Anthropic's Claude models are now generally available in Microsoft Foundry running on NVIDIA GB300 Blackwell Ultra GPUs on Microsoft Azure, the three companies confirmed on June 29, 2026. The deployment sits on GB300 NVL72 rack-scale systems wired with NVIDIA Quantum-X800 InfiniBand networking, the same configuration NVIDIA has been positioning as its premier substrate for agentic workloads. It is the first time Anthropic's frontier models have been offered to Azure-native enterprises on Blackwell Ultra silicon at general-availability scale.
The rollout is the operational delivery of the strategic partnership Microsoft, NVIDIA, and Anthropic first announced in November, when the three committed to expanding enterprise access to Claude on NVIDIA accelerated computing. That announcement was structural; this one is the product. Customers building inside Microsoft Foundry can now point agentic workloads at Claude without leaving the Azure perimeter, and without negotiating a separate Anthropic contract.
The technical pitch is inference economics. NVIDIA's Dave Salvator, writing in the company blog post, framed it bluntly: "Having great inference performance and efficiency reduces total cost of ownership and drives positive company results." GB300 NVL72 packs 72 Blackwell Ultra GPUs into a single liquid-cooled rack with NVLink and Quantum-X800 fabric between them, a configuration designed specifically to keep long-running agent sessions from hemorrhaging tokens and time on inter-node traffic.
Key facts
- 01Claude models are now generally available in Microsoft Foundry running on NVIDIA GB300 Blackwell Ultra GPUs on Azure.
- 02The deployment uses GB300 NVL72 systems with NVIDIA Quantum-X800 InfiniBand networking for agentic workloads.
- 03The rollout builds on the Microsoft–NVIDIA–Anthropic partnership announced in November to expand enterprise Claude access.
- 04NVIDIA's Secure Agent Workspace Reference Design provides the governance blueprint for running Claude agents on Azure.
- 05General availability was announced June 29, 2026; NVIDIA GTC Berlin registration runs October 20–22.
The deeper integration goes beyond hosting. NVIDIA said it is working with Anthropic to wire NVIDIA tools directly into the Anthropic stack, exposing what NVIDIA calls verified agent skills — pre-built, accelerated capabilities that Claude agents can invoke against a customer's existing data and systems. The goal stated by NVIDIA is that enterprises "can embed AI agents deeply into their business and use them as the operating system for the organization." That is the agentic thesis stated about as plainly as a vendor can state it.
Governance is the other half of the announcement. Running autonomous agents inside a regulated enterprise raises the obvious problems: which identity is the agent acting as, what credentials does it hold, what network can it reach, and what happens when a policy needs to change at runtime. NVIDIA is shipping the Secure Agent Workspace Reference Design as the answer — a blueprint that pushes identity, network access, credentials, and runtime policy down to the infrastructure layer rather than leaving them in application code.
For Anthropic, the deal is a meaningful distribution win at a moment when its US export situation has been turbulent. Recent AI Chat Daily coverage tracked the US clearing Anthropic to ship Mythos 5 to roughly 100 firms and agencies, and Austria lobbying the EU to host the company after US export curbs tightened. A first-class slot inside Microsoft's enterprise cloud, on NVIDIA's newest hardware, gives Anthropic a deeply embedded enterprise channel that does not depend on those export dynamics shifting in its favor.
For Microsoft, offering Claude on Foundry alongside OpenAI models is the continuation of a multi-model strategy that has been building all year. Foundry customers can now select between frontier providers without leaving Azure, which matters for enterprises that have standardized on Azure identity, networking, and compliance but want optionality on the model layer. The competitive question is whether the same enterprises that pay for ChatGPT Enterprise will also stand up Claude-based agent fleets inside Foundry — and how Microsoft prices the two against each other.
For NVIDIA, the announcement is one more data point that Blackwell Ultra is shipping into real enterprise workloads, not just hyperscaler training clusters. GB300 NVL72 is the architecture NVIDIA needs to defend against custom accelerators from the hyperscalers themselves, and lighting it up as the substrate for Claude-based agents in Azure is exactly the kind of reference deployment that justifies the premium. NVIDIA also used the post to open registration for GTC Berlin, running October 20–22.
The caveats are the ones that always attach to enterprise agentic rollouts. General availability of the infrastructure is not the same as production deployment by customers, and the verified-skills catalog will need to grow substantially before Claude agents can plausibly act as an organizational operating system in the way the marketing copy suggests. Pricing for GB300-backed Claude inference on Foundry was not disclosed in the announcement, and total cost of ownership at scale remains the open question every enterprise buyer will ask first.
The broader signal is that the agentic build-out is now a three-vendor stack — model provider, cloud, accelerator — and the deals binding those three together are getting tighter, not looser. Anthropic on GB300 in Foundry is the latest expression of a pattern in which frontier model access is increasingly bundled with specific silicon on specific clouds. Enterprises gain a turnkey path to production. They lose, in exchange, the leverage of swapping any one layer cleanly. That tradeoff is the shape of the AI enterprise market in 2026.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
