Enterprise AI Agents
Anthropic's Claude AI Models Now Generally Available on Microsoft Azure
Key Points
Hosting options: Claude is generally available on Microsoft Foundry hosted on Azure, with an additional option to host on Anthropic infrastructure that is still in preview, a split designed to meet varying procurement, compliance, and governance requirements.
Billing and invoicing: Claude Consumption Units (CCUs) are used for usage billing, with token usage converted at per-model rates and billed via the Azure Marketplace; eligible spend can count toward a Microsoft Azure Consumption Commitment.
Hardware specifics: The GB300 NVL72 rack-scale system packs 72 Blackwell Ultra GPUs and 36 Grace CPUs, with about 37TB of memory, 130TB/s NVLink bandwidth, and FP4 Tensor Core performance up to 1,440 petaflops with sparsity.
Claude model lineup in Foundry: 11 Claude models are listed in Microsoft Foundry, including Opus 4.8 and Sonnet 4.6, with Mythos and Fable currently unavailable.
Governance features: NVIDIA-verified agent skills and the NVIDIA Secure Agent Workspace Reference Design provide infrastructure-level controls for identity, networking, credentials, and runtime policy when deploying Claude agents on Azure.
Anthropic's Claude AI models are now generally available on Microsoft Azure, running on NVIDIA's most powerful chips to date. The launch, announced June 29, marks the first time Claude has ever run on NVIDIA hardware — specifically the GB300 Blackwell Ultra platform inside Microsoft Foundry, according to Wccftech.
The rollout is the product of a $45 billion three-way alliance between Microsoft, NVIDIA, and Anthropic, formed in November 2025. Dario Amodei, Anthropic's CEO, said the deal makes Claude "the only frontier model available on all three of the world's most prominent cloud services" — AWS, Google Cloud, and now Azure.
NVIDIA's Most Powerful Chips Drive Claude's Azure Debut
Claude runs on the GB300 NVL72, a rack-scale system that packs 72 Blackwell Ultra GPUs and 36 Grace CPUs into a single unit. It carries about 37TB of memory and delivers up to 1,440 petaflops of FP4 performance, according to Neowin. A petaflop is one quadrillion math operations per second — enough to run very large AI models fast.
The system also uses Quantum-X800 InfiniBand networking, which keeps data moving between chips at extremely high speeds with very low delay. Let's Data Science notes that this hardware "materially changes the cost-performance tradeoffs," enabling more complex, multi-step AI agent work that was previously too slow or too expensive to run at scale.
11 Claude Models Listed, Two Hosting Options Available
Microsoft Foundry now lists 11 Claude models, including Opus 4.8, Sonnet 4.6, and Haiku 4.5. Two models — Mythos and Fable — are not available. Mythos is restricted to defensive cybersecurity research only, according to Technetbooks.
Enterprises can choose between two hosting setups. The first hosts Claude directly on Azure infrastructure and is fully live. The second hosts Claude on Anthropic's own infrastructure and is still in preview. Microsoft designed the split to meet different compliance, data residency, and procurement needs. Billing runs through the Azure Marketplace using Claude Consumption Units, priced at $0.01 per unit, and eligible spend can count toward a Microsoft Azure Consumption Commitment.
NVIDIA Secure Workspace Addresses the 'Agentjacking' Threat
The launch comes weeks after a security firm disclosed "Agentjacking" — a new attack where hackers hijack autonomous AI agents to perform unauthorized actions. To counter this, the deployment uses the NVIDIA Secure Agent Workspace, a reference design that provides infrastructure-level controls over identity, networking, credentials, and runtime policy, according to Technetbooks.
Justin Boitano, NVIDIA's VP of Enterprise Computing, said Claude on GB300 gives organizations "the performance, scale and security needed for production." Not everyone is convinced. Some safety advocates warn that the Secure Agent Workspace may give enterprises a false sense of security over models whose behavior is still hard to predict fully.
A Strategic Bet to Rival OpenAI Inside Microsoft's Ecosystem
The alliance effectively gives Microsoft a second top-tier AI model inside its own cloud. Market analysts note this hedges Microsoft's heavy bet on OpenAI by bringing in a direct competitor on native Azure infrastructure, according to LinkedIn. Anthropic has reportedly filed for a confidential IPO at a valuation near $965 billion, driven by projected annual revenue of $47 billion.
Jensen Huang, NVIDIA's CEO, framed the GB300 as the essential "operating system" for enterprise AI, one built to run "autonomous and specialized sub-agents" across business domains. For enterprises already standardized on Azure, the integration means they can build, govern, and scale Claude-powered agents without leaving the Microsoft environment.
Sentiment
Mostly PositivePublishers
10
Articles
13
Reach
23
