Google Cloud ships Agent Compute Engine, llm-d 1.0 and TPU updates

Google Cloud launched Agent Compute Engine, a compute service that runs autonomous AI agents in hardware-isolated microVM sandboxes with sub-second activation, plus gateway and registry tools for governing them. It also open-sourced llm-d 1.0, a Kubernetes-native stack for serving large language models across TPU and GPU fleets, and TorchTPU with Meta, which brings Google TPUs into PyTorch. GKE gained a preview CPU startup boost with up to 2x faster cold starts, and Oracle will become the first hyperscaler to offer TPUs outside Google Cloud, from 2027.