The AWS billing bug that surfaced on the night of 16 July sent shockwaves through the cloud computing community: customers opened their dashboards to find estimated charges running into the millions, billions, and in at least one case, the trillions. None of the figures were real. But the hours of confusion that followed were.
Amazon confirmed that a defect inside its billing computation subsystem corrupted unit pricing estimates, hitting the Cost Explorer tool and spreading across the AWS Billing and Cost Management platform. According to Yahoo Tech, the defect first appeared at 7:38 PM PDT on 16 July. By Friday morning, Amazon’s own status page acknowledged that a rollback attempt had failed: ‘the rollback of a recent change did not resolve the issue.’
Amazon was careful to draw a hard line between the phantom figures and anything touching real money. The erroneous estimates ‘do not reflect actual usage and charges,’ the company said. The defect, it stressed, was isolated to the estimation layer and did not affect the actual billing pipeline, invoices, or payment processing.
From Bangalore to Bolsover: How Far the AWS Billing Bug Spread
The reach was global. The Guardian reported affected customers from Bangalore to Bolsover, with some receiving estimated monthly bills as high as $1.5 trillion. Screenshots shared by customers on Reddit showed a range of figures, with one estimate running to close to $2.5 billion for a single month’s usage. The Guardian and the Reddit screenshots describe different figures, likely reflecting different customers or different moments in the glitch’s progression; both outlets confirmed the estimates bore no relation to actual spend.
Among those caught in the chaos was Dan Harvey, head of marketing at Learning Through Landscapes, a Hampshire-based charity whose school grounds audit app normally costs less than £1 per month on AWS. Harvey told The Guardian he entered ‘full panic mode’ and worked with tech support to investigate before establishing that no real charges had been applied.
His situation captures a broader anxiety the incident exposed: for smaller organisations and individual developers, a phantom bill of even a few million dollars can trigger genuine alarm, credit-card freezes, or internal escalations that eat hours of productive time, regardless of whether the figure is real.
Not the First Time AWS Billing Estimates Have Gone Wrong
Amazon has been here before. Yahoo Tech notes that a previous bug involving Lightsail snapshots generated similarly phantom billions in estimated charges, with affected users’ real bills left entirely unaffected. That incident did not produce the same global wave of screenshots and social media panic, but it foreshadowed the same structural fragility: a disconnection between the estimation layer and the actual billing pipeline that, when it breaks, breaks visibly.
The recurrence raises a reasonable question about how Amazon quality-controls changes to the billing computation subsystem, given the scale of alarm even a purely cosmetic error can cause across its customer base.
Amazon’s spokesperson did not return a request for comment by the time of publication.
The restoration timeline is more drawn out than initially suggested. The AWS Health Dashboard indicated, as of 18 to 19 July 2026, that relevant billing operations in at least one AWS region had been suspended while the company worked to restore normal service, with full restoration expected to take several months.
That timescale matters. Customers who rely on Cost Explorer for real-time budget monitoring, cost allocation, or internal chargebacks will be operating without reliable estimates for a sustained period. For finance teams running monthly cloud cost reviews, the absence of that tooling is a practical problem even if the underlying invoices remain accurate.
Amazon has not disclosed what specific change to the billing computation subsystem triggered the defect, nor what safeguards will prevent a recurrence. The company’s next scheduled billing cycle will be the first real test of whether the fix holds, and whether the estimation layer and the actual billing pipeline remain properly separated under normal load.
