Google Cloud cuts idle AI agent costs on Kubernetes
Google Cloud has donated three open-source projects to the Cloud Native Computing Foundation to reduce compute costs for AI agents running on Kubernetes. Agents sit idle 90-95% of the time, but Agent Substrate addresses this by suspending agents, saving memory snapshots, and freeing compute until work resumes, delivering a reported 10x cost reduction versus dedicated VMs. Google also open-sourced gVisor for agent security sandboxing, and is collaborating with Anyscale on KubeRay and OpenRL to bring distributed training and reinforcement learning to platform engineering teams.
