How the GPU price index works
The index is the typical price of a GPU per GPU-hour: a median over direct providers, from live prices, with a minimum of three providers.
Published Oct 19, 2026
The index
For every direct provider (hyperscalers excluded), the cheapest comparable on-demand price per GPU-hour is taken from its latest live retrieval within the last three days. The index is the median of these prices.
Prices far below or above the median (below a third or above three times) are treated as outliers and not counted. With fewer than three providers there is no index; single offers stay visible.
Cheapest current offer
A separate figure: the single lowest comparable on-demand price of any provider and source, hyperscalers included. It is often far below the index, because one provider can be cheap while the typical price is higher.
Changes and charts
Changes over one, seven and thirty days compare only providers that had a price on both days. A provider joining or leaving does not move the number.
Charts start once fourteen days of index values exist.
What is not counted
«from» prices, prices for the GPU only (without vCPU and RAM), offers without a stated pricing model and community capacity are shown, but not counted.