AI Tokens: The New Measure of AI’s Economic Impact

Earlier this year, OpenAI CEO Sam Altman stated, “We see a future where intelligence is a utility, like electricity or water, and people buy it from us on a meter.” This idea reflects a significant shift in how AI might be integrated into everyday life. If this future materializes, “AI tokens” might transition from niche tech discussions to a common aspect of daily operations. AI tokens could become a defining measure in a new industrial era, akin to kilowatt-hours in electricity, accounting for AI usage.

AI tokens represent the units of measurement for tasks completed by AI models. Each prompt read, each outcome generated, and each function executed by AI models is translated into tokens. The complexity of tasks demands more tokens, thus impacting pricing. While AI companies typically offer flat-rate subscriptions for consumers, they’re increasingly billing businesses and developers based on token usage.

As companies intensify their AI deployment, they’ve noticed the escalating costs associated with token-based pricing. The tech sector witnessed periods of unrestricted AI use, termed “tokenmaxxing,” leading firms like Uber and Amazon to implement limits to curb expenses, an approach referred to as “tokenminimizing.” These tokens, beyond usage tracking, provide a digital footprint for economists and researchers studying AI’s economic effects.

A working paper by Nicola Borri, Aleh Tsyvinski, and Yukun Liu employs data from 380 trillion AI tokens to examine AI’s evolving impact on financial markets. Their inquiry revolves around assessing which companies’ stock prices correlate with the fluctuations in the overall AI consumption.

By analyzing broad AI usage trends, rather than tracking individual company consumption, the economists apply classic finance theory, suggesting stock prices reflect investor anticipations of AI’s future impact. Although predictions may falter — evidenced by historical financial bubbles — stock prices offer insight into market expectations of AI’s role in the economy.

They discovered that as AI usage increased, financial markets reacted differently to various companies. Those seen as major beneficiaries received a higher “AI premium,” transcending traditional tech stocks. Industries including airlines, utilities, manufacturers, retailers, and banks are perceived as potential AI beneficiaries.

The study utilizes data from OpenRouter, which aggregates AI model access and has become a key tool for managing token costs. This platform provides rich, anonymized data on AI token usage, representing approximately 2% of global AI use from January 2024 to April 2026. The researchers merged growth, spending, and user statistics into a comprehensive measure termed the “AI Factor.” Their findings highlight stock sensitivity to AI consumption changes, revealing varied impacts across industries.

Despite the intriguing insights, the study carries various caveats. Being a working paper, it awaits peer review. OpenRouter’s data might portray a skewed picture of AI usage due to its predominantly sophisticated user base. Also, even if investor expectations align with AI potential, such optimism might already be included in present stock prices.

Although not financial advice, the study heralds a new avenue for tracking AI’s economic footprint using AI tokens. It invites further exploration into AI measurement possibilities, emphasizing the importance of robust data in addressing AI’s economic queries.