I'm working with a relatively new Azure tenant, and every Anthropic Claude model I've checked in Azure AI Foundry shows no usable quota or capacity. This is more than a low TPM limit: I can't deploy or run the models in a practical way. I already submitted a quota-increase request, but it was denied. We're a small business without an account team or a pre-negotiated Azure commitment.
For people who have actually gotten Claude running in Foundry, what is the practical route? Is pay-as-you-go still realistic for a new or low-spend tenant, or do you effectively need provisioned throughput, a reservation, or another commitment? If a first support request was denied, what changed when a later request succeeded—more Azure spending history, a particular SKU or region, switching from standard to provisioned capacity, or filing a capacity-specific ticket?
I've already checked East US, East US 2, and Sweden Central, and both the V1 and V2 Anthropic deployment options appear unusable. I'm also interested in whether any region and model combination still has on-demand capacity, or whether Claude availability is essentially limited to certain US regions and enterprise customers. I'd prefer to keep the workload within Azure for billing, identity, and data-governance reasons, although direct Anthropic access is an option if Foundry is effectively unavailable.
What approaches have worked for small organizations, and which steps are just likely to waste time?
4 Answers
The fastest answer may be to use Anthropic directly rather than treating Foundry availability as a deployment problem. Direct API access or a coding tool that accepts an API key can avoid Azure’s model-specific capacity limits. That won’t satisfy requirements around consolidated Azure billing, tenant identity, or regional governance, but if those are preferences rather than hard requirements, it may save weeks of support tickets.
A different route worth checking is Azure Databricks model serving. It may expose Claude through a separate serving and quota path instead of the exact Foundry allocation, so it could be worth verifying before giving up. I would confirm the specific model, region, pricing, and whether the workspace still requires Anthropic capacity approval before investing time in setting it up.
That’s the most plausible alternative I’ve heard, but I’d want confirmation that it works for a new tenant and isn’t just another interface over the same restricted quota. I also need to know which Claude models are actually available there.
If the workload must stay in Azure, the realistic enterprise path may be an Azure account team, CSP partner, or reseller that can discuss a commitment or provisioned-throughput arrangement. Simply resubmitting the same quota form for another region is unlikely to help when multiple regions and both Anthropic hosting options already show zero capacity. Any request should clearly describe the production use case, expected token rates, required regions, compliance constraints, and willingness to use a paid commitment.
That seems to be the practical route if pay-as-you-go is effectively gated. The missing piece is whether a small company can access that process through a partner without first having substantial Azure spend.
Anthropic capacity on Azure appears to be unusually constrained right now, and the available quota can depend on internal tenant or usage tiers. Some people have reportedly received increases after building Azure spend history and submitting another request, but there doesn’t seem to be a reliable self-service way to force that change. A new tenant without an account team may simply be competing for the same limited pool as much larger customers.
That matches what I’m seeing. I’m mainly trying to determine whether anyone has actually moved a small tenant off the zero-capacity tier without an account manager, rather than just receiving a slightly larger limit.
I agree that direct Anthropic is the clean fallback, especially for coding workloads. I’m still checking whether Foundry can be made viable because keeping billing, access control, and data handling in Azure matters for this production use case.