Market
AI compute capacity and availability signals, plus a directional H100-equivalent market rate. Verified per-provider GPU prices live in the Provider Directory.
Where AI compute capacity can actually be requested, and which way the blended market rate is moving — built from public provider data and normalized estimates, not exchange-traded prices. Per-provider list prices live in the Provider Directory.
ComputeTape tracks directional AI compute market signals, not exchange-traded prices. The capacity board and market rate combine public provider pricing, observed marketplace data, availability checks, and normalized H100-equivalent estimates. Use them as market indicators, not investment advice.
Anchor price
The market's anchor: the on-demand median to rent one H100 GPU for an hour, taken from reviewed public provider pricing and timestamped — a sourced market indicator, not an exchange price. Median across 7 reviewed providers, as of 2026-07-01.
What it costs to rent one H100 GPU for an hour on demand — the market’s anchor price. On-demand median · CT-H100 · Flat day
Access
Availability and access signals for major AI compute providers — where capacity can actually be requested. Updated weekly. Last refreshed July 1, 2026 at 6:00 AM EDT.
| Provider | H100 | H200 | B200 / GB200 | MI300X | Large Cluster | Quota Friction | Status |
|---|---|---|---|---|---|---|---|
| AWS | Yes | Yes | Emerging | No | Strong | High | Constrained |
| Azure | Yes | Yes | Emerging | Some | Strong | High | Constrained |
| Oracle | Yes | Yes | Emerging | No | Strong | Medium | Limited |
| CoreWeave | Yes | Yes | Emerging | No | Strong | Medium | Available / tight |
| Lambda | Yes | Some | No | No | Moderate | Low/Med | Available |
| RunPod | Yes | Some | No | Some | Lower | Low | Variable |
| Crusoe | Yes | Some | Emerging | No | Strong | Medium | Expanding |
| Nebius | Yes | Some | Emerging | No | Strong | Medium | Expanding |
Directional signal
A single directional estimate of the blended market rate for an H100-equivalent GPU-hour, normalized from public provider data. A market indicator, not an exchange price or investment advice. Last updated July 1, 2026 at 6:00 AM EDT.
The enterprise compute cost, model serving cost, and GPU availability score remain available in Compute College lessons and the methodology reference.
See the H100 Index →Providers
Per-provider on-demand H100 list prices — each verified against the provider's public price page, normalized to a per-GPU rate, and sortable — are maintained on the Provider Directory. This page no longer repeats that table.