Unable to deploy Claude Sonnet 4.6 in Azure AI Foundry. All supported regions show "No Quota Available". Need guidance.

Gopi Danusu 0 Reputation points
2026-07-21T10:35:58.13+00:00

I'm trying to use the Anthropic models (specifically Claude Sonnet 4.6) in Azure AI Foundry, but every supported region shows "No Quota Available" or doesn't allow deployment.

Is there any way to use Claude Sonnet 4.6 through Azure AI Foundry despite this limitation?

Some specific questions:

  • Is there a workaround if all regions report no available quota?
  • Can quota be requested or increased for Anthropic models?
  • Are there any regions where Claude Sonnet 4.6 is currently available?
  • If Azure AI Foundry isn't an option due to capacity constraints, what is the recommended way to access Claude Sonnet 4.6 while still integrating with Azure services?

Any guidance or recent experience would be greatly appreciated.

Foundry Tools
Foundry Tools

Formerly known as Azure AI Services or Azure Cognitive Services is a unified collection of prebuilt AI capabilities within the Microsoft Foundry platform


2 answers

Sort by: Most helpful
  1. Karnam Venkata Rajeswari 5,090 Reputation points Microsoft External Staff Moderator
    2026-08-04T22:43:48.27+00:00

    Hello @Gopi Danusu ,

    Welcome to Microsoft Q&A .Thank you for reaching out to us.

    The behavior being observed can occur when the selected Claude deployment configuration does not align with the available allocation for that specific model setup. In addition to quota availability, factors such as hosting option, model version, deployment type, subscription eligibility and Marketplace onboarding can also influence whether a deployment can be created successfully.

    Regarding if there is a workaround if all regions report “No Quota Available” - A useful first step is to validate the Claude Sonnet 4.6 deployment configuration itself.

    Please follow these steps:

    1. Open: Azure AI Foundry > Model Catalog > Claude Sonnet 4.6
    2. Select: Deploy > Custom settings It is recommended to use Custom settings for troubleshooting, as the default deployment flow may automatically select a hosting configuration that does not have available allocation for the chosen model version.
    3. Review the following:
      • Hosted on value shown on the model card
      • Model version
      • Deployment type
      • Region scope
    4. If the deployment is currently using: Version 2: Hosted on Azure and the deployment blade reports “No Quota Available”, check whether the alternative supported hosting option is available: Version 1: Hosted on Anthropic infrastructure
    5. Retry the deployment after selecting the supported configuration. Quota availability is associated with a specific model version and hosting configuration. If the selected configuration does not have allocation available, switching Azure regions alone may not resolve the issue.

    Regarding regions where Claude Sonnet 4.6 is currently available -

    Claude Sonnet 4.6 availability is determined by the combination of:

    • Model version
    • Hosting option
    • Deployment type
    • Subscription eligibility

    Current deployment examples reference East US 2, and Claude deployments have also been referenced in regions such as Sweden Central. However, model availability can change over time, so the deployment experience within Azure AI Foundry and the official region availability documentation should be treated as the authoritative source.

    Recommended validation:

    1. Open the Claude Sonnet 4.6 model page.
    2. Start the deployment workflow.
    3. Review the deployment options available for the subscription.
    4. Confirm the supported deployment scope and locations displayed by Foundry.

    Please also check for the following-

    Subscription and billing

    • Active Azure subscription
    • Supported subscription type
    • Active billing configuration

    Marketplace

    • Marketplace requirements have been completed
    • Required Claude marketplace offer has been accepted

    Permissions

    • Contributor or Owner access on the target resource group
    • Deployment creation permissions
    • Marketplace purchase permissions where required

    Resource providers

    Verify provider registration using:
    az provider show \
    --namespace Microsoft.CognitiveServices \
    --query registrationState
    

    Expected Result - Registered

    If the issue remains after completing the above validations, sharing the deployment hosting option, model version, deployment settings, location, and any correlation or Activity Log details will allow further investigation into whether the behavior is related to deployment configuration, entitlement, regional availability, Marketplace onboarding or actual quota allocation.

    The following references might be helpful , please check them out

     

    Please let us know if the response was helpful

     

    Thank you

    Was this answer helpful?

    0 comments No comments

  2. AI answer

    2026-07-21T10:36:33.72+00:00

    For Claude Sonnet 4.6 in Azure AI Foundry, there is no self-service workaround when all supported regions show No Quota Available. Deployment and inference consume subscription quota, and once quota is exhausted, new deployments require either reallocating TPM from existing deployments or requesting more quota.

    Supported actions:

    1. Check whether quota is actually available for the model and region.
      • In the Foundry portal, quota availability by region can be viewed for the subscription.
      • If quota exists on other deployments of the same model, reduce or reassign TPM from those deployments to free capacity.
    2. Request a quota increase.
      • Anthropic models are explicitly supported by the quota increase request form: https://aka.ms/oai/stuquotarequest.
      • Requests are processed in order received, and priority goes to customers actively using existing quota.
      • In the Foundry portal, the Request quota action requires Owner or Contributor on the subscription. Some model/region combinations might still not support increases.
    3. Verify subscription eligibility before treating this as only a quota issue.
      • Claude models require a paid Azure subscription with an active pay-as-you-go billing method.
      • Unsupported subscription types include Cloud Solution Provider subscriptions, student/free trial/startup credit-based accounts, sponsored subscriptions using only Azure credits, and Enterprise Accounts located in South Korea.
      • Anthropic models also require a billing account in a country or region where Anthropic offers the models for purchase.
    4. Verify deployment location support.
      • Claude models support Global Standard deployments.
      • Hosted on Azure versions of claude-sonnet-5 and claude-opus-4-8 also support Data Zone Standard (US).
      • For exact Azure regions by deployment type, use the published region availability page. If a region is unsupported, deployment will fail regardless of quota.

    About region availability for Claude Sonnet 4.6:

    • The exact currently supported Azure regions are not listed in the provided material.
    • The supported way to confirm them is the Region availability by deployment type page referenced in the Claude model documentation.
    • One troubleshooting guide for the Bicep/Terraform starter mentions eastus2 as the QuickStart default, and notes that swedencentral commonly hosts Haiku, Sonnet, and Opus, while westus2 hosts Sonnet and Opus, but availability can change.

    If Azure AI Foundry is blocked by capacity constraints:

    • The documented Azure path is to keep using Foundry and either reallocate TPM or request more quota.
    • For Azure integration, the Claude on Foundry starter kit is the recommended setup path when deployment is available. It provisions a Foundry account and project and wires the Anthropic SDK and Claude Code CLI to call the deployment over Microsoft Entra ID without API keys.
    • If deployment continues to fail even when quota should exist, check for subscription eligibility, billing support, Marketplace access, and permissions. Common permission requirements are Contributor or Owner on the resource group for deployment and Cognitive Services User for Entra ID-based invocation.

    If the symptom is specifically quota appears full with no live deployments, soft-deleted Foundry or AI Services accounts can reserve TPM quota for up to 48 hours. In that case, list deleted accounts with az cognitiveservices account list-deleted -o table and purge accounts no longer needed. This is potentially destructive because purging permanently removes the deleted resource.


    References:

    AI-generated content may be incorrect. Read our transparency notes for more information.

    Was this answer helpful?

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.