ARTICLE AD BOX
Microsoft and NVIDIA just announced that Anthropic’s Claude models are now widely available in Microsoft Foundry. These models run on Azure and use NVIDIA’s top-tier GB300 Blackwell Ultra GPUs, giving businesses another powerful choice for building AI agents and production-ready generative AI apps.
This rollout lands at a moment when many companies are graduating from AI pilots to handling more complicated agent-based systems. These bigger, more demanding projects need speed, reliable security, strong governance, and the ability to scale. With Claude now in Foundry, developers can work through their current Azure accounts, using the same controls they already know for authentication, billing, networking, governance, and data management.
On the hardware side, NVIDIA says Claude in Microsoft Foundry runs on their GB300 NVL72 systems with Quantum-X800 InfiniBand networking. This combination targets enterprise-grade AI agents that can cross business functions and tackle complex tasks.
This move builds on the partnership the three companies kicked off in November 2025. Under that deal, Microsoft said Foundry customers would get access to Anthropic’s Claude models, while NVIDIA and Anthropic would work together on both model and hardware optimization. There’s a lot of money tied up here too—NVIDIA pledged up to $10 billion and Microsoft, $5 billion, in Anthropic. Anthropic, for its part, committed $30 billion for Microsoft’s Azure cloud services as part of the broader agreement. The deal opened up Microsoft’s AI ecosystem beyond just OpenAI, while also boosting Anthropic’s reach into enterprise markets thanks to Azure.
Earlier, Microsoft had talked about its first large-scale Azure cluster for OpenAI workloads, built from more than 4,600 NVIDIA GB300 NVL72 systems with Blackwell Ultra GPUs and connected by NVIDIA’s latest InfiniBand. Those clusters were designed for faster model training and high-throughput inference.
Now with Claude on the same infrastructure, Microsoft is framing Foundry as a versatile, multi-model platform. Businesses can pick from cutting-edge AI models based on what fits their use case, how much performance or governance they need, and so on. Microsoft says Claude is a good fit for things like coding, workflow automation, complex reasoning, heavy document analysis, and business decision support.
Developers get access to Claude in Foundry through the Messages API, and they can use things like prompt caching, extended reasoning, and tool streaming. All inference runs inside Azure, with customers able to choose between global or US data zones if data residency is an issue.
The launch also makes Microsoft’s enterprise AI stack stronger. Foundry Agent Service, Microsoft IQ, and broader Copilot tie-ins are all part of this. Foundry Agent Service can plug Claude in as the reasoning engine for multi-step planning, tool use, and complex execution across enterprise tools.
Really, this all reflects a bigger trend. Cloud providers, chipmakers, and AI model developers are teaming up to lock in computing power and get better performance out of these massive models. For companies, it’s more choice, but also more complexity—there are real questions now around cost, data residency, matching the right model to the right job, integration, and more.
As for pricing, Microsoft and NVIDIA aren’t sharing specific details about running Claude on the GB300 Blackwell Ultra systems. Microsoft says usage is billed through Claude Consumption Units as part of the regular Azure billing setup.
English (US) ·