aicalcus.com
Tech & Developer4 min read

Cloud Cost Optimization: How Engineering Teams Cut AWS Bills by 40-60%

The average company wastes 32% of its cloud spend. These are the five cuts that engineering teams make first — and the ones that actually move the number.

JOJames Okafor·
Cloud Cost Optimization: How Engineering Teams Cut AWS Bills by 40-60%

Flexera's 2025 State of the Cloud report found that organizations waste an average of 32% of cloud spend. For a company spending $100,000/month on AWS, that's $32,000/month in recoverable costs — without changing anything about application performance.

Here are the five cuts that consistently deliver the biggest savings.

Cut 1: Reserved Instances and Savings Plans (20-40% savings)

On-demand instances are the most expensive way to run consistent workloads. Reserved Instances and Compute Savings Plans offer:

CommitmentSavings vs. On-Demand
1-year, no upfront30-40%
1-year, all upfront40-50%
3-year, all upfront50-65%

Rule of thumb: Any instance that runs > 720 hours/month (i.e., effectively 24/7) should be covered by Reserved Instances or Savings Plans.

Common mistake: Teams buy Reserved Instances for specific instance types (m5.xlarge) when Compute Savings Plans are more flexible — they apply to any instance type in a region.

Estimated savings for a company spending $80,000/month on EC2 with 70% baseline utilization: $16,000-25,000/month from RI/SP conversion.

Cut 2: Right-Sizing Instances (10-20% savings)

AWS Cost Explorer and Compute Optimizer analyze CPU/memory utilization and flag over-provisioned instances. In practice:

  • 40-60% of EC2 instances run at < 20% average CPU utilization
  • The default response (provision for peak) results in massive idle capacity
  • Moving from m5.2xlarge to m5.xlarge (same workload, lower peak provisioned): saves 50% on that instance

Implementation: Run AWS Compute Optimizer across your fleet. Start with the top 20 instances by monthly cost. Right-size any showing < 20% average CPU over 14 days.

Cut 3: Eliminate Unattached Resources (5-15% savings)

The easiest savings. These resources accumulate silently:

Resource typeTypical monthly costOften-found count
Unattached EBS volumes$8-100/volume50-200 in large accounts
Unallocated Elastic IPs$3.65/IPOften dozens
Unused load balancers$16-22/monthMultiple
Stale snapshots$0.05/GBOften terabytes
Orphaned NAT Gateways$32+/monthMultiple

A monthly audit script that flags unattached resources and sends a Slack alert to the account owner reduces this waste by 80%.

Cut 4: S3 Storage Tiering (10-30% savings on storage)

Default S3 stores everything in Standard (most expensive). Intelligent-Tiering moves objects automatically based on access patterns:

Storage classCost per GB/monthWhen to use
Standard$0.023Frequently accessed
Standard-IA$0.0125Infrequent access
Glacier Instant$0.004Archive, < 1hr retrieval
Glacier Deep Archive$0.00099Long-term, 12hr retrieval
Intelligent-TieringAutomaticUnknown patterns

For buckets with > 1TB of data not accessed frequently, moving to Intelligent-Tiering typically saves 40-60% on storage costs within 90 days as objects migrate to lower tiers.

Cut 5: Data Transfer Cost Reduction (variable, often 10-20%)

Data egress from AWS is expensive and often overlooked:

Transfer typeCost
EC2 to internet$0.09/GB
Cross-region$0.02/GB
Cross-AZ$0.01/GB
To CloudFront$0.00/GB
Via S3 Transfer Acceleration$0.04/GB additional

Fix: Put CloudFront in front of every S3 bucket and any public-facing origin. CloudFront egress is 40-80% cheaper than direct S3/EC2 egress, and cached content has zero origin cost.

Cross-AZ traffic: Often invisible because it's in the VPC. A 3-tier app with traffic traversing AZs repeatedly on every request can generate significant cross-AZ charges at scale. Co-locating components in a single AZ where HA isn't critical reduces this to zero.

The Savings Roadmap

PriorityCutImplementation effortTypical savings
1RI/Savings PlansLow — purchase in console20-40% of compute
2Unattached resourcesLow — audit script5-15% of total
3Right-sizingMedium — testing required10-20% of compute
4S3 tieringLow — lifecycle rules10-30% of storage
5Data transferHigh — architecture change10-20% of data costs

Use the Cloud Cost Calculator to model your potential savings before starting.

Get weekly AI cost benchmarks & productivity data

Join 4,200+ founders, developers, and creators. No spam, unsubscribe anytime.

#aws#cloud-cost#optimization#infrastructure#cost-reduction