Customers of Amazon Web Services received a rude shock ahead of the weekend as they received alerts showing massive bills from the cloud giant, some of which were in the billions and even trillions of US dollars.
AWS customers posted screenshots of their billing dashboards, with some whose normal usage is in the tens of dollars per month being told that they would owe billions or even more.
The issue appears to have affected customers worldwide, and continued on Saturday.
AWS's Cost Explorer, Billing and Cost Management Console and the Cost and Usage Reports which feeds into enterprise customers' finance systems, were all affected.
Datadog staff engineer and AWS Hero AJ Stuyvenberg was one of several AWS customers to share billing estimates.
Alright 55T (estimated 103T) is the highest AWS bill I've seen so far, who can beat it? pic.twitter.com/Z95W0H1Hz4
— AJ Stuyvenberg (@astuyve) July 17, 2026
AWS acknowledged the fault, and added that "the displayed billing estimates do not reflect actual usage and charges."
Engineers traced the root cause to a unit pricing error within the estimated billing computation subsystem, which is tthe internal engine responsible for calculating customers' estimated costs. and which appears to have received an update prior to the problem occurring.
AWS first paused estimated billing computations altogether, to stop the inflated numbers climbing any further.
The cloud giant then attempted two parallel fixes: reverting to the last known accurate billing data, and rolling back a recent change to the computation subsystem.
Unfortunately, the rollback attempt failed to resolve the issue, forcing engineers back to the drawing board and further delaying recovery.
It has now started backfilling corrected data, while acknowledging that progress has been slower than first anticipated.
AWS said the "majority" of affected accounts were remediated by July 18 at 11pm Australian time.
In February this year, AWS posted a rebuttal for a Cost Explorer incident reported by Financial Times, which took place in December last year, and which the publication attributed to artificial intelligence and the Kiro coding agent.
This was not the case, AWS said.
Instead of AI, AWS said the service interruption was due to user error, stemming from a misconfigured role.

iTnews State of Data & AI Breakfast
Forrester's AI Forum Sydney
The 2026 iAwards
Integrate 2026
Security Exhibition & Conference



