Nebius Makes Live Preemptible GPU Prices Public Across Regions

On Oct. 21, Nebius announced that customers could publicly view live prices for its preemptible GPU capacity online and through marketplace partners, making rates available by GPU model and region.

The prices are listed on Nebius’ website and through Shadeform, Silicon Data and GPU Finder. A public API provides rates for each GPU and region along with the time each rate takes effect; it does not require an account. The rates describe a specific GPU in a specific location at a particular time, so the effective time is part of the price information.

Nebius announced its dynamic spot-pricing model on Sept. 22, and the model took effect Oct. 8. Rates reflect capacity and demand for each GPU type in each region and may change as often as every 15 minutes. The maximum spot rate is one cent below the current on-demand rate for the same GPU.

A price snapshot at 5:29 a.m. ET (09:29 UTC) on Oct. 21 showed an H200 GPU-hour at $0.886 in us-central1 and $4.49 in eu-west1. The figures show the regional difference at that moment; they are not continuing quotes. A GPU-hour is the listed price for one GPU for one hour, and the snapshot does not establish what a customer would pay later.

Preemptible virtual machines use capacity that has already been allocated but is temporarily idle. That capacity can be reclaimed. Customers can set a maximum price, and an instance stops if the spot rate exceeds the limit, while its data is retained. The service is suited to flexible workloads such as experimentation, data processing and short fine-tuning jobs, where the work can accommodate an interruption.

“Companies should be able to see what compute costs and access it on terms that fit their work,” Nebius founder and CEO Arkady Volozh said.