AWS billed one customer $1.7 billion, then blamed a 'unit pricing' bug

AWS's own postmortem: a 'unit pricing' bug nobody could roll back on the first try

On July 17, AWS's Billing and Cost Management Console began showing estimated costs with no relationship to real usage — accounts that normally sat at a few dollars a month were suddenly quoted in the billions. AWS's own incident log, relayed by a reader who'd been tracking the status page, laid out what the company actually knew:

"We continue to work to resolve the issue affecting estimated cost and usage data displayed in the Billing and Cost Management Console. We have identified the root cause as an issue with unit pricing within the estimated billing computation subsystem and we are working on a mitigation. The displayed billing estimates do not reflect actual usage and charges. There are no customer actions required at this time." — AWS's own incident log, relayed by a reader

The scale varied by account in ways that had nothing to do with real usage: one reader's routine budget alert read $2.8 million; another watched a single Glacier archive get quoted at nearly $242 billion. None of it touched real invoices — the estimate feed had simply detached from AWS's actual metering pipeline, which kept billing normally underneath the noise. AWS tried a rollback next; the status page said so directly:

"The rollback of a recent change did not resolve the issue and we are continuing to investigate multiple mitigation paths. Estimated bill updates remain paused." — a reader tracking AWS's live status updates

The bytes-for-gigabytes theory that actually explains the math

Buried well below the panic, one comment offered a plausible mechanism for how a billing system produces numbers a billion times too large:

"Apparently what used to be GB of storage consumed is confused with Bytes of storage consumed, leading to a cool off by 2*30 error." — a commenter who traced the unit-conversion math

A gigabyte is 2^30 bytes, so mixing the two units up produces almost exactly the billion-times inflation people were reporting — consistent with AWS's own admission that the fault sat in "unit pricing" logic, not in the real usage-tracking systems underneath it.

'If you owe the bank a billion, that's the bank's problem' — except one reader called that bluff

The thread reached for a decades-old joke about debt and scale so often it became a running bit: owe a little, it's your problem; owe enough, and it becomes the lender's. One reply refused to let the joke stand unexamined:

"This joke only works if you actually impose a cost on AWS… If they just serve you a bill for no reason, it's still your problem." — a commenter who ran the joke's own logic to its conclusion

A live AWS job posting made the 'vibe-coded billing' joke look less like a joke

Multiple commenters guessed, half-joking, that an AI coding agent had shipped the bug. Then someone went looking for evidence instead of just guessing:

"Wanna bet the description of this job post will be updated by the end of the day? 'Software Development Engineer II, AWS Invoicing'… Our platforms are powered by generative AI, large language models, knowledge graphs, and agentic architectures that dynamically compose specialized agents based on context." — a reader who found Amazon's own recruiting page for the invoicing team

Nothing in AWS's own incident notes names a cause beyond "unit pricing," so the connection stays circumstantial. But a live job ad describing agentic AI actively building the invoicing platform was enough to make the joke theory feel a lot less like a joke.

The guardrail that a billing system this size evidently didn't have

Past the jokes, one reply pushed for a very specific engineering question:

"…I'm also a little surprised this didn't trip a circuit breaker. For something as non-real-time as billing, I'm surprised they don't have an automated kill switch that pauses the billing system and fires a page if variance in bills spikes." — an engineer who spec'd out a variance-based kill switch for billing systems

An engineer who'd spent a stretch of their career on the other side of that math had already made the point personally:

"At some point my role was to reduce our startup's AWS bill. I managed to keep 7 figures on our books instead of handing it to AWS… 'Well-architected' is for the hyperscalers' balance sheet, not yours." — an engineer who cut seven figures off their own startup's AWS bill

AWS spends real effort telling customers, through its own Well-Architected framework, to build exactly the kind of anomaly detection that its own billing dashboard evidently didn't have for itself.