Cloud GPU prices › About this data
Cloud GPU prices
Cloud providers rent out servers with GPUs by the hour, and the price of that hour is one of the few public numbers on what AI computing costs. Once a day we record the list price of every GPU server type we can tie to a GPU model on AWS and Azure, and divide it by the number of GPUs, so an NVIDIA H100 on one provider can be compared with an H100 on the other.
- Where it comes from
AWS and Azure publish their prices in a public price file and a public price service. We matched each server type to its GPU from the provider's product pages.
Who reports it: AWS and Azure publish these prices themselves, in a price file and a price service that anyone can read. We matched each server type to its GPU from the provider's own product pages.
Official sources
- How often we update it
- Every day
- Last updated
- (4 hours ago)
- Coverage
- Daily snapshots since November 17, 2026: on-demand prices on AWS (US East) and on-demand and spot prices on Azure (East US), Linux only.
- Worth knowing
- These are list prices for Linux in one region of each provider (AWS US East, N. Virginia; Azure East US). Large customers pay less under private contracts, and a listed price does not mean a server is available. Spot prices are for capacity the provider can take back at any time; AWS does not put its spot prices in the price file, so the spot list is Azure's only. This is not an index of the GPU rental market: specialist GPU clouds are not covered yet.
See the Cloud GPU prices page · Use this data in your AI or app (how connecting works) · Information, not investment advice (disclaimer).