Unable to Create GPU VM (NVIDIA H100/A100/T4) in Azure — Quota and Availability Issues

The main blockers shown here are subscription eligibility, quota, and regional capacity.

• GPU VM families such as NC-series can start with a regional vCPU quota of 0. For the NC family, the quota is initially set to 0 in each region, and a quota increase must be requested for that VM family in an available region.
• Some subscription types do not support these GPU SKUs. NC-series documentation states these SKUs aren’t available to trial or Visual Studio Subscriber Azure subscriptions, and the subscription level might not support selecting or deploying them.
• Free subscriptions aren’t eligible for quota increases. The documented guidance for confidential GPU VMs also says Azure free accounts don’t have enough quota for the required compute cores.
• Even with quota, deployment can still fail because of regional hardware availability. Azure guidance for GPU workloads says to check GPU instance availability by region, and if the required size is unavailable, switch region or choose another GPU VM size.

For the specific questions:

1. Does Azure allow H100 or A100 deployments for this subscription type?
   The available documentation does not confirm Azure Startup Sponsorship eligibility for H100 or A100 GPU VM deployment. What is supported by the documentation is that subscription level can block GPU SKU selection or deployment, and some subscription types cannot receive quota increases.
2. Which GPU VM families and regions are suitable?
   The documentation only supports checking this per region and per subscription:
   • Check Products available by region for VM availability.
   • In Azure portal, go to Subscription > Usage + quotas to see which GPU quota families are present and request quota where supported.
   • If a GPU size is unavailable in the current region, try another region or another GPU VM size.
3. Is there a specific quota request process for H100/A100 families?
   Yes. The supported process is to request a regional quota increase for the required GPU VM family. The documentation references quota increase requests for the target VM family in the target region.
4. Could the subscription still have GPU restrictions even with active credits?
   Yes. The documentation explicitly states that subscription level can prevent selecting or deploying GPU SKUs. Active credits do not by themselves guarantee GPU family eligibility.
5. How to verify actual regional capacity and deployment eligibility after quota approval?
   Use this sequence:
   1. In Azure portal, open Subscriptions > Usage + quotas and confirm the target GPU family has nonzero quota in the target region.
   2. Check Products available by region for the VM family.
   3. Start VM creation in that same region and verify the size is selectable under See all sizes.
   4. If quota is sufficient but deployment still fails, treat it as a likely capacity issue in that region and retry later, switch region, or choose another size.
6. Recommended alternatives if H100/A100 are restricted
   The available documentation lists these GPU VM families as suggested alternatives for GPU workloads:
   • NDv2 for demanding AI, machine learning, simulation, and HPC workloads.
   • NCads_H100_v5 for Azure Applied AI training and batch inference workloads.
   • NCasT4_v3 for AI services, real-time inferencing, and interactive graphics/visualization.

Important constraint for the requested OS:

• For confidential GPU VMs, the qualified guest OS image listed is Ubuntu 22.04 LTS. The available documentation does not support Ubuntu 24.04 LTS for confidential GPU VM images.

Practical next steps:

1. In Subscriptions > Usage + quotas, identify the exact GPU family needed in the target region and submit a quota increase request for that family.
2. Check Products available by region for the same family and region.
3. If the subscription cannot request or receive quota for that family, use a supported subscription purchase option such as pay-as-you-go.
4. If quota is approved but the VM size is still unavailable or allocation fails, switch region or use an alternative GPU family such as NDv2 or NCasT4_v3.
5. If the goal is specifically a confidential GPU VM, use a supported image such as Ubuntu 22.04 LTS.

References:

• Azure Confidential GPU options
• 'NC' sub-family GPU accelerated VM size series
• Troubleshoot AKS AI toolchain operator add-on errors
• Windows virtual machine deployment FAQs
• Modernize your NP-series virtual machines by May 31, 2027
• can't create GPU instance & can't request quota in Machine learning - Microsoft Q&A