7 Practical Ways to Reduce AWS Cloud Costs

Before you start: find out where the money goes

Open Cost Explorer, set the last three months, and group by Service. In most accounts four or five lines (EC2, RDS, EBS, data transfer and S3) make up the bulk of the bill. Then group by Usage type inside the biggest service; that is where surprises like "NatGateway-Bytes" or "EBS:SnapshotUsage" show up.

If resources are not tagged, start now with three tags: environment (production, staging, dev), application and owner. Activate them as cost allocation tags in the Billing console so spend can be grouped by team or product from next month on.

1. Remove unused and forgotten resources

Nothing depends on these, which makes them the safest savings on the bill.

# Volumes not attached to any instance
aws ec2 describe-volumes --filters Name=status,Values=available \
  --query 'Volumes[].{id:VolumeId,gb:Size,created:CreateTime}' --output table

# Elastic IPs not associated with anything
aws ec2 describe-addresses --query 'Addresses[?AssociationId==null].PublicIp'

# Your snapshots, oldest first
aws ec2 describe-snapshots --owner-ids self \
  --query 'sort_by(Snapshots,&StartTime)[].{id:SnapshotId,gb:VolumeSize,date:StartTime}' --output table

2. Right-size oversized servers

AWS Compute Optimizer analyzes CloudWatch metrics and recommends smaller or different instance types. Memory is not collected by default, so install the CloudWatch agent first if you want memory-aware recommendations; otherwise you are guessing at half the picture.

• Size to the 95th percentile of the busiest realistic period, and keep 20–30% headroom.
• Move to newer instance generations, which usually cost the same or less for more performance.
• Test AWS Graviton (ARM) instances for Linux workloads. AWS cites up to 40% better price-performance; most Node.js, Python, Java and .NET applications run unchanged, but test first.
• Check RDS instances too; databases are frequently the most oversized line on a bill.

3. Optimize storage

{
  "Rules": [{
    "ID": "archive-then-expire-logs",
    "Filter": { "Prefix": "logs/" },
    "Status": "Enabled",
    "Transitions": [{ "Days": 30, "StorageClass": "DEEP_ARCHIVE" }],
    "Expiration": { "Days": 730 }
  }]
}

If you do not know how often data is read, S3 Intelligent-Tiering moves objects between tiers automatically for a small monitoring fee per object.

4. Commit to reserved capacity for steady workloads

If part of your usage runs 24 hours a day, every day, you are paying on-demand rates for something predictable. A one- or three-year commitment earns a discount.

AWS advertises discounts of up to 72% for Savings Plans and up to 90% for Spot compared with on-demand; real savings depend on term, payment option and instance type. Cost Explorer's Savings Plans recommendations show the commitment that fits your last 30 or 60 days of usage.

5. Schedule non-production resources

A week has 168 hours; office hours (12 hours a day, 5 days a week) are 60. Running dev and test only during office hours cuts their compute hours by about 64%.

The Instance Scheduler on AWS solution starts and stops EC2 and RDS instances based on a tag such as Schedule=office-hours. Give developers a simple way to override it for late nights, or they will find a way around it.

6. Reduce data transfer and networking charges

Networking charges are easy to miss because they never show up as a "server."

• NAT gateway processing. Every gigabyte through a NAT gateway costs about $0.045. Servers in private subnets that read 2 TB a month from S3 through a NAT gateway pay about $92 a month for nothing. An S3 gateway endpoint carries that traffic with no data processing charge.
• Data transfer out. Put Amazon CloudFront in front of public content; it caches close to users and its outbound pricing is often lower than serving directly.
• Cross-zone traffic. Traffic between Availability Zones is charged in each direction (about $0.01 per GB each way). Very chatty services spread across zones can quietly add up; keep tightly coupled components close where reliability allows.

7. Monitor continuously and give cost an owner

• Create AWS Budgets with alerts at 80% and 100% of expected monthly spend, sent to a person, not a shared inbox.
• Turn on Cost Anomaly Detection so a runaway resource is flagged within a day, not at month-end.
• Review costs monthly with each application owner, using the tags from the first step.
• Track a unit cost the business understands, such as AWS cost per active customer or per 1,000 orders, so growth and waste are easy to tell apart.

Saving money without hurting reliability

• Change one thing at a time and watch the metrics for a week before the next change.
• Never remove production redundancy to save money: Multi-AZ databases and spare capacity exist for the day something fails.
• Test instance changes in staging first, especially moves to Graviton or a new family.
• Use Spot only for work that can be interrupted and retried.
• Keep backup retention aligned with your recovery requirements, not the cheapest setting.

If the bill jumped after a migration, our AWS cloud migration guide explains why copying server sizes 1:1 is a common cause.