Kubernetes savings calculator

How this estimate is calculated

It is neither a measurement of your environment nor audited market data: it is a simple model with conservative assumptions that I use as a starting point in assessments. Each lever is applied in sequence to what the previous one left, so the same saving is never counted twice.

1. 1.Consolidation and rightsizing (Karpenter). Compares your average CPU utilization with 65%, which a well-packed cluster can sustain. The difference is slack; I assume consolidation captures 50% of it.
2. 2.Spot. Counts only the interruption-tolerant share not yet on Spot, at a 65% discount to On-Demand. AWS advertises discounts of up to 90%, but they vary by instance type, region and time.
3. 3.Shutting down non-production off-hours. Applies a 60% saving to the dev/staging share, since nights and weekends are about 64% of the week.
4. 4.Graviton. Assumes 70% of workloads run on ARM without rework and each costs 20% less per unit of work. AWS advertises larger price-performance gains; I use a conservative value.

Realistic range

The low end is 60% of the estimate and the high end is 115%, never above 60% of the bill. Applying everything takes engineering effort and not every workload is eligible.

What the calculator does not consider

Storage, network and managed-service costs, discounts you already negotiated (Savings Plans, Reserved Instances), availability requirements, migration cost and the specifics of your workloads. That is why the result is an order of magnitude, and a technical assessment measures the real number.