What sets the price
- Marketplace vs dedicated. Marketplaces and community clouds rent other people's servers and are usually the cheapest per hour. Dedicated servers in Tier III datacenters cost more but come with a contract and predictable uptime.
- Term. Committing for months lowers the rate. At Strata reserved terms are 3% to 14% below on demand.
- SXM vs PCIe. H100 SXM with NVLink costs more than H100 PCIe and is the one to pick for multi-GPU training.
Strata H100 prices
| Term | Per GPU-hour | 8×H100 node per month |
|---|---|---|
| On demand | $3.23 | $18,863 |
| 3 months | $3.13 | $18,297 |
| 6 months | $3.04 | $17,731 |
| 12 months | $2.91 | $16,977 |
| 24–36 months | $2.78 | $16,222 |
Monthly cost = price × 8 GPUs × 730 hours. Prices as of 2026-10-23.
H100 prices by provider
| Provider | H100 on demand, per GPU-hour |
|---|---|
| Strata GPU (dedicated) | $3.23 |
| RunPod Community Cloud | $2.69 |
| RunPod Secure Cloud | $3.99 |
| Lambda (8-GPU instance) | $3.99 |
| CoreWeave (8-GPU node) | $6.16 |
| AWS p5.48xlarge (8× H100) | $6.88 |
List prices from each provider's pricing page, checked 2026-10-23. 8-GPU node prices divided by 8.
How the market moved
H100 rates fell to about $1.70 an hour in late 2025 and climbed back above $3 by September 2026 as demand for inference grew. See GPU rental prices in 2026 for the full story.

