HPA vs VPA vs KEDA: Which Kubernetes Autoscaler Actually Cuts Your Cloud Bill

Short summary

A practical comparison of Kubernetes' three main autoscalers: HPA scales pod count for traffic swings, VPA adjusts per-pod resource requests to fix over-provisioning, and KEDA scales event-driven workloads to zero. The key insight is that pod autoscaling alone doesn't save money — you need node-level autoscaling (Cluster Autoscaler or Karpenter) to actually shrink your bill. Most mature setups combine all three since they address different forms of waste.

• •HPA scales replica count, VPA right-sizes pods, KEDA scales to zero on events
• •Pod autoscaling alone won't cut costs — node autoscaling must reclaim freed capacity
• •Match the autoscaler to your waste type: over-provisioning→VPA, fixed replicas→HPA, idle workloads→KEDA

Generated with AI, which can make mistakes.