Anthropic's Claude models are now generally available in Microsoft Foundry, Microsoft said in the June recap published July 7 on its Foundry developer blog, running on Azure infrastructure built from NVIDIA Blackwell Ultra systems linked by InfiniBand under a three-way partnership between Microsoft, NVIDIA and Anthropic.
This is not a marketplace listing or a thin proxy. Foundry exposes Claude through the full Messages API, including prompt caching, extended thinking and tool streaming, with deployments such as claude-sonnet-4-6 sitting alongside OpenAI and open-weight models in the same catalog. Foundry Agent Service can use Claude as the reasoning core of multi-step agents. Billing lands as Claude Consumption Units consolidated into the regular Azure invoice, with Global and US data zones and a zero data retention option for sensitive workloads.
The same update moved agent publishing to Microsoft 365 Copilot and Teams to general availability, so agents "move through one governed publishing pipeline instead of separate rebuilds per surface," as the post puts it. Autopilot agents entered public preview with their own Entra Agent ID, a productivity license, and access to email, calendar, OneDrive and Teams, designed to sit in group chats rather than one-on-one sessions.
Why it matters: Microsoft has invested more than anyone in OpenAI, and it is now selling OpenAI's chief rival as a generally available, Azure-billed service on premium NVIDIA silicon. That is the clearest signal yet that frontier models are commoditizing on top of cloud distribution. For enterprises, the switching cost between labs collapses to a deployment name and a rate card. For Anthropic, it is enterprise distribution at Azure scale without building that channel itself.
The Gulf link is real. Governments and banks across the UAE, Saudi Arabia and Qatar run heavily on the Microsoft stack, with Azure regions in the UAE and Qatar and Microsoft 365 embedded across ministries and enterprises. Claude access through existing Azure enterprise agreements removes the procurement and security-review friction that has kept many regulated Gulf organizations on a single model family. One caution: the announced data zones are Global and US only, so workloads bound by in-country residency rules need a compliance pass before regulated data touches a Claude endpoint, even on an Azure bill.
The practical takeaway: builders on Azure can now A/B Claude against GPT models behind one API surface, one identity layer and one invoice, and should benchmark their highest-value workloads this quarter rather than waiting for contract renewal. Operators negotiating Azure commitments gain leverage, since consumption can shift between labs without replatforming. The ones who should worry are vendors whose pitch was privileged access to a single frontier model. That moat drained on July 7.