We hereby declare September to be scalability month! As the world prepares for a surge of agentic fleets, we are shoring up our AI infrastructure and orchestration offerings to gracefully — and quickly — respond to that demand, all while maintaining workload isolation and security, and keeping costs in check. Read on to learn how these enhancements manifest across Google Cloud’s compute, network, storage, and orchestration offerings, plus new ways customers are using Google Cloud AI infrastructure, and third-party industry validation of our strategy.

Product, technology, and tools updates

Google Kubernetes Engine updates: The GKE team is all about improving the scalability of the platform, and in September, those improvements came in many shapes and sizes:

• New feature: Need an execution runtime with higher density for your agentic workloads? We engineered the new open-source GKE Agent Substrate to run millions of sandboxes with 10x higher density than standard container runtimes. Agent Substrate also delivers sub-500ms resume operations at over 500 suspend/resume activations per second with a native zero-trust kernel and network isolation. 

• Product update: GKE now has scale-to-zero capabilities built-in. No need to configure complex components to scale your workloads down, thanks to the HPA with the Autoscaling Metric and support for KEP-2021, which do the job for you, out of the box. Read the blog to learn more. 

• Product update: Further, the GKE HPA (with the above-mentioned Autoscaling Metric) now lets you scale up and down based on custom PromQL metrics, in addition to standard metrics, allowing you to trigger workloads according to conditions that are meaningful and unique to your business. Read more here. 

• New feature: Yet another scalability feature is GKE Pod snapshots, which lets you save the running state of your workload, including CPU and GPU memory, and restore it on demand. According to internal tests, GKE Pod snapshots can reduce AI inference start-up by as much as 89%. Learn more here.

• New migration tool: Finally, if you’ve always wanted to migrate your container workloads from AWS EKS to GKE but feared a daunting, high-friction engineering endeavor, we’ve just launched GKE agentic migration, a purpose-built agent plugin that replaces brittle, ad-hoc prompting with an AI-assisted migration pipeline protected by deterministic guardrails. Designed as a compilation of agent skills and a local Model Context Protocol (MCP) server, it uses AI to translate complex AWS EKS IaC and Kubernetes manifests directly into GKE landing zones. Get started with the onboarding guide.

• Feature updates: Reinforcement learning (RL) and evaluation workloads are a beast: In a standard agentic RL loop, an LLM policy generates actions like code snippets on GPUs and executes them inside isolated CPU sandboxes to observe a reward signal. However, when scaling up this loop to support tens of thousands of parallel rollouts, infrastructure bottlenecks emerge, for instance idle accelerators, image cardinality, and a saturated control plane. To help, we developed GKE Agent Sandbox optimized for RL, plus an Agent Sandbox RL orchestration SDK and native integrations for popular RL gyms and harnesses. All are now generally available, and you can learn more here.

Storage updates: AI trains and creates lots of data, and that data has to live somewhere — in block storage systems, file systems, object stores and databases. We announced enhancements to our storage portfolio to help this critical layer roll with the agentic punches:  

• Product update: Filestore agent volumes offer high-performance, elastic, persistent file storage for agentic workloads. Thanks to its tight integration with GKE Agent Substrate and GKE Agent Sandbox, Filestore agent volumes automatically allocates and attaches a dedicated, isolated file workspace to GKE agent sandboxes in milliseconds. Request access to the preview here. 

• New product: If you run generative AI and RAG data layers  — think Milvus, Pinecone, Qdrant, Vespa, Redis, and in-memory context caching — you may want to take a look at the M4N family of VMs, now GA, which offers the highest per-core IOPS and throughput of leading hyperscalers. Paired with Google Cloud's custom Titanium offload architecture and paired with Hyperdisk Extreme, M4N instances deliver up to 25,000 MiB/s (25 GiB/s) of aggregate host storage performance and up to 1 million IOPS. 

• New product: Another new Compute Engine product, Z4D, is GA, and a strong storage solution for AI/ML training and inference workloads. When configured as a bare metal instance, Z4D provides both the high local SSD (LSSD) capacity and low latency required by agentic microVMs, so you can run thousands of isolated sandboxes per host with native performance and efficiency. Learn about Z4D machines here.

• Product update: Today’s AI training and inference pipelines create data faster than most storage management systems can keep up, creating challenges for teams trying to understand their storage estates. A new version of Storage Intelligence advisor makes it easier to answer the question: "What’s in my buckets?" and quickly identify unexpected changes. Then, enhanced batch operations let you automate bulk changes across your buckets — say, move storage classes, mass-delete stale or temporary data, or apply metadata, tagging, retention, or encryption changes. Learn about the latest in Storage Intelligence advisor here.

• New product: Last but not least, a new version of AlloyDB for PostgreSQL brings together pioneering Google infrastructure — Colossus distributed file system, and Jupiter network — to power a new, no-compromises database architecture for the agentic era. 

Practitioner guides and how-tos

• How-to: Wish you could make GPUs and TPUs scattered around the globe behave as a single pool behind a single entry point? In this blog, we show you how to do just that. At the edge, the multi-cluster GKE Inference Gateway focuses on global, multi-region traffic distribution and high availability. Beneath that, the LLM-d router handles complex, memory-aware scheduling algorithms to keep utilization high. This architecture is deliberately runtime-, model-, and accelerator-agnostic, and in tests, routing traffic through the multi-cluster GKE Inference Gateway added less than 1% overhead.

• Guide: Using or planning to use GKE on TPUs for AI model training or inference? Training massive Large Language Models (LLMs) or running high-throughput inference serving represents a significant investment in specialized AI hardware, such as Cloud TPUs and GPUs. To get the most out of every dollar spent, you need to understand workload lifecycle metrics in GKE. This detailed guide explains how to turn opaque cluster behaviors into actionable telemetry.

Customer and partner updates

• GKE customer: Learn why gaming startup SeaVerse relies on GKE Agent Sandbox for its multi-tenant workloads, and how the platform helped it decrease its infrastructure costs by 60%. 

Research, reports and deep-dives

In case you missed it, we’re also thrilled to share that Google has been named a leader, including achieving the highest score on either product or strategy, in three key analyst reports from Gartner and Forrester. 

• Google is a leader in The Forrester Wave™: Public Cloud Platforms, Q3 2026: Highest overall score of any cloud provider!
• Google named a Leader in 2026 Gartner® Magic Quadrant™ for Strategic Cloud Platform Services: Positioned furthest for “Completeness of Vision” of all vendors evaluated.
• Google is a Leader in the 2026 Gartner Magic Quadrant for Container Management: Positioned highest in “Ability to Execute” of all vendors evaluated.
• Google Cloud achieved a Gold rating in the latest SemiAnalysis ClusterMax 3.0 report, which evaluates the reliability, performance, support, pricing, and security of GPU providers globally. See the full report for more.

August 2026

Product, technology, and tools updates

• Product update: Filestore, Google Cloud’s first-party, secure, scalable NFS file service, has emerged as a popular storage platform for AI and agentic workflows, and now, it’s even better suited to the task, with a new backend storage layer built directly on Colossus, Google’s foundational distributed storage system. This new backend lets you provision IOPS independently from storage capacity, and is deeply integrated with GKE. In AI environments, this can help you service so-called agentic swarms — large groups of agents that need to read and write to a common dataset — without a drop off in performance. For more, check out the blog post. 

• New feature: gVisor sandboxes are now available in distributed Ray clusters on GKE. In partnership with Anyscale, we introduced an experimental library for Ray that brings gVisor, Google’s open-source application kernel, directly into distributed Ray clusters. gVisor provides lightweight environments with stronger isolation than ordinary containers, plus fast startup times and low memory overhead. To try out these sandboxing capabilities on GKE, head over to the Ray sandboxing User Guide.

• Product update: Looking for high-performance, easy-to-use infrastructure on which to run a personal AI agent, but don’t want to spend a lot of money? New Cloud Run instances are dedicated, singleton compute runtimes on Cloud Run that won’t shut down when the agent is idle. Better yet, the cost to run a Cloud Run instance with 1 vCPU and 1 GiB of memory continuously for 30 days is just $5.70.  

Practitioner guides, documentation and how-tos

• How-to guide: Big news in Model Context Protocol (MCP) land: As of the 2026-07-28 specification, the protocol core is “completely stateless. The handshake is gone. The initialize / initialized handshake (SEP-2575) and the logical Mcp-Session-Id header (SEP-2567) have been removed entirely. Instead, every request is now self-describing and independent.” Whoa. Learn more about the changes that the latest MCP specification brings, and more importantly, how to implement them, in this Google Developers blog.  
• Guide: Real-time AI systems make a mess of traditional network load balancing techniques. “Instead of handling isolated requests, the backend has to manage a continuous, live bidirectional stream. You’re dealing with a constant stream of audio chunks, transcripts, model outputs, and synthesized speech flowing back and forth simultaneously.” Things only get worse when the user gets involved. “The server has to immediately halt its current speech generation, pivot to update the context, maybe trigger a new tool, and start drafting a different response; this must be done without dropping the connection.” For a new approach to managing load in the AI era, read Scaling real-time AI agents with session-aware load balancing.
• How-to: Learn how to build an elastic, scalable LLM inference platform on GKE, even with a mix of different GPU accelerators. The proposed architecture combines Capacity Advisor and Compute Advisor, plus high-performance storage like RunAI:model streamer or GCPFuse with parallel downloads. Get all the details here.
• Documentation: The thing about hosts with GPUs or TPUs is that you can’t use live migration to update them, setting up a maintenance challenge. In this new docs page, learn how to update accelerator-equipped hosts according to your tolerance for downtime for your training and inference workloads.    
• Documentation: Advanced Compute Images, or ACIs, are standardized image stacks for AI/ML and HPC infrastructure, so you don’t need to manually build your own custom images. In this new docs page, learn how to create an ACI image using the Google Cloud CLI, console, or SchedMD's Slurm workload manager. 
• Guide: AI workloads are notoriously difficult to architect, resource-intensive, and bursty, which can also lead to scaling bottlenecks and large pools of underutilized — or misutilized — compute resources. A new blog outlines the three main ways to achieve dynamic capacity management in Google Cloud: 1) scheduling capacity for planned downtime; 2) maintaining automated fallback capacity for unplanned downtime; and 3) relying on GKE’s core orchestration capabilities to automate resource allocation. 

Customer and partner updates

• Business orchestration software provider UiPath was dealing with spiky workloads, and wanted more predictable costs. To get there, it re-architected its infrastructure, moving from isolated clusters to a shared Google Cloud GPU fleet that included both A3 VM instances (NVIDIA H100 GPUs) for training with G4 VM instances (NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs) for inference. You can read more about their architecture here. 

• Mirendil, an frontier AI lab focused on accelerating AI development, announced that it is using AI Hypercomputer with both TPUs and NVIDIA GPUs to support its model pre-training and post-training applications. 

