Google Agent Gateway for Cross-Cloud Opens in Limited Preview: What It Governs on EKS and AKS and What It Costs

Published:

Google Cloud has opened a Limited Preview of Agent Gateway for cross-cloud, a way to apply the same governance rules to AI agents running on Amazon EKS, Azure AKS, Red Hat OpenShift or bare-metal Kubernetes as to agents on Google’s own platform. Google’s release notes list it under November 12, 2026. It is free during the preview and gated behind an access request, and Google’s documentation labels it a Pre-GA offering that “might have limited support”. An analyst firm that reviewed the plan before launch says the key question, whether the same policy gives the same result on a rival cloud, has not been shown yet.

What is Agent Gateway for cross-cloud?

Per Google’s documentation, the feature extends Gemini Enterprise Agent Platform governance to CNCF-conformant Kubernetes clusters outside Google Kubernetes Engine. The examples named are on-premises OpenShift, bare-metal clusters, Amazon EKS and Azure AKS. Policies from Model Armor and Identity-Aware Proxy (IAP) are defined centrally in Google Cloud, and a proxy in the customer’s cluster enforces them on each request. AI Weekly’s summary of Google’s notes adds that each agent gets a verifiable identity, agents and MCP servers are registered in Agent Registry, and LLM, MCP and agent-to-agent traffic is routed through local proxies.

How does it work, and who runs what?

The documentation describes three layers: a Google-hosted management plane (Agent Registry, Model Armor, IAP, Workload Identity Federation and Google Cloud Observability), a control plane with an in-cluster controller that syncs services and workloads with Google Cloud, and Envoy proxies that the customer runs in its own Kubernetes namespaces, each with an identity sidecar.

  • Google manages: the management plane, the cross-cloud routing configuration and the standard Helm charts for the proxies.
  • The customer handles: forwarding traffic into the proxies through its own service mesh or network, injecting Kubernetes service account tokens for identity federation, allowing outbound connections from in-cluster components to Google Cloud APIs, and protecting its Google Cloud project and registered cluster resources from accidental deletion.

What does it cost and who can use it?

The gateway is free during the Limited Preview. Products used alongside it, including Model Armor, Cloud Logging and Cloud Trace, are billed at standard rates. Access is by request form linked from the documentation, so it is not open to every customer. Google has not published pricing for after the preview, and the analyst firm Futurum noted before launch that cross-cloud pricing was unpublished.

What are the known limits?

Google lists three known issues in the documentation:

  • The failOpen: true setting on extension bindings does not apply if the proxy cannot reach Google Cloud, or if a binding points to a Model Armor template that does not exist.
  • A gateway can report ready up to two minutes before it accepts connections on port 8443.
  • In-cluster service URLs must be unique across all clusters registered to the same project.

Futurum’s October 30 analysis, written before the preview opened, raises a bigger question: Google has not shown that one policy produces the same decision on a runtime it does not own, or explained how enforcement behaves on a rival’s cloud if the management plane is unreachable. Futurum advises buyers to treat that parity as a purchase gate and to keep registries and policy models portable. That is an analyst opinion, not a Google statement.

Why does it matter for businesses?

Companies that run AI agents on more than one cloud have had to apply access rules separately on each. A single policy layer is the promise here, and the preview is the first chance to test it. For Google’s latest model pricing, see our report on Gemini 4 Hydrogen. For the cloud background, our AWS, Azure and Google Cloud comparison covers market share and pricing, and ChatGPT Enterprise on AWS Marketplace shows another route to enterprise AI on AWS.

Sources: Google Cloud release notes; Google Cloud, Agent Gateway for cross-cloud overview; AI Weekly (dates the launch November 11); Futurum Group. All read on November 16, 2026.

Like our coverage? Add stepphase.com as a preferred source on Google to see more of our stories in Top Stories.

Saif Ali Tai
Saif Ali Taihttps://xplorely.com/
I'm Saif Ali Tai, a tech blogger and software engineer with 8+ years of experience building apps, and I run StepPhase as its editor. I cover technology news, AI tools and security stories that matter to everyday users. Articles here are drafted with AI assistance and checked against the sources linked in each post.

Related articles

Recent articles