capabilities of the newly announced Gemma 4.

Eliminating RL compute bottlenecks

Reinforcement learning (RL) is a key driver of AI compute demand and RL jobs involve sequential processing for sampling, reward, and training that can leave GPU and TPU accelerators idle between these RL steps. To streamline RL, we are adding new GKE capabilities in preview:

RL Scheduler solves for the "straggler effect" and inter-batch tail latency, maximizing throughput via intelligent routing.
RL Sandbox provides kernel-level isolation for tool-calling and reward evaluation with millisecond-scale provisioning. Easy integration with RL sampling and reward steps.
RL Observability and Reliability dashboards offer the deep visibility required to troubleshoot and optimize the entire RL loop instantly, out of the box.

Review the RL on GKE recipe, specifically the implementations for Verl and NeMo RL.

Intent-based autoscaling on custom metrics

Traditionally, scaling AI workloads based on application health has imposed a "custom metric tax." To scale the system on anything but basic compute or memory utilization, organizations have to manage complex monitoring systems and IAM roles. This creates operational risk: if your external observability stack fails, your autoscaling breaks along with it.

Intent-based autoscaling eliminates this overhead via native custom metrics support for GKE’s Horizontal Pod Autoscaler (HPA). This agentless architecture bypasses external dependencies by sourcing metrics directly from Pods, hardening reliability while cutting costs. Crucially, it drops reaction times from 25 seconds to just 5 seconds—a 5x performance gain for near-instantaneous infrastructure elasticity.

New workloads, same mission

For over a decade, GKE has set the standard for scalable infrastructure. As we enter the era of agentic and autonomous AI, our mission remains the same: eliminating operational friction so you can focus on innovation. The capabilities we are announcing at Next ‘26 — from GKE hypercluster and the Agent Sandbox, to ultra-fast inference and intent-based autoscaling — give you the secure, efficient, and powerful engine you need to succeed with your ambitious AI workloads. To learn more about using GKE for your AI workloads, check out GKE Inference Quickstart.

Introducing the Google Cloud Knowledge Catalog

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

Traditional data catalogs were built as manual inventories for technical users, focusing on table structures rather than the deep context that AI agents need. When agents lack business semantics and data relationships, this triggers hallucinations, high latency, and stale insights.

To address this problem, we are evolving Dataplex into a dynamic, always-on Knowledge Catalog that serves as the universal context engine for your enterprise, helping agents execute complex tasks with accuracy.

Customers like Bloomberg Media are already using the Knowledge Catalog to power agents with trusted context:

“By unifying Bloomberg Media’s enterprise metadata and business context through the Knowledge Catalog, we successfully launched our Data Access AI Agent. This internal solution empowers stakeholders across the organization to intuitively explore our data lake, translating complex business inquiries into instant, AI-driven narratives. Crucially, by grounding our AI in trusted institutional context, we ensure confidence in the accuracy and quality of every insight generated."

— William Anderson, CTO, Bloomberg Media

The Knowledge Catalog operates on three foundational pillars:

Aggregation: Unifying context and resolving conflicting definitions
Enrichment: Generating continuous meaning and mapping relationships
Search: Empowering agents with high-precision retrieval

Aggregation: Unifying context across your data estate

To build true context, you must bring it together from everywhere. The Knowledge Catalog aggregates native context across your Google and partner data platforms, semantic models, and third-party catalogs, unifying them into a single, governed source of truth.

Broad metadata aggregation (GA): To build a truly comprehensive context engine, you must leave no silo behind. The Knowledge Catalog automatically harvests technical metadata across your foundational systems — including BigQuery, AlloyDB, Spanner, Cloud SQL, Firestore (Preview), and Looker (Preview). It also supports integrations with third-party databases and partner catalogs like Atlan, Collibra, Datahub, Ab Initio, and Anomalo, ensuring that even your legacy metadata is brought into the agentic fold.
Enterprise connectivity (Preview): To truly understand your operations, semantic context must cover all the key systems in your enterprise. Using Google Cloud Lakehouse, we are interconnecting these systems with context federation, and Knowledge Catalog gains full and immediate visibility to applications, operating systems, and AI platforms — including Palantir, Salesforce Data360, SAP, ServiceNow, and Workday. For example, your SAP data products are automatically mapped to the Knowledge Catalog.
BigQuery measures (Preview): We are redefining data consistency by embedding programmatic business logic directly into the SQL engine. BigQuery measures ensure every calculation is universally reusable and mathematically accurate. The Knowledge Catalog acts as the ultimate aggregator, pulling BigQuery measures and LookML together into a single, governed semantic foundation.
Data products (GA): Data products package data assets and context that grounds agents and makes them reliable in production. These self-contained blocks include built-in intent, SLAs, and governance constraints, providing the essential building blocks to scale complex AI use cases.

Enrichment: Generating meaning through continuous learning

The Knowledge Catalog provides continuous data enrichment — going beyond manual curation to actively mine structured schemas, query logs, and BI semantic models while extracting entity relationships from unstructured data. We are delivering this continuous enrichment where your teams work:

Smart Storage and Object Context API (Preview): Built natively into Google Cloud Storage (GCS), Smart Storage automatically tags, embeds, and enriches files with metadata as soon as they land in your buckets. By integrating this intelligence feature with the Knowledge Catalog, unstructured data is instantly discoverable by agents.
Deep multimodal metadata extraction (Preview): For collections of complex unstructured data, the Knowledge Catalog natively integrates with Gemini to identify useful business information and automatically build pipelines that extract entities and map complex business relationships directly from unstructured content.
Automated context curation (Preview): The Knowledge Catalog automatically generates natural language descriptions, including business glossaries, for datasets, data products, relationships, and verified SQL patterns that allow both humans and agents to interact with data without guesswork. By inferring these hidden relationships and intent-based patterns, it constructs a dynamic, evolving map of how data actually relates to the business.
Verified queries and semantic guardrails (Preview): One leading cause of AI failure is hallucinated logic and guessed SQL joins. To prevent this, the catalog provides verified SQL patterns and pre-generated natural language questions.

Search: Unleashing agents with high-precision, secure retrieval

Creating a massive context layer is great, but in the agentic era, search has evolved to be the new query path. When autonomous agents are working on your behalf, they are iterating incredibly fast. The hardest problems at enterprise scale are speed, relevance, global reach, and security.

High-precision semantic search (GA): The Knowledge Catalog uses a hybrid search stack leveraging decades of Google innovation. Built on the same advanced query-rewriting and machine-learning technologies that power Google Search, it delivers the sub-second latency and pinpoint relevance that agents need. When an agent receives a prompt, the catalog instantly ranks and returns the right context to agents in real-time.
Access control-aware search: The ability to find the right data and its corresponding context is critical in the agentic era; if an agent retrieves the wrong context, it hallucinates. To gain trust, our global search respects metadata access permissions as defined in the source systems, ensuring agents can only retrieve and act on the assets they are explicitly authorized to see.
Measurable context evaluation: To ensure long-term accuracy, we are augmenting our search capabilities with a robust evaluation framework. This transforms context construction from a guessing game into a measurable engineering discipline. It allows your teams to quantitatively test and iterate on various context construction strategies, ensuring continuous optimization of the relevance and quality of the context feeding your agents.

With foundational data products, high-precision search, and guardrails in place, we can deploy advanced AI reliably. A prime example is the Deep Research Agent in Gemini Enterprise, powered by the Knowledge Catalog (Preview). Now natively powered by the Knowledge Catalog, this agent synthesizes live business data, internal documents, and web research to answer highly complex questions. It delivers deterministic precision and deep citations, executing tasks in minutes that previously required weeks of manual effort.

Stop forcing your agents to guess the unwritten rules of your business. Build the context once, and unleash your agents to do the rest.

Get started today with Knowledge Catalog.

The future of data lakehouse: Open and interoperable for the agentic era

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

Traditional lakehouses were engineered for the era of reporting, not the high-velocity, multimodal demands of AI agents. To bridge this gap, architecture must evolve into an AI-native foundation — one that replaces batch processing with continuous feedback loops and live data streams. This shift gives agents the reliable context they need to transform raw data into action and unlock all enterprise data (structured and unstructured) across cloud boundaries.

Today, we announced our next-generation cross-cloud Lakehouse that delivers four core breakthroughs:

Fully managed Iceberg storage with enterprise-grade features, giving you the benefits of open-source flexibility plus performance, scale, governance, and multimodal processing.
New cross-cloud interoperability, bringing Google’s high-performance, scalable foundation and AI capabilities to your data, supporting an expansive data ecosystem.
A high-performance Apache Spark experience, accelerating your data science workloads with exceptional performance and your choice of developer environments.
AI-powered, always-on context, enabling AI agents to reason in real time across operational and analytical data.

This agentic-first lakehouse approach can deliver an estimated 117% ROI with payback in under six months. Spotify is already unlocking innovation with Google Cloud’s Lakehouse.

“Spotify is leveraging Google Cloud’s Apache Iceberg products as part of our efforts to build a truly modern data lakehouse that removes the silos between our data lakes and warehouses. This architecture provides us with an interoperable and abstracted storage interface, allowing our teams to process the same data across BigQuery, Dataflow, and other open-source engines without duplication. It will simplify our governance and unlock the ability to innovate at a scale that was previously impossible.”
— Ed Byne, Product Manager, Spotify

Accenture, a key partner in this journey, sees this as a fundamental shift in how enterprises operate:

“To reinvent enterprise operations, organizations must collapse the data boundaries that fragment their intelligence. By utilizing the Google Cloud Lakehouse and 'zero-copy' innovation, we can help customers activate agentic AI with surgical precision. Whether leveraging high-performance Apache Spark for complex data science or delivering scale for industries like retail and life sciences, this AI-native foundation transforms trapped data into real-time action."
— Scott Alfieri, Global Lead, Accenture Google Business Group

Openness without compromise

With Google Cloud’s unique, vertically integrated infrastructure, you get the open-source flexibility of Apache Iceberg backed by a fully integrated, managed data-to-AI experience. You can manage all your multimodal data with unified governance, and get your data estate ready for agents with always-on context. By connecting your Iceberg tables directly to engines like BigQuery and Managed Service for Apache Spark, you can accelerate your AI workloads in real time.

