Skip to content
Neural Daily

ToolsNews

Google Agent Compute Engine gives each agent a microVM

🇭🇺 Olvasd el magyarul →

Google Cloud unveiled Agent Compute Engine, the compute service that already runs Gemini's work agents. Every agent gets a hardware-isolated microVM, and customers can try it in preview in the coming months.

By: Mikola Máté Patrik · · 2 min read

Google Cloud graphic titled The infrastructure behind Gemini for business, with blue and green concentric rings on black
Image: Google Cloud · Press kit · Official Google Cloud blog hero image (og:image) · Source

What was announced?

On October 22, at AGNTCon + MCPCon, Google Cloud introduced Agent Compute Engine (Google Cloud), a compute service built specifically for AI agents that already powers Gemini as an agent for work tasks. It will reach customers in preview in the coming months. No pricing has been announced.

How it works

  • Each agent gets its own microVM with hardware isolation: a Cloud Hypervisor boundary, its own guest kernel and root access.
  • MicroVMs boot quickly, so an agent loop or eval sweep doesn't wait on infrastructure.
  • Copy-on-write memory sharing runs multiple agents on a single OS template. When an agent is thinking or waiting on a human, its microVM pauses and later resumes quickly from a snapshot.
  • Security builds on Google's Titanium IPU. Every microVM gets a cryptographically verified ID and short-lived credentials, and service credentials are injected at egress, outside the microVM.
  • Each microVM has its own storage to persist workspaces and agent context, and Agent Gateway integration adds observability and policy checks.

The Kubernetes option: Agent Substrate

For teams on GKE or multiple clouds, Google offers Agent Substrate, an open-source runtime. Google claims up to 10x higher density than traditional compute; in a 200,000-agent test it sustained 500 activations per second with suspend and resume under 500 ms. It runs on any Kubernetes cluster and works with any agent harness, including Claude Code, OpenClaw and Hermes.

Governance everywhere

Google is also extending Gemini for business agent governance to Cloud Run, GKE, other clouds and on-prem environments. An OS-level sandbox for managed devices, covering coding agents like Claude Code and Antigravity CLI, is coming to preview soon.

Why it matters

Agents run code, drive browsers and often handle untrusted input. Google is offering customers the same isolated execution model it uses for its own Gemini agents. The specifics, including pricing, regions and limits, will come with the preview.

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles