Amazon Insiders Horrified at “Catastrophically Expensive” Internal AI Usage

Amazon is facing a financial reckoning as employees discover the true cost of utilizing AI tools, according to reports from The Financial Times. The tech giant's internal AI experiments have led to "catastrophically expensive" cost overruns, with some instances costing millions of dollars. In one instance, Amazon spent $1.8 million on Anthropic's Claude Sonnet AI tool, which was tasked with a simple job - matching author details with product listings on the ecommerce site.

The project exceeded its allocated budget by a staggering 860 percent and took five months to be detected.

Amazon insiders warn that such incidents are not isolated and that coding mistakes that were once "trivially cheap" have now become "catastrophically expensive." The company also experienced unexpected costs related to building financial auditing tools, with $541,000 wasted, and an additional $134,000 spent on improving delivery speeds across its logistics network.

This spending spree is part of a broader issue within the business world as companies grapple with the costs of deploying AI tools. The "tokenmaxxing" era, which encouraged widespread use of AI agents, has led to detrimental consequences such as one company reportedly burning through $500 million in Claude usage fees in a single month. To address the spending concerns, AI companies are transitioning to token-based billing instead of flat subscription rates.

Amazon's own internal reckoning began with an internal leaderboard ranking employees based on their AI tool usage, which was eventually shut down. While Amazon claims to be running a tight ship, a spokesperson stated that isolated examples of AI usage do not reflect the overall company's approach.