Skip to main content
Spot pricing is dynamic pricing for preemptible virtual machines (VMs) and Nebius AI Cloud services that run on preemptible VMs. For each preemptible GPU platform in each region, the current spot price is shared across preemptible VM capacity and changes based on how much capacity is actually in use. To control your maximum price per GPU hour, you can create a pricing policy. A pricing policy defines how much you’re willing to pay for preemptible resources that you create on a specific platform. To protect against price increases, set a max price in a pricing policy: resources that use the policy stop if the spot price exceeds that max price. If you don’t want to set a max price, you can choose to follow the current spot price when creating a preemptible resource. For detailed instructions, see Managing pricing policies.

​
How spot pricing works

Each preemptible GPU platform in each region has a spot price and a minimum price.
  • Spot price: The current price for one GPU hour. This price may change as often as every 15 minutes based on capacity that is actually in use.
  • Minimum price: The lowest possible spot price for one GPU hour on the platform in that region.
  • Max price: The value that you set in a pricing policy. It’s the highest spot price at which your resources that use the policy can run.
Spot prices can change before and during resource usage. The current spot price becomes the applicable price when the resource starts using preemptible VM capacity. During use, the applicable price can change at the next 15-minute update. Before incurring charges, you can check the current spot price and minimum price on the Billing → Pricing → Pricing history page in the web console. When you create a new preemptible resource, the web console shows the current spot price for the selected resource based on its GPU count and an estimated hourly cost. The estimate assumes the spot price doesn’t change during that hour, so the actual charge can differ if the spot price changes. Spot prices aren’t capped and can rise above the regular on-demand service rates used for pay-as-you-go billing. A max price in a pricing policy is a firm ceiling: if the spot price exceeds the max price, preemptible resources that use the policy are stopped.
Spot pricing doesn’t guarantee resource availability. Preemptible resources may be stopped if capacity is needed elsewhere or if the current spot price exceeds the max price set in the selected pricing policy.
For more information about preemptible VM behavior, see Preemptible virtual machines.

​
Pricing policies

A pricing policy belongs to a project and applies to one GPU platform. When a resource uses the policy, the spot price for the resource’s region is compared with the policy’s max price. The max price is a per-GPU-hour price limit, not the price that you pay by default. While the spot price is lower than or equal to the max price, resources that use the policy are charged at the current spot price. Setting the max price higher than the current spot price doesn’t raise the price for you or other users. If the spot price exceeds the max price, your resources are stopped, and you’ll never pay more than the max price you’ve set in your pricing policy for GPU usage. If the max price is lower than the current spot price, the policy is shown as Blocked in the web console. Resources that use a blocked policy can’t start until the spot price decreases and you restart them, or until you increase the max price set in the policy.

​
Pricing examples

For example, for a preemptible VM with 8 GPUs:
  • The spot price stays within the max price. If the max price is $3.00 per GPU hour and the VM runs for 30 minutes, but the spot price changes after 15 minutes, the computing cost is calculated for each pricing interval:
    • First 15 minutes at $2.00 per GPU hour: 8 GPUs × 0.25 hours × $2.00 = $4.00
    • Next 15 minutes at $2.50 per GPU hour: 8 GPUs × 0.25 hours × $2.50 = $5.00
    Total: $4.00 + $5.00 = $9.00
  • The spot price exceeds the max price. If the max price is $3.00 per GPU hour and the spot price rises from $2.00 to $3.20 after 15 minutes, the VM stops when the new price applies:
    • First 15 minutes at $2.00 per GPU hour: 8 GPUs × 0.25 hours × $2.00 = $4.00
    • After 15 minutes, the VM stops because the spot price exceeds the max price.
    Total: $4.00
  • No max price is set. If you follow the spot price, the spot price is $3.00 per GPU hour when you create a VM and the VM runs for 45 minutes, the VM keeps running as the spot price changes:
    • First 15 minutes at $3.00 per GPU hour: 8 GPUs × 0.25 hours × $3.00 = $6.00
    • Next 15 minutes at $3.20 per GPU hour: 8 GPUs × 0.25 hours × $3.20 = $6.40
    • Next 15 minutes at $2.90 per GPU hour: 8 GPUs × 0.25 hours × $2.90 = $5.80
    Total: $6.00 + $6.40 + $5.80 = $18.20

​
Choosing a pricing option

When you create or start a preemptible VM, choose how the VM accepts the current spot price:
Following the spot price sets no upper limit for GPU usage. The spot price can exceed the regular on-demand service rates, and the resource is not stopped when the price rises.

​
Capacity allocation

When demand for a preemptible GPU platform is higher than available capacity, the max price helps determine allocation priority. Requests are prioritized by max price, not by the order in which they are submitted. A resource with a higher max price can be prioritized over resources with lower max prices for the same GPU platform and region. Only running VMs use capacity and affect the spot price. Requests that don’t start, including requests with a high max price, don’t change the price. Running more of your own preemptible VMs uses capacity and affects the spot price for all users on the same GPU platform and region, just like any other running preemptible VMs. You pay for that capacity while it runs. When your VMs stop, the released capacity becomes available to all users, and you don’t get priority for it.

​
Services that use spot pricing

Spot pricing applies to preemptible VMs with GPUs. Depending on the service and resource type, this can include: If a service doesn’t let you select preemptible GPU capacity, spot pricing isn’t available for that resource.

​
Usage and invoices

Spot prices can change during a billing period. Your invoice shows the total consumption for the billing period and doesn’t list every spot price change separately. To check detailed usage for each price interval, export usage data in the FinOps Open Cost and Usage Specification (FOCUS) 1.2 format. The export contains the effective hourly price for VM usage and the time period for each charge. For more information, see Exporting usage data as CSV.

​
See also