Storage & Data Transfer

Introducing Google Cloud’s U4 compute: Enabling ultra-low latency trading

Wed, 07 Oct 2026 16:00:00 +0000

In capital markets, operational success is built on predictable performance and low latency, where even tiny variations in packet timing alter queue priority and trade execution. At the same time, both exchange venues and their participants face mounting constraints in traditional on-premises data centers — from power and physical rack space limits to lengthy hardware procurement cycles. Increasingly, global capital markets seek the speed and determinism of physical co-location combined with the dynamic scalability and automation of the cloud.

At Google Cloud Next ‘26, we announced the Ultra Low Latency (ULL) Solution, now generally available, providing high-frequency trading workflows that run in the cloud with an ultra-low latency network and agility. The ULL Solution includes the new U4 machine family, also generally available today.

The ULL Solution is based on three infrastructure pillars:

Hardware-level networking: Scaleable, hardware-based multicast data distribution for reliable market-data feeds
Advanced networking observability: Dynamic network traffic capture with hardware-level timing accuracy, facilitating consistent auditing, market replays, and absolute trade validation without impacting primary traffic performance
Deterministic high-performance compute: Processing of latency-critical execution tiers in a highly predictable, consistent amount of time, every single time

Taken together, the ULL Solution’s compute, storage, networking, and observability provide financial exchanges and market participants with a number of technical capabilities:

Bare metal performance: Dedicated bare metal compute provides direct access to physical host resources, minimizing jitter and delivering predictable, low-latency execution for market-data feeds and order routing.
High-performance storage options: Local Titanium SSDs handle real-time transaction and tick logging on the host, complemented by scalable Google Cloud Hyperdisk for persistent market data archives and analytics.
Physical traffic isolation: The ULL trading fabric’s redundant A/B multicast market feeds are accessed through two independent and dedicated Titanium adapters, while an independent third Titanium adapter offloads telemetry, management, and provides access to Google Cloud services.
Hardware-accelerated multicast distribution: The ULL network architecture supports hardware-level multicast feed ingestion. This allows participants to stream high-throughput market data directly to low-latency trading applications, bypassing traditional hypervisor-level virtual switches.
Accelerated packet processing: Support for OpenOnload and DPDK enables Linux user-space networking to deliver predictable unicast and multicast packet handling while minimizing application code changes.
Precision timing and UTC synchronization: Integration with Google Cloud’s Firefly clock synchronization system allows the solution to consistently achieve sub-10 nanosecond network-level timestamping and better synchronization to UTC than the sub-100 microsecond regulatory requirement for financial exchanges.
Built-in telemetry: 24/7 low-latency packet capture and seamless out-of-band packet brokering for regulatory compliance and real-time analytics helps ensure deep visibility without impacting primary traffic performance.
24-7 market ready: The solution is designed for continuous, round-the-clock trading readiness by isolating production workloads in a dedicated primary zone for live trading, while routine cloud maintenance and qualification testing occur in a secondary zone for updates and testing.

Compute in the ULL Solution is delivered by the new U4 machine family, which brings predictable performance and ultra-low latency compute in three specialized machine series: dual-socket bare metal instances with three physical NICs — U4P for exchange operators and U4C for market participants — alongside U4S high-performance VMs for operators, participants, and service providers.

Machine Series	CPU Platform	CPU Cores	Memory	NICs	Storage Options
U4P (Bare Metal)	Intel 5th Gen Xeon (Emerald Rapids)	120 Physical Cores	512 GB or 768 GB	3 Physical (1 Standard + 2 ULL)	Local Titanium SSD (12 TiB), Hyperdisk
U4C (Bare Metal)	Intel 5th Gen Xeon (Emerald Rapids)	120 Physical Cores	512 GB or 768 GB	3 Physical (1 Standard + 2 ULL)	Local Titanium SSD (12 TiB), Hyperdisk
U4S (VM)	Intel 6th Gen Xeon (Granite Rapids)	2 to 288 vCPUs	Up to 2,232 GB	Multi-vNIC (Up to 200 Gbps)	Local Titanium SSD (18 TiB), Hyperdisk

Machine Series

CPU Platform

CPU Cores

Memory

NICs

Storage Options

U4P

(Bare Metal)

Intel 5th Gen Xeon (Emerald Rapids)

120 Physical Cores

512 GB or 768 GB

3 Physical (1 Standard + 2 ULL)

Local Titanium SSD (12 TiB), Hyperdisk

U4C

(Bare Metal)

Intel 5th Gen Xeon (Emerald Rapids)

120 Physical Cores

512 GB or

768 GB

3 Physical (1 Standard + 2 ULL)

Local Titanium SSD (12 TiB), Hyperdisk

U4S

(VM)

Intel 6th Gen Xeon (Granite Rapids)

2 to 288 vCPUs

Up to 2,232 GB

Multi-vNIC (Up to 200 Gbps)

Local Titanium SSD (18 TiB), Hyperdisk

We developed the U4 machine family to enable the world’s most technically demanding markets to run within Google Cloud and benefit from cloud services and scale. An example of this is our ongoing collaboration with CME Group, through which we are migrating listed derivatives markets to Google Cloud. Here, the U4C and U4P bare metal instances deliver direct physical co-location latency parity, providing predictable low-latency clock precision, and native hardware-multicast feed ingestion. This architecture demonstrates that core exchange systems and trading strategies can run in the cloud with the speed, consistency, and control that financial markets require.

"Migrating the world’s leading derivatives marketplace to the cloud requires uncompromising performance, latency, and reliability. Google Cloud’s Ultra Low Latency Solution and the U4 instance families represent a breakthrough for financial market infrastructure. By combining deterministic, ultra-low latency networking and dedicated bare metal compute with the benefits of the cloud, market participants can execute and ingest market data and execute with the speed of traditional co-location and the agility of the cloud." - Pearce Peck-Walden, Managing Director Markets Engineering, CME Group

Powered by Google Cloud-native infrastructure

The ULL Solution is available in select private Google Cloud regions, where trading teams can leverage core cloud infrastructure benefits such as rapid automated resource provisioning, on-demand capacity scaling, and dynamic fleet management. Trading teams can execute ultra-low latency trades on a dedicated, isolated ULL network, and offload data streams over an independent network interface to Google Cloud services such as BigQuery and Gemini Enterprise.

For exchange participants managing tick-to-trade workflows or processing real-time market data, microsecond variations are critical. As bare metal instances, U4P and U4C machine series don’t add any virtualization overhead for latency-sensitive execution tiers. Leveraging Google Cloud's Titanium offload system architecture, these bare metal instances provide direct access to the host resources. Compute Engine instances enable users to provision dedicated hardware through standard Google Cloud APIs, orchestrate deployments with infrastructure as code, apply Cloud Next Generation Firewall (NGFW) policies, and collect rich telemetry.

The U4S VM series complements these bare metal offerings for applications such as pre-trade risk validation and real-time analytics. Deployed alongside U4P and U4C instances, these VMs shorten transit hops across the trading architecture while offering elastic scaling from 2 to 288 vCPUs and networking up to 100 Gbps.

“Our testing of the U4 instances validated that Google Cloud delivers the low latency and determinism required for our demanding exchange trading workloads. Having both bare metal and VM options allows us to evaluate the value of that performance relative to its cost, helping us align the right infrastructure to each of our trading strategies.” - Corbin Kidd, EVP/CTO, Prime Trading

"In our tick-to-trade performance testing, the U4C bare metal instance demonstrated exceptional speed, showing that cloud compute and ultra-low latency networking can comfortably satisfy microsecond trading demands. But, what sets the U4 series apart isn't just raw processing speed, it’s structural determinism. When we stress-tested Google Cloud's U4 bare metal family, the performance curve remained virtually flat, which confirms that these cloud compute layouts can now offer the strict determinism institutional market participants require." - Christopher Wilson, Global Head of Experience Modernization, 28Stone

Looking to the future, this low-latency architecture serves as the foundation for algorithmic trading, allowing firms to feed real-time markets directly into AI-driven trading models.

Storage and placement for trading workloads

The U4 family supports Google Cloud storage options and placement policies, helping trading teams record trades at speed and minimize network latency between machines:

Local Titanium SSD for fast logging: For real-time trade logging and data caching without slowing down execution, U4P and U4C provide up to 12 TiB of direct-attached NVMe SSDs, and U4S VMs scale up to 18 TiB.
Hyperdisk for data archives: For historical market data, backtesting models, and regulatory archives, you can attach Hyperdisk Balanced and Hyperdisk Extreme volumes across all U4 series, scaling up to 32 volumes and 512 TiB per instance.
Placement policies: To reduce physical network hops, compact placement policies keep related machines close together in the data center. Alternatively, spread placement policies place machines on separate racks to improve resiliency.

Getting started with the ULL Solution

With the ULL Solution, financial exchange operators, exchange participants, and trading service providers can execute trades with the speed and reliability they need today while positioning their businesses to leverage continuously evolving AI and infrastructure capabilities for the trading of tomorrow. The ULL Solution is available in select Google Cloud regions. To learn more, check out the public documentation, and reach out to our team to request access.

Enabling Cloud Storage end-to-end checksums for improved data integrity and durability

Thu, 01 Oct 2026 16:00:00 +0000

At Google Cloud, we know that you count on us to maintain the durability and integrity of your data at all times, both at rest and in transit. And now we’re making it easier for developers to take advantage of native data integrity features in Cloud Storage, by enabling end-to-end checksumming by default in all the Cloud Storage SDKs.

Like in any disk-based storage system, bits can flip anywhere in their journey, from the application all the way down to the disk. Cloud Storage has always let clients provide a checksum of the object data being uploaded, and receive a checksum of the data being downloaded. Also since its inception, Cloud Storage stores a checksum for every object in its metadata, regardless of how the object was uploaded into Cloud Storage.

But until recently, ensuring end-to-end data integrity required extra work on the part of developers to calculate and provide checksums to Cloud Storage. Cloud Storage always calculates the crc32 (32-bit cyclic redundancy check) of data it receives and ensures data stored on disk matches this checksum. When a client request includes the object’s checksum, Cloud Storage ensures that this checksum also matches. However, when an upload request doesn’t include a checksum, that upload is vulnerable to a bit flip while the data is in-flight, prior to the server-side checksum computation. Not all customers and clients enable client-side checksums by default, leaving data in this phase unprotected.

To address this gap, the latest version of all Cloud Storage SDKs now internally checksums data being uploaded and passes this checksum to Cloud Storage, if it’s not provided by the application. The SDKs also support verifying the object’s checksum when an object is being downloaded.

Finally, there are many use-cases where applications download select ranges of objects instead of the full object. When using Cloud Storage SDKs with our gRPC API to perform a range read, the SDKs take advantage of gRPC’s built-in end-to-end range checksum, using it to verify the data it receives.

We highly recommend updating to our latest SDK versions to take advantage of these important integrity features.

And now, let’s peek under the hood

Ensuring continuous “chain-of-custody” between the data and its associated checksum from your application down to the disk platter, with no gap where a bit flip could go unnoticed, is quite challenging. And it’s critical to get this right: at our current scale of hundreds of thousands of Cloud Storage frontends, bit flips aren’t theoretical and do happen from time to time.

For instance, consider this simple example: when Cloud Storage receives your data in its frontend, this data gets encrypted with per-object encryption keys. This involves a data copy: the plaintext data is passed through an encryptor into a new memory buffer containing ciphertext. Extremely rarely, a bit in the source or destination memory buffer flips during this process. However, at our scale, extremely rare things happen routinely.

In this situation, we maintain chain-of-custody by reversing the whole process: after encrypting the data (1), we calculate a checksum that protects the ciphertext. Then we decrypt the ciphertext (2), and if the resulting plaintext doesn’t match the original (3), we throw everything away and start over. This adds up to a lot of extra CPU time spent on encryption and checksumming, but it’s a necessary step to ensure data integrity.

Another challenge is how data gets broken up and aggregated as it passes through layers of our stack. As data gets uploaded to Cloud Storage, it gets split up into chunks, each of which has its own checksum. To manage data efficiently at scale, Cloud Storage groups thousands of chunks together into a storage unit we call a shard file. These gigabyte-sized files are how Cloud Storage ultimately delivers data to our cluster-level storage system, Colossus. Internally, Colossus uses Reed Solomon encodings to spread data across many disks and protect against the failures of individual disks, machines, and racks. This requires chopping up the shard file data into blocks, each of which is again protected by a checksum.

To maintain chain-of-custody of the data as it goes through all these transformations, we take advantage of some nifty properties of cyclic redundancy checks (CRCs), for example, concatenation. When you have two data buffers that each have their own CRC, you can cheaply compute the CRC of the two concatenated buffers without having to re-checksum the data. This comes in handy in many situations, such as when concatenating chunks together into shard files: Colossus can cheaply determine the CRC of the entire shard file from its constituent chunks and store that in its metadata.

Ultimately, the data lands on disks managed by our “D” file server (our network attached disks). D stores inline checksums for each range of data within a Colossus block. Whenever data is read from the disk, it is verified at several layers: The Colossus client verifies the data it reads against D’s inline checksums, and the Cloud Storage frontend reads data chunk-by-chunk, verifying each chunk against its checksum before sending it to the client. These chunk-level checksums are what enable our gRPC protocol to provide a checksum for a range read that can be verified by our SDKs, all without losing chain-of-custody.

Chain of Custody: Maintaining Data integrity across Data transformations

The above image shows the data integrity handoff across multiple layers under the hood of Google Cloud Storage. On reads, checksums are verified inline at multiple layers to prevent silent corruptions.

Client passes full object checksum to Cloud Storage Frontends.
Data is split into chunks and individual chunk level checksums are computed.
Shard level checksums are computed based on concatenated chunk level CRCs.
Shards are stored across disk blocks with another level of block level inline checksums.

Here on the Cloud Storage team, we remain dedicated to maintaining the highest standards of data integrity for our customers. By making end-to-end checksumming the default in our SDKs and maintaining chain-of-custody throughout our internal storage stack, data remains exactly as intended from the moment of upload to the final download. This continuous vigilance reflects our commitment to protecting your data at any scale. To take full advantage of these protections, we recommend updating to the latest version of our SDKs.

What’s new in AI infrastructure and orchestration in September

Wed, 30 Sep 2026 16:00:00 +0000

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.

Storage-optimized Z4D machine family, now GA, is designed for IO-intensive workloads

Mon, 28 Sep 2026 07:00:00 +0000

Today, we’re excited to announce the general availability of our next-generation Storage-optimized Z4D machine series in Google Compute Engine with both Virtual Machine (VM) and bare-metal instances.

We built Z4D for IO-intensive and business-critical workloads that require large local storage capacity and high storage performance, including SQL, NoSQL, KVrocks and vector databases, data analytics and data search. Powered by 5th Gen AMD EPYC processors (Turin) and paired with the latest enhancements in Titanium, Z4D provides up to 84,000 GiB of Local SSD (LSSD) storage. It improves the performance of these demanding workloads by up to 40% compared to the prior-generation Z3 instances, so you can increase your applications throughput while right-sizing your cloud investment. Z4D’s large LSSD capacity also makes it a strong storage solution for AI/ML training and inference workloads and running distributed parallel file systems at scale.

Z4D VMs and bare-metal instances

The Z4D VM portfolio lets you rightsize your infrastructure and scale your clusters to meet workloads requirements by providing large total local SSD capacity and high local SSD capacity per vCPU. Z4D offers two different VM types: the Z4D-highmem-standardlssd VM type, which includes seven VM shapes and offers 219 GiB of LSSD per vCPU. These VMs are optimized for data analytics (OLAP), and SQL databases like MySQL and Postgres. The Z4D-highmem-highlssd VM type includes seven different VM shapes, with 438 GiB of LSSD per vCPU and is optimized for distributed databases, data streaming, large parallel file systems and data search. In addition, you can easily scale your existing Z3-based workloads by expanding into Z4D clusters.

Z4D bare metal instances give you direct access to the physical hardware without a virtualization layer. This reduces latency for latency-sensitive workloads, custom hypervisors and workloads with specific licensing needs. Further, Z4D bare metal will be the very first AMD-based instance to support Nutanix Cloud Clusters (NC2), a hybrid multi-cloud platform that works across several cloud providers. Z4D bare-metal instances deliver the LSSD capacity and low latency that agentic AI architectures using microVMs require, allowing developers to run thousands of isolated sandboxes per host with native performance and efficiency.

