Your AWS credits are ending. Here's what your real bill will be.
For a year or two the invoice says $0 and nobody looks at it. Then the credits expire, or run out early, and a bill that grew quietly the whole time lands on your runway in one month. We find it before it lands.
Fig. 1
Illustrative: $6,400 a month of usage growing 5% a month, with $80k of credits expiring in month 14.
Why a $0 bill lies to your runway spreadsheet.
Credits pay the bill automatically, so the gross number on the AWS invoice is the one to watch. In the example above, $80,000 of credits were supposed to last 14 months. Growth used them up by month 10, and the first real invoice was over $10,000.
Try it with your own numbers in the runway calculator. It shows the month your credits really run out and how many months of runway the cliff costs you.
Prices: AWS list prices, us-east-1, Oct 2026.
Where startup AWS bills usually leak.
| Line item | Why it grows | Typical cost |
|---|---|---|
| NAT gateways | One per availability zone, per environment, running all month, plus a per-GB charge on traffic | $32.85 /mo each + $0.045 /GB |
| Public IPv4 addresses | Charged since 2024 whether used or idle | $3.65 /mo each |
| EKS control planes | One per cluster; much more on extended support for old Kubernetes versions | $73 /mo, or $438 on extended support |
| Staging and preview environments | Sized like production and left on at night and weekends | often 30 to 60% of the bill |
| CloudWatch logs | Debug logging left on, kept forever | grows with traffic |
| RDS | Instances sized for launch day; Multi-AZ on non-production | varies |
Sources: Amazon VPC pricing, Amazon EKS pricing. The staging figure is what we tend to see, not an AWS statistic.
Our credits still have a year left. Is it too early?
It's the best time. Changes are cheaper to make before the bill is real, and you get to plan your fundraise around the true number. The review takes about a week and needs read-only billing access.
Will cutting costs make the app slower or less reliable?
We change one thing at a time and check latency and error rates after each change. Removing idle resources carries almost no risk; resizing a busy database is scheduled with your team and has a rollback plan.
Can't we just buy a Savings Plan?
Sometimes that's the right first step. But committing to a year of spend on resources you don't need locks the waste in. We trim first, then commit to what's left.
Do you resell AWS or take a cut of anything?
No. We're not an AWS reseller and we don't get paid by any tool we recommend. Your AWS contract stays exactly where it is.