Methodology

What the numbers on this site mean, and what they don't.

What GPU Index tracks

We publish hourly snapshots of GPU cloud list prices and, where we can observe it, whether capacity appears available. Historical data is kept — we don't overwrite past observations.

How fresh is the data?

Numbers on this site refresh about once per hour (UTC). The timestamp at the top of each page shows when the latest snapshot was computed. Expect up to one hour of lag.

Prices

All prices are shown as USD per GPU-hour so you can compare across providers. We normalize GPU names (e.g. H100-SXM-80GB) because each provider labels hardware differently.

Prices come from a mix of public provider catalogs, marketplace listings, and published rate cards. A low listed price does not guarantee you can rent at that rate right now.

Availability

Not every provider publishes capacity information. Where we have a signal, we show it; where we don't, we label it unknown rather than guessing.

Cheapest available only includes offers where we saw a recent availability signal. Cheapest listed includes all published prices, including catalog entries that may not be rentable on demand.

Limitations

API

JSON endpoints mirror the data on this site. Access is free and rate-limited. No account required.