• Replenit, a retail CRM provider, built its AI decision engine in Google Cloud, using BigQuery, Gemini Enterprise Agent Platform, and open-source Gemma models that it runs on Cloud TPUs. This latter combination provided Replenit with 90% lower pipeline costs than their previous cloud provider, the company reports. Read the full case study for more. 

• Malachyte architected its AI-powered e-commerce recommendation platform on top of Bigtable, Managed Service for Apache Kafka, Pub/Sub, Compute Engine, and last but not least, GKE. See how it all comes together in this blog.

July 2026

Product, technology, and tools updates

• Product update: Google Cloud Managed Lustre is now GA, and available in four distinct performance tiers that deliver throughput ranging from 125 MB/s, 250 MB/s, 500 MB/s, to 1000 MB/s per TiB of capacity — with the ability to scale up to 8 PB of storage capacity. The Managed Lustre solution is powered by DDN’s EXAScaler, combining DDN's decades of leadership in high-performance storage with Google Cloud's expertise in cloud infrastructure.

• Product update: C4N network and storage optimized VMs are now GA. C4N is our first network- and block-storage-optimized VM series built to eliminate data-transfer bottlenecks. Powered by 5th Gen Intel Xeon Scalable processors and built on Google's Titanium offloading hardware, it achieves 400 Gbps network bandwidth, 95 million packets per second (MPPS), and up to 25 GiB/s of block storage throughput when paired with Hyperdisk Extreme.

• New feature: GKE Dataplane V2 up to 15K Nodes with Network Policies (GA). This capability enables standard GKE clusters to scale up to 15,000 nodes while maintaining full active Network Policy enforcement, supporting the massive infrastructure needs of large enterprise and AI/ML customers.

• New feature: Co-operative time-slicing in llm-d. If you’re running reinforcement learning (RL) workloads, you can now interleave independent RL jobs onto shared physical hardware, increasing aggregate accelerator duty cycles from a ~40% baseline up to 70% without impacting model convergence or accuracy. 

• New AI security tool: Looking to secure your AI supply chain on GKE, deploy AI workloads safely, and cut down on shadow AI? We open-sourced k8s-aibom, a lightweight, unprivileged Kubernetes controller that continuously monitors container clusters to automatically detect running AI runtimes (like vLLM and Triton) and generate standard CycloneDX Machine Learning Bill of Materials (ML-BOMs). Check out the k8s-aibom project and get involved.

Practitioner guides and how-tos

• How-to guide: On July 27, Google announced Day 0 support for Moonshot AI’s Kimi K3 2.8-trillion-parameter open-weight model, the day weights were released. Whichever your preferred deployment path — via Model Garden, custom orchestration, or GKE with llm-d recipes — this guide offers detailed step-by-step instructions to help you evaluate and pilot Kimi K3 in Google Cloud. 

• How-to guide: Google Kubernetes Engine (GKE) managed DRANET supports both GPUs and TPUs. There are several configurations to use this implementation, including standard cluster (where you have full control) and autopilot cluster (where Google does the heavy configs for you). Take a deeper dive in the hands-on lab, GKE Autopilot clusters with TPUs, GKE managed DRANET and Gemma 4.

• How-to guide: Learn to run Ray on TPUs, not GPUs. In Part 1 of this two-part series, we discuss TPU slices (hint: Ray thinks of them as just another accelerator on which to schedule), then walk through Ray’s various AI libraries (Part 2).

• How-to guide: Evaluate TPUs for sample workloads using a new microbenchmark suite that helps you accurately assess whether a device is achieving its theoretical performance specifications, and to identify specific performance gaps or architecture-specific bottlenecks. Dive in here. 

• How-to guide: Scale your agents without killing your budget. Learn how GKE orchestration can help you safely pack more agents onto a fixed compute footprint with GKE Agent Sandbox and Pod snapshots. Whether your goal is performance or cost optimization, we teach you how to turn the right dials for optimal agent efficiency. 

• Technical blueprint: Inside the optimization of Mistral 3 large inference on Ironwood. This blog outlines how one Google team optimized Mistral 3 large MoE model inference on Google’s Ironwood (TPU v7x), achieving a 1.5x performance gain. They did so with hybrid sharding, replacing linear VPU summations with tree reductions, optimizing GMM/MLA kernels, and adopting asynchronous scheduling. As a result, they boosted throughput by up to 48% while maintaining benchmark accuracy neutrality. Read the full blog here.

Research, reports and deep-dives

• Report: Google was named a Leader in the inaugural GartnerⓇ Magic Quadrant™ for AI Infrastructure, positioned highest for ‘Ability to Execute’ and furthest for ‘Completeness of Vision’. Gartner called out Google’s proprietary scalable compute, integrated AI Hypercomputer architecture, and the scale of our AI compute capacity as key strengths. Download a copy here.

• Report: We recently surveyed more than 1,400 senior IT leaders for our State of AI Infrastructure report, and a resounding pattern emerged: The gap between AI ambition and infrastructure reality is widening. In fact, 83% of organizations say they require infrastructure upgrades to support production-grade agentic AI. Read the accompanying blog to understand how adapting your infrastructure to meet the demands that agentic applications place on your systems will help you move from pilot to production.

June 2026

Product, technology and tool updates

• Product update: Protecting sensitive data used with AI is a critical part of advanced and secure cloud infrastructure. Confidential Computing cryptographically protects data in use in hardware-based Trusted Execution Environments (TEEs) with verifiable data integrity, and is now available on the accelerator-optimized G4 machine series, featuring NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs. Get started with Confidential G4 VMs and Confidential G4 GKE Nodes.

• Developer resource: The new TPU Developer Hub is the place to go for model builders, optimizers, and developers to learn to unlock the full performance of Google Cloud TPUs. Read more in this blog. 

• New product: Scale your AI workloads with the new OpenTelemetry-Based TPU AI Telemetry Collector Agent. For the first time, you can route high-fidelity TPU hardware telemetry to Google Cloud Monitoring, Google Managed Prometheus, or your own self-hosted Grafana stack.

Practitioner guides and how-tos

• How-to guide: Learn how to build high availability into an AI inference workload running on GKE Inference Gateway with TPUs, Cloud Storage FUSE and Dynamic Resource Allocation (DRA). This blog provides an overview, or you can get all the technical details in the hands-on codelab.

• How-to guide: Did you know you can connect your AI agents to unstructured data in Cloud Storage via Model Context Protocol (MCP)? In this blog, learn about why would want to do that from three customer examples, then how to do it, choosing either a fully managed service, or a self-managed local server for more customization and control. 

Research, reports and deep-dives

• Report: According to an independent benchmark report, GKE Inference Gateway outperforms the next leading managed Kubernetes service with 15.7% higher throughput, 92.8% shorter wait times, and 62.6% lower inter-token latency. This performance can be attributed to its use of prefix caching, which optimizes LLM performance by storing the KV cache (activation states) of long, repetitive prompt prefixes. Learn more in the blog. 

• Architecture deep dive: A closer look at the cold start problem, this time for TPUs and GKE, and how the Run:ai Model Streamer can help change the dynamic. 

Customer and partner updates

• Customer win: Leveraging GKE, BigQuery, Cloud SQL, and Gemini Enterprise Agent Platform, Pager Health is eliminating operational fragmentation to deliver a simplified, personalized U.S. healthcare experience that transforms lives.

• Customer win: Trustpilot, the customer review platform, built a high-volume streaming pipeline using fine-tuned Gemma models with Dataflow and Gemini Enterprise Agent Platform running on cost-optimized A2 VMs using A100 GPUs, as well as optimized version of vLLM maintained by Gemini Enterprise Agent Platform.

May 2026

Product, technology and tool updates

• Product update: GKE Agent Sandbox is now generally available.

• New open-source project: Agent Substrate is a new open-source project aimed at continuing to push the limits of agentic infrastructure density

• New feature: Google AI Edge Portal, a solution for testing and benchmarking on-device machine learning (ML) at scale, now supports benchmarking and debugging on-device LLMs. Read more here. 

• Product deep dive: We went into depth about Cloud Storage Rapid, a new family of high-performance storage offerings for AI workloads. At launch, offerings include Rapid Bucket (formerly Rapid Storage), a high-performance zonal object storage offering, and Rapid Cache (formerly Anywhere Cache), which accelerates reads on-demand and colocates compute and data for workloads in existing buckets. 

Research, reports and deep dives

• Architecture deep dive: Google Global Infrastructure VP Bikash Koley and Engineering Fellow Arjun Singh provide a high-level overview of the challenges that AI workloads pose to network infrastructure, and discuss the deep enhancements we’ve made to our data center fabrics, WAN, and global networks to better support them. 

• Architecture deep dive: We unveiled a new cluster-level reliability model for developing frontier AI models on TPUs, ditching instance-level reliability 

Customer and partner updates

• Customer win: Visual media provider Imgix serves more than 8 billion images and videos from AI Hypercomputer equipped with G4 VMs powered by NVIDIA RTX PRO 6000 Blackwell GPUs.

Spanner Omni, the deploy-anywhere version of Spanner, is now generally available, ready to power your most demanding production workloads in your on-premises data center or on other clouds. 

With Spanner, Google pioneered the distributed SQL market over a decade ago, combining the horizontal scalability of NoSQL with the ACID compliance and strong consistency of a traditional relational database. Since then, Spanner has evolved into an interoperable multi-model database that simplifies complex workloads and powers agentic AI, combining SQL, graph, key-value, full-text search, vector search, and analytical processing with a columnar engine, all in a single database.

When we debuted Spanner Omni at Google Cloud Next ’26, we untethered our distributed database from Google Cloud and brought it directly to your infrastructure. This generated incredible interest from both the enterprise customers and developer community.

Spanner Omni delivers the same core capabilities as the fully managed Spanner service, with the added freedom to deploy it wherever you need it. Whether you’re running virtual machines or Kubernetes in your private data centers, running a multi-cloud deployment that spans multiple clouds, or testing locally on a laptop, Spanner Omni brings Google-grade consistency, availability, scale, and interoperable multi-model capabilities directly to your next agentic AI applications.

Attio accelerates agentic application velocity with Spanner Omni

Spanner Omni enables teams to build once and deploy anywhere, providing application portability across Google Cloud, on-premises and other clouds. Attio, an AI-native CRM company based in London, has built its platform on Spanner. Attio migrated its agentic application environment to a full-featured and containerized Spanner Omni + managed Spanner.

