About the rate card
A rate card for one GPU
One GPU, the NVIDIA H100, priced by every provider that publishes a number, with the page each number came from.
What the card covers
The rate card lists the public price of renting an H100 by the hour: on-demand rows on the home page, spot and reserved rows on their own page, and the same prices cut by server size and by provider. Each row names the configuration, the billing unit the provider states, the page it was read from and the date.
Beside the prices sit a few tools built on the same rows: a memory-fit page that counts how many H100s the weights of a model need, a calculator that totals a job from a provider’s own listing, and comparisons with a handful of other GPUs.
What it leaves out
It does not rank providers on quality, speed, support or availability, because the data holds none of those. It does not show price history: there is one reading, dated on every page. It does not include storage, network egress or tax, and it does not guess a price for a size a provider does not list.
How the rows are treated
Rankings and statistics use only rows that are in effect today and priced by the provider. Marketplace figures set by hosts are shown with a chip and never ranked. Prices announced for a later date are shown separately and counted once they take effect. Spot and reserved rows are never mixed into the on-demand numbers.
Who runs this
H100 Rate Card is published by the team behind Runcrate, an AI cloud. We list Runcrate in the tables where it has a price, marked as ours, and sort by price like every other row.