Anthropic Claude goes generally available on NVIDIA GB300 Blackwell Ultra in Azure

Anthropic's Claude models became generally available on NVIDIA GB300 NVL72 systems through Microsoft Foundry on June 29, 2026, giving Azure-based enterprises a high-performance foundation for deploying autonomous AI agents at scale. The launch marks the production-grade fruition of a partnership between Microsoft, NVIDIA, and Anthropic first announced in November 2025.

What's new

Claude is now available in the Microsoft Foundry catalog running on NVIDIA GB300 NVL72 systems — NVIDIA's highest-density Blackwell Ultra hardware — interconnected with NVIDIA Quantum-X800 InfiniBand networking. The configuration is designed for low-latency, high-throughput inference on long-horizon agentic tasks.

The deployment includes support for the NVIDIA Secure Agent Workspace Reference Design, a governed infrastructure blueprint that applies controlled identity, network access, credential injection, and runtime policy at the infrastructure layer. Enterprises can embed Claude agents as specialized sub-agents within their existing Azure infrastructure without building custom security architecture from scratch.

Key specifics:

• Hardware: NVIDIA GB300 NVL72 systems (Blackwell Ultra generation)
• Networking: NVIDIA Quantum-X800 InfiniBand
• Access: Microsoft Foundry catalog (Azure-native)
• Agent support: NVIDIA verified agent skills enable domain-specific capabilities across business functions
• Security: NVIDIA Secure Agent Workspace Reference Design for governed autonomous deployment

Context

Microsoft, NVIDIA, and Anthropic announced a partnership in November 2025 to bring Claude models to NVIDIA accelerated computing infrastructure on Azure. At the time, Claude was available through earlier Blackwell hardware. The GB300 Blackwell Ultra is NVIDIA's current flagship data center platform, offering substantially higher memory bandwidth and compute density than prior generations.

Microsoft Foundry is the managed model platform inside Azure that gives enterprise customers unified access to frontier models alongside Azure's broader AI services. Anthropic has also made Claude available through Amazon Bedrock and its own Claude Platform on AWS, but the Foundry integration targets organizations already committed to the Azure ecosystem.

The NVIDIA Secure Agent Workspace Reference Design, which now supports Claude, provides a blueprint for running autonomous agents in environments where security and compliance requirements make ad hoc infrastructure impractical.

Why it matters

GB300 NVL72 systems are among the densest AI inference platforms commercially available. For enterprises running high-traffic Claude workloads — particularly long-context agentic sessions with many tool calls — moving from earlier hardware to GB300 can meaningfully improve throughput and reduce cost-per-token, though NVIDIA did not publish specific performance benchmarks in today's announcement.

The Secure Agent Workspace integration is notable for regulated industries. Healthcare, financial services, and defense contractors building autonomous agents face strict requirements around data handling and access control. Having a Microsoft and NVIDIA-certified reference design that covers those layers reduces time-to-production for teams that would otherwise spend months on security review before running Claude agents in a sensitive environment.

The broader signal is consolidation: three of the most important names in enterprise AI infrastructure — Microsoft for cloud, NVIDIA for silicon, Anthropic for frontier models — are now jointly GA on a single hardware stack. Competitors deploying on alternative silicon or cloud configurations will face direct comparison.