Anthropic and Microsoft announced general availability of Claude in Microsoft Foundry on June 29, 2026. Customers can choose between Global and US data zones for inference processing. The model catalog shows Claude Opus 4.8 and Haiku 4.5 as deployable. The authentication is Azure-native. The billing draws down existing Microsoft Azure Consumption Commitments. For a US-based customer that is a working integration. For a European customer with GDPR data-residency requirements it is a deployment option that does not meet compliance constraints yet.
I run ten production platforms. Three call Claude through AWS Bedrock in us-east-1 because the customers are US-based and Bedrock was the fastest path to a working integration. If one of those customers were EU-based and required EU data residency I would deploy Bedrock in eu-west-1 and the data-processing addendum would specify Frankfurt. The Foundry GA announcement does not give me that option because Claude models accessed through Azure AI Foundry are currently delivered through Anthropic-managed inference infrastructure rather than fully Azure-native regional processing, meaning even when an EU region such as Sweden Central or Germany West Central is selected, inference execution does not yet occur entirely within Azure-operated data center boundaries.
The partnership announced compute but the deployment is still third-party
Anthropic committed to purchase $30 billion of Azure compute capacity and to contract additional compute capacity up to one gigawatt when Microsoft, NVIDIA, and Anthropic announced their partnership in November 2025. Claude in Microsoft Foundry accelerated by NVIDIA GB300 GPUs on Azure builds on the strategic partnership. The compute runs in Azure data centers on NVIDIA Blackwell systems. The inference orchestration is still Anthropic-managed.
That distinction matters because OpenAI models on Azure are first-party: Microsoft operates inference, data stays within the Azure trust boundary, and EU data zone deployments are available. Claude on Foundry is a third-party marketplace offering where Anthropic operates the inference and is the data processor and SLA provider. The compute is real estate and it is in Azure. The inference layer is not Azure-operated yet, which means the data-residency guarantee that applies to Bedrock and Vertex does not extend to Foundry.
When estimate.pro calls GPT-4o through Azure OpenAI Service in Sweden Central the request stays inside Azure's EU trust boundary because Microsoft operates the entire inference stack. When a Foundry customer in Sweden Central calls Claude the prompts and outputs are processed using Azure's global infrastructure to dynamically route traffic to the data center with best availability for each request, and data may be processed for inferencing in any Azure location even though data stored at rest remains in the designated geography.
The region listing says coming 2026 and it is already September
Anthropic's regional compliance page lists Microsoft Foundry in Europe as "Coming 2026". A Microsoft Q&A question asking for a more specific timeline has been open since April with zero answers. The announced deployment types for Claude on Foundry include Global Standard and US Data Zone. The EU Data Zone that would match the deployment options available on Bedrock and Vertex is not live yet and the public timeline is a year with no month attached.
The 120-route construction ERP at coenconstruction.com uses Azure OpenAI Service because the Azure tenant, authentication, billing, and governance were already in place when we added LLM calls in 2024. If the customer were EU-based and we wanted to use Claude the Foundry path would not meet data-residency requirements. The workaround would be to deploy Bedrock in Frankfurt or Vertex in Belgium and integrate a second cloud provider for a single model.
That is not a configuration change. It is a migration that touches authentication, cost allocation, network policies, and vendor onboarding. The third-party deployment documentation scopes its data residency and compliance guidance specifically to Vertex AI and Bedrock deployments, stating "this section applies when using Vertex AI or Bedrock" and noting that "inference runs in your cloud tenant," meaning the data residency guarantees that European enterprises rely on when using Claude through AWS Bedrock or Google Vertex AI do not extend to Foundry today.
The model works. The region does not. That is not a bug—it is the deployment timeline.
GA means the API is live not that every customer can use it
When a vendor announces general availability the expectation is that the service is production-ready for all customer segments. Claude models reached GA on Microsoft Foundry with Azure-native billing and governance, but no European data zone exists. The service is production-ready for US customers and customers without strict data-residency requirements. It is not production-ready for EU customers in regulated industries where GDPR Article 44 requires that data transfers outside the EU use adequacy decisions or appropriate safeguards.
The Foundry integration solves real procurement and billing friction for Azure customers. It does not solve the jurisdiction problem that determines whether a customer can legally route production data through the endpoint. When a model ID changes the integration stops working until the code updates. When a region is listed as "coming 2026" with no further detail the integration cannot start working until the region deploys, and September is close enough to the end of the year that "2026" is starting to mean "maybe Q1 2027."
The model shipped. The region is in progress. The compute capacity is contracted and running. The inference orchestration is Anthropic-managed until the Azure-native path finishes rolling out. That architecture makes sense for a vendor scaling rapidly on hyperscaler infrastructure. It does not make sense for a compliance officer evaluating whether a deployment meets GDPR data-residency requirements today.