"At Attio, we are building the world's most advanced agentic application environment on Spanner to deliver on our vision for an AI-native CRM platform. Spanner Omni has been a major win for us by delivering Spanner capabilities and performance across all our environments, allowing our agents to bring complex, mission-critical workflows such as the newly announced Spanner queues, to production. With Spanner Omni, we have been able to accelerate our production velocity at a truly enhanced level of scale and confidence that was not possible earlier." - Alexander Christie, Co-Founder & CTO, Attio

Built-in AI capabilities for any environment

Spanner Omni extends Spanner's converged multi-model foundation directly to your private infrastructure and third-party clouds, providing critical capabilities for AI workloads:

• Vector search and Spanner Graph: Natively store and index vector embeddings alongside your relational tables. With support for KNN and ANN search, you can combine semantic similarity with structured SQL filters. Spanner Graph integrates property graphs directly into the engine, allowing you to trace complex entity relationships and connect graph traversals with vector search.

• Model Context Protocol (MCP) support: Integrate directly with agentic systems using MCP Toolbox. This open standard allows autonomous agents to inspect schemas, retrieve relevant context, and use Spanner Omni as an operational memory layer across multi-cloud deployments.

What’s new with Spanner Omni

Since the preview, we’ve been working to expand Spanner Omni’s capabilities, and refine the pricing model for production-ready deployments.

Enterprise features for production deployments

The GA release unblocks the full suite of enterprise-grade capabilities required for mission-critical production workloads, including:

• Robust security and governance: Support for advanced enterprise security, including TLS encryption, authentication and authorization, and audit logging

• Data protection: High-performance backup and restore capabilities to safeguard your data against “fat-finger” deletion or data corruption

• Worker nodes: Dedicated, stateless compute nodes unique to Spanner Omni that are designed to offload background and resource-intensive operations from primary Spanner Omni servers, so they can focus on handling core database workloads

• Enterprise-grade support: Direct access to Google Cloud Customer Care to keep your critical workloads running smoothly 

Flexible licensing and industry-standard pricing

To support you at every stage of development, Spanner Omni offers two distinct licensing tiers designed to fit your scale and budget:

• Developer Edition (free): Tailored for development, testing, and prototyping in non-commercial, non-production environments, it includes all core Spanner features, allowing you to build and validate your applications before scaling to production. The Developer Edition has a default license. The default license lasts for 90-days and includes all features as the commercial edition, except backup features and worker nodes. When used in a single server deployment of 4 vCPUs or less, the default license does not expire and backup-restore is also supported. In case of more than 4vCPUs the default license can be extended beyond 90 days by filling out this form.
• Commercial Edition (paid): Designed for commercial, production workloads, this edition provides the full suite of Spanner Omni capabilities backed by enterprise support. It follows a highly industry-standard, predictable vCPU-based annual subscription model. We also offer a proof-of-concept license for pre-production evaluation at a discounted price. 

Spanner Omni vs. fully managed Spanner on Google Cloud

Our goal with Spanner Omni is to provide parity with the fully managed Spanner service on Google Cloud. However, as self-managed software, deploying and operating Spanner Omni differs from fully managed Spanner on Google Cloud in several ways: 

• Self-managed operations: You are responsible for all day-to-day operations, including routine maintenance, version upgrades, and infrastructure monitoring.

• Availability and SLAs: Because Spanner Omni runs on customer-managed infrastructure, Google does not provide availability SLAs. However, deploying according to our recommended reference architectures will help you achieve comparable high availability.

• Integration with Google Cloud: To ensure it can run anywhere, Spanner Omni excludes capabilities available in managed Spanner that rely on Google Cloud-specific capabilities, such as native integrations with BigQuery, Knowledge Catalog, or Gemini Enterprise and other Google Cloud services.

• Feature-parity roadmap: While some feature gaps exist today between Spanner Omni and fully managed Spanner, we are actively developing updates to close these gaps in future releases.

Get started today

Whether you’re modernizing on-prem legacy systems or building a resilient multi-cloud architecture, Spanner Omni is ready to help you scale. 

• For more information, visit the Spanner Omni website to explore documentation and use cases. 

• To use the Spanner Omni commercial edition with full features, please contact your Google Cloud account team or reach out to us at https://cloud.google.com/consulting/spanner-omni. 

• For developing and testing for non-commercial, non-production purposes, download the Spanner Omni developer edition.

Today, we’re expanding our ecosystem of managed remote MCP servers by introducing the Google Cloud CLI remote MCP server in preview.

Powered by the popular gcloud and bq (BigQuery) command-line tools, this new server gives AI agents immediate, broad access to command-line operations for managing Google Cloud infrastructure and working with advanced BigQuery workflows securely and seamlessly. 

Why CLI matters for AI agents

Agents are increasingly performing complex cloud operations, but standardizing how they interact with backend systems remains a challenge. The Google Cloud CLI remote MCP server bridges this gap by packaging the versatility of hundreds of gcloud and bq commands into one single MCP server. This results in two strong benefits for the agent:

• Higher-level abstractions: CLI commands package complex multi-step workflows, validation checks, and high-level operations into unified commands rather than requiring multi-step API orchestration.

• Leverages model training: LLMs are heavily pre-trained on public command-line documentation, syntaxes, and usage examples, making CLI invocation intuitive and highly accurate for models.

The benefits of putting CLI behind remote MCP

Managing cloud infrastructure with AI agents traditionally requires installing and maintaining Google Cloud CLI binaries inside agent execution environments. The Cloud CLI remote MCP server bridges CLI capabilities with MCP benefits by providing an isolated execution sandbox on Google Cloud infrastructure. This solves key infrastructure challenges:

• Simplified dependency and runtime management: For teams building custom agents, maintaining local CLI versions and dependencies across dev, test, and production environments creates operational overhead. Remote MCP eliminates local installations and runtime maintenance.

• Access for web-based agent endpoints: Web-hosted agent platforms and web interfaces (such as Gemini Enterprise and other hosted enterprise agent platforms) run in environments where users cannot control or install local packages. Remote MCP enables secure, managed access to Google Cloud CLI operations directly from these surfaces.

Enterprise-grade security and governance

Connecting an AI agent to your infrastructure requires strict, enterprise-ready safeguards. This remote server leverages Google Cloud's standard identity and governance frameworks to keep your environments secure:

• Zero ambient credentials: The server isolates execution in a network-restricted proxy boundary with no ambient credentials. Authentication and authorization are handled through Agent Identity, OAuth 2.0, and Identity and Access Management (IAM).

• Strict policy enforcement: Every command executed through the remote MCP server is run with the permissions of the authenticated caller identity. Both standard IAM permissions and organization policy service constraints are strictly enforced against downstream target resources.

• Advanced protection with Model Armor: To minimize the risks associated with AI tool calling, the Cloud CLI remote MCP server integrates with Model Armor. You can proactively screen LLM prompts and responses to protect against risks like prompt injection and malicious inputs.

• Cloud audit logging: The Cloud CLI remote MCP server can be configured to log every tool invocation to Audit Logs (Data Access logs under cloudcli.googleapis.com/mcp). Security teams can gain full visibility into caller identities, OAuth clients, and IAM authorization decisions (mcp.googleapis.com/tools.call) without exposing sensitive command payloads or personally identifiable information (PII).

Connecting to the Google Cloud CLI Remote MCP Server

Integrating cloud management into your agents no longer requires packaging Google Cloud CLI binaries, managing local execution runtimes, or maintaining dependencies inside agent container images. Because the Google Cloud CLI remote MCP server implements the standard Model Context Protocol, any MCP-compatible agent platform or orchestration runtime can connect immediately via standard configuration:

Your agent immediately gains access to execute gcloud and bq commands in a secure, network-isolated cloud sandbox.

Authentication is handled via keyless Agent Identity for hosted Google Cloud platforms, or standard OAuth 2.0 for external runtimes. For authentication options, see the MCP Authentication Guide.

Bringing infrastructure management to agents

The Cloud CLI remote MCP server exposes two powerful tools, run_gcloud_command and run_bq_command, giving your AI agents broad, immediate access to Google Cloud operations through natural language.

Managing cloud infrastructure with run_gcloud_command

With run_gcloud_command, agents can execute the full breadth of gcloud operations to manage, diagnose, and secure your Google Cloud environment. An example follows:

• Observability and incident diagnostics: An agent streamlines incident diagnostics by automating command execution and reducing context-switching across tools.

Extending BigQuery operations with the run_bq_command tool

While the BigQuery MCP server already helps organizations analyze and explore data using AI agents, with the introduction of run_bq_command, agents can now tackle advanced BigQuery tasks such as resource allocation, job monitoring, and task scheduling by unlocking the full scope of bq CLI functionality. Key capabilities include:

• Automating scheduled queries: An agent utilizes BigQuery Data Transfer Service configurations to schedule queries automatically. 

• Job and resource management: Gain deep insight into query execution details, including processed data volume, slot usage, and execution plans, as well as managing reservations.

• Access and permissions control: Data administrators and owners can inspect and update table permissions directly through the agent.

Pricing and availability

The Google Cloud CLI MCP server is available today in public preview. There is no additional charge to use the MCP server itself. You pay only for the GCP resources you create and any applicable data transfer costs.

To get started, enable the Cloud CLI Execution API (`cloudcli.googleapis.com`) in your Google Cloud project, grant the required MCP Tool User (`roles/mcp.toolUser`) IAM role to your agent or user identity, and configure your MCP client to connect to `cloudcli.googleapis.com/mcp`.

• Use the Google Cloud CLI Remote MCP Server Guide

• Google Cloud CLI MCP Reference

• Google Cloud Remote MCP Servers Overview

• Full List of Google OneMCP Servers

• Set up authentication to Google and Google Cloud MCP servers

• Gemini Enterprise Agent Platform & Agent Identity Overview

• Automate Google Cloud with Cloud CLI Remote MCP Server

For years, Swift was perceived mainly as a UI language tied to Apple client devices. With the release of Swift 6 and strict concurrency checking, Swift has matured into a premier systems with cloud engineering language that matches Go and Rust in memory safety while running with predictable, reference-counted CPU performance.

To support this ecosystem, Google engineering has launched the official Google Cloud API Client Libraries for Swift. Built from the ground up for Swift 6.2+, this new SDK uses the latest non-blocking Swift NIO event loops, HTTP/2 multiplexing, gRPC transport, and zero-cost compile-time data race safety.

In this article, we'll walk you through all you need to know to get started, and to understand how the Server Side Cloud Swift SDK, or google-cloud-swift, works.

The rise of server-side Swift and cloud-native concurrency

Traditional backend languages often force a compromise between developer velocity and system resource usage. Managed runtimes rely on heavy garbage collectors that induce tail-latency spikes under heavy traffic, while systems languages can slow down feature iteration.

Server-side Swift balances both ends of the spectrum. Using Automatic Reference Counting (ARC), Swift reclaims memory deterministically without stop-the-world pauses. More importantly, Swift 6 introduces compile-time concurrency checking. When you share state between async tasks across a cloud microservice, the compiler enforces that types conform to Sendable. Data races are caught in your editor before a binary ever compiles or reaches production.

At the network layer, every request to Google Cloud APIs runs over event-driven, non-blocking sockets that scale across multicore Linux server environments without spawning system threads per connection.

