EKS Rightsizing Tools: Amazon EKS Rightsizing Software and EKS Cost Optimization Tools Compared

Best for: Platform teams that will install and run the autoscaler themselves

The upstream VPA is still the first thing to try on EKS. It recommends and applies CPU and memory requests from observed usage, and on Kubernetes 1.35 and later it can resize running pods in place instead of evicting them. Start in Off mode, read the recommendations for two weeks (the open source Goldilocks dashboard makes them easier to read), then let it act. The catch on EKS is ownership: AWS documents how to deploy it but does not manage or support it, so upgrades and failures are yours.

Best for: EKS clusters whose managed node groups were sized once and never revisited

Karpenter is the node half of rightsizing on EKS. Instead of scaling a fixed node group, it reads what pending pods need and launches the cheapest instance types that fit, including spot, then consolidates underused nodes. AWS built it and it is the engine inside EKS Auto Mode. Self-managed, it costs nothing but the engineering time to run it well. It does not touch pod requests, so pair it with VPA or a pod-level tool.

Best for: Teams that want AWS to own node provisioning, patching and core add-ons

Auto Mode is AWS running Karpenter, the VPC CNI, storage and load balancing for you, on Bottlerocket nodes with no SSH access. It removes the node-rightsizing work rather than paying a vendor for it. The price is the thing to model: the fee follows the on-demand rate even when you pay spot or commitment prices, so a heavily discounted fleet pays proportionally more. It does not change pod requests.

Best for: Seeing pod-level and unused EKS cost in the Cost and Usage Report

Split cost allocation data adds pod-level CPU and memory cost to your Cost and Usage Report, based on the amortized cost of each instance and each pod's share of it, and since October 2025 it imports up to 50 Kubernetes labels per pod as cost allocation tags. It changes nothing in the cluster, but it is the most honest before-and-after number for a rightsizing project because it comes from the bill. Someone still has to query and chart it.

Best for: EKS estates that want nodes, spot and pod requests managed by one controller

CAST AI rightsizes both layers on EKS: it takes over node provisioning, moves eligible workloads to spot with fallback to on-demand, bin-packs pods and resizes requests. It installs a read-only agent first and estimates savings before you grant write access. Its own pricing page is a quote form; the AWS Marketplace listing is where the numbers are, and a 500-CPU cluster on Growth lists at about $42,000 a year.

Best for: Large EKS estates that want continuous pod rightsizing and keep their node tooling

ScaleOps publishes its own EKS workload optimization guide and runs self-hosted, as ScaleOps Cloud or air-gapped. It works at the pod layer: continuous request rightsizing, replica tuning and placement, alongside Karpenter rather than instead of it. The public price is steep at the bottom end, so it fits estates where pod waste clearly exceeds $50,000 a year. Get the metering period of the per-vCPU charge in writing.

Best for: Teams that want ML pod rightsizing priced per vCPU with no minimum, including Auto Mode clusters

StormForge learns request and limit settings from weeks of usage and can resize pods in place. CloudBolt publishes a guide to running it with EKS Auto Mode, which is a sensible pairing: Auto Mode handles nodes, StormForge handles the requests that decide how many nodes Auto Mode launches. The pay-as-you-go meter has no minimum, which makes it the cheapest paid pod automation here at small scale.

Best for: Smaller EKS estates that want automated rightsizing with a free entry tier

PerfectScale supports EKS, AKS, GKE, OpenShift and Rancher and has the only published free tier sized for a real cluster among the automated vendors: 300 vCPUs a month. That makes it the easiest paid-grade tool to trial on one EKS cluster without a procurement cycle, and its per-vCPU paid rates are low enough to compare directly with the VPA engineering time they replace.

Best for: Namespace chargeback and rightsizing recommendations across EKS clusters

Kubecost is the cost allocation tool most EKS teams already run, and IBM publishes a free Amazon EKS cost monitoring listing on AWS Marketplace. It allocates cost per namespace, deployment and label and produces request recommendations, which makes it the natural measurement layer next to Karpenter and VPA. It does not apply anything itself.

Best for: Finance and FinOps teams proving what EKS rightsizing saved on the AWS invoice

CostAnalyst reads your AWS bill, including split cost allocation data for EKS, read-only, allocates EKS spend to teams and products next to the rest of your cloud and SaaS spend, and flags anomalies. It sits outside the cluster the rightsizing tool changes, so the before-and-after number finance sees does not come from the vendor being paid on it.