Preemptible VMs with Spot pricing

Access GPU capacity at market-driven prices, with more choice in how you run and spend.
Dynamic spot prices create opportunities to pay less. Set a maximum price or follow the market to take advantage of them on your terms.
See the dynamic market price and use it to get quick access to GPU capacity for non-mission-critical workloads.
Run experiments, batch inference or data processing jobs on preemptible resources using VMs, serverless runtime or managed Kubernetes.
Access market-driven prices across GPU platforms and regions.
Set your maximum price for workloads where price is the priority.
Follow the market price to avoid price-triggered interruptions.
Note that both options use preemptible capacity. Setting a higher maximum or following the price does not reserve GPUs or guarantee continuous runtime.
Test model configurations, compare approaches and run evaluations without requiring a fixed completion time.
Run queued preprocessing, transformation and other GPU jobs that can retry or resume after interruption.
Use spot capacity for shorter fine-tuning jobs or workflows designed to save progress and recover.
Process offline inference jobs that can be retried and do not need an immediate response.
Access now in three steps
Choose your GPU and region where capacity is available
Set your maximum price or follow the market
Launch your workload
Access GPU capacity at market-driven prices, with complete control over what you’re willing to pay.
Does spot pricing apply to all Preemptible VMs?
Yes, with one exception: NVIDIA L40S preemptible instances are billed at the current flat rate. All other Preemptible Virtual Machines on the Nebius AI Cloud use dynamic spot pricing. You’ll see “preemptible” in the console and documentation.
How often can the price change?
Spot prices are dynamic and may increase or decrease based on available capacity, cluster utilization, customer demand and customer pricing preferences. Current spot prices are displayed in the price calculator in the Nebius AI Cloud console and may vary during active use, including as frequently as every 15 minutes.
Does my maximum price limit my total bill?
No. It limits the spot price per GPU-hour. Total charges depend on GPU count, runtime and other billable resources.
What happens when my Preemptible VM is interrupted?
The Preemptible VM stops. Data on attached persistent volumes is retained, but data on local SSDs is lost. Save progress regularly to persistent storage.
Will my VM restart automatically?
Standalone preemptible VMs do not restart automatically. Restart them manually or arrange automation, subject to capacity and your pricing policy. Serverless and Managed Service for Kubernetes workloads follow the recovery behavior of their own service, so check its documentation.
Do charges continue after a VM stops?
Retained resources, such as attached storage, can continue to incur charges until you delete them.
Which GPU models are available with spot pricing?
Spot pricing supports NVIDIA H100, H200, B200, B300 and RTX PRO 6000 GPUs. Supported configurations and available capacity vary by region. Check the Nebius console for current options and prices.
How does spot pricing compare with other Nebius pricing options?
Spot prices vary by GPU type, region and demand. Visit our GPU pricing page to compare pricing options and choose the right fit for your workload, including instances for workloads that require predictable capacity or cannot tolerate interruption.
How can I use preemptible resources?
You can use preemptible capacity as standalone Preemptible VMs, through a serverless runtime, or as nodes in Managed Service for Kubernetes. The same spot prices apply across all three. Use VMs when you want direct control over the machine, Serverless AI when you want to run jobs without managing infrastructure, and Managed Service for Kubernetes when you already orchestrate workloads in a cluster.
Subscribe to updates