SEATTLE – When a screenshot of an AWS billing portal showing nearly $2.5 billion in estimated charges began circulating on Reddit Thursday night, the initial response was disbelief. The number was genuine, in the limited sense that Amazon’s billing computation system had actually generated it. It was also wrong. The platform that processes cloud invoices for a third of the global internet had misfired, manufacturing phantom charges that bore no relationship to actual customer usage.
Amazon confirmed Friday that a bug in its billing computation subsystem had displayed dramatically inflated estimates to an unknown number of AWS customers. The company’s status page attributed the failure to “a recent change” to the billing layer. Screenshots shared on Reddit showed figures ranging from several million to approximately $2.5 billion. Amazon spokesperson Aisha Johnson confirmed the figures “do not reflect actual usage and charges,” meaning no customer was actually billed for the phantom amounts. She declined to comment beyond the status page.
The harder problem was the timeline. Amazon attempted a rollback after identifying the issue. The rollback did not work. By Friday morning, the status page acknowledged the billing portal still displayed inaccurate figures and said engineers estimated the issue would persist for several more hours. The company did not disclose how many accounts were affected, whether any accounts had been suspended due to anomalous spending alerts, or what specifically the “recent change” had done to the billing subsystem.
AWS billing works through a computation layer that aggregates real-time usage data across dozens of cloud service types, including storage, compute, data transfer, machine learning inference, and other metered services, converting that usage into running cost estimates. Because AWS pricing is dynamic, consumption-based, and tiered, actual cloud bills for large enterprise customers can reach seven or eight figures per month without representing anything unusual. That design creates a specific vulnerability: a bug that inflates numbers to $2.5 billion is distressing in a way a $25,000 overcharge is not, because $2.5 billion is implausible for most customers but not obviously implausible to an IT team watching AI workloads scale.
The incident landed against a backdrop of accelerating cloud spending. IBM’s worst single-session stock drop in its history came two weeks ago when the company disclosed that enterprise customers were moving technology budgets toward AI infrastructure rather than IBM software. Cloud spending is growing fast enough that unusual billing figures have become harder to dismiss. When customers are already watching their AWS bills rise sharply quarter over quarter, a billing alert showing an extraordinary figure is not immediately legible as a system error.
Microsoft’s $2.5 billion AI deployment arm, launched earlier this month to place engineers inside Fortune 500 clients, followed Amazon’s own $1 billion AI infrastructure commitment by two days. Both companies are betting that AI workloads will scale large enough to require dedicated deployment support, and both are running infrastructure expansions to service that demand. The billing environment is a direct function of that growth: larger workloads generate larger bills, and customers monitoring those bills in real time are more exposed to the operational disruption that comes when the billing layer misfires.
The rollback failure is the detail Amazon has said least about. AWS engineering practices are built around fast deploy and fast revert: the ability to push a code change, identify a problem, and roll it back without extended service disruption. When a rollback fails, it typically means the change interacted with a downstream dependency in an unexpected way, or that the reversal introduced a different problem. Either scenario implies the billing subsystem change touched something broader than a simple configuration update. Amazon has acknowledged the failure without explaining it.
AWS remains the world’s largest cloud provider by revenue. AI chip demand has pushed TSMC to forecast growth above 40 percent for 2026, a measure of the same underlying trajectory that is driving AWS customers to expand their deployments. The billing console is one of the primary tools those customers use to manage their cost exposure. An incident that makes the console untrustworthy, even temporarily, undermines a tool that influences real budget decisions in real time.
According to TechCrunch, the issue began appearing late Thursday and was still unresolved Friday morning. Companies with automated cost governance systems – rules that pause or restrict deployments when spending estimates exceed predefined thresholds – would have been particularly exposed. If a billing alert fires on a phantom figure, the automation that responds to it does not know the figure is false. The operational disruption that follows is real even when no money changes hands.
What Amazon has not addressed is whether any AWS accounts were suspended as a result of the anomalous billing figures, and how it will verify this after the fact. Cloud accounts can be restricted automatically when spending alerts cross configured thresholds, and that restriction mechanism does not distinguish between genuine and spurious figures. Amazon has confirmed the numbers were wrong. It has not confirmed what those numbers triggered before anyone confirmed they were wrong. The status page remained unresolved into Friday evening.
