Spot GPU Droplets vs On-Demand GPU Dropletspublic

Last verified 11 Aug 2026

DigitalOcean Droplets are Linux-based virtual machines (VMs) that run on top of virtualized hardware. Each Droplet you create is a new server you can use, either standalone or as part of a larger, cloud-based infrastructure.

GPU Droplets are available with on-demand or spot capacity. Both tiers provide the same GPU configurations, but they differ in availability, pricing, and the possibility of reclamation.

Note

Spot GPU Droplets are in public preview. Spot GPU Droplet pricing is separate from on-demand pricing and may change daily based on available idle GPU capacity.

On-Demand GPU Droplets Spot GPU Droplets
Guaranteed capacity. After you create an on-demand GPU Droplet, its capacity remains available until you destroy it. As-available capacity. Spot GPU Droplets use available idle GPU capacity, so DigitalOcean does not guarantee availability.
Fixed pricing. On-demand GPU Droplets have a fixed hourly rate. Variable pricing. Spot GPU Droplets always cost the same as or less than the equivalent on-demand GPU Droplets, but pricing may change daily based on available capacity. Check the current rate in the DigitalOcean Control Panel.
No capacity-based reclamation. DigitalOcean does not reclaim on-demand GPU Droplets to reallocate capacity. Capacity-based reclamation. DigitalOcean may reclaim a Spot GPU Droplet with at least two hours’ notice. We send the notice by email to both the team account and the account that created the Droplet.
Stable workloads. On-demand GPU Droplets are best for production, latency-sensitive, or stateful workloads that require continuous capacity. Fault-tolerant workloads. Spot GPU Droplets are best for fault-tolerant workloads that can save progress, restart, and move to another instance.

On-demand GPU Droplets are best for workloads including:

  • Production inference
  • Latency-sensitive applications
  • Stateful applications
  • Workloads that require stable, continuous capacity

Spot GPU Droplets are best for fault-tolerant workloads including:

  • Batch model training
  • Batch or asynchronous inference
  • Hyperparameter tuning
  • Rendering

We can't find any results for your search.

Try using different keywords or simplifying your search terms.