Where to use the Swift SDK

The Server Side Cloud Swift SDK is engineered for server, container, and automated DevOps environments.

When you build high-throughput microservices with Swift web frameworks like Hummingbird or Vapor, google-cloud-swift provides native access to Cloud Storage, AI, Identity and Access Management (IAM), and over one hundred other Google Cloud services. You can containerize your executable on Linux and deploy directly to Cloud Run, Google Kubernetes Engine (GKE), or Compute Engine VMs.

Because the SDK compiles on macOS, and Linux, you can develop the backend in your preferred development environment, and then seamlessly deploy to production. And using Swift on both the frontend and backend allows you to share application-specific types across both.

The SDK also excels at platform engineering and DevOps automation. You can author cross-platform CLI utilities and data rotation scripts that run on your developer laptop or inside CI/CD pipelines. These tools authenticate automatically against Google Cloud using Application Default Credentials (ADC) or Workload Identity Federation.

If you're building an iOS, iPadOS, or visionOS app for the Apple App Store, you should not embed google-cloud-swift directly into your client bundle. Shipping Google Cloud service account keys or administrative credentials inside a client binary creates security risks. For direct client-side features, use the Firebase SDK for Apple Platforms to handle user authentication, real-time Firestore sync, and client-side security rules, or route requests through your own Cloud Run backend API.

Getting started with your IDE and packages

Because google-cloud-swift treats Linux and macOS as first-class citizens, you can develop on Apple hardware with Xcode or on Linux workstations with Visual Studio Code and swiftly.

To install the official Swift compiler on Linux workstations using the swiftly CLI installer, run:

Alternatively, you can download prebuilt toolchain tarballs directly from official Swift Downloads for Ubuntu, Debian, Fedora, or Amazon Linux. Note that google-cloud-swift requires Swift 6.2 or later, so verify your compiler version with swift --version after installation.

To resolve Swift Package Manager bare repository trust warnings when cloning across Linux filesystems, configure Git before building with git config --global safe.bareRepository all.

Add the required packages to your Package.swift manifest:

On macOS you need to change the platforms directive:

Notice how Document().with { ... } avoids verbose temporary variables or mutating setters by providing a clean, thread-safe configuration closure.

Networking, transport, and authentication

The repository splits infrastructure primitives into modular packages under packages/:

• swift-google-cloud-auth: Implements Application Default Credentials discovery, service account JWT signing, external account exchange for Workload Identity Federation, and API keys.

• swift-google-cloud-wkt: Provides idiomatic Swift types for Google Protocol Buffer well-known types, including nanosecond-precision Timestamp representations that bridge cleanly to Swift's Date.

• swift-google-cloud-gax: Handles Google API Extensions such as automated retry loops, exponential backoff, and pagination state machines.

When you initialize any client library without arguments, Credentials.default() automatically scans your environment (GOOGLE_APPLICATION_CREDENTIALS, quota project variables, or the local Google Cloud CLI configuration) and authenticates connections over gRPC or HTTP/2.

If you need to programmatically override credentials with an API key or attach custom access headers, you can pass explicit configuration options. For example, you could modify the previous example to use the following:

The autogenerated client ecosystem

Google Cloud operates a vast ecosystem of APIs whose schemas update regularly. The teams supporting google-cloud-swift use code generators to automatically update the client libraries with the latest features and with new APIs. Using code generators produces stable APIs, without disruptive breaking changes. While the releases are on a fixed cadence, please contact Cloud Customer Care if you need a particular feature or API urgently.

Whether you need to rotate keys in Secret Manager or invoke multimodal inference models via the Gemini API on Gemini Enterprise Agent Platform, the generated SDKs follow consistent naming and async/await signatures.

These generated clients offer more than plain unary RPC wrappers. They also offer wrappers that simplify application development. For example, iterating over long results involves fetching pages of results with one RPC, iterating over the page of results, and then preparing a new request to retrieve the following page. Using the generated clients this becomes an asynchronous iterator. This example shows how to query project secrets using GoogleCloudSecretManagerV1:

The pagination response returns an asynchronous sequence (AsyncSequence). You can iterate over items with for try await while the client library fetches subsequent pages in the background over non-blocking NIO channels.

Where to go next

With google-cloud-swift, server-side Swift developers can write end-to-end cloud infrastructure with compile-time race safety, native async/await ergonomic APIs, and zero OS thread congestion.

To inspect the source code, open issues, or contribute new veneers, visit the official repository at googleapis/google-cloud-swift.

Want to discuss server-side Swift architectures or Cloud Run containerization? Join the Google Developer Program to continue the conversation.

This is the first of a two-part series exploring how Google Cloud is bringing the foundational values of a high-performance parallel filesystem–TB/s throughput, sub-ms latency at high client scale, and POSIX support–to a broader set of use cases and users.

Historically, due to the cost and special purpose nature of parallel filesystems, colder data had to be stored outside of the filesystem and AI developers have had to maintain separate, slower environments for writing code, compiling libraries, and managing repositories. This fragmentation increases the toil of manual data staging, dataset copying, and managing disjointed namespaces.

Google Cloud Managed Lustre is solving these problems through our 6 cents/GB*month Dynamic Tier and by optimizing Managed Lustre performance for a range of development tasks and workloads – making Managed Lustre a “One-Stop Shop” for high-performance AI and HPC workloads.

Lower Cost: More Lustre for Less with the Dynamic Tier

The Managed Lustre Dynamic Tier provides sub-ms latency for hot data, which allows you to store all of your data in a single namespace, and costs only 6 cents/GB*month.

• Throughput, capacity scale and client scale: Throughput scales linearly with capacity up to 80 PB, while sub-ms latency for hot data remains stable as you scale to tens of thousands of clients.

• Single-flat fee: Predictable pricing. No independent charges for disk media types, data movement within the namespace, or metadata IOPS.

• Read Latencies: Sub-ms latencies for High-Performance Cache (SSD).  The Capacity Pool (“HDD”) is built on Google Cloud Hyperdisk throughput, which has an average read latency of 10 to 30 ms.

Recommended workloads for Dynamic Tier

• Multi-Epoch Training and/or Training with Optimized Fetch Sizes: Hot data is promoted to the High Performance Cache (SSD) after the first run. Larger data prefetch will allow you to take advantage of the Dynamic Tier cost structure and gain from low-latency SSD.

• Write-Heavy Checkpointing: Bursty checkpoint writes land directly in the High Performance Cache. Older checkpoints are transparently demoted to the Capacity Pool (HDD).

• Rapid Checkpoint Restore:  New checkpoints are written to the High Performance Cache, enabling low-latency checkpoint restores.

• Interactive Snappiness for Developers: Low-latency tasks like git cloning, compiling libraries, or running notebooks benefit from a local-disk feel (~300µs average read latencies) on the same shared workspace hosting large training sets.

Frictionless development: Lustre as a one-stop shop for developer’s workloads

In addition to Managed Lustre’s scalability for large AI and HPC workloads (checkpoint/restart/data-loading), it also meets the demands for interactive work, meaning developers can start on Managed Lustre and stay on Managed Lustre throughout the entire workload lifecycle:

Unified Foundation & Interactive Performance

Consolidates the AI and HPC lifecycle into a single namespace, providing a "local disk" feel for interactive work (Read more about the latency benefits of Managed Lustre experienced by Salesforce and others).

• Latency: ~300µs average read latency—delivering up to 4x better responsiveness than alternative distributed file systems.

• Accelerated Setup: Untar the Linux kernel in ~2 minutes (4.7x faster than alternative file solutions), run a 20-worker parallel git clone of Python in ~40 seconds, compile Python in ~200s.

High-Concurrency Broadcast & Cluster Startup

Managed Lustre maximizes GPU ROI by preventing storage bottlenecks during cluster initialization. 

When thousands of worker nodes attempt to read the exact same file simultaneously (such as a shared model checkpoint, base weights, or container layer), traditional distributed file systems can choke on localized hotspotting, leaving high-cost GPU clusters idle for minutes.

• Improves Aggregate Throughput for a large number of clients reading the same file: Demonstrates a 67% improvement over alternative file solutions.

• Parallel Loading: Imports libraries like PyTorch across 4,000+ processes in under 60 seconds.

Run One-Stop Shop Workflows for Yourself

Here is the code for the tests we’ve run, so that you can perform your own testing.

Low latency for interactive access

We used fio to emulate small, low-concurrency reads and writes:

¹ Storage system specs: 500 MBps per TiB tier of Managed Lustre, 108,000 GiB capacity. Zonal Filestore at 102,400 GiB capacity. Average throughput of 36.7 GB/s to 2,048 client VMs reading the same 40 GiB file.

Accelerated setup

How to run Linux untar

In the above use case, you will want to take care to avoid the metadata performance tax that can come from running as root (Namely, tar issues chown and chmod calls to make extracted files’ owner+permissions match the ones recorded in the archive.).  If you still wish to run as root (and have verified that this approach is compatible with your setup), you may specify `--no-same-owner --no-same-permissions` in order to ensure that extracted files maintain root as owner and have root's default file permissions. In other words, it makes extraction as root behave like extraction as non-root (by ignoring the owner+permissions in the archive).

How to run Python gitclone

How to run Python compile

High scale distribution

Aggregate throughput for distributing one large file to many nodes

Run the below on each client VM:

Parallel loading of libraries across many processes

Run the below on a selected client VM:

Looking ahead and next steps

By eliminating the manual data staging tax and lowering entry costs with the Dynamic Tier, Google Cloud Managed Lustre is evolving from an elite, single-purpose engine into a highly versatile, unified storage fabric for the entire AI lifecycle.

In the second part of this series, we will focus on upcoming object integration features. Stay tuned!

Next steps

• Run the benchmarks yourself (if you haven’t already): Deploy a Google Cloud Managed Lustre instance using the Google Cloud console and run tests provided above to benchmark your own workloads.

• Explore the Dynamic Tier: Read the Google Cloud Managed Lustre Documentation to learn more.

• Stay tuned for Part 2: In the next installment of this series, we will dive deep into upcoming object integration features and how they further simplify AI and HPC storage.

• Get started with centralizing your development-to-training lifecycle on Google Cloud Managed Lustre!

