AWS cost optimization checklist: 15 ways to reduce your AWS bill

The sample account used for the figures

The savings below are computed on a fictitious account, typical of a mid-sized company running a web application in the Paris Region (eu-west-3). Your amounts will differ; the method applies as is.

• Production: 12 m7i.xlarge instances running around the clock, 4 of them below 10% CPU.
• Non-production: 10 m7i.large instances running around the clock, used during office hours.
• Storage: 6 TB of attached gp2 EBS volumes, 800 GB of detached gp2 volumes, 3 TB of snapshots older than six months, 15 TB of rarely read S3 Standard objects.
• Network: two VPCs (production and non-production) with one NAT gateway per Availability Zone, six in total; 1.2 TB a month of S3 traffic going through the production NAT; 12 public IPv4 addresses no longer used.
• Other: 1.5 TB of CloudWatch logs with no retention period, 2 Application Load Balancers with no traffic, one EKS cluster in extended support, one unused Interface VPC endpoint across 3 zones.

AWS public on-demand prices for eu-west-3, taken from the AWS Price List API on 23 October 2026, in US dollars (the currency of AWS prices), on the basis of 730 hours a month. No Savings Plan or Reserved Instance is in place at the start.

Wave 1: clean-up, with no risk to production

These levers target resources that are no longer used. They need no load test and no maintenance window, only a check of who owns each resource before deleting it. The details per resource type are in the unused AWS resources checklist.

Start with detached EBS volumes and public IP addresses: since February 2024, AWS charges USD 0.005 an hour for every public IPv4 address, attached or not, which is USD 3.65 a month per address. An Elastic IP reserved for a server deleted two years ago is still being paid for.

CloudWatch logs with no retention are a cost that only grows: every gigabyte ingested stays stored forever. A retention of 30, 90 or 365 days depending on use stops the increase.

Moving from gp2 to gp3 belongs here because it happens live, with no instance stop: gp3 costs 20% less and includes 3,000 IOPS and 125 MiB/s whatever the size. As gp2 gives 3 IOPS per GB, a volume above 1 TB must get on gp3 the IOPS it would otherwise lose (USD 0.0058 per IOPS-month in eu-west-3).

# Unattached EBS volumes (status available)
aws ec2 describe-volumes --filters Name=status,Values=available \
  --query 'Volumes[].[VolumeId,Size,VolumeType,CreateTime]' --output table

# Elastic IP addresses not associated
aws ec2 describe-addresses \
  --query 'Addresses[?AssociationId==`null`].[PublicIp,AllocationId]' --output table

# Log groups with no retention, with their stored volume
aws logs describe-log-groups \
  --query 'logGroups[?retentionInDays==`null`].[logGroupName,storedBytes]' --output table

# Set a 90-day retention
aws logs put-retention-policy --log-group-name /aws/lambda/my-api --retention-in-days 90

# gp2 volumes to migrate, then live migration
aws ec2 describe-volumes --filters Name=volume-type,Values=gp2 \
  --query 'Volumes[].[VolumeId,Size,Iops]' --output table
aws ec2 modify-volume --volume-id vol-0123456789abcdef0 --volume-type gp3

Wave 3: commitment and architecture

Once the baseline is stable, commitment brings the last big saving. On an m7i.large in eu-west-3, a one-year no-upfront Compute Savings Plan brings the hourly price from USD 0.1176 down to USD 0.08919, 24% less; over three years, 47%. The choice between plan types and Reserved Instances is covered in Savings Plans vs Reserved Instances.

• Network: each NAT gateway costs USD 36.50 a month before a single byte, then USD 0.05 per GB processed. An S3 or DynamoDB gateway endpoint is free and takes that traffic out of the NAT. The full calculation is in AWS NAT gateway cost.
• S3 storage: a lifecycle rule moves rarely read objects to Standard-IA (USD 0.0131/GB-month) or Glacier Instant Retrieval (USD 0.005/GB-month) instead of USD 0.024 in Standard. If access patterns are unpredictable, S3 Intelligent-Tiering does it on its own, for USD 0.0025 a month per 1,000 monitored objects.
• Graviton: at the same size, an m7g costs 19% less than an m7i in eu-west-3. The saving is real but needs Arm images and a test campaign: it is a project, not a setting.

# Last month's cost by service
aws ce get-cost-and-usage --time-period Start=2026-09-01,End=2026-10-01 \
  --granularity MONTHLY --metrics UnblendedCost \
  --group-by Type=DIMENSION,Key=SERVICE

# Compute Savings Plan recommendation over 60 days of history
aws ce get-savings-plans-purchase-recommendation --savings-plans-type COMPUTE_SP \
  --term-in-years ONE_YEAR --payment-option NO_UPFRONT \
  --lookback-period-in-days SIXTY_DAYS

What the FinOps audit automates

Silamir's AWS FinOps audit collects read-only data from Cost Explorer, Savings Plans, Compute Optimizer and Trusted Advisor, then looks for unused resources region by region:

• unattached EBS volumes, unassociated Elastic IPs, snapshots older than six months, AMIs older than a year, logs with no retention, VPC endpoints with no traffic over 30 days;
• Savings Plans and Reserved Instances coverage and utilisation, purchase recommendations;
• right-sizing recommendations, and enabling Compute Optimizer when it is off.

Each action states the expected monthly saving, the resources concerned by their identifier and the command to apply it, sorted by return on investment. The audit costs USD 299, billed by AWS, and also covers security, resilience and seven other domains. It changes nothing: it recommends, you decide. To compare with a consultancy, see what a Well-Architected review costs.

Where does your AWS account stand?

Ten questions, three minutes, no access to your account: a security score, a FinOps score and the article to read for each of your three weakest points. Your answers stay in your browser.