Spot Instance

Spot (or preemptible) instances are spare cloud capacity sold at steep discounts — often 60-90% below on-demand — with the caveat that the provider can reclaim them at short notice, typically with 30 seconds to two minutes of warning.

They are ideal for workloads that tolerate interruption: batch processing, distributed training with checkpointing, rendering farms, and stateless workers sized as a fleet. A common pattern pairs a stable on-demand base with a large spot layer that absorbs peak or opportunistic work.

The engineering requirements are fault tolerance and fast drainage: jobs must checkpoint or retry, and schedulers should treat the two-minute warning as a hard deadline. Teams that build for interruption routinely cut compute costs by half or more; teams that do not build for it simply inherit outages with a discount.