optimizing-costs

Start with quick wins: delete unattached EBS volumes and old snapshots, stop unused dev instances, and right-size instances under 40% utilization. Then purchase Savings Plans or Reserved Instances for baseline workloads to get 40-72% discounts.

Reserved Instances offer the highest discount (up to 72%) but lock instance types, suiting stable databases. Savings Plans are more flexible across instance types, regions, and services like Lambda and Fargate, making them better for variable compute workloads.

Deploy Kubecost or OpenCost for namespace-level cost visibility, then label workloads with team, product, and environment labels. Set resource requests equal to average usage and enforce quotas with ResourceQuota and LimitRange objects.

Use spot instances for fault-tolerant, stateless workloads like CI/CD runners, batch jobs, and ML training with checkpointing, saving 70-90%. Avoid them for databases, real-time services, or jobs without graceful interruption handling.

Low utilization usually means over-commitment or workload migration to different instance types or regions. Monitor utilization weekly targeting above 95%, exchange Convertible RIs, or sell unused RIs on the AWS Reserved Instance Marketplace.

Integrate Infracost into pull requests to estimate Terraform cost changes before merging. Combine this with cascading budget alerts at 50%, 75%, 90%, and 100% thresholds sent to Slack or PagerDuty.