OpenAI CEO Sam Altman has articulated a vision of artificial intelligence as a utility, with customers purchasing it through metered usage. This concept has prompted industry observers to consider whether AI tokens—the fundamental units of measurement for AI model operations—could emerge as a defining metric for what some are calling the AI age, comparable to how kilowatt-hours measure electricity consumption.
Tokens represent the basic chunks of text and data that AI models process. Every time a user inputs a prompt or an AI system generates output, token consumption is tracked and measured. While consumer-facing AI companies typically offer flat-rate subscription models, they increasingly charge businesses and developers based on token usage. This shift has created financial pressure on enterprises, leading some to adopt cost-management strategies dubbed “tokenminimizing” after an earlier period of unrestricted AI experimentation called “tokenmaxxing.”
Economists Nicola Borri, Aleh Tsyvinski, and Yukun Liu have published a working paper analyzing 380 trillion AI tokens processed between January 2024 and April 2026 through OpenRouter, a platform aggregating access to multiple AI models. They tracked how overall AI consumption correlates with stock market performance across different industries. Their analysis created a metric they termed the “AI Factor,” measuring weekly growth in tokens, spending, and active users.
The researchers discovered that companies whose stock prices moved most strongly with increases in AI consumption—exhibiting what they called an “AI premium”—subsequently earned significantly higher returns, outperforming companies perceived as less benefited by roughly 0.64 percentage points weekly. Notably, this premium extends beyond technology companies to include airlines, utilities, manufacturers, retailers, financial institutions, and waste management firms. The effect proved strongest among U.S. and European companies and when analyzing frontier AI models.
The study’s broader significance may lie in establishing a new research methodology. By utilizing token data as a digital paper trail, economists can potentially track AI adoption and economic impact with unprecedented precision in real time, offering advantages over traditional methods like surveys and earnings calls. However, the analysis carries important limitations: it relies on unreviewed research using OpenRouter data that likely skews toward sophisticated, cost-conscious users rather than average consumers, and stock market movements may not accurately predict actual technological and economic outcomes.
Article Attribution | Read More at Article Source
Article summary produced by Claude AI