AWS's billing console briefly told customers who usually spend cents or dollars that they owed billions, and in one screenshot, 7.1 trillion dollars. Nobody owed it, but you shouldn't wave this away as harmless.
Late Thursday into Friday, AWS customers opened the Billing and Cost Management console and saw numbers that looked less like cloud bills than sovereign debt. Amazon confirmed the problem on Friday, July 17. The console was showing some customers estimated charges in the millions, billions and, in a few cases, trillions of dollars even though their usage hadn't changed overnight.
According to TechCrunch, screenshots posted by AWS customers showed estimated bills from a few million dollars to hundreds of millions, including one close to 2.5 billion dollars. WIRED reported that Bill Radjewski, who runs CollegeFootballData.com and said his AWS account had never topped two cents a month, received an alert showing more than 1.5 billion dollars in usage fees and a projected August 1 bill above 3 billion dollars. One Reddit screenshot cited by WIRED put current service fees at 7.1 trillion dollars, more than twice Amazon's market value.
Nobody owes any of it. AWS said the estimates did not reflect actual usage or charges, and told customers no action was required. That distinction matters. Real invoices are one thing. Estimated billing data is another. But for a founder watching runway, payroll and cloud spend at the same time, the console is still the place you go when you need to know whether something is burning money right now.
The Bug Was In The Estimate
AWS traced the problem to what it called an issue with unit pricing inside the estimated billing computation subsystem. The timing is specific: the AWS Health Dashboard said incorrect estimated billing data began showing on Thursday, July 16, at 10:38pm Eastern, or 7:38pm Pacific. The company began investigating hours later and then tried to roll back a recent change to the billing computation subsystem.
It didn't work.
By Friday morning, AWS said the rollback had not resolved the issue, according to TechCrunch. That is not just a process detail. A rollback that fails usually means the bad data has already moved somewhere else, or the change being rolled back was not the whole problem. AWS later said it had paused estimated billing computations while it worked through recalculating the numbers.
The Guardian reported the same root cause from AWS's status updates: unit pricing inside the estimated billing computation system. It also put names and faces on the panic. Dan Harvey of the UK charity Learning Through Landscapes said an app that usually cost less than one pound a month showed a 7.8 billion dollar bill. A student in Delhi who usually paid 1.28 dollars saw 10.9 billion dollars. These weren't abstract cloud accounts. They were people opening a dashboard and wondering whether their card, company or project had just been destroyed.
The Trust Problem Is Real
Here's the thing founders should care about: this wasn't some forgotten settings page. It was the billing console. That's the tool companies use to catch runaway spend before it becomes a real invoice.
Startups wire AWS Budgets, Cost Explorer checks and Slack or PagerDuty alerts into daily operations because cloud usage can spike fast. An AI job can run longer than planned. A database can replicate badly. A misconfigured service can sit there all weekend charging money while nobody is looking. You don't need a lecture on FinOps to understand the point. You need the warning system to tell the truth.
When the warning system screams 1.5 billion dollars, the first reaction is panic. The second is disbelief. The third is worse: people start treating the dashboard as noise. That is where a fake trillion-dollar bill can still do real damage. It won't bankrupt a customer once AWS corrects it, but it can make the next alert easier to ignore when the number is merely large rather than absurd.
AWS remains the infrastructure layer under a huge share of the startup and AI market StartupFortune covers, from small side projects to model-training workloads. Founders don't just rent compute from Amazon. They rely on Amazon's accounting machinery to decide when to scale down, when to cap spend and when to call an engineer in the middle of the night. On July 17, that machinery failed in public.
Amazon has not said how many accounts were affected. WIRED reported that Amazon spokesperson Aisha Johnson referred questions to the AWS Health Dashboard, which described the issue as global. TechCrunch reported that Amazon did not answer whether any AWS accounts had been suspended or paused because of the bug.
That unanswered question matters. If the only impact was a set of ridiculous estimates in a console, this ends as an embarrassing AWS incident. If budget alerts, internal automations or customer operations acted on those estimates, the blast radius is larger. AWS can recompute the numbers. It still has to explain why a pricing error could turn cents into trillions in a tool customers use precisely because they cannot afford surprises.
Also read: Moonshot AI's Kimi K3 Beats Claude and GPT Rivals to Top a Coding Benchmark • India's Smartphone Shipments Just Posted Their Worst June Quarter in Six Years • Emirates NBD becomes first Middle East bank to settle dollars on blockchain
