EKS Cost Optimization: Cut Your Kubernetes Bills on AWS

Amazon EKS comes with its own $0.10/hour cluster fee — $73/month before a single pod runs. But the cluster fee is the least of your problems. The real EKS cost driver is over-provisioned node groups. Here's how to cut EKS bills by 40–70%.

EKS Cost Components

Component	Cost	Optimization Potential
EKS Cluster fee	$0.10/hr ($73/mo)	Low — fixed cost
EC2 nodes	Varies (largest cost)	High (40–80%)
EBS volumes (PVs)	$0.10/GB/mo	Medium
Load balancers (ALB/NLB)	$16–$22/mo each	Consolidation
Data transfer	$0.01–$0.09/GB	Architecture
NAT Gateway	$0.045/GB processed	VPC endpoints

Karpenter: Replace Cluster Autoscaler

Karpenter (now a CNCF project, AWS-originated) makes smarter node provisioning decisions than Cluster Autoscaler. Key advantages: it selects the optimal instance type and size for each pending pod batch (not just scaling a fixed node group), enables Consolidation (rebalancing pods onto fewer nodes when possible), and responds faster to scale events.

Consolidation alone typically reduces node count by 20–35% on mature clusters with mixed workload sizes. Enable it with consolidationPolicy: WhenUnderutilized in your NodePool spec.

Graviton Nodes: 20% Better Price/Performance

AWS Graviton3 (arm64) EC2 instances offer 20–40% better price/performance than x86 equivalents. For most containerized Linux workloads, migration is straightforward — build your containers as multi-arch (amd64 + arm64) images using Docker buildx, then add Graviton instance types to your Karpenter NodePool.

# Karpenter NodePool with Graviton preference
spec:
  requirements:
    - key: karpenter.k8s.aws/instance-family
      operator: In
      values: ["m7g", "c7g", "m6g", "c6g", "m5", "c5"]
    - key: kubernetes.io/arch
      operator: In
      values: ["arm64", "amd64"]
    - key: karpenter.sh/capacity-type
      operator: In
      values: ["spot", "on-demand"]

Spot Node Groups for EKS

Run batch jobs, non-critical services, and stateless web tiers on Spot nodes. Use Karpenter's karpenter.sh/capacity-type: spot requirement and set weight to prefer Spot when available. Add the AWS Node Termination Handler to ensure graceful pod eviction on Spot reclaim.

// FAQ

Should we use EKS managed node groups or self-managed?

Managed node groups for most use cases — AWS handles node OS updates and patches. Self-managed nodes only make sense if you need custom AMIs or instance types not supported in managed groups. The operational overhead of self-managed rarely justifies the marginal flexibility.