Amazon has reportedly seen its AI costs skyrocket, exceeding its budget by a staggering 860%. Amazon counters that it's a small case and the 200 million yen figure is negligible.
Sources say that at an employee meeting held at Amazon at the end of July 2026, employees were told that 'unplanned expenses' had occurred as a result of attempting to switch their work from traditional programming to the use of AI models. According to sources, despite introducing AI to cut costs, the implementation failed, and AI costs exceeded the project budget by 860%.
Amazon finds cases of AI causing runaway spending on tech projects
Amazon accidentally spent $1.8 million using Claude for menial coding task, went 860% over budget — 'catastrophically expensive' coding blunders discovered in internal Amazon AI usage metrics | Tom's Hardware
https://www.tomshardware.com/tech-industry/artificial-intelligence/amazon-accidentally-spent-usd1-8-million-using-claude-for-menial-coding-task-went-860-percent-over-budget-catastrophically-expensive-coding-blunders-discovered-in-internal-amazon-ai-usage-metrics
According to sources, at an Amazon staff meeting held at the end of July 2026, employees were given a presentation about a case study in which Amazon used Anthropic's AI, Claude Sonnet. In that project, $1.8 million (approximately 289 million yen) was spent to use Claude Sonnet to match product listings with author information on the group's e-commerce website, but the implementation ultimately failed.
Sources say the expenditure exceeded the project budget by 860%, and it took five months to discover. While it hasn't been revealed exactly what processes led to the increased costs, employees have been told that coding errors that would have resulted in minimal costs under the old system now incur 'catastrophic costs' due to the dramatic increase in token spending once the team began using AI models to complete tasks.
In another instance, Amazon incurred unexpected expenses of approximately $541,000 (approximately 86 million yen) in connection with the development of financial auditing tools. Furthermore, in a separate project using AI to improve the delivery speed of its logistics network, an unexpected expenditure of $134,000 (approximately 21 million yen) occurred, and it took more than two weeks for the overspending to be discovered. To prevent such situations, Amazon says it is working on building 'guardrails' to automatically monitor and limit the amount of AI used.
Furthermore, regarding its internal AI platform 'MeshClaw,' Amazon set a goal that 'more than 80% of developers use the AI weekly,' and tracked token consumption through internal rankings. As a result, it has been reported in the past that some employees inflated their token consumption by having the AI perform even low-priority tasks in order to make a good impression on management.
Amazon employees were found to be inflating their AI usage to impress their superiors and improve their rankings - GIGAZINE
In an internal presentation, Amazon explained, 'As teams learn from each other's experiences through trial and error, small-scale, individual cases will inevitably arise. To highlight only a few of these cases and portray them as representative of the general use of AI at Amazon does not accurately reflect how AI is being used across the entire company.' It has also been pointed out that the company spent more than $1.8 million on failed AI projects, but these excessive AI expenses amount to less than 0.1% of its monthly revenue.
Related Posts:
in AI, Posted by log1e_dh