Today, we announced four new innovations to make your Iceberg experience on Google Cloud even stronger:

Fully managed Iceberg storage with read/write interoperability: Experience unified Apache Iceberg tables, managed via the Lakehouse runtime catalog (formerly BigLake metastore). This provides read and write interoperability between BigQuery and Managed Service for Apache Spark including Iceberg-compatible OSS engines like Spark, Trino, and Flink, and third-party engines like Databricks and Snowflake (Preview).
The power of BigQuery with Iceberg: Access advanced runtimes, automatic table management, partitioning, multi-table transactions, and history-based optimization for REST catalog tables (Preview) and Iceberg tables managed by BigQuery Catalog (GA).
A unified multimodal foundation: Use BigQuery ObjectRefs (GA) to merge unstructured data in Cloud Storage with structured data in Iceberg. This simplifies multimodal analysis and manages conversational insights through BigQuery AI.
Unified management and governance: Enhance enterprise trust with open lakehouse governance (Preview) via Knowledge Catalog (formerly Dataplex). Securely ground your agents with business context using end-to-end data lineage, search, quality profiling, and table-level access controls for your Iceberg estate.

Cross-cloud power without the friction

Cross-cloud is today’s enterprise reality, and your agents need a scalable solution to work across data, no matter where it is. Production cross-cloud data access often fails to scale due to high egress overheads and performance bottlenecks. So we are introducing a new high-performance, scalable cross-cloud experience by bringing Google’s AI capabilities to your AWS and Azure data that delivers similar price-performance characteristics to cloud-native solutions:

A new AI-native cross-cloud lakehouse: Lakehouse cross-cloud interconnect and cross-cloud caching (Preview) provides BigQuery and Managed Service for Apache Spark (formerly Dataproc) with high-performance access to AWS Iceberg data at scale. This new capability, powered by high-throughput, low-latency, cross-cloud connectivity and advanced cross-cloud query processing innovations, delivers price-performance characteristics similar to AWS-native data platform solutions. You can run Gemini-powered use cases, such as building agents in Gemini Enterprise and BigQuery AI functions, over your Amazon S3 Iceberg data.
An interoperable ecosystem powered by open standards: To help you easily discover and analyze all your enterprise data across any engine or cloud, we are launching Lakehouse catalog federation (Preview) for AWS Glue, Databricks, SAP, and Snowflake, with Confluent Tableflow coming later this year. This foundation enables simple access to data across clouds through BigQuery and Managed Spark, supported by an expanding partner ecosystem that includes bi-directional access for Databricks, Oracle Autonomous Database, and Snowflake pipeline support for dbt, serving with Clickhouse, and catalog integrations with Atlan and Datahub (preview). And, advanced Lakehouse Governance (Preview), guarantees that security protocols and access permissions are immediately enforced throughout this unified environment.

High-performance Spark, built for enterprise scale

Managed Service for Apache Spark offers a unified, high-performance experience that accelerates everything from data engineering to agentic AI development. It empowers data teams to extract maximum value from their enterprise data without friction by delivering key advantages:

Frictionless, agentic data science: Customers can gain a highly flexible data science environment, integrating Colab Enterprise, Gemini Enterprise, and local IDEs with BigQuery and managed Spark. This allows developers to run Python, Spark, and SQL on a single, unified copy of data, eliminating movement and optimizing engine choice while the Lakehouse runtime catalog Iceberg REST catalog endpoint (Preview) automates table management.
Better Spark processing: Lightning Engine for Apache Spark delivers up to 2x the price-performance over the leading high-speed Spark alternative. This engine uses vectorized execution, intelligent caching, and optimized I/O to provide industry-leading performance on Iceberg, Parquet, and Delta formats without requiring any code changes.

Built for the scale and speed required by agentic AI

Google Cloud’s Lakehouse is a high-performance, real-time foundation businesses can use to scale in the agentic era. We use AI to discover hidden relationships within your enterprise data to provide 24/7 curated context for your agents, making it easier to activate them instantly using databases with new integrations. This includes:

Always-on context for agents: Knowledge Catalog (formerly Dataplex) builds a unified foundation by aggregating business context from your entire data landscape, including Iceberg. It delivers continuous enrichment by learning how your enterprise actually uses data, using Smart Storage to automatically map complex relationships within unstructured files. To guarantee trust and relevance, it uses access-control-aware, high-precision search powered by Google Search innovations. This instantly identifies trusted context and feeds it to AI agents, ensuring they deliver reliable, grounded results.
Ready-to-use BigQuery and Looker agents: BigQuery provides built-in Conversational Analytics, a Data Engineering Agent, and a Data Science Agent (Preview) that work with your cross-cloud, multimodal data. Looker Conversational Analytics provides insights in natural language to your business users. You can build your own agents with your lakehouse as the foundation with Google-native tools like Agent Developer Kit (ADK) and Model Context Protocol (MCP).
Real-time analysis and agentic activation of operational data: Integrate operational data into your lakehouse. Spanner, AlloyDB, and Cloud SQL support real-time change replication into BigQuery (GA) and Iceberg (Preview). Analytical data in Iceberg can also be served with low latency using AlloyDB and Spanner (Preview).

Get started today

Built for the AI era, our cross-cloud Lakehouse delivers uncompromising open Iceberg storage, enables multi-cloud interoperability, and equips your AI agents with always-on context for closed-loop activation. Learn more and start building today.

Introducing Google Cloud Fraud Defense, the next evolution of reCAPTCHA

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

The agentic web — where autonomous AI agents reason, plan, and execute complex transactions using the open web and industry standard protocols — aims to create an autonomous customer experience. While these agents can significantly enhance online interactions, they also introduce new abuse and fraud vectors, creating unique challenges for security platforms.

This rise in sophisticated automation requires a fundamental shift in risk management. Today at Google Cloud Next, we are launching Google Cloud Fraud Defense, a trust platform for the agentic web. As the next evolution of reCAPTCHA, Fraud Defense is a comprehensive platform designed to verify the legitimacy of bots, humans, and AI agents, providing businesses with the intelligence needed to secure their digital interactions and commerce.

Agentic activity in the Fraud Defense dashboard.

As part of our mission to enable a safe agentic web, Fraud Defense introduces a powerful suite of capabilities that allow customers to measure and control agentic activity on their websites. By using the same global signals that protect Google’s own ecosystem, businesses can now enable trusted experiences for both human users and AI agents alike.

Creating policies for agentic traffic in the Fraud Defense policy engine.

These new capabilities include:

Agentic activity measurement: A new dashboard to help you measure and understand agentic activities. We are integrating with industry standards such as Web Bot Auth and SPIFEE, as well as using traditional methods, to identify, classify, and analyze agentic traffic, and connecting agent and human identities to better understand risk and trust.
Agentic policy engine: To provide you with granular control at different stages of the end user interaction across the entire journey, Fraud Defense’s agentic policy engine allows you to allow and block agents and users based on conditions that include risk scores, automation types, and agent identity.
AI-resistant challenge: As we identify potentially fraudulent behavior from agents, we enable application providers to deter and mitigate malicious requests by requesting humans to be in the loop using the new QR code-based challenge. This AI-resistant mitigation challenge to prove human presence is designed to make automated fraud economically unviable.

New QR-code challenge in a shopping website.

reCAPTCHA will continue to be the core bot defense pillar of the broader Fraud Defense platform. Existing reCAPTCHA customers are automatically Fraud Defense customers, with no migration required, no action needed, and no change to pricing. Your existing site keys and integrations remain exactly as they are today.

The trust platform for the agentic web

At Google Cloud, we believe preventing fraud and abuse in the agentic web should fundamentally result in a simpler customer experience. Fraud Defense uses a three-pronged approach to help enable a safe agentic web and drive business growth:

1. Preventing evolving threats
We protect your business with the same fraud intelligence that secures many of Google’s services. As threats shift from bot automation and invalid traffic to agent takeover and large-scale, AI-driven synthetic identity fraud, Fraud Defense identifies emerging threats before they reach your site.

This unrivaled visibility, built upon a massive fraud intelligence graph that already protects 50% of Fortune 100 companies and over 14 million domains globally, provides a level of collective immunity and verified trust that local data alone can not match.

2. Securing the customer journey
Attackers don’t target endpoints in isolation; they target digital journeys. This is even more true in the agentic web as agents are being tasked to perform end to end journeys. Fraud Defense provides a unified view of risk — from registration and login to payment and checkout.

By correlating telemetry across the entire lifecycle, our unified trust model identifies complex, multi-stage fraud campaigns that disconnected point solutions miss. This holistic view has demonstrated a 51% average reduction in account takeover (ATO) by accurately distinguishing between legitimate customer activity and sophisticated abuse.

3. Accelerating business growth
In the agentic economy, friction kills conversion. Fraud Defense is designed to be invisible for the majority of users, replacing disruptive puzzles with silent background verification. By using our intelligent trust model, we allow you to surgically block malicious bots, humans and agents, while confidently welcoming legitimate users, including AI shopping assistants that drive a projected 25% increase in average order value, according to the 2025 Shopify Retail Report.

Learn more about how Fraud Defense works

We invite you to join us at Next ‘26 to talk about new capabilities designed to help protect you as you continue your journey on the agentic web. While you’re there, be sure to attend our breakout session and visit our demo pod, where you can see Fraud Defense in action and learn more directly from our experts. We look forward to meeting you there and discussing how we can safeguard your organization’s future in this changing landscape.

To take the next step in your journey to the agentic web, please check out the Fraud Defense website and log into the console. You can follow all of our security announcements at Next ‘26 here.

What’s new in the Agentic Data Cloud: Powering the System of Action

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

Companies are shifting from gen AI that simply answers questions to autonomous agents that perceive, reason, and act on their behalf. Attempting to scale these agents on legacy stacks exposes structural failures that can lead to fractured governance, a persistent trust gap, and broken reasoning loops, all while causing costs to spiral.

To solve this, we’re introducing the Agentic Data Cloud: an AI-native architecture that evolves the enterprise data platform from a static repository into a dynamic reasoning engine. It closes the gap between thinking and doing, allowing AI agents to act on your business data and context. While last-generation systems of intelligence were built only for human scale, the Agentic Data Cloud is a System of Action, built for agent scale.

Leading organizations are already using the Agentic Data Cloud to deliver tangible value:

Vodafone has launched hundreds of agents to deliver uninterrupted service to their customers, which is expected to save them millions of euros every year.
American Express is moving a core on-premises data warehouse and hundreds of production applications to BigQuery to power trusted agentic commerce at scale.
Virgin Voyages is using over 1,000 specialized AI agents, including one that slashes mass itinerary rebooking from six hours to just 11 minutes.

Today, we’re announcing three new innovation areas powering our Agentic Data Cloud:

A universal context engine that provides agents with trusted business context to drive higher accuracy.
Agentic-first practitioner experiences to evolve the role of data practitioners and developers as orchestrators of agents.
An AI-native, cross-cloud lakehouse that eliminates data silos by connecting your entire data estate.

