Why is EKS so expensive?
EKS charges a fixed fee of $0.10 per hour for each cluster’s control plane, which is about $73 a month. That fee is easy to see, so it gets blamed most often. For a small cluster it can be a large share of the bill. As a cluster grows, the worker nodes that run your pods, plus networking, idle capacity and logging, usually matter more.
The fixed control plane fee
- Standard support: $0.10 per hour per cluster, about $73 a month if the cluster runs all month.
- Extended support: $0.60 per hour per cluster for Kubernetes versions past standard support, about $438 a month. That is six times the standard rate.
The fee applies whether the cluster runs one pod or a thousand. Test clusters are a common source of surprise charges, because the fee keeps running after the work is finished.
Worker nodes are usually the biggest cost
The control plane is managed by AWS, but your worker nodes are not. You pay separately for them:
- EC2 instances are billed by the instance type, the time they run and the storage attached to them.
- AWS Fargate bills per vCPU-hour and per GB-hour of memory for each pod. Pod sizes are rounded up, so Fargate can cost more per unit of compute than well-packed EC2 nodes.
- EKS Auto Mode adds a management fee on top of each instance’s EC2 price.
Capacity is often wasted in two ways. Pods that request more CPU and memory than they use leave nodes mostly empty. Node-level agents also use capacity before your apps do. These DaemonSets include the networking plugin, kube-proxy, logging and monitoring agents, along with the resources the kubelet reserves on each node.
Networking charges add up
Networking is the cost most people overlook:
- NAT gateway: about $0.045 per hour, which is roughly $33 a month if it runs all month, plus about $0.045 per GB processed. Each Availability Zone with a NAT gateway adds another one.
- Load balancers: an Application or Network Load Balancer costs about $0.0225 per hour (about $16 a month) plus capacity units. Each Kubernetes
LoadBalancerservice can create one. - Cross-AZ traffic: about $0.01 per GB in each direction between Availability Zones.
- Public IPv4 addresses: about $0.005 per hour for each address, including the ones that nodes and load balancers use.
Prices are for us-east-1 and change over time. Check the AWS pricing pages before you plan a budget.
Logs and monitoring
CloudWatch Logs charges for ingestion, at about $0.50 per GB in us-east-1. Control plane logs, Container Insights and verbose application logs all add to that amount. Set a retention period on each log group so old logs do not accumulate.
Extended support for older versions
Each Kubernetes minor version gets 14 months of standard support. After that, the cluster moves to extended support for another 12 months, at the higher rate. Keeping clusters on current versions avoids that increase. Plan upgrades before a version reaches the end of standard support.
A fixed-cost example
This table shows the fixed items for one cluster that runs all month in us-east-1. It leaves out worker nodes, data transfer and logs, which depend on your workload.
| Item | Rate | Roughly per month (730 hours) |
|---|---|---|
| EKS control plane, standard support | $0.10 per hour | $73 |
| One NAT gateway | $0.045 per hour | $33, plus data charges |
| One Application Load Balancer | $0.0225 per hour | $16, plus capacity units |
| One public IPv4 address | $0.005 per hour | About $3.65 |
Use the AWS Pricing Calculator to estimate the rest for your own node sizes and traffic.
Is EKS more expensive than other options?
It depends on what you compare it with:
- Self-managed Kubernetes on EC2 has no control plane fee, but you run the control plane yourself. A highly available control plane usually means three control-plane nodes, plus etcd backups, upgrades and the time to operate them.
- Amazon ECS has no control plane fee. It is simpler to run, but it is AWS-specific rather than Kubernetes.
- Azure Kubernetes Service (AKS) has a free tier with no control plane charge. Its paid tiers add charges.
A fair comparison counts the whole bill, including the engineering time to run each option. For a deeper comparison, read the Kubernetes vs. EKS guide.
How to bring the cost down
- Right-size requests. Set CPU and memory requests close to what pods actually use, and check usage over time.
- Use Spot capacity for fault-tolerant workloads. Spot can be interrupted, so keep it away from critical services.
- Use Savings Plans when your usage is steady.
- Consider Graviton instances. AWS publishes better price-performance for them, so test your workload on them.
- Remove idle capacity. Tools such as Karpenter can consolidate nodes that are underused.
- Delete unused resources. Remove unused
LoadBalancerservices, NAT gateways and clusters. - Compare ECS for workloads that do not need Kubernetes.
Next steps
- Read the EKS setup guide for the teardown order that avoids leftover load balancers.
- Compare the ECS container deployment, which has no control plane fee.
- Set up billing alerts with the guide to avoiding unexpected AWS charges.