Methodology: GPU cloud hourly prices
GPU cloud hourly prices
- Conditions
What we track. Public hourly prices of cloud GPUs by provider, GPU model, memory size and rental mode; on-demand, spot and community capacity are separate records.
What counts as a value. The US-dollar price per GPU hour the provider's public pricing shows for that exact configuration on the date we read it, kept as published. A multi-GPU price is not substituted for a single-GPU price, and reserved prices are left out.
What it is not. A public price, not a completed rental, a benchmark or an invoice. Public rates do not guarantee available capacity; storage, networking, CPU bundles and service guarantees vary by provider and are not included.
When a value is Unknown. When the page shows no price for that configuration or the price cannot be tied to it without doubt. A value we have not recorded is shown as Unknown, never as zero, and no value is carried over from another day.
How often. Every price is read once a day.
GPU Public Price Watch. Separate spot-price medians for H100 80 GB, A100 40 GB and A100 80 GB on the date, with the number of offers (N) and the range; on-demand and community prices are separate series. Each offer counts once; the median is not weighted by market share.
Version. Version 1 of this method, in effect since 27 September 2026; description revised 15 October 2026.
- Coverage
- On : 42 cloud GPU offers from 4 providers.
- Region
- the United States