Google launches Agent Compute Engine for AI scaling

Google announced Agent Compute Engine, a service providing each AI agent with a hardware-isolated micro virtual machine. According to The-decoder, the service already runs Gemini and will enter a preview phase in the coming months. The system uses Google's Titanium chip to secure these microVMs. It also includes the open-sourced Agent Substrate, a Kubernetes runtime designed to increase agent density on single servers.

Why it matters

For Swiss businesses, these technical specs translate to lower overhead for scaling AI operations. If you deploy agents for customer service or internal data processing, the cost of idle compute often eats into your margins. Google aims to solve this by pausing inactive VMs and sharing memory. This means you can run more agents on the same hardware without sacrificing security.

• Lower operational costs for high-volume agent deployments.
• Predictable spending through integrated token tracking.
• Simplified security compliance using existing identity providers.

What to watch

Check your current identity provider compatibility. Google plans to support Ping Identity and SailPoint for these controls. If your business relies on these for security, you should verify how they will integrate with your managed endpoints. You need to know if your existing access rules will carry over to the new agent environment without manual reconfiguration.

High-volume AI deployments will move from fixed infrastructure costs to variable, density-based pricing within a year. This requires Google to maintain hardware reliability at scale.