Anthropic and NVIDIA Bring Claude to Microsoft Azure on GB300 Blackwell Ultra

Anthropic’s Claude models are now generally available through Microsoft Foundry, hosted on Microsoft Azure and accelerated by NVIDIA GB300 Blackwell Ultra infrastructure. The deployment gives Azure customers direct access to Claude through the authentication, billing, networking, and governance systems already used across their Microsoft cloud environments.

According to the official NVIDIA announcement, Claude is running on NVIDIA GB300 NVL72 systems connected through NVIDIA Quantum X800 InfiniBand networking. The infrastructure is designed to support demanding inference workloads, including autonomous agents, specialized sub agents, coding systems, research tools, and business applications that require sustained reasoning across multiple tasks.

The first models available through the Azure hosted service are Claude Opus 4.8 and Claude Haiku 4.5. Developers can access both through the Claude Messages API, with support for capabilities including prompt caching and extended thinking. Anthropic says it plans to continue expanding model and feature availability inside Microsoft Foundry over time.

Claude runs inside the Azure environment with support for Microsoft Entra ID authentication, Azure role based access controls, consolidated billing, networking policies, and existing governance tools. Customers can also select where inference is processed, including a United States data zone for organizations with data residency requirements. Anthropic operates the inference service and acts as the data processor, while eligible customers can apply Claude usage against existing Microsoft Azure commitments.

“Claude models bring strong reasoning, coding and enterprise capabilities that are valuable for complex technical work.”
— Justin Boitano, NVIDIA Vice President and General Manager of Enterprise Computing

Microsoft is positioning Claude as a production ready option for companies building agentic applications through Foundry Agent Service. Claude can serve as the reasoning engine behind systems that plan tasks, call external tools, access enterprise information, generate software, analyze documents, and coordinate workflows across multiple business platforms.

The hardware foundation is equally important. Each GB300 NVL72 rack combines 72 Blackwell Ultra GPUs with 36 NVIDIA Grace CPUs, 37 TB of fast memory, up to 130 TB per second of internal NVLink bandwidth, and Quantum X800 InfiniBand connectivity for communication between racks. Microsoft previously disclosed an Azure production cluster containing more than 4,600 GB300 GPUs, demonstrating the scale at which the company is deploying Blackwell Ultra infrastructure. Microsoft has not stated that the entire 4,600 GPU cluster is dedicated specifically to Claude.

This deployment also builds on the strategic partnership announced by Microsoft, NVIDIA, and Anthropic in November 2025. That agreement included plans to scale Claude on Azure, optimize Anthropic models for NVIDIA architecture, and expand Claude availability across Microsoft enterprise services. Anthropic also committed to purchasing $30 billion of Azure computing capacity and securing up to 1 gigawatt of additional infrastructure.

NVIDIA is also integrating its software and enterprise tools into Anthropic’s wider technology stack. Verified agent skills can give Claude specialized abilities for industries and technical workflows, while the NVIDIA Secure Agent Workspace Reference Design provides a blueprint for controlling agent identity, credentials, network access, and runtime policies at the infrastructure level.

The announcement arrives as Blackwell Ultra becomes increasingly focused on agentic inference rather than conventional single response AI workloads. Our previous coverage of the NVIDIA GB300 agentic AI benchmark showed the platform delivering significantly higher concurrent agent capacity than the previous H200 generation. The infrastructure also connects with NVIDIA’s broader Quantum X networking strategy, which we examined through its co packaged optics deployment at Lambda.

Anthropic is simultaneously expanding its relationships across the wider infrastructure market. The company recently partnered with Micron around memory, storage, and internal Claude deployments, as detailed in our coverage of the Micron and Anthropic AI infrastructure agreement. These partnerships show that scaling frontier AI now requires coordinated development across accelerators, memory, storage, networking, cloud software, and enterprise governance.

Claude arriving as a generally available Azure hosted service is more significant than another model appearing in a cloud catalog. Microsoft is combining Anthropic’s reasoning capabilities with the procurement, security, identity, compliance, and billing systems that large organizations already use. That removes several of the operational barriers that often prevent experimental AI projects from reaching production.

For NVIDIA, the deployment gives Blackwell Ultra another major enterprise workload and reinforces the company’s position as a complete AI infrastructure provider. GB300 performance will matter, but the larger advantage comes from controlling the accelerator, rack architecture, networking fabric, deployment tools, and agent governance framework.

The competitive focus is also shifting. Enterprises are no longer evaluating AI models only by benchmark scores. They are increasingly examining how reliably those models can run autonomous workflows, access business systems, meet data requirements, and operate within established cloud controls. Claude on Azure directly targets that transition.

Will Claude running natively on Azure make Anthropic a stronger enterprise competitor to OpenAI, Google, and other frontier AI providers?