Google Cloud has introduced a purpose-built compute service designed to host AI agents, with security isolation as its central selling point.

The company announced Agent Compute Engine on October 22, 2026, during the AGNTCon + MCPCon events. The offering is a compute service engineered specifically for running AI agents at scale.

What Google Actually Built

Agent Compute Engine runs agent workloads inside hardware-isolated microVMs. Each microVM receives a dedicated guest kernel and root access.

The microVM detail is worth unpacking. Cloud workloads usually run either in containers, which share the kernel of the machine they sit on, or in traditional virtual machines, which boot their own kernel but start slowly. A microVM is a stripped-down virtual machine — hence the dedicated guest kernel — designed to start with container-like speed while keeping the isolation profile of a full VM.

That isolation matters because agents are active participants rather than passive services. They execute code, invoke tools, and handle data on behalf of users.

Google says the service is tuned for the fast churn and heavy demands typical of agent workloads. In practical terms, agents can launch quickly, pause, and then resume later.

The Density Trick and the Security Layer

The second headline feature is copy-on-write memory sharing, a long-standing technique from operating system design applied here to agent fleets. When many agents start from the same base, they share the same memory until one of them modifies something. At that point, only the altered portion is duplicated into that agent's private copy.

According to Google, this approach has produced benchmarks of 10x density, meaning the company claims it can pack far more agents onto the same hardware than would otherwise be possible. For workloads built around many short-lived sessions, how much fits on a single machine feeds directly into per-agent infrastructure cost.

On the security side, Google is leaning on its Titanium IPU technology — an IPU, or infrastructure processing unit, is a dedicated chip that handles networking and other infrastructure tasks so host CPUs are freed for workloads — which the company positions as an added layer of protection against adversarial threats.

Google frames the service as part of its Gemini Enterprise Agent Platform, where security and governance for AI agents sit at the center of the platform's messaging.

Already in Use, at Least In-House

Agent Compute Engine is not a lab prototype awaiting its first customer. Google says it already operates the service internally for its own Gemini agents.

Customers will have to wait a little longer. The service is expected to enter a preview phase soon — in cloud-industry terms, a testing window before general availability in which features and terms can still change — though Google has not committed to a firm public date in the announcement.

Why Agents Need Their Own Infrastructure

A chatbot answers a question and stops. An agent may browse, write code, run that code, check the result, and try again — all without a human clicking anything.

That behavior creates two problems for a cloud provider. The first is scale, because agents can appear and disappear constantly. The second is containment, because an agent holding root access in a poorly isolated environment poses a risk to everything around it.

Agent Compute Engine is Google's answer to both problems at once. Fast boots and pause-and-resume handle the churn, while hardware isolation and Titanium IPU handle the containment.

What This Means

The 10x density claim is the number to watch. If preview customers see similar results, it could meaningfully change the cost of running large fleets of agents. The announcement also lands in a market where other major cloud providers, including Amazon Web Services and Microsoft, have been building their own services for creating and running AI agents — putting the infrastructure beneath those agents in play as well.

Timing is the other thing to watch. Google has said the preview is coming soon, and how quickly the service moves from preview to broad availability will show how ready it really is outside Google's own walls.

For now, the most convincing endorsement comes from Google itself. Its Gemini agents already live in these microVMs, and the company is clearly hoping customers will want to move into the building next.