Google Launches Agent Compute Engine ACE for AI Agents

Google announced on October 22, 2026 the launch of Agent Compute Engine (ACE), a Google Cloud compute service designed for AI agents. The service aims to give each agent a secure, persistent, and elastically scalable dedicated environment.

ACE provides each AI agent with a hardware-isolated microVM to safely execute code, invoke tools, operate browsers, and carry out multi-step tasks. Unlike traditional VMs or containers, ACE is optimized for agent lifecycle patterns, with fast startup, an independent guest kernel, root permissions, and security enforced by a hardware-isolated Cloud Hypervisor boundary. The sandbox does not store credentials, reducing the risk of misuse or data leakage.

A key feature is persistence with automatic suspend and resume. Idle agents can automatically suspend and release compute resources, then quickly restore files and memory state when needed. Google said that in tests involving 200,000 agents, suspension and resume took less than 500 milliseconds, and shared memory plus snapshot replication can significantly increase the number of agents that a single server can support.

ACE initially supports Gemini, the Gemini platform, Antigravity, and Google's internal model evaluation and reinforcement learning workflows. Google said developers will eventually be able to use ACE through Google Cloud, controlling image, identity, network access, and lifecycle for their agent environments, while Google handles underlying resource scheduling and scaling.