Glossary of GPU prices
Short definitions of the terms used in the price area.
Published Oct 19, 2026
Prices
GPU-hour: one GPU for one hour. Index: median of the cheapest comparable on-demand price of each direct provider. Cheapest current offer: the lowest comparable on-demand price of any provider.
Comparable offer: a full price (GPU with vCPU and RAM) with a stated pricing model, no «from» price, no outlier.
Pricing models
On-demand: no minimum term. Spot: discounted spare capacity that can be reclaimed. Reserved: fixed term (1 month, 1 year, 3 years), lower price per hour.
Providers
Direct provider: operates or is responsible for the GPUs and is the contracting party. Hyperscaler: large general cloud (for example AWS, Microsoft Azure, Google Cloud); counted in the cheapest offer, not in the index.
Marks
«from»: minimum price. «GPU only»: price without vCPU and RAM. «location not stated»: the provider does not say where the offer runs. «not yet verified»: compliance facts not yet confirmed by a second check. «new»: GPU first seen in the last 30 days. «announced»: announced by the manufacturer, no offers yet.
Compliance
European-owned: owner and ultimate parent in Europe. US parent or control: the provider or its ultimate parent is a US company or controlled by one. DPA: data processing agreement. SLA: service level agreement. Contracting entity: the company you sign with.