Google Cloud Adds Isolated MicroVMs for AI Agents, Preview Coming

Google Cloud announced Agent Compute Engine on Thursday, Oct. 22, a hardware-isolated microVM service already used for its Gemini work agent, with a customer preview planned in the coming months.

The service is designed for custom agents that carry work across multiple steps. Those agents can write and run code, use tools and browsers, and continue tasks over time. Their workloads may pause and resume repeatedly, so Agent Compute Engine can suspend idle microVMs and restore them later. It combines copy-on-write memory sharing with checkpoint-and-restore to handle that cycle.

Each microVM has its own guest kernel and root access inside a hardware-isolation boundary built with Cloud Hypervisor. The design also uses Google Cloud’s Titanium infrastructure processing unit (IPU), while service credentials stay outside the microVM.

Google announced Agent Substrate alongside the managed service, but it is a separate open-source runtime for Google Kubernetes Engine (GKE). Agent Substrate carries a claim of up to 10x higher compute density than traditional compute. A benchmark involving 200,000 agents covered 500 activations per second and suspend-and-resume latency below 500 milliseconds. Those figures apply to Agent Substrate, not Agent Compute Engine.

Agent Substrate can run on Kubernetes clusters and supports agent frameworks including Claude Code, OpenClaw and Hermes. The two products offer different ways to run agents: Agent Substrate is open source, while Agent Compute Engine is the managed service slated for customer preview.

Google Cloud has not announced pricing or a firm calendar date for the Agent Compute Engine preview.