In addition to 84,000 GiB of local SSD storage, Z4D VMs and bare metal instances offer up to 384 vCPUs and up to 3,072 GiB of memory. Z4D instances are based on Titanium SSDs, which offload local storage processing from CPU resources to deliver real-time data processing, low-latency, high-throughput storage performance and enhanced storage security. Z4D delivers up to 15,600K random read IOPS and up to 75,600 MiB/s sequential read throughput, improving the LSSD storage performance by up to 70% compared to Z3. Z4D also reduces write latency by up to 25% and improves mixed read-write IOPS by up to 30% without increasing the IO latency vs Z3. At the same time, Z4D instances provide the connectivity and storage performance that enterprise and AI/ML workloads need by doubling the networking throughput compared to Z3 and offering up to 400 Gbps of standard networking bandwidth.

What customers and partners are saying

“Elastic is committed to delivering best-in-class performance with the Elasticsearch Platform that powers observability, security, search, and AI solutions. Performance and cost efficiency are critical for teams running these workloads at scale and our initial testing of the new Z4D virtual machines shows up to 50% better indexing throughput compared to previous generation Z3 VMs. We look forward to bringing these benefits to Elasticsearch users deploying on Google Cloud.” - Yuvraj Gupta, Principal Product Manager, Elastic

"Migrating to the Z4D instance reduced our processing runtime by ~70% while reducing overall costs. This improvement enables Immunai to process large-scale immune data significantly faster and turn it into biological insights that support pharma companies in making better-informed decisions throughout drug discovery and development." - Guy Yachdav, Senior Director of Software Engineering, immumeai

"We are thrilled to expand our technical collaboration with Google Cloud and to deepen our strategic partnership with AMD to bring Nutanix Cloud Clusters (NC2) to the new Z4D bare metal instances. This marks a significant milestone for our customers, as NC2 on Z4D will be a first-of-its-kind offering — the very first AMD metal instance on which NC2 is supported. By uniting AMD's cutting-edge compute performance with Google Cloud's robust infrastructure and the Nutanix hybrid cloud platform, we are delivering unprecedented flexibility, scale, and choice to empower enterprise workloads." - Saveen Pakala, Vice President of Product Management, Nutanix

"Redis powers real-time data infrastructure and AI workloads for thousands of organizations worldwide. Flex extends that to larger datasets without all-RAM economics — local SSD handles the scale while memory provides the speed. We compared Google Cloud's new Z4D storage-optimized VMs to our current generation C3D instances using our Flex benchmark suite on the same flash-heavy workloads. The results were impressive: up to 4.3 times higher throughput and up to 77% lower latency on our smaller shapes. Z4D also sustained nearly 1.8 million operations per second at full RAM hit ratio. The more data we served from flash, the more that advantage grew — which is exactly the profile Flex is built for. Z4D gives us a clear path to deliver the same performance tier on a smaller footprint, and we're looking forward to expanding our testing as Z4D moves toward GA.” - Benjamin Renaud, CTO, Redis

“Shopify looks at what's best for our fleet, and Z4D gives us high CPU density alongside fast locally attached storage and fast networking. Shopify's user experience depends on how fast data can be served, and AI shopping agents query storefronts far more aggressively than people do. These shapes give us the throughput to keep up. We saw roughly 20% better throughput on Z4D than on Z3.” - Brad Dietrich, Distinguished Engineer, Shopify

"Google's Z4D VMs are a significant leap forward. In our testing we observed up to 60% better performance than previous Gen 2 VMs, and up to 36% better performance than Z3 VMs. With high local SSD density and strong cost-efficiency together with improved system reliability, Z4D gives Silk customers faster, more predictable performance for their most demanding workloads." - Adik Sokolovski, Chief R&D Officer, Silk

"Connecting AI with petabytes of fresh data means we're searching more than ever before. We're excited about the new Z4D instance types. Compared to Z3, they gave us a 40% throughput improvement on real query and indexing workloads — which directly translates to faster and cheaper web-scale search for our customers" - Ben Linsay, Engineer, turbopuffer

Enhanced maintenance experience

Both Z4D VMs and bare-metal instances make it easier for you to plan ahead and schedule maintenance operations at a time of your choosing by providing notice from the system several days in advance of a required maintenance. Z4D VMs further enhance the maintenance experience by allowing you to live-migrate an instance during maintenance events for VMs with 42,000 GiB or less of local SSD storage. Z4D VMs with 84,000 GiB of local SSD and Z4D bare metal instances are terminated and restarted while preserving your data through the planned maintenance events.

Support for Hyperdisk

Z4D VMs and bare metal support Hyperdisk, Google Cloud’s workload-optimized block storage that lets you optimize the performance for each workload by independently tuning the storage performance and capacity for each instance.

Specifically, they are compatible with Hyperdisk Balanced, Hyperdisk Throughput, and Extreme Hyperdisk storage for scalable, high-performance network-attached storage, supporting up to 512 TiB of capacity per instance. For general-purpose workloads, Hyperdisk Balanced, with up to 320K IOPS per instance, offers a mix of performance and cost-efficiency. Hyperdisk Extreme delivers ultra-low latency and supports up to 500K IOPS and 12,500 MiB/s throughput per Z4D VM and bare metal instance, making it well-suited for demanding database workloads.

Get started with Z4D today

Z4D VMs are available today in select regions worldwide and Z4D bare metal instances are in preview - reach out to your account team for additional information and access. To start using Z4D instances, select Z4D under the Storage-Optimized machine family when creating a new VM or GKE node pool in the Google Cloud console. Learn more at the Z4D machine series page. Contact your Google Cloud sales representative for more information on regional availability.

Storage Intelligence advisor: Know what changed in your storage estate and act on it

Fri, 25 Sep 2026 16:00:00 +0000

The volume of data being generated today brings both opportunity and massive operational complexity. Most teams that operate at scale don't discover issues until they appear on an invoice — and by the time an unusual access pattern shows up as a line item, it has often been running for weeks. Understanding what happened means exporting inventory, joining it against access logs, and hoping someone still remembers which service account belongs to which job.

That workflow was manageable in the past, but today’s AI training and inference pipelines create data faster than governance systems can classify it, and read data in patterns that shift from week to week.

Today we're announcing two new features for Google Cloud Storage: the general availability of Storage Intelligence advisor along with expanded capabilities in storage batch operations. Advisor tells you what changed in your storage estate and what to do about it. Batch operations can carry that decision out across millions of objects. These features are available now to all Storage Intelligence customers.

Storage Intelligence advisor in cloud console.

Storage Intelligence advisor makes reporting easy

For the last decade, answering "what’s in my buckets?" has been a data engineering project. Export your inventory, load it somewhere queryable, join it against usage, build dashboards, and then maintain them. Storage Intelligence delivers visibility without the engineering overhead. Teams are voting with their workloads: the number of customers using Storage Intelligence to analyze datasets of over 1 billion objects has more than doubled this year.

There are two ways to run a large storage estate. Teams can leverage daily activity data and metadata snapshots to build exactly the pipelines they need –Storage Intelligence still gives you that option – but most teams would prefer not to build pipelines if they don’t have to. They want to be told what changed in their storage environment and what to do about it. Storage Intelligence advisor is for them.

What Advisor gives you on day one

Storage Intelligence advisor brings visibility into your storage without having to perform any setup. Advisor starts from a curated set of findings. There's no schema to design, no pipeline to manage, and no dashboard to assemble. Enable Storage Intelligence on an organization, folder, or project, and charts and findings appear for the buckets in that scope.

Shipt can now more quickly detect anomalies with Storage Intelligence advisor:

"Before Storage Intelligence advisor, tracking critical usage metrics and catching anomalies [in Google Cloud Storage] required heavy engineering and complex data pipelines. Now, with native, out-of-the-box dashboards, we can instantly identify usage spikes and drill down into the details. Having the visibility to immediately remediate unintended usage — without any configuration — has turned what used to be a major effort into a simple, self-service task." - Charley King, DataOps-DevOps Engineer, Shipt (a subsidiary of Target.com)

Once it’s installed, Advisor immediately starts analyzing the Cloud Storage estate, scanning for anomalies and optimization opportunities including:

A spike in Class A or B operations against Coldline or Archive data. Cold storage is cheap to use but expensive to access.
A spike in 429 errors. Where a request pattern is outrunning limits, timeouts follow.
A spike in cross-region egress.
Total consumption rising above a long-term trend.

Each finding is baselined from your project's own activity and metadata and works from daily snapshots of your storage usage, so a spike on one day is surfaced within 24 hours, not a line item you discover at the end of the month. In the last 30 days, over 6,000 findings have been generated across hundreds of customers.

Take a runaway analytics job that issues millions of daily reads against Archive storage. Without Storage Intelligence advisor, this surfaces as a retrieval-fee weeks later on a bill.

Advisor identifies the anomaly against your project’s baseline, attributes it to the responsible bucket, prefix, and service account, and points at the controls that apply: bulk-transition the affected objects to Cloud Storage Standard to stop retrieval charges, enable Autoclass so tiering follows real access patterns, or tighten access with Managed Folders so the job can’t reach data it was never meant to access.

Act on findings with storage batch operations

Most storage recommendations go unactioned because carrying them out is a lot of work. Updating retention policies or storage classes across billions of objects means handling throttling, partial failures, and retries. Storage batch operations removes that work. Execution is fully managed and serverless, with progress tracking and automatic retries built in, so a recommendation becomes a policy-driven job rather than a project.

Palo Alto Networks had this to say about batch operations:

"Object retention locks were essential for our security guardrails, but managing them across billions of objects was once a non-starter. Storage Intelligence changed that. Today, our team uses storage batch operations to seamlessly update retention policies on demand across our entire fleet." - Kurtis Nusbaum, Senior Principal Software Engineer, Palo Alto Networks

Because Storage Intelligence advisor and batch operations are part of the same Storage Intelligence subscription so customers can now quickly identify issues with Advisor and easily remediate those issues with batch operations.

Batch operations enables the following:

Remediating operational spikes: Bulk-transition high-traffic Archive or Coldline objects to Standard as soon as the pattern is detected, curbing retrieval and operation charges immediately.
Containing runaway growth. Mass-delete stale or temporary data across specific prefixes when the advisor flags above-trend storage growth.
Enforcing fleet-wide consistency. Apply metadata, tagging, retention, or encryption changes uniformly across massive object sets, with no dedicated compute to provision.

We also expanded and enhanced the existing capabilities of batch operations, making it easier to execute actions at scale:

Multi-bucket processing: Run a single job across up to a thousand buckets per project, rather than executing it bucket-by-bucket.
Dry-run validation. Simulate your transformations using dry-run mode before modifying live data. A dry run helps you safely preview a job's impact (including affected object counts, total size, and potential errors) before committing to permanent changes.
Advanced filters powered by Storage Insights datasets: Use Common Expression Language (CEL) expressions to select objects directly by specifying conditions that match fields in Insights datasets. For example, you can filter objects across your buckets by storage class, object size, creation date, or custom attributes.

Below is a CLI example demonstrating how to create a batch operations job using advanced filters. This job deletes all temporary objects belonging to the Standard storage class present in a user's "analytics" buckets.

code_block: <ListValue: [StructValue([('code', 'gcloud storage batch-operations jobs create bulk-delete-temp-objects \\\r\n --description="Bulk delete temporary objects in analytics buckets" \\\r\n --target-project="my-project-id" \\\r\n--insights-dataset-config="projects/my-project-id/locations/us-central1/datasetConfigs/my-dataset" \\\r\n --bucket-filters="name.startsWith(\'analytics-\')" \\\r\n --object-filters="storageClass == \'STANDARD\' && name.endsWith(\'.temp\')" \\\r\n --delete-object'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7fb520fbdf90>)])]>

The evolution of storage management

Storage management shouldn’t be a reactive effort reserved for quarterly reviews and post-incident fire drills. It should be continuous, proactive, and contextual.

Storage Intelligence advisor and batch operations help to surface what changed and enable insights and action at scale. As Storage Intelligence gets better at recognizing which findings matter, Cloud Storage can carry more of the operating load for teams that need to manage storage at scale.

Storage Intelligence advisor and enhanced storage batch operations are generally available today.

To get started, enable Storage Intelligence on a project or org. If you haven't used Storage Intelligence before, a 30-day trial is available at no cost.

Introducing Filestore agent volumes: fully managed storage for agent workspaces

Tue, 15 Sep 2026 16:00:00 +0000

From running build tools, to data analysis pipelines, to collaborative research, executing data-driven tasks is essential for any enterprise agent.

Today, platform teams often stitch together custom workarounds to address agent storage requirements, which could include shuttling state back and forth between agent sandboxes and centralized storage or manually managing local disks and/or self-hosted file systems. However, as agent fleets scale, these approaches force difficult trade-offs between cold-start latency, operational complexity, and the cost of idle, pre-allocated storage.

As organizations scale agent sandboxes to thousands or even millions of concurrent sessions, storage must evolve to overcome these trade-offs and meet the needs of these dynamic workloads, which require strict workspace isolation, instant session resumption, elastic pay-per-use economics, and fluid multi-agent collaboration.

To meet these emerging demands, we’re expanding our AI storage portfolio and announcing availability of Filestore agent volumes, a new, fully managed capability purpose-built to deliver high-performance, elastic file storage for scaling agentic workloads on Google Cloud.

Purpose-built storage for AI agent workspaces

Autonomous agents require isolated runtime environments to safely execute dynamic code, install third-party packages, and run tools without putting host infrastructure or tenant data at risk. While Agent Substrate on GKE and GKE Agent Sandbox provide the dedicated compute environments needed to run high-density agent fleets, those sandboxes also need dedicated persistent workspaces to operate on.

Filestore agent volumes within Google Cloud Filestore, give you purpose-built agentic storage to complement your agentic compute via a dynamic provisioning architecture designed specifically for the scale and elasticity of AI agent fleets. Co-designed with Agent Substrate to support agentic fleets at scale, Filestore agent volumes provide GKE sandboxes with instantaneous access to isolated, persistent file storage. When configured to leverage Filestore, GKE storage management happens behind the scenes: Every time GKE launches a sandbox for a new agent task, Filestore automatically allocates and attaches a dedicated, isolated file workspace to that environment in milliseconds. Platform teams don't need to manually create, attach, or tear down storage volumes for individual agent runs; instead, the system handles the entire volume lifecycle automatically as your agent fleet scales up and down. The result is an efficient, end-to-end infrastructure solution for cost-effective agent management that provides:

Granular isolation and enterprise guardrails: Agent platforms face security and data leakage risks when running untrusted, autonomous code. Filestore agent volumes enforce strict boundary controls and granular access permissions per workspace, ensuring agents operate exclusively within their designated directories and keeping dynamic toolchains strictly isolated across tenants.
Sub-second session resumption: Traditional storage provisioning approaches can introduce cold-start latency that stalls interactive agent sessions. Agent volumes attach and detach in milliseconds, making it possible for orchestrators to aggressively suspend idle sandboxes to save compute costs, and resume instantly when new tasks or user inputs arrive.
Smart lifecycle economics and pay-per-use pricing: Pre-allocating fixed-size, high-performance storage for thousands of short-lived or intermittent agent tasks can create massive storage waste. With agent volumes, platforms pay only for the storage capacity consumed and benefit from automatic lifecycle tiering. This means you get high performance without wasted spend: When your agents aren’t actively reading/modifying code or analyzing datasets, you can automatically shift idle workspace state to lower-cost storage.
Multi-agent collaboration: Coordinating multi-agent swarms can result in brittle data-passing pipelines and risk of file collisions. Built with native Read-Write-Many (RWX) support and POSIX file locking, agent volumes allow orchestrators to attach a single shared workspace across multiple agents. Collaborating agents can safely co-author, test, and review project files concurrently with file-level consistency and protection against write conflicts.

Powering next-generation agentic workloads

By providing an elastic, high-performance, and isolated file tier, Filestore agent volumes unlock a wide spectrum of agentic workloads and use cases in production:

Software engineering and coding sandboxes: Agentic coding platforms can spin up thousands of isolated workspaces where agents safely install libraries, write multi-file patches, run build tools, and execute unit tests, all leveraging standard POSIX file semantics with no need for storage-specific customization.
Collaborative multi-agent swarms: Complex workflows, such as a lead orchestrator delegating tasks to dedicated research, code generation, and validation sub-agents, can directly share a unified file tree. RWX support allows agents to co-author and review project files concurrently without write conflicts.
Interactive long-horizon workflows: For user-in-the-loop applications (such as agents that require asynchronous user approval or run multi-hour data analysis pipelines), platforms can suspend idle agent sandboxes to minimize compute waste, then resume execution on demand with sub-second responsiveness.

Get started today

If you are building an Agent-as-a-Service platform, scaling coding assistants, or deploying enterprise agent fleets, your storage tier should accelerate your innovation — not hinder it.

Filestore agent volumes are now available to all Google Cloud customers for non-production workloads. GA support for production workloads is available via allowlist. This new offering features out-of-the-box integrations with Agent Substrate on GKE and GKE Agent Sandbox to help you build responsive, scalable, and cost-efficient agent platforms today.

To request access to Filestore agent volumes, submit this form and visit the Filestore documentation and GKE documentation to learn more.

Unlocking the future of shared storage: Filestore on Colossus

Wed, 05 Aug 2026 13:00:00 +0000

Today, enterprise storage must be as agile, elastic, and responsive as the workloads it supports. Filestore, Google Cloud’s first-party, secure, scalable NFS file service, can service a wide-range of enterprise use cases as well as cutting-edge AI and agentic workflows. Today, we’re sharing a major enhancement: Filestore now incorporates a cloud-native backend storage layer built directly on Colossus, Google’s foundational distributed storage system. Leveraging Colossus, Filestore can deliver even greater flexibility and scalability to support the most demanding modern workloads.

Colossus: A foundation of global scale

As a first-party service, Filestore is positioned to leverage the best of Google’s infrastructure-level innovation. By leveraging Colossus, Filestore now utilizes the same infrastructure DNA that powers Google’s biggest global services, including YouTube, Gmail, and Gemini. This platform-level upgrade enables superior scalability and operational efficiency compared to legacy VM-based architectures, providing a robust foundation for your data.

This also allows Filestore to decouple storage capacity and performance. Now you can provision IOPS independently to precisely meet workload demands without having to over-provision capacity, supporting a broad range of workloads — from small developer environments to massive datasets.

This decoupled scale is especially powerful when applied to containerized environments. Through the Filestore CSI driver, Filestore delivers persistent, high-performance storage for GKE workloads with enterprise-grade reliability and availability. To further optimize resource utilization at scale, Filestore multishares for GKE lets you segment a single Filestore instance into many smaller shares, starting at just 10 GiB. This means AI teams can scale out their GKE clusters efficiently, carving up large high-performance instances into smaller, project-specific shares to support thousands of concurrent containers, without sacrificing performance or increasing TCO.

Shared storage for agentic swarms

The combination of decoupled performance and deep GKE integration enables a critical new use case: high-concurrency workspaces for AI agent swarms, where multiple specialized, autonomous AI agents collaborate in parallel to accomplish complex goals. In an agentic workflow, multiple agents often need to read from and write to a common dataset simultaneously to maintain state and share context. Filestore file shares act as these common workspaces leveraging the NFS protocol to provide consistent file system access across the swarm.

Backed by Colossus, Filestore can support millions of agents, working independently or securely collaborating with shared data. By utilizing NFS file locking, Filestore helps ensure strict data consistency, preventing conflicts even as swarms grow in size and complexity. This allows AI agents to maintain a unified view of their environment, enabling more complex reasoning and collaboration.

Operational agility and enterprise-grade control

Beyond performance, this enhancement improves operational agility. Now, you can independently scale IOPS via Custom Performance settings, optimizing costs in real-time without having to rebuild clusters. Furthermore, the distributed storage layer provides faster failure recovery and zero-downtime capacity changes compared to traditional monolithic storage architectures. Filestore is also providing security at scale via deep integration with Google Cloud IAM, NFS User IDs and Group IDs (UIDs/GIDs), and IP Access Control Lists (ACLs).

Building for the next era of data

This Filestore update reflects our commitment to meeting the changing needs of AI-era workflows and lays the foundation for future enhancements. By providing a first-party, deeply integrated service that evolves with your needs, we are helping you maximize your business potential flexibly and cost-efficiently. Get started today by visiting the Filestore page.

C4N, now GA: Delivering cloud’s highest per vCPU network and block storage I/O for x86 workloads

Wed, 08 Jul 2026 20:00:00 +0000

September 15, 2026: C4N Instances with Titanium Local SSD (c4n-*-lssd) are now available in preview.

As organizations scale modern workloads — from high-throughput databases and network/security appliances to real-time analytics and AI/ML inference — network and block storage performance can quickly become a bottleneck. Standard virtual machines often struggle to balance compute efficiency with the high-volume data-transfer demands of these applications.

At Google Cloud Next ‘26, we announced C4N in preview, our first network- and block-storage-optimized Google Compute Engine instance that’s purpose-built to eliminate I/O bottlenecks for demanding enterprise applications, and today, it is generally available. Built on Google's custom-designed Titanium offload architecture, C4N instances offload network and storage tasks to dedicated hardware to unlock incredible performance and compute efficiency. C4N offers up to 400 Gbps of network bandwidth and a market-leading 95 million packets per second (MPPS) — nearly 33% higher network bandwidth per vCPU and 224% faster packet processing performance than comparable Intel-based offerings at other hyperscalers. This performance makes C4N a great fit for network-intensive applications such as virtual appliances (e.g., next-gen firewalls, virtual routers, load balancers, DDoS mitigation), large-scale data analytics, telco applications (5G UPF), distributed compute and CPU-based AI/ML workloads.

Paired with Hyperdisk Extreme, our high-performance block storage, C4N also delivers Compute Engine’s highest block storage performance, scaling up to 25 GiB/s of storage bandwidth and 1M IOPS — nearly 33% higher storage bandwidth and 39% more IOPS per vCPU versus comparable Intel-based offerings, making them a strong choice for large-scale databases, high-performance file systems, in-memory databases, and other workloads that benefit from high block storage performance.

Engineered specifically to deliver predictable, high-throughput I/O performance for networking, packets-per-second-bound and storage-optimized applications, C4N allows customers to scale network, storage, and compute resources more precisely to meet specific workload requirements, delivering significant TCO benefits by eliminating the need to over-provision resources just to meet I/O demands.

C4N is powered by 5th Gen Intel® Xeon® Scalable processors (code-named Emerald Rapids).

“Google Cloud’s introduction of C4N highlights how infrastructure innovation and a strong silicon foundation can help customers address increasingly data-intensive workloads. With Intel Xeon and Custom Infrastructure Processing Unit (IPU), C4N delivers the performance and efficiency needed for demanding network optimized environments.” – Srini Krishna, Intel Fellow, Data Center products, Intel

What’s new: Scaling massive data layers with C4N

Our network-optimized C4N instances are designed to deliver predictable, high-performance I/O at scale. By providing consistent bandwidth, packet-processing performance (PPS), and IOPS scaling across all VM shapes and sizes, C4N helps ensure your most demanding data workloads run reliably. To achieve this, we have built deep resiliency into every layer of our infrastructure — from the host and fabric layers to redundant top-of-rack (ToR) switches — delivering continuous performance for your applications.

Compared to general-purpose C4 VMs, the network-optimized C4N delivers significant performance gains across both network and block storage vectors.

Next-generation network performance

Superior VM-to-VM network bandwidth: Achieves up to 400 Gbps of VM-to-VM network bandwidth (an almost 4x increase in BW-per-vCPU over standard C4) and supports up to 50 Gbps single-flow bandwidth between C4N instances routed within the same VPC network. This provides non-blocking data delivery for high-throughput single-stream and multi-stream applications.
Enhanced VM-to-internet performance: Benefits from an 8x increase in internet egress network bandwidth, reaching up to 200 Gbps. It also features a nearly 32x increase in internet egress packet processing performance, scaling up to 48 MPPS.
Optimized I/O for smaller shapes: Keeps your cloud bill lean by delivering up to 25–50 Gbps of network bandwidth specifically for 2–16 vCPU shapes, great for accelerating I/O-bound tasks without needing to over-provision compute. Furthermore, these smaller shapes introduce predictable, steady-state baseline bandwidth limits to provide consistent performance at a lower cost.
Enhanced out-of-the-box networking: gVNIC interfaces on C4N now start with more Tx/Rx queues by default, scaling with vCPUs up to a maximum of 64 (in comparison to 16 queues on C4/C4D).
Shorter Google Cloud Storage transfer times: C4N VMs now offer up to a 2x increase in bandwidth to retrieve and store large volumes of data from Cloud Storage, boosting performance for analytics, AI/ML, and backup workloads.

Better yet, this performance is available out of the box, with no add-ons. Designed for high performance from the get-go, C4N offers maximum performance without needing to purchase or configure premium add-ons like Tier_1 networking.

Dynamic storage performance with Hyperdisk

The C4N instance family, when combined with Hyperdisk, can help dynamically tune storage performance, latency, and throughput independently of your compute instance sizing to deliver high block storage performance for your applications. C4N supports the complete Hyperdisk portfolio, including Hyperdisk Balanced, Balanced High Availability, Extreme, Throughput, and ML block storage options.

Hyperdisk Extreme: C4N with Hyperdisk Extreme provides low-latency, high-speed data access for modern databases and enterprise AI applications, with up to 25 GiB/s of block storage throughput and nearly 1M IOPS, a 2x increase in storage performance over C4. Also, exclusive to network optimized machine series such as C4N, we now offer Hyperdisk Extreme across all machine sizes — even down to the smallest 2 vCPU sizes.
Hyperdisk Balanced: Delivering the highest throughput and IOPS for general-purpose block storage in the Compute Engine portfolio, Hyperdisk Balanced on C4N scales up to 20 GiB/s of block storage throughput and nearly 640K IOPS. This makes it a highly cost-effective option for running storage-intensive applications at scale.

Together, C4N’s network and storage optimizations combine for tremendous impact in real-world applications:

Web serving: Up to 1.5x additional Nginx requests per second compared to C4 for typical web request sizes (100–300Kb), significantly boosting capacity for network-bound web applications
Databases: Up to 45% better queries per second (QPS) for MySQL when data resides primarily on disk than equivalent C4 VMs

What customers are saying

Industry leaders are already proving that workload-optimized infrastructure is the engine for transformation. Here is how our customers are leveraging the network-optimized power of C4N:

“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 compute family, we’ve found the ideal engine for Ericsson On-Demand. The C4N’s architectural focus on network-optimized compute allows our 5G Core-as-a-Service to reach unprecedented throughput levels — like our recent 1 Tbps milestone — while maintaining the carrier-grade reliability our customers expect. It’s no longer just about cloud-native; with C4N, we are delivering network-native performance in a public cloud environment.” - Eric Parsons, VP, Head of Ericsson On-Demand, Ericsson

“Teradata's Autonomous Knowledge Platform unifies production-grade AI, analytics, and data into a single integrated system — providing the context, governance, and performance backbone autonomous AI demands at scale. Customers rely on Teradata to run mission-critical, highly I/O-intensive workloads where performance and cost control directly determine value.

Google Cloud C4N instances are well suited for these demanding workloads, delivering strong price-performance and supporting more efficient, optimized deployments. By leveraging C4N on Google Cloud, Teradata Cloud can help customers accelerate from insight to action — scaling enterprise intelligence with confidence and driving greater impact from their data and AI investments” - Kevin Dougherty, Senior Director of Product Management, Core Platform, Teradata

“With the next-generation network and storage bandwidth of C4N VMs, Google Cloud NetApp Volumes will unlock new levels of performance to support our customers’ most demanding AI workloads. By collaborating to extend Google Cloud NetApp Volumes support for the C4N VM family, Google and NetApp are deepening our partnership to address real customer challenges. Together, we’re delivering data-in-place AI and analytics solutions that simplify architectures, maximize performance, and turn data into impact.” - Pravjit Tiwana, Senior Vice President and General Manager of Cloud Storage and Services, NetApp

"Most Compute Engine instances ship with a single high-speed network interface. The new C4N doubles the bandwidth potential with two 200 GbE interfaces. That architectural shift is significant. It means we can dedicate both networks entirely to storage traffic, doubling the available bandwidth for data-intensive workloads, and achieving 2x storage performance over the previous generation. The C4N was announced just weeks ago and is already active in Sycomp's test environment, ensuring our customers can evaluate the latest GCP capabilities without delay. Google Cloud’s published maximum hyperdisk balanced performance for the C4N is 20 GiB/s. In our tests, with three storage servers Sycomp achieved 58.5 GiB/s on read and 58.6 GiB/s on write, with ten C4N storage servers we achieved 195 GiB/s read and write — 97% of the theoretical ceiling with zero platform-specific tuning. That's a strong starting point, and there's measurable room to close the remaining gap through configuration work we can finetune. The C4N isn't just faster — it changes the price-performance equation for storage workloads on Google Cloud." - Scott Fadden, Senior HPC Solutions Architect, Sycomp

“At ClipperDB Technologies, our mission is to drive down the cost and drive up the performance of large-scale Spark analytics. Google Cloud’s C4N instances are the perfect compute engine for our fully native architecture. C4N’s substantial increase in network bandwidth per vCPU combined with large memory configurations and 5th Generation Intel Xeon processors align with ClipperDB’s precise parallel cloud-store prefetching and caching, concurrent dataflow native batch pipelines, streaming no-copy exchange, and cloud store checkpoint fault tolerance to radically accelerate and cost reduce Spark workloads with disaggregated Cloud Storage datalakes.

The results speak for themselves: across industry-standard TPC-DS benchmarks, ClipperDB+C4N delivered over 3x lower cost per query and up to 11x faster analytics, all while maintaining 100% Spark compatibility. We can’t wait to see customers dramatically improve their Spark workload price-performance with C4N coupled with Clipper DB Accelerator." - John Busch, CEO, ClipperDB Technologies

A deeper look at C4N shapes and specs

C4N instances are available in nine different sizes ranging from 2-192 vCPUs and up to 1.5 TB of DDR5 memory, offering predefined shapes in high-cpu, standard, and high-mem configurations. For applications that benefit from caching and high-speed, low-latency local storage, C4N VM instances are equipped with up to 12 TiB of latest Titanium SSDs. C4N with Local-SSD are in preview. For workloads that require direct access to the machine's resources (e.g., hypervisors, container platforms), where nested virtualization does not meet the workload’s performance requirements, or have special performance monitoring or licensing needs, we are introducing C4N bare metal shapes. Coming soon, these native bare metal shapes will offer the same network and storage I/O performance as their virtual machine counterparts. Google Cloud customers can use C4N instances with Compute Engine and Google Kubernetes Engine (GKE), with support for other services coming soon.

Name	vCPUs	Memory (GB)	Local Storage (GiB)	Network Bandwidth		Hyperdisk Extreme Bandwidth (MiB/s)	Hyperdisk Extreme IOPS
Name	vCPUs	Memory (GB)	Local Storage (GiB)	VM-VM (Gbps)	VM-Internet (Gbps)	Hyperdisk Extreme Bandwidth (MiB/s)	Hyperdisk Extreme IOPS
C4n-highcpu	2 - 192	4 - 384	N/A	25 - 400	7 - 200	1.000 - 25,000	80,000 - 1M
C4n-standard	2 - 192	7 - 720	N/A	25 - 400	7 - 200	1.000 - 25,000	80,000 - 1M
C4n-standard-lssd	4 - 192	15 - 720	375 - 12,000	30 - 400	7 - 200	1.000 - 25,000	100,000 - 1M
C4n-highmem	2 - 192	15 - 1,488	N/A	25 - 400	7 - 200	1.000 - 25,000	80,000 - 1M
C4n-highmem-lssd	4 - 192	31 - 1,488	375 - 12,000	30 - 400	7 - 200	1.000 - 25,000	100,000 - 1M

C4N machine series performance and specifications

How to get started

Whether you’re hosting heavy-duty distributed databases, running network virtualization appliances, or orchestrating large-scale data pipelines for AI, C4N is engineered to provide the throughput, scale, and efficiency your business demands. C4N instances are now generally available via on-demand, as Spot VMs, and via reservations. You can also take advantage of further cost savings by purchasing Committed Use Discounts (CUDs) or FlexCUDs in one- and three-year terms in the us-central1 (Iowa), us-east1 (South Carolina), us-east5 (Ohio), us-west1 (Oregon) and europe-west2 (London). For more information visit Network Optimized Machine Type.

Ready to establish a high-performance launchpad for innovation? Head straight to the Google Cloud console to spin up a C4N VM under the “Network Optimized” machine family. Stay up-to-date on regional availability by visiting our regions and zones page or contact your Google Cloud sales representative for more information.

Google Cloud named Leader in the 2026 Gartner® Magic Quadrant™ for AI Infrastructure

Wed, 08 Jul 2026 16:00:00 +0000

In the agentic era, AI is evolving from answering questions to reasoning and taking action. Companies who want to lead in this next phase of AI need computing infrastructure that’s designed and optimized for these new requirements, helping them innovate faster, deliver compelling user and customer experiences, and optimize for cost and energy efficiency — all at massive scale.

Today, we are pleased to announce that Google has been named a Leader in the inaugural Gartner^Ⓡ Magic Quadrant™ for AI Infrastructure, positioned highest for ‘Ability to Execute’ and furthest for ‘Completeness of Vision’. We believe their findings validate our dedication to solving these challenges internally and for our customers.

