LearnGlossary

What is a spot instance?

Updated 25 October 2026 · Published by Zoorik

A spot instance is a cloud virtual machine (VM) that runs on spare capacity, for less than the on-demand price. The cloud can evict it (take it back) when it needs that capacity. AWS warns you two minutes before, and Azure up to 30 seconds before, on a best-effort basis.

AWS warns 2 minutes before eviction, Azure up to 30 seconds Two lanes on one time axis of seconds before the capacity is taken back, with ticks at 120, 90, 60, 30 and 0. AWS Spot Instance: the VM runs on spare capacity, and a solid interruption notice window runs from 120 seconds, 2 minutes, to 0. Then the VM is terminated, stopped or hibernated. Azure Spot VM: the VM runs on spare capacity, and a dashed Scheduled Events notice window, up to 30 seconds and best effort, runs from 30 seconds to 0. Then the VM is deallocated or deleted. A red line at 0 marks the moment the capacity is taken back. Illustration. AWS warns 2 minutes before eviction, Azure up to 30 seconds interruption notice2 minruns on spare capacityAWS Spot Instanceterminated, stoppedor hibernatedScheduled Events noticeup to 30 s, best effortAzure Spot VMdeallocatedor deletedcapacity taken back1209060300seconds before the capacity is taken back
Illustration: the notice times are the ones AWS and Azure publish, and the span before each notice has no set length.

Spot instances vs on-demand

  • On-demand runs until you stop it. Spot runs while spare capacity lasts.
  • On AWS, spot prices shift slowly with supply and demand. The on-demand rate doesn’t.
  • AWS says spot saves up to 90% off on-demand prices (October 2026). Prices change.
  • Azure gives spot VMs no service level agreement (SLA).

Spot pools on AKS

A spot instance in Azure is an Azure Spot VM. An AKS spot node pool is a node pool made of them.

  • It can’t be the default pool.
  • Its nodes carry the taint kubernetes.azure.com/scalesetpriority=spot:NoSchedule, so only pods with a matching toleration run there.
  • After an eviction, the Cluster Autoscaler adds nodes that pods need.

Why it matters for your bill

  • Spot reduces each node’s price. It suits work that can stop and restart, like batch jobs and test environments.
  • Requests that match real use mean fewer nodes.
  • An evicted Azure Spot VM still bills for its disks under Deallocate, its default policy. AKS spot pools default to Delete.
  • Mark a cluster as non-production and turn on its spot switch. Coral can then plan spot nodes in pools that fall back to on-demand when spot runs out.
  • The free assessment works out your own figure.

Read next: How to reduce Kubernetes cost: requests, then nodes

Sources: Amazon Web Services, “Spot Instances” (opens in a new tab) (Amazon EC2 User Guide); Amazon Web Services, “Amazon EC2 Spot Instances” (opens in a new tab); Microsoft Learn, “About Azure Spot Virtual Machines” (opens in a new tab); Microsoft Learn, “Add an Azure Spot node pool to an Azure Kubernetes Service (AKS) cluster” (opens in a new tab).

Find your own number

One line in your cloud console. One PDF report with your estimated saving at list price.

Estimate your saving

Talk to the builders

Thirty minutes with the people who build Coral.

Book a demo (opens in a new tab)

Or write to founders@zoorik.com