- Google announced Agent Compute Engine, a service that gives each AI agent its own hardware-isolated micro virtual machine secured by Google’s Titanium chip; the service already powers Gemini and opens for public preview in the coming months.
- In an internal test with 200,000 agents, suspending and resuming a VM took less than 500 milliseconds, and Google says the architecture lets it fit ten times as many agents on a single server.
- Idle agent VMs pause automatically to free up server capacity, while shared memory between VMs cuts the cost of running each one.
- Google is adding token tracking, model routing, and support for third-party identity providers including Ping Identity and SailPoint, with the controls extending to Cloud Run, GKE, and managed endpoints.
What Happened
Google unveiled Agent Compute Engine, a new infrastructure service that runs every AI agent inside its own hardware-isolated microVM, The Decoder reported on October 22, 2026. The service already underpins Gemini internally and will open for preview in the coming months.
Why It Matters
Virtual machines are a major cost factor in running agentic systems at scale, and Google is positioning Agent Compute Engine as an efficiency play: by pausing idle agent VMs automatically and sharing memory between them, the company says it can run ten times as many agents per server than before.
Technical Details
Each agent’s microVM is secured by Google’s custom Titanium security chip. In a test involving 200,000 agents, suspend-and-resume operations completed in under 500 milliseconds. The service is also tied to Google’s newly open-sourced Agent Substrate, a Kubernetes-based runtime, building on Google Cloud’s broader $750 million push into agentic AI rollouts with consulting partners.
Who’s Affected
Enterprise customers running agents through Cloud Run, Google Kubernetes Engine, or managed endpoints gain access to new oversight tools, including token tracking and model routing, plus support for enterprise identity providers like Ping Identity and SailPoint. The move follows Google’s recent addition of security hooks to its Gemini API managed agents.
What’s Next
Google has not given a firm date for general availability beyond “the coming months,” and enterprises evaluating large-scale agent deployments will likely watch the preview period to see whether the promised cost and isolation benefits hold up outside Google’s internal Gemini usage.
