GPU Cloud Index: providers compared by GPU, price per hour and billing

The index lists what nine providers publish for renting a GPU: one row per provider, GPU and way of renting it. Each price is converted to US dollars per GPU per hour so rows can be compared, and the provider’s own wording sits under it. Every row links to the page it came from and shows the date it was checked.

How to read it

• The rental model matters as much as the GPU. The same H100 appears as an on-demand VM, a community or marketplace listing, an interruptible spot instance, a reserved cluster and a serverless worker. These are different products with different risks, so they are never merged into one “price”.
• “From” means the provider says “from”. Runpod labels Community Cloud prices that way, and Hyperstack labels reservations “starting from”. We keep the qualifier.
• Marketplace rows are snapshots. Vast.ai hosts set their own prices, and they change all the time. The row shows the lowest single-GPU on-demand listing in Vast.ai’s public offers feed at the time we checked, plus the median, so you can see the spread.
• Announced is not current. Lambda’s public price list takes effect on 16 November 2026; until then its current on-demand prices are visible only to logged-in users. Those rows show “Not published” as the current price, with the announced figure labelled separately.
• Unit size. CoreWeave mostly sells 8-GPU instances, so the per-GPU figure is the instance price divided by eight, and the row shows the minimum you must rent.
• Extras are not included. Storage, egress and tax are excluded unless the row says otherwise. Modal also bills CPU and memory separately from the GPU rate.

Choosing from here

• If you specifically need H100 capacity, start with renting an H100.
• If you are still deciding between on-demand, spot, reserved and serverless, see where to rent a GPU for AI.
• Bursty inference that should scale to zero is covered in serverless GPU platforms compared.
• The three most-compared self-serve options are side by side in Runpod vs Lambda vs Vast.ai.
• If price is the deciding factor, the cheapest GPU cloud by GPU class pulls the lowest published figure per GPU from this index.

Inclusion

A provider is listed when it publishes a price for at least one GPU offering. Listing and row order follow our methodology and cannot be bought. If a figure is wrong or out of date, tell us.