Amazon Contends With Unplanned Overspending on AI

July 30, 2026 — PYMNTS summarizes a Financial Times report (Thursday Jul 30) citing multiple sources familiar with the matter.

Amazon senior engineers told colleagues at a staff meeting that switching tasks from conventional programming to AI models caused unplanned spending described as “catastrophically expensive.”

Incidents cited (FT sources):

• $1.8 million on matching author details with eCommerce listings using Anthropic Claude Sonnet; deployment failed; 860% over budget; five months to detect
• ~$541,000 unanticipated costs on financial auditing tools
• (Unite.AI / HeadlinesBriefing also cite ~$134K logistics-speed AI overrun)

Amazon spokesperson statement to PYMNTS:
“As with any new technology, we’re experimenting, learning and improving how we use it, including how we drive cost efficiencies.”
“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.”

Primary FT article is paywalled; PYMNTS, Unite.AI, HeadlinesBriefing, and TradingKey independently relay consistent dollar figures attributed to FT. Treat figures as FT-attributed, not independently audited.