On-demand, spot and reserved GPUs
The three pricing models of GPU cloud capacity, what they mean for a buyer and how the price index treats them.
Published Oct 19, 2026
On-demand
Capacity billed by the hour or second, without a minimum term. You start and stop when you want. It is the most expensive model per hour and the most flexible one.
The index of every GPU is built from on-demand prices only, because they are published by most providers and can be compared directly.
Spot
Spare capacity at a discount. The provider can reclaim it at short notice, so a job must tolerate interruptions (checkpoints, retries).
Spot prices are shown separately. A spot index appears only when at least three providers publish a spot price.
Reserved
Capacity booked for a fixed term, for example one month, one year or three years, often with prepayment. The price per hour is lower, the commitment is higher.
Reserved terms are listed as their own pricing models, so a one-year price is never compared with an on-demand price.
What to check before booking
Minimum term and notice period, prepayment, whether the price covers the whole server or the GPU only, storage and data transfer, and where the capacity runs.