※この投稿は米国時間 2026 年 9 月 22 日に、Google Cloud blog に投稿されたものの抄訳です。

レジリエンスを備えたソフトウェア サプライ チェーンは、最新のデリバリーの基盤であり、継続的インテグレーションと継続的デリバリー（CI / CD）パイプラインを保護することこそが、イノベーションを安全に推し進める原動力となります。

Wiz の最新のクラウド脅威ハイライト レポートによると、2026 年上半期の注目すべきサプライ チェーン攻撃は、2025 年下半期と比較して 2 倍以上に急増しています。

プライベート クラウドにおいて、ソースコードが最も脆弱な部分にならないようにすることが重要です。ソフトウェア サプライ チェーンの脅威に、より効果的に対処できるよう、Google Cloud Secure Source Manager（SSM）では、統一された認証と認可の仕組みを使用して、ソースコードと CI / CD システムを管理できます。

Google は、開発と CI / CD ワークフローを簡素化し、セキュリティを高める 2 つの新機能の一般提供を開始しました。

1. CI / CD システムへの不正アクセス: 攻撃者は、デプロイ スクリプトをたった 1 つ改ざんするだけで、CI / CD パイプラインをマルウェアの拡散ルートに変貌させてしまいます。このリスクを軽減するため、SSM は、バージョン管理システムからビルドやアーティファクトのシステム、デプロイツールに至るまで、たとえ企業ネットワークが侵害された場合でも、CI / CD システムへの不正アクセスをブロックできるようになりました。

2. 権限を持つユーザーによるコードの不正な変更: 新しい Code Owners システムは、ファイル単位、ブランチ単位で pull リクエストの承認者セットを管理し、よりきめ細かい Identity and Access Management（IAM）を提供します。Code Owners は、コードの記述、編集、レビューを行うエンジニアを支援します。リポジトリ内のファイルとディレクトリに、ファイル単位またはブランチ単位で追加の保護ルールを適用します。

主な機能

CI / CD パイプラインへのソースコード変更をはじめとして、新しい Code Owners 機能はきめ細かいマージガードを提供します。CODEOWNERS ファイルをリポジトリにチェックインすることで、必要な承認者を極めて詳細に指定できます。

• パスごとの承認者セット: 柔軟な glob スタイルのパス指定子を使用して、一致するファイルへの変更に対して、指定されたユーザーセットの 1 人以上による承認を必須にすることができます。

• ブランチ固有のガバナンス: セキュリティ ルールとデプロイルールをブランチ間でスムーズに管理できます。同じファイル内で main や dev に異なるオーナーを定義できるため、既存の CODEOWNERS ソリューションで発生するマージ コンフリクトを解消できます。詳しくは、こちらのドキュメントをご覧ください。

• ネスト可能な複数ファイルのオーナー設定: 5,000 行にも及ぶような巨大なルートファイル 1 つに縛られる必要はありません。CODEOWNERS ファイルはサブディレクトリにネストできます。SSM は「ローカル優先」ロジックを採用しているため、サブチームがそれぞれのフォルダのオーナー権限を持ちつつ、ルート管理者はリポジトリ全体に対する拒否権を維持できます。

• 独立した承認セクション: [SectionName][count] 構文（例: [Security Team][2]）により、1 つの pull リクエスト（PR）において、複数の部門からの個別の承認を同時に義務付けることが可能になります。PR 自体はチームのメンバーによってレビューされても、セキュリティ チームのメンバー 2 人による承認が揃うまではマージできないようになっています。

ソースコードの準備が整ったら、SSM の新しい Developer Connect 連携により、CI / CD システムとランタイムが異なるプライベート ネットワークにある場合でも、セキュアかつ簡単に接続できるようになります。

プライベート CI / CD ブループリントのアーキテクチャは、セキュアな接続経路を採用しています。Secure Source Manager が Private Service Connect に接続し、そこからさらに Cloud Build へと接続されます。リポジトリ、ビルドプール、アーティファクト ストレージはすべてプライベート ネットワーク内に配置され、VPC Service Controls（VPC-SC）が多層防御を提供することで、プロキシ エンドポイントへのアクセスを制限します。

次のステップ

ネットワークを保護するには、新しいプライベート ネットワークの統合ガイドに沿って、Developer Connect を使用して SSM を Cloud Build に接続します。

pull リクエストの承認を保護するには、ルート CODEOWNERS ファイルを作成して、一律の IAM「承認者」ロールを、ファイル固有のオーナー権限に置き換えます。

- Google Cloud、プロダクト マネージャー、Logan Henriquez

※この投稿は米国時間 2026 年 9 月 19 日に、Google Cloud blog に投稿されたものの抄訳です。

AI は、かつてないペースでソフトウェア開発を加速させています。しかし、コード生成が拡大するにつれて、コードのセキュリティを確保する課題も増大しています。特に、AI を利用した新たな脆弱性の悪用が懸念されています。こうした課題に対処するため、Google AI およびインフラストラクチャ チームは、セキュリティへの取り組み方を変革しています。この記事では、Google が開発した新しい AI ネイティブなエージェント手法について説明します。この手法は、高精度で網羅的な脆弱性スキャンとパッチ適用を、Google のソフトウェア開発ライフサイクルに体系的かつ直接組み込むものです。Google は、インフラストラクチャにデプロイされる数億行に及ぶコードのあらゆる変更を継続的にスキャンすることで、毎月数百件もの脆弱性がコードベースや本番環境に混入するのを未然に防ぎ、グローバル ネットワーク、AI インフラストラクチャ、そしてユーザーを保護しています。

ソリューション アーキテクチャと実装

従来、テクノロジー業界は、処理に時間がかかり、十分なコンテキストを把握できない単発の大規模なセキュリティ スキャンに頼ってきました。その結果、脆弱性の発見が遅れることがよくあります。Google のアプローチでは、これとは異なり送信前スキャンに焦点を当てています。AI エージェントを活用し、スタックのあらゆるレイヤにおいて、コードのチェックインをリアルタイムで評価します。デベロッパーの使い慣れたツールに送信前スキャンを統合することで、セキュリティは、ルール チェッカーや可読性レビューなどのソフトウェア開発ツールと同じように、開発プロセスにおける日常的なルーティン作業になります。また、AI の観点から見ると、個々のコード変更をスキャンする方が、大規模な 1 回限りのスキャンを実行するよりもはるかに少ないコンテキストで済むため、スキャンの効果が大幅に向上します。

ローカライズされた脅威モデルの重要性

この取り組みのために、Google はオープンソースのマルチ エージェント レビュー ハーネスである Mantis を進化させ、ローカライズされた堅牢な脅威モデルのコホートと照合することで、セキュリティ エージェントの精度を高めました。脅威モデルは、コードから切り離された静的なドキュメントに依存するのではなく、リアルタイムのコードベース メタデータを使用します。スキャン エージェントは、パッケージやライブラリ間の依存関係コールグラフを使用し、脅威モデルのコンテキストを拡張および洗練させることで、精度をさらに向上させます。脅威モデルを継続的な脆弱性スキャンの一部に組み込むことで、開発者が脅威や依存関係を常に更新するようになり、モデルを最新の状態に保つことができます。ローカライズされた正確な脅威モデルデータを使用することは、精度の劇的な向上に直結しており、場合によっては偽陽性率をわずか 3% にまで抑えることができます。

専門のトリアージ エージェントが開発を加速

コード チェックインの一環として脆弱性スキャンを行う場合、エンジニアリングの生産性を妨げないよう、コードを生成する開発者やエージェントに迅速に対応する必要があります。低レイテンシで回答を得るために、2 段階の検証プロセスを実行しています。まず、迅速な軽量スキャンを実行し、その結果を専門のトリアージ エージェントで検証します。このエージェントは、コードの実際の構造をプログラムでチェックし（抽象構文ツリーの解析、コールグラフの走査、事前インデックス化されたドメイン安全ルールを使用）、脆弱なパスが実際に攻撃者によって到達可能であることを証明します。このエージェントは 92% 以上の精度を達成し、1 分以内に作業を完了します。次に、夜間の統合テストの一環として、送信後のスキャンが第 2 の防御層として機能し、オフピーク サイクルを使用して、複数の変更にわたって導入された可能性のある脆弱性をテストします。

バグ修正エージェントがループを閉じる

脆弱性を見つけることは、解決への道のりの半分にすぎません。Google のソリューションの最後のコンポーネントは、スキャン結果と生成された実証コード（脆弱性がどのように再現されるかを示すコード スニペット）を使用して、Google の社内コーディング標準に合致した正確な修正案を自律的に生成する自動バグ修正エージェントです。エージェントは、元の変更リクエストのレビューの一環として、人間によるレビューのために修正を送信します。これにより、検出から解決までの時間がさらに短縮されます。

学んだことと行動喚起

継続的なスキャンをソフトウェア開発ライフサイクルに直接組み込むことは、Google においてまさにゲーム チェンジャーとなりました。AI が提示する修正案は広く採用されており、数多くの脆弱性がコードベースに混入するのを未然に防いでいます。セキュリティ向上を目指す組織であれば、次の原則に従うことで、同様の AI ネイティブなアプローチを導入することが可能です。

1. システムを分離する: バイアスを防ぐため、開発、スキャン、トリアージの各エージェントのハーネス、ルール、コンテキストを分離します。軽量な AI スキャンと決定的かつ構造的な検証を組み合わせることで、レイテンシを短縮し、精度を向上させます。

2. コンテキストを賢く使用する: 既存の脅威モデルをエージェントにフィードします。偽陽性を減らすには、正確なコンテキストが不可欠です。また、最新の脅威モデルを使用することで、送信前スキャンの真陽性率が向上し、チームのセキュリティ ポスチャーの最低基準が引き上げられます。

3. 優れたハーネスを構築する: 基盤となるモデルの選択は重要ですが、マルチ エージェント ハーネスを使用すると、モデル選択のばらつきを補うことができるため、大きな効果が得られます。

4. 修正を自動化する: エージェントを使用して、人間参加型の修正を提案し、解決までの時間をさらに短縮します。

独自の AI ネイティブなセキュリティ変革を始めたい場合は、Mantis がオープンソースとして公開されていますので、ぜひ活用してそのメリットをお役立てください。また、サイバーセキュリティの基礎や、このエージェント型パイプラインを支える他のプラットフォーム（Google Cloud、Gemini Enterprise、Trillium および Ironwood TPU で実行される Gemini モデル）についても詳しく学ぶことができます。また、Google Cloud のセキュアなソフトウェア開発のライフサイクル（SDLC）の取り組みに不可欠な要素として、エージェント型の脆弱性スキャンと修復が Google Cloud のお客様をどのように保護しているかについても、ヒントを得ることができます。

^{この成果の実現に多大な貢献をした主要メンバーである、Stella Voutsina（リード プログラム マネージャー）、Yulong Zhang（シニア スタッフ セキュリティ エンジニア、Mantis）、Nick Galloway（スタッフ セキュリティ エンジニア、Mantis）に心からの敬意を表します。}

- Google、上級エンジニア、Andrés Lagar-Cavilla

- バイス プレジデント兼エンジニアリング フェロー、Parthasarathy Ranganathan

※この投稿は米国時間 2026 年 9 月 17 日に、Google Cloud blog に投稿されたものの抄訳です。

フランスを拠点とする大手多国籍通信プロバイダの Orange では、エンジニアリング チームがデリバリー バックログを一時中断して、全員でクラウド費用のクリーンアップに 1 日を費やす日があります。ランキングが表示され、素敵な特典が用意されています。経験豊富な実務担当者が初心者を指導するため、初心者は実際に作業しながら仕事を覚えます。一日の終わりには、プロジェクトの支援者（スポンサー）も結果を確認できます。

