Sam Altman wants to sell you intelligence the way your utility company sells you electricity, and a new economic study tracking 380 trillion AI tokens suggests the market already believes him. The AI economy is no longer just a Silicon Valley story, and Wall Street has been quietly placing its bets for a while now. The rest of us are just catching up.
Tokens: The Meter Running in the Background of Everything
Here's the thing about AI tokens: most people have never heard of them, and they're already everywhere. As NPR's Planet Money reports, tokens are the tiny chunks of text and data that AI models read and generate every time you type a prompt or an AI completes a task. Think of them as the atomic unit of AI work. The more work the model does, the more tokens it burns through.
Right now, average consumers mostly deal with flat-rate subscriptions. You pay your $20 a month to ChatGPT, you don't think about it. But businesses and developers are being charged based on exactly how many tokens they use, and those bills have been getting eye-watering. Planet Money notes that after a period of what some in the industry dubbed 'tokenmaxxing,' companies like Uber and Amazon started slapping guardrails on AI use to stop the bleeding. There is apparently now something called 'tokenminimizing,' which is exactly what it sounds like, and the fact that we have named both of these things tells you everything about where we are in the AI adoption curve.
380 Trillion Tokens Walk Into an Economics Paper
A group of economists decided to do something clever with all this token data. According to NPR, Nicola Borri, Aleh Tsyvinski, and Yukun Liu published a working paper analyzing 380 trillion AI tokens processed through OpenRouter between January 2024 and April 2026. OpenRouter, for the uninitiated, is essentially a one-stop shop where developers can access hundreds of AI models from OpenAI, Anthropic, Google, and others through a single interface, rather than signing individual contracts with each company. That represents roughly two percent of monthly global AI usage.
The economists took that token data, combined it with weekly growth in spending and active users into a single broad measure they call the 'AI Factor,' and then asked a genuinely interesting question: which companies' stock prices move most strongly when overall AI consumption goes up? The answer to that question is, it turns out, considerably more interesting than you might expect.
The 'AI Premium' Is Real, and It's Not Only About Nvidia
Here's where it gets good. NPR reports that companies whose stock prices were most sensitive to increases in overall AI consumption subsequently outperformed the least AI-correlated companies by about 0.64 percentage points per week. The researchers call this the 'AI Premium.' Over time, that gap compounds into something significant, and it represents what the market collectively believes about AI's future impact.
The more surprising finding is that this premium extends well beyond the obvious tech plays. Researcher Aleh Tsyvinski told NPR directly: 'The story of AI is no longer just a Silicon Valley story. Financial markets already see Main Street being impacted.' That is a significant statement. It means Wall Street isn't just betting on the chip makers and the model builders. It's pricing AI's effects into a wide range of companies and industries that most people wouldn't immediately associate with the technology. Which companies, exactly? The paper doesn't name names at the individual stock level in the way Planet Money's coverage presents it, but the directional finding is clear: AI exposure is now a factor across the broader economy, not just the tech sector.
Why Tokens as Data Matter More Than Tokens as Currency
The economists are upfront that stock markets can be catastrophically wrong. They invoke the long history of financial bubbles themselves, which is a refreshing level of intellectual honesty from people publishing a paper about stock market signals. But the real prize in this research isn't necessarily the investment thesis. It's what the methodology makes possible going forward.
As NPR explains, economists have historically had to rely on surveys, earnings calls, and company announcements to study how new technologies spread through an economy. Those methods are slow, self-reported, and easy to game. Token data is none of those things. It's granular, it's near real-time, and it's a direct behavioral signal rather than what some PR team decided to say on an investor call. Tsyvinski and his co-authors are essentially showing that AI tokens could become the data infrastructure for studying the AI economy itself, the same way kilowatt-hour data tells us how electricity consumption tracks with industrial output. If that research pipeline develops, we will eventually know far more precisely which industries AI is actually transforming, and which ones are just talking about it in press releases.
Altman's Utility Vision and What It Actually Means
Sam Altman's quote, cited by NPR at the top of their piece, deserves more scrutiny than it usually gets. 'We see a future where intelligence is a utility, like electricity or water, and people buy it from us on a meter.' Read that slowly. He is describing a world where the ability to think through complex problems, generate content, write code, and make decisions gets metered and billed by a private company. Your monthly invoice will include AI tokens the way it currently includes kilowatt-hours.
The electricity comparison is instructive in ways Altman probably doesn't intend. Electricity became a utility in the 20th century because it was too essential to leave entirely to unregulated markets. We built regulatory frameworks, public utilities, rural electrification programs. We decided, as a society, that access to electricity was not something that should be purely subject to whatever pricing a private company felt like charging. The AI industry is currently operating as if none of those historical lessons apply to them. Maybe they're right. But 380 trillion tokens into the AI age is probably a good time to start asking the question.
The Dingo Take
Look, the research here is genuinely interesting and the economists are doing real work. Following the token trail to study AI's economic effects is a smarter methodology than asking companies to self-report during earnings calls, and the finding that Wall Street sees Main Street exposure to AI is worth paying attention to. None of that is in dispute.
What deserves a harder look is the frame around all of it. The 'AI as utility' vision is presented in Planet Money's piece with a kind of cheerful inevitability, as if the only question is when it happens rather than what kind of utility we're talking about, who controls it, who regulates it, and who gets left out. Token pricing is already creating situations where companies like Uber and Amazon are rationing AI use because the costs are spiraling. What happens to smaller businesses, or to individuals who can't afford the premium tier, when intelligence itself is metered?
A kilowatt-hour of electricity costs roughly the same whether you're a billionaire in Manhattan or a teacher in rural Ohio because we built systems to make that true. Nobody has built those systems for AI tokens yet, and the industry titans currently designing this new economy have shown approximately zero interest in doing so. The research is pointing us toward a future we should be building deliberately. Right now we're mostly just watching it happen.
