Amazon Web Services customers worldwide woke on July 17, 2026 to erroneous bills reaching as high as $1.5 trillion after a glitch in the cloud provider's estimated billing computation subsystem produced wildly inflated charges across its billing and cost management console. The Guardian

The erroneous figures began displaying at 3:38am UK time on Friday, July 17. AWS attributed the fault to "an issue with unit pricing within the estimated billing computation subsystem," according to The Guardian. After approximately an hour and a half of investigation, AWS shut off the bill estimation system entirely. The company said it expected "full resolution to take multiple hours as we work through recomputing the estimated billing data," in a statement also reflected on its health status page.

The scale of the miscalculation varied widely but was uniformly absurd relative to actual usage. Bharath, an AWS customer who posted on X, saw his bill reach $1.5 trillion, with the console reporting usage up 745,728,201,771% over the previous month. Dan Harvey, head of marketing at the UK charity Learning Through Landscapes, received a $7.8 billion bill for an app that normally costs less than a pound per month. Andrea Zuvich, a historian in Bolsover, Derbyshire who runs The Seventeenth Century Lady website, was billed $245 billion against her usual $15 monthly charge. Sachin, a student in Delhi whose typical monthly bill runs $1.28, saw a figure of $10.9 billion. Another unnamed customer reported being billed $256 billion.

AWS apologised "for any confusion and concern around these costs."

The Guardian reported that Amazon had been approached for comment at the time of publication.

The incident zeroes in on a specific architectural vulnerability in AWS's billing stack: the estimated billing computation subsystem, which generates provisional cost projections distinct from final invoiced charges. When the unit-pricing component within that subsystem failed, it cascaded inflated figures into every customer's console view. That AWS engineers took roughly ninety minutes to identify and disable the estimation pipeline, rather than rolling back instantly, points to the complexity of isolating a faulty component within a real-time cost computation architecture that processes usage data across global regions and service tiers.

Several elements matter for those who manage AWS infrastructure or cloud budgets. First, the glitch affected estimated billing data displayed in the console, not final invoiced amounts. Customers were not charged the trillion-dollar figures. But the estimation system is what organisations rely on for real-time cost monitoring, budget alerts, and spend governance. If those figures feed into automated workflows, alerting thresholds, or third-party FinOps tooling, the downstream noise could be considerable even though no money moved. Second, the duration of the outage, with AWS projecting "multiple hours" for recomputation, means that cost visibility for affected customers was degraded for a meaningful window during the business day.

The affected customers span a telling range: a student in Delhi, a small charity in the UK, an independent historian running a niche website. These are the long-tail AWS customers for whom a billing anomaly of this magnitude is not a back-office curiosity but a moment of genuine alarm, particularly if automated payment methods or credit limits are tied to the account. The absence of large enterprise cases in the reported accounts may reflect who is most likely to post publicly about a billing shock, rather than the actual distribution of the error.

The broader question for cloud customers is one of dependency. Estimated billing is a monitoring tool, and when that tool produces output that is not merely inaccurate but off by several orders of magnitude, it exposes the fragility of the trust model between cloud provider and customer. AWS's billing console is, for many organisations, the single source of truth for cloud spend between invoice cycles. This incident makes clear that the estimation layer can fail catastrophically and globally, and that detection-to-mitigation time is measured in hours, not minutes.