Cloud GPU rental prices: H100, H200, B200 per GPU-hour

This table lists GPU rental prices per GPU-hour for data-center GPUs (Nvidia H100, H200, B200, B300 and GB200, AMD MI300X and MI325X, plus L40S and A100) as each cloud publishes them on its own public pricing page: AWS, Google Cloud, Microsoft Azure, CoreWeave, Lambda, Nebius, Crusoe, RunPod, DigitalOcean, Vultr and Hyperstack.

Every row is one provider, one GPU and one offering. Where a provider prices a whole instance (for example an 8-GPU machine), we divide the listed instance price by the GPU count and say so in the price basis. Each row links its pricing page and shows the date we checked it.

How to read the numbers:

• These are list prices from each provider’s public page on the checked date. They exclude taxes, negotiated or volume discounts, storage, networking and egress.
• On-demand, spot and reserved or commitment prices are labelled separately. Spot capacity can be reclaimed; reserved prices require a booking or contract.
• Cloud GPU prices change often, and some providers announce new rates days ahead. Re-check the price on the provider’s own page before you buy.
• This is a rental table. It does not show purchase prices for GPUs, and it does not cover AI model API token prices.

We re-check the providers three times a week. Sources and methods follow our editorial standards. For chip specs, see AI chips; for the facilities these GPUs run in, see AI data centers.