Azure Foundry: Inability to Deploy Anthropic Claude Sonnet 5 Due to Zero Quota Allocation

RB-0330 0 Reputation points
2026-09-29T18:02:31.14+00:00

Problem descriptionI am trying to deploy Anthropic Claude Sonnet 5 Version 2, Hosted on Azure, in Microsoft Foundry. Deployment fails with “Insufficient quota.”In Foundry Management Center → Quota, Claude Sonnet 5 currently shows:Global Standard: 0/0 TPMUS Data Zone Standard: 0/0 TPMOther Anthropic Claude models also show 0/0 TPM.The Foundry resource is in a newly created direct Microsoft Pay-As-You-Go Azure subscription with an active payment method. The Anthropic Claude Sonnet 5 CCU Marketplace offer has also been accepted.Microsoft’s current Claude quota documentation lists Claude Sonnet 5 on Pay-As-You-Go with default limits of 40 RPM, 40,000 ITPM, and 8,000 OTPM for Global Standard and US Data Zone Standard. However, my PAYG subscription is receiving 0/0 TPM. [learn.microsoft.com]Microsoft’s Claude deployment documentation also states that a paid Azure subscription with an active pay-as-you-go billing method is supported, while CSP, Free Trial, Student, credit-based, and certain Sponsored subscriptions are not supported. [learn.microsoft.com]EnvironmentDirect Microsoft Pay-As-You-Go Azure subscriptionActive payment methodMicrosoft Foundry resource in West US 3Claude Sonnet 5 CCU Marketplace offer acceptedClaude Sonnet 5 Version 2, Hosted on AzureIntended deployment: Global StandardExisting CSP subscription is not being used for this Foundry environmentWhat I've already triedInitially attempted this using an existing CSP subscription, then created a new direct Microsoft Pay-As-You-Go subscription because CSP subscriptions are unsupported for Claude. [learn.microsoft.com]Created a new Foundry resource under the direct PAYG subscription.Accepted the Anthropic Claude Sonnet 5 CCU Marketplace offer.Verified the Foundry quota page.Attempted to deploy Claude Sonnet 5 Hosted on Azure using Global Standard.Submitted the Anthropic quota request form.The quota request was denied due to regional capacity and instructed me to expand my regional choices.The Anthropic Global Standard quota request form did not provide a region-selection field.Current status / questionsI am trying to determine why an eligible direct PAYG subscription is receiving 0/0 TPM when Microsoft’s current documentation lists nonzero default limits for Claude Sonnet 5 PAYG subscriptions. [learn.microsoft.com]Specifically:Can Microsoft verify that the Anthropic/Foundry quota backend is correctly recognizing this subscription as Pay-As-You-Go?Is the 0/0 TPM state caused by missing initial Claude quota enablement/allocation, or is it solely a capacity constraint?If it is a capacity constraint, which Foundry region or deployment scope currently has actual capacity for new Claude Sonnet 5 Hosted on Azure allocations, rather than simply listing the model as available?Does recreating my Foundry resource in another region affect Global Standard initial quota allocation, given that Claude Global Standard quota is documented as subscription-level and shared across regions? [learn.microsoft.com]How should I “expand my regional choices” as instructed by the quota rejection when the Anthropic Global Standard quota form does not provide a region-selection field?If backend investigation is required, please escalate this to the team responsible for Anthropic Claude subscription eligibility, initial quota provisioning, or capacity allocation.I am willing to recreate the Foundry resource in another region or use US Data Zone Standard if necessary, but I would like to know which option currently has actual capacity before rebuilding the environment.

Azure OpenAI in Foundry Models
0 comments No comments

1 answer

Sort by: Most helpful
  1. Divyesh Govaerdhanan 11,890 Reputation points MVP Volunteer Moderator
    2026-09-29T18:14:15.25+00:00

    Hello RB-0330,

    Welcome to Microsoft Q&A,

    Your setup isn't wrong. West US 3 is a supported region for claude-sonnet-5, and PAYG subscriptions are documented to get 40 RPM / 40,000 ITPM / 8,000 OTPM by default for this model on Global Standard, so the 0/0 you're seeing isn't a config mistake.

    Claude models go through a separate initial capacity-gating step beyond the published default-limits table. The table shows what your subscription tier is entitled to, but that allocation still needs to be turned on for your specific subscription before it shows as usable quota. This is a known pattern, several other eligible PAYG subscriptions have hit the same 0/0 starting point, including ones asking this exact question for other Claude models, and it wasn't a configuration issue on their end either.

    What to do:

    1. Resubmit the quota request form: https://aka.ms/oai/stuquotarequest. It has no region dropdown, so put your subscription ID, the model (claude-sonnet-5), deployment type (Global Standard), and West US 3 directly in the notes field.
    2. Hosted on Azure (version 2) and Hosted on Anthropic infrastructure (version 1) draw from separate quota pools for the same model. If one stays at 0, try deploying the other version from the model card, it's a genuinely different allocation, not a workaround.
    3. A denial for "regional capacity" means there's no free backend capacity for that model/region combination right now, not that the request was invalid. claude-sonnet-5 is also supported in eastus, eastus2, centralus, southcentralus, westcentralus, and westus, so if West US 3 specifically is the constraint, requesting one of those may move faster.
    4. If quota is still 0/0 after resubmitting, open a support request through Help + Support in the portal and reference your quota request ticket number, that's the documented path when self-service allocation doesn't resolve it.

    Reference: Claude model quotas and rate limits and Deploy and use Claude models in Microsoft Foundry

    Please click Accept Answer and upvote if this helped.

    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.