Cloud bills accumulate quietly. Resources provisioned for a specific project outlive their usefulness by months. The quick wins below are common, and most of them can be identified quickly with a structured review.
Where savings usually come from
Savings opportunities generally fall into three categories: idle resources, right-sizing opportunities, and commitment purchase gaps. The mix varies by workload maturity and account age, so measure your own account rather than relying on a rule of thumb.
1. Idle Elastic IPs
Elastic IP addresses are free when attached to a running instance or network interface. When they are unattached — allocated but not associated with any running resource — AWS charges approximately $0.005 per hour, or around $3.65 per month per address. That sounds trivial, but accounts that have been running for years often have dozens of unattached Elastic IPs that nobody realized they were paying for.
Find unattached Elastic IPs in your account:
# List all unassociated Elastic IPs
aws ec2 describe-addresses \
--query 'Addresses[?AssociationId==null].[PublicIp,AllocationId,Tags]' \
--output table
# Check across all regions
for region in $(aws ec2 describe-regions --query 'Regions[].RegionName' --output text); do
echo "=== $region ==="
aws ec2 describe-addresses --region $region \
--query 'Addresses[?AssociationId==null].[PublicIp,AllocationId]' \
--output table
done2. Unattached EBS volumes
When an EC2 instance is terminated, its EBS volumes are not deleted by default — unless the instance was launched with DeleteOnTermination: true for the root volume (which it is, by default) and data volumes (which it is not, by default). Data volumes left behind after instance termination continue to accrue storage charges indefinitely.
Before deleting unattached volumes, snapshot them first. Storage charges for snapshots are significantly lower than for provisioned EBS volumes, and you retain the ability to restore if the data turns out to be needed.
# List unattached EBS volumes with size and cost estimate aws ec2 describe-volumes \ --filters Name=status,Values=available \ --query 'Volumes[*].[VolumeId,Size,VolumeType,CreateTime,Tags[?Key==`Name`].Value|[0]]' \ --output table
3. Over-provisioned RDS instances
RDS instances are frequently over-provisioned. A database started as db.r5.4xlarge during a high-load period never gets right-sized when load drops. Multi-AZ deployments that were required for a compliance audit remain in place after the audit closes even when the database serves internal tooling that does not require that level of availability.
Use CloudWatch metrics to understand actual utilization before making right-sizing decisions. Look at CPUUtilization, DatabaseConnections, and FreeableMemory over a 30-day lookback period before deciding on a new instance type.
# Get average CPU utilization for an RDS instance over 30 days aws cloudwatch get-metric-statistics \ --namespace AWS/RDS \ --metric-name CPUUtilization \ --dimensions Name=DBInstanceIdentifier,Value=my-database \ --start-time $(date -u -d '30 days ago' '+%Y-%m-%dT%H:%M:%SZ') \ --end-time $(date -u '+%Y-%m-%dT%H:%M:%SZ') \ --period 86400 \ --statistics Average \ --output table
4. NAT Gateway data processing costs
NAT Gateway charges appear in two forms: hourly charges ($0.045/hour per NAT Gateway) and data processing charges ($0.045 per GB processed). The data processing charge catches many teams off guard because it applies to all traffic that flows through the NAT Gateway — including traffic between services in different Availability Zones.
The most common cause of unexpectedly high NAT Gateway costs is cross-AZ traffic. If your application in us-east-1a communicates with a service in us-east-1b via a NAT Gateway in us-east-1a, you pay both the cross-AZ data transfer charge and the NAT Gateway processing charge. The fix is to route intra-VPC traffic via private IP addresses — which avoids both charges entirely — rather than routing it through the NAT Gateway.
5. S3 storage class optimization
S3 Standard is the most expensive storage class and the default for all new objects. Objects that are rarely accessed — log archives, database dumps, old report outputs — should move to cheaper storage classes automatically via lifecycle policies. S3 Intelligent-Tiering is the right choice when you genuinely do not know the access pattern of your objects: it monitors access and moves objects between tiers automatically, without retrieval charges.
For buckets where access patterns are predictable, use explicit lifecycle rules: transition to Standard-IA after 30 days, Glacier Instant Retrieval after 90 days, and Glacier Flexible Retrieval after 180 days. The storage cost difference between S3 Standard and Glacier Flexible Retrieval is roughly 85%.
6. Savings Plans vs Reserved Instances
For workloads with stable, predictable compute usage, commitment-based discounts — Reserved Instances or Savings Plans — are the highest-leverage cost reduction available. Discounts range from 30–60% compared to On-Demand pricing for a 1-year commitment, and 40–72% for a 3-year commitment.
Compute Savings Plans are almost always preferable to EC2 Reserved Instances for most teams. They apply automatically across EC2, Lambda, and Fargate regardless of instance type or region — which means they remain useful even as your workload evolves. EC2 Reserved Instances are appropriate only when you have a specific instance type in a specific region that you are confident will not change for the commitment period.
The AWS Cost Explorer Savings Plans recommendation page analyzes your actual usage and recommends a commitment size. Start with a 1-year, no-upfront Compute Savings Plan at the coverage level the recommendation engine suggests — then review coverage quarterly and top up as workload grows.
The quick wins — idle resources and unattached storage — are worth addressing before committing to Reserved Instances or Savings Plans. You do not want to commit to paying for compute capacity you are about to right-size downward.