ComputeTape H100 Index · CT-H100
What does an on-demand H100 cost per hour?
The ComputeTape H100 Index is the median verified on-demand H100 GPU list price across sourced providers — a single number you can cite, chart, and embed. Directional, not exchange-traded.
Median on-demand list price across 6 sourced providers · as of July 21, 2026 · methodology
- Spot median
- $2.46 normalized estimate
- Blended H100e
- $5.41 normalized estimate
- Weekly move
- Flat
Use it
Embed, cite, or download the index
Embed the current daily number in your own page, cite it in a piece, or pull the full history as CSV or JSON.
Embed the current card
A card refreshed with the daily build and attributed back to this page. Choose a theme to match your site.
Cite this
Download the data
Download the current canonical provider rows or the daily benchmark history.
Provenance
The 6 prices behind the median
Each is a verified on-demand H100 list price, normalized to a per-GPU hourly rate. Full per-provider detail and sorting are in the Provider Directory.
| Provider | On-demand $/H100-hr | Region | Confidence | Source |
|---|---|---|---|---|
| Hyperstack | $3.20 | Multi-region | High | Price page |
| Crusoe | $3.90 | Multi-region | High | Price page |
| Lambda | $4.29 | Multi-region | High | Price page |
| CoreWeave | $6.16 | us | High | Price page |
| AWS | $6.88 | us-east-1 | High | Price page |
| Azure | $12.29 | eastus | High | Price page |
How it's built
How the index is built
The headline is a raw median of verified on-demand H100 list prices — one per provider, each checked against the provider's public price page and normalized to a per-GPU hourly rate. The spot and blended H100-equivalent figures are normalized estimates derived from public provider data, shown for context and clearly labeled as estimates — never presented as quoted prices. See the full methodology.
ComputeTape tracks directional AI compute market signals, not exchange-traded prices. Use the index as a market indicator, not investment advice. More signals — enterprise compute cost, model serving cost, GPU availability — available on the Market page.