Skip to content
All posts

Google Cloud Addresses the Infrastructure Needs of Autonomous AI Agents

The Brief: Google Cloud has introduced Agent Compute Engine, a purpose-built compute service for running autonomous AI agents. The service currently powers Gemini's work-related agent tasks and will enter preview in the coming months.

Google also announced expanded Gemini for business governance capabilities to manage agents across Google Cloud, multicloud, hybrid environments, and managed devices. The update includes Agent Substrate, an open-source execution runtime for Google Kubernetes Engine (GKE), and additional Agent Gateway capabilities for token management, model routing, security policies, and third-party integrations.

These developments address the computing, security, identity, and administrative requirements of enterprise agent deployments.

Learn full details of the announcement about Agent Compute Engine, Agent Substrate, and Google's expanded AI agent governance capabilities at cloud.google.com.

Google Cloud branding and the headline “The infrastructure behind Gemini for business” displayed beside overlapping blue and green circular graphics on a black backgroundSource: Google Cloud

Google Cloud Launches Agent Compute Engine and Expanded AI Agent Governance

Analyst Perspective: Google Cloud is extending its AI infrastructure to accommodate software that does more than respond to prompts. Autonomous agents need computing environments where they can execute code, interact with applications, access enterprise resources, and continue working across multiple steps. These activities introduce infrastructure requirements that differ from conventional application workloads.

Agent Compute Engine gives Google a dedicated service for this type of execution. Its design accommodates the frequent starting, pausing, and resuming associated with agent activity, while its security architecture provides isolated environments for individual sessions. This makes the service relevant to organizations developing agents that perform more involved business tasks.

The addition of Agent Substrate also gives Google Cloud customers another way to deploy these workloads through Kubernetes. This combination is an important development in Google's enterprise AI portfolio because it connects specialized computing resources with existing infrastructure choices and operational requirements.

Google Cloud Develops Agent Compute Engine for Autonomous Workloads

Agent Compute Engine provides each Gemini agent with a dedicated computing environment through microvirtual machines (microVMs). These environments support activities such as writing and executing code, operating browsers, invoking tools, and coordinating enterprise workflows.

Google designed the service to accommodate the frequent activity changes associated with autonomous agents. Its microVMs boot rapidly and use copy-on-write memory sharing to run multiple agents from a common operating system template. When an agent enters an idle period, the host can pause its microVM and restore it when the agent needs to resume execution. This helps conserve computing resources without requiring agents to restart their sessions.

Security features include hardware isolation through Google's Titanium infrastructure processing unit and Cloud Hypervisor. Each microVM receives its own guest kernel, root access, cryptographically verified identity, and short-lived credentials. Hyperdisk also provides persistent storage for execution workspaces and agent context across sessions.

Google Cloud Brings Agent Substrate to Kubernetes Deployments

Agent Substrate on Google Kubernetes Engine gives platform teams an open-source execution runtime for Kubernetes environments. It caters to organizations that need customization, multicloud portability, or integration with existing Kubernetes workloads.

The runtime separates agent execution from machine management. When an agent becomes idle, Agent Substrate captures its active memory and returns the machine to the available resource pool. The system restores the agent when it needs to continue working.

Google reports that Agent Substrate can deliver up to 10 times higher compute density than traditional compute while maintaining kernel isolation and identity-based network security. In benchmark testing involving 200,000 agents, it sustained 500 activations per second with suspend and resume latency below 500 milliseconds.

Its security capabilities include microVM or gVisor isolation, individual agent identities, and granular network policies enforced through egress proxies. The runtime also supports frameworks such as Claude Code, OpenClaw, and Hermes, with Filestore agent volumes providing shared storage for multi-agent workflows.

Google Cloud Extends Agent Governance Across Cloud and Endpoint Environments

Google is extending its Gemini for business governance capabilities through four services: Agent Gateway, Agent Registry, Agent Identity, and agent observability.

Agent Gateway manages communication between clients, agents, tools, and other services. It supports policies covering sensitive data protection, agent attack prevention, and custom security requirements.

Agent Registry maintains an inventory of approved agents, tools, MCP servers, skills, models, and endpoints. Agent Identity assigns each agent a unique identity using the SPIFFE open standard to support authentication and access control.

Google also provides OpenTelemetry-compatible observability through standardized metrics, logs, and distributed traces. Governance coverage will extend across Agent Runtime, Cloud Run, GKE, Agent Compute Engine, multicloud environments, and managed endpoint devices. Additional Agent Gateway capabilities will introduce token consumption controls, model routing, extensible policies, and integrations with security partners.

Google Cloud Connects Agent Execution with Enterprise Control

Google Cloud's latest developments complement its existing portfolio of Gemini services, cloud infrastructure, Kubernetes, and enterprise security products.

Agent Compute Engine adds dedicated execution resources, while Agent Substrate accommodates organizations that prefer Kubernetes-based deployments. The governance services connect these capabilities with the administrative requirements of enterprise IT.

Businesses developing coding assistants, customer service agents, research tools, and automated internal workflows could benefit from this combination.

Security and Integration Challenges for Autonomous Agents

Managing permissions across multiple agents, applications, and cloud environments will require careful configuration. Organizations will need clear access policies, accurate agent inventories, and reliable monitoring.

Integration with existing identity providers and security systems will also influence deployment complexity. Google can help address these requirements through its policy controls, identity services, observability capabilities, and security partner integrations.

The Road Ahead for Secure and Scalable Agent Deployment

The next consideration is how effectively Google brings these services into production environments. Agent Compute Engine's preview, cross-cloud governance, endpoint controls, and additional Agent Gateway capabilities will provide important indicators of adoption.

Demand for dedicated agent infrastructure is expected to increase as businesses move from experimental deployments to automated operations.

Google's ability to deliver consistent security, manageable computing costs, and compatibility across different environments will determine how valuable this infrastructure becomes to enterprise customers.

Strengthen your brand voice through expert-led video and digital content. Get in touch with our team at the Collab Collective to learn how we help technology providers reach the right audience.