NVIDIA H100 price per hour
The cheapest on-demand NVIDIA H100 (80 GB) we found is $2.20/GPU-hr at DeepInfra (managed endpoint). Across 12 listings, prices run up to $8.00. Hopper generation and still the most widely available training and inference GPU.
| Provider | $/GPU-hr | Type | Variant | Notes | |
|---|---|---|---|---|---|
| DeepInfra | $2.20 | Managed endpoint | 80GB | Dedicated deployment, billed per minute | source |
| RunPod | $2.89 | Cloud VM | PCIe | source | |
| Vast.ai | $3.23 | Marketplace | SXM | Median of 7 live 1x offers on Oct 12; lowest was $1.80 (host-set prices) | source |
| Crusoe | $3.90 | Cloud VM | HGX | source | |
| Modal | $3.95 | Serverless | SXM5 | Billed per second ($0.001097/s) | source |
| Lambda | $3.99 | Cloud VM | SXM | Instance size not stated on page | source |
| RunPod | $3.99 | Cloud VM | SXM | source | |
| Together AI | $3.99 | Cloud VM | HGX cluster | Preemptible $1.99; 91–180 day reserved $3.19 | source |
| Nebius | $4.50 | Cloud VM | HGX | Up to 35% less with multi-month reservations | source |
| CoreWeave | $6.16 | Cloud VM | HGX 8x | $49.24 per 8-GPU instance-hour | source |
| Baseten | $6.50 | Managed endpoint | 80GB | Billed per minute ($0.10833/min) | source |
| Fireworks | $8.00 | Managed endpoint | 80GB | On-demand deployment | source |
Per-GPU prices for 8-GPU instances are the instance price divided by 8. Reserved and preemptible prices are lower and are listed in the notes where the provider publishes them. Last checked 2026-10-12.