Read the full report: 2026 Gartner Magic Quadrant™ for AI Infrastructure

Building on the infrastructure foundation powering Gemini

Today’s model and serving architectures require a fundamental rethinking of how silicon and software interact. We realized early on that the platform we envisioned couldn’t be bought off the shelf — we had to invent it. For over a decade, our infrastructure engineers and Google DeepMind researchers have worked shoulder to shoulder to co-design the entire stack for Gemini, YouTube, and Search. We make those innovations, together with popular third party and open source software, available to our customers through Google Cloud. Today our integrated stack serves 9 out of 10 frontier AI labs; capital markets firms like Citadel Securities; and enterprises like Mercedes Benz.

At the hardware layer, Gartner recognized our commitment to custom silicon as a core strength. Earlier this year we shared two new advancements in custom silicon, our 8th generation TPUs, engineered to solve enterprise scaling and memory bottlenecks at a systems level:

TPU 8t, the training powerhouse: Purpose-built to optimize training timelines, TPU 8t packs 9,600 chips into a single superpod, delivering the high-density compute required for frontier models with nearly 3x the compute performance per pod over the previous generation.
TPU 8i, the inference engine: Engineered to handle the collaborative, iterative work of specialized agents, TPU 8i breaks the memory wall for real-time agentic workflows, with 288 GB of high-bandwidth memory and 384 MB of on-chip SRAM — 3x more than the previous generation — keeping a model's active working set entirely on-chip.

While our TPU platforms push the boundaries of what is possible, we know that one size doesn't fit all. Different customers have different workloads, different requirements, and different use cases. So, we also partner deeply with NVIDIA to deliver the latest accelerated computing platforms as highly performant, reliable and scalable services in Google Cloud. We will be among the first to deliver A5X instances based on the next-generation Vera Rubin platform when it becomes available later this year, enabling customer choice. We also work closely with NVIDIA to integrate GPUs into many Google Cloud software services to give our customers easier access to accelerated computing.

To enable even more flexibility, we continue to contribute to open-source projects across the orchestration, inference engines, and framework layers through llm-d and vLLM. We also recently announced TorchTPU, which gives PyTorch developers portability without complex code rewrites while maximizing the performance of their deployment.

Get more performance per dollar on AI Hypercomputer

As your infrastructure investment grows, you need to balance raw performance and cost to make AI applications economically viable. Taking a ‘buy now, integrate later’ approach to AI is becoming unsustainable. By combining pre-integrated hardware and open software frameworks that feature flexible consumption models, we deliver a unified system engineered for better performance per dollar across training, reinforcement learning, and inference.

Gartner recognized our integrated AI Hypercomputer as a core strength. This AI-optimized infrastructure is engineered to drastically improve your performance per dollar:

A massive compute cluster is only as effective as the storage system feeding it data. Google Cloud Managed Lustre, powered by our new C4NX instances and Hyperdisk Exapools, now delivers 10 TB/s of bandwidth — up to 20x faster than other hyperscalers — while Rapid Buckets transforms object storage with up to 20 million operations per second, helping ensuring large-scale training checkpoints and recoveries happen near-instantly.
Our Virgo Network provides a high-bandwidth scale-out fabric capable of connecting more than one million TPUs across multiple data center sites into a training cluster, or up to 960,000 GPUs across multiple sites without performance degradation — transforming globally distributed infrastructure into a unified supercomputer.
GKE Inference Gateway enables scaling models in production with near-zero latency by combining LLM-aware routing, caching, and the disaggregated serving capabilities of llm-d, increasing throughput by up to 40% while reducing serving costs up to 30%.

Run AI on a fluid infrastructure at virtually any scale

In the agentic era, infrastructure cannot be a rigid, static constraint. It must be an intelligent resource that adapts to the shifting priorities of your business, scaling up with demand and down to zero when agents are idle, with consistent, reliable performance. On AI Hypercomputer, you can:

Train smarter and faster, using Cluster Director and Google Kubernetes Engine to scale up to 130,000 nodes. At the same time, squeeze up to 97% productivity (Goodput) out of every accelerator using TPU 8t together with software co-designed with Google DeepMind and integrated open-source frameworks — from JAX to Pathways and Pallas.
Enable secure, low latency agent execution with GKE Agent Sandbox. Because agents need to scale, GKE Agent Sandbox can sense agent bursts and respond rapidly — provisioning up to 300 sandboxes per second per cluster, then instantly scale back when agents sit idle, optimizing compute costs.
Run distributed enterprise and AI workloads consistently across multicloud, edge, and on premises environments with Cross-Cloud Network and Cloud WAN. This approach delivers low-latency, policy-driven connectivity across Google’s private global backbone spanning over 10+ million kilometers of fiber and over 200 countries and territories, with up to 40% higher performance than public internet routing.

Take the next steps on your journey with AI Hypercomputer

From frontier models, to billion user applications, AI Hypercomputer gives you the purpose-built hardware, open software, and flexible consumption models you need to improve AI performance, cost, and developer productivity. We are honored to see decades of experience building scalable, affordable and reliable AI systems rewarded with a leadership position in Gartner’s research.

You can download a complimentary copy of the 2026 Gartner Magic Quadrant™ for AI Infrastructure on our website.

Enhanced data resilience with cross-region backups in Backup and DR Service

Wed, 24 Jun 2026 17:00:00 +0000

To maintain business continuity, you need a robust data-backup strategy. While multi-region backups offer the highest availability, many organizations want a more cost-effective way to protect their data against a regional outage, but still adhere to data residency requirements.

Building on our foundation of multi-region backup protection, we are excited to announce the general availability (GA) of cross-region backups for our Backup and DR Service. With this update, your backup destination is no longer tethered to your source: The backup regions can be entirely distinct from the region where the primary workload is located. This decoupling is essential for protecting against localized regional outages while maintaining granular control over where your data lives.

This capability is now fully available for Compute Engine instances, Disks and Filestore, with support for Cloud SQL and AlloyDB to follow.

Why cross-region backups?

While multi-region backups provide near-zero downtime, they can carry higher infrastructure costs that you don’t always need for every application. Cross-region backups bridge this gap by offering:

Optimizing costs: Selectively designate recovery regions, offering a granular alternative to standard multi-region deployments while enabling control and maximizing value.
Simplifying compliance: Navigate complex data residency laws (like GDPR) by choosing exactly which geopolitical boundary your backup should reside in.
Improving regional resilience: Protect against localized disasters by placing a restorable copy of your data in a completely different geographical region.

Here’s how it works

We designed implementing a cross-region backup strategy to be intuitive and integrated into your existing workflows:

1. Create a backup vault: Set up a backup vault in a region different from your source resource.

2. Configure the backup plan: Create a backup plan in the resource's region but select the vault located in the secondary region.

3. Attach and automate: Attach the plan to your resource. Backup and DR handles the rest, moving your data directly to a regional backup vault, outside of the source region.

In short, cross-region backups deliver the resilience and cost-efficiency organizations need without compromising on complex compliance and data residency standards. By allowing direct backups to secondary regions, you get a robust layer of protection against localized disasters, with more autonomy to select secondary locations compared to the pre-defined boundaries of multi-region deployments. We encourage you to explore these new capabilities and experience how Backup and DR can enhance your data resilience strategy.

Resources

Storage Insights datasets: Enabling org-wide operational discovery with activity insights

Tue, 09 Jun 2026 16:00:00 +0000

As enterprise storage footprints scale to billions of objects, AI applications and agentic workloads are fundamentally shifting the role of storage from a passive repository to the foundation of the data platform. This is driven by a surge in unstructured model data and the billions of actions performed on those objects, including session logs and audit trails. To manage this and answer questions about cost, operations, and security, storage and platform admins need to go beyond knowing what data they have, to understanding exactly how it is being accessed, moved, and modified.

To help, we're excited to announce activity insights within Storage Insights datasets. Now generally available, these new views provide visibility into the operational details of your Google Cloud Storage assets, enabling data-driven cost optimization and faster troubleshooting. For example, with activity insights, you can answer questions like:

Are my objects located in the right storage classes within my buckets?
What regions is my bucket interacting with the most so I can assess if it is optimally located?
Where are there errors across operations on my storage estate and why?

Answering these questions confidently is the key to unlocking cost optimizations and reclaiming engineering time. Storage Insights datasets, a feature of Storage Intelligence for Cloud Storage, provides daily metadata and frequent activity insights (typically within four hours of the activity) so you have better visibility into your storage estate. While Storage Intelligence is a unified management product with capabilities like Bucket relocation, Batch operations and Gemini Cloud Assist, this blog focuses on how you can leverage Storage Insights datasets for operational optimization.

What are Storage Insights datasets?

Storage Insights datasets deliver an automated, query-ready BigQuery index of your entire storage estate, complete with raw metadata and activity insights, replacing manual, error-prone data collection. Storage Insights datasets can be customized in scope: create a dataset for your entire org, a specific folder, a project, or a set of projects, or even specific buckets. The dataset then refreshes with regular updates, giving you a comprehensive view of your storage.

From static metadata to live intelligence

Storage Insights datasets are your go-to tool for understanding your storage metadata, acting as an inventory management tool, scanning object metadata (storage class, location, age, custom metadata) and organizing it into a powerful, queryable BigQuery-linked dataset. This is crucial for knowing what data you have (learn more about how to optimize storage spend with Storage Insights datasets here).

But what if you also knew how and when that data is being used?

Storage Insights datasets now offers a set of new views that capture:

Object-level activity, including writes, updates, deletes, and errors
Bucket-level aggregate activity, including total object operations, a breakdown by type of operations, total errors and most active prefixes
Bucket-level regional traffic activity, including ingress and egress bytes per region that interact with your bucket
Project-level aggregate activity, including total object operations, a breakdown by type of operations and total errors

This data flows directly into new BigQuery views within your dataset so you can run analytics queries for specific insights, interact with the data via Gemini or simply connect it to powerful Looker dashboards for visualization.

This moves you from a static snapshot to a dynamic, queryable analysis of your data's entire lifecycle. It's the difference between knowing what's in your warehouse and knowing what’s used and when.

Three ways to use activity insights immediately

Here’s what you can do, starting today, with activity insights in Storage Intelligence datasets.

1. Right-size your storage estate

The challenge: You have terabytes of data in Standard or Nearline class storage that you believe is cold. But without proof, moving it to Coldline or Archive class is risky. What if a critical process still needs to read it once per quarter?
The solution: With the new Storage Intelligence views that surface activity insights, you can now identify buckets that have had minimal read/write activity over the last 30, 60, or 90 days.
The outcome: Apply or fine-tune lifecycle policies to transition this data to more cost-effective storage classes.

For example, here’s a SQL query to order all the buckets in your estate with little to no activity in the last six months:

code_block: <ListValue: [StructValue([('code', 'SELECT name, location, project, totalRequests\r\nFROM\r\n `[project]`.`[dataset]`.`bucket_activity_view`\r\nWHERE\r\n snapshotEndTime >= TIMESTAMP(DATE_SUB(DATE_TRUNC(CURRENT_DATE(), MONTH), INTERVAL 5 MONTH))\r\n AND snapshotEndTime < CURRENT_TIMESTAMP()\r\nORDER BY totalRequests ASC\r\n\r\n//Running queries in Datasets accrues BQ query costs, refer to the pricing page for further details.'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7fb519129910>)])]>

2. Architect for global performance with data-driven bucket placement

The challenge: Your team set up a multi-region bucket to serve a global application. But a year later, is that still the right architecture? What if 99% of your traffic is now coming from a single region?
The solution: Analyze the access patterns in your new bucket_region_activity_view table. You can easily pinpoint which regions are driving read and write activity for the bucket.
The outcome: Make data-driven decisions to co-locate your bucket with your compute. You might find that changing a multi-region bucket to a single-region one (or vice-versa) can lead to significant cost-savings and even improve performance.

For example, here’s a SQL query to break down the egress and ingress traffic pattern for a bucket across regions:

code_block: <ListValue: [StructValue([('code', "SELECT\r\n requestLocation,\r\n bucketLocation,\r\n SUM(requestBytes) AS total_request_bytes,\r\n SUM(responseBytes) AS total_response_bytes\r\nFROM\r\n `[project]`.`[dataset]`.`bucket_region_activity_view` \r\nWHERE\r\n name = '[bucket name]'\r\nGROUP BY\r\n requestLocation,\r\n bucketLocation;\r\n\r\n\r\n//Running queries in Datasets accrues BQ query costs, refer to the pricing page for further details."), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7fb519f74d10>)])]>

Shipt, a retail technology platform and same-day delivery service, has been using Storage Intelligence capabilities to inform their data location decisions:

“Storage Intelligence enables us to efficiently manage over 2 billion objects, delivering cost and performance optimization. With Insights datasets, we detected and analyzed egress charges from multi-region buckets, identifying opportunities to improve efficiency by co-locating compute and storage. By leveraging the Bucket Relocate capability, we seamlessly moved 1.3 Petabytes of data from multi-region to regional storage, achieving substantial cost savings while maintaining uninterrupted application performance and data pipeline continuity.” - Ron Cuirle, Director of Engineering - Cloud Platforms, Shipt

3. Demystify and resolve operational hotspots

The challenge: Your team sees a spike in 429 (too many requests) errors. In a massive environment, this is rarely just a performance hiccup — it’s expensive! These errors trigger automatic retries, which often lead to a cycle of high-frequency, billable operations that drive up your Class A costs. Pinpointing exactly which object or prefix is causing this can be a time-consuming troubleshooting nightmare.
The solution: The new Storage Insights datasets views provide granular details on these errors, right in BigQuery. You can query for 429 errors and see exactly which objects and prefixes are under pressure.
The outcome: Additionally, you can pinpoint the cause of your 429 errors, moving your team from troubleshooting to resolution.

For example, here’s a SQL query to analyze 429s occurring across your estate, where they are happening and why:

code_block: <ListValue: [StructValue([('code', 'SELECT\r\n requestOperation,\r\n errorReason,\r\n objectName,\r\n bucketName,\r\n requestCompletionTimestamp,\r\n project\r\nFROM\r\n `[project]`.`[dataset]`.`object_events_view` \r\nWHERE\r\n responseStatus = 429\r\nORDER BY\r\n requestCompletionTimestamp DESC;\r\n\r\n\r\n//Running queries in Datasets accrues BQ query costs, refer to the pricing page for further details.'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7fb519f75c50>)])]>

Getting started

As your organization grows with Google Cloud, the scale of your data will only increase. Stop relying on archival data and start optimizing your organization’s storage estate. Cloud Storage Storage Insights datasets with activity insights turn massive data estates from complex operational challenges into clearly understood, highly optimized assets.

To get started, check out use our pre-configured Looker Studio template here to connect to your dataset for quick analysis and value:

For example: View the trend for Total Reads on your bucket over time

Or, analyze the ingress and egress traffic patterns for your bucket:

Ready to turn insight into action?

Enable Storage Intelligence today in the Google Cloud console.
Configure your dataset today and query your data directly in BigQuery or connect to our Looker template for quick and easy visualization.
Reference our videos for more information on Storage Intelligence and How to Get Started.
Read more about how to optimize your Cloud Storage footprint with Storage Insights datasets.

Connecting AI agents with unstructured data using Google Cloud Storage MCP Servers

Tue, 02 Jun 2026 17:00:00 +0000

Google Cloud Storage (GCS) is a foundational component of the modern agentic tech stack and the preferred home for unstructured data at scale. As enterprises deploy agents in production, the critical focus has shifted to turning data into context and building secure, standardized integrations to access context. This is the core of smart storage: making unstructured data inherently agent-ready by turning passive objects into rich context for reasoning. Whether it’s automating complex financial workflows or diagnosing system failures in seconds, AI success now depends on how seamlessly agents can leverage this intelligence to make smart, high-stakes decisions.

In this blog, we will share three examples of agents built by customers using GCS, and then share how you can securely and reliably connect your agents to GCS using Model Context Protocol (MCP). Combined with smart storage features like auto annotations and object contexts, GCS MCP server makes the whole agent deployment process easy and simple.

Real-world agent success on Google Cloud Storage

We are seeing incredible innovation from customers leveraging MCP and Google’s agentic tech stack to solve complex business problems:

Palo Alto Networks built the Strata Co-Pilot agent, a screen-aware AI assistant that guides network security administrators through complex configuration flows—either by highlighting steps or executing them directly. The agent is powered by the Gemini Live API, with GCS serving as its “historical memory” connected via the GCS MCP server.
Airwallex developed an AI Assistant that understands user context, answers questions, and executes workflows on their behalf. For example, it can smartly analyze expense policy documents and generate detailed approval workflows - a task that would normally take hours to do manually. GCS and GCS metadata are used by the agent to store documents and the extracted information, respectively.