Enabling agents with a universal context engine

AI is only as smart as its context. If an agent doesn't understand your definition of "margin" or the intricate relationships in your supply chain, it’s forced to guess. In the age of the agentic enterprise, data alone is not enough, and the old model of governance is insufficient. This is why we evolved the Dataplex Universal Catalog into the Knowledge Catalog, which maps and infers business meaning across your entire data estate, using a rigorous framework of aggregation, continuous enrichment, and search.

Here’s how it works:

Aggregation: To build true context, you must bring it together from everywhere. We are aggregating native context across your Google Cloud and partner data platforms. This includes third-party catalogs, applications, operating systems, and AI platforms like Palantir, Salesforce Data360, SAP, ServiceNow, and Workday (Preview). By using our Lakehouse, your third-party data assets are automatically mapped to the Knowledge Catalog. For Google Cloud sources, we are automating business logic with the new LookML Agent (Preview) and BigQuery measures (Preview), which embeds that business logic natively into the platform.
Continuous enrichment: The Knowledge Catalog delivers continuous enrichment by analyzing usage logs across your organization and profiling data behind the scenes. It learns how your enterprise actually uses data, not just what it is. This extends to unstructured data. The moment a file lands in Google Cloud Storage, our Smart Storage (Preview) instantly tags and enriches images and, soon, PDF objects. The Knowledge Catalog also identifies useful collections of unstructured data and uses Gemini to automatically generate missing schemas, mapping complex relationships so your AI is no longer flying blind.
Search and retrieval: Creating a massive context layer is great, but in the agentic era, search has evolved to be the new query path. The hardest problems at enterprise scale are speed, relevance, global reach, and security. To solve this, the Knowledge Catalog uses a sophisticated hybrid search stack built on Google Search innovation. To deliver relevance, it combines semantic and lexical matching with intelligent, machine-learning-based re-ranking. To deliver trust, we are enforcing your security permissions natively with access-control-aware search, so agents can only retrieve and act on the assets they are authorized to see. This high-precision infrastructure instantly identifies trusted context and feeds it to specialized agents.

The Knowledge Catalog now powers the Deep Research Agent (Preview). Part of the suite of Google-made agents available in Gemini Enterprise, this agent can perform multi-step reasoning across Google Cloud data platforms, such as BigQuery, as well as internal documents and web assets to answer complex questions with citations and precision that previously required weeks of manual effort.

Agentic-first practitioner experiences

As we move to this new architecture, the data practitioner role shifts from writing manual pipelines to orchestrating intent-driven engineering.

We’re accelerating this transition with the Google Cloud Data Agent Kit (Preview). Rather than introducing a new interface, we are launching a portable suite of skills, tools, environment-specific extensions, and built-in plugins, that drop into the environments developers love. By meeting practitioners where they already build — including VS Code, Gemini CLI, Codex, and Claude Code — we turn your IDE, notebook, or agentic terminal into a native data environment. This enables your environment to autonomously orchestrate a wide range of business outcomes, automatically selecting the right frameworks (e.g., dbt, Apache Spark, or Apache Airflow) and generating production-ready code based on Google’s gold standards.

This kit doesn't just connect tools — it injects high-performance capabilities directly into the developer's flow, scaling to petabytes without moving data. In fact, the Data Agent Kit features the same skills and tools that power our own out-of-the-box agents, including:

Data Engineering Agent (GA): Builds complex pipeline transformations from scratch and enforces governance rules to keep bad data out of production.
Data Science Agent (GA): Automates the model lifecycle — from wrangling to training — scaling across BigQuery Dataframes and Serverless Apache Spark.
Database Observability Agent (Preview): Acts as a 24/7 guardian for your infrastructure, diagnosing root causes and executing database remediations.

To help ensure the smooth execution of agents, Google Cloud has fully embraced Model Context Protocol (MCP), which provides a secure, universal interface that allows any agent to discover and use your data assets across our core engines, including: BigQuery, Spanner (Preview), AlloyDB, Cloud SQL (GA), and Looker MCP (Preview). MCP for Google Cloud uses our security stack, governing agent interactions based on your existing IAM policies, VPC Service Controls, and data residency requirements.

We’re also reimagining the business user experience with Conversational Analytics, now supported across BigQuery (GA), Cloud SQL, Spanner, AlloyDB (Preview), and Looker (GA). Organizations can simply publish these custom analytical agents in Gemini Enterprise, enabling employees to chat with live data in a familiar interface. By removing the technical barriers, we’ve eliminated the weeks spent waiting for manual reports, allowing businesses to move at the speed of thought.

A cross-cloud foundation built for agentic scale

For an agent to act, it must have a fundamentally open foundation. If an agent is blocked by cross-cloud latency or trapped in a proprietary walled garden, its autonomy is broken. That’s why we are introducing a truly borderless, cross-cloud Lakehouse that liberates your data wherever it resides by:

Connecting analytical estates: We’re integrating Cross-Cloud Interconnect (CCI) directly into our data plane. By combining CCI’s dedicated, high-speed private networking with Apache Iceberg REST Catalog, we’re enabling connectivity across clouds that is low latency and eliminates massive egress fees. As a result, agents can use data across AWS and Azure as if it were local to Google Cloud with seamless cross-cloud access.
Ending proprietary silos: We’re championing open federation to end the era of proprietary catalogs by launching bi-directional federation (Preview). Powered by the Iceberg REST Catalog, engines can now read directly from Databricks Unity Catalog on Amazon S3 (Preview), Snowflake Polaris (Preview), and the AWS Glue Data Catalog on Amazon S3 (Preview). This is reinforced by enhanced Lakehouse Governance (Preview), which ensures your security policies and access controls apply instantly across this borderless environment.
Unlocking operational data: We’re announcing Spanner Omni (Preview), unchaining the most scalable, globally consistent database on the planet. For the first time, you can run the Spanner engine anywhere — across clouds, on-premises, or on your laptop — with the same capabilities used to run Google.
Bridging the insight to action gap: We’re also closing the gap between insight and action. Most "unified" data platforms force the creation of complex ETL pipelines that block agents from accessing your real-time data. With Lakehouse federation for AlloyDB (Preview), we’re removing these pipelines by providing protocol-level, zero-ETL synchronization to give agents access to deep analytical history with low latency in operational transactions.

Automating the future

Moving to agent scale generates orders of magnitude more workloads. To support this, we are announcing four major performance breakthroughs:

Lightning Engine for Apache Spark delivers up to 2x the price-performance over the proprietary market alternative.
Managed Lustre delivers up to 10 terabytes-per-second of throughput to make sure data moves quickly enough for demanding models.
Bigtable now supports an in-memory tier that delivers sub-millisecond read latency for real-time applications. This means you can finally eliminate separate, side-by-side caching layers.
BigQuery fluid scaling helps lower costs by up to 34% on average for autoscaling workloads, scaling up resources instantly when agents act, and scaling back when they don't.

Build your success on a System of Action

The era of passive observation is over. The future belongs to the System of Action, made possible by Google’s Agentic Data Cloud.

Ready to see what an Agentic Data Cloud can do for your business?
Sign up for a strategy workshop today on how to get your data ready to fuel autonomous agents through a System of Action.

Want to learn more about an Agentic Data Cloud?
Read our blueprint for turning passive data into proactive action.

Introducing Virgo Network, Google’s scale-out AI data center fabric

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

The AI era requires a fundamental rethink of physical cloud architecture — networking, in particular. With foundational model parameters growing exponentially, traditional general-purpose networks are reaching their breaking points. To fuel the next decade of machine learning, Google designed Virgo Network, a new megascale AI data center fabric that embraces a "campus-as-a-computer" philosophy, and that underpins our AI Hypercomputer.

Legacy network designs simply cannot handle some of the constraints of modern AI:

Massive scale: Training demands now exceed the power and space of a single data center, requiring unified, multi-data-center domains.
Explosive bandwidth growth: Because foundational model training is heavily network-bound, the required bandwidth per accelerator has surged significantly over the last few years, creating throughput and congestion bottlenecks for older architectures.
Synchronized bursts: Intense, millisecond-level traffic spikes (figure 1) put immense pressure on network buffers. The outcome is that even a single "straggler" node can throttle the entire cluster’s performance.
Low latency: ML serving requires fast, consistent response times to deliver real-time inference, making strict latency control a critical architectural constraint.

Figure 1: Sub-millisecond line-rate bursts of an AI training workload

Reimagining the data center network

Meeting the demands of the AI era requires a fundamental shift away from general-purpose network design towards a specialized flat, low-latency network architecture. To address the unique scale and latency constraints, we leverage our proven Jupiter network for north-south traffic and are introducing a new fabric for east-west communication. The resulting architecture consists of three distinct and specialized layers that operate as one unified compute domain:

Scale-up domain: A high-bandwidth, low-latency interconnect fabric designed for tightly coupled communication between accelerators within a single pod.
Scale-out accelerator fabric (east-west): A dedicated accelerator-to-accelerator remote direct memory access (RDMA) fabric optimized for massive horizontal scale across pods. This layer is engineered for deterministic latency and maximum resilience, to provide high “goodput” for the ML workload.
Jupiter front-end network (north-south): A high-capacity fabric that provides fast, reliable access to distributed storage and general-purpose compute resources. It ensures that data access does not become a bottleneck for training and serving workloads, and is also used to scale-across multiple sites for very large training runs.

This architectural decoupling provides key strategic advantages:

Independent evolution: We can evolve and upgrade each network domain independently, preventing system-wide disruptions while accelerating the innovation cycle.
Dedicated scale-out bandwidth: A non-blocking network delivers massive bisectional bandwidth to accelerators for critical training tasks.
ML and network co-design: The network is built in lockstep with each new generation of ML accelerators, helping ensure the fabric is matched to the hardware it supports.

Figure 2: Data center network architecture

Introducing Virgo Network: Megascale data center fabric

Virgo Network is a scale-out fabric designed for the extreme requirements of modern AI workloads. Built on high-radix switches that reduce network layers by allowing more ports per switch, it employs a flat, two-layer non-blocking topology. Compared with traditional datacenter networks, this significantly reduces latency by minimizing network tiers. It features a multi-planar design with independent control domains to connect accelerators (figure 3). The accelerator racks also connect with the Jupiter north-south fabric to access compute and storage services. Together, this streamlined architecture delivers the massive bisection bandwidth and deterministic low latency necessary for both distributed training and serving workloads.

