Companies Scale Back AI 'Tokenmaxxing' Trend Amid Rising Costs

Many companies are now scrutinizing their artificial intelligence investments, scaling back the previous "tokenmaxxing" trend where the goal was to generate the maximum amount of AI-produced output. This approach has proven costly without a corresponding boost in productivity, prompting a strategic pivot.

"Tokenmaxxing" refers to maximizing the use of tokens, the fundamental units of text processed by AI systems. The cost of these tokens can quickly accumulate, especially with more advanced AI models. Vincent Gusdorf, head of AI analytics at Moody's Ratings, stresses the need for a more disciplined approach and warns that it is too easy to create unnecessary content with AI.

Initially, the trend fueled revenue growth for leading AI developers like OpenAI and Anthropic, with companies like Nvidia and Meta encouraging high usage. However, the reality of high expenses and limited tangible benefits has led to a backlash. Microsoft CEO Satya Nadella has noted that customers may be paying double – both for token consumption and for sharing their proprietary data.

Instead of using the most advanced models for all tasks, companies are now seeking AI "model routing" tools. These systems direct simpler queries to cheaper, more efficient models, while complex tasks are sent to more powerful alternatives. The emergence of cost-effective open-source models, particularly from China, also offers alternatives that challenge the more expensive U.S. offerings.