Amazon has faced significant financial overruns when integrating artificial intelligence models into its own workflows. Small code errors that were previously cheap in traditional systems now lead to unplanned losses of millions of dollars due to the specifics of AI service pricing.
Amazon has faced significant financial overruns when integrating artificial intelligence models into its own workflows. Small code errors that were previously cheap in traditional systems now lead to unplanned losses of millions of dollars due to the specifics of AI service pricing.
This was reported by the Financial Times, citing an internal meeting of the company's employees, where senior engineers discussed budget control issues.
According to the publication's sources, attempts to transfer routine tasks from conventional programming to the use of AI models have led to a significant overestimation of costs, which the developers themselves called "catastrophically expensive" at the presentation. Engineers are now urgently working on creating automated fuses for future projects.
One of the most notable cases occurred while trying to automate the mapping of author data to listings on an e-commerce platform using Anthropic’s Claude Sonnet model. Although the project failed to roll out, it cost the company $1.8 million. Cost overruns were 860% of the original budget, and it took management five months to uncover the financial hole.
Other examples discussed at the meeting included unexpected costs of about $541,000 during the development of financial audit tools, as well as a loss of $134,000 in a logistics AI project. The latter mistake was only noticed by developers two weeks later.
“It’s very difficult to understand how much anything related to artificial intelligence costs,” one senior Amazon employee admitted to the publication.
Official representatives of the corporation, in response to a journalist's request, noted that these cases occurred only in a few teams and do not reflect the overall picture of how the company's staff uses artificial intelligence. Amazon emphasized that they continue to experiment and learn to work with new technology to optimize processes.
Previously, Amazon's internal AI tools caused at least two outages at the company that lasted more than a day.
