Anthropic Claude deployment fails with InternalServerError despite valid ModelProviderData - Microsoft.SaaS resource never created (India MCA, countryCode=IN)

Sabarish Dhanarajan 10 Reputation points

SUMMARY

=======

Deployment of Anthropic Claude models on Azure AI Foundry consistently fails with InternalServerError after PUT is accepted (HTTP 201). The Microsoft.SaaS/resources list remains empty throughout, indicating the Anthropic organization / SaaS subscription provisioning never completes on the Azure-Anthropic integration path.

SUSPECTED ROOT CAUSE

====================

Based on testing, I believe the failure is caused by the combination of:

  • modelProviderData.organizationName = <your-organization-name>
  • modelProviderData.countryCode = IN
  • Subscription billing country = IN (MicrosoftCustomerAgreement)

The Marketplace / Anthropic SaaS subscription creation appears to be silently rejecting the request at the eligibility layer for this organization name and / or country code combination. No Microsoft.SaaS resource is ever created, and no actionable error is surfaced to the customer.

Please confirm:

  1. Whether countryCode = "IN" is a supported value for Anthropic model deployments on Azure AI Foundry.
  2. Whether there are any restrictions on the organizationName field (length, characters, registered entity match, etc.) that could be causing the Marketplace step to fail.
  3. Whether subscriptions billed in India under MCA are eligible for Anthropic Marketplace offers.

ENVIRONMENT

===========

Subscription type: MicrosoftCustomerAgreement (MCA)

Billing country: IN

Resource group: <your-resource-group>

User: <your-user-email>

ACCOUNTS TESTED

===============

  1. <sweden-central-account-name> (Sweden Central)
  2. <eastus2-account-name> (East US 2)

Both accounts: Kind=AIServices, SKU=S0.

Both deployments failed identically.

MODEL DEPLOYMENT REQUEST

========================

Model: claude-sonnet-4-6

Format: Anthropic

Version: 1

SKU: GlobalStandard

Capacity: 2000

API version tried: 2025-10-01-preview (also tested 2025-06-01)

versionUpgradeOption: OnceNewDefaultVersionAvailable

modelProviderData provided:

industry: Technology

organizationName: <your-organization-name>

countryCode: IN

PREREQUISITES VERIFIED

======================

Microsoft.SaaS provider: Registered

Microsoft.CognitiveServices provider: Registered

Microsoft.SaaS/resources list: EMPTY (before, during, and after deployment attempts)

REPRODUCIBLE BEHAVIOR

=====================

  1. PUT deployment request -> HTTP 201 Accepted
  2. provisioningState = "Creating" for approximately 6 minutes
  3. Silent transition to provisioningState = "Failed"
  4. No actionable statusMessage in the deployment resource
  5. Microsoft.SaaS/resources list remains empty

ACTIVITY LOG ERROR (identical in both regions)

==============================================

{

"status": "Failed",

"error": {

"code": "ResourceOperationFailure",

"message": "The resource operation completed with terminal provisioning state 'Failed'.",

"details": [

 {

 "code": "InternalServerError",

 "message": "Internal Server Error."

 }

]

}

}

FAILURE TIMESTAMPS (UTC)

========================

Sweden Central attempt: 2026-05-27T08:58:14Z (failed)

Sweden Central attempt: 2026-05-27T09:17:37Z (failed, retry after Microsoft.SaaS registered)

East US 2 attempt: 2026-05-27T09:36:33Z (failed)

WHAT I HAVE ALREADY TRIED

=========================

  1. Switched API version from 2025-06-01 to 2025-10-01-preview (resolved InvalidModelProviderData, but exposed underlying InternalServerError).
  2. Registered Microsoft.SaaS resource provider (was NotRegistered initially); confirmed Registered before retry.
  3. Deleted and recreated the deployment.
  4. Tested a second region (East US 2) with a fresh Cognitive Services account - identical failure.
  5. Verified modelProviderData is correctly populated in the request body and echoed back in the deployment resource response.
  6. Verified Microsoft.CognitiveServices provider is Registered.

WHAT I NEED FROM SUPPORT

========================

  1. Confirm whether countryCode = "IN" and / or India-billed MCA subscriptions are eligible for Anthropic Marketplace offers.
  2. Confirm whether the organizationName field has any format or validation requirements that could be causing the SaaS subscription creation to fail.
  3. If the configuration is supported in principle, please investigate and apply the backend fix for the Anthropic organization provisioning step.
  4. Share the underlying Marketplace / SaaS error that is not currently surfaced to the customer in the activity log.

This is currently blocking our AI integration work. Appreciate your help.

  1. SRILAKSHMI C 19,195 Reputation points • Microsoft External Staff • Moderator

    Hi @Sabarish Dhanarajan

    I will check this further with the backend engineering team and get back to you with an update. Kindly allow us some additional time for investigation.