Orange はこれを FinOps Clean Days と呼んでいます。ゲーム要素を取り入れたハッカソンと組み合わせることで、100 人以上の社員が参加する FinOps コミュニティは、社内におけるネット プロモーター スコアで 70 を超える評価を獲得しました。

これらの数字は、業界全体が苦戦しているある課題を示唆しています。最近の State of FinOps レポートでは、エンジニアに行動を起こさせることが、組織が直面する最大の課題の一つとして挙げられています。単なる認識から実際の行動へと移すには、FinOps のアカウンタビリティを醸成し、チームに真剣に取り組んでもらう方法を見つけなければなりません。

つまり、FinOps はビジネスにおける「変革」の課題と言えます。そして、ビジネスの変革に関する課題には、既知の解決策があります。Orange の FinOps リードである Camille Marini 氏に、同社がどのようにしてこれらの課題を克服し、AI の導入と ROI を加速させたのか、また、他の組織がどのようにして同じ道をたどることができるのかについて詳しくお話を伺いました。

Clean Days が効果的な理由

Orange は、FinOps チームの立ち上げ当初から、2 つの原則を掲げています。1 つ目は、Cloud FinOps は責任の共有であり、プロジェクトのすべての関係者がそれぞれの方法で関与する、という点です。そして 2 つ目は、その責任の共有を実現する唯一の道はコミュニケーションと、意図的な変革の取り組みである、という点です。

「FinOps の実践において、組織全体で責任の共有というコンセプトを重視しました」と Marini 氏は語ります。「これは、クラウド セキュリティに対する当社のアプローチと非常によく似ています。責任ある効果的な AI への支出と使用を実現するには、プロジェクトのすべての関係者が、それぞれの方法で FinOps に関与していることをチームに理解してもらう必要がありました。」

これらの原則に基づき、Orange は Google Cloud コンサルティングのサポートを受けて FinOps 実践コミュニティを立ち上げました。このコミュニティは、標準化されたコミュニケーション チャネルで運営されたため、その手法は中心メンバーの枠を超えて広く浸透しました。また、会議では具体的な最適化案や請求の更新情報を共有し、常に「すぐに行動に移せる内容」に徹したことで、すべてのセッションが参加者にとって価値あるものとなりました。

Clean Days は、アジャイルなチームが実際にどのように機能しているかを見極めたうえで考案されました。デプロイが絶え間なく行われるアジャイル環境では、最適化作業がスプリントより優先されることはほとんどありません。デリバリーの優先事項、バックログ、日常業務で、利用できる時間は真っ先に埋まってしまいます。そこで Orange は、専用の時間を確保し、それを共同作業にして、楽しいイベントに仕立て上げました。

このアプローチがなぜ効果的なのかは、McKinsey の「変革の 4 つの構成要素」によって説明がつきます。このフレームワークでは、大規模な組織変更には、次の 4 つの領域にわたるアクションが必要であるとしています。

• 納得と理解:「自分に何が期待されているかを理解し、それに同意している。」

• 公式なメカニズム:「構造、プロセス、システムによって変化が促されている。」

• ロール モデリング:「リーダーや同僚の行動が変化している。」

• 人材とスキル:「新しい方法で行動するためのスキルと機会が与えられている。」

Orange の手法をこれらのブロックにマッピングすると、パターンが見えてきます。ゲーム要素と報酬により、エンジニアたちにお手本となる同僚が生まれます。ランキングによってさまざまな行動が可視化され、スポンサーはすぐに成果を確認できます。経験豊富な実務者が初心者を指導することで、コミュニティ自体を通じて人材とスキルが育成されます。定期的なセッションで最適化と請求に関する最新情報を共有することで、資金の使途を把握しているという確信が生まれます。

100 人を超える場合

100 人の熱心なコミュニティを築くことは、それだけでも十分に大きな成果です。しかし、数千人のエンジニアを抱える組織では、中央の FinOps チームが全員に直接連絡することはできません。リーダーにとっての課題は、Orange のようなコミュニティがもたらすもの、つまり意識改革、当事者意識の共有、行動の習慣化を、FinOps チームが直接アプローチできない人々に、どのように浸透させていくかです。

ここで、AI エージェントの出番です。AI エージェントは 2 つの主なメリットをもたらし、FinOps チームの能力を高めます。これまで人間が対応せざるを得なかった、複雑で時間のかかる作業をエージェントが引き受け、社内の誰もが摩擦なしに FinOps に取り組めるようになります。

エージェントをチームに定着させるには、自社が抱える具体的な課題に焦点を当てた戦略が必要です。こうした課題は多くの場合、認知負荷の高さ、責任の所在のあいまいさ、あるいは優先事項の競合などから生じています。まずは、FinOps ライフサイクルにおいてエンゲージメントが低下するポイントを特定します。

• 認識のギャップ: チームが支出の影響を把握していない場合、分析情報エージェントが、日常的に使用するツールにリアルタイムの費用データを push できます。

• リソースのギャップ: エンジニアがバックログの消化に追われているなら、修復エージェントがすぐに効果の出る対策を特定し、そのままマージできる状態のコード変更として提示してくれます。

• 複雑さのギャップ: レポート作成が面倒なルーティン作業のように感じられる場合、オーケストレーション エージェントがデータを収集してプロセスを簡素化できます。

信頼から始め、自律性を追加する

賢明な進め方は、段階を踏んで進めることです。Orange のように、コミュニティ プラクティスを確立します。次に、情報提供と提案を行う読み取り専用エージェントを導入します。コミュニティ全体でこれらが定着して初めて、システムに直接変更を加えるエージェントの構築へと進むべきです。直接的なアクションには運用上のリスクが伴うため、慎重に管理してください。しかし同時に、そこにこそ、最も大きな成果が眠っているのも事実です。

どのように構築するかは、誰が構築を担うかによって異なります。最小限のコードで迅速にデプロイしたいチームの場合は、Gemini Enterprise app がエージェントを作成するためのノーコード環境を提供します。きめ細かい制御を必要とするデベロッパー向の場合は、Gemini Enterprise Agent Platform（旧称 Vertex AI）が、Agent Development Kit（ADK）などのフレームワークで構築されたエージェントをリリースして管理するための高度なツールを提供します。

Cloud FinOps は、一元化されたレポートから、実際の開発現場で即座にアクションを起こすフェーズへと移行しつつあります。この変革に成功している組織は、まず「文化の醸成」から始め、次に AI エージェントを活用することで、一つのチームの限界を超えた組織全体への拡大を実現しています。

Orange の数値はコミュニティの取り組みから得られたものです。その基盤を構築することが、最初に真似るべき部分です。この取り組みをさらに拡大する準備ができたら、Google Cloud コンサルティングがコミュニティ プラクティスの形成を支援します。また、Gemini Enterprise アプリを活用すれば、最初の読み取り専用エージェントを手軽にチームへ導入することができます。

- Google、AI Transformation および FinOps コンサルタント、Samuel Moss

- Google、AI Transformation コンサルタント、Celine Devie

We hereby declare September to be scalability month! As the world prepares for a surge of agentic fleets, we are shoring up our AI infrastructure and orchestration offerings to gracefully — and quickly — respond to that demand, all while maintaining workload isolation and security, and keeping costs in check. Read on to learn how these enhancements manifest across Google Cloud’s compute, network, storage, and orchestration offerings, plus new ways customers are using Google Cloud AI infrastructure, and third-party industry validation of our strategy.

Product, technology, and tools updates

Google Kubernetes Engine updates: The GKE team is all about improving the scalability of the platform, and in September, those improvements came in many shapes and sizes:

• New feature: Need an execution runtime with higher density for your agentic workloads? We engineered the new open-source GKE Agent Substrate to run millions of sandboxes with 10x higher density than standard container runtimes. Agent Substrate also delivers sub-500ms resume operations at over 500 suspend/resume activations per second with a native zero-trust kernel and network isolation. 

• Product update: GKE now has scale-to-zero capabilities built-in. No need to configure complex components to scale your workloads down, thanks to the HPA with the Autoscaling Metric and support for KEP-2021, which do the job for you, out of the box. Read the blog to learn more. 

• Product update: Further, the GKE HPA (with the above-mentioned Autoscaling Metric) now lets you scale up and down based on custom PromQL metrics, in addition to standard metrics, allowing you to trigger workloads according to conditions that are meaningful and unique to your business. Read more here. 

• New feature: Yet another scalability feature is GKE Pod snapshots, which lets you save the running state of your workload, including CPU and GPU memory, and restore it on demand. According to internal tests, GKE Pod snapshots can reduce AI inference start-up by as much as 89%. Learn more here.

• New migration tool: Finally, if you’ve always wanted to migrate your container workloads from AWS EKS to GKE but feared a daunting, high-friction engineering endeavor, we’ve just launched GKE agentic migration, a purpose-built agent plugin that replaces brittle, ad-hoc prompting with an AI-assisted migration pipeline protected by deterministic guardrails. Designed as a compilation of agent skills and a local Model Context Protocol (MCP) server, it uses AI to translate complex AWS EKS IaC and Kubernetes manifests directly into GKE landing zones. Get started with the onboarding guide.

• Feature updates: Reinforcement learning (RL) and evaluation workloads are a beast: In a standard agentic RL loop, an LLM policy generates actions like code snippets on GPUs and executes them inside isolated CPU sandboxes to observe a reward signal. However, when scaling up this loop to support tens of thousands of parallel rollouts, infrastructure bottlenecks emerge, for instance idle accelerators, image cardinality, and a saturated control plane. To help, we developed GKE Agent Sandbox optimized for RL, plus an Agent Sandbox RL orchestration SDK and native integrations for popular RL gyms and harnesses. All are now generally available, and you can learn more here.

Storage updates: AI trains and creates lots of data, and that data has to live somewhere — in block storage systems, file systems, object stores and databases. We announced enhancements to our storage portfolio to help this critical layer roll with the agentic punches:  

• Product update: Filestore agent volumes offer high-performance, elastic, persistent file storage for agentic workloads. Thanks to its tight integration with GKE Agent Substrate and GKE Agent Sandbox, Filestore agent volumes automatically allocates and attaches a dedicated, isolated file workspace to GKE agent sandboxes in milliseconds. Request access to the preview here. 

• New product: If you run generative AI and RAG data layers  — think Milvus, Pinecone, Qdrant, Vespa, Redis, and in-memory context caching — you may want to take a look at the M4N family of VMs, now GA, which offers the highest per-core IOPS and throughput of leading hyperscalers. Paired with Google Cloud's custom Titanium offload architecture and paired with Hyperdisk Extreme, M4N instances deliver up to 25,000 MiB/s (25 GiB/s) of aggregate host storage performance and up to 1 million IOPS. 

