Could AI Tokens Become the ‘Kilowatt-Hour’ of the AI Age?

A growing number of economists are proposing that AI tokens — the fundamental units of data that artificial intelligence models read and generate — could become the standard unit of measurement for tracking AI’s spread through the economy, much like kilowatt-hours measure electricity consumption. The concept, detailed in a new working paper and reported by NPR’s Planet Money, would give researchers and policymakers a powerful new tool for quantifying AI usage, productivity gains, and economic impact as the technology becomes deeply embedded in everyday business operations.

A Vision of Intelligence as a Utility

At the heart of the token-as-measurement concept is a bold vision articulated by OpenAI CEO Sam Altman. Speaking at BlackRock’s U.S. Infrastructure Summit in March 2026, Altman declared: “We see a future where intelligence is a utility, like electricity or water, and people buy it from us on a meter.” He invoked the phrase “too cheap to meter,” borrowed from the nuclear energy industry, suggesting that AI access could one day be nearly as ubiquitous and affordable as electricity.

This vision is already taking shape in the business world. While consumer AI products like ChatGPT, Claude, and Gemini still largely offer flat-rate subscriptions, companies and developers are increasingly charged based on the number of tokens they consume. Every interaction — reading a prompt, generating text, completing a multi-step task — is metered, creating a natural measurement mechanism similar to how utilities track kilowatt-hours or gallons of water.

The NBER Working Paper: Measuring the AI Premium

A newly published working paper from the National Bureau of Economic Research (NBER Working Paper No. 35451, July 2026) offers a groundbreaking look at how token data can be used to understand AI’s economic impact. Authored by Nicola Borri of Luiss University, Aleh Tsyvinski of Yale and NBER, and Yukun Liu of the University of Rochester, the study analyzes 380 trillion AI tokens processed through OpenRouter — a platform that routes requests across more than 400 AI models — between January 2024 and April 2026. That dataset represents approximately 2% of monthly global AI consumption.

The researchers construct what they call an “AI Factor” from weekly growth in tokens, dollars, and active users, then estimate firm-level “AI Betas” by measuring how individual stock returns move with changes in overall AI consumption. Their key finding: companies whose stock prices were most sensitive to increases in AI usage earned significantly higher returns. A value-weighted long-short strategy of high AI beta versus low AI beta firms delivered approximately 64.1 basis points (0.641%) per week — what the paper terms the “AI Premium.”

“The story of AI is no longer just a Silicon Valley story,” Tsyvinski told NPR. “Financial markets already see Main Street being impacted.”

AI’s Reach Extends Well Beyond Tech

Perhaps the most striking finding is that the AI premium is not confined to technology stocks. The researchers identified significant AI exposure across airlines, cruise lines, utilities, industrial manufacturers, retailers, banks, and even waste management companies — suggesting Wall Street expects AI to reshape virtually every sector of the economy. The effect is strongest for companies in the United States and Europe, and significantly weaker in China and other emerging markets.

The paper also identifies the S&P 500 companies with the highest AI premiums: AppLovin, Carvana, Lumentum, Expand Energy, and Baker Hughes. At the other end, the companies whose stock prices appear most negatively exposed to AI growth include Moderna, Estée Lauder Companies, ON Semiconductor, Skyworks Solutions, and Aptiv.

The Rise and Fall of Tokenmaxxing

The token economy has already generated its own cultural phenomena. In early 2026, tech workers at companies including Meta, Amazon, and Uber engaged in what became known as “tokenmaxxing” — using AI for unnecessary tasks to climb internal leaderboards and earn titles like “Token Legend.” But the party didn’t last. As Business Insider reported, companies soon began slamming the brakes. Amazon closed an internal dashboard tracking AI usage after employees gamed the system. Uber capped staff use of AI coding tools after blowing its AI budget. Meta implemented cost controls, ushering in an era of what industry observers dubbed “tokenminimizing.”

By mid-2026, the shift from usage maximization to cost efficiency was in full swing. The rise of agentic AI — autonomous models that execute complex, multi-step tasks — is accelerating this trend. By 2026, agentic models accounted for more than half of all tokens consumed on OpenRouter, up from a small share in 2024, intensifying both the promise and the cost pressures of the token economy.

Why AI Tokens Matter as an Economic Metric

Beyond the financial findings, the paper’s most enduring contribution may be methodological. AI tokens leave behind a digital paper trail that allows economists to track AI adoption in near real-time — a level of granularity that was simply unavailable during previous technological revolutions like the advent of electricity or the internet.

If token-based metrics become widely adopted, they could enable:

• Real-time tracking of AI adoption across industries and geographies
• Granular productivity measurement connecting specific AI usage to economic output
• Cross-industry comparisons of AI integration
• New financial instruments such as AI token futures or consumption-tracking ETFs

Caveats and Open Questions

The research comes with important caveats. The NBER paper is a working paper that has not yet been peer-reviewed. OpenRouter’s data likely overrepresents sophisticated, cost-conscious heavy users rather than average consumers. And stock prices reflect investor expectations, which can be spectacularly wrong — as the long history of financial bubbles attests. The researchers themselves caution that the AI premium may already be priced into markets, and that buying or selling based on these findings is not a recommended financial strategy.

Still, as Altman’s vision of metered intelligence moves closer to reality, the questions multiply: Will token-based pricing become the dominant model for consumer AI, or remain primarily for enterprise customers? How will regulators approach AI tokens — as a utility commodity, a financial instrument, or something entirely new? Can the data sources be expanded to provide a more representative picture of global AI usage?

What to Watch For

The tokenization of AI consumption is still in its early stages, but the trajectory is clear. As agentic AI grows and usage-based pricing becomes more common, the number of tokens flowing through the economy will only increase. Whether AI tokens ultimately become as familiar a metric as the kilowatt-hour depends on how quickly researchers, businesses, and policymakers embrace them — but the groundwork is being laid now.

One thing is certain: the era of AI measurement has begun.