Snap's Job Optimization Agent analyzes Flink and Spark job specs, metadata, and historical metrics stored on GCS across thousands of jobs to find optimization opportunities, generate cost estimates, and tune configurations. Using this agent, Snap is already seeing investigation time reduced from 30 minutes to 30 seconds!

In all these three agents, the GCS MCP server handles data operations as well as enforces standard RBAC and access policies.

Connecting agents to GCS using MCP

MCP has rapidly emerged as the universal standard for connecting agents to data sources, but building custom servers from scratch is often a slow, distracting process that diverts focus from innovation. This path introduces significant development overhead and risk, as it forces you to manage everything from authentication and error handling to keeping pace with GCS’s evolving capabilities. To solve this, GCS offers two powerful MCP server options — Remote and Local — allowing you to offload the foundational plumbing and focus on creating value.

1. Remote MCP server: Fully-managed
Connecting your agents to the Cloud Storage MCP server requires zero infrastructure deployment. By simply pointing your agent configuration to the managed endpoint, you gain immediate access to your unstructured data on GCS, allowing you to scale your agentic workloads effortlessly without the burden of operational overhead.

Because the Cloud Storage MCP server follows the open MCP standard, it works seamlessly with major agentic frameworks like ADK and is compatible with MCP clients. You can easily connect clients like Google Antigravity and Anthropic’s Claude by adding a Custom Connector in the settings. Simply point it to your Cloud Storage MCP endpoint, and you are ready to start building — no complex configuration files required.

Connecting an agent to storage requires robust security and governance. GCS MCP server is built on Google Cloud's standard identity, observability, and security frameworks:

Identity-first security: Authentication is handled entirely through Identity and Access Management (IAM) rather than shared keys. This ensures agents can only access data (buckets and objects) explicitly authorized by the user.
Full observability: To track agent activity, every request and action taken via these MCP servers is logged in Cloud Audit Logs. This provides security teams with a record of every interaction, maintaining visibility alongside ease of access.
MCP security - content scanning: You can optionally configure the MCP endpoint with Google’s content security service, Google Cloud Model Armor. This allows you to implement security controls against common MCP attack vectors—such as direct and indirect prompt injection attacks, MCP Tool poisoning attacks, and malicious URL/SQL injections—as well as prevent the leakage of sensitive data.

Cloud Storage MCP servers are perfect for most production use cases; however, as with all remote servers, you lose the capability to fully customize your MCP tools.

2. Local MCP Server: Self-managed for controlled customization
While the Remote server handles standard data access, Local MCP is the right choice when you need to build custom tools specific to your business logic. For example, if your agent needs to perform specialized data transformations—such as redacting PII or adding context from another internal system—whenever it reads a file from GCS, a Local MCP server allows you to define those unique capabilities

The GCS Local MCP server is an open-source GitHub repository of Google-maintained tools that provides you with a reliable bridge to your data. Here are a few tips to keep in mind while designing custom tools:

Provide precise, clear descriptions to minimize incorrect invocations by the models
Implement model-friendly error handling for models to understand their mistakes and self-correct

The GCS Local MCP is now also a part of the MCP Toolbox for Databases, a single open-source repository containing connectors for major data services such as GCS, BigQuery, AlloyDB, Spanner, and Cloud SQL, making it easier to monitor and manage your data ecosystem. The Toolbox offers simplified development with reduced boilerplate code, enhanced security through OAuth2 and OIDC, and end-to-end observability with OpenTelemetry integration.

Get started

Whether you are optimizing an existing process like Snap or automating workflow creations like Airwallex, your unstructured data is one of your agent's greatest assets.

Explore the generally available GCS Remote MCP Server.
Check out our GCS Local MCP GitHub repository to start building custom tools today, or use it as part of MCP Toolbox for Databases.
Reach out to us to discuss your Agent use case with GCS data.

Cloud Storage Rapid: Turbocharged object storage for AI and analytics

Mon, 11 May 2026 17:00:00 +0000

At Google Cloud Next ’26 we announced Cloud Storage Rapid, a family of object storage capabilities for data-intensive workloads like AI and analytics. Out of the gate, Cloud Storage Rapid consists of 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.

Cloud Storage Rapid is our response to the generational shift in how organizations build with AI. Teams are training trillion-parameter models, deploying inference at global scale, and building autonomous agents that reason over vast amounts of enterprise data. While accelerators like GPUs and TPUs often get the spotlight, they have a critical dependency: storage.

Storage is the engine that feeds accelerators during training, and the fast-access layer that makes real-time inference responsive. But as models scale, storage performance can be a bottleneck. Every time an AI/ML cluster waits on a data read or a checkpoint write stalls, you are paying for expensive compute cycles that aren't doing useful work.

Historically, AI/ML practitioners have had to choose between the specialized performance of a niche, zonal storage system, and the reliability and scale of a global object store like Google Cloud Storage. Many developers value Cloud Storage for its simplicity, scalability, reliability, and cost-effectiveness, but as the AI era has progressed, they’ve been throwing hotter and hotter workloads at it, running training and inference workloads with thousands of GPUs and TPUs. We’ve reached a performance tipping point that traditional object storage was never meant to handle. The Rapid family provides multiple options for co-locating compute workloads directly with high-performance zonal storage. It minimizes I/O bottlenecks that can block accelerators, so that your GPUs and TPUs stay fully saturated and productive. In this blog, let’s take a closer look at Cloud Storage Rapid’s capabilities.

Rapid Bucket

Rapid Bucket (GA), helps Cloud Storage meet the evolving demands of massive-scale generative AI, analytics, and other high-performance workloads. It does so by leveraging Colossus, the Google distributed storage system that powers Gemini and YouTube, to provide massive read/write performance and ultra-low latency in a dedicated object storage zonal bucket.

Lightning-fast performance
By combining the sub-millisecond latency of block-like storage, the throughput of a parallel filesystem, and the scalability and ease of use of object storage, Rapid Bucket provides high performance from the same Cloud Storage that you know and love.

Highlights include:

Ultra-low latency: Achieve up to 20 million queries per second and sub-millisecond latency.
Massive scalability: Rapid Bucket delivers 15+ TB/s of aggregate read throughput from a single Rapid zonal bucket.
New semantics: Enable higher performance with new capabilities such as native appends, unlimited readers (while writing!), and vectored reads.

Optimized for AI and analytics
You can use Rapid Bucket for a variety of demanding scenarios, including AI/ML data preparation, training, checkpointing, batch and streaming analytics processing, and optimizing distributed database architectures.

Key benefits include:

Optimized accelerator utilization: With Rapid Bucket, we observed 50% reduced blocked GPU time and up to 2.5x faster data loading for multi-modal training runs.
Faster checkpointing: Rapid Bucket makes checkpoint restores up to 5x faster and writes 3.2x faster compared to traditional object storage. This ensures faster recovery from workload interruptions, minimizes wasted accelerator time, and increases overall efficiency.

>5x faster checkpoint restores with Rapid Bucket

>3.2x faster checkpoint writes with Rapid Bucket

You can get started with Rapid Bucket here.

Rapid Cache

Originally announced at Cloud Next ‘25, Rapid Cache accelerates bandwidth for AI/ML workloads like data prep, training, and bursty model loading for inference, delivering an aggregate read throughput of 2.5 TB/s for your existing buckets — with no code changes. For inference workloads, we’ve observed that Rapid Cache provides up to 2.1x (114%) accelerated model load, resulting in 47% TCO savings.

When combined with multi-region buckets, customers can flexibly access GPUs and TPUs distributed across regions in a geo, while maintaining a single bucket namespace. This eliminates the need for manually orchestrated data movements between buckets, while benefitting from zonally co-located high performance.

New: Rapid Cache ingest on write
Customers at some of the world’s largest frontier AI/ML labs told us that they were looking for ways to accelerate reads immediately after a write, such as checkpoint restore workloads or a data prep pipeline that then feeds training. Before, caching the data required an initial read to trigger ingestion, which was served directly from the bucket at standard performance.

Rapid Cache’s new ingest on write feature solves this by simultaneously writing data to the Rapid Cache as it is being written to a Cloud Storage bucket. This proactive approach eliminates the initial cache-miss penalty, and helps workloads benefit from an immediate cache hit on the very first read. This provides up to 2.2x faster checkpoint restore times, allowing training clusters to recover faster from interruption.

To enable ingest on write, simply modify the ingestion criteria of your existing Rapid Cache.

Rapid Cache’s simplicity and performance has resulted in explosive adoption. In just one year since General Availability, customers have deployed thousands of Rapid Caches with a 20x growth in caches deployed, In fact,Rapid Cache serves up to 20% of Cloud Storage’s global egress. Cutting-edge AI/ML customers deploy their workloads on Rapid Cache, including Anthropic who uses Rapid Cache to improve the resilience of their cloud workload by co-locating data with TPUs in a single zone and providing dynamically scalable read throughput up to 2.5TB/s.

Case study: Thinking Machines Lab
Thinking Machines Lab is an artificial intelligence research and product company. Its mission is to make AI systems that are adaptable and customizable, building a future where everyone has access to the knowledge and tools to make AI work for their unique needs and goals.

At Next ‘26, James Sun, Member of Technical Staff at Thinking Machines Lab, spoke at our session, Cloud Storage Rapid: Turbocharged object storage for AI & Analytics, where he presented about the needs of the data-hungry AI/ML workloads that Thinking Machines Lab runs for high-performance storage at scale.

Thinking Machines runs diverse workflows: data processing in Dataflow, Kafka, and Spark, multi-model training, and serving Tinker — a flexible API for fine-tuning open source models. Thinking Machines' workloads run on Google Cloud Storage, Sun explained. Running these data-intensive AI/ML workloads at such a large scale introduces significant infrastructure challenges.

The first is managing a hub and spoke data architecture, where data processing hubs are located in one primary region while training GPUs are spread across multiple regions. Historically, this has made manual data movement and lifecycle management a major operational pain point. Furthermore, Thinking Machines Lab's workloads such as data prep and pretraining workflows, which rely on massive-scale Spark workloads to prepare their multi-modal datasets, often spike from cold to hot instantly. Previously, these surges led to disruptive 429 errors, which stalled data processing and loading, and interrupted critical training cycles.

To minimize these bottlenecks, Thinking Machines Lab integrated Rapid Cache across their AI/ML pipeline, to positive results.

“Rapid Cache has become a core foundation of our AI/ML data infrastructure, supporting our critical workflows, from data prep and pretraining to training and model loading. By acting as a crucial bandwidth shield and booster, it enables us to scale our data-intensive workloads across our entire fleet without compromise, providing us with the on-demand high bandwidth and consistent stability that we need to innovate at speed.” - James Sun, Member of Technical Staff, Thinking Machines Lab

In short, Cloud Storage and Rapid Cache provides Thinking Machines Lab with:

Easy, instant, scalable, on demand bandwidth: The team now achieves stable read throughput peaks of over 1.8TB/s.
Enhanced stability: Rapid Cache has greatly reduced tail-end latencies and 429 errors, providing the consistent performance needed for multi-modal training.
Fleet-wide scalability: Combined with multi-region buckets, they can now scale data-intensive workloads across their entire fleet, meeting the demands of a rapidly growing compute scale without the hassle of manual data movement while benefiting from zonally colocated storage for high performance.
Operational efficiency: The use of Hierarchical Namespace (HNS) has optimized their massive Spark workloads for data preparation, by supporting fast directory renames, along with providing the ability to ramp QPS more quickly as they scale out clusters. Rapid Cache’s "ingest on write" capability helps ensure immediate cache hits for checkpoint restores.

Choose your rocket ship

Whether you are running data preparation, massive-scale training, or low-latency inference, Cloud Storage Rapid delivers high performance together with the reliability and scalability that Cloud Storage is known for.

Rapid Bucket delivers the highest Cloud Storage throughput and queries per second as well as the lowest latency for read/write use cases, such as analytics, AI training, checkpointing, and model serving. This helps to reduce storage bottlenecks and increase compute utilization.
Rapid Cache provides higher read bandwidth and tail latency stabilization in existing buckets, without code changes. Key use cases include AI training, checkpoint restores, and serving, as well as accelerator optionality via multi-region buckets.

Get started with the Cloud Storage Rapid family today!

Storage innovations to accelerate your AI workloads at Next ‘26

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

At Google Cloud Next, we are announcing innovations across every layer of our storage stacks — performance, intelligence, and management — to ensure your data is as fast and as useful as the AI models, apps and agents you are building.

Why it matters: Storage is no longer just a place to keep data. When training AI models, storage is the engine that feeds data-hungry accelerators. During AI inference, it’s the access layer that makes it responsive, acting as the source for the context that AI agents need to be effective. When storage performance falls short, accelerators sit idle, agents respond slowly, and data remains invisible to AI models.

But storage performance is only half the battle; you also need storage that’s smart. With the help of Google’s AI models integrated directly into the storage layer, you’re no longer just storing bits, but data that has full context about its content. In this new era of smart storage, raw data becomes a valuable asset that’s ready to use by a variety of downstream AI and enterprise applications.

What’s new:

High-performance storage infrastructure: New Rapid family of features in Cloud Storage for high-performance object storage; delivering 10x performance enhancements plus a new cost-effective Dynamic tier for Google Cloud Managed Lustre.
Smart Storage: Unlocking unstructured data with automated metadata annotation, and AI agent connectivity via MCP.
Storage Intelligence: Streamlined data management through zero-configuration dashboards, aggregated activity views, and enhanced batch operations.
Enhanced ecosystem: Expanded capabilities across Google Cloud NetApp Volumes, Filestore for GKE, and our backup and data protection portfolio.

Let’s take a deeper look at the storage enhancements we are unveiling this week.

Storage infrastructure that keeps up with AI

As AI models scale, getting data from the storage to the compute layer fast enough can be a bottleneck. New storage capabilities move performance directly into the storage layer, reducing total cost of ownership (TCO) and keeping accelerators fully utilized.

Cloud Storage Rapid
Cloud-based object storage like our Cloud Storage is scalable and cost-effective, but bottlenecks can stall AI jobs and waste expensive compute cycles. Every time a training cluster waits on a read or a checkpoint write stalls, you're paying for accelerators that aren't doing useful work.

Cloud Storage Rapid marks a fundamental shift in designing AI infrastructure: you no longer have to choose between the reliability of object storage and the high performance of a specialized AI storage system. Cloud Storage Rapid lets you leverage the industry-leading durability, massive distributed scale, and cost-effective auto-tiering of object storage, while simultaneously achieving extreme throughput, frequent I/Os, and ultra-low latency. With native integrations into PyTorch and JAX, Cloud Storage Rapid is optimized out-of-box for the most popular AI/ML ecosystem frameworks, so that your data preparation, training, and inference workloads run on a high-performance and reliable foundation.

The Cloud Storage Rapid family consists of two offerings: Rapid Bucket and Rapid Cache.

Rapid Bucket, now generally available, leverages Colossus, the Google distributed storage system that powers Gemini and YouTube, to deliver more than 15 TB/s of bandwidth, 20 million requests per second, and sub-millisecond latency in a single zonal bucket. With access via high-performance gRPC and S3-compatible APIs, Rapid Bucket increases accelerator utilization for multi-modal training with 50% reduced GPU blocked time and 2.5x faster data loading. Checkpoint restores are 5x faster and checkpoint writes are 3.2x faster compared to traditional object storage, reducing workload interruptions and wasted GPU time.

Checkpoint writes are 3.2x faster and restores are 5x faster with Rapid Bucket

Rapid Cache, formerly Anywhere Cache, accelerates bandwidth for bursty workloads like model loading for inference, delivering an aggregate read throughput of 2.5 TB/s for existing buckets, with no code changes. The new ingest-on-write feature provides up to 2.2x faster checkpoint restores, allowing training clusters to recover faster from interruptions.

Rapid Cache’s combination of simplicity and performance has resulted in strong adoption, including from cutting-edge AI/ML customers like Thinking Machines Lab.

“Rapid Cache has become a core foundation of our AI/ML data infrastructure, supporting our critical workflows, from data prep and pretraining to training and model loading. By acting as a crucial bandwidth shield and booster, it enables us to scale our data-intensive workloads across our entire fleet without compromise, providing us with the on-demand high bandwidth and consistent stability that we need to innovate at speed.” - James Sun, Member of Technical Staff, Thinking Machines Lab

Google Cloud Managed Lustre
The Lustre parallel file system is the industry standard for organizations whose AI training and inference workloads require high throughput and sub-millisecond latency, and is trusted by AI labs and HPC centers worldwide to feed thousands of accelerators simultaneously and keep them saturated under pressure. Google Cloud Managed Lustre brings that capability as a fully managed service, and with today's announcements, it is the most performant managed Lustre offering available in any cloud.

