Amazon’s Claude Bill 9x Over-Budget Following ‘Catastrophically Expensive’ Coding Project

Amazon’s internal use of Anthropic’s Claude Sonnet for a routine coding task appears to have cost the company $1.8 million, bringing it 9.6 times over budget. The project was supposed to link author data with Amazon product listings. Not only is it unclear whether the project was ever completed, but its wildly out-of-control budget went unnoticed for about five months, according to a paywalled Financial Times report.

The internal presentation cited in the FT’s report allegedly identified additional Claude costs, including $541,000 for financial auditing tools and $134,000 for logistics work, intended to speed up deliveries. Together, these bills clearly demonstrate weak spending controls around internal AI use.

Amazon has reportedly told its staff that it will begin monitoring and limiting its AI spending more closely. One part of the presentation was defensive, however, saying that “Cherry-picking small, isolated examples where teams are learning from one another and portraying them as business as usual doesn’t reflect how teams across Amazon are using AI.”

Amazon’s AI spending puts into perspective previous reports that AI-driven layoffs are apparently costing firms rather than saving them money.

Undoubtedly, AI tools can reduce labor for some tasks, but they can also lead to large bills when teams lack clear token limits and don’t review their work along the way.