Claude reaches GA on Microsoft Foundry, but EU teams still have no compliant data zone

Anthropic’s Claude models reached general availability on Microsoft Foundry in late June 2026, which turns Claude from an Azure preview into something an Azure-committed team can buy, bill, and govern through the contract it already has. The more revealing part is what the launch does not cover: as it stands, the only data-residency-scoped way to run Claude on Foundry is a US data zone, which leaves enterprises bound by EU residency rules without a compliant deployment.

What shipped

Microsoft and Anthropic moved Claude Opus 4.8 and Claude Haiku 4.5 to general availability on Microsoft Foundry on June 29, 2026, with Claude Sonnet 5 following days later at introductory pricing reported at $2 and $10 per million input and output tokens through August 31. All three run under a “Hosted on Azure” option that keeps inference on Azure infrastructure end to end, alongside a separate “Hosted on Anthropic infrastructure” option for teams that want the fuller API surface.

The GA milestone is mostly a procurement and governance story. Claude now sits behind Azure’s native identity, networking, and billing, so usage lands on a single consolidated invoice and, for customers with a Microsoft Enterprise Agreement, draws down an existing Azure commitment. Access runs through Entra ID rather than a separate Anthropic key, and consumption is metered as Claude Consumption Units through the Azure Marketplace. For a developer inside an Azure shop, the practical shift is that adopting Claude no longer means a separate vendor contract, a new billing relationship, or a second identity system. It becomes a model you switch on inside the environment you already run.

Why Europe can’t deploy it yet

The constraint is not the model, it is where the bytes are processed and who processes them. Foundry offers Claude in two deployment shapes. Global Standard is available in East US2 and Sweden Central but routes inference globally, so it does not pin processing to a region. Data Zone Standard, the option that does scope processing for residency, is documented as US-only for Claude Sonnet 5, Opus 4.8, and Haiku 4.5. There is no European data zone for Claude on Foundry today.

Compounding that, Microsoft’s own documentation is explicit that Anthropic remains an “independent data processor” for prompts and outputs regardless of hosting choice, and sells and operates the models as a non-Microsoft product. That is a different legal posture than OpenAI models on Azure, which Microsoft operates directly and offers inside EU data zones. As InfoQ reported, the gap is already blocking real deployments: a practitioner cited a large Dutch bank client that will not run Anthropic models through Foundry for this reason, and Anthropic’s compliance materials list European Foundry availability as “coming 2026” with no firm date.

Where this lands in the market

Reaching GA closes the procurement objection that kept Claude out of many Azure enterprises: buyers who could not add a separate AI vendor can now consume it under governance and billing they already trust. That is a meaningful unlock for Anthropic in a Microsoft-heavy enterprise base, and it is the kind of distribution that turns a model from a pilot into a default.

The catch is that, on Azure, choosing a model now carries a compliance geography rather than only a price and quality tradeoff. A US-headquartered team can treat Claude and OpenAI’s models as near-interchangeable options behind the same Foundry controls. A regulated European team cannot, because one vendor is operated inside an EU data zone and the other is not. For that buyer, the practical decision is less about benchmark scores and more about which model has a lawful place to run.

What’s worth watching

1. When a European data zone lands. Anthropic’s “coming 2026” note is the single gating item for EU regulated buyers. A firm date, or its slippage, decides whether those enterprises wait or standardize on a Microsoft-operated model in the meantime.
2. Whether residency, not capability, drives the choice. If EU teams pick OpenAI on Azure mainly because it has a compliant data zone, it suggests model selection in regulated sectors is becoming a legal question first and a technical one second.
3. How pricing settles after the intro window. Sonnet 5’s promotional rate runs through August 31. Where it lands afterward will signal how aggressively Anthropic is buying share inside the Azure channel.

The plain takeaway is that going GA solved Claude’s enterprise buying problem and exposed a geography problem in the same week. For an Azure team in the US, Claude is now a first-class option under familiar controls. For a regulated team in Europe, the model is available but not yet deployable, and that distinction, not the benchmark leaderboard, is what decides the choice for now. Stackmaven’s follow-up coverage will revisit the European data-zone timeline on or around October 3.

Sources cited

1. Anthropic: Claude in Microsoft Foundry is now generally available claude.com
2. Microsoft Learn: Claude models in Microsoft Foundry (deployment types and regions) learn.microsoft.com
3. Microsoft Learn: Data, privacy, and security for Claude models in Microsoft Foundry learn.microsoft.com
4. InfoQ: Claude Reaches GA on Microsoft Foundry: European Enterprises Cannot Deploy It www.infoq.com