Figure 3: Megascale data center fabric (Virgo Network)

Virgo Network is the foundation of our next-generation accelerator designs and delivers the following advantages:

Massive fabric scale: Virgo Network can link 134,000 chips (TPU 8t) with up to 47 petabits/sec of non-blocking bi-sectional bandwidth in a single fabric.
Generational performance leap: With up to 4x the bandwidth per accelerator (TPU 8t) over the previous generation, Virgo Network delivers the bandwidth you need to get the full power of every chip.
Predictable low latency: Virgo Network delivers 40% lower unloaded fabric latency for TPUs compared to previous generation leading to more predictable performance for latency sensitive AI workloads.

Improving reliability at scale

In a system supporting hundreds of thousands of chips, hardware failures are inevitable. Because a single faulty component can disrupt a synchronized training job, reliability at scale is a primary focus. To maximize workload goodput, we designed the Virgo Network architecture around fault isolation, deep observability, and the rapid mitigation of hangs and stragglers.

At this scale, system-wide resilience requires a solid network foundation. Virgo Network integrates independent switching planes that provide robust fault isolation, protecting cluster-wide goodput from being degraded by localized hardware failures.

Figure 4: How fail-stop and fail-slow impact MTTR

Building on this foundation, we optimize the software and orchestration stack to maximize mean-time between interruptions (MTBI) and minimize mean-time to recovery (MTTR) through two primary areas:

Observability: Reliability at scale requires high-fidelity visibility. We use sub-millisecond telemetry to monitor network systems. This deep visibility allows us to detect transient congestion, optimize buffer management, and pinpoint the root causes of slowdowns across the hardware and software stack.
Identifying stragglers and hangs: Proactive monitoring is critical for identifying nodes that are experiencing performance degradation (stragglers) or that have stopped responding completely (hangs). By rapidly localizing these bottlenecks, with automated straggler and newly added hang detection, we accelerate the training job and protect it from localized slowdowns.

The foundation of the AI Hypercomputer

Virgo Network is a reimagined scale-out data center network custom-built for the stringent demands of modern AI workloads. This flat, multi-planar architecture unifies accelerators across pods into a single compute domain, addressing the bandwidth and scale limitations of traditional networks. By providing robust fault isolation directly at the hardware level, Virgo Network serves as the foundation for system-wide resilience, protecting synchronized workloads from localized hardware faults.

Ultimately, Virgo Network delivers the scale, predictable latency, and reliability necessary to accelerate the agentic AI era. To learn more about how we are building infrastructure for the future of AI, visit our AI infrastructure solutions page, explore the technical documentation, or attend the dedicated breakout session at Google Cloud Next.

What’s new in Cloud Run at Next ‘26

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

From vibe-coded and large-scale apps to AI models and agents, Cloud Run delivers on-demand compute with zero overhead and pay-per-use pricing for all of your workloads. Last year, the number of external active developers and applications on Cloud Run doubled, with more new customers and apps coming to Cloud Run in 2025 than in its first 6 years combined!

Today, we’re announcing new features and improvements to Cloud Run to help you run your workloads:

Build and deploy full-stack apps in Google AI Studio with Cloud Run, Firestore, and user authentication.
Build, scale, govern, and optimize reliable AI agents with the all new Gemini Enterprise Agent Platform and Cloud Run.
Enable easy deployments from developers and agents with Cloud Run's fully managed remote MCP server.
Combine high-performance inference and serverless compute with NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs on Cloud Run.

Empowering the new era of developers

For decades, software development has had an inaccessible learning curve, but thanks to AI, anyone can be a digital builder. With Cloud Run, you can go from prototype to deployed app in seconds.

Build full-stack apps in Google AI Studio
AI Studio now supports full-stack applications that can run server-side code, a Firestore database, and user authentication. Deploy your vibe-coded apps with a single click to Cloud Run, now generally available.

Cloud Run's fully managed remote MCP server
To make it even easier for developers or agents to deploy code, we are launching an official remote Cloud Run MCP (Model Context Protocol) server, giving you the tools to manage and deploy apps. Now GA.

Billing caps
Soon, you’ll be able to define your maximum spend per month. If your bill reaches this amount, your Cloud Run resources will be de-activated.

"Cloud Run has been one of the best technical choices we made in our deployments platform. It is our primary target, powering us to over 1 million live projects being hosted on Replit." - Scott Kennedy, VP of Engineering, Replit

Embracing the agentic era

AI agents are just like people in that they need access to a compute environment to perform their tasks. For a cloud-based AI agent to take complex actions, it can use Cloud Run’s on-demand compute service.

Cloud Run integration with Gemini Enterprise Agent Platform
Through its integration with Cloud Run, Agent Platform helps agents transition from experimental environments into fully managed, production-grade systems without having to rebuild them. Now in preview with select customers.

Cloud Run instances
Traditionally, Cloud Run services, jobs, or worker pools have been opinionated ways to manage Cloud Run infrastructure. Now, we are giving you access to the underlying primitive: you can create individual Cloud Run instances. Coupled with Cloud Storage volume mounts, these instances are ideal for hosting long-running background agents like OpenClaw in one simple command:

code_block: <ListValue: [StructValue([('code', 'gcloud run instances create \\\r\n --image alpine/openclaw:latest \\\r\n --port 18789 \\\r\n --memory 4Gi \\\r\n --default-url \\\r\n --add-volume mount-path=/home/node/.openclaw,type=cloud-storage,bucket=$BUCKET_NAME'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f93c8a1f8e0>)])]>

This functionality is available in preview to select customers.

Cloud Run sandboxes
Agents often need a safe place to execute code or other commands as quickly as possible. Coming soon, while processing a single request, you will be able to very quickly spin up an ephemeral sandbox that’s strictly isolated from your agent code using a built-in sandbox tool:

code_block: <ListValue: [StructValue([('code', 'app.post(\'/execute\', (req, res) => {\r\n const escapedCode = req.body.code.replace(/"/g, \'\\\\"\');\r\n\r\n exec(`sandbox do -- /usr/bin/python3 -c "${escapedCode}"`, (e, stdout, stderr) => {\r\n res.send({ stdout, stderr });\r\n });\r\n\r\n});'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f93c8a1f940>)])]>

"Cloud Run’s concurrency model has been instrumental in simplifying our AI workloads for our customer service AI tool, Lumi. Using Cloud Run alongside Gemini and AlloyDB, we’ve created a unified action layer that enables real-time call summarization and flow guidance, leading to improved first-call resolution rates and faster onboarding for our contact center team." - Edward Wright, Head of Engineering, VirginMedia O2 UK

Automatic scaling for high-demand applications

Cloud Run automatically scales to meet your demand, making it a great fit for large customers that need to serve and respond instantly to heavy traffic spikes.

SSH support for Cloud Run
Developers can now gain secure shell access (SSH) directly into a running Cloud Run container, enabling advanced troubleshooting and inspecting the container's file system on the fly, now in preview with select customers. Open a secure interactive shell session with simple command:

code_block: <ListValue: [StructValue([('code', 'gcloud run services ssh SERVICE'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7f93c8a1f9a0>)])]>

Cloud Run service bindings
Coming soon, unlock seamless service-to-service communication for your scalable microservices architectures with Cloud Run service bindings.

“Cloud Run’s serverless architecture empowers us to meet exponentially growing demand through near-instant scaling, while its streamlined developer experience simplifies building and running our applications.” - Mimi Chen, Member of Technical Staff, Anthropic

Running AI models

From serving custom models with GPUs to training and fine-tuning models with jobs, you can use Cloud Run for your most AI-intensive workloads.

Support for NVIDIA RTX PRO 6000 Blackwell GPU on Cloud Run
We’re bringing the serverless experience to high-end inference with support for NVIDIA RTX PRO™ 6000 Blackwell GPUs on Cloud Run, now GA. This means you can serve up to 70B+ parameter models without having to manage any underlying infrastructure, including scaling to zero when the resource is not in use.

Ephemeral disk
With per-instance temporary disk storage, workloads can process large files or use scratch space without eating up your container memory. Now in preview, ephemeral disk storage is created when an instance starts and deleted when it stops.

"Cloud Run has fundamentally changed how we manage our model deployments. By moving to a usage-based, scale-to-zero model, we’ve eliminated idle GPU costs for low-traffic models. We are now running over 17 model variants in production across multiple regions, each independently deployable and isolated, without the burden of capacity planning or fleet management." - Ajay Nair, Global VP, Elastic

On-demand compute for every workload

Whether you’re a seasoned software developer or a vibe coder looking to deploy the next viral app, Cloud Run delivers on-demand compute for everyone and every workload. Get started today.

Introducing Gemini Enterprise Agent Platform, powering the next wave of agents

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

In the early days of generative AI, building safe and reliable business tools took massive engineering effort and a high tolerance for trial and error. We helped solve that with Vertex AI, our trusted AI development platform. But today, we’re managing a different level of complexity, with agents interacting across multiple systems — and often without security and governance guardrails.

To move toward a truly autonomous enterprise, one where agents can act with the same independence and reliability as a member of your team, you need a foundation that can sustain that level of trust.

What’s new: Today, we’re launching Gemini Enterprise Agent Platform — our new, comprehensive platform to build, scale, govern, and optimize agents. It’s the evolution of Vertex AI, bringing the model selection, model building, and agent building capabilities that customers love, together with new features for agent integration, DevOps, orchestration, and security.

Agent Platform provides a single destination for your technical teams to build agents that can transform your products, services, and operations. These agents can be seamlessly delivered to your employees through the Gemini Enterprise app, all while remaining tightly integrated with your IT operations to help ensure control, governance, and security as you scale.

The platform also provides first-class access to more than 200 of the world’s leading models through Model Garden. This includes our latest first-party breakthroughs like Gemini 3.1 Pro, Gemini 3.1 Flash Image, and Lyria 3, alongside our open models like Gemma 4. And, of course, customers have full flexibility to use the best model for the job with support for third-party models like Anthropic’s Claude Opus, Sonnet and Haiku.

Moving forward, all Vertex AI services and roadmap evolutions will be delivered exclusively through the Agent Platform, rather than as a standalone service, to power the next generation of agent development.

Why Agent Platform matters for your business: Agent Platform helps you move from managing individual AI tasks to delegating business outcomes with total confidence. You can:

Build: Choose the right environment for the job — from the low-code, visual interface of the new Agent Studio, to the code-first logic of the upgraded Agent Development Kit (ADK). We’ve simplified the entire lifecycle with AI-native coding capabilities to help you ship production-grade agents faster.
Scale: Clear the path to production with the re-engineered Agent Runtime. This supports long-running agents that maintain state for days at a time and are backed by Memory Bank for persistent, long-term context.
Govern: Establish centralized control with Agent Identity, Agent Registry, and Agent Gateway. These capabilities help ensure every agent — whether built on Agent Platform or sourced from our partner ecosystem — has a trackable identity and operates within enterprise-grade guardrails.
Optimize: Guarantee quality with Agent Simulation, Agent Evaluation, and Agent Observability. These tools provide full execution traces and a real-time lens into agent reasoning to help ensure your agents always hit their goals.

Get started with Agent Platform: Visit Agent Platform in the Google Cloud console to explore new features and start building today.

Keep reading for a deeper look at our latest releases and how Agent Platform helps you deliver the production-ready agents you can trust at every stage of the journey.

How customers are achieving more with Gemini Enterprise Agent Platform

"Burns & McDonnell uses Agent Platform to transform how organizational knowledge is applied across the enterprise. Using ADK, we are building an AI agent that turns decades of project data into real-time, actionable intelligence. Agent Platform enables this innovation to scale responsibly by combining deterministic business rules with probabilistic reasoning — making AI a trusted operational capability, not just a productivity tool. With Agent Platform, we aren’t just managing knowledge; we are activating experience to drive faster, more confident decisions." – Matt Olson, Chief Innovation Officer, Burns & McDonnell

“Color Health uses Agent Platform to power our Virtual Cancer Clinic, delivering end-to-end care. By building our Color Assistant with the Agent Development Kit (ADK) and scaling it via Agent Runtime, we are helping more women get screened for breast cancer. The Color Assistant engages users to check screening eligibility, connects them to clinicians, and helps schedule appointments. The power of the agent lies in the scale it enables — helping us reach more people and respond to individual risk and eligibility in real time.” – Jayodita Sanghvi, PhD., Head of AI Platform, Color

“By rebuilding Comcast’s Xfinity Assistant with Agent Development Kit (ADK), we’ve moved beyond simple scripted automation to conversational generative intelligence that delivers personalized troubleshooting and self-service support to our customers. Agent Runtime has been a massive accelerator, allowing us to deploy a sophisticated multi-agent architecture that increases digital containment while ensuring secure, grounded interactions via Gemini. We aren't just reducing repeat interactions by solving customers’ issues the first time; we're redefining the customer experience at scale.” – Rick Rioboli, Chief Technical Officer, Connectivity & Platforms, Comcast

“Geotab uses Agent Platform to rapidly accelerate our AI Agent Center of Excellence. Google's Agent Development Kit (ADK) provides the flexibility to orchestrate various frameworks under a single, governable path to production, while offering an exceptional developer experience that dramatically speeds up our build-test-deploy cycle. For Geotab, ADK is the foundation that allows us to rapidly and safely scale our agentic AI solutions across the enterprise” – Mike Branch, Vice President, Data & Analytics, GeoTab

"Gurunavi uses Agent Platform to power 'UMAME!', an AI restaurant discovery app that leverages Memory Bank to achieve a deep understanding of user context. Unlike conventional prompt-based systems, our agent remembers a user's past actions and preferences to proactively present the best options. This eliminates the need for manual searches and creates a seamless experience that will improve user satisfaction by 30% or more. We view this memory function as a non-negotiable feature for the future of new culinary experiences.” – Toshiaki Iwamoto, CTO, Gurunavi

"At L'Oréal, Beauty Tech is not just a support function — it is a powerful catalyst to create the beauty that moves the world. To live up to that ambition, we decided to build our own proprietary Beauty Tech Agentic Platform, powered by Google Cloud. Leveraging Agent Development Kit (ADK), we are leading a fundamental shift: moving from deterministic workflow automation to autonomous, outcome-oriented agent orchestration. Our agents are not locked in a vacuum — through Model Context Protocol (MCP), they are securely connected to our single sources of truth, including our Beauty Tech Data Platform and core operational applications. Google Cloud gives us the resilience, the multi-LLM flexibility, and the enterprise-grade trust framework we need to scale this platform globally, while keeping human oversight at the center." – Etienne BERTIN, Group CIO, L'Oréal

“Payhawk uses Agent Platform to transform our AI agents from simple task executors into genuine financial assistants. By leveraging Memory Bank, we have moved from stateless interactions to long-term context retention. Our agents now act like dedicated team members, autonomously recalling user-specific constraints and history. For example, our Financial Controller Agent now remembers a user’s habits to auto-submit expenses, reducing submission time by over 50%. This shift allows our agents to anticipate needs based on past behavior rather than just reacting to prompts.” – Diyan Bogdanov, Principal Applied AI Engineer, Payhawk

"PayPal uses Agent Platform to rapidly build and deploy agents in production. Specifically, we use Agent Development Kit (ADK) and visual tools to inspect agent interactions, and manage multi-agent workflows. This provides the step-by-step visibility we need to visualize the flow of intent and payment mandates. Finally, Agent Payment Protocol (AP2) on Agent Platform provides the critical foundation for trusted agent payments. helping our ecosystem accelerate the shipping of secure agent-based commerce experiences." – Nitin Sharma, Principal Engineer, AI, PayPal

Build AI agents

Build agents quickly and easily by empowering your developers, business users and everyone in between to build and deploy agents at scale.

Build smarter agents, faster

A major upgrade to ADK: More than six trillion tokens are processed monthly on Gemini models through ADK. Unlock more powerful reasoning by organizing agents into a network of sub-agents. This new, graph-based framework allows you to define clear, reliable logic for how agents work together to solve complex problems.
Workspaces are secure-by-design: Give agents a hardened, sandboxed environment to run bash commands and manage files safely, isolated from your core systems.
Multimodal streaming: Bring human-like stability to real-time interactions with multimodal support for live audio and video cues.

Connect your agents to the enterprise

Securely access any system: Use plug-and-play architecture with Native Ecosystem Integrations to connect agents to your internal data and tools without custom coding.
Automate background operations: Activate your data in BigQuery and Pub/Sub with Batch & Event-driven agents. This way, you can run massive, asynchronous tasks like content evaluation or data analysis in the background.

Go from idea to production in hours

Enable AI-driven development: A programmatic interface for coding agents to access Google’s complete suite of agentic capabilities, allowing them to build, evaluate, and deploy production-ready agents on your behalf.
Bringing agent building directly to Agent Studio: Now, you can move seamlessly from building simple prompts to deploying complex agents in Agent Studio. Once you're ready for deep customization, export your logic directly into ADK to continue development in a full-code environment.
Get a head start with pre-built agents: Access a curated set of agent templates in Agent Garden — including code modernization, financial analysis, economic research, invoice processing, and more — that serve as immediate building blocks for your multi-agent systems.

Scale AI agents

To move from a proof-of-concept to a live environment, you need a platform that can handle the performance, state, and security requirements of real-world work.

Powering high-performance agent execution

The latest Agent Runtime: Our revamped Agent Runtime delivers sub-second cold starts and allows you to provision new agents in seconds.
Support for multi-day workflows: You can now deploy long-running agents that run autonomously for days at a time. This allows your agents to manage complex, multi-step workflows and deep reasoning tasks that require extended persistence, like managing a sales prospecting sequence.
Autonomous action with security-by-design environments: Agent Sandbox provides a hardened environment to safely execute model-generated code and perform computer use tasks like browser-based automation without risk to your host systems.
Agent-to-agent orchestration: Enables agents to seamlessly delegate tasks to one another, including support for complex, generative, and deterministic orchestration patterns. This ensures that for critical flows such as compliance, your agents follow well-specified paths every time.

Move beyond temporary session data to high-accuracy context

Personalize interactions: Agent Memory Bank dynamically generates and curates long-term memories from conversations. Using new Memory Profiles, agents can recall high-accuracy details with low latency, ensuring context is never lost.
Link AI interactions to your existing records: Store and manage history using Agent Sessions. With Custom Session IDs, you can use your own unique identifiers to track sessions and map them directly to your internal database and CRM records.
Enable real-time, human-like interactions: Using the WebSocket protocol for Bidirectional Streaming, you can help ensure your agents are highly responsive during live customer or employee interactions, processing audio and video without lag.

Govern AI agents

Govern with a secure-by-design architecture that applies enterprise rigor to every agent in your fleet – from the ones you build on Agent Platform to the ones you source from our partner ecosystem.

Manage all of your agents through a single source of truth for identity and access.

Assign every agent a verifiable identity: Agent Identity improves the security posture of your agents by ensuring every agent receives a unique cryptographic ID. This creates a clear, auditable trail for every action an agent takes, mapped back to defined authorization policies.
Maintain a central library of approved tools: Our new Agent Registry provides a single source of truth for your enterprise. It indexes every internal agent, tool, and skill, simplifying discovery and ensuring only governed, approved assets are available to your users.
Manage your agent fleet from one control point: Agent Gateway acts as the air traffic control for your agent ecosystem. It provides secure, unified connectivity between agents and tools across any environment, while enforcing consistent security policies and Model Armor protections to safeguard against prompt injection and data leakage.

Use AI-powered insights to detect hidden risks and suspicious behavior before they impact your business.

Detect suspicious behavior in real-time: Agent Anomaly Detection uses statistical models and an LLM-as-a-judge framework to flag unusual reasoning. This works alongside Agent Threat Detection to provide visibility into malicious activity, such as reverse shells or connections to known bad IP addresses.
Uncover vulnerabilities automatically: A new Agent Security dashboard, powered by Security Command Center, unifies threat detection and risk analysis. It allows your teams to map relationships between agents and models, automate asset discovery, and scan for vulnerabilities in the underlying operating system and language packages.

Optimize AI agents

Agent Platform gives you the visibility needed to understand how your AI is performing, making it easy to refine their logic and get smarter over time.

Test your agents before they ship

Simulate realistic conversations: Use Agent Simulation to test agents against human-like synthetic user interactions and virtualized tools in a controlled environment. Agents are automatically scored based on task success and safety across multi-step conversations.

Monitor and improve in production

Track live performance: Use Agent Evaluation to continuously score agents against live traffic using multi-turn autoraters that can evaluate the logic of an entire conversation, not just a single response. With turnkey dashboards and Agent Observability, you can visually trace complex reasoning to debug issues as they happen.
Automate agent refinement: Instead of manually digging through logs, Agent Optimizer automatically clusters real-world failures and suggests refined system instructions to improve accuracy.

Detailed technical guides and a full list of updates are available in our updated documentation and release notes. Agent Platform is the new standard for enterprise agent development, built to help you move from experimentation to production-scale impact, starting today.

