Google Cloud BlogAlex Zakonov6 min readintermediate
Agent Substrate brings high-density, scalable, trusted infrastructure to GKE
Summary
Agent Substrate is an open‑source runtime for AI agents that runs on GKE. It uses Cloud Hypervisor microVMs or gVisor sandboxes to give kernel‑level isolation, a custom control‑ and data‑plane that can suspend/resume agents in <500 ms, and a “zero‑idle” model that packs >1 000 dormant agents per host (≈10× density vs. containers). GKE integration adds custom ComputeClasses, spot/on‑demand pools,…
- Secure‑by‑default isolation via hardware‑isolated microVMs or gVisor, plus an egress/ingress gateway for fine‑grained network policies.
- Sub‑second activation: pre‑warmed workers enable <500 ms resume and hundreds of suspend/resume cycles per second.
- Active‑only compute economics: idle agents are snapshotted to local disk/Cloud Storage, freeing RAM/CPU; >1 000 dormant agents per host → ~10× higher compute density.
- Kubernetes remains the control plane for node health, autoscaling, and multi‑zone scheduling; the agent‑native data plane handles fast suspend/resume.
Running large fleets of autonomous AI agents requires both strong isolation (to prevent code‑escape) and efficient resource use (agents spend most of their time idle). Agent Substrate’s snapshot‑based suspend/resume and kernel‑level sandboxing aim to make million‑agent scale feasible on standard Ku…
5/10




