Google's Agent Compute Engine Gives Every Gemini Agent Its Own Isolated Micro VM
Google has introduced Agent Compute Engine, a runtime environment built specifically for AI agents, and expanded governance tools for managing agents across cloud, hybrid, and on-premises systems.
Google has unveiled Agent Compute Engine, a runtime environment designed specifically for artificial intelligence agents, giving each Gemini agent its own hardware-isolated micro virtual machine. The move marks a shift in how cloud infrastructure is being tailored to autonomous software that acts on behalf of users rather than simply responding to queries.
The core of the announcement is isolation. By assigning every agent a separate micro virtual machine with hardware-level separation, Google aims to contain the actions of each agent so that one agent's behaviour cannot affect another's data or operations. That architecture is intended to address a central concern for organisations deploying autonomous systems: what happens when an agent runs code, accesses tools, or makes decisions without a human in the loop.
Alongside the runtime, Google is expanding its governance capabilities so that companies can manage agents across cloud, hybrid, and on-premises environments. The governance layer is meant to give organisations oversight of how agents are deployed and operated, regardless of where the underlying infrastructure sits. For enterprises that keep sensitive workloads outside public cloud environments, the hybrid and on-premises support is a significant part of the offering.
The launch reflects a broader shift in the technology industry. Cloud providers are moving from hosting applications and data to hosting autonomous agents that execute tasks, call other services, and operate with a degree of independence. That change raises questions about security, accountability, and control that go beyond traditional software deployment. Isolating each agent in its own micro virtual machine is one answer to the security question, while governance tools address the accountability side.
For Google, the Agent Compute Engine also strengthens the position of its Gemini models in the enterprise market. Companies evaluating agentic systems weigh not only model capability but also the infrastructure that surrounds it: how agents are sandboxed, audited, and managed at scale. By building a dedicated runtime and governance framework, Google is signalling that it intends to compete on those operational foundations as much as on the models themselves.
The announcement does not specify pricing, availability dates, or which governance features will be included at launch. Those details will matter for organisations deciding whether to build agent-based workflows on Google's platform or on competing infrastructure. What is clear is that the infrastructure layer for AI agents is becoming a distinct product category, with isolation and oversight as its defining features.
7
