Our team cut six figures off annual cloud spend without downsizing a single workload. Every tactic below is something we actually shipped, not a slide deck idea. If you run production workloads on AWS, GCP, or Azure, at least 5 of these will save you money this quarter.

1. Find the zombies

Every cloud account has unallocated Elastic IPs, orphaned EBS volumes, idle load balancers, and unused NAT gateways. Run a sweep monthly. The most common zombie we find is an unattached 1TB gp3 volume costing 96 USD a month. Multiply by a hundred.

2. Right size before you reserve

Reserved Instances and Savings Plans are a trap if you have not right sized first. We saved 38 percent on EC2 alone by moving from m5.4xlarge (mostly idle) to m6i.2xlarge. Reservation came after, on the new footprint.

3. Spot instances for stateless workloads

CI runners, batch jobs, dev environments, and async workers should run on Spot. Spot pricing is 60 to 90 percent cheaper than On Demand. The interruption rate in 2026 is under 5 percent for most instance families.

4. S3 lifecycle policies

Most logs and backups never need hot storage. Move anything older than 30 days to S3 IA, anything older than 90 days to Glacier Instant Retrieval. The pattern we ship by default:

5. Graviton migration

AWS Graviton3 and Graviton4 are 20 to 40 percent cheaper per compute unit than x86 for most workloads. Most container images run on ARM with zero code change. We migrated 240 services over 6 months and saved 31 percent on EC2 and ECS combined.

6. Egress is the silent killer

Cloud to cloud egress is 0.09 USD per GB. Cross region is the same. CDN egress is 0.08 USD per GB. We saved 180,000 USD a year by moving all static assets behind CloudFront and fronting the API with a regional cache layer.

7. Autoscaling that actually works

Most teams set fixed min capacity and forget. We replaced fixed minimums with predictive autoscaling and reduced idle EC2 hours by 42 percent. The trick: scale on a schedule for known patterns (deploy windows, business hours) and on metrics for the unknown.

8. Logs are usually over retained

Default log retention is forever. We changed every log group to 30 days hot, 90 days in S3, then deleted. CloudWatch Logs are expensive. S3 + Athena queries are 90 percent cheaper.

9. Commit based pricing on data transfer

AWS CloudFront and Azure Front Door offer committed pricing tiers. If you serve over 100TB a month, the tier pricing pays for itself in the first week.

10. Buy a FinOps tool, not a consultant

Vantage, CloudHealth, and Anaconda Cloud cost 30 to 60K a year but find waste that pays for them in the first month. We found 22 percent more waste in the first 60 days than we ever found manually.

Savings by tactic (real numbers, 12 month period)

TacticAnnual SavingsEffortRisk
Zombie cleanup14,200 USD1 dayNone
Right sizing112,000 USD2 weeksLow
Spot for stateless78,400 USD3 weeksLow
S3 lifecycle23,600 USD1 weekNone
Graviton migration186,000 USD6 monthsMedium
Egress optimization180,000 USD2 monthsLow
Predictive autoscaling96,500 USD1 monthLow
Log retention policy34,800 USD2 weeksNone
CDN committed pricing52,000 USD1 dayNone
FinOps tooling64,000 USD2 weeksNone
Total841,500 USD

Monthly cost per workload tier

WorkloadBeforeAfterReduction
API tier28,400 USD17,200 USD39%
Background workers14,600 USD5,400 USD63%
Data pipeline22,800 USD15,600 USD32%
Storage11,200 USD6,800 USD39%
Observability9,400 USD5,200 USD45%
Total86,400 USD50,200 USD42%

Related video: AWS re:Invent 2025: What is new in AWS cost optimization (AWS Events)

How to start this week

Run a zombie sweep today (1 hour, saves 5 to 20K a year). Right size your top 10 instances this week. Move S3 buckets to lifecycle policies by Friday. Buy a FinOps tool on Monday. The compound effect of these four moves alone is usually 15 to 25 percent of your bill, and the team will thank you for it within a quarter.