Challenge
A growing Kubernetes estate on AWS was over-provisioned and under-governed — idle capacity, oversized resource requests and missing guardrails meant the bill kept climbing ahead of actual usage.
What we did
- Implemented Karpenter for just-in-time, right-sized node autoscaling
- Enforced cost-control and resource policies as code with Kyverno
- Right-sized workload requests and limits against observed usage
- Consolidated workloads to eliminate idle capacity
Results
~−30%
AWS compute cost
2x
Cluster utilization
Automated
Policy guardrails (Kyverno)
Just-in-time
Node autoscaling (Karpenter)