Google launches Agent Compute Engine: a hardware-isolated microVM per agent
Google Cloud announced Agent Compute Engine at AGNTCon+MCPCon: hardware-isolated microVMs built for agent workloads that already power Gemini work tasks, with customer preview expected in the coming months. The design stresses fast pause/resume and no long-lived credentials inside the sandbox, alongside broader Gemini for business agent governance.
Google Cloud says it is launching Agent Compute Engine at AGNTCon + MCPCon—a purpose-built compute service that already powers Gemini as an agent for work tasks and will be available in preview in the coming months with hardware-isolated runtime sandbox security. It is also extending Gemini for business agent governance so controls can follow agents wherever they run.
Per Google, each Gemini agent gets its own high-performance secure computer. MicroVMs boot quickly and use copy-on-write memory sharing plus checkpoint/restore to pause idle sessions between tool calls. Security leans on Titanium (IPU) and a Cloud Hypervisor boundary; sandboxes are described as holding no user credentials or API keys, with short-lived credentials injected on egress through a policy proxy. The deep dive says the platform already supports interactive sessions, autonomous software-engineering workspaces, and large-scale evaluation and RL workloads inside Google.
Availability: Google says Agent Compute Engine is “targeted for preview in the coming months.” This article does not invent a GA date or price. Where secondary coverage cites density or latency figures not restated in the Google Cloud posts, the official wording takes priority.
Sources
- Introducing Agent Compute Engine and expanding agent governance — Google Cloud
- Agent Compute Engine deep dive — Google Cloud