Google Cloud launches Agent Compute Engine and expands agent governance tools

Google Cloud announces Agent Compute Engine, a purpose-built compute service for AI agents that already powers Gemini as an agent for work tasks, with preview availability planned in the coming months. It provides microVMs with hardware-isolated sandboxing via Titanium IPU, copy-on-write memory sharing, and pause/resume via checkpoint and restore. The company also announces Agent Substrate, an open-source Kubernetes-native execution runtime on GKE, and expanded Gemini for business agent governance including Agent Gateway, Agent Registry, and Agent Identity.

Key takeaways

• Agent Compute Engine gives each Gemini agent its own microVM with hardware-isolated Cloud Hypervisor boundary and its own guest kernel.
• Idle microVMs are paused via fast checkpoint and restore and resumed from snapshots while agents wait on humans or external systems.
• Agent Substrate is open source on GKE and reports up to 10x compute density and 500 activations per second in a 200,000-agent benchmark.
• Agent Gateway, Agent Registry, and Agent Identity extend governance, with agents assigned SPIFFE-based unique secure IDs.

Why it matters

Running autonomous agents at scale raises isolation, cost, and auditability problems that traditional compute was not designed for. The announcement describes concrete infrastructure and governance approaches to these problems, including open-source options for portability.

Read with care

Source claims not independently checked; performance figures come from Google's own benchmarks and the service is not yet generally available.