Cloud GPU Provider Inventory Report 2026: Who Has the Most GPUs?

Which cloud GPU provider actually has inventory when you need it? We track 5,124 live instances across 54 providers in real time. The distribution is far more concentrated than most AI teams realize: four providers (GCP, RunPod, AWS, Azure) hold 75% of all listed instances.

The Full Provider Inventory Table

Key Findings

1. GCP Dominates Instance Count — But Not Price

GCP lists 2,044 instances — 39.9% of the entire tracked market. But most of those are older GPUs (T4, L4, V100) across many regions and configurations. For H100 at competitive prices, Vast.ai ($0.53/hr) and Verda ($0.80/hr) are 2-3x cheaper than GCP ($1.41/hr).

2. RunPod Has the Widest GPU Selection

RunPod lists 32 distinct GPU models — twice as many as the next competitor (Shadeform with 24). If you need a specific GPU — A40, RTX 6000 Ada, H200, B200, or even an older RTX 3090 — RunPod is the most likely place to find it.

3. AWS Has the Narrowest Selection

Despite having 619 instances, AWS only offers 7 GPU models. Hyperscalers optimize for consistency, not variety. If your workload needs a specific GPU not in their lineup — or needs the cheapest H100 — a specialized provider is a better fit.

4. The Long Tail: 46 Providers Share 10%

The bottom 46 providers — including Datacrunch, Scaleway, OVHcloud, TensorDock, and 42 others — collectively account for ~10% of listed inventory. Many of these offer the best prices for niche GPUs (e.g., A40, L40S, MI300X) or specific regional requirements.

H100 Inventory: Where to Find It Cheapest

The H100 is available across 10+ providers. Here is the lowest on-demand price available from each provider as of April 2026: