Amazon has encountered significant cost overruns on internal AI projects, the Financial Times reports. One initiative — an automated system to match author data with product listings on the marketplace, using Anthropic’s Claude Sonnet model — cost $1.8 million, which is 860 % above the original budget.
The overspending wasn’t detected until five months later, and the project was never deployed.
According to the publication, this wasn’t an isolated incident. Amazon spent an additional 541,000onafinancialaudittoolandanother134,000 on an AI project aimed at optimizing logistics — both well above expected costs. During an internal meeting, Amazon engineers described such cases as “catastrophically expensive,” and one employee noted that “it’s very difficult to determine the real costs of AI‑driven projects.”
The Financial Times also recalls that Amazon previously shut down its internal KiroRank system, which incentivized employees to use AI agents more actively. Following this, company leadership urged staff not to adopt artificial intelligence “just for the sake of using AI.” Rising expenses have reportedly prompted the company to rethink its approach to deploying these tools.
In response, Amazon stated that the Financial Times report relies on a few isolated examples and does not reflect how most teams are using AI within the company.
About the Author
