Compare AI hosting and GPU cloud providers

Every value comes from the provider's entry in our dataset and its dated sources. GPU prices are USD per GPU-hour, listed per pricing basis (on-demand, community cloud, serverless, spot or a marketplace floor) and per node size where two providers sell the same one. In the rows for one GPU model, a lowest price is marked only between prices for that GPU, on the same basis and for the same number of GPUs, for example “Lowest on-demand price” or “Lowest 8-GPU on-demand price”. Plan prices are standard or renewal prices; an introductory price appears only next to the renewal price it leads to.

A row with a blue rule at its label has at least two different values. “Lowest of 3”, “Highest of 3” and “Finest of 3” compare that one field across the providers, using only its verified values. On the entry GPU price row that is each provider's cheapest listed GPU-hour, whatever the GPU model, so it is not a like-for-like GPU price; on the egress row it is the price per GB. “Not verified” means no official source confirmed the value; “Not listed” means the provider does not list it; “Price not republished” means the provider's terms do not allow us to show its prices, so we link to its live pricing. Prices exclude VAT and sales tax. Affiliate relationships never change the order or the highlights.

MethodologyAffiliate disclosure