ReleaseBytes
gcp Google Cloud Blog ·

Agent Substrate brings high-density, scalable AI agent infrastructure to GKE

blogaigcpgaarchitectgcp-gkegcp-cloud-storage
feature announcement

Google has announced the general availability of Agent Substrate on Google Kubernetes Engine (GKE), an open-source agent execution runtime designed for large-scale autonomous AI agents. It offers 10x higher density than standard container runtimes by using hardware-isolated microVMs or gVisor sandboxes and a zero-trust kernel with network isolation. The solution delivers sub-500ms resume operations and supports massive bursts while optimizing for compute efficiency. Developers leveraging GKE can now build highly secure and scalable agent platforms, integrating with services like Google Cloud Storage and Filestore.

  • →Agent Substrate for GKE Generally Available
  • →Addressing Core Challenges in AI Agent Scalability
  • →Secure by Default with High Efficiency and Fast Activation
  • →Optimized for Google Cloud Infrastructure
Features (1) ›
  • Agent Substrate for GKE Generally Available

    Agent Substrate is now generally available on Google Kubernetes Engine, providing an open-source, secure-by-default execution runtime for autonomous AI agents. It is engineered to run millions of sandboxes with 10x higher density than standard container runtimes, addressing the unique infrastructure challenges of scaling agent platforms.

Enhancements (2) ›
  • Secure by Default with High Efficiency and Fast Activation

    Agent Substrate enforces security through hardware-isolated Cloud Hypervisor microVMs or gVisor sandboxes, paired with egress proxies for granular network control. It features a dedicated control plane and data plane for sub-500ms resume operations at high rates, and optimizes compute economics by suspending idle actors to free up resources.

  • Optimized for Google Cloud Infrastructure

    The platform leverages GKE for node management, autoscaling, and reliability, while handling high-frequency suspend-resume locally. It maximizes machine obtainability and flexibility on Google Cloud with custom ComputeClasses, native support for Google Axion processors, and optional integration with Filestore agent volumes for stateful workspaces.

Notes (1) ›
  • Addressing Core Challenges in AI Agent Scalability

    The platform tackles issues such as opaque trust boundaries when running untrusted code, friction in tool access, managing massive bursts of agent sandboxes, and wasted compute resources from idle agents. It decouples agent execution from machine management, offering a purpose-built execution layer for agentic workloads.

Read the original announcement →

https://cloud.google.com/blog/products/containers-kubernetes/agent-substrate-available-on-gke/

Related releases