The Average Company Wastes 35–45% of Their AWS Bill. Here's How to Fix It.

AWS Compute Savings Plans offer up to 66% off On-Demand rates in exchange for a 1- or 3-year commitment to a dollar/hour spend level. Unlike Reserved Instances, Savings Plans apply automatically to any EC2 instance type, size, region, or OS — maximum flexibility. Start with Compute Savings Plans at 1-year, no-upfront: ~40% discount with zero cash commitment. Analyze your last 30 days of EC2 spend, commit to 70% of your consistent baseline. Use aws ce get-savings-plans-purchase-recommendation or the Cost Explorer Savings Plans recommendation tool.

AWS Compute Optimizer analyzes 14 days of CloudWatch metrics and recommends optimal instance types. Enable it free in your account. For each recommendation: check P99 CPU and memory over 14 days. Downsize if P99 CPU < 40%. Use Graviton (ARM) instances — they're 20% cheaper than x86 for equivalent performance (m7g vs m6i). A typical 100-instance cluster with average 20% CPU has 60+ instances that could be halved in size: $60,000–$120,000/year saved.

Spot instances run on unused AWS capacity at up to 90% discount vs On-Demand — but can be interrupted with 2 minutes notice. Best for: batch jobs, ML training, CI/CD build agents, data processing, stateless auto-scaling workers. Use Spot with Auto Scaling Groups (mixed instances policy: Spot + On-Demand). Implement Spot interruption handling with aws ec2 describe-spot-instance-requests and graceful shutdown hooks. A team running $100K/year in batch workloads on On-Demand can save $60K–$90K by moving to Spot.

S3 Standard ($0.023/GB) → S3-IA ($0.0125/GB at 30 days) → S3 Glacier Instant ($0.004/GB at 90 days) → S3 Glacier Deep Archive ($0.00099/GB at 180+ days). Apply lifecycle policies to all log buckets, artifact buckets, and backup buckets. Rule of thumb: if data isn't accessed in 30 days, it should move to S3-IA; 90+ days → Glacier. A team storing 100TB of logs in S3 Standard ($2,300/month) can move to a tiered policy: same 100TB at ~$600/month. Annual savings: $20,400.

Dev and staging environments don't need to run 24/7. An EC2 instance running 168 hours/week can run 56 hours (business hours M–F) instead — 67% reduction. Use AWS Instance Scheduler, Lambda functions, or AWS Systems Manager to automate start/stop based on a schedule. Also: tag all non-production resources with Environment=dev and configure a tag-based policy to stop at 8 PM and start at 7 AM. Average team saves 60–70% on non-production EC2 and RDS.

Common RDS waste: (1) Multi-AZ enabled for dev databases ($2× cost, not needed); (2) Oversized instance class for actual query load; (3) On-Demand instead of Reserved. For production: use RDS Reserved Instances (1-year, no-upfront = 40% discount). For variable-load databases: migrate to Aurora Serverless v2 (scales to zero when idle, ~$0.12/ACU-hour). For dev/staging: single-AZ, smallest instance class that fits workload, stopped nights/weekends.

AWS charges $0.01/GB for cross-AZ traffic and $0.045/GB for NAT Gateway processing. At scale these compound. Tactics: (1) Use VPC Endpoints for S3 and DynamoDB (free — eliminates NAT Gateway charges for those services); (2) Co-locate compute and data in same AZ where possible; (3) Use CloudFront for internet egress ($0.0075/GB vs $0.09/GB direct = 92% cheaper for high-volume content); (4) Enable S3 Transfer Acceleration only when needed (0.08/GB surcharge). A company transferring 1PB/year to internet saves $82,000 routing through CloudFront.

AWS Trusted Advisor (free for all accounts) identifies: idle EC2 instances, unattached EBS volumes ($0.10/GB/month when unattached), unused Elastic IPs ($0.005/hour when unattached), idle load balancers ($0.025/hour + $0.008/LCU-hour). Run Trusted Advisor monthly. Also: use Cost Explorer with "Unblended Cost by Service" + filter by zero-utilization tags. Average audit finds $2,000–$15,000/month in idle resources that can be terminated immediately.

AWS Graviton3 (m7g, c7g, r7g) instances are 20% cheaper than equivalent x86 instances (m6i, c6i, r6i) for the same performance. Most Linux workloads, containerized apps, and web services run on Graviton without code changes. EKS, ECS, and Lambda all support ARM64. Migration path: update instance type in Terraform/CDK, rebuild Docker images for linux/arm64 (multi-arch build with Docker Buildx). Companies spending $500K/year on EC2 save $100K by migrating to Graviton.

If you're spending $1M+/year on AWS, you're eligible for the Enterprise Discount Program — a private pricing agreement with AWS where you commit to a minimum spend over 1–3 years in exchange for a percentage discount on all AWS services. Request an EDP review through your AWS account team. Typical discounts: 10% for $1M/year commitment, 15–20% for $3M+/year, 20–25% for $10M+/year. Always negotiate: start at 25%, settle for 15–18%.