Cross-cloud infrastructure at Next ‘26

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:

1. 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. 
2. 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. 
3. 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.
4. 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.