Managed Lustre now delivers up to 10 TB/s of throughput — a 10x increase since last year and 4–20x higher than managed Lustre offerings from other hyperscalers for a single instance. Powered by C4NX VMs and Hyperdisk Exapools, Managed Lustre writes and restores checkpoints 2.6x faster when compared to other Google Cloud storage solutions.

The new Dynamic tier ($0.06/GB-month) delivers the low-latency performance required for intense AI workloads like training and checkpointing. By serving data from persistent disk rather than relying on object-based caching, we eliminate a performance cliff — helping ensure your data remains responsive and your accelerators stay productive. A single SKU provides simple, predictable billing without the hidden complexity of traditional data tiering.

“By integrating Managed Lustre we eliminated the typical onboarding bottlenecks, allowing us to hit the ground running with the inferencing workload. This high-throughput, low-latency storage keeps our B200 GPUs fully saturated, driving a substantial performance gain in LLM inference over the H200. For our customers, this translates directly into faster, more responsive AI agents that can handle complex reasoning at a fraction of the previous latency.” - Lavnaya Karanam, Software Engineering PMTS, Salesforce

Smart Storage: Context for the AI era

The beauty of an object storage system like Cloud Storage has long been its simplicity: the system knows the object’s name, its size, and when it was created. But if you want to understand the object’s content — what entities it references, whether it contains sensitive PII, or whether it’s relevant to a pending query — you need to use custom pipelines, separate databases, and bespoke enrichment systems.

AI has changed the equation. To fine-tune a model, you need to select the right objects from the get-go, from a corpus of millions. Building an agent requires retrieving the right context for each decision. To meet a compliance obligation, you need to know what every file contains up front, before it becomes a liability. In each case, the bottleneck isn’t compute or model quality — it’s the inability to describe, find, and act on objects at scale.

To bridge that gap between stored and usable data, last year we introduced Smart Storage, a set of capabilities built directly into Cloud Storage that makes every object self-describing. New Smart Storage capabilities include:

Automated annotations, which eliminates the need to build and maintain custom annotation pipelines. With Smart Storage enabled, Cloud Storage can now automatically generate context — including image annotations — so your data is discoverable and usable from the moment it lands. You pay to annotate the data once at write time, and every downstream system can use those annotations immediately for the life of the object.
Cloud Storage MCP server lets you read, write, and analyze Cloud Storage data using the standard MCP protocol.

Smart Storage enables these capabilities, and others, thanks to its object context, now generally available. This metadata substrate adds structured, mutable, IAM-governed context to every object. Customers write their own tags and classifications; Google's annotation pipelines automatically attach labels, extracted entities, and compliance signals.

With Smart Storage, ML teams can select training datasets from semantic criteria without building retrieval pipelines. AI agents can ground their reasoning in enterprise data without a separate retrieval layer.

Storage Intelligence: Data management at AI scale

As data estates grow to hundreds of petabytes, storage costs can spike without warning, and security blind spots can multiply across billions of objects. To manage this, teams have to stitch together multiple tools just to answer basic questions about their own data.

Last year we launched Storage Intelligence to give enterprises a unified management experience built directly into Cloud Storage. Today, 70% of our largest customers use Storage Intelligence, each of whom manage over 50 billion objects.

Storage Intelligence provides a single view across your entire project or organization, with unique capabilities like bucket relocations across regions. Today, we're making it significantly more powerful with:

New zero-configuration dashboards instantly surface cost anomalies and integrate Security Command Center’s Data Security Posture Management (DSPM) data governance feature, to detect critical security vulnerabilities across Cloud Storage — no setup required.
New object events and bucket activity tables in Insights Datasets now drive deeper cost analysis and accelerate operational tasks. You can use these insights to perform a wide range of analyses, from optimizing bucket placement based on egress patterns to quickly troubleshooting 429 errors by finding the impacted objects.
Enhanced batch operations make it even simpler to act on billions of objects with new change ACL and storage class operations, and support for multi-bucket operations.

Enhancing the storage ecosystem

Beyond our core storage offerings, we are streamlining how enterprises migrate to and protect data in the cloud.

Google Cloud NetApp Volumes: With the launch of Flex Unified, NetApp Volumes now provides a unified enterprise storage platform that bridges the data center and the cloud, provisioning 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.
Filestore for GKE: Developers building AI workloads on Google Kubernetes Engine (GKE) can start small, with shares as small as 100 GiB, and scale capacity and IOPS independently. At the same time, tighter integration to the Colossus distributed file system provides more scale and enterprise capabilities.
Data protection: Google Cloud Backup and DR now features agentic AI capabilities that can autonomously audit your backup estate and remediate coverage gaps, with new GA integrations for AlloyDB and Filestore.

Where to start

As you navigate today’s generational AI shift, you need a storage foundation to support ever-larger, more intelligent, and autonomous models. With new high-performance and intelligent storage layers, plus enhanced storage management tools and a deeper data protection bench, Google Cloud’s storage platforms lets you understand and use your enterprise data in ways that weren’t previously possible, allowing you to:

Reduce the AI data bottleneck: Saturate compute and accelerate ROI. Keep your expensive GPUs and TPUs fully productive with high-throughput storage that delivers the extreme performance required for large-scale training and inference.
Build agent-ready data foundations: Shift from building custom pipelines to an active knowledge base where self-describing objects let AI agents instantly reason over data without manual prep.
Minimize blind spots across exabytes: Replace fragmented management tools with zero-configuration dashboards and datasets to instantly surface cost anomalies and security risks across billions of objects.
Embrace the storage ecosystem: Streamline migration and protection. Bridge your data center to the cloud, scale containerized apps, and automate data resilience with agentic AI.

Visit the Google Cloud Storage console to explore these new features, read more about Cloud Storage Rapid, or explore our Next '26 storage sessions.

Cross-cloud infrastructure innovation for the agentic enterprise

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

The era of agentic AI is accelerating from human- to machine-speed operations, while also creating profound stress on legacy technology infrastructure. This new reality pushes foundational systems to their limits: agents generate thousands of internal messages and complex queries, spawning more agents, all of which can rapidly overwhelm traditional networks and databases, and expose new security vulnerabilities.

Unlocking AI's full potential in the era of agents requires a secure, adaptive foundation. We call it cross-cloud infrastructure for the agentic enterprise – and at Google Cloud Next ‘26, we’re launching a powerful set of new innovations across four areas:

What’s new:

Fluid compute: Google Compute Engine and Kubernetes services work together to enable cost-effective, high-speed AI agents and enterprise workloads with new compute and orchestration capabilities.
Secure cross-cloud connectivity: Agent Gateway, Cloud Armor, and other tools deliver a secure, governed, and simplified networking foundation for AI agents, including observability of agentic traffic across clouds.
Unified data layer: Smart Storage, Knowledge Catalog, and other innovations transform passive data archives into dynamic reasoning engines, giving AI agents the context they need to execute.
Digital sovereignty: Confidential External Key Management and new features in Google Distributed Cloud bring Google’s leading models and AI enablers wherever your data lives.

Let’s take a closer look at all the news for each of these four areas.

Fluid compute

Agentic workloads are dynamic and unpredictable, impacting both traditional enterprise applications and the AI agents themselves. Fluid compute is enabled by Google Compute Engine and Google Kubernetes services working together to dynamically adapt and shift weight in real-time, enabling cost-effective, high-speed AI agents and operational enterprise workloads for all customers.

While our AI Hypercomputer delivers raw power for large-scale AI model training, fluid compute addresses the needs of operational workloads and agents. As agents move toward reasoning and reinforcement learning, CPUs are reclaiming a central role, excelling at the "branchy" logic, complex control flows, and secure execution sandboxes (like those for agentic orchestration, RL, SLM inference, and RAG) that agent workflows demand. CPUs also provide the critical isolation needed for secure agent execution, complementing the parallel processing strength of GPUs and TPUs used in training.

We are introducing new CPU families, GKE capabilities, and Hyperdisk block storage capabilities to run traditional workloads and AI agents securely at scale, including:

Google C4N Series: These VMs help ensure your enterprise workloads don't slow down under the demands of agentic AI by processing up to 95 million packets per second, up to 40% faster than other leading hyperscalers. This eliminates I/O bottlenecks for demanding workloads like security appliances, streaming media, and open source databases, even when utilizing smaller instance sizes.
Google M4N Series with Hyperdisk Extreme: M4N removes data pipeline bottlenecks and eliminates overprovisioning to deliver industry leading per-core IOPS and throughput required to handle massive data I/O from agents, analytics, and mission-critical databases. M4N provides 26.57 GB of RAM per vCPU, allowing you to scale mission-critical workloads cost-effectively on fewer cores. For example, M4N with Hyperdisk Extreme reduces Oracle workload total cost of ownership by over 20% compared to leading hyperscale clouds.
GKE Agent Sandbox: This solution secures agents with trusted gVisor isolation and handles demand spikes, launching up to 300 sandboxes per second, per cluster. Backed by the only managed sandbox technology available among leading hyperscale clouds, it achieves up to 30% better price-performance than competitors when running AI agents on GKE Agent Sandbox with Google Axion N4A.

“Wayfair's AI strategy is built on years of systematic infrastructure modernization on Google Cloud — migrating our core eCommerce engine and databases off legacy systems, decomposing monolithic services into cloud-native architecture, and unifying our data and analytics platform. That foundation is what makes everything else possible. Today, Gemini Enterprise Agent Platform is powering everything from catalog enrichment to generative shopping experiences that help customers create a home that's just right for them — and it's the same foundation preparing us for the agentic era, where AI doesn't just assist but actively drives discovery, personalization, and commerce across every customer touchpoint and across our business.” - Fiona Tan, Chief Technology Officer, Wayfair

Explore all our latest compute innovations in this blog.

Secure cross-cloud connectivity

Agentic AI replaces predictable human requests with autonomous “reasoning loops,” in which agents call other agents that, in turn, call LLMs, triggering massive, sudden surges in compute and machine-to-machine traffic. This shift creates unique challenges for network predictability and security of non-human identities. Optimized for agentic AI, our Cross-Cloud Network moves data across diverse environments, connecting employees, customers, and agents with visibility and security. New in Cross-Cloud Network are:

Agent Gateway: Governs and orchestrates your enterprise agentic traffic as the “air traffic controller” in Gemini Enterprise Agent Platform. It natively understands agent protocols like MCP and A2A to inspect and govern every agent interaction. By integrating with Google and third-party identity and AI safety services, it enables deep inspection to verify access, block attacks, and protect sensitive data, maintaining compliance across your core business.
Cloud Network Insights: Delivers broad visibility across your hybrid and multi-cloud infrastructure to drive faster troubleshooting and network resolutions. Continuously monitor your end-to-end agent, network and web performance across Google Cloud, AWS, Azure, data centers, internet applications, and agentic workloads. Using synthetic traffic analytics, Cloud Network Insights provides hop-by-hop network path visibility to help you pinpoint the source of degradations and is coupled with AI-powered insights from Gemini Cloud Assist to deliver more autonomous operations.
Enhanced Cloud Next Generation Firewall (NGFW) and Cloud Armor: Provides machine-speed, AI-powered protection to combat the rapid explosion of AI-generated polymorphic malware and zero-day exploits. Cloud NGFW advanced malware sandbox delivers real-time inline prevention of AI-generated threats, while Cloud Armor managed rules provides automated protection against both known and unknown Common Vulnerabilities and Exposures (CVEs). Together with Model Armor, these services analyze the intent and content of AI agent communications.

Discover more about how we optimized networking for AI in and outside of the data center.

Unified data layer

AI agents are only as powerful as the data they can access and the context they’re given. More applications and platforms are using structured and unstructured data, but it can be difficult to catalog, find, and act on that data at scale, leading to less effective agent interactions. To close the gap, your agents need all of your data brought together into a cohesive, queryable knowledge engine, or unified data layer. This way, your agents can identify and access accurate sources. At Next ‘26, we’re enhancing the unified data layer with:

Smart Storage: This solution transforms dark data into a powerful knowledge asset for AI agents and training by embedding new semantic intelligence directly into your data objects. With new Google Cloud Storage capabilities like automated annotation, entity extraction, and semantic search, your agents can instantly find and use the specific data they need — whether it's hidden in spreadsheets, PDFs, or other unstructured formats across your entire organization. This significantly speeds up the development and deployment of your AI solutions. Learn more about storage innovations to accelerate your AI workloads.
Knowledge Catalog: Knowledge Catalog maps business meaning across your entire data estate, providing a grounded source of truth so agents can deliver the most accurate results. This foundation enables AI training and inferencing and doesn’t require you to migrate your data; your agents interact with it directly, wherever it lives, with full context and governance, making modernization easier.

Part of our Agentic Data Cloud, Smart Storage and Knowledge Catalog can take your data from a passive archive into a dynamic reasoning engine.

“AI is critical to making our customers’ smart home and security solutions more intelligent and convenient. By leveraging Google Cloud’s Smart Storage, we auto-annotate rich metadata delivered in BigQuery. We’ve scaled and accelerated our data discovery and curation efforts, speeding up our AI development process from months to weeks, continuously delivering innovations that build trust and enhance the overall home experience.” - Brandon Bunker, VP of Product, AI, Vivint

Digital sovereignty

In the agentic era, digital sovereignty is a fundamental requirement for public sector and enterprise customers looking to accelerate innovation — without sacrificing control. There’s no one-size-fits-all solution, which is why we’ve designed a comprehensive set of offerings to meet different sovereign AI needs anywhere: public cloud, on-premises, or hybrid. New capabilities in our sovereign AI portfolio include:

Confidential External Key Management: Organizations can use Confidential External Key Management to maintain complete possession, custody, and control of your encryption keys and the policies that govern them. Confidential External Key Management leverages Confidential Compute to host the key management endpoint in a tamper-proof environment within Google Cloud. You are in control and determine where your keys are stored, who can access them, and under what circumstances. Even highly privileged Google administrators cannot access your keys without authorization, which you can revoke at any time. Your data, your control.
Gemini on Google Distributed Cloud: With Gemini on GDC, companies can securely deploy Gemini in sensitive environments, while meeting data sovereignty needs. Your choice of deployment models includes managed software on your connected hardware or a fully disconnected, air-gapped solution. You can now scale with Google's leading AI capabilities even in the most restricted, high-security environments — from powerful Gemini models to advanced coding, search, and other agentic capabilities.

In addition, Google Distributed Cloud supports an end-to-end AI stack, combining our latest-generation AI infrastructure with Gemini models to accelerate and enhance all your sovereign AI workloads. This stack includes:

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.
New VM families: New A4 family offerings provide the ability to handle the most demanding inference tasks, delivering a 2.25x increase in peak compute. Memory-Optimized M2 and M3 brings the high memory-to-vCPU ratios needed for massive ERP and data analytics workloads on-premises.
Enhanced storage: Eliminate storage bottlenecks with 6x storage capacity per zone and a 10x performance boost, giving you the ability to do AI reasoning on-premises. Now, your data infrastructure moves at the speed of AI reasoning.

"Our customers demand high-performance, private AI inference without the risks of multi-tenancy. Google Distributed Cloud allows us to provide dedicated, low-latency environments that meet strict sensitive data requirements. With the ability to run Gemini on B200s and B300s, we can significantly increase inference speeds and provide the token throughput our clients need to scale." - Dave Driggers, CEO & Co-founder, Cirrascale Cloud Services

Transforming vision into reality

When these product areas converge, your infrastructure evolves into a high-performing, secure, adaptive foundation for the agentic era. We're not just offering tools; we're providing the architectural blueprint to enable enterprises and the public sector to rapidly embrace the full power of AI and agents with confidence.

To learn more about key industry trends for AI Infrastructure, read our State of Infrastructure in the Agentic AI Era report.

New GKE Cloud Storage FUSE Profiles take the guesswork out of configuring AI storage

Wed, 08 Apr 2026 16:30:00 +0000

In the world of AI/ML, data is the fuel that drives training and inference workloads. For Google Kubernetes Engine (GKE) users, Cloud Storage FUSE provides high-performance, scalable access to data stored in Google Cloud Storage. However, we learned from customers that getting the maximum performance out of Cloud Storage FUSE can be complex.

Today, we are excited to introduce GKE Cloud Storage FUSE Profiles, a new feature designed to automate performance tuning and accelerate data access for your AI/ML workloads (training, checkpointing, or inference) with minimal operational overhead. With these profiles, tuned for your specific workload needs, you can enjoy high performance of Cloud Storage FUSE out of the box.

