Claude models reach GA on NVIDIA GB300 Blackwell Ultra in Microsoft Foundry

Microsoft and NVIDIA announced general availability of Anthropic's Claude models running on NVIDIA GB300 Blackwell Ultra GPUs within Microsoft Foundry. The integration leverages NVIDIA's latest Blackwell Ultra architecture to accelerate Claude inference, letting Azure-native enterprises build autonomous and domain-specific AI applications with frontier Anthropic models directly on Azure infrastructure.

The move is notable for its cross-vendor character: Anthropic models (historically most associated with AWS Bedrock, given Amazon's investment) now run at GA on Microsoft's cloud atop Nvidia's newest silicon. It reflects the increasingly multi-cloud reality of frontier model distribution, where labs place their models wherever enterprise demand exists rather than locking to a single provider.

For Azure customers, the appeal is combining Foundry's enterprise tooling and governance with best-in-class Claude capabilities and Blackwell Ultra performance. GB300 offers substantial memory and throughput advantages for large-context and agentic workloads, which pairs naturally with the newly launched Sonnet 5 and Opus 4.8.

The strategic subtext is the intensifying three-way dance between the clouds (AWS, Azure, Google Cloud), the chipmaker (Nvidia), and the labs (Anthropic). With Google capping Meta's Gemini access over compute shortages and signing a GPU deal with SpaceX, compute allocation has become a competitive weapon. Anthropic's presence across both AWS and Azure hedges its infrastructure dependency and broadens its enterprise reach ahead of its IPO.