H100 cost per hour - Last verified July 2026
Nvidia H100 cost per hour across 17 GPU cloud vendors
Per-GPU per-hour H100 rates pulled from the public pricing pages on the 17 vendor pages on this site. Sorted cheapest first. Quote-only vendors are listed at the bottom and we do not infer rates for them.
| Vendor | Pricing model | Configuration | Per GPU per hour |
|---|---|---|---|
| Vast.ai | marketplace | Marketplace floor, interruptible | $1.490 |
| RunPod | on-demand | Secure Cloud, per-GPU | $2.990 |
| Together AI | on-demand | Reserved cluster, 91-180 day, per-GPU | $3.090 |
| Hyperstack | on-demand | On-demand | $3.200 |
| Crusoe Cloud | on-demand | On-demand, per-GPU on 8x | $3.900 |
| Modal | per-second | Per-second, hourly equiv | $3.950 |
| Lambda Labs | on-demand | On-demand, per-GPU on 8x | $3.990 |
| DigitalOcean GPU Droplets | on-demand | On-demand, per-GPU (1 or 8x) | $4.410 |
| Replicate | per-second | Per-second, hourly equiv | $5.490 |
| Paperspace | on-demand | Per-GPU on 8x | $5.950 |
| CoreWeave | on-demand | HGX H100 8x node, per-GPU | $6.155 |
| AWS EC2 | on-demand | On-demand, 8x per node, per-GPU equiv | $6.880 |
| Oracle Cloud (OCI) | on-demand | Bare metal, per-GPU equiv | $10.000 |
| Google Cloud | on-demand | On-demand, per-GPU equiv | $11.060 |
| Azure | on-demand | On-demand, per-GPU equiv | $12.290 |
| Fly.io GPU | quote-only | Quote only - contact Fly.io GPU sales | Quote only |
| OVHcloud GPU | quote-only | Quote only - contact OVHcloud GPU sales | Quote only |
How to read this table
Per-GPU per-hour means the rate divided across all GPUs in the smallest billable unit (typically an 8-GPU HGX node). Hyperscaler on-demand list rates are the published rate before any negotiated discount, Savings Plan, Reserved Instance, or Capacity Blocks commitment. Specialist clouds (CoreWeave, Lambda, Together AI) usually publish reservation rates that are 30 to 60 percent below their on-demand. Marketplace floor (Vast.ai) is interruptible.
Last verified July 2026.