• New product: Another new Compute Engine product, Z4D, is GA, and a strong storage solution for AI/ML training and inference workloads. When configured as a bare metal instance, Z4D provides both the high local SSD (LSSD) capacity and low latency required by agentic microVMs, so you can run thousands of isolated sandboxes per host with native performance and efficiency. Learn about Z4D machines here.

• Product update: Today’s AI training and inference pipelines create data faster than most storage management systems can keep up, creating challenges for teams trying to understand their storage estates. A new version of Storage Intelligence advisor makes it easier to answer the question: "What’s in my buckets?" and quickly identify unexpected changes. Then, enhanced batch operations let you automate bulk changes across your buckets — say, move storage classes, mass-delete stale or temporary data, or apply metadata, tagging, retention, or encryption changes. Learn about the latest in Storage Intelligence advisor here.

• New product: Last but not least, a new version of AlloyDB for PostgreSQL brings together pioneering Google infrastructure — Colossus distributed file system, and Jupiter network — to power a new, no-compromises database architecture for the agentic era. 

Practitioner guides and how-tos

• How-to: Wish you could make GPUs and TPUs scattered around the globe behave as a single pool behind a single entry point? In this blog, we show you how to do just that. At the edge, the multi-cluster GKE Inference Gateway focuses on global, multi-region traffic distribution and high availability. Beneath that, the LLM-d router handles complex, memory-aware scheduling algorithms to keep utilization high. This architecture is deliberately runtime-, model-, and accelerator-agnostic, and in tests, routing traffic through the multi-cluster GKE Inference Gateway added less than 1% overhead.

• Guide: Using or planning to use GKE on TPUs for AI model training or inference? Training massive Large Language Models (LLMs) or running high-throughput inference serving represents a significant investment in specialized AI hardware, such as Cloud TPUs and GPUs. To get the most out of every dollar spent, you need to understand workload lifecycle metrics in GKE. This detailed guide explains how to turn opaque cluster behaviors into actionable telemetry.

Customer and partner updates

• GKE customer: Learn why gaming startup SeaVerse relies on GKE Agent Sandbox for its multi-tenant workloads, and how the platform helped it decrease its infrastructure costs by 60%. 

Research, reports and deep-dives

In case you missed it, we’re also thrilled to share that Google has been named a leader, including achieving the highest score on either product or strategy, in three key analyst reports from Gartner and Forrester. 

• Google is a leader in The Forrester Wave™: Public Cloud Platforms, Q3 2026: Highest overall score of any cloud provider!
• Google named a Leader in 2026 Gartner® Magic Quadrant™ for Strategic Cloud Platform Services: Positioned furthest for “Completeness of Vision” of all vendors evaluated.
• Google is a Leader in the 2026 Gartner Magic Quadrant for Container Management: Positioned highest in “Ability to Execute” of all vendors evaluated.
• Google Cloud achieved a Gold rating in the latest SemiAnalysis ClusterMax 3.0 report, which evaluates the reliability, performance, support, pricing, and security of GPU providers globally. See the full report for more.

August 2026

Product, technology, and tools updates

• Product update: Filestore, Google Cloud’s first-party, secure, scalable NFS file service, has emerged as a popular storage platform for AI and agentic workflows, and now, it’s even better suited to the task, with a new backend storage layer built directly on Colossus, Google’s foundational distributed storage system. This new backend lets you provision IOPS independently from storage capacity, and is deeply integrated with GKE. In AI environments, this can help you service so-called agentic swarms — large groups of agents that need to read and write to a common dataset — without a drop off in performance. For more, check out the blog post. 

• New feature: gVisor sandboxes are now available in distributed Ray clusters on GKE. In partnership with Anyscale, we introduced an experimental library for Ray that brings gVisor, Google’s open-source application kernel, directly into distributed Ray clusters. gVisor provides lightweight environments with stronger isolation than ordinary containers, plus fast startup times and low memory overhead. To try out these sandboxing capabilities on GKE, head over to the Ray sandboxing User Guide.

• Product update: Looking for high-performance, easy-to-use infrastructure on which to run a personal AI agent, but don’t want to spend a lot of money? New Cloud Run instances are dedicated, singleton compute runtimes on Cloud Run that won’t shut down when the agent is idle. Better yet, the cost to run a Cloud Run instance with 1 vCPU and 1 GiB of memory continuously for 30 days is just $5.70.  

Practitioner guides, documentation and how-tos

• How-to guide: Big news in Model Context Protocol (MCP) land: As of the 2026-07-28 specification, the protocol core is “completely stateless. The handshake is gone. The initialize / initialized handshake (SEP-2575) and the logical Mcp-Session-Id header (SEP-2567) have been removed entirely. Instead, every request is now self-describing and independent.” Whoa. Learn more about the changes that the latest MCP specification brings, and more importantly, how to implement them, in this Google Developers blog.  
• Guide: Real-time AI systems make a mess of traditional network load balancing techniques. “Instead of handling isolated requests, the backend has to manage a continuous, live bidirectional stream. You’re dealing with a constant stream of audio chunks, transcripts, model outputs, and synthesized speech flowing back and forth simultaneously.” Things only get worse when the user gets involved. “The server has to immediately halt its current speech generation, pivot to update the context, maybe trigger a new tool, and start drafting a different response; this must be done without dropping the connection.” For a new approach to managing load in the AI era, read Scaling real-time AI agents with session-aware load balancing.
• How-to: Learn how to build an elastic, scalable LLM inference platform on GKE, even with a mix of different GPU accelerators. The proposed architecture combines Capacity Advisor and Compute Advisor, plus high-performance storage like RunAI:model streamer or GCPFuse with parallel downloads. Get all the details here.
• Documentation: The thing about hosts with GPUs or TPUs is that you can’t use live migration to update them, setting up a maintenance challenge. In this new docs page, learn how to update accelerator-equipped hosts according to your tolerance for downtime for your training and inference workloads.    
• Documentation: Advanced Compute Images, or ACIs, are standardized image stacks for AI/ML and HPC infrastructure, so you don’t need to manually build your own custom images. In this new docs page, learn how to create an ACI image using the Google Cloud CLI, console, or SchedMD's Slurm workload manager. 
• Guide: AI workloads are notoriously difficult to architect, resource-intensive, and bursty, which can also lead to scaling bottlenecks and large pools of underutilized — or misutilized — compute resources. A new blog outlines the three main ways to achieve dynamic capacity management in Google Cloud: 1) scheduling capacity for planned downtime; 2) maintaining automated fallback capacity for unplanned downtime; and 3) relying on GKE’s core orchestration capabilities to automate resource allocation. 

Customer and partner updates

• Business orchestration software provider UiPath was dealing with spiky workloads, and wanted more predictable costs. To get there, it re-architected its infrastructure, moving from isolated clusters to a shared Google Cloud GPU fleet that included both A3 VM instances (NVIDIA H100 GPUs) for training with G4 VM instances (NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs) for inference. You can read more about their architecture here. 

• Mirendil, an frontier AI lab focused on accelerating AI development, announced that it is using AI Hypercomputer with both TPUs and NVIDIA GPUs to support its model pre-training and post-training applications. 

• Replenit, a retail CRM provider, built its AI decision engine in Google Cloud, using BigQuery, Gemini Enterprise Agent Platform, and open-source Gemma models that it runs on Cloud TPUs. This latter combination provided Replenit with 90% lower pipeline costs than their previous cloud provider, the company reports. Read the full case study for more. 

• Malachyte architected its AI-powered e-commerce recommendation platform on top of Bigtable, Managed Service for Apache Kafka, Pub/Sub, Compute Engine, and last but not least, GKE. See how it all comes together in this blog.

July 2026

Product, technology, and tools updates

• Product update: Google Cloud Managed Lustre is now GA, and available in four distinct performance tiers that deliver throughput ranging from 125 MB/s, 250 MB/s, 500 MB/s, to 1000 MB/s per TiB of capacity — with the ability to scale up to 8 PB of storage capacity. The Managed Lustre solution is powered by DDN’s EXAScaler, combining DDN's decades of leadership in high-performance storage with Google Cloud's expertise in cloud infrastructure.

• Product update: C4N network and storage optimized VMs are now GA. C4N is our first network- and block-storage-optimized VM series built to eliminate data-transfer bottlenecks. Powered by 5th Gen Intel Xeon Scalable processors and built on Google's Titanium offloading hardware, it achieves 400 Gbps network bandwidth, 95 million packets per second (MPPS), and up to 25 GiB/s of block storage throughput when paired with Hyperdisk Extreme.

• New feature: GKE Dataplane V2 up to 15K Nodes with Network Policies (GA). This capability enables standard GKE clusters to scale up to 15,000 nodes while maintaining full active Network Policy enforcement, supporting the massive infrastructure needs of large enterprise and AI/ML customers.

• New feature: Co-operative time-slicing in llm-d. If you’re running reinforcement learning (RL) workloads, you can now interleave independent RL jobs onto shared physical hardware, increasing aggregate accelerator duty cycles from a ~40% baseline up to 70% without impacting model convergence or accuracy. 

• New AI security tool: Looking to secure your AI supply chain on GKE, deploy AI workloads safely, and cut down on shadow AI? We open-sourced k8s-aibom, a lightweight, unprivileged Kubernetes controller that continuously monitors container clusters to automatically detect running AI runtimes (like vLLM and Triton) and generate standard CycloneDX Machine Learning Bill of Materials (ML-BOMs). Check out the k8s-aibom project and get involved.

Practitioner guides and how-tos

• How-to guide: On July 27, Google announced Day 0 support for Moonshot AI’s Kimi K3 2.8-trillion-parameter open-weight model, the day weights were released. Whichever your preferred deployment path — via Model Garden, custom orchestration, or GKE with llm-d recipes — this guide offers detailed step-by-step instructions to help you evaluate and pilot Kimi K3 in Google Cloud. 

• How-to guide: Google Kubernetes Engine (GKE) managed DRANET supports both GPUs and TPUs. There are several configurations to use this implementation, including standard cluster (where you have full control) and autopilot cluster (where Google does the heavy configs for you). Take a deeper dive in the hands-on lab, GKE Autopilot clusters with TPUs, GKE managed DRANET and Gemma 4.

• How-to guide: Learn to run Ray on TPUs, not GPUs. In Part 1 of this two-part series, we discuss TPU slices (hint: Ray thinks of them as just another accelerator on which to schedule), then walk through Ray’s various AI libraries (Part 2).

• How-to guide: Evaluate TPUs for sample workloads using a new microbenchmark suite that helps you accurately assess whether a device is achieving its theoretical performance specifications, and to identify specific performance gaps or architecture-specific bottlenecks. Dive in here. 