    Thank you!

  2. Sabarish Dhanarajan 10 Reputation points

    Hi @SRILAKSHMI C ,

    Thank you for picking this up and for engaging the backend engineering team.

    To help speed up the investigation, here are some additional diagnostic details the engineering team may find useful:

    KEY FINGERPRINT OF THE FAILURE

    ==============================

    • PUT deployment request returns 201 Accepted
    • provisioningState = "Creating" for approximately 6 minutes
    • Silent transition to provisioningState = "Failed"
    • Activity log shows: ResourceOperationFailure / InternalServerError with no actionable message
    • Microsoft.SaaS/resources list remains EMPTY before, during, and after the attempt
    • Identical failure pattern reproduced in two regions: Sweden Central and East US 2
    • Reproduces with both API versions 2025-06-01 and 2025-10-01-preview
    • ModelProviderData is correctly accepted (industry, organizationName, countryCode all echoed back in the deployment resource)

    SPECIFIC QUESTIONS FOR THE BACKEND TEAM

    =======================================

    1. Is countryCode = "IN" currently a supported value for Anthropic model deployments on Azure AI Foundry?
    2. Are subscriptions billed in India under MicrosoftCustomerAgreement eligible for Anthropic Marketplace offers, or is there an eligibility restriction at the Marketplace / SaaS subscription layer?
    3. Could the team check the underlying Marketplace / SaaS subscription creation log for my subscription? The error appears to be silently swallowed before reaching the activity log.
    4. If this is an eligibility issue rather than a provisioning bug, could you confirm officially so we can plan accordingly?

    TIMELINE

    ========

    This is currently blocking our AI integration work, so any indication of expected investigation time would be appreciated. Happy to provide any additional logs, trace IDs, or run further reproduction steps if it helps the engineering team.

    Thank you again for your help.

  3. SRILAKSHMI C 19,195 Reputation points • Microsoft External Staff • Moderator

    Hi @Sabarish Dhanarajan

    Thank you for providing the details. I will review the information with our backend/Product Group team for further investigation. Please allow some time for us to analyze the issue. Once I receive an update from the team, I will let you know as soon as possible. Thank you for your patience.

    Thank you!

  4. SRILAKSHMI C 19,195 Reputation points • Microsoft External Staff • Moderator

    Hi @Sabarish Dhanarajan

    could you please provide the details asked in private message.

    Thank you!

Sign in to comment

1 answer

  1. AI answer

    1. Anthropic Claude models on Microsoft Foundry currently require an Enterprise or MCA-E subscription. MicrosoftCustomerAgreement (MCA) subscriptions that are not MCA-E are listed as ineligible. This is surfaced in the documentation as a “Subscription eligibility error” with the note that “Claude models require an Enterprise or MCA-E subscription.” Because of this requirement, an MCA subscription billed in India that is not MCA-E is not eligible for Claude usage in Foundry, regardless of countryCode or region. The behavior described (deployment accepted, then failing with InternalServerError and no SaaS resource created) is consistent with a backend eligibility failure on the Marketplace side.
    2. The documented prerequisites and troubleshooting guidance for Claude on Foundry do not define any special validation rules for organizationName or countryCode in modelProviderData (such as length, character set, or legal-entity matching). The only documented eligibility constraints are:
      • Subscription type: must be Enterprise or MCA-E.
      • Region: deployments must be in East US 2 or Sweden Central.
      • Marketplace access enabled for the subscription.
      There is no documentation indicating that countryCode = "IN" is unsupported or that organizationName formatting would cause deployment failure.
    3. For the scenario described (MCA subscription, billing country IN, Claude deployment in East US 2 / Sweden Central), the supported and documented checks are:
      • Ensure the subscription is Enterprise or MCA-E (non-MCA-E MCA subscriptions are not eligible).
      • Ensure Azure Marketplace access is enabled for the subscription for partner models.
      • Ensure deployment is in a supported region (East US 2 or Sweden Central).
      If these conditions are not met (for example, the subscription is MCA but not MCA-E), Claude deployments can fail even when the PUT returns 201 and the modelProviderData is valid. The internal Marketplace/SaaS error details are not exposed in the public documentation.

    Given the current documentation:

    • countryCode = "IN" is not listed as unsupported.
    • organizationName restrictions are not documented as a cause of failure.
    • The blocking factor for Anthropic Claude on Foundry is subscription eligibility: only Enterprise and MCA-E subscriptions are supported.

    To proceed, use an eligible Enterprise or MCA-E subscription that meets the Marketplace access and region requirements, then redeploy the Claude model.

    References:

    AI-generated content may be incorrect. Read our transparency notes for more information.
Sign in to answer

Your answer