Free calculator
EC2 idle cost estimator
An instance averaging under 5% CPU is usually doing nothing anyone would miss. Under 20% and it is a rightsizing candidate, because dropping one size halves the bill for it. Put your numbers in and see what the quiet ones cost you.
Common SME types listed.
CloudWatch → EC2 → CPUUtilization, 14-day average.
These instances cost
$280.32/month
$3,363.84/year · ≈ £219/month
Estimated waste
$140.16/month
$1,681.92/year · ≈ £109/month
Under 20% average CPU — strong rightsizing candidate. One size down halves the cost.
On-demand Linux pricing, 730 hrs/month. Same conservative thresholds our scanner uses: idle <5% CPU counts the full cost; 5–20% counts half (one size down). Burstable (t-family) instances at low CPU may be sized correctly — check credit usage before acting.
Averages hide the truth — one busy instance masks three idle ones. The free scan checks 14 days of CloudWatch data per instance, in every region, and names the idle ones.
Estimate the whole billHow idle instances happen (to everyone)
- →The experiment that never ended. A proof of concept box, a load test rig, an instance someone spun up to try something. It worked, the project moved on, and the instance stayed. Nobody owns it, so nobody stops it.
- →The worker whose job moved. A queue consumer or cron box whose workload moved to Lambda or a container. The migration shipped and the old instance stayed on, just in case.
- →Sized for launch-day traffic. Instances provisioned for the traffic you hoped for, still running at that size for the traffic you have. Halving an instance size halves its cost, exactly.
- →Dev environments that never sleep. A staging stack used 40 hours a week bills for 168. Scheduling off-hours is a ~70% saving with zero architectural change.
- →Previous-generation types. m3, c3, r3 and t2 instances still running. The same workload on a current generation type such as m7g, c7g or t4g runs 10% to 20% cheaper and faster.
Frequently asked questions
What counts as an idle EC2 instance?
A common, conservative definition: under 5% average CPU over two weeks, with low network traffic. At that level the instance is usually a forgotten experiment, an old worker whose queue moved elsewhere, or a "temporary" box from two years ago. Stopping it recovers close to 100% of its cost.
How much can rightsizing save?
Every instance size step doubles the price, so an m5.2xlarge costs twice an m5.xlarge. An instance averaging under 20% CPU can usually drop at least one size and halve its cost. AWS Compute Optimizer is free and gives per-instance recommendations from your actual usage.
Is low CPU always waste?
No. Memory-bound workloads such as databases, caches and JVM apps can be perfectly busy at 10% CPU, and burstable t-family instances are designed to sit idle. That is why the scanner treats a CPU threshold as one signal among several and looks at network traffic and instance family before flagging anything.
Should I stop or terminate an idle instance?
Stop it first, because stopping keeps the EBS volumes and you can start it again if something breaks. You carry on paying for the storage but not the compute. If nothing complains for a month, snapshot the volumes and terminate.
What about dev and staging instances running 24/7?
A dev box used 8 hours a day, 5 days a week sits idle 76% of the time. Switching it off overnight and at weekends saves roughly 70% of its cost, and either AWS Instance Scheduler or a small Lambda on a cron will do it.