• How-to guide: Scale your agents without killing your budget. Learn how GKE orchestration can help you safely pack more agents onto a fixed compute footprint with GKE Agent Sandbox and Pod snapshots. Whether your goal is performance or cost optimization, we teach you how to turn the right dials for optimal agent efficiency. 

• Technical blueprint: Inside the optimization of Mistral 3 large inference on Ironwood. This blog outlines how one Google team optimized Mistral 3 large MoE model inference on Google’s Ironwood (TPU v7x), achieving a 1.5x performance gain. They did so with hybrid sharding, replacing linear VPU summations with tree reductions, optimizing GMM/MLA kernels, and adopting asynchronous scheduling. As a result, they boosted throughput by up to 48% while maintaining benchmark accuracy neutrality. Read the full blog here.

Research, reports and deep-dives

• Report: Google was named a Leader in the inaugural GartnerⓇ Magic Quadrant™ for AI Infrastructure, positioned highest for ‘Ability to Execute’ and furthest for ‘Completeness of Vision’. Gartner called out Google’s proprietary scalable compute, integrated AI Hypercomputer architecture, and the scale of our AI compute capacity as key strengths. Download a copy here.

• Report: We recently surveyed more than 1,400 senior IT leaders for our State of AI Infrastructure report, and a resounding pattern emerged: The gap between AI ambition and infrastructure reality is widening. In fact, 83% of organizations say they require infrastructure upgrades to support production-grade agentic AI. Read the accompanying blog to understand how adapting your infrastructure to meet the demands that agentic applications place on your systems will help you move from pilot to production.

June 2026

Product, technology and tool updates

• Product update: Protecting sensitive data used with AI is a critical part of advanced and secure cloud infrastructure. Confidential Computing cryptographically protects data in use in hardware-based Trusted Execution Environments (TEEs) with verifiable data integrity, and is now available on the accelerator-optimized G4 machine series, featuring NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs. Get started with Confidential G4 VMs and Confidential G4 GKE Nodes.

• Developer resource: The new TPU Developer Hub is the place to go for model builders, optimizers, and developers to learn to unlock the full performance of Google Cloud TPUs. Read more in this blog. 

• New product: Scale your AI workloads with the new OpenTelemetry-Based TPU AI Telemetry Collector Agent. For the first time, you can route high-fidelity TPU hardware telemetry to Google Cloud Monitoring, Google Managed Prometheus, or your own self-hosted Grafana stack.

Practitioner guides and how-tos

• How-to guide: Learn how to build high availability into an AI inference workload running on GKE Inference Gateway with TPUs, Cloud Storage FUSE and Dynamic Resource Allocation (DRA). This blog provides an overview, or you can get all the technical details in the hands-on codelab.

• How-to guide: Did you know you can connect your AI agents to unstructured data in Cloud Storage via Model Context Protocol (MCP)? In this blog, learn about why would want to do that from three customer examples, then how to do it, choosing either a fully managed service, or a self-managed local server for more customization and control. 

Research, reports and deep-dives

• Report: According to an independent benchmark report, GKE Inference Gateway outperforms the next leading managed Kubernetes service with 15.7% higher throughput, 92.8% shorter wait times, and 62.6% lower inter-token latency. This performance can be attributed to its use of prefix caching, which optimizes LLM performance by storing the KV cache (activation states) of long, repetitive prompt prefixes. Learn more in the blog. 

• Architecture deep dive: A closer look at the cold start problem, this time for TPUs and GKE, and how the Run:ai Model Streamer can help change the dynamic. 

Customer and partner updates

• Customer win: Leveraging GKE, BigQuery, Cloud SQL, and Gemini Enterprise Agent Platform, Pager Health is eliminating operational fragmentation to deliver a simplified, personalized U.S. healthcare experience that transforms lives.

• Customer win: Trustpilot, the customer review platform, built a high-volume streaming pipeline using fine-tuned Gemma models with Dataflow and Gemini Enterprise Agent Platform running on cost-optimized A2 VMs using A100 GPUs, as well as optimized version of vLLM maintained by Gemini Enterprise Agent Platform.

May 2026

Product, technology and tool updates

• Product update: GKE Agent Sandbox is now generally available.

• New open-source project: Agent Substrate is a new open-source project aimed at continuing to push the limits of agentic infrastructure density

• New feature: Google AI Edge Portal, a solution for testing and benchmarking on-device machine learning (ML) at scale, now supports benchmarking and debugging on-device LLMs. Read more here. 

• Product deep dive: We went into depth about Cloud Storage Rapid, a new family of high-performance storage offerings for AI workloads. At launch, offerings include Rapid Bucket (formerly Rapid Storage), a high-performance zonal object storage offering, and Rapid Cache (formerly Anywhere Cache), which accelerates reads on-demand and colocates compute and data for workloads in existing buckets. 

Research, reports and deep dives

• Architecture deep dive: Google Global Infrastructure VP Bikash Koley and Engineering Fellow Arjun Singh provide a high-level overview of the challenges that AI workloads pose to network infrastructure, and discuss the deep enhancements we’ve made to our data center fabrics, WAN, and global networks to better support them. 

• Architecture deep dive: We unveiled a new cluster-level reliability model for developing frontier AI models on TPUs, ditching instance-level reliability 

Customer and partner updates

• Customer win: Visual media provider Imgix serves more than 8 billion images and videos from AI Hypercomputer equipped with G4 VMs powered by NVIDIA RTX PRO 6000 Blackwell GPUs.

Today, we’re expanding our ecosystem of managed remote MCP servers by introducing the Google Cloud CLI remote MCP server in preview.

Powered by the popular gcloud and bq (BigQuery) command-line tools, this new server gives AI agents immediate, broad access to command-line operations for managing Google Cloud infrastructure and working with advanced BigQuery workflows securely and seamlessly. 

Why CLI matters for AI agents

Agents are increasingly performing complex cloud operations, but standardizing how they interact with backend systems remains a challenge. The Google Cloud CLI remote MCP server bridges this gap by packaging the versatility of hundreds of gcloud and bq commands into one single MCP server. This results in two strong benefits for the agent:

• Higher-level abstractions: CLI commands package complex multi-step workflows, validation checks, and high-level operations into unified commands rather than requiring multi-step API orchestration.

• Leverages model training: LLMs are heavily pre-trained on public command-line documentation, syntaxes, and usage examples, making CLI invocation intuitive and highly accurate for models.

The benefits of putting CLI behind remote MCP

Managing cloud infrastructure with AI agents traditionally requires installing and maintaining Google Cloud CLI binaries inside agent execution environments. The Cloud CLI remote MCP server bridges CLI capabilities with MCP benefits by providing an isolated execution sandbox on Google Cloud infrastructure. This solves key infrastructure challenges:

• Simplified dependency and runtime management: For teams building custom agents, maintaining local CLI versions and dependencies across dev, test, and production environments creates operational overhead. Remote MCP eliminates local installations and runtime maintenance.

• Access for web-based agent endpoints: Web-hosted agent platforms and web interfaces (such as Gemini Enterprise and other hosted enterprise agent platforms) run in environments where users cannot control or install local packages. Remote MCP enables secure, managed access to Google Cloud CLI operations directly from these surfaces.

Enterprise-grade security and governance

Connecting an AI agent to your infrastructure requires strict, enterprise-ready safeguards. This remote server leverages Google Cloud's standard identity and governance frameworks to keep your environments secure:

• Zero ambient credentials: The server isolates execution in a network-restricted proxy boundary with no ambient credentials. Authentication and authorization are handled through Agent Identity, OAuth 2.0, and Identity and Access Management (IAM).

• Strict policy enforcement: Every command executed through the remote MCP server is run with the permissions of the authenticated caller identity. Both standard IAM permissions and organization policy service constraints are strictly enforced against downstream target resources.

• Advanced protection with Model Armor: To minimize the risks associated with AI tool calling, the Cloud CLI remote MCP server integrates with Model Armor. You can proactively screen LLM prompts and responses to protect against risks like prompt injection and malicious inputs.

• Cloud audit logging: The Cloud CLI remote MCP server can be configured to log every tool invocation to Audit Logs (Data Access logs under cloudcli.googleapis.com/mcp). Security teams can gain full visibility into caller identities, OAuth clients, and IAM authorization decisions (mcp.googleapis.com/tools.call) without exposing sensitive command payloads or personally identifiable information (PII).

Connecting to the Google Cloud CLI Remote MCP Server

Integrating cloud management into your agents no longer requires packaging Google Cloud CLI binaries, managing local execution runtimes, or maintaining dependencies inside agent container images. Because the Google Cloud CLI remote MCP server implements the standard Model Context Protocol, any MCP-compatible agent platform or orchestration runtime can connect immediately via standard configuration:

Your agent immediately gains access to execute gcloud and bq commands in a secure, network-isolated cloud sandbox.

Authentication is handled via keyless Agent Identity for hosted Google Cloud platforms, or standard OAuth 2.0 for external runtimes. For authentication options, see the MCP Authentication Guide.

Bringing infrastructure management to agents

The Cloud CLI remote MCP server exposes two powerful tools, run_gcloud_command and run_bq_command, giving your AI agents broad, immediate access to Google Cloud operations through natural language.

Managing cloud infrastructure with run_gcloud_command

With run_gcloud_command, agents can execute the full breadth of gcloud operations to manage, diagnose, and secure your Google Cloud environment. An example follows:

• Observability and incident diagnostics: An agent streamlines incident diagnostics by automating command execution and reducing context-switching across tools.

Extending BigQuery operations with the run_bq_command tool

While the BigQuery MCP server already helps organizations analyze and explore data using AI agents, with the introduction of run_bq_command, agents can now tackle advanced BigQuery tasks such as resource allocation, job monitoring, and task scheduling by unlocking the full scope of bq CLI functionality. Key capabilities include:

• Automating scheduled queries: An agent utilizes BigQuery Data Transfer Service configurations to schedule queries automatically. 

• Job and resource management: Gain deep insight into query execution details, including processed data volume, slot usage, and execution plans, as well as managing reservations.

• Access and permissions control: Data administrators and owners can inspect and update table permissions directly through the agent.

Pricing and availability

The Google Cloud CLI MCP server is available today in public preview. There is no additional charge to use the MCP server itself. You pay only for the GCP resources you create and any applicable data transfer costs.

To get started, enable the Cloud CLI Execution API (`cloudcli.googleapis.com`) in your Google Cloud project, grant the required MCP Tool User (`roles/mcp.toolUser`) IAM role to your agent or user identity, and configure your MCP client to connect to `cloudcli.googleapis.com/mcp`.

• Use the Google Cloud CLI Remote MCP Server Guide

• Google Cloud CLI MCP Reference

• Google Cloud Remote MCP Servers Overview

• Full List of Google OneMCP Servers

• Set up authentication to Google and Google Cloud MCP servers

• Gemini Enterprise Agent Platform & Agent Identity Overview

• Automate Google Cloud with Cloud CLI Remote MCP Server