EKS Auto Mode Pricing: What the EKS Auto Mode Fee Costs and How It Compares to Karpenter, CAST AI and Spot Ocean

EKS Auto Mode costs the EC2 price of every instance it launches plus a management fee of about 12 percent of that instance's on-demand rate, billed per second with a one-minute minimum, on top of the standard $0.10 per hour EKS cluster fee. AWS's own examples put the fee at $0.03672 an hour on a c6a.2xlarge and $0.07344 on a c6a.4xlarge. The fee does not change when the instance runs on spot, a Reserved Instance or a Savings Plan, which is the part most cost estimates miss.

Everything attributed to AWS below was read on aws.amazon.com/eks/pricing and in the EKS user guide in September 2026. Vendor prices come from each vendor's own AWS Marketplace listing, sold under its own seller name. The arithmetic is ours, and it is shown so you can redo it for your fleet.

How much does EKS Auto Mode cost?

EKS Auto Mode is priced per instance, for as long as the instance runs. You pay three things:

The 12 percent is not a figure AWS states as a percentage. It falls out of every example on the pricing page: $0.03672 on a $0.306 c6a.2xlarge is 12.0 percent, and the same ratio holds for the m5a examples AWS quotes. For an m5.large at $0.096 an hour on demand in US East, the fee is about $0.0115 an hour.

Does EKS Auto Mode cost more?

Yes, in license terms EKS Auto Mode costs more than running the same nodes yourself, because self-managed nodes carry no management fee at all. Whether it costs more in total depends on what you would otherwise pay engineers to do, and on how much of your fleet runs at a discount. Here is the fee alone at three fleet sizes, on m5-class instances at $0.048 per vCPU-hour on demand, 730 hours a month:

That works out to roughly $4.20 per vCPU per month. Compute-optimized c6a instances are cheaper per vCPU, so the fee is lower there, about $3.35 per vCPU a month. Memory-heavy and GPU instances cost more per vCPU, so the fee is higher.

Is the Auto Mode fee charged on spot instances and Savings Plans?

Yes. AWS states that you can use On-Demand, Reserved Instances, Compute Savings Plans and Spot Instances with Auto Mode, but "the EKS Auto Mode charges are independent of the EC2 instance purchase option." The fee is pegged to the on-demand price, whatever you actually pay.

That changes the math for discounted fleets. If your nodes run on spot at 35 percent of the on-demand price, a 12 percent fee is about a third of what you pay for the instance itself. On a three-year Compute Savings Plan, which AWS advertises at up to 66 percent off, the fee is again a third or more of the discounted instance cost. A team that moved most of its EKS capacity to spot to cut the bill will feel Auto Mode far more than a team running on demand.

EKS Auto Mode vs Karpenter: which costs less?

Self-managed Karpenter costs less in fees, because it has none. It is open source under Apache 2.0, and Auto Mode uses it internally. What you pay instead is the work: running the controller, upgrading it with each Kubernetes version, managing its IAM role, picking and patching node AMIs, and keeping the VPC CNI, CoreDNS, kube-proxy and storage drivers current. Auto Mode takes all of that, runs Bottlerocket nodes with no SSH access, and replaces every node within 21 days.

A practical rule: if one platform engineer already runs Karpenter on your clusters as part of their week, Auto Mode's fee is paying for something you have. If nobody does, and your node groups have not been resized since they were created, $25,000 a year at 500 vCPUs is often less than the waste Karpenter would remove and far less than a new hire. The same trade-off, with the paid vendors included, is laid out in our Karpenter alternatives comparison.

How does EKS Auto Mode pricing compare to CAST AI and Spot Ocean?

CAST AI and Spot Ocean are not add-ons to Auto Mode. They replace the node layer, so you pick one of the three. Both publish public rate cards on AWS Marketplace even though their own sites show a quote form. At the same fleet sizes:

The inputs: CAST AI lists Growth and GrowthPro at $1,000 a month plus $0.00694444 per managed CPU per hour, which is $5.00 per CPU for a 720-hour month; the full card is in our CAST AI pricing breakdown. Flexera lists Ocean on its Spot.io contract at $1.415 per 100 vCPU-hours on a 12-month term, about $10.33 per vCPU a month, and separately a dimension at $0.28 per dollar of savings; the listing does not say which applies to a given cluster, so ask.

