July 17, 2026
Anthropic’s Claude models are officially generally available in Microsoft Foundry on Azure, giving enterprises a direct path to building production AI agents without leaving the cloud environment they already run on.
The announcement, posted by Azure Product Lead Steve Sweetman on June 29, 2026, positions the move as a fix for a problem that has nothing to do with model quality. Most enterprise AI projects don’t stall because the AI isn’t good enough. They stall because of what surrounds them: procurement approvals, governance reviews, networking setup, and data handling requirements that can drag a pilot project out for months.
With Claude now sitting natively inside Foundry, none of that has to be solved from scratch. Companies can access Claude through the Azure account they already use, along with the authentication, billing, networking, governance, and data controls their teams already trust. Instead of setting up new infrastructure, teams can go straight to building.
What changes for developers and IT teams
Developers reach Claude through the Messages API, with access to prompt caching, extended thinking, and tool streaming built in. Teams building AI agents can lean on Foundry Agent Service, which uses Claude as the reasoning layer to handle multi-step planning, tool use, and task execution across a company’s existing systems.
Inference runs inside Azure itself, and customers can pick between Global and US data zones depending on their data residency needs. Anthropic still operates the inference and serves as the data processor and SLA provider, but everything else routes through familiar Azure tools. That includes signing in with Microsoft Entra ID, applying Azure role-based access controls, and tracking usage the same way teams already track every other Azure service.
For workloads that need extra caution, zero data retention is available, meaning prompts and completions aren’t kept by Anthropic once the API call is finished. On the billing side, Claude usage now shows up as Claude Consumption Units on a single line of the Azure bill, which simplifies how finance and procurement teams handle the purchase.
Why the infrastructure matters as much as the model
The announcement makes a point of separating two different questions: is the model powerful enough, and can a company actually deploy it at scale? Claude runs on NVIDIA Blackwell Ultra systems connected through InfiniBand networking, part of a broader partnership between Microsoft, NVIDIA, and Anthropic first announced in November 2025. That infrastructure is built specifically for inference performance rather than training, which matters for companies running agents around the clock rather than occasional batch jobs.
A model router feature can automatically send queries to whichever Claude model fits the task best, a step Microsoft says can cut costs by up to 50% while also improving how satisfied users are with the responses. Everything running through Foundry is also monitored by Foundry Control Plane, which evaluates agent responses on an ongoing basis and can block outputs that violate a company’s rules before a user ever sees them.
Early adopters are already in production
Several companies quoted in the announcement describe moving past pilot projects into live systems. Gary Ballabio, Vice President of Partnerships at Bolt, said running Anthropic’s models on Azure has delivered the throughput and reliability their enterprise customers expect. NVIDIA’s Justin Boitano said Claude is now running on Foundry with NVIDIA GB300 GPUs lets more organizations run specialized AI agents at production scale. Momentic co-founder and CEO Jeff An said the company now serves millions of tokens per minute through the setup, and Everstar’s Matt Huang described using the combination to compress a nuclear safety analysis that would normally take 200 human days down to a single day.
Claude in Foundry is aimed squarely at the workloads growing fastest in enterprise AI right now: coding and software development, multi-step agent workflows, and document-heavy business analysis. With Microsoft IQ layered on top, agents can also pull in live enterprise context, and tools like the agent optimizer inside Foundry Agent Service can tune how an agent is prompted. Hence, it performs better regardless of which model is doing the work underneath.
For businesses evaluating whether to build their next AI agent on Azure or elsewhere, this general availability milestone removes one of the bigger points of friction: getting a frontier model approved, secured, and billed inside a cloud environment that’s already been vetted.
How Primotech Can Help
Getting Claude running inside Foundry is one thing. Getting it wired into your existing Azure environment, governed correctly, and actually adopted by your teams is another. Primotech works as an Azure partner on exactly this kind of transition, from cloud migration and consulting to hands-on setup of the IaaS and PaaS layers an AI agent needs to run reliably.
If your team is still weighing whether to bring Claude into a live Azure environment, Primotech managed cloud services can handle the security, access controls, and ongoing monitoring so your developers can focus on building the agent itself instead of the infrastructure underneath it. That includes setting up Entra ID authentication, role-based access, and the kind of governance guardrails enterprises need before anything touches production data.
Reach out to Primotech to talk through what a Claude-on-Foundry rollout would look like for your business.
