A manufacturing client reduced their AWS bill from $42,000/month to $17,000/month in 90 days using strategies 1, 3, 5, and 7 from this list.
Strategy 1: Right-Size EC2 Instances
Over-provisioning is the #1 source of wasted cloud spend. Most EC2 instances run at under 20% CPU utilization. Use AWS Compute Optimizer to get machine-learning-driven right-sizing recommendations.
aws compute-optimizer get-ec2-instance-recommendations and look for instances with "OVER_PROVISIONED" findings. Right-sizing a single m5.4xlarge to m5.xlarge saves ~$150/month per instance.Strategy 2: Reserved Instances & Savings Plans
On-demand pricing is the most expensive way to run predictable workloads. Switch to:
- 1-year Reserved Instances → 40% savings vs on-demand
- 3-year Reserved Instances → up to 60% savings
- Compute Savings Plans → most flexible, covers EC2 + Lambda + Fargate
Strategy 3: Eliminate Idle & Zombie Resources
This is free money. Run this audit monthly:
# Find unattached EBS volumes (paying for storage with no instance)
aws ec2 describe-volumes \
--filters Name=status,Values=available \
--query 'Volumes[*].[VolumeId,Size,CreateTime]'
# Find unused Elastic IPs (charged even when unattached)
aws ec2 describe-addresses \
--query 'Addresses[?!InstanceId].[AllocationId,PublicIp]'
# Find stopped EC2 instances still incurring EBS charges
aws ec2 describe-instances \
--filters Name=instance-state-name,Values=stopped
Strategy 4: S3 Storage Intelligent-Tiering
S3 Standard-IA and Glacier can save 60–90% on infrequently accessed data. Enable S3 Intelligent-Tiering to let AWS automatically move objects to the right tier:
aws s3api put-bucket-intelligent-tiering-configuration \
--bucket my-bucket \
--id EntireBucket \
--intelligent-tiering-configuration '{
"Id": "EntireBucket",
"Status": "Enabled",
"Tierings": [
{"Days": 90, "AccessTier": "ARCHIVE_ACCESS"},
{"Days": 180, "AccessTier": "DEEP_ARCHIVE_ACCESS"}
]
}'
Strategy 5: Spot Instances for Non-Critical Workloads
Spot Instances offer up to 90% discount vs on-demand. Use them for:
- CI/CD build agents (GitHub Actions self-hosted runners)
- Batch processing jobs and data pipelines
- Dev/test environments (use Instance Termination protection patterns)
- Kubernetes worker nodes in mixed node groups
Strategy 6: NAT Gateway Optimization
NAT Gateway costs are often invisible until your bill arrives. At $0.045/GB processed, a busy NAT Gateway can cost thousands per month. Solutions:
- Use VPC Endpoints (free) for S3 and DynamoDB traffic
- Use Interface Endpoints for ECR, Secrets Manager, SSM
- Route developer traffic through a bastion host instead of NAT Gateway
Strategy 7: Lambda & Serverless Optimization
Lambda is billed per GB-second. Tune memory allocation with AWS Lambda Power Tuning (open source) to find the optimal memory setting — sometimes increasing memory actually reduces cost by shortening execution time.
Strategy 8: RDS & Database Optimization
- Switch dev/test RDS to stop/start schedule (saves 70% on non-prod)
- Use Aurora Serverless v2 for variable-traffic databases
- Enable RDS Proxy for Lambda → RDS connections (reduces connection overhead)
- Migrate read-heavy tables to ElastiCache
Strategy 9: CloudWatch Logs Retention
CloudWatch Logs storage costs add up. Set retention policies on all log groups — most teams don't need more than 30–90 days of logs in CloudWatch (archive to S3 for long-term):
# Set 30-day retention on all log groups
aws logs describe-log-groups \
--query 'logGroups[?!retentionInDays].logGroupName' \
--output text | \
xargs -I {} aws logs put-retention-policy \
--log-group-name {} --retention-in-days 30
Strategy 10: Implement Tagging & Budget Alerts
You can't optimize what you can't measure. Implement mandatory tagging and budget alerts from day one:
Mandatory Tags on Every Resource
Environment— production, staging, developmentTeam— which team owns this resourceProject— the business project/initiativeCostCenter— for chargeback/showback
Set AWS Budgets with alerts at 80% and 100% of your monthly target. Add anomaly detection to catch unexpected spikes within hours, not at month-end.