The new Gemini Enterprise: one platform for agent development, orchestration, and governance

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

The first wave of AI changed how we find information; the next wave is changing how we get work done. Today, we’re enhancing our most powerful AI tools and bringing them together under one roof.

Gemini Enterprise is now an end-to-end system for the agentic era, built for agents that can execute complex, multi-step work processes. It combines access to frontier AI models, an intuitive UI, a secure development framework, and the ability to successfully deploy agents at scale.

At Google Cloud Next, we’re launching an expanded Gemini Enterprise portfolio that includes several key components:

Gemini Enterprise Agent Platform is our new developer platform and evolution of Vertex AI. It brings together our full suite of models, development, and tuning services, with new features for businesses to build, scale, govern, and optimize agents that can work autonomously to execute complex business workflows. Learn more here.
Gemini Enterprise app lets teams discover, create, share, and run AI agents in a single, secure environment. The Gemini Enterprise app is built on the Agent Platform, meaning all governance, security, and identity capabilities come standard, and it integrates with your enterprise data – including data in third-party systems–through connectors.
An open partner ecosystem to discover and deploy a wide range of third-party agents from leaders like Oracle, Salesforce, and ServiceNow – all within a secure and governed environment.

As organizations start building and managing thousands of agents, they need new ways to provide oversight and minimize sprawl. Gemini Enterprise enables IT departments to manage agent permissions and activities securely. It provides the same level of oversight and auditability found in essential business applications like payroll or quarterly financial reporting.

Gemini Enterprise’s unified approach provides multiple advantages across an organization:

For developers: Agent Platform is the foundation for the agentic enterprise built on our leading infrastructure and deeply integrated with our data and security capabilities. It’s where technical teams can build innovative agents that transform your products and services, alongside agents that scale your internal operations. With tools like low-code Agent Studio, support for Agent Development Kit (ADK), and protocols like A2A and MCP, Agent Platform is the key to building, scaling and optimizing agents across the entire organization.
For IT teams: We provide a single control plane for governance in Agent Platform, so every employee can use and share agents with full IT visibility. Both no-code and pro-code agents are managed through a consistent model for identity, security, and auditing – preventing AI sprawl without slowing down innovation.
For knowledge workers: Gemini Enterprise app is the front door for AI for every employee in your organization. Knowledge workers can build their own agents with our no-code Agent Designer or leverage a wide range of partner-built agents from Google Cloud Marketplace. And these agents can be easily shared across the organization with full governance and compliance controls built-in. Inbox in Gemini Enterprise provides a central location to manage and monitor agents, including long-running agents.

Gemini Enterprise Agent Platform: securely orchestrating the agentic enterprise

At the heart of the Gemini Enterprise experience is our new Agent Platform. It’s the central place where developers can build, scale, govern, and optimize fleets of agents. Today, we're introducing new capabilities across four key areas:

Build: An enhanced Agent Development Kit (ADK) helps you unlock more powerful reasoning by organizing agents in a network of sub-agents. With ADK’s new graph-based framework, you can orchestrate complex processes with new ecosystem integrations and support for MCP.
Scale: With our improved Agent Runtime, your agents can be both fast and persistent, and with agent-to-agent orchestration, they can seamlessly delegate tasks to one another. The launch of Memory Bank and Memory Profiles also ensure agents operate with long-term contextual memory, allowing them to recall user preferences and past history across different sessions.
Govern: Our secure-by-design architecture gives IT teams centralized control over their entire agent lifecycle. Agent Identity assigns every agent a unique cryptographic ID for complete traceability and auditing, while Agent Gateway provides air traffic control to ensure interactions between agents and data is secure. Model Armor then protects against activities like prompt injection, tool poisoning, and sensitive data leakage.
Optimize: To move beyond trial-and-error, the new Agent Simulation tool allows you to stress-test your agents against real-world scenarios before they ship. Agent Observability also helps ensure agents function exactly as intended, providing a real-time view into safety and performance.

Any agent built on Agent Platform can then be instantly surfaced in the Gemini Enterprise app for your employees to access. By building on a foundation where security and identity are included by default, you can stop worrying about managing risk and start focusing on managing agent-led business outcomes.

Learn more: What's new in Agent Platform

The Gemini Enterprise app: your hub for everyday work

In the Gemini Enterprise app, teams can discover, create, share, collaborate, and run agents in a single environment. With data governance and compliance built in, it gives your teams a safe place to leverage agents grounded with your business data. It also provides a simple interface for users to engage, and democratizes the power of agent creation, orchestration, and scaling to every individual in the organization.

New capabilities available in the Gemini Enterprise app include the ability to:

Innovate with no-code agents: Using the newly-integrated Agent Designer, anyone can create agents to automate complex, multi-system processes with simple, natural language. This includes schedule or trigger-based agents, including those triggered by another business application, even via a partner connector.
Centralize agent oversight: Inbox in Gemini Enterprise provides a unified hub for managing agents at scale, including long-running agents executing complex workflows. Users can monitor active agents, adjust configurations, and receive real-time status alerts directly through email and chat, ensuring automated processes stay on track.
Establish persistent team memory: Moving beyond individual chats, Projects creates a shared workspace for teams and agents. It centralizes the context and history of a given topic, ensuring that critical project intelligence remains a permanent company asset even as team members or roles change.
Collaborate in a dynamic workspace: Canvas provides teams with an interactive editor to co-create and edit in Google Docs and Slides. We’ve also added Microsoft 365 interoperability, allowing you to export the Docs and Slides you create within Canvas into common Microsoft Office formats.

Learn more: What's new in Gemini Enterprise

The agent economy: amplified by our partner ecosystem

No organization can build every agent they need from scratch. The agentic enterprise thrives on an open, interoperable ecosystem. That's why we are helping to make it easier than ever to discover and deploy specialized agents from our partners.

Benefits include:

Activating third-party agents from within the Gemini Enterprise app: We are now bringing the full catalog of partner-built agents from Google Cloud Marketplace directly into the Agent Gallery. You no longer need to leave your workflow to find and deploy powerful third-party agents. The app provides a secure request-and-approval gateway that connects employee needs to IT-validated solutions.
A growing library of trusted solutions: Agent Gallery features an expansive collection of validated agents from SaaS leaders and innovative startups. For example, you can use specialized agents from partners like Adobe, Salesforce, ServiceNow, and Workday–directly within the Gemini Enterprise app.
Deploy with confidence: Every partner agent in the gallery has been validated by Google Cloud to meet our requirements for security and interoperability. This allows you to benefit from a world of innovation, while maintaining the centralized governance and control your business demands.

Learn more: Partner-built agents available in Gemini Enterprise

Gemini Enterprise brings the best of Google AI to every employee, every developer, and every customer — empowering them with cutting-edge AI agents. Get started with Gemini Enterprise today.

What’s new with compute: Scaling core and agentic workloads

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

At Google Cloud Next, we’re announcing a range of compute capabilities to enable your core general purpose and AI workloads for the agentic world with higher performance and lower costs.

Why it matters: IT leaders and builders are faced with balancing compute investments and resources between agentic AI and the general purpose use cases, including the web servers, databases, and enterprise applications that drive everyday customer experiences.

On one side, agents can place unpredictable demand on compute infrastructure, often scaling exponentially. A single user interaction can instantaneously kick off hundreds of concurrent, high-throughput, and low-latency tasks. On the other side, general-purpose workloads generate and hold the data required to fuel the agentic world. Relying on static and siloed infrastructure to run them can risk performance bottlenecks and spiraling costs, leaving your organization unable to respond to surges in demand.

Consider a global travel application where a simple vacation search instantly triggers a massive orchestration of agentic inventory checks, dynamic pricing models, and AI-driven personalized itineraries. Without a modern architecture, this sudden surge in demand can overwhelm the core booking database and bring business to a halt.

We address this with fluid compute, Google Cloud infrastructure that adapts to your general-purpose and agentic workflows, enabling both to win by flexing in performance, capacity, and scale, all in real time. This dynamic flexibility relies directly on the automated orchestration of Google Kubernetes Engine (GKE) and our new Agent Sandboxes to instantly provision secure, isolated execution environments at machine speed.

Let’s take a deeper look at the new compute capabilities announced at Next ‘26.

Run AI and general purpose workloads together

Agentic planning and reinforcement learning depend on highly fluid compute to process unpredictable bursts of autonomous tasks. Relying on static infrastructure to isolate agent-generated code can create severe provisioning delays and heavily inflate your cloud budget. You can remove these bottlenecks by adopting an adaptive cloud foundation. Leveraging GKE Agent Sandboxes empowers your teams to securely launch thousands of execution environments. Pairing these scalable sandboxes with efficient Google Axion processors helps your organization optimize total cost of ownership while fueling artificial intelligence innovation.

Here’s what’s new in Google Cloud compute launches and announcements:

Google Axion N4A is GA: Harness the agility of Google’s custom Arm-based Axion CPUs and achieve up to 2x better price-performance than comparable current-generation x86-based VMs for cost-sensitive workloads such as Java applications, scale-out web servers, and SaaS built by startups, enterprises and partners. Learn more here.
GKE Agent Sandbox, with Axion N4A for price performance, is GA. As the industry’s only native sandbox service among hyperscalers, GKE Agent Sandbox offers scalable and low-latency infrastructure designed for agents to safely execute untrusted code and tool calls without sacrificing performance. With Google Axion, you can build agents on leading infrastructure without compromising on cost or choice. GKE Agent Sandbox running on Google Axion N4A instances provides up to 30% better price-performance than the next leading hyperscale cloud provider. Try GKE Agent Sandbox here.
Google Axion C4A.metal, our first Axion bare metal instance, is in preview: C4A.metal instances power Android development, automotive simulation, CI/CD pipelines, security workloads, and custom hypervisors, without the performance overhead and complexity of nested virtualization. C4A.metal will be GA this summer; learn more here.
C4 instances offer expanded support for Intel Xeon 6 (Granite Rapids) across all shapes: Achieve high-performance for AI workloads like LLM inference and vector search by using Intel AMX with native FP16 support to increase throughput and reduce latency, offering 13% better price-performance versus comparable Intel Xeon 6-based VMs from another leading hyperscaler. C4 VMs are available with Intel Xeon 6 processors across all shapes. Learn more here.
Flexible CUDs expanded support is GA: Shift spend across regions and VM families while optimizing for TCO, with flexible committed use discounts, now with support for a wider range of VM families and services, including memory-optimized (M1-M4) and HPC-optimized (H3, H4D) VM families, as well as Cloud Run. Learn more here.

