Kubernetes Cost Optimisation: 8 Strategies That Actually Work

Kubernetes is the most powerful infrastructure platform available — and one of the easiest to overspend on. Teams routinely run K8s clusters at 15–20% utilisation while paying for 100%. After optimising infrastructure for a dozen clients, here are the eight strategies that consistently deliver the biggest savings.

Strategy 1: Right-Size Your Resources

Most teams set resource requests and limits conservatively — and never revisit them. A pod requesting 2 CPU and 4GB RAM that actually uses 0.3 CPU and 600MB RAM is wasting 85% of its allocated capacity. Use Kubernetes VPA (Vertical Pod Autoscaler) in recommendation mode for 2 weeks, then apply the suggested values.

apiVersion: autoscaling.k8s.io/v1
kind: VerticalPodAutoscaler
metadata:
  name: api-vpa
spec:
  targetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: api
  updatePolicy:
    updateMode: 'Off'  # recommendation mode — won't change pods automatically
  resourcePolicy:
    containerPolicies:
      - containerName: api
        minAllowed: { cpu: 50m, memory: 64Mi }
        maxAllowed: { cpu: 2, memory: 2Gi }

The 8 Strategies

1. <strong>Right-size requests and limits.</strong> Use VPA recommendations. Typically saves 20–40% immediately.
2. <strong>Use Spot/Preemptible instances for non-critical workloads.</strong> Spot instances are 60–90% cheaper. Use them for batch jobs, CI runners, and stateless services with proper disruption budgets.
3. <strong>Implement cluster autoscaling.</strong> Scale down during off-peak hours. A dev/staging cluster running overnight is wasted spend.
4. <strong>Use Reserved Instances for baseline capacity.</strong> Commit to 1-year Reserved Instances for your predictable baseline. 30–40% discount vs on-demand.
5. <strong>Consolidate small clusters.</strong> Running 10 small clusters is more expensive than 2 medium ones. Namespace isolation is sufficient for most use cases.
6. <strong>Implement KEDA for event-driven scaling.</strong> Scale to zero when queues are empty. Huge savings for batch and async workloads.
7. <strong>Optimise container image sizes.</strong> Smaller images mean faster pulls, less storage cost, and faster scaling events.
8. <strong>Use Kubecost or OpenCost for visibility.</strong> You cannot optimise what you cannot see. Namespace-level cost visibility changes team behaviour immediately.