Anthropic Claude Models Now Available in Microsoft Foundry on NVIDIA GB300 NVL72 Infrastructure

NVIDIA has posted that Anthropic’s Claude models are now generally available in Microsoft Foundry on Azure, running on the company’s GB300 NVL72 platform — one of the most powerful AI server configurations currently deployed at commercial scale. The announcement, made via NVIDIA’s official social media account, places Claude on infrastructure that Microsoft and NVIDIA have been building out for demanding AI inference workloads.

The claim of general availability in Microsoft Foundry comes directly from the NVIDIA post. It has not, at the time of writing, been independently corroborated by official Anthropic documentation, so it should be treated as verified via social announcement rather than through multiple independent sources.

What Is the GB300 NVL72?

The GB300 NVL72 is not a single server — it’s an entire rack unit. Each rack packs 72 NVIDIA Blackwell Ultra GPUs alongside 36 NVIDIA Grace CPUs, all liquid-cooled and managed through smart rack systems. Microsoft’s Azure documentation describes the configuration as delivering 1.44 exaflops of FP4 Tensor Core performance per rack domain, and around 1.1 million large language model inference tokens per second per rack. Those are large numbers by any measure.

Networking runs over NVIDIA Quantum-X800 InfiniBand, providing 800 gigabits per second of connectivity per GPU. That bandwidth matters for the kind of large-scale inference workloads — chiefly agentic AI tasks, where models must reason across multiple steps — that Microsoft and NVIDIA say the platform is designed to handle.

Azure’s announced large-scale cluster contains over 4,600 NVIDIA Blackwell Ultra GPUs in total, according to NVIDIA’s blog, making it what the company describes as the first supercomputing-scale production cluster of GB300 NVL72 systems for commercial AI inference.

Claude on Azure: What’s Been Announced

According to the NVIDIA post, Claude models — Anthropic’s family of AI assistants — are now accessible through Microsoft Foundry, Microsoft’s platform for deploying and customising AI models within enterprise Azure environments. The tweet references autonomous or agentic AI capabilities as part of the context, though the full wording is only partially visible in available sources and should be treated with appropriate caution.

Microsoft Foundry is aimed at enterprise customers who want to run foundation models at scale with control over deployment, fine-tuning, and integration into existing business systems. Placing Claude on GB300 NVL72 infrastructure positions Anthropic’s models alongside other high-end AI offerings available through Azure.

It’s worth being clear about what hasn’t been confirmed: the specific terms of the arrangement between Anthropic and Microsoft, the pricing structure for Claude access through Foundry, and the precise customer segments being targeted are not detailed in the available sources.

The Broader Azure AI Build-Out

This announcement sits within a wider pattern of Microsoft and NVIDIA deepening their infrastructure partnership. Azure’s NDv6 GB300 VM series, built on the same GB300 NVL72 platform, is described in NVIDIA’s blog as part of a production-scale supercomputing cluster — not a pilot or preview, but live infrastructure for AI workloads.

Jensen Huang, NVIDIA’s chief executive, said at a recent event: “The age of agentic AI is here, and enterprises need infrastructure that can keep up with the demands of reasoning at scale.”

The liquid cooling in GB300 NVL72 racks is a practical detail that reflects how power-dense modern AI hardware has become. Air cooling alone can’t handle the thermal output of 72 Blackwell Ultra GPUs in a single rack; liquid systems are now standard for this tier of deployment.

Questions That Remain Open

Several things aren’t yet clear. Anthropic has not, in the sources available, independently confirmed the general availability announcement or provided detail on how Claude in Microsoft Foundry differs from Claude accessed through Anthropic’s own API or through Amazon Bedrock, where the company has a separate cloud partnership.

Pricing for Claude via Microsoft Foundry has not been detailed in the available sources. And while NVIDIA’s post references agentic AI capabilities, the specific Claude model versions — whether Claude 3.5 Sonnet, Claude 3 Opus, or others in the family — available through Foundry are not specified in the tweet or the supporting documentation reviewed for this article.

Microsoft has not issued a separate press release corroborating the announcement at the time of publication.

What This Means for Kent Residents

For Kent businesses and public sector organisations already using Microsoft Azure — and there are many, from NHS trusts to logistics firms operating around the port of Dover — the general availability of Claude through Microsoft Foundry means access to Anthropic’s AI models through an existing enterprise cloud relationship, running on faster, higher-capacity infrastructure than previous Azure AI offerings. UK enterprises using Azure will pay in sterling, though Microsoft has not published Foundry pricing for Claude at the time of writing. For individual consumers, the change is unlikely to be immediately visible, but it reflects the ongoing expansion of enterprise AI capacity that increasingly shapes the software tools people use at work.