EKS vs ECS: total cost comparison

The EKS vs ECS question shows up in every AWS-native organization sooner or later. Both run containers. Both integrate with the AWS platform. Both support Fargate and EC2 launch types. The honest cost comparison is mostly about cluster overhead, with the bulk of the bill coming from worker compute that's priced identically across both. This post walks through where the cost difference actually shows up and when each makes financial sense.

The cluster overhead: $73/month difference

ECS has no control plane fee. The cluster resource (aws_ecs_cluster) is free. You pay only for the compute the cluster runs.

EKS charges $0.10/hour per cluster (about $73/month) for the managed control plane. AWS runs the API server, etcd, scheduler, and controller manager across multiple availability zones. The cost is the same regardless of how many pods you run on the cluster.

For organizations with many clusters (separate dev/staging/prod, or separate per-team clusters), the EKS overhead multiplies fast. A company running 10 EKS clusters pays $730/month just for control planes. The same 10 ECS clusters cost $0.

For full pricing details, see the catalog pages for aws_eks_cluster and aws_ecs_cluster.

Worker compute: identical pricing

This is the part most cost comparisons get wrong. EKS and ECS workers run on the same underlying AWS compute at the same prices.

EC2 launch type

Both EKS and ECS support running tasks/pods on EC2 instances. Same instance types, same on-demand prices, same Reserved Instance and Savings Plan eligibility. A 3-node cluster of m6i.large costs the same on EKS as on ECS: about $315/month of EC2 charges.

The only differences are at the orchestration layer. EKS uses node groups (aws_eks_node_group) or Karpenter to provision nodes. ECS uses capacity providers and Auto Scaling Groups. Both add or remove EC2 instances as workloads scale. The instances themselves are billed identically.

Fargate

Both EKS and ECS support Fargate for serverless task execution. Pricing is identical across both: about $0.04048/vCPU-hour and $0.004445/GB-memory-hour.

A 1 vCPU / 2 GB task running 24/7 on Fargate costs $35-40/month on either EKS or ECS. Fargate Spot is available on both at the same discounted rate (roughly 70% off Fargate's on-demand price).

So where does cost actually differ?

Three places, and none of them are big at typical production scale.

First, EKS Auto Mode adds a ~12% management fee on EC2 instances when enabled. ECS doesn't have an equivalent feature; capacity providers do similar work for free. For a $5,000/month EC2 footprint on EKS Auto Mode, that's $600/month in Auto Mode fees. Optional.

Second, EKS network plugins (VPC CNI by default) consume secondary IP addresses on EC2 instances. This can require larger subnets and, in some configurations, more or larger instances to fit the pod density needed. ECS networking is simpler and doesn't have this constraint.

Third, EKS observability (Container Insights with enhanced metrics) is per-pod by default and bills via CloudWatch. ECS Container Insights is per-task and typically lower volume. The difference is usually small but can be 5-10% at high pod density.

A real cost comparison: 10-service production fleet

Take a typical production workload running 10 services with rough sizing:

• 30 vCPUs total compute across all services (some on-demand, some scaled)
• 60 GB total memory
• Running 24/7 with autoscaling
• 3-AZ deployment for HA
• One CloudWatch log group per service

On EKS Standard mode

• Control plane: $73/month
• EC2 workers (mix of m6i.large with autoscaling): ~$1,800/month
• EBS for node root disks: ~$60/month
• CloudWatch Container Insights: ~$80/month
• VPC Endpoints (ECR, S3, KMS in 3 AZs): ~$65/month
• Total: ~$2,078/month

On ECS with EC2 launch type

• Cluster: $0
• EC2 workers (same mix): ~$1,800/month
• EBS for node root disks: ~$60/month
• CloudWatch Container Insights: ~$70/month
• VPC Endpoints: ~$65/month
• Total: ~$1,995/month

On ECS with Fargate

• Cluster: $0
• Fargate tasks: ~$2,400/month (higher per-vCPU than EC2)
• CloudWatch Container Insights: ~$70/month
• VPC Endpoints: ~$65/month
• Total: ~$2,535/month

The cluster overhead is 1-3% of the total. The bigger decisions are EC2 vs Fargate (which applies equally to both EKS and ECS) and what sizing/Reserved Instances you commit to.

