Claude is now generally available in Microsoft Foundry, running on NVIDIA GB300

Anthropic’s Claude models reached general availability in Microsoft Foundry on June 29, 2026. (Source: Microsoft Azure Blog, 2026-06-29) For Azure-committed teams, this turns Foundry into a first-class place to run Claude — without leaving your existing Azure identity, billing, and data-residency setup.

Key facts:

• GA date: June 29, 2026.
• Two models are available today: Claude Opus 4.8 and Claude Haiku 4.5. (Source: Claude.com blog, 2026-06-29)
• The models run on NVIDIA GB300 Blackwell Ultra GPUs — specifically GB300 NVL72 systems with Quantum-X800 InfiniBand networking. (Source: NVIDIA Blog, 2026-06-29)
• You call Claude through the Messages API, with prompt caching and extended thinking supported.
• There are two deployment options: hosted on Azure (Azure authentication, billing, governance, and a US data zone for residency) and hosted on Anthropic (the full API feature set, previously Foundry Preview).
• Microsoft Enterprise Agreement customers can draw Claude usage against their Azure commitments, on a single consolidated invoice.
• Anthropic operates the inference and acts as the data processor — even on the Azure-hosted path.
• This builds on the Microsoft, NVIDIA, and Anthropic partnership announced in November 2025.

What actually shipped

The capability is not new — these are the same Claude Opus 4.8 and Claude Haiku 4.5 you already call from Anthropic’s API. What changed is the procurement and governance surface. You now reach Claude inside the Foundry catalog, authenticate with Azure identity, and pay through your Azure invoice instead of a separate Anthropic contract.

For teams building agents, Foundry Agent Service can use Claude as the reasoning core to orchestrate multi-step planning, tool use, and task execution across enterprise systems. On the NVIDIA side, the deployment adds NVIDIA verified agent skills and a Secure Agent Workspace Reference Design for governed autonomous agents.

Microsoft cited early production users: Bolt (“Running Claude models on Azure has given us the sustained throughput and reliability our enterprise customers expect”), Everstar (compressed a nuclear-safety analysis from ~200 human-days to one day), and Momentic (serving millions of tokens per minute for test automation). (Source: Claude.com blog, 2026-06-29)

What it changes if you build on Azure

This is the Azure mirror of OpenAI’s Codex and GPT-5.5 landing on Amazon Bedrock. The pattern is the same: a frontier provider’s models become billable and governable inside a hyperscaler you’re already committed to.

• If you’re Azure-native: moving Claude calls into Foundry means one invoice, Azure RBAC and networking controls, and usage that counts against your EA commitment. The model string and Messages API stay familiar.
• Watch the two-path feature gap. The hosted-on-Azure path gives you Azure governance plus a US data zone; the hosted-on-Anthropic path is the one with the full feature set. Pick deliberately if you depend on newer API features — and keep the model ID and host in a config flag so you can switch.
• Data flow is honest, not magic. Even on Azure, Anthropic runs the inference and is the data processor. “Runs on Azure” is a residency and billing statement, not “your tokens never leave your tenant.”
• No new benchmark. There are no fresh capability numbers here — Opus 4.8 is Opus 4.8. The value is operational, so evaluate it as a procurement decision, not a model upgrade.

If you’re already running Claude through Claude Code as a daily driver or weighing Opus 4.8 against the rest of the field, this mainly changes where the bill lands and who governs the traffic — useful if your security team mandates Azure, neutral otherwise.

Sources

• Claude in Microsoft Foundry is now generally available — Microsoft Azure Blog (2026-06-29)
• Claude in Microsoft Foundry is now generally available — Claude by Anthropic (2026-06-29)
• Claude Meets Blackwell Ultra: Anthropic’s Models Now Run on NVIDIA GB300 in Azure — NVIDIA Blog (2026-06-29)