Fargate vs Karpenter for EKS: A Cost Decision Guide

For EKS cost, choose Karpenter when you run at steady scale and can keep nodes packed, because it provisions right sized EC2, consolidates underused nodes, and runs a large share of capacity on Spot. Choose Fargate when workloads are small, spiky, or isolated, because removing node management is worth its per pod premium there. Both add the EKS control plane fee of about 0.10 US dollars per cluster per hour. The crossover is utilization plus Spot tolerance: if you will keep nodes busy and use Spot, Karpenter is cheaper; if you will not, Fargate is.

Last updated: June 2026. Written by Fredrik Filipsson and reviewed by Morten Andersen, built on our See, Cut, Lock, Run method.

This article sits in our complete guide to Kubernetes cost optimization, the cluster pillar it links up to, and connects to the broader cloud cost optimization playbook. Once you choose Karpenter, the next decision is Spot safety, covered in our companion guide on how to use Spot instances safely with Karpenter. Picking the compute model is a Cut step decision that sets the EKS cost ceiling before any tuning.

What are Fargate and Karpenter on EKS?

AWS Fargate is a serverless compute engine that runs each EKS pod on managed capacity billed per vCPU and memory, with no nodes to provision. Karpenter is an open source node autoscaler that provisions, right sizes, and consolidates EC2 nodes for your pods, including Spot capacity. With Fargate you never see a node; you pay for the resources each pod requests. With Karpenter you run real EC2 nodes that Karpenter sizes to the pending pods and removes when underused. Both run on EKS and both carry the EKS control plane fee.

How does each one bill?

Fargate bills per pod vCPU hour and GB hour, while Karpenter bills the underlying EC2 nodes it provisions, and both add the EKS control plane fee. Per the AWS Fargate pricing, Fargate charges for vCPU and memory from image pull to pod termination, rounded up, with a one minute minimum. Karpenter compute is plain EC2, so you get the full On-Demand, Spot, and Savings Plan rate card, per the Amazon EKS pricing. The EKS control plane is about 0.10 US dollars per cluster per hour during standard support, roughly 73 US dollars per month, rising to about 0.60 US dollars per hour on extended support versions. Verify current rates in those pages before modeling, because AWS pricing changes.

Fargate vs Karpenter, side by side

The rates above reflect AWS pricing structure as we read it in June 2026. Confirm exact figures for your region and configuration in the linked AWS documentation, because prices change.

When is Fargate cheaper?

Fargate is cheaper when workloads are small, bursty, short lived, or strictly isolated, so that running EC2 nodes would leave you paying for idle capacity and node management you do not want. Batch jobs, low traffic services, per tenant isolation, and teams without platform engineers to tune nodes all fit. Because Fargate bills per pod, the idle node gap disappears and there is nothing to bin pack. The trade is the per pod premium and no Spot, so the moment your fleet is large and steady, that premium starts to dominate and Karpenter pulls ahead.

When is Karpenter cheaper?

Karpenter is cheaper once you run at steady scale and can keep nodes busy, because it provisions exactly the EC2 your pending pods need, consolidates underused nodes within minutes, and runs a large share of capacity on Spot at a steep discount to On-Demand. The raw EC2 rate, packed densely and discounted with Spot and Savings Plans, beats Fargate's per pod price for any meaningful sustained fleet. The break even is utilization and Spot tolerance: a well tuned Karpenter cluster running dense nodes with heavy Spot is the cheapest way to run EKS. Get the consolidation settings right with our guide on how to tune Karpenter consolidation for maximum savings.

Which should you choose?

Choose by scale and Spot tolerance. If you run a large, steady core and your workloads tolerate Spot interruptions, Karpenter is the lower total cost because you keep dense packing and heavy Spot. If you run small, bursty, or strictly isolated workloads where node operations are not worth it, Fargate is the lower total cost despite the per pod premium. Many estates run both: Fargate profiles for the long tail and isolated tenants, Karpenter for the heavily tuned core. The wrong choice in either direction is the expensive one, Fargate across a large steady fleet, or an untuned Karpenter cluster with loose requests and no Spot.

Frequently asked questions

Is Fargate or Karpenter cheaper for EKS?

Karpenter is usually cheaper at scale because it provisions right sized EC2 nodes, packs pods densely, and can run a large share of capacity on Spot. Fargate is cheaper only at small or spiky scale where node management overhead and idle nodes would otherwise outweigh its per pod premium. The crossover is utilization: if you will keep nodes busy, Karpenter wins; if you will not, Fargate wins.

How much does EKS cost on top of Fargate or Karpenter?

Both add the EKS control plane fee of about 0.10 US dollars per cluster per hour during standard support, roughly 73 US dollars per month, rising to about 0.60 US dollars per hour for extended support versions. Fargate then bills per vCPU hour and GB hour for each pod, while Karpenter bills the underlying EC2 nodes. Confirm current rates in the AWS pricing documentation.

Can Karpenter use Spot instances?

Yes. Karpenter natively provisions EC2 Spot capacity and falls back to On-Demand when Spot is unavailable, which is the single largest lever for cutting EKS compute cost. Fargate does not offer Spot for EKS pods, so Spot heavy workloads strongly favor Karpenter.

Can I run Fargate and Karpenter in the same cluster?

Yes, and many estates do. Run Fargate profiles for small, bursty, or isolated workloads where node management is not worth it, and let Karpenter provision EC2 for the large, steady core where dense packing and Spot deliver the lowest rate. Match each workload to the cheaper model rather than forcing one across the whole cluster.

The short version

Fargate wins at small or spiky scale because it removes node management and bills per pod; Karpenter wins at steady scale because it packs right sized EC2 and runs heavy Spot. The crossover is utilization and Spot tolerance, and many estates run both. When you want each workload modeled and placed on the cheaper model, that is what our rightsizing and waste elimination service delivers.