How a B2B SaaS Platform Halved Kubernetes Compute Costs with Karpenter and Right-Sized Workloads

A 120-engineer SaaS company cut EKS compute spend by 52% in 90 days by replacing Cluster Autoscaler with Karpenter, enforcing VPA-driven right-sizing, and enabling consolidation. The lesson: most Kubernetes waste is invisible until you measure it.

Containers & K8s · advanced · 12 min read · 2222 words · By Sandip Das · Published

Composite scenario: built from common industry patterns; the figures are illustrative.

What this covers

  1. The Situation
  2. The Problem
  3. Constraints
  4. Options Considered
  5. The Solution
  6. Implementation
  7. Results
  8. What Went Wrong Along the Way
  9. Lessons Learned
  10. Apply It Yourself

Key numbers

MetricBeforeAfter
Monthly EC2/EKS compute cost$310,000$149,000
Average CPU request utilisation18%61%
Node count (production cluster)180 nodes74 nodes
Spot instance share of compute0%58%
API p99 latency (30-day avg)610 ms598 ms

Key components

  • Karpenter NodePool with consolidation — Replaces Cluster Autoscaler and continuously bin-packs running workloads onto fewer, cheaper instances using the disruption consolidation policy.
  • Vertical Pod Autoscaler (Off mode) — Generates resource request recommendations without evicting pods, feeding a PR-based workflow that keeps humans in the loop for right-sizing decisions.
  • Spot NodePool for batch and workers — Routes interruption-tolerant workloads to Spot capacity across six EC2 instance families, delivering the largest single chunk of cost reduction.
  • Kubecost Community Edition — Provides per-namespace and per-deployment cost allocation that surfaces waste and tracks savings against the baseline.
  • PodDisruptionBudgets — Prevent Karpenter from over-consolidating the API tier during low-traffic windows that precede traffic spikes.
  • AWS Node Termination Handler — Intercepts EC2 Spot interruption notices and cordons nodes, giving pods the full 2-minute window to checkpoint state and terminate gracefully.

The Situation

This is a composite scenario assembled from patterns observed across multiple production Kubernetes environments; the organisation, team structure, and figures are illustrative but internally consistent.

The organisation in question is a 120-engineer B2B SaaS company running a multi-tenant analytics platform. Their backend is entirely on AWS: three EKS clusters (production, staging, and a shared tooling cluster), roughly 400 microservices written in Go and Python, and a handful of GPU-backed ML inference services. The platform team — six engineers — owns the clusters. Individual product squads own their own Helm charts and resource manifests. Monthly AWS spend was approximately $480,000, with EC2 and EKS node costs accounting for $310,000 of that.

---

The rest of this case study — the trade-offs, the scaling numbers and the build prompt — is available to LearnXops members. Create an account or see membership.

Topics: kubernetes, karpenter, cost-optimization, eks, autoscaling, containers

Related case studies

Browse all case studies