01
Cluster Consolidation
Merge under-utilized EKS clusters. Each consolidated cluster eliminates one $73/month control-plane fee and frees operational overhead.
Pricing Guide · Containers
AWS EKS & Container Pricing Guide.
EKS charges for the control plane. ECS does not. Fargate prices the data plane differently from managed nodes. ECR adds another line. This guide breaks down the container stack and the levers that move the bill.
$2.4B+
AWS spend reviewed
500+
Engagements
38%
Average reduction
$340M+
Client savings
The Container Stack
Three orchestrators, four billable lines.
Containers on AWS run through three orchestrators (EKS, ECS, App Runner) and two data-plane options (managed EC2 nodes, Fargate). The economics differ in non-obvious ways. The container bill aggregates control-plane fees, data-plane compute, registry storage, networking, and Load Balancer charges. Most container bills are 70-85% data-plane compute — which means the negotiation is mostly an EC2 negotiation.
| Component | Pricing | Savings Plan? |
|---|---|---|
| EKS Control Plane | $0.10/hour per cluster (~$73/mo) | No |
| ECS Control Plane | Free | n/a |
| EKS Auto Mode (compute) | ~12% premium on EC2 + EC2 list | Compute SPs cover EC2 portion |
| Managed Node Group EC2 | EC2 list pricing | Yes — Compute SPs, RIs, Spot |
| Fargate (ECS/EKS) | vCPU-hours + GB-hours | Compute SPs |
| ECR Storage | $0.10/GB-month | No |
| ECR Data Transfer Out | Standard egress rates | No |
EKS versus ECS economics
EKS carries the $73/month/cluster control-plane fee plus higher operational overhead. ECS has no control-plane fee. For small estates (under 10 clusters), the EKS premium is rounding error. For estates with hundreds of EKS clusters — which we see in microservice-heavy organizations — the cumulative control-plane spend hits six figures annually. Cluster consolidation under fewer, larger EKS control planes saves real money. Most multi-cluster estates we audit have 30-50% more clusters than they need.
The bin-packing problem
The largest single inefficiency in container estates is poor bin-packing. Most production Kubernetes clusters run at 25-40% CPU utilization on the node fleet. Right-sized requests and limits, combined with cluster autoscalers tuned to consolidate workloads, can lift utilization to 60-75% without affecting performance. The 35-point utilization swing translates directly to 35% less node-fleet spend. Karpenter — AWS's open-source provisioner — handles this better than Cluster Autoscaler and is now broadly supported.
Spot on Karpenter
Karpenter makes Spot-on-EKS mechanical. Pool diversification across instance families and sizes drives Spot interruption rates below 3% in most production workloads. Stateless services behind PodDisruptionBudgets run reliably on Spot. Karpenter handles graceful interruption automatically. For workloads we have moved to Spot via Karpenter, the data-plane line falls 60-75% on the affected portion of the fleet.
Optimization Levers
Where container savings actually live.
02
Karpenter Migration
Replace Cluster Autoscaler with Karpenter. Better bin-packing, faster scale-up, automatic Spot diversification.
03
Spot Node Groups
Move stateless workloads to Spot. 60-75% off list on the affected fleet, with under 3% interruption rate when properly pooled.
04
Right-Size Requests
CPU and memory requests are usually 2-4x actual usage. VPA recommendations or Goldilocks-style analysis lifts utilization.
05
Graviton Node Groups
Karpenter happily provisions arm64 nodes for arm64-compatible workloads. 20% discount on node compute with no orchestration change.
06
Fargate for Spike
Provision baseline on managed nodes with Compute SP coverage. Burst onto Fargate (or Fargate Spot) for spikes. Lower total cost than fully provisioned.
Frequently Asked
Questions on container pricing.
01Should we move from EKS to ECS to save the control-plane fee?+
For most enterprise estates, no. The EKS control-plane fee is dwarfed by data-plane spend. The portability and ecosystem of Kubernetes outweigh the control-plane line. The exception is large fleets of small clusters where consolidation is not feasible — those occasionally justify ECS. We model it per estate; the answer is rarely a wholesale migration.
02Is Karpenter really worth the migration from Cluster Autoscaler?+
Yes for most production estates. Karpenter is faster, packs better, handles Spot diversification natively, and removes the node-group sprawl that Cluster Autoscaler encourages. The migration is well-documented and reversible. We have seen Karpenter migrations cut node-fleet spend 25-40% with no change to workload requests.
03How do Compute Savings Plans apply to container workloads?+
Compute Savings Plans cover Fargate and EC2 node compute interchangeably. The baseline portion of your container fleet — the part that runs 24/7 — should be covered by Compute SPs. The bursty portion runs on Spot or unreserved capacity. See our Savings Plans Optimization service for the model we use.
04What about App Runner?+
App Runner is convenient for small services but priced at a meaningful premium versus Fargate or EC2 nodes for the same workload. It rarely makes economic sense for production at scale. We map App Runner workloads to Fargate-on-ECS during cost reviews; the migration is straightforward and pays back quickly.
05Can container spend be negotiated inside an EDP?+
Container spend is EC2 and Fargate spend — both of which are qualifying spend inside an EDP. The EDP compute tier discount applies, stacked with Compute Savings Plan coverage. We negotiate container-specific concessions inside EDP including Graviton migration credits and Spot-friendly commit structures. See our EDP negotiation service.
Containers are mostly
an EC2 negotiation.
500+ engagements. $340M+ client savings. We tune bin-packing, migrate to Karpenter, push Spot adoption, and negotiate the EDP tier underneath.