Your AWS bill is leaking. Here is where.

You cannot fix what you cannot see. Before changing anything, spend an afternoon getting visibility. AWS gives you most of what you need at no extra cost.

• Open Cost ExplorerGroup last month's costs by service, then by usage type. Two or three lines usually make up most of the bill.
• Tag everythingAdd tags like project, environment and owner to your resources, then turn them on as cost allocation tags. Now you can see what each product or client really costs.
• Set budgets and alertsCreate a monthly budget with an email alert at 80% and 100%, and switch on Cost Anomaly Detection so unusual spikes reach you in days, not at the end of the month.

1. Servers that run all night for nobody

Development, testing and staging machines often run 24 hours a day, seven days a week, while people only use them during office hours. Schedule them to stop in the evening and at weekends. That alone removes well over half of their running hours.

2. Machines that are bigger than they need to be

Instances are often sized for a launch-day guess and never revisited. If a server's CPU and memory sit mostly idle, move it down a size. AWS Compute Optimizer suggests right-sizing based on your actual usage.

3. Paying full price for work that never stops

On-demand pricing is the most expensive way to run something that is always on. For steady workloads, Savings Plans can cut compute costs by up to 72% in exchange for a one or three year commitment. For work that can be interrupted, such as batch jobs, Spot capacity can be up to 90% cheaper.

4. Forgotten leftovers

Every account collects them: disks still attached to nothing, snapshots from projects that ended long ago, load balancers with no traffic and IP addresses nobody uses. Since February 2024, AWS also charges for every public IPv4 address, so a handful of forgotten ones now shows up on the bill.

5. Old storage settings

Many volumes still use the older gp2 type. The newer gp3 type costs up to 20% less per gigabyte and lets you set performance separately, and you can switch without downtime.

6. Files that never move to cheaper storage

Old backups, logs and uploads often sit in the most expensive S3 storage class forever. Add lifecycle rules that move them to cheaper classes as they age, or use S3 Intelligent-Tiering when you cannot predict how often files are read.

7. The NAT Gateway surprise

Private servers that reach the internet through a NAT Gateway pay for every gigabyte that passes through it, and that includes traffic to S3 and DynamoDB in the same region. Add VPC gateway endpoints for those two services. They cost nothing and take that traffic off the meter.

8. Data moving around more than it needs to

Data leaving AWS for the internet costs money, and so does data moving between availability zones. Serve images and downloads through a CDN such as CloudFront, keep chatty services in the same zone where it is safe to do so, and watch for services copying the same data back and forth.

9. Logs kept forever

New CloudWatch log groups are set to never expire by default. Months of debug logs nobody reads add up quietly. Set a retention period on every log group, for example 30 days for routine logs, and archive only what you are required to keep.

10. Older machine types

Newer instance generations usually give more performance for the same money. For many workloads, AWS's own Graviton processors offer up to 40% better price performance than comparable x86 instances. Test your application on them when you next upgrade.

Most teams can plug the biggest leaks in a single focused week without touching anything risky in production.

1. MondayTurn on budgets, anomaly alerts and cost allocation tags. List the top ten cost lines.
2. TuesdayDelete unattached volumes, old snapshots, idle load balancers and unused IP addresses (after checking with their owners).
3. WednesdaySchedule non-production servers to stop outside working hours. Set retention on every log group.
4. ThursdayRight-size the obviously oversized instances. Move gp2 volumes to gp3. Add S3 lifecycle rules and gateway endpoints.
5. FridayCompare the new daily run rate with last week, then decide how much steady usage is safe to cover with Savings Plans.

A clean-up feels great for a month. Without a few habits, the bill drifts back up. The teams that stay lean do three simple things:

• Someone owns the bill. One named person reviews it every month and follows up on anything unusual.
• Cost is part of building. New resources must be tagged, and the expected monthly cost is mentioned when a new service is planned.
• Waste is cleaned up on a schedule. A monthly check for idle and unattached resources catches leftovers before they pile up.

Thinking about adding AI to your workflows next? Read this first, so you know where it saves money and where it does not.