Explanation
Why the waste happens and who it affects.
Production VMs and scale sets that run around the clock for months or years at a stable size are the usage that Azure Reservations and the Azure savings plan for compute are designed to discount, and Microsoft's FinOps guidance puts the savings from commitment discounts at up to 72% compared with list prices. When no commitment covers that baseline, every hour is paid at the on-demand rate.
This gap is different from an underused or mis-scoped commitment: here the commitment simply does not exist, or it covered last year's fleet and was never extended as new workloads, regions and VM families were added. It persists when nobody owns commitment purchasing, or because teams defer the decision while an architecture is still changing. Azure Advisor flags it with the High-impact recommendations 'Consider virtual machine reserved instance to save over the on-demand costs' and 'Consider purchasing a savings plan to unlock lower prices'.
Billing model
The pricing dimensions that drive this cost.
Steady VM compute can be billed three ways.
- Pay-as-you-go compute
- Billed per hour of running VM at the list rate for the size and region
- Reserved VM Instance
- A 1- or 3-year commitment for a VM size and region, applied automatically each hour to matching running VMs; covers compute only, not Windows or other software meters
- Savings plan for compute
- A 1- or 3-year fixed hourly spend commitment applied across eligible compute services and regions; usage above the commitment is billed at pay-as-you-go rates
- Use-it-or-lose-it hours
- Reserved hours or hourly savings plan commitment with no matching usage in scope expire and do not roll over
How to detect
4 checks to find it in your estate.
- Measure VM commitment coverage: the share of eligible VM cost with PricingCategory Standard (on-demand) versus Committed, using the FinOps hub coverage query or cost exports grouped by pricing model
- Review Azure Advisor for 'Consider virtual machine reserved instance to save over the on-demand costs' and 'Consider purchasing a savings plan to unlock lower prices', and the purchase recommendations shown in the Reservations and savings plan purchase blades
- Identify VM sizes, families and regions whose running instance count has stayed flat over the last 30 to 90 days; the steady floor is the commitment candidate
- Check which steady VMs run on series that can no longer be reserved: from July 1, 2026, one-year reservations can't be bought or renewed for older series such as Av2, D, Dv2, Dsv2, F and Fsv2, and one- and three-year reservations can't be bought for Dv3, Dsv3, Ev3 and Esv3; these need a savings plan or a move to a newer series
How to fix
5 ways to remove the waste.
- Buy Reserved VM Instances for stable VM sizes and regions, with instance size flexibility so the discount also applies to other sizes in the same size group; use shared or management group scope so coverage follows usage across subscriptions
- Cover compute that changes size, family or region, or runs on series no longer eligible for reservations, with a savings plan for compute sized to the steady hourly spend
- Commit only to the stable baseline and leave bursty, short-lived or soon-to-be-retired capacity on pay-as-you-go or Spot
- Account for flexibility limits: reservations bought from February 1, 2027 for savings-plan-eligible services such as VMs cannot be exchanged, savings plans cannot be canceled or refunded, and refunds are capped at USD 50,000 per 12-month rolling window
- Rightsize before committing, and re-run coverage analysis monthly and after migrations so new steady workloads are picked up
Documentation
Vendor references for pricing and configuration.
- FinOps best practices for computelearn.microsoft.com
- Prepay for Azure virtual machines to save moneylearn.microsoft.com
- What are savings plans?learn.microsoft.com
- Cost recommendations - Azure Advisorlearn.microsoft.com
- Transition guide for retired Azure Reserved VM Instanceslearn.microsoft.com
- Self-service exchanges and refunds for Azure Reservationslearn.microsoft.com