GKE Agent Sandbox provides the industry’s most scalable and low-latency infrastructure for agentic AI. Built with gVisor kernel isolation—the same technology securing Gemini—it allows you to safely execute untrusted code and tool calls without sacrificing performance. Spin up to 300 isolated sandboxes per second and confidently deploy secure agent infrastructure that keeps up with your business.

With support for up to 65,000-node clusters, integration with AI Hypercomputer, and GPU and TPU support, GKE makes it easy to run ML, HPC, and other workloads that benefit from specialized hardware accelerators.

GKE inference capabilities with gen AI-aware scaling and load balancing techniques help to reduce serving costs by over 30%, tail latency by 60%, and increase throughput by up to 40% compared to other managed and open source Kubernetes offerings.

GKE runs on a secure-by-design foundation, now enhanced with always-on essential security enabled by default. Get instant visibility into cluster misconfigurations, risks, and agentless scanning for critical vulnerabilities directly in the GKE security dashboard. Protect data-in-use with hardware-based encryption on Confidential GKE Nodes and isolate workloads with GKE Sandbox. Use built-in IaC scanning to proactively detect misconfigurations in your Terraform plans before they reach production.

Fleets and Teams can be used to organize clusters and workloads and assign resources to multiple teams easily to improve velocity and delegate ownership. Team scopes let you define subsets of fleet resources on a per-team basis, with each scope associated with one or more fleet member clusters.

You might choose multiple clusters to separate services across environments, tiers, locales, teams, or infrastructure providers. Fleets strive to make managing multiple clusters as easy as possible.