Amazon Claude overruns reveal AI cost risks

Amazon internal reports show AI projects have driven unexpected spending spikes, with one failed Claude Sonnet deployment producing a $1.8 million bill. The system was meant to match author details with Amazon listings, but it ran 860% over its allocated budget before the issue was detected some five months after it began.

Other overruns included a $541,000 additional cost from a project building a financial auditing tool and a $134,000 extra expense for a logistics system intended to reduce delivery times. Amazon said in an internal presentation that teams are experimenting, learning and improving how they use AI, including cost efficiency, and said isolated examples do not reflect broader use across the company.

The expenses are small relative to Amazon’s latest quarterly revenue of more than $181 billion, but the reports highlight how token-based AI usage can escalate quickly. AWS previously limited AI coding bot access after outages tied to coding mistakes, and Amazon also dropped an internal leaderboard that tracked which employees used AI the most as spiraling costs prompted a rethink.