Catalog
ND/NC/NV families as on-demand when we have a published hour. Low-priority labeled.
Provider review / verified 2026-09-05
AzureGPU review
Azure sells ND/NC/NV GPU VMs next to the Microsoft estate: Entra, Blob, Azure ML. If your company already signed the EA, the GPU hour is a rounding error in a political sense. If you are an indie lab, it is a trap.
Models listed4
From /GPU/hr$0.98
Regions in sample1
Business modelHyperscaler
Our reading
Who this provider is for
Azure GPUs are for Microsoft shops. I will not talk you out of ND-H100 if your data cannot leave the tenant. I will talk you out of Azure if you are a five-person lab with a Hugging Face repo and a dream. The catalog hour is usually not the win. The win is staying inside a compliance boundary you already paid lawyers for.
Rent it if Enterprises on Azure ML / Fabric / Entra who need NVIDIA in the same tenant.
Skip it if Everyone shopping a 4090. Startups with no Microsoft relationship hunting a cheap H100.
How we compare hoursStrengths
- Enterprise identity, networking, and compliance already in the building
- ND-family NVIDIA SKUs for people who must stay on Azure
- Reservations and hybrid benefit games if you already play them
- The procurement department knows the vendor
Limits
- On-demand GPU hours vs neoclouds: usually a loss
- Quota and regional stock, same hyperscaler tax
- Spot/low-priority is not on-demand
- You will meet a portal. It will not be RunPod
Same standard, every cloud
What we measured on Azure
Four checks we apply to every provider. This is not a score, and it is not a paid ranking.
Price signal
Hyperscaler band. EA discounts are real and invisible to a public table, we do not invent them.
Product shape
Azure VM + GPU. Platform first, GPU second.
Still verify
Quota, region, VM size. Your EA price is not our catalog price. Confirm both.
- Model
- Hyperscaler
- Headquarters
- Redmond, US
- Setup
- ND/NC/NV VMs, Azure ML
- On-demand
- Yes, quota permitting
- Spot
- Low-priority / spot, labeled
- Why people stay
- Microsoft estate
- Why people leave
- The GPU hour + portal
- Compare with
- AWS, GCP, CoreWeave
Catalog listings
Azure catalog sample
Published on-demand hours first. Spot and waitlists are labeled. Confirm the live rate before you provision.
EA gravity
If Microsoft already takes a percentage of your IT spend, adding ND-instances can be ‘free’ in the only way CFOs understand. That is not the same as cheap. GPUBeacon shows a public-shaped hour. Your EA is a private conversation. I will not fake it into a green cell.
Azure ML is not a GPU
Managed training services are a product. ND VMs are a product. Compare them separately. I rent VMs when I want nvidia-smi. I use managed services when I want Microsoft to own the plumbing. Mixing them in a blog post is how you get a surprise bill.
How I actually rent it
- 01
Confirm you must stay on Azure
If not, open Lambda or CoreWeave. If yes, continue.
- 02
Quota + region + ND size
The SKU is not ‘H100.’ It is a VM size.
- 03
Compare public vs EA
Our table is public. Your account manager is not. Use both.
FAQ
Azure questions
Is Azure cheaper than AWS for GPUs?+
It depends on EA, region, and SKU. Public list vs public list is a coin flip by generation. Both lose to many neoclouds on raw hours.
Azure vs CoreWeave?+
Azure if the tenant is the constraint. CoreWeave if the GPU cluster is the product and data can move.
Can startups use Azure GPUs?+
Yes, with credits. Credits end. Plan the hour after credits or you will do a midnight migration.
Who should skip Azure GPUs?+
Anyone who is not already a Microsoft shop and who does not have a compliance gun to their head.
Sources
Where these numbers come from
Prices and SKUs are from the GPUBeacon catalog sample. Ranking rules are on the methodology page. Outbound links may be affiliate. They never change the hour we display.
GPUs
Keep comparing
Other providers to compare
Same fields, different product. Compare the GPU hour before you lock a brand.
