Glonce

Google announces Agent Compute Engine: one microVM per AI agent

infrastructureThe Glonce editorsThe Decoder🔥75Glonce rating75Impact70Novelty70Relevance100Surprise40Players95Credibility80
Google announces Agent Compute Engine: one microVM per AI agent

Google announced Agent Compute Engine, a service that gives each AI agent its own isolated virtual machine. According to the article, the service already powers Gemini and will open for preview in the coming months. Every agent runs inside a microVM secured by Google's Titanium chip, the company says.

The design aims at cost. When an agent idles, its VM pauses automatically and frees up the machine for others. Shared memory between VMs cuts what each one costs to run. The Decoder's own view is that VMs are a big cost factor when providing agentic systems, so Google is going after efficiency here.

The same technology is behind the open-sourced Agent Substrate, a Kubernetes runtime. Google says it can fit ten times as many agents on a single server. In a test with 200,000 agents, suspend and resume took less than 500 milliseconds, according to Google.

On the oversight side, Google is adding token tracking, model routing, and support for third-party identity providers such as Ping Identity and SailPoint. Those controls will extend to Cloud Run, GKE, and managed endpoints.

Key facts

  • Agent Compute Engine gives each AI agent its own isolated microVM, secured by Google's Titanium chip, the company says.
  • The service already powers Gemini and will open for preview in the coming months.
  • Idle agents have their VMs paused automatically, and shared memory between VMs lowers the running cost of each one.
  • Google says its open-sourced Kubernetes runtime, Agent Substrate, fits ten times as many agents on a single server; in a 200,000-agent test, suspend and resume took under 500 milliseconds, per Google.
  • Oversight additions: token tracking, model routing and third-party identity provider support (Ping Identity, SailPoint), extending to Cloud Run, GKE and managed endpoints.

Why it matters

Running many agents is expensive when each one needs its own machine. Google's answer is to isolate every agent in a microVM while cutting the cost of each: idle VMs pause and free their host, and VMs share memory. The Decoder reads this as Google targeting efficiency because VMs are a big cost factor for agentic systems. The fact that the service already powers Gemini makes it more than a concept announcement.

Who it affects

Teams that run or plan to run large numbers of AI agents, especially on Google Cloud, are the obvious audience. The oversight additions touch organisations that manage agent identity and spending: token tracking, model routing, and support for third-party identity providers such as Ping Identity and SailPoint. The controls are slated to extend to Cloud Run, GKE, and managed endpoints.

How to use it

Agent Compute Engine is not open yet. Google says a preview will open in the coming months. The open-sourced Agent Substrate, a Kubernetes runtime built on the same technology, is the piece named as already public. The source gives no preview date, pricing or regions.

How solid is it

The figures are all Google's own. The ten-times density gain is Google's claim, and the source does not say what baseline it is measured against. The sub-500-millisecond suspend and resume time comes from a Google test with 200,000 agents; no independent verification is mentioned. The Titanium security claim is likewise attributed to the company. No Google spokesperson or executive is quoted.

Risks and caveats

This is an announcement, not a shipped product: the preview has not opened. The source gives no cost figures or savings percentages for the shared-memory or pause mechanisms. It also does not say whether the oversight controls are available now or when they will reach Cloud Run, GKE and managed endpoints. Treat the efficiency and speed numbers as vendor claims until outside testing exists.