Reducing Infrastructure Costs by 40% on Kubernetes

The wake-up call

Our monthly AWS bill hit six figures before we took cost optimization seriously. The cloud makes it easy to spend money — every engineer can provision resources with a click. But bringing costs down requires discipline and the right tooling.

Spot instances

The single biggest cost saver: use spot instances for non-critical workloads. We run 70% of our cluster on spot instances with a fallback to on-demand:

nodeSelector:
  eks.amazonaws.com/capacityType: SPOT

# Fallback via pod topology spread constraints
topologySpreadConstraints:
  - maxSkew: 1
    topologyKey: eks.amazonaws.com/capacityType
    whenUnsatisfiable: ScheduleAnyway

Resource right-sizing

Most teams over-provision resources by 2-3x. Use the Vertical Pod Autoscaler in recommendation mode to get accurate resource requests:

# Install VPA
kubectl apply -f https://github.com/harjjotsinghh

# Get recommendations
kubectl describe vpa my-service-vpa

Cluster autoscaling

Don't run a fixed-size cluster. Use the Cluster Autoscaler to scale nodes up and down based on pod resource requests. This alone saved us 15% on compute costs.

Waste elimination

• Remove unused load balancers (they cost $20/month each just existing)
• Delete untagged resources (they're invisible in cost reports)
• Use S3 Intelligent-Tiering for infrequently accessed data
• Right-size RDS instances (we were running db.r5.4xlarge when db.r5.xlarge was sufficient)

The result

After six months of optimization work, our monthly infrastructure bill dropped by 40% — from six figures to a much more reasonable number. And we maintained the same reliability SLA.