Google Cloud introduces Agent Compute Engine

Google Cloud announced Agent Compute Engine, a compute service built specifically for running AI agents, at the AGNTCon plus MCPCon event. The service already powers Gemini's own agent features internally and will be available to customers in preview in the coming months. It gives each agent a hardware-isolated microVM rather than a shared container or a traditional virtual machine.

Google says the microVMs boot quickly, use copy-on-write memory sharing so many agents can run on one machine template, and can be paused and resumed from snapshots when an agent is idle, such as while it waits on a human. Security is anchored in Google's Titanium infrastructure processing unit, with each microVM holding no persistent credentials and getting a short-lived, cryptographically verified identity instead. Each microVM also gets its own persistent disk for saving agent context across sessions.

Alongside Agent Compute Engine, Google is extending its existing Gemini for business agent-governance tools, including Agent Gateway, Agent Registry, and Agent Identity, across more environments: Google Cloud runtimes, other clouds, on-premises systems, and local coding agents such as Claude Code. It is also releasing an open-source Kubernetes-based runtime called Agent Substrate, which Google says sustained 500 agent activations per second with under 500ms resume latency in a 200,000-agent benchmark test, a figure that comes from Google.

Why it matters

Purpose-built, secure compute for agents addresses a real infrastructure gap as companies move from one-off AI pilots to large fleets of autonomous agents.

Source: Google Cloud