Here’s what customers are saying:

Unity: Unity is redefining the economics of real-time AI with Unity Vector. By migrating its on-demand feature processor workloads to Google Axion N4A instances, Unity achieved a 20% improvement in cost efficiency without sacrificing latency. As Unity Vector scales to meet increasing demand, the move to N4A instances ensures that Unity continues to deliver industry-leading performance at a sustainable cost.

Deutsche Börse: A leading German market infrastructure provider, Deutsche Börse migrated and modernized dozens of core financial applications onto Google Compute Engine VMs, including latest generation C4 and C4D instances, supporting latency-sensitive Oracle databases and post-trade processing at scale, and boosting release speed, operational agility, and resilience. This delivered the consistent performance they needed to process millions of financial transactions every day and they achieved 58% faster time to market and 33% lower TCO.

WP Engine: WP Engine powers millions of digital experiences where every millisecond matters. By running GKE clusters on C4D and N4D instances, WP Engine has seen up to a 60% reduction in latency for mobile-optimized REST APIs and up to 51% faster processing for data-rich application requests.

eDreams ODIGEO: Operating a high-volume, AI-powered travel platform where every millisecond dictates the customer experience, eDreams ODIGEO migrated its foundational Java-based ecommerce modules on GKE to Axion virtual machines. This immediately eliminated weeks of manual code optimization, delivered a massive 75% improvement in P95 latency with zero code changes, and unlocked price-performance to scale their global services far more cost-effectively than their legacy x86 infrastructure could.

Chainguard: Prioritizing absolute isolation for their foundational software build system, Chainguard deployed the new Axion C4A bare metal instances. This allowed them to establish a strong hypervisor security boundary for package builds, secure their development pipeline with architectural parity, and ensure robust protection, all without compromising build performance.

Run I/O and latency-sensitive workloads together

Both AI and core workloads depend on the ability to store, read, and move data as a single, high-performance operation. Traditionally, these stages are slowed by network and storage limits tethered to vCPU counts, which can starve AI models of the data they need to function. You can remove these constraints by leveraging accelerated Hyperdisk performance for rapid data access and high-performance networking for consistent transit. By allowing your data pipelines to scale independently of compute, your AI training and I/O-sensitive workloads have the dedicated bandwidth they need to remain stable under peak demand.

C4N is in preview: Running high-volume network applications such as concurrent mobile app requests and real-time inventory updates can risk bottlenecks during peak traffic. Maximize your throughput with C4N, featuring Titanium adapters that offload complex packet processing to deliver a market-leading 95 million packets per second — a 40% performance advantage for high-traffic network applications compared to other leading hyperscalers. Designed to rapidly transfer large datasets, C4N provides nearly 400 Gbps of VM-to-VM bandwidth, a 4x improvement in bandwidth-per-vCPU, and achieves an 8x increase in egress network bandwidth through internet gateways compared to C4 VMs. C4N with Hyperdisk Extreme also provides the low-latency, high-speed data access that modern databases and enterprise AI applications need, with 25 GiB/s of block storage throughput and nearly 1M IOPS. Sign-up here for C4N preview access.
M4N is in preview: Running memory-intensive databases can push organizations to overprovision compute cores (vCPU) to meet memory speeds, driving up software licensing fees. We introduced the new M4N series to solve this exact problem. Running Oracle workloads on M4N with Hyperdisk Extreme can reduce TCO by over 20%, enabling you to run Oracle more efficiently, with 26.57 GiB of RAM per vCPU for scale and on far fewer cores. Paired together, M4N with Hyperdisk Extreme delivers the highest per-core IOPS and throughput for high-memory instances across leading hyperscalers. Sign-up for the preview here.
Announcing Z4D: Optimize I/O-intensive workloads and remove network-based storage bottlenecks with new Z4D instances. By securing up to 84 TiB of high-performance local SSD directly on the node, organizations can process massive datasets for SQL, NoSQL, and vector databases. Z4D provides up to 400 Gbps of VM-to-VM bandwidth, matching both C4N and M4N. Z4D virtual machines and bare metal instances will be in preview soon.

Here is what customers are saying:

Ericsson: 5G Core workloads are inherently network-heavy, demanding high-throughput packet processing and deterministic latency that standard public cloud instances often struggle to maintain at scale. By leveraging the Google Cloud C4N, they’ve found the ideal choice for network performance to power Ericsson On-Demand. C4N’s architectural focus on network-optimized compute allows its 5G Core-as-a-Service to reach unprecedented throughput levels, like its recent 1 Tbps milestone, while maintaining the carrier-grade reliability its customers expect.

Teradata: Teradata’s Autonomous Knowledge Cloud enables the world’s largest enterprises to activate enterprise intelligence and turn trusted data into measurable business outcomes. Customers rely on Teradata to run mission‑critical, highly I/O‑intensive analytics at scale where performance and efficiency directly determine value. C4N instances are well suited for these demanding workloads, delivering strong price‑performance and supporting more efficient, optimized deployments. With C4N, Teradata can help customers accelerate insights, scale with confidence, and drive greater impact from their data and AI investments.

Handle demanding storage requirements

Foundational workloads such as web servers, applications, and databases hold the data required to fuel the agentic world. Siloing this critical information on rigid hardware creates bottlenecks that can completely stall enterprise modernization. Imagine a global retail brand running a holiday promotion, but the inventory database times out and drops customer requests because the legacy hardware couldn’t process the sudden flood of agentic queries.

Organizations require the highest performing database hosts backed by high performance IOPS and throughput per vCPU to ensure non-blocking data delivery. Moving these applications to modern cloud infrastructure dramatically improves total cost of ownership and operational throughput. Through strategic cloud migrations, customers can eliminate the architectural walls that stall modernization and unlock their data for AI. Here is what is new in fluid compute for throughput- and capacity-sensitive workloads:

Hyperdisk Balanced improvements. Hyperdisk Balanced enables fast and efficient block storage for general purpose workloads, including applications and relational databases. With Hyperdisk Balanced you can drive up to 2.4 GiB/s and 160K IOPS per volume, higher than general-purpose block storage offerings from other hyperscalers, all while achieving lower mean latency than alternatives. With Hyperdisk Balanced High Availability you can now achieve a 4x performance improvement for high availability databases like SQL Server or PostgreSQL by dynamically routing full disk performance to the active VM, removing the need to overprovision storage. Leverage zero-downtime encryption key rotation and consistency groups for instant snapshots, making it easier to stay more secure. With these capabilities, you can deliver lower TCO, higher performance, and workload resilience for your general-purpose workloads. Learn more here.
Hyperdisk ML performance improvements and Hyperdisk Exapools are GA: With 2 TiB/s of aggregate throughput (up from 1.2 TiB/s), Hyperdisk ML helps eliminate AI storage bottlenecks, offering more than 200x higher throughput per disk than competitive offerings, so your valuable accelerator clusters never sit idle. This allows you to maximize AI compute ROI while powering the next generation of intelligent agents. In addition, for large-scale training needs, Hyperdisk Exapools offer the highest aggregate block storage performance and capacity, per AI cluster, of any hyperscaler. Learn more about Hyperdisk ML and Exapools here and here.
Announcing Z4M: Access up to 168 TiB of local SSD coupled with up to 400 Gbps of network bandwidth, support for RDMA, and bare-metal shapes to run distributed parallel file systems and large-scale AI/ML workloads. Z4M will be integrated with Cluster Director with the option to be colocated with accelerators to provide fast and low-latency access to data. Z4M VMs and bare metal instances are expected to be in preview in Q3 2026.

Here is what customers are saying:

Shopify: During Black Friday weekend sales, Shopify processed over $14.6B and tracked 136 million packages for 81M buyers using its Shop App built on Compute Engine’s Z-series backed storage — without compromising speed or reliability.

HubX: Operating a massive portfolio of AI-powered mobile applications where rapid model loading dictates the user experience, HubX deployed Hyperdisk ML on GKE to eliminate severe I/O bottlenecks. Leveraging this specialized storage layer allowed HubX to support hundreds of concurrent readers and accelerate pod initialization times by 30x during peak traffic surges, drastically reducing idle accelerator costs and helping ensure their complex inference workloads scaled as expected.

Fluid infrastructure for the agentic era

Now, your foundational workloads and agents no longer need to compete for capacity or performance. With Google Cloud’s fluid compute, you get adaptive cloud infrastructure that prevents bottlenecks and enables both your foundational and AI workloads to collaborate and thrive.

Ready to get started? Head straight to the Google Cloud console to spin up a VM for your next big project. Or start planning your migration by checking out Migration Center's AI-powered toolsets to perform cost estimates, create a business case, and evaluate your modernization options.

Inside the eighth-generation TPU: An architecture deep dive

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

At Google, our TPU design philosophy has always been centered on three pillars: scalability, reliability, and efficiency. As AI models evolve from dense large language models (LLMs) to massive Mixture-of-Experts (MoEs) and reasoning-heavy architectures, the hardware must do more than just add floating point operations per second (FLOPS); it must evolve to meet the specific operational intensities of the latest workloads.

The rise of agentic AI requires infrastructure that can handle long context windows and complex sequential logic. At the same time, world models have emerged as a necessary evolution from current next-sequence-of-data architectures, which means newer agents are simulating future scenarios, anticipating consequences, and learning through "imagination" rather than risky trial-and-error. The eighth-generation TPUs (TPU 8t and TPU 8i) are our answer to these challenges, ensuring that every workload, from the first token of training to the final step of a multi-turn reasoning chain, is running on the most efficient path possible. They are built to efficiently train and serve world models like Google DeepMind’s Genie 3, enabling millions of agents to practice and refine their reasoning in diverse simulated environments.

TPU 8: Specialized by design

Recognizing that the infrastructure requirements for pre-training, post-training, and real-time serving have diverged, our eighth-generation TPUs introduce two distinct systems: TPU 8t and TPU 8i. These new systems are key components of Google Cloud's AI Hypercomputer, an integrated supercomputing architecture that combines hardware, software, and networking to power the full AI lifecycle. While both systems share the core DNA of Google’s AI stack and support the full AI lifecycle, each is built to address distinct bottlenecks and optimize efficiency for critical stages of development. Additionally, by integrating Arm-based Axion CPU headers across our eighth-generation TPU system, we’ve removed the host bottleneck caused by data preparation latency. Axion provides the compute headroom to handle complex data preprocessing and orchestration, so that TPUs stay fed and don’t stall.

TPU 8t: The pre-training powerhouse

Optimized for massive-scale pre-training and embedding-heavy workloads, TPU 8t utilizes our proven 3D torus network topology at an even larger scale of 9,600 chips in a single superpod. TPU 8t is designed for maximum throughput across hundreds of superpods, ensuring that training runs stay on schedule.

