Claude Meets Blackwell Ultra: Anthropic’s Models Now Run on NVIDIA GB300 in Azure

Summary

Anthropic’s Claude models have reached general availability inside Microsoft Foundry on Azure, where they run on NVIDIA GB300 Blackwell Ultra GPUs. The deployment uses GB300 NVL72 racks connected by NVIDIA Quantum-X800 InfiniBand, giving Azure customers direct access to the inference performance needed for autonomous, domain-specific agents.

The arrangement extends the November partnership among Microsoft, NVIDIA and Anthropic. NVIDIA is embedding its accelerated-computing tools into the Anthropic software stack so that Claude agents can acquire verified skills tied to particular business domains. Enterprises can therefore treat the agents as reusable components that operate across internal systems rather than as isolated chat interfaces.

To keep these agents inside controlled boundaries, Microsoft and NVIDIA supply the Secure Agent Workspace Reference Design. The blueprint defines infrastructure-level controls for identity, network access, credentials and runtime policy, allowing organizations to run agent fleets without exposing the underlying Azure and NVIDIA resources.

The combination is intended to lower the cost of running large-scale agent workloads while meeting enterprise requirements for governance and repeatability.

Why it matters

This update is highly relevant as it provides enterprises with access to state-of-the-art AI models backed by next-generation computing power via a major cloud provider. The emphasis on secure, governed environments for AI agents aligns perfectly with the EU and Dutch focus on safe, compliant, and transparent AI deployment.