Clarification needed – Custom Neural model training and processing region for resource in Germany West Central

Document AI 0 Reputation points
2026-08-17T13:55:18.5366667+00:00

Our Document Intelligence resource is located in Germany West Central. Public documentation states that Custom Neural model training is only supported in a specific list of regions, which does not include Germany West Central. However, we are able to start Custom Neural training directly on our Germany West Central resource without any region error.

Could you please confirm, in writing:

  1. Where does the actual training computation take place when we run Custom Neural training on a Germany West Central resource – is it genuinely performed in Germany West Central, or silently routed to another region?
  2. When we call the Analyze operation against our Germany West Central resource endpoint after training, is the document processing (inference) performed exclusively in Germany West Central?

This is needed for our GDPR data residency documentation, so a resource-specific written confirmation would be very helpful.

Azure Document Intelligence in Foundry Tools

1 answer

Sort by: Most helpful
Most helpful Newest Oldest
  1. SRILAKSHMI C 19,730 Reputation points Microsoft External Staff Moderator
    2026-09-01T16:42:34.6533333+00:00

    Hello @Document AI

    Thank you for reaching out to Microsoft Q&A.

    We understand that you require a clear confirmation of the training and inference processing locations for your Document Intelligence resource in Germany West Central, specifically for GDPR data-residency documentation.

    According to the current Microsoft documentation, Custom Neural model training is officially supported only in the regions listed in the Custom Neural model documentation, and Germany West Central is not currently included in that list. Microsoft also documents that a model trained in one of the supported regions can be copied to another region and used there.

    At the same time, the Document Intelligence data, privacy, and security documentation states that incoming documents are processed in the same region where the Document Intelligence resource was created. It also states that, for customer-trained models, the trained models are stored in Azure Storage in the same region.

    Microsoft's disaster recovery guidance further states that once a Document Intelligence resource is created in a region, the resource and its operations remain associated with that Azure region.

    However, the public documentation does not currently explain the apparent discrepancy in your scenario—namely, why Custom Neural training can be initiated successfully on a Germany West Central resource even though Germany West Central is not listed as a supported Custom Neural training region. The documentation also does not provide a resource-specific statement identifying the backend compute location of that particular training job.

    Therefore, we would not want to make an assumption that the training computation is performed in Germany West Central solely because the training request was initiated through a Germany West Central resource.

    For the Analyze/inference operation, the published Document Intelligence privacy documentation does state that incoming data is processed in the same region as the Document Intelligence resource. Based on that documentation, processing submitted to your Germany West Central Document Intelligence resource is associated with Germany West Central.

    Please refer this

    Data, privacy, and security for Document Intelligence

    I Hope this helps. Do let me know if you have any further queries.

    If this answers your query, please do click Accept Answer and Yes for was this answer helpful.

    Thank you!

    Was this answer helpful?

    0 comments

Your answer

Answers can be marked as 'Accepted' by the question author and 'Recommended' by moderators, which helps users know the answer solved the author's problem.