Amazon Slows Down AI: 'Tokenmaxxing' is Costly for Tech Giants

Amazon recently made the decision to shut down its internal ranking system, known as Kirorank, in response to excessive and costly use of artificial intelligence. This leaderboard, created by employees to encourage the use of AI, was abandoned after 'tokenmaxxing' practices were identified. This practice involved using AI for mundane tasks to artificially inflate token usage, leading to significant costs for Amazon.

An Amazon spokesperson clarified that Kirorank was merely an experimental tool and not an official one. The company now aims to focus on a more judicious adoption of AI, emphasizing innovation and operational efficiency.

A leaked employee memo from Dave Treadwell, Amazon's senior vice president, had already urged employees to stop using AI just for the sake of using AI. Amazon claims to have measured token usage to understand costs and efficiency but has discouraged the use of these metrics to evaluate developer productivity.

Amazon is not alone in this approach. Meta also shut down a similar leaderboard in April, and Uber has expressed similar concerns regarding AI costs. Other companies like Microsoft, Salesforce, and DoorDash have also begun to limit their use of AI due to rising costs and disappointing results.

However, the use of generative AI continues to grow, with Google Gemini reaching 3.2 quadrillion tokens per month in May 2026. This increase is partly due to the rising popularity of coding tools and agentic AIs.

Jackie Rees Ulmer from Ohio University believes that while the reduction of AI usage by some companies may not be enough to burst the generative AI bubble, it highlights the importance of targeting applications that deliver real value. Companies are still trying to understand how to use AI effectively while strengthening essential human skills.