When EKS makes more sense despite the $73/month

Five scenarios where EKS is the right choice even though ECS is cheaper at the cluster layer:

You need Kubernetes ecosystem tools

Helm charts, Karpenter, Argo CD, Flux, Istio, Linkerd, Crossplane, OpenTelemetry operators. All of these have first-class Kubernetes support and either don't work on ECS or require significant custom integration. If your team is investing in Kubernetes-native tooling, EKS is the obvious choice.

You're standardizing across clouds

Kubernetes runs on AWS, Azure, GCP, on-premises. ECS runs on AWS only. For organizations with multi-cloud requirements or wanting to avoid AWS lock-in, the $73/month per cluster is a small price for portability.

Your team has Kubernetes expertise

Hiring and onboarding engineers who already know Kubernetes is easier than training people on ECS-specific constructs. Senior platform engineers expect Kubernetes by default.

You have many workloads per cluster

The $73/month is fixed per cluster, regardless of how many pods run in it. A cluster with 200 pods amortizes the overhead to $0.37/pod. A cluster with 10 pods pays $7.30/pod. ECS's per-cluster math is the same except the starting point is $0 instead of $73.

You need fine-grained scheduling

Kubernetes has pod affinity, node taints, topology spread constraints, custom schedulers. ECS has placement constraints and capacity providers but a simpler model. For complex scheduling requirements (GPU sharing, NUMA awareness, multi-tenant isolation), Kubernetes wins.

When ECS makes more sense

Three scenarios where ECS is the better choice:

Small or simple workloads

For applications with a handful of services and no Kubernetes ecosystem dependency, ECS is simpler operationally. The $73/month overhead and the larger conceptual surface of Kubernetes aren't worth the cost for a 5-person team running 10 services.

You want minimum AWS-native overhead

ECS integrates more deeply with AWS-specific services. IAM roles for tasks, secrets via Secrets Manager, ELB integration: all simpler in ECS than the EKS equivalents (IRSA, external-secrets operator, AWS Load Balancer Controller).

Many small clusters

Organizations running per-team or per-environment clusters at small scale find EKS's per-cluster fee disproportionate. 20 small EKS clusters cost $1,460/month in control plane fees alone. The same 20 ECS clusters cost $0.

The decision in practice

For most organizations, the choice is decided by factors other than cost. Team expertise, ecosystem fit, and operational preferences usually drive the decision. The $73/month per cluster is rarely the deciding factor.

Where cost matters more is in the sub-decisions:

• Fargate vs EC2 launch type. For variable workloads, Fargate wins. For sustained workloads with RI/SP commitments, EC2 wins.
• Spot vs on-demand. For fault-tolerant workloads, Spot saves 60-90% on either platform.
• Right-sizing. Most container workloads are overprovisioned by 2-4x. Vertical Pod Autoscaler on EKS or task definition right-sizing on ECS recovers significant cost.

How to estimate before committing

c3x estimates both EKS and ECS workloads from Terraform:

c3x estimate --path . --usage-file c3x-usage.yml

For EKS, c3x reads aws_eks_cluster, aws_eks_node_group, and any aws_eks_fargate_profile resources. For ECS, c3x reads aws_ecs_cluster, aws_ecs_service, and the linked task definitions. Both produce per-resource cost breakdowns suitable for direct comparison.

For full per-resource pricing details, see the catalog pages: aws_eks_cluster, aws_ecs_cluster, aws_ecs_service, and aws_instance (for the worker node prices).

FAQ

Where to go from here

Three follow-ups depending on which direction this nudges you:

• If you're going with EKS: see the EKS catalog page for cost optimization tips specific to Kubernetes on AWS.
• If you're going with ECS: the ECS service catalog covers Fargate vs EC2 trade-offs in more detail.
• For broader Terraform cost estimation including Kubernetes workloads, see how to estimate AWS costs from Terraform.