Nebius Group N.V. has made live rates for its preemptible GPU capacity public across regions, giving AI cloud customers a way to compare current compute costs before choosing a service that can be interrupted.
On Oct. 21, the company began showing prices on its website and through a public API that requires no account. Shadeform, Silicon Data and GPU Finder also carry the listings. The API supplies the GPU platform, region and effective time for each rate, while Nebius offers up to 30 days of price history.
In an Oct. 21 snapshot taken at 5:29 a.m. ET (09:29 UTC), Nebius listed an H200 GPU-hour at $0.886 in us-central1, versus $4.49 in eu-west1. A GPU-hour is the pricing unit for one GPU used for an hour. Those figures captured listed rates at one moment, rather than prices guaranteed for later use.
Nebius uses demand and spare capacity in a region to set each GPU’s spot rate. It may revise rates every 15 minutes; the highest allowed spot rate for a GPU is $0.01 less than its current on-demand rate. The company announced the dynamic spot model Sept. 22, before it took effect Oct. 8.
The potential savings come with interruption risk. Customers may set a maximum rate, and a virtual machine stops if the spot price exceeds that limit. Nebius can also reclaim the temporarily idle capacity used by the service, making it suited to work that can tolerate a pause.
Nebius says data on attached persistent storage survives an interruption, while data on local solid-state drives does not. Customers may continue paying for retained storage after a machine stops. The public listings give them more information for scheduling flexible work, while the rate and availability can still change during use.
