Updated 25 October 2026 · Published by Zoorik
An on-demand instance is a cloud virtual machine (VM) you pay for by how long it runs, with no commitment. Azure calls this pay-as-you-go pricing.
On-demand instance pricing
- AWS bills Linux and Windows instances by the second, with a 60-second minimum. SUSE Linux Enterprise Server bills in full hours.
- Azure bills each full minute a VM runs, so in Azure’s example 6 minutes and 45 seconds bills as 6 minutes.
- On Azure, a VM that shows “Stopped” still bills for its cores. Only “Stopped (deallocated)” ends that charge.
On-demand instances vs reserved
- AWS Reserved Instances and Azure Reservations trade a one-year or three-year commitment for a discount.
- You pay for it even when it’s unused.
- Usage outside the commitment bills at the on-demand rate, as does usage after an AWS Reserved Instance expires.
In October 2026, Azure says reservations can reduce costs by up to 72% from pay-as-you-go prices. AWS quotes the same top figure for Savings Plans. Prices change.
Why it matters for your bill
- Each node in an AKS node pool or EKS node group is a VM, billed at the on-demand rate for its size.
- A commitment or spot price can replace that rate.
- A spare node keeps billing until it’s removed.
- Coral drains spare nodes so your autoscaler can remove them.
- It rebalances CPU and memory requests to match real use, so the same pods fit on fewer nodes.
- The free assessment works out your own figure.
Read next: Kubernetes node size best practices: large or small nodes
Sources: AWS, “Amazon EC2 On-Demand Pricing” (opens in a new tab); AWS, “Reserved Instances for Amazon EC2 overview” (opens in a new tab); Microsoft Learn, “What are Azure Reservations?” (opens in a new tab); Microsoft Azure, “Linux Virtual Machines Pricing” (opens in a new tab).