Here are some key advancements of TPU 8t over prior-generation TPUs:

The SparseCore advantage: Central to TPU 8t is the SparseCore, a specialized accelerator designed to handle the irregular memory access patterns of embedding lookups. While the Matrix Multiply Unit (MXU) handles matrix math, the SparseCore offloads data-dependent all-gather operations, amongst other collectives, preventing the zero-op bottlenecks that often plague general-purpose chips.
VPU/MXU overlap and balanced scaling: TPU 8t is designed to maximize provisioned FLOPs utilization. By implementing more balanced Vector Processing Unit (VPU) scaling, the architecture minimizes exposed vector operation time. This allows for better overlapping of quantization, softmax, and layernorms with the matrix multiplications in the MXU, helping the chip stay busy rather than waiting on sequential vector tasks.
Native FP4: TPU 8t introduces native 4-bit floating point (FP4) to overcome memory bandwidth bottlenecks, doubling MXU throughput while maintaining accuracy for large models even at lower-precision quantization. By reducing the bits per parameter, the platform minimizes energy-intensive data movement and allows larger model layers to fit within local hardware buffers for peak compute utilization.

Figure 1: TPU 8t ASIC block diagram

Virgo Network topology and up to 4x data center network increase: To support the massive data requirements of TPU 8t, we introduced Virgo Network. This new networking architecture enables up to 4x increased data center network (DCN) bandwidth on TPU 8t training over DCN. Virgo Network is a scale-out fabric designed for the extreme requirements of modern AI workloads. Built on high-radix switches that reduce network layers by allowing more ports per switch, it employs a flat, two-layer non-blocking topology. Compared with traditional datacenter networks, this significantly reduces latency by minimizing network tiers. It features a multi-planar design with independent control domains to connect TPU 8t chips. The TPU 8t racks also connect with the Jupiter north-south fabric to access compute and storage services. Together, this streamlined architecture delivers the massive bisection bandwidth and deterministic low latency necessary for enabling the world's largest training clusters with high availability.

With 2x scale-up bandwidth on the inter-chip interconnect (ICI) and up to 4x raw scale-out DCN bandwidth compared to the previous generation, TPU 8t drastically reduces data bottlenecks. Then, to further accelerate the development of frontier models, we scale distributed training beyond a single cluster. With JAX and Pathways, we can now scale to more than 1 million TPU chips in a single training cluster. Virgo Network can link over 134,000 TPU 8t chips with up to 47 petabits/sec of non-blocking bi-sectional bandwidth in a single fabric. This fabric delivers over 1.7K ExaFlops with near-linear scaling performance.

Figure 2: TPU 8t rack level connectivity to Virgo fabric

Faster storage access: We are introducing TPUDirect RDMA and TPU Direct Storage in TPU 8t. TPU Direct RDMA enables direct data transfers between the TPU's memory (HBM) and the Network Interface Cards (NICs), bypassing the host CPU and DRAM. This reduces latency and host system bottlenecks, increasing the effective bandwidth for TPU-to-TPU communication. Similarly, TPUDirect Storage bypasses CPU host bottlenecks by enabling direct memory access between the TPU and high-speed managed storage like 10T Lustre, effectively doubling the bandwidth for massive data transfers. This architecture allows the silicon to ingest training data at line rate, ensuring that the MXUs stay fully saturated even when processing large multimodal datasets.

By combining Managed Lustre 10T and TPUDirect Storage to route hundred-petabyte datasets directly to the silicon, TPU 8t prevents training delays caused by data ingestion bottlenecks. This delivers 10x faster storage access compared to training on seventh-generation Ironwood TPUs.

Figure 3: The top diagram shows the data transfer path without TPUDirect Storage. The bottom diagram shows TPU 8t data transfer with TPUDirect Storage between 2 TPU 8t chips and TPUDirect Storage with Managed 10T Lustre storage.

TPU 8i: The sampling and serving specialist

Optimized for post-training and high-concurrency reasoning, we designed TPU 8i with our highest on-chip SRAM, a new Collectives Acceleration Engine (CAE), and a new serving-optimized network topology called Boardfly.

Large on-chip SRAM: With 3x more on-chip SRAM over the previous generation, TPU 8i can host a larger KV Cache entirely on silicon, significantly reducing the idle time of the cores during long-context decoding.

Figure 4: TPU 8i ASIC block diagram

The Collectives Acceleration Engine (CAE): To solve the sampling bottleneck, TPU 8i uses the CAE, which aggregates results across cores with near-zero latency, specifically accelerating the reduction and synchronization steps required during auto-regressive decoding and "chain-of-thought" processing. For each TPU 8i chip, there are two Tensor Cores (TC) on-core dies and one CAE on the chiplet die, replacing four SparseCores (SCs) on core dies in previous-generation Ironwood TPU. By integrating a specialized CAE, TPU 8i further reduces the on-chip latency of collectives by 5x. Lower latency per collective operation means less time spent waiting, directly contributing to higher throughput required to run millions of agents concurrently.
Boardfly ICI topology: While the 3D torus allows connecting thousands of chips to be used in cohesion, a large mesh does have more hops between chips and higher all-to-all latencies. For 8i, we changed how the chips connect together in fully connected boards that are then aggregated into groups. Utilizing a high-radix design, we connect up-to 1,152 of these chips together, reducing the network diameter and the number of hops a data packet must take to cross the system. By slashing the hops required for all-to-all communication (the heart of MoE and reasoning models), Boardfly achieves up to a 50% improvement in latency for communication-intensive workloads.

Figure 5: TPU 8i hierarchical Boardfly topology building up from a building block of four fully connected chips into a fully connected group of eight boards, with 36 of such groups fully connected into a TPU 8i pod

Boardfly consists of the following elements, and its topology is hierarchical by nature:

Building Block (BB): Each tray forms a four-chip ring using internal ICI links, providing 16 external connections for broader networking.
Group (G): Eight boards are fully connected via copper cabling to create a localized group, utilizing 11 of the available external links for intra-group communication.
Pod structure: The final architecture scales to 36 groups (up to 1,024 active chips) linked through Optical Circuit Switches (OCS), ensuring a maximum latency of seven hops for any chip-to-chip communication.

Deep dive: The Boardfly vs. torus math

Why move away from the torus for TPU 8i? It comes down to network diameter.

In a 3D torus, nodes are arranged in a grid where each dimension wraps around like a ring. To reach the furthest possible chip in a 8 x 8 x 16 (1024-chip) configuration, a packet must traverse half the distance of each ring:

3D torus = 8/2(X) + 8/2(Y) + 16/2(Z) = 16 hops

While the torus is highly efficient for the neighbor-to-neighbor communication typical of dense training, it creates a latency tax for all-to-all communication patterns. In the era of reasoning models and MoE, where any chip may need to talk to any other chip to route a token, this hop count matters.

Boardfly’s high-radix topology is inspired by Dragonfly topology principles. By increasing the number of direct optical long-haul links between groups of boards, we flatten the network. For that same 1024-chip pod, Boardfly reduces the network diameter from 16 hops down to just seven.

This 56% reduction in network diameter translates directly to lower tail latency, so that the TPU 8i CAE isn't left waiting for data to arrive from across the pod.

Figure 6: A visual representation of the maximum seven-hop ICI network diameter via optical circuit switch on TPU 8i pod

TPU 8t and TPU 8i at a glance

Feature	TPU 8t	TPU 8i
Primary Workload	Large-scale pre-training	Sampling, serving, and reasoning
Network Topology	3D torus	Boardfly
Specialized Chip Features	SparseCore (Embeddings) & LLM Decoder Engine	CAE (Collectives Acceleration Engine)
HBM Capacity	216 GB	288 GB
On-Chip SRAM (Vmem)	128 MB	384 MB
Peak FP4 PFLOPs	12.6	10.1
HBM Bandwidth	6,528 GB/s	8,601 GB/s (~1.3x of TPU 8t)
CPU Header	Arm Axion	Arm Axion

Software enablement: A performance-first AI stack

Hardware is only as powerful as the software that drives it. The eighth generation of TPUs are built on the same performance-first stack we pioneered with the seventh-generation Ironwood TPUs, designed to make custom kernel development accessible without sacrificing the abstraction of high-level frameworks. This stack includes:

Pallas and Mosaic: We provide first-class support for Pallas, our custom kernel language that lets you write hardware-aware kernels in Python. This enables you to squeeze every drop of performance out of the TPU 8i CAE and the TPU 8t SparseCore.
Native PyTorch experience: We're thrilled to share that native PyTorch support for TPUs is now in preview. If you're currently building and serving models on PyTorch, we've made it easier than ever to start using TPUs. You can bring your existing models to our TPUs just as they are, complete with full support for the native features you rely on, such as Eager Mode.
Portability: The same JAX, PyTorch, or Keras code that runs on Ironwood scales to this generation. Accelerated Linear Algebra (XLA) handles the complex translation of the Broadly topology and CAE synchronization behind the scenes, allowing you to focus on your model, not the interconnect.

Generation over generation: The performance leap

Our commitment to co-designing hardware and software continues to pay dividends. When compared to seventh-generation Ironwood TPU, the eighthgeneration TPUs deliver massive gains:

Training price-performance: TPU 8t delivers up to 2.7x performance-per-dollar improvement over Ironwood TPU for large-scale training.
Inference price-performance: TPU 8i delivers up to 80% performance-per-dollar improvement over Ironwood TPU, particularly at low-latency targets for large MoE models.
Energy efficiency: Both chips deliver up to 2x better performance-per-watt, critical for scaling the next generation of AI sustainably.

Looking ahead

To empower Google Cloud customers pioneering the next wave of innovation, we designed TPU 8t and TPU 8i as two distinct, specialized systems tailored to the multifaceted future demands of the AI lifecycle. TPU 8t and 8i are both purpose-built for the most demanding serving and training workloads, fully integrating with the AI Hypercomputer software stack: JAX, PyTorch, vLLM, XLA, and Pathways. This specialization and ground-up redesign, all in deep collaboration with Google Deepmind, delivers exceptional price-performance and power efficiency.

The modularity of our eighth-generation architecture provides a clear and unique roadmap for the future. Just as every major shift in the computing landscape has required infrastructure breakthroughs, so does the agentic era. Reasoning agents that plan, execute, and learn within continuous feedback loops cannot operate at peak efficiency on hardware that was originally optimized for traditional training or transactional inference; their operational intensity are fundamentally distinct. Our eighth-generation TPU infrastructure has evolved to meet these specific requirements head-on.

To learn more about the eighth-generation TPU family: