AWS cost · EKS

Why EKS is expensive: the $73/month control plane

The EKS control plane is the cost that starts before you do anything. Spin up a cluster and you are billed $0.10/hour for it - about $73/month - before a single node joins or a single pod runs. On its own that is easy to justify. The problem is that clusters multiply: one per environment, one per team, a couple of proofs of concept nobody tore down. I have run four at once, and that is already about $292/month in control-plane fees before you count the nodes underneath them.

Every EKS cluster bills $0.10/hour - roughly $73/month - just for the managed control plane, whether it runs one pod or none. After 14 months on an old Kubernetes version the rate jumps to $0.60/hour on extended support. The waste is rarely one cluster; it is the idle and forgotten ones spread across regions and environments.

How much does the EKS control plane cost?

At $0.10/hour on standard support, the flat fee scales with how many clusters you keep alive - and only counts the control plane, not the nodes, load balancers, or data transfer billed on top:

What you pay for before any nodes run

The fee buys the managed Kubernetes control plane - the API server, etcd, and the scheduler - run and patched by AWS across Availability Zones so you do not operate it yourself. That is a fair deal for a cluster doing real work. It is a bad deal for a cluster doing nothing, because the meter does not care. An empty cluster left over from a migration or a demo bills the same $73/month as a production one.

The extended-support jump to $0.60/hour

Each Kubernetes version gets 14 months of standard support at $0.10/hour. After that it moves to extended support at $0.60/hour - six times the rate, about $438/month per cluster - until you upgrade to a supported version. A fleet left on an old version quietly reprices itself. Keeping clusters current is not only a security habit; it keeps you on the cheaper tier.

How to find your clusters and what they cost

Clusters are regional, so the expensive ones are usually the ones in a region you never open. List them everywhere, then check each for node groups - a cluster with none is a strong candidate for deletion:

# list clusters in one region (repeat per region you use)
aws eks list-clusters --region eu-central-1 --query 'clusters'

# for each cluster: version, status, and support period
aws eks describe-cluster --region eu-central-1 --name my-cluster \
  --query 'cluster.{Version:version,Status:status}'

# a cluster with no node groups is often forgotten
aws eks list-nodegroups --region eu-central-1 --cluster-name my-cluster \
  --query 'nodegroups'

The nodes and networking usually dwarf the control-plane fee on a busy cluster - idle oversized worker nodes and CloudWatch log ingestion are where the real EKS money tends to hide. But the control plane is the one you pay even when nothing is running, so it is the first thing to prune.

Frequently asked questions

More AWS cost breakdowns