Kubernetes FinOps: Why Your Cluster Costs 3× What It Should

Kubernetes clusters are where cloud budgets go to hide. Utilization of 20–30% is typical in the estates we audit — meaning two-thirds of spend buys idle capacity. The fixes are systematic and boring, which is why they work.

Start with the requests audit

Developers set resource requests defensively, then never revisit them. Compare requested vs. actual usage per workload over 30 days. We routinely find pods requesting 4 CPUs that use 200 millicores. Rightsizing requests alone typically recovers 30–40% of cluster capacity.

The autoscaling stack that works

• Horizontal Pod Autoscaler on real signals (queue depth, RPS), not just CPU
• Karpenter or cluster autoscaler for node provisioning that matches pod shapes
• Vertical Pod Autoscaler in recommendation mode to keep requests honest over time

Spot instances are free money for the right workloads

Stateless services, batch jobs and CI runners tolerate interruption. With proper disruption budgets and multi-instance-type node pools, spot capacity runs at 60–80% discount with negligible operational pain. Keep stateful and latency-critical workloads on on-demand.

Make cost visible per team

Namespace-level cost allocation with showback dashboards changes behavior faster than any policy. When a team sees their service costs $14k/month, the requests audit happens without being asked.

The compounding effect

These practices stack: rightsizing × autoscaling × spot typically halves cluster cost. One client's EKS bill dropped from $87k to $41k monthly — with better performance, because scheduling improved too.