Before (manual tuning)

code_block: <ListValue: [StructValue([('code', 'apiVersion: v1\r\nkind: PersistentVolume\r\nmetadata:\r\n name: serving-bucket-pv\r\nspec:\r\n accessModes:\r\n - ReadWriteMany\r\n capacity:\r\n storage: 64Gi\r\n persistentVolumeReclaimPolicy: Retain\r\n storageClassName: ""\r\n claimRef:\r\n name: serving-bucket-pvc\r\n mountOptions:\r\n - implicit-dirs\r\n - metadata-cache:ttl-secs:-1\r\n - metadata-cache:stat-cache-max-size-mb:-1\r\n - metadata-cache:type-cache-max-size-mb:-1\r\n - file-cache:max-size-mb:-1\r\n - file-cache:cache-file-for-range-read:true\r\n - file-system:kernel-list-cache-ttl-secs:-1\r\n - file-cache:enable-parallel-downloads:true\r\n - read_ahead_kb=1024\r\n csi:\r\n driver: gcsfuse.csi.storage.gke.io\r\n volumeHandle: BUCKET_NAME\r\n volumeAttributes:\r\n skipCSIBucketAccessCheck: "true"\r\n gcsfuseMetadataPrefetchOnMount: "true"\r\n---\r\napiVersion: v1\r\nkind: PersistentVolumeClaim\r\nmetadata:\r\n name: serving-bucket-pvc\r\nspec:\r\n accessModes:\r\n - ReadWriteMany\r\n resources:\r\n requests:\r\n storage: 64Gi\r\n volumeName: serving-bucket-pv\r\n storageClassName: ""\r\n–--\r\napiVersion: v1\r\nkind: Pod\r\nmetadata:\r\n name: gcs-fuse-csi-example-pod\r\n annotations:\r\n gke-gcsfuse/volumes: "true"\r\nspec:\r\n containers:\r\n # Your workload container spec\r\n ...\r\n volumeMounts:\r\n - name: serving-bucket-vol\r\n mountPath: /serving-data\r\n readOnly: true\r\n serviceAccountName: KSA_NAME \r\n volumes:\r\n - name: gke-gcsfuse-cache # gcsfuse file cache backed by RAM Disk\r\n emptyDir:\r\n medium: Memory \r\n - name: serving-bucket-vol\r\n persistentVolumeClaim:\r\n claimName: serving-bucket-pvc'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7fb520eab450>)])]>

After (Cloud Storage FUSE mount options, CSI configs, and file cache medium automatically configured!)

code_block: <ListValue: [StructValue([('code', 'apiVersion: v1\r\nkind: PersistentVolume\r\nmetadata:\r\n name: serving-bucket-pv\r\nspec:\r\n accessModes:\r\n - ReadWriteMany\r\n capacity:\r\n storage: 64Gi\r\n persistentVolumeReclaimPolicy: Retain\r\n storageClassName: gcsfusecsi-serving\r\n claimRef:\r\n name: serving-bucket-pvc\r\n csi:\r\n driver: gcsfuse.csi.storage.gke.io\r\n volumeHandle: BUCKET_NAME\r\n---\r\napiVersion: v1\r\nkind: PersistentVolumeClaim\r\nmetadata:\r\n name: serving-bucket-pvc\r\nspec:\r\n accessModes:\r\n - ReadWriteMany\r\n resources:\r\n requests:\r\n storage: 64Gi\r\n volumeName: serving-bucket-pv\r\n storageClassName: gcsfusecsi-serving\r\n–--\r\napiVersion: v1\r\nkind: Pod\r\nmetadata:\r\n name: gcs-fuse-csi-example-pod\r\n annotations:\r\n gke-gcsfuse/volumes: "true"\r\nspec:\r\n containers:\r\n # Your workload container spec\r\n ...\r\n volumeMounts:\r\n - name: serving-bucket-vol\r\n mountPath: /serving-data\r\n readOnly: true\r\n serviceAccountName: KSA_NAME \r\n volumes: \r\n - name: serving-bucket-vol\r\n persistentVolumeClaim:\r\n claimName: serving-bucket-pvc'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7fb520eaa550>)])]>

The trouble with optimizing Cloud Storage FUSE

Optimizing Cloud Storage FUSE for high-performance workloads is a multi-dimensional problem. Historically, users had to navigate manual configuration guides that could span dozens of pages. And as AI/ML has evolved, Cloud Storage FUSE’s capabilities have also increased, with new mount options available to accelerate your workloads. The "right" settings were never static; they depended heavily on a variety of dynamic factors:

Bucket characteristics: The total size of your dataset and the number of objects significantly impact metadata and file cache requirements.
Infrastructure variability: Optimal configurations change based on whether you are using GPUs, TPUs, or general-purpose compute.
Node resources: Available RAM and Local SSD capacity determine how much data can be cached locally to minimize expensive round-trips to Cloud Storage.
Workload patterns: A training workload (high-throughput reads of large datasets) requires different tuning than a checkpointing workload (bursty, high-throughput writes) or a serving workload (latency-sensitive model loading).

In fact, many customers leave available performance on the table or face reliability issues (e.g., Pod Out-of-Memory kills) due to unoptimized or misconfigured Cloud Storage FUSE settings.

Introducing Cloud Storage FUSE Profiles for GKE

GKE Cloud Storage FUSE Profiles simplify this complexity with pre-defined, dynamically managed StorageClasses tailored for specific AI/ML patterns. Instead of manually adjusting dozens of mount options, you simply select a profile that matches your workload type.

These profiles operate on a layered model. They take the base best practices from Cloud Storage FUSE and add a GKE-specific intelligence layer. When you deploy a Pod using a profile, GKE automatically:

Scans your bucket (or a specific directory) to understand its size and object count.
Analyzes the target node to check for available RAM, Local SSD, and accelerator types.
Calculates optimal cache sizes and selects the best backing medium (RAM or Local SSD) automatically.

We are launching with three primary profiles:

gcsfusecsi-training: Optimized for high-throughput reads to keep GPUs and TPUs fed with data.
gcsfusecsi-serving: Optimized for model loading and inference, with automated Rapid Cache integration.
gcsfusecsi-checkpointing: Optimized for fast, reliable writes of large multi-gigabyte checkpoint files.

Using GKE Cloud Storage FUSE Profiles delivers several benefits:

Simplified tuning: Replace complex, error-prone manual configurations with three simple, purpose-built StorageClasses.
Dynamic, resource-aware optimization: The CSI driver automatically adjusts cache sizes based on real-time environment signals, so that you can maximize performance without risking node stability.
Accelerated read performance: The serving profile automatically triggers Rapid Cache, placing your data closer to your compute for faster cold-start model loading.
Granular performance insights: Gain visibility into automated tuning decisions through structured logs that detail exactly why specific cache sizes and mediums were selected for your Pod.

Using GKE Cloud Storage FUSE Profiles inference profile, we were able to reduce model loading time for a Qwen3-235B-A22B workload on TPUs (480GB) from 39 hours to just 14 minutes, helping customers achieve the maximum benefit of Cloud Storage FUSE GCSFuse out-of-the-box.

How to use Cloud Storage FUSE Profiles on GKE

To get started, ensure your cluster is running GKE version 1.35.1-gke.1616000 or later with the Cloud Storage FUSE CSI driver enabled.

1. Identify the StorageClass

GKE comes pre-installed with the profile-based StorageClasses. You can verify them with:

code_block: <ListValue: [StructValue([('code', 'kubectl get sc -l gke-gcsfuse/profile=true'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7fb520eab9d0>)])]>

2. Create your PV and PVC

When creating your PersistentVolume, point it to your Cloud Storage bucket. GKE automatically initiates a bucket scan to determine the optimal configuration.

code_block: <ListValue: [StructValue([('code', 'apiVersion: v1\r\nkind: PersistentVolume\r\nmetadata:\r\n name: gcs-pv\r\nspec:\r\n accessModes:\r\n - ReadWriteMany\r\n capacity:\r\n storage: 5Gi\r\n persistentVolumeReclaimPolicy: Retain \r\n storageClassName: gcsfusecsi-training\r\n mountOptions:\r\n - only-dir=my-ml-dataset-subdirectory # Optional\r\n csi:\r\n driver: gcsfuse.csi.storage.gke.io\r\n volumeHandle: my-ml-dataset-bucket\r\n---\r\napiVersion: v1\r\nkind: PersistentVolumeClaim\r\nmetadata:\r\n name: gcs-pvc\r\nspec:\r\n accessModes:\r\n - ReadWriteMany\r\n resources:\r\n requests:\r\n storage: 5Gi\r\n storageClassName: gcsfusecsi-training\r\n volumeName: gcs-pv'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7fb520eabd90>)])]>

3. Create your Deployment

Once your Persistent Volume Claim (PVC) is bound, simply consume it in your Deployment as you would any other volume. GKE mounts the volume with the precise settings your hardware and dataset require.

code_block: <ListValue: [StructValue([('code', 'apiVersion: apps/v1\r\nkind: Deployment\r\nmetadata:\r\n name: my-deployment\r\nspec:\r\n replicas: 3\r\n selector:\r\n matchLabels:\r\n app: my-app\r\n template:\r\n metadata:\r\n labels:\r\n app: my-app\r\n annotations:\r\n gke-gcsfuse/volumes: "true"\r\n spec:\r\n serviceAccountName: my-ksa\r\n containers:\r\n - name: my-container\r\n image: busybox\r\n volumeMounts:\r\n - name: my-gcs-volume\r\n mountPath: "/data"\r\n volumes:\r\n - name: my-gcs-volume\r\n persistentVolumeClaim:\r\n claimName: gcs-pvc'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7fb520ea92d0>)])]>

After it's deployed, the CSI driver automatically calculates optimal cache sizes and mount options based on your node's resources, such as GPUs or TPUs, memory, Local SSD, the bucket or sub-directory size, and the sidecar resource limits.

Get started today

GKE Cloud Storage FUSE Profiles remove the guesswork from configuring your cloud storage for high performance. By moving from manual "knob-turning" to automated, workload-aware profiles, you can spend less time debugging storage throughput and more time building the next generation of AI.

Ready to get started? GKE Cloud Storage FUSE Profiles are generally available in version 1.35.1-gke.1616000. Explore the official documentation to configure Cloud Storage FUSE profiles in GKE for your AI/ML workloads!

Accelerate model downloads on GKE with NVIDIA Run:ai Model Streamer

Thu, 04 Dec 2025 17:00:00 +0000

As large language models (LLMs) continue to grow in size and complexity, the time it takes to load them from storage to accelerator memory for inference can become a significant bottleneck. This "cold start" problem isn't just a minor delay — it's a critical barrier to building resilient, scalable, and cost-effective AI services. Every minute spent loading a model is a minute a GPU is sitting idle, a minute your service is delayed from scaling to meet demand, and a minute a user request is waiting.

Google Cloud and NVIDIA are committed to removing these barriers. We’re excited to highlight a powerful, open-source collaboration that helps AI developers do just that: the NVIDIA Run:ai Model Streamer now comes with native Google Cloud Storage support, supercharging vLLM inference workloads on Google Kubernetes Engine (GKE). Accessing data for AI/ML from Cloud Storage on GKE has never been faster!

The chart above shows how quickly the model streamer can fetch a 141GB Llama 3.3-7 70B model from Cloud Storage as compared to the default vLLM model loader (lower is better).

Boost resilience and scalability with fewer cold starts

For an inference server running on Kubernetes, a "cold start" involves several steps: pulling the container image, starting the process, and — most time-consuming of all — loading the model weights into GPU memory. For large models, this loading phase can take many minutes, with painful consequences such as slow auto-scaling and idling GPUs as they wait for the workload to start up.

By streaming the model into GPU memory, the model streamer slashes potentially the most time-consuming part of the startup process. Instead of waiting for an entire model to be downloaded before loading, the streamer fetches model tensors directly from object storage and streams them concurrently to GPU memory. This dramatically reduces model loading times from minutes to seconds.

For workloads that rely on model parallelism— where a single model is partitioned and executed across multiple GPUs— the model streamer goes a step further. Its distributed streaming capability is optimized to take full advantage of NVIDIA NVLink, using high-bandwidth GPU-to-GPU communication to coordinate loading across multiple processes. Reading the weights from storage is divided efficiently and evenly across all participating processes, with each one fetching a portion of the model weights from storage and then sharing its segment with the others over NVLink. This allows even multi-GPU deployments to benefit from faster startups and fewer cold-start bottlenecks.

Performance and simplicity

The latest updates to the Model Streamer introduce first-class support for Cloud Storage, creating an integrated and high-performance experience for Google Cloud users. This integration is designed to be simple, fast, and secure, especially for workloads running on GKE.

For users of popular inference servers like vLLM, enabling the streamer is as simple as adding a single flag to your vLLM command line:

--load-format=runai_streamer

Here’s how easy it is to launch a model stored in a Cloud Storage bucket with vLLM:

code_block: <ListValue: [StructValue([('code', 'vllm serve gs://your-gcs-bucket/path/to/your/model \r\n--load-format=runai_streamer'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7fb51969cf90>)])]>

The NVIDIA Run:ai Model Streamer is a key component for Vertex AI Model Garden's large model deployments. With container image streaming and model weight streaming, we have been able to significantly improve the first deployment and autoscaling experience for our users, and the efficiency of NVIDIA GPUs.

When running on GKE, the Model Streamer can automatically use the cluster's Workload Identity. This means you no longer need to manually manage and mount service account keys, simplifying your deployment manifests and enhancing your security posture. The following deployment manifest shows how to launch a container serving Llama3 70B on GKE. We have added the model loader distributed option to accelerate loads when model parallelism > 1:

code_block: <ListValue: [StructValue([('code', 'apiVersion: apps/v1\r\nkind: Deployment\r\n…\r\n spec:\r\n serviceAccountName: gcs-access\r\n containers:\r\n - args:\r\n - --model=gs://your-gcs-bucket/path/to/your/model \r\n - --load-format=runai_streamer\r\n \t\t- --model-loader-extra-config={"distributed":true}\r\n\t\t…\r\n command:\r\n - python3\r\n - -m\r\n - vllm.entrypoints.openai.api_server\r\n image: vllm/vllm-openai:latest\r\n ….'), ('language', ''), ('caption', <wagtail.rich_text.RichText object at 0x7fb519f26210>)])]>

That’s it! The streamer handles the rest, auto-tuning streaming concurrency to match your VM’s performance. For more details, see the documentation on optimizing vLLM model loading on GKE.

Combining NVIDIA Run:ai Model Streamer with Cloud Storage Anywhere Cache

Anywhere Cache provides zonally co-located SSD-backed caching for data stored in a regional or multi-regional Cloud Storage bucket. Reducing latency by up to 70% and providing up to 2.5 TB/s of read throughput, Anywhere Cache is a great solution for scale-out inference workloads where the same model is downloaded multiple times across a series of nodes. Together, Anywhere Cache server-side acceleration, along with the NVIDIA Run:ai Model Streamer’s client-side acceleration, create an easy-to-manage, extremely performant model-loading system.

Get started today

The NVIDIA Run:ai Model Streamer is evolving into a critical piece of the AI infrastructure puzzle, enabling teams to build faster, more resilient, and more flexible MLOps pipelines on GKE.

To learn more about how to use the model streamer on GKE see our GKE NVIDIA Run:ai Guide.
For detailed instructions on using the streamer with vLLM, see the official vLLM documentation.
To learn more and contribute to the model streamers ongoing development check out the NVIDIA Run:ai Model Streamer project on GitHub.

Reducing TCO for AI inferencing with external KV Cache on Managed Lustre

Fri, 31 Oct 2025 16:00:00 +0000

The demand for AI inference infrastructure is accelerating, with market spend expected to soon surpass investment in training the models themselves. This growth is driven by the demand for richer experiences, particularly through support for larger context windows and the rise of agentic AI. As organizations aim to improve user experience while optimizing costs, efficient management of inference resources is paramount.

According to an internal experimental study of large model inferencing, external key-value caches — KV Cache or, “attention caches” — on high-performance storage like Google Cloud Managed Lustre, can reduce total cost of ownership (TCO) by up to 35%, allowing organizations to serve the same workload with ~40% fewer GPUs by offloading prefill compute to I/O. In this blog, we explore the core challenges of managing long-context AI inference and detail how Google Cloud Managed Lustre provides the high-performance external storage solution required to achieve these significant cost and efficiency benefits.

About KV Cache

During the inference phase, a KV Cache is a critical optimization technique for the efficient operation of Transformer-based large language models (LLMs).

