AKS Cost Optimization: The Real Fixes Under the Hood – NAP, VPA, Spot, and Commitments

AKS clusters often carry the worst utilization metrics in cloud compute — but there’s no single setting that fixes it. This post breaks down the real levers: Node Auto Provisioning vs. Cluster Autoscaler, pod rightsizing with VPA, Spot VM strategy, GPU sharing, ARM adoption, and why commitments should come last, not first — with real examples from Caudalie and Mercedes-Benz.io.