Kubernetes Cost Optimization: 10 Strategies to Reduce Cloud Spend

Why Kubernetes Costs Spiral

Kubernetes makes resource allocation easy — which means over-provisioning is common. The three biggest cost drivers:

1. Over-specified resource requests: Pods request 2 CPU and 4GB RAM but average 0.2 CPU and 500MB. You pay for 2 CPU.
2. Under-utilised node pools: Cluster nodes running at 20% utilisation because the cluster was sized for peak.
3. Always-on workloads that could scale to zero: Batch jobs, dev/staging environments, and low-traffic services running 24/7.

Strategy 1: Right-Size Pod Resource Requests

Resource requests determine which node a pod is scheduled on — and which node you pay for. Oversized requests waste reserved capacity.

Use the Vertical Pod Autoscaler (VPA) in recommendation mode:

kubectl apply -f vpa.yaml  # VPA in "Off" mode (recommendations only)
kubectl describe vpa my-deployment
# Shows: Recommended CPU: 250m, Memory: 512Mi
# Your current request: CPU: 2000m, Memory: 4096Mi

Adjust requests based on VPA recommendations. Typical finding: 40–70% of pods are over-provisioned.

Strategy 2: Use Spot / Preemptible Instances

Spot instances (AWS) / Preemptible VMs (GCP) cost 60–80% less than on-demand. They can be reclaimed with 2-minute notice — so they are appropriate for:

• Stateless web services (with multiple replicas)
• Batch processing jobs
• CI/CD runners
• Dev and staging environments

Use node pools with mixed instance types: on-demand for critical pods, Spot for everything else.

Strategy 3: Cluster Autoscaler

Automatically adds and removes nodes based on pending pods and utilisation:

# AWS EKS Cluster Autoscaler
clusterAutoscaler:
  enabled: true
  scaleDownUtilizationThreshold: 0.5  # Remove nodes at 50% utilisation
  scaleDownDelay: "10m"

This eliminates paying for idle nodes during off-peak hours.

Strategy 4: Horizontal Pod Autoscaler (HPA)

Scale deployments based on CPU/memory or custom metrics:

apiVersion: autoscaling/v2
kind: HorizontalPodAutoscaler
metadata:
  name: api-hpa
spec:
  scaleTargetRef:
    apiVersion: apps/v1
    kind: Deployment
    name: api
  minReplicas: 2
  maxReplicas: 20
  metrics:
  - type: Resource
    resource:
      name: cpu
      target:
        type: Utilization
        averageUtilization: 70

Strategy 5: KEDA for Scale-to-Zero

KEDA (Kubernetes Event-Driven Autoscaling) can scale workloads to zero replicas when there is no traffic — and back up when needed:

• Ideal for background workers, queue processors, scheduled jobs
• Scale to zero = zero cost for idle workloads
• Triggers: queue depth, cron schedule, HTTP traffic, custom metrics

Strategy 6: Namespace Resource Quotas

Prevent any single team or service from consuming the entire cluster:

apiVersion: v1
kind: ResourceQuota
metadata:
  name: team-quota
  namespace: team-frontend
spec:
  hard:
    requests.cpu: "10"
    requests.memory: "20Gi"
    limits.cpu: "20"
    limits.memory: "40Gi"
    persistentvolumeclaims: "10"

Strategy 7: Scheduled Scaling for Dev Environments

Scale dev/staging to zero at night and weekends:

# Using kube-downscaler or KEDA cron trigger
schedule: "0 18 * * MON-FRI"  # Scale down at 6pm
scaleUpSchedule: "0 8 * * MON-FRI"  # Scale up at 8am

Savings: dev environments typically represent 15–25% of K8s spend.

Strategy 8: Delete Orphaned Resources

Run monthly cleanup:

# Find unused PersistentVolumeClaims
kubectl get pvc --all-namespaces | grep Released

# Find deployments with zero replicas
kubectl get deployments --all-namespaces | grep "0/0"

# Find unused namespaces
kubectl get namespaces  # Review and delete stale ones

Strategy 9: Use Cost Visibility Tools

You cannot optimise what you cannot see:

• Kubecost — shows cost per namespace, deployment, and pod
• OpenCost — CNCF open-source cost monitoring
• AWS Cost Explorer — EKS cost breakdown by tag
• Datadog Cost Management — real-time K8s spend with anomaly detection

Strategy 10: Reserved Instances for Baseline Load

Use Reserved Instances (1-year or 3-year) for the stable baseline of your node pool. Use Spot for burst capacity. A mix of 60% Reserved + 40% Spot typically delivers 40–50% total savings vs all on-demand.

Need a Kubernetes cost audit? Talk to our cloud team → or contact us to book your free K8s cost review.