The key innovation of the Transformer was the complete elimination of sequential processing (recurrence), which was achieved by introducing the self-attention mechanism to allow every element in a sequence to instantaneously and dynamically compare itself to and assess the relevance of every other element (a global, all-at-once evaluation). Within this self-attention mechanism, the model computes Key (K) and Value (V) vectors of all preceding tokens in the sequence. To generate the next token during the inference phase, the model needs the K and V vectors of all the previous tokens.

This is where the KV Cache comes into play. The KV Cache stores these K and V vectors after the initial context processing (known as the "prefill" stage), thereby avoiding the redundant, costly re-computation of the context sequence when generating subsequent tokens. By eliminating this re-computation, the KV Cache vastly speeds up the overall inference process. While smaller caches can fit in high-bandwidth memory (HBM) or host DRAM — up to a few TBs of memory may be available in a single multi-accelerator server — managing a KV Cache for contexts across multiple concurrent users that exceed the memory capacity often requires external or hierarchical storage solutions.

These large contexts can make the "prefill" computation — the calculation that an AI model performs when processing a large context window — very expensive:

For a large context of 100K or more tokens, the prefill computation may cause the time to first token (TTFT) to increase to tens of seconds.
Prefill computation requires a high number of floating-point operations (FLOPs). KV Cache reuse saves these costs and makes additional resources available on the accelerator.

The growth of agentic AI is likely to make the challenge of managing a long context even greater. Unlike a simple chatbot, agentic AI is built for action. It moves beyond conversation to solve problems proactively, completing tasks on your behalf. To do this, it actively gathers context from a wide range of digital sources. Agentic AI may, for example: check live flight data, pull a customer's history from a database, research topics on the web, and/or keep organized notes in its own files. Agentic AI thereby builds a rich understanding of its environment, but often increases context lengths and their associated KV Cache size.

The key to managing performance costs at scale is to ensure that the accelerator is utilized as fully as possible. High-performance, scale-out storage provides the required greater throughput per accelerator and therefore translates into lighter resource requirements.

External KV Cache on Google Cloud Managed Lustre

We believe that Google Cloud Managed Lustre should be your primary storage solution for external KV Cache. On GPUs, Lustre is assisted by locally attached SSDs. And on TPUs, where local SSDs are not available, Lustre’s role is even more central.

A recent LMCache blog post by Google’s Danna Wang, “LMCache on Google Kubernetes Engine: Boosting LLM Inference Performance with KV Cache on Tiered Storage,” demonstrates the foundational value of host-level offloading. Our Managed Lustre strategy is the next evolution of this host-offloading concept. While Local SSDs and CPU RAM are effective node-local tiers, they are fixed in size and cannot be shared. Managed Lustre provides a parallel file system to act as the massive, high-throughput external storage, making it a great solution for large-scale, multi-node, and multi-tenant AI inference workloads where the cache exceeds the capacity of the host machine.

Here’s an example of how the performance gains of Managed Lustre can reduce your TCO:

In an experiment with a 50K token context and a high cache hit rate (about 75%), using Managed Lustre improved total inference throughput by 75% and reduced MTTF by more than 40% compared to using KV Cache in host memory alone (further detail below).
TCO analysis yielded a 35% savings from using an external attention/KV Cache for a workload processing 1 million Tokens per Second (TPS) and leveraging A3-Ultra VMs and Managed Lustre, when compared to a workload leveraging no external storage.
Our experiment demonstrated that with configuration tuning and an improvement in KV Cache software to adopt more I/O parallelism, Managed Lustre can substantially improve inference performance.

Total Cost of Ownership: Analysis

When evaluating a KV Cache solution, it's critical to consider the TCO, which includes not just compute and storage costs but also operational expenses and potential savings. Our analysis shows that a high performance storage-backed KV Cache, like one built on Managed Lustre, provides a compelling TCO advantage compared to purely memory-based solutions.

Cost savings

After taking incremental storage costs into account, we project that the TCO for a file-system-backed KV Cache solution, processing 1m TPS, is 35% lower compared to a memory-only solution. This makes it a more scalable and economically viable option for large-scale AI inference deployments.

The primary TCO benefit comes from a more efficient utilization of expensive compute resources. By offloading KV Cache to a high-performance storage solution, you can achieve a higher inference throughput per accelerator. This means that fewer accelerators are needed for the same workload: You can handle a specific number of queries per second with ~40% fewer accelerators, resulting in direct cost savings.

TCO model assumptions

The TCO calculation includes several key components:

Storage costs (list price): These are the costs of Managed Lustre. Testing used the 1000 MB/s per TiB Performance Tier. The TCO model includes sufficient Lustre capacity (73 A3-Ultra machines, with 18 TiB Lustre capacity per machine) to hit the 1m TPS target rate.
Compute costs (list price): A3-Ultra VMs each with 8x H200s GPUs and 8x 141 GB HBM (spot prices will be lower).

Performance benchmarks

Our experiments demonstrated Google Cloud Managed Lustre’s ability to deliver the high-performance I/O necessary with a state-of-the-art LLM. These experiments serve Deepseek-R1 on a Google Cloud A3-Ultra machine (8x H200s; 8x 141GB HBMs). The experiments ran a synthetic serving workload with a 50K token context and a high cache-hit rate (about 75% hit rate) with a total KV Cache size of about 3.4TiB. The memory-only baseline uses 1 TiB host memory for KV Cache. We experimented with two variants of Managed Lustre at high and low I/O parallelism. For high I/O parallelism, we utilized 32 I/O worker threads to read KV Cache data from Lustre in parallel.

Lustre improved total inference throughput by 75% and reduced the mean time to first token by greater than 40% compared to using KV Cache in host memory alone.

Ready to optimize your inference workloads?

To get started with an external KV Cache solution that solves the capacity limits of long context windows and delivers significant performance gains on your large-scale LLMs, follow these steps:

1. Provision your infrastructure; create a Managed Lustre instance:

Provision your Lustre file system in the same region and zone as your target accelerators (GPUs or TPUs) for optimal low-latency access.
Deploy your inference engine: Deploy your LLM using a high-performance inference server like vLLM or a similar framework that supports an external KV Cache or paged-attention architecture.

2. Configure for performance

Once you’ve mounted Managed Lustre, you must configure your inference engine software to leverage the high-performance storage:

Implement direct I/O: Configure your application to access Managed Lustre using the o_direct flag. This bypasses the general-purpose file system cache, allowing the inference engine to manage the critical host memory more effectively.
Tune I/O parallelism: Depending on your inference KV Cache software, its out-of-the-box storage I/O parallelism may not be ideal. You may need to tune the KV Cache software to read KV chunk files with enhanced parallelism to maximize performance.

To take the next step, read the documentation about how to get started with Managed Lustre.

From dark data to bright insights: The dawn of smart storage

Tue, 14 Oct 2025 16:00:00 +0000

Organizations interested in AI today have access to amazing computational power with Tensor Processing Units (TPUs) and Graphical Processing Units (GPUs), while foundational models like Gemini are redefining what's possible. Yet for many enterprises a critical obstacle to AI is the data itself, specifically unstructured data. According to Enterprise Strategy Group, for most organizations, 61% of their total data is unstructured, the vast majority of which sits unanalyzed and unlabeled in archives, so-called "dark data." But with the help of AI, this untapped resource is an opportunity to unlock a veritable treasure trove of insights.

At the same time, when it comes to unstructured data, traditional tools only scratch the surface, and subject matter experts must build massive, manual preprocessing pipelines and define the data’s semantic meaning. This prevents any real analysis at scale, preventing companies from using even a fraction of what they store.

Now imagine a world where your unstructured data isn't just stored, but understood. A world where you can ask complex questions of data such as images, videos, and documents, and get interesting answers in return. This isn't just a futuristic vision — the era of smart storage is upon us. Today we are announcing new auto annotate and object contexts features that use AI to generate metadata and insights on your data, so you can then use your dark data for discovery, curation, and governance at scale. Better yet, the new features relieve you from having to build and manage your own object-analysis data pipelines.

Leveraging AI to transform dark data

Now, as unstructured data lands in Google Cloud, it's no longer treated as a passive object. Instead, a data pipeline leverages AI to automatically process and understand the data, surfacing key insights and connections. Two new features are integral to this vision: auto annotate, which enriches your data by automatically generating metadata using Google’s pretrained AI models, and object contexts, which lets you attach custom, actionable tags to your data. Together, these two features can help transform passive data into active assets, unlocking use cases such as rapid data discovery for AI model training, streamlined data curation to reduce model bias, enhanced data governance to protect sensitive information, and the ability to build powerful, stateful workflows directly on your storage.

Making your data smart

Auto annotate, currently in a limited experimental release, automatically generates rich metadata (“annotations”) about objects stored in Cloud Storage buckets by applying Google's advanced AI models, starting with image objects. Getting started is simple: enable auto annotate for your selected buckets or an entire project, pick one or more available models, and your entire image library will be annotated. Furthermore, new images are automatically annotated as they are uploaded. An annotation’s lifecycle is always tied to its object’s, simplifying management and helping to ensure consistency. Importantly, auto annotate operates under your control, only accessing object content to which you have explicitly granted permissions. Then, you can query the annotations, which are available as object contexts, through Cloud Storage API calls and Storage Insights datasets. The initial release uses pretrained models for generating annotations: object detection with confidence scores, image labeling, and objectionable content detection.

a sample of generated annotations for an object

Then, with object contexts, you can attach custom key-value pair metadata directly to objects in Cloud Storage, including information generated by the new auto annotate feature. Currently in preview, object contexts are natively integrated with Cloud Storage APIs for listing and batch operations, as well as Storage Insights datasets for analysis in BigQuery. Each context includes object creation and modification timestamps, providing valuable lineage information. You can use Identity and Access Management (IAM) permissions to control who can add, change, or remove object contexts. When migrating data from Amazon S3 using Cloud Storage APIs, existing S3 Object Tags are automatically converted into contexts.

In short, object contexts provide a flexible and native way to add context to enrich your data. Combined with a smart storage feature like auto annotations, object contexts convert data into information, letting you build sophisticated data management workflows directly within Cloud Storage.

Now, let’s take a deeper look at some of the new use cases these smart storage features deliver.

1. Data discovery

One of the most significant challenges in building new AI applications is data discovery — how to find the most relevant data across an enterprise's vast and often siloed data stores. Locating specific images or information within petabytes of unstructured data can feel impossible. Auto annotate automatically generates rich, descriptive annotations for your data in Cloud Storage. Annotations, including labels and detected objects, are available within object contexts and fully indexed in BigQuery. After generating embeddings for them, you can then use BigQuery to run a semantic search for these annotations, effectively solving the "needle in a haystack" problem. For example, a large retailer with millions of product images can use auto annotate and BigQuery to quickly find 'red dresses' or 'leather sofas', accelerating catalog management and marketing efforts.

2. Data curation for AI

Building effective AI models requires carefully curated datasets. Sifting through data to ensure it is widely representative (e.g., "does this dataset have cars in multiple colors?") to reduce model bias, or to select specific training examples (e.g., “Find images with red cars”), is both time-consuming and error-prone. Auto annotate can identify attributes like colors and object types, to automate selecting balanced datasets.

For instance, an autonomous vehicle company training models could use petabytes of on-road camera data to recognize traffic signs, using auto annotate to identify and extract images that contain the word ‘Stop’ or 'Pedestrian Crossing'.

Vivint, a smart home and security company, has been using auto annotate to find and understand their data.

“Our customers trust us to help make their homes and lives safer, smarter, and more convenient, and AI is at the heart of our product and customer experience innovations. Cloud Storage auto annotate’s rich metadata delivered in BigQuery helps us scale our data discovery and curation efforts, speeding up our AI development process from 6 months to as little as 1 month by finding the needle-in-a-haystack data essential to improve our models.” - Brandon Bunker, VP of Product, AI, Vivint

3. Governing unstructured data at scale

Unstructured data is constantly growing, and manually managing and governing that data to identify sensitive information, detect policy violations, or categorize it for lifecycle management is a challenge. Auto annotate and object contexts help solve these data governance and compliance challenges. For example, a retail customer can use auto annotate to identify and flag images containing visible customer personally identifiable information (PII) such as shipping labels or order forms. This information, stored in object context, can then trigger automated governance actions such as moving flagged objects to a restricted bucket or initiating a review process.

BigID, a partner building solutions on Cloud Storage, reports that using object contexts is helping them manage their customers’ risk:

“Object contexts gives us a way to take the outputs of BigID's industry-leading data classification solutions and apply labels to Cloud Storage objects. Object contexts will allow BigID labels to shed light onto data in Cloud Storage: identifying objects which contain sensitive information and helping them understand and manage their risk across AI, security, and privacy." - Marc Hebrard, Principal Technical Architect, BigID

The future is bright for your data

At Google Cloud, we’re committed to building a future where your data is not just a passive asset but an active catalyst for innovation. Don't keep your valuable data in the dark. Bring your data to Cloud Storage and enable auto annotation and object contexts to unlock its full potential with Gemini, Vertex AI, and BigQuery.

You can start using object contexts today, and reach out to us for an early look at auto annotate. Once you have access, simply enable auto annotate for selected buckets or on an entire project, pick one or more available models, and your entire image library will be annotated. You can then query the annotations that are available as object contexts through Cloud Storage API calls and Storage Insights datasets.

To learn more, read about our end-to-end vision in a showcase paper with Enterprise Strategy Group: Illuminating Dark Data With Smart Storage from Google Cloud.

Power your enterprise applications in the cloud with unified block and file storage

Tue, 14 Oct 2025 16:00:00 +0000

Migrating enterprise applications to the cloud requires a storage foundation that can handle everything from high-performance block workloads to globally distributed file access. To solve these challenges, we’re thrilled to announce two new capabilities for Google Cloud NetApp Volumes: unified iSCSI block and file storage to enable your storage area network (SAN) migrations, and NetApp FlexCache to accelerate your hybrid cloud workloads. These features, along with a new integration for agents built with Gemini Enterprise, can help you modernize even your most demanding applications.

Run your most demanding SAN workloads on Google Cloud

For decades, enterprises have relied on NetApp for both network attached storage (NAS) and SAN workloads on-premises. We’re now bringing that same trusted technology to a fully managed cloud service, allowing you to migrate latency-sensitive applications to Google Cloud without changing their underlying architecture.

Our unified service is engineered for enterprise-grade performance, with features including:

Low latency engineered for your most demanding applications
Throughput that can burst up to 5 GiB/s with up to 160K random IOPS per volume
Independent scaling of capacity, throughput, and IOPS to control costs
Integrated data protection with NetApp Snapshots for rapid recovery and ransomware defense

iSCSI block protocol support is available now via private preview for interested customers.

Accelerate your hybrid cloud with NetApp FlexCache

For organizations with distributed teams and a hybrid cloud strategy, providing fast access to shared datasets is critical. NetApp FlexCache, a new capability for Google Cloud NetApp Volumes, provides high-performance, local read caches of remote volumes. This helps distributed teams access shared datasets as if they were local, and supports compute bursting for workloads that need low-latency data access, improving productivity and collaboration across your entire organization. FlexCache is available now in preview via an allowlist.

aside_block: <ListValue: [StructValue([('title', 'Try Google Cloud for free'), ('body', <wagtail.rich_text.RichText object at 0x7fb520ea9250>), ('btn_text', ''), ('href', ''), ('image', None)])]>

Bring your enterprise data to Gemini Enterprise

We’re also announcing that Google Cloud NetApp Volumes now serves as a data store for Gemini Enterprise. This integration unlocks new possibilities for retrieval-augmented generation (RAG), allowing you to ground your AI models on your own secure, factual, enterprise-grade data. Your data remains securely governed in NetApp Volumes and is quickly available for search and inference workflows, without the need for complex ETL or manual integrations.

Additional enhancements for your cloud environment

Google Cloud NetApp Volumes has several other new capabilities to help you modernize your data estate:

NetApp SnapMirror: You can now quickly replicate mission-critical data between on-prem NetApp systems and Google Cloud, providing a zero recovery point objective (RPO) and near-zero recovery time objective (RTO).
High-performance for large volumes: For applications with massive datasets such as HPC, AI, and EDA, we now offer large-capacity volumes that scale from 15TiB to 3PiB, with over 21GiB/s of throughput per volume.
Auto-tiering: To help you manage costs, built-in auto-tiering dynamically moves infrequently accessed data to lower-cost storage, with cold data priced at just $0.03/GiB for the Flex service level. As a turnkey, integrated feature, auto-tiering is transparent to any application built on Google Cloud NetApp Volumes, and can support a tiering threshold of anywhere from 2-183 days, with dynamically adjustable policy support.

Get started

Whether you’re migrating your enterprise SAN data, powering AI with Gemini Enterprise, or running high-throughput EDA workloads, Google Cloud NetApp Volumes can help you modernize your data estate. To learn more and get started, explore the product documentation.