Kubernetes and Container Cost Tools

A Kubernetes cost estimate that starts from node count is wrong in two directions at once. It comes in too low, because a cluster bill is not just nodes; it is nodes plus a control-plane fee, plus a cloud load balancer for every Service of type LoadBalancer, plus a root disk on every node, plus egress and cross-zone traffic. On a small cluster those can exceed the compute.

And it comes in too high per unit of useful work, because pods request far more than they use. Nodes are provisioned against requests, so you pay for the request, not the usage, and typical clusters run at a third to a half of what was asked for. That gap is the single largest source of Kubernetes waste and it is invisible to anything that starts from node count.

So this calculator starts where a cluster actually starts: pod requests. It subtracts the kubelet reservation and daemonsets to get real allocatable capacity, packs the requests onto nodes, and only then prices the result, which is why its node count comes out higher than a simple division and its bill comes out higher than the compute line.

It covers EKS, GKE and AKS. The control-plane fee is $0.10 per cluster-hour on all three, so the difference between them is node pricing and free-tier credits, not the managed service itself.

The serverless container runtimes sit here too, because the decision they answer is the same decision. Fargate and Cloud Run remove the node layer entirely and with it the requested-versus-used gap, and they charge a premium per unit of compute to do it. Whether that premium is worth paying depends on the utilisation figure the Kubernetes calculator makes you confront: at 30% utilisation the premium is often cheaper than the waste it removes, and at 70% it is not.