Updated October 25, 2026
· ~3 min read
Choosing EKS is only half the decision; the other half is who runs the nodes. EKS Auto Mode, launched at the end of 2024 and steadily extended since, runs Karpenter, the AWS Load Balancer Controller and the EBS CSI driver off-cluster, launches Bottlerocket nodes for pending pods, consolidates them, and charges a per-instance fee for doing so. Karpenter run yourself does much of the same scheduling with no fee and full control. Managed node groups are the older, simpler model of Auto Scaling groups that EKS helps you update. And Fargate runs pods with no nodes at all, with limits.
AWS reports that Karpenter inside Auto Mode scales out 43% faster and consolidates up to 69% faster after optimisations shipped through 2025 and 2026.
For an m5a.xlarge the Auto Mode fee is about $0.021 an hour against an On-Demand price of about $0.17 — roughly 12%, and a smaller share of the effective price once Savings Plans apply to the EC2 part. What it buys is the work that most often goes wrong: keeping Karpenter, the load balancer controller and the storage driver compatible with each Kubernetes version, patching node operating systems, and replacing nodes on schedule. For a team without a dedicated platform engineer that is cheap. For a team with several, who already run Karpenter well, it may not be.
Auto Mode can be enabled on an existing cluster and run alongside managed node groups, so you can move workloads gradually: enable it, create a NodePool, cordon and drain an existing node group while pods reschedule onto Auto Mode nodes, then remove the group. Check anything that assumed a specific AMI or host agent — Auto Mode nodes run Bottlerocket — and remove the self-managed copies of the controllers Auto Mode now runs, so two of them are not fighting over the same load balancers.
Question 1 of 2
Do you need custom AMIs, host-level agents or node configuration Auto Mode does not allow?
Use EKS Auto Mode for new clusters unless you have a specific reason not to — it removes node patching, the controllers that most often break during upgrades, and most capacity planning, for a fee of roughly a tenth of a typical instance's On-Demand price. Run Karpenter yourself when you need control Auto Mode does not give you (custom AMIs, unusual node configuration, your own add-on versions) and have the people to own it. Keep managed node groups for small, static clusters or workloads that need predictable nodes. Use Fargate on EKS only for isolated, low-volume pods — it does not support DaemonSets or GPUs.
Production EKS Platform
A production-ready Kubernetes platform on EKS with Karpenter, GitOps, service mesh, full observability stack, and security hardening: built to run real workloads.
AWS Cost Optimization & FinOps Sprint
Reduce your AWS bill by 20–60% while improving performance and reliability.
We work with startups and growing companies on AWS architecture decisions. Tell us about your situation and we will recommend the right approach.
As of October 25, 2026. Provided as is, for general information, and accurate to the best of our knowledge on the date shown. Not legal, compliance, tax, financial or other professional advice. Prices, limits and vendor services change; check current vendor documentation before relying on anything here. Terms of use.