Google launches ACE Agent Compute Engine, providing an independent microVM for each AI agent
Google has released the Agent Compute Engine (ACE), which provides each AI agent with a hardware-isolated microVM capable of executing code, calling tools, and controlling a browser. When an agent is idle, its compute environment is automatically suspended and resources are released; when needed, file and memory state are quickly restored. In a test with 200,000 agents, pausing and resuming took less than 500 milliseconds. The service was initially used to support Gemini, the Gemini platform, Antigravity, and Google's internal model evaluation and reinforcement learning workflows. In the future, developers will be able to control agent environment images, identities, network access, and lifecycles themselves through Google Cloud.
Why read this
AI score 69
The original article describes ACE's isolation approach, suspend/resume latency, and internal use cases, allowing readers to judge the design trade-offs of agent runtime environments.
Topics
Body · AI translation
Google Launches ACE Service: Providing AI Agents with "Dedicated Workstations" That Automatically Clock Out and Wake Up in Under 500 Milliseconds
2026-10-23 10:55 IT Home - Guyuan
Thanks to IT Home user Huanan Wu Yanzu for the tip-off!
IT Home, October 23: Google yesterday (October 22) issued an announcement launching the Agent Compute Engine (ACE), a Google Cloud computing service built specifically for AI agents, with the goal of giving every agent a secure, persistent, and elastically scalable "dedicated computer."
According to the official announcement, the ACE service will provide each AI agent with an independent, hardware-isolated microVM (micro virtual machine), enabling it to securely execute code, call tools, operate a browser, and complete multi-step tasks.
Citing a blog post, IT Home explains that unlike traditional virtual machines or containers, ACE is specially optimized for the lifecycle characteristics of agents: it can start quickly, comes with an independent guest kernel and root privileges, and ensures security through the hardware-isolated Cloud Hypervisor boundary. Google says credentials are not stored inside the sandbox, reducing the risk of agent misuse or data leakage.
The highlight of the ACE service is that it is "persistent and pausable": when an agent is idle, its computing environment can automatically suspend and release resources; when it needs to continue working, it can quickly restore file and memory state.
Google says that in tests with 200,000 agents, pausing and resuming took less than 500 milliseconds, and that shared memory and snapshot replication can greatly increase the number of agents a single server can host.
The service was initially used to support Gemini, the Gemini platform, Antigravity, and Google's internal model evaluation and reinforcement learning processes. In the future, developers will be able to use ACE through Google Cloud to control their own agent environments' images, identities, network access, and lifecycles, while Google handles underlying resource scheduling and scaling.