On October 22, at AGNTCon + MCPCon, Google Cloud introduced Agent Compute Engine (ACE), a purpose-built compute service for AI agents. Google Cloud describes it as a hardware-isolated microVM runtime with rapid boot, checkpoint and resume, per-microVM Hyperdisk persistence, and governance hooks through Agent Gateway. DigIntoAI has not tested ACE.
What Google Cloud says ACE does
According to the Google Cloud blog, ACE gives each agent a high-performance, secure computer for writing and running code, tools, browsers, and enterprise workflows. Google Cloud says the same stack already supports Gemini agent workloads for work tasks, and that customers will be able to use the service for their own custom agents. Availability is framed as preview in the coming months.
Security claims in the post include a hardware-isolated Cloud Hypervisor boundary with its own guest kernel, short-lived credentials, and cryptographically verified microVM IDs. For prior DigIntoAI coverage of Google Cloud agent and Gemini product choices, see Google Cloud Gemini Agent and Gemini 4 Argon.
Agent Substrate on GKE
Google Cloud also announced Agent Substrate on Google Kubernetes Engine (GKE) as an open-source execution runtime for teams that want customization, multi-cloud portability, or an existing Kubernetes standard. The blog says it uses a zero-trust design that decouples agent execution from machine management, snapshots idle agents, and returns capacity to the pool. Google Cloud positions it as extending agent-first design and governance across runtimes. Related Google open-infra reading on DigIntoAI includes Google TorchTPU open source.
We have not tested Agent Compute Engine or Agent Substrate.