Fees are not the whole comparison. Auto Mode automates nodes and nothing else. CAST AI also moves workloads to spot with fallback to on-demand and resizes pod requests, and Ocean fills spot and existing commitments before on-demand. If either lowers your EC2 bill by more than its fee exceeds Auto Mode's, it wins on total cost. The only way to know is to measure the EC2 line before and after a trial, on the invoice.

EKS Auto Mode vs Fargate: which is cheaper?

For steady, packable workloads, Auto Mode on EC2 is usually cheaper than Fargate. Fargate charges per pod for the CPU and memory the pod requests, $0.04048 per vCPU-hour and $0.004445 per GB-hour for Linux x86 in US East, with no node to manage and no fee on top. Four vCPUs and 16 GB on Fargate come to about $0.233 an hour. The same shape as an m5.xlarge under Auto Mode is $0.192 plus a $0.023 fee, about $0.215, and it can run several pods if they fit. Fargate wins for small, spiky or isolated workloads where a node would sit mostly empty, and it cannot use spot on EKS the way EC2 can.

Does EKS Auto Mode rightsize pods?

No. Auto Mode chooses instances from the requests your pods declare, so if every deployment asks for twice the CPU it uses, Auto Mode will faithfully launch twice the instances and charge its fee on all of them. Request rightsizing is a separate layer: the open source Vertical Pod Autoscaler, or commercial tools such as StormForge, ScaleOps and PerfectScale. StormForge's owner CloudBolt publishes a guide to running it on Auto Mode clusters. Our EKS rightsizing tools comparison covers both layers, with the AWS Marketplace price of each option and the EKS-specific catch.

Fixing requests first also shrinks the Auto Mode bill, because the fee scales with the instances launched. Cut requested CPU by 30 percent and, once consolidation catches up, the fee drops by about the same.

Is EKS Auto Mode worth it?

EKS Auto Mode is usually worth it for teams that run mostly on-demand, have no one who owns node management today, and value AMI patching and add-on upgrades being someone else's job. It is harder to justify for large fleets that already run Karpenter well, and for fleets that live on spot or three-year Savings Plans, where a fee pegged to the on-demand price takes a large bite out of the discount.

Migrating is also real work: node groups become Auto Mode NodePools, and any launch template customization, SSH-based debugging or long-running node assumption has to go. If your infrastructure lives in Terraform, an AI coding agent can draft the NodePool and NodeClass rewrites from your existing modules, which turns a week of manifest editing into a review.

What should I check before turning on EKS Auto Mode?

1. Your purchase mix. Pull the share of EKS node hours on spot, Savings Plans and Reserved Instances. The higher it is, the larger the fee relative to what you pay.
2. Your instance families. The fee is a share of each family's on-demand price. Model it on your real mix, not on m5.
3. Node lifetime. Nodes are replaced within 21 days. Stateful sets, long batch jobs and anything with local state need disruption budgets that tolerate it.
4. Who owns pod requests. Auto Mode does not change them. Budget for VPA or a pod rightsizing tool, or accept today's overprovisioning at a 12 percent markup.
5. How you will measure it. Turn on split cost allocation data for EKS in the Billing console first, so each team's pod cost and unused instance cost are in the Cost and Usage Report before and after the switch.

The short version

EKS Auto Mode adds about 12 percent of the on-demand price to every instance it manages, charged even on spot and committed capacity, on top of the $0.10 per hour cluster fee. At 500 always-on vCPUs on m5-class instances that is roughly $25,000 a year, against $0 for self-managed Karpenter, about $42,000 for CAST AI Growth and about $62,000 for Spot Ocean on their AWS Marketplace cards. Whichever you choose, check the result on the bill rather than in the vendor's dashboard: CostAnalyst connects your AWS billing read-only and gives each team its EKS cost allocation, so the fee and the savings sit on the same page finance reads.