Price instances sitting below a CPU threshold plus their attached disks and idle addresses, and what scheduling would save.
An idle instance costs more than its compute line, because attached disks bill in full whether the instance runs or not and reserved public addresses are charged precisely when nothing is using them. Scheduling is costed separately: it does not delete anything, it just stops paying for the roughly 680 hours a month that a 50-hour workload is not needed. Stopping an instance removes the compute charge but not the storage charge, which is why teams report far smaller savings than they expected after a shutdown campaign. Confirm which meters keep running while stopped on your provider — the behaviour differs between compute, disk, address and licence charges.
Idle Resource Cost
idle cost = idle instances × rate × 730 + idle instances × attached GB × storage rate + unattached addresses × address rate × 730; scheduling saving = schedulable instances × rate × (730 − needed hours).
idle cost = idle instances × rate × 730 + idle instances × attached GB × storage rate + unattached addresses × address rate × 730; scheduling saving = schedulable instances × rate × (730 − needed hours). An idle instance costs more than its compute line, because attached disks bill in full whether the instance runs or not and reserved public addresses are charged precisely when nothing is using them. Scheduling is costed separately: it does not delete anything, it just stops paying for the roughly 680 hours a month that a 50-hour workload is not needed.
Stopping an instance removes the compute charge but not the storage charge, which is why teams report far smaller savings than they expected after a shutdown campaign. Confirm which meters keep running while stopped on your provider — the behaviour differs between compute, disk, address and licence charges.
This calculator takes 12 inputs: Instances in scope, Instances below the CPU threshold, Idle CPU threshold, Instance rate, Hours in the month, Hours per week the workload is needed, Instances that could be scheduled off, Attached storage per instance, Block storage rate, Unattached public addresses, Idle address rate, Total monthly cloud spend. The pre-filled defaults are a realistic starting point — replace them with figures from your own environment for a result you can act on.
No. Database replicas, log shippers and licence servers can sit near zero CPU while being critical. A safe rule combines low CPU with negligible network traffic, no active sessions and an owner confirmation, which is why the tag and owner data matters as much as the metric.