What changed
Google Cloud's AI Infrastructure team released several new tools and updates in October, including the Agent Compute Engine, AI Hypercomputer enhancements, and open-source projects like llm-d 1.0 and TorchTPU. These updates aim to improve scalability, security, and efficiency for AI and agentic workloads.
What this means in practice
The Agent Compute Engine provides hardware-isolated microVM sandboxes with governance capabilities to secure and scale autonomous AI agents. AI Hypercomputer updates include new A5X instances powered by NVIDIA hardware, offering enhanced performance and sovereign AI controls. The llm-d 1.0 open-source project unifies distributed LLM inference across mixed TPU and GPU fleets, while TorchTPU integrates Google TPUs into the PyTorch ecosystem for training and deployment.
Limitations and availability
The GKE CPU startup boost and Wiz Runtime Sensor add-on are in preview, meaning they are not yet generally available. The AI Hypercomputer updates are available for enterprises using NVIDIA GTC Berlin 2026 configurations. Open-source projects like llm-d 1.0 and TorchTPU are now available for developers. The Google Distributed Cloud air-gapped version supports NVIDIA HGX B300 systems and includes frontier AI models and Gemini for isolated environments. The ML Drift framework is open-sourced for edge inference and supports multiple hardware platforms.
