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—the models can't be deployed or used practically.
I already submitted a quota increase request, but it was denied. We're a small business without an account team, negotiated commitment, or provisioned-throughput reservation. I've checked East US, East US 2, and Sweden Central, with the same result. Both the V1 Anthropic-hosted and V2 Azure-hosted model options appear unavailable.
For organizations that have successfully deployed Claude through Foundry, what actually worked? Is pay-as-you-go still realistic for a new, low-spend tenant, or is provisioned throughput, a reservation, or a larger commitment effectively required? Did additional Azure spend, a particular model or region, a different type of support ticket, or involvement from a reseller or account manager make the difference?
I'd also like to know whether any region still has on-demand Claude capacity and which approaches are a waste of time. Direct Anthropic access is an option, but keeping billing, identity, and data handling within Azure is preferable for this regulated production use case.
3 Answers
Another option is Azure Databricks Model Serving, which may expose Claude through a different service and quota path than Foundry’s native deployments. Before investing in a workspace, confirm whether the service uses a separate capacity pool and which Claude models are actually available. If the same Anthropic restrictions apply, going directly through Anthropic with an API key may be faster.
If you don’t already have an account manager, getting Claude capacity through Foundry can be difficult. The practical enterprise route seems to be building Azure spend, working through an account team or partner, and then requesting a tier or quota change. Provisioned throughput or another commitment may be more realistic than repeatedly submitting the same pay-as-you-go request.
That’s the concern. I have no account manager and the first quota request was denied, so I’m trying to determine whether there is a workable path for a small tenant or whether an Azure partner or reseller is effectively required.
Anthropic capacity appears to be unusually limited across Azure right now, including for larger customers. Foundry may have internal quota tiers that increase with usage, but Microsoft may not be able to manually raise them on demand. Some people have found the Anthropic-hosted V1 versions available when the Azure-hosted V2 versions were not, so it’s worth checking both—but availability is not guaranteed.
I checked both V1 and V2, and neither has usable capacity on this subscription. East US, East US 2, and Sweden Central all show the same problem, so this may simply be a tenant stuck at the lowest quota tier.
I’m considering Databricks, but I’d want to verify that it really uses a separate quota pool rather than inheriting the same subscription-level restriction. Direct Anthropic access is the fallback, although it loses the Azure billing and tenant integration I was aiming for.