Amazon uncovers ‘catastrophically expensive’ AI cost overruns

Amazon Amazon.com, Inc. ↘ 0.10% has uncovered cases of runaway spending on internal AI projects, including one that exceeded its budget by 860% and went undetected for five months, according to a Financial Times report published Thursday.ft

The most expensive incident involved a project built on Anthropic's Claude Sonnet model that was designed to automate the mapping of author data to product listings on Amazon's e-commerce platform. The project ran up $1.8 million in costs before anyone noticed the overrun — and it never shipped. Two additional cases flagged by engineers included roughly $541,000 in unexpected costs on a financial audit tool and $134,000 lost on a logistics AI project meant to improve delivery speeds.aiweekly+2

'Catastrophically Expensive'

Engineers at the company described the pattern as "catastrophically expensive" during an internal review meeting, with one noting it is "very difficult to understand how much anything related to artificial intelligence costs". The warning highlights a structural challenge facing enterprises deploying large language models: unlike traditional software, where the cost of a coding mistake is typically negligible, inefficient code running against generative AI models can trigger massive computing consumption, causing costs to spiral.tradingkey+1

Amazon's Response

Amazon said the cases involved only a few teams and do not represent the company's broader AI usage. The company's senior engineers have since warned internal teams about the risks and moved to impose spending controls.unite+1

The revelations come on the eve of Amazon's quarterly earnings report and amid broader investor scrutiny of the company's AI spending. Amazon has committed hundreds of billions of dollars to AI infrastructure, and the internal cost failures add a new dimension to questions about whether those investments are yielding proportional returns.linkedin

Industry-Wide Concerns

The incidents underscore a growing challenge across the technology sector. As companies rush to embed generative AI into workflows, the absence of robust financial guardrails has created conditions where small errors — a misconfigured prompt loop, an inefficient API call pattern — can generate enormous bills before traditional cost-monitoring systems raise an alert. For Amazon, a company whose cloud division sells cost-management tools to other enterprises, the internal stumbles carry particular irony.aiweekly+1