The AI token economy could become one of the most important ways to measure artificial intelligence use, spending and economic influence.
Every time an AI system reads a prompt, generates an answer or completes a task, it processes units known as tokens. These units work quietly in the background, but they are increasingly shaping how companies charge for AI services and how businesses control their technology expenses.
Researchers are now exploring another possible use for tokens: measuring how artificial intelligence is spreading through the global economy.
A new working paper based on 380 trillion tokens suggests that AI usage data could help economists track which companies and industries investors expect to benefit from the technology.
The findings are preliminary and have not been peer-reviewed. However, they offer an early look at how token data could become a powerful economic indicator.
What Are AI Tokens?
AI tokens are small pieces of information that artificial intelligence models process when they receive instructions or produce responses.
A token may represent a complete word, part of a word, punctuation or another type of data. The number of tokens used depends on the length and complexity of a task.
A short question followed by a brief answer may require relatively few tokens. A request involving a long document, detailed research or complex reasoning can use substantially more.
In general, the more work an AI model performs, the more tokens it processes.
Many consumer AI services still charge flat monthly subscription fees. Businesses and software developers, however, are often billed according to the number of tokens they use.
This makes tokens similar to metered units in other industries. Electricity is measured in kilowatt-hours, mobile data is measured in gigabytes and cloud computing is often priced according to storage or processing capacity.
The AI token economy could eventually develop a comparable standard for measuring and selling access to artificial intelligence.
AI Token Economy Creates New Business Costs
Token-based pricing gives companies flexibility because they only pay for the AI resources they consume.
However, costs can rise quickly when thousands of employees and automated systems use AI throughout the working day.
Many technology companies initially encouraged workers to experiment freely with generative AI. That approach helped organizations discover new uses for the technology, but it also produced unexpectedly large bills.
Businesses are now introducing controls to limit unnecessary AI consumption. Some are directing employees toward cheaper models for routine work while reserving the most powerful systems for complex assignments.
Companies such as Uber and Amazon have reportedly introduced measures intended to control rising token expenses.
The trend suggests that AI cost management could become a major corporate priority. Organizations may need to monitor tokens in the same way they currently track cloud spending, electricity consumption or software subscriptions.
AI Usage Leaves a Valuable Data Trail
Tokens do more than measure how much artificial intelligence costs. They also create a digital record of how frequently AI systems are being used.
That record could help researchers study AI adoption with greater speed and accuracy than traditional methods.
Economists have historically relied on surveys, corporate statements, investment reports and employment statistics to understand the impact of major technologies.
Those sources remain useful, but they can be delayed, incomplete or influenced by what companies choose to disclose.
Token data could provide a more immediate view of AI activity. Researchers may be able to observe changes in consumption almost as they happen and compare those changes with movements in productivity, employment, investment and financial markets.
Studying 380 Trillion AI Tokens
Economists Nicola Borri, Aleh Tsyvinski and Yukun Liu analyzed data covering approximately 380 trillion AI tokens used between January 2024 and April 2026.
The information came from OpenRouter, a platform that allows developers to access hundreds of AI models through a single interface.
Instead of creating separate accounts with companies such as OpenAI, Anthropic or Google, users can compare and access different models through OpenRouter.
This can help developers select lower-cost systems for simple assignments and more advanced models for demanding work.
OpenRouter’s position between users and multiple AI providers has also given it access to extensive information about model usage. The data used by the researchers was anonymized to protect user identities.
The economists estimated that the dataset represented roughly 2% of monthly global AI usage.
Researchers Build an AI Factor
The researchers combined weekly changes in token consumption, spending and active users to create a broad measurement of artificial intelligence activity.
They called this measurement the AI Factor.
The team then examined how the stock prices of different companies responded when overall AI activity increased or decreased.
Their goal was not to determine how many tokens individual businesses consumed. Instead, they studied which stocks moved most strongly alongside broader changes in AI adoption.
Stock prices often reflect what investors expect to happen in the future. By examining those movements, the researchers sought to understand which businesses financial markets believe will benefit from growing AI use.
However, market expectations are not guarantees. Investors can misjudge technologies, overvalue companies and contribute to speculative bubbles.
AI Token Economy Reveals a Market Premium
The researchers found that companies whose shares were most sensitive to increased AI consumption later recorded stronger returns.
They described this performance gap as an AI premium.
According to the paper, companies viewed as the biggest potential beneficiaries of artificial intelligence outperformed those considered the least likely to benefit by approximately 0.64 percentage points per week.
The finding does not prove that AI caused the higher returns. It shows that investors appeared to reward companies whose performance was closely connected to rising artificial intelligence activity.
The market may have been correctly identifying future winners. It may also have been driven by enthusiasm that will not produce the expected business results.
For that reason, the findings should not be treated as an investment strategy.
AI Benefits May Extend Beyond Technology
One of the paper’s most notable findings is that the apparent AI premium was not limited to technology companies.
Investors appeared to expect artificial intelligence to support businesses across a wide range of industries.
These included airlines, cruise operators, retailers, banks, utilities, industrial manufacturers, energy companies and waste-management providers.
The results suggest that markets increasingly view AI as a general-purpose technology rather than a product that benefits only software developers and chipmakers.
Companies outside Silicon Valley may use AI to automate administrative tasks, improve customer service, forecast demand, reduce operational costs and make faster business decisions.
The researchers found that the AI premium was strongest among companies in the United States and Europe. It was considerably weaker in China and other emerging markets.
Stocks were also more responsive to increased use of advanced frontier models than to growth in less capable AI systems.
Companies With the Highest AI Premium
The researchers identified several S&P 500 companies whose stock performance showed strong sensitivity to changes in the AI Factor.
The five companies with the highest measured AI premiums were:
- AppLovin
- Carvana
- Lumentum
- Expand Energy
- Baker Hughes
The list contains businesses from several industries, supporting the argument that investors expect AI to influence much more than the traditional technology sector.
The five companies with the lowest measured AI premiums were:
- Moderna
- Estée Lauder Companies
- ON Semiconductor
- Skyworks Solutions
- Aptiv
These rankings reflect market behavior during the period studied. They do not prove that the companies at the top will succeed because of AI or that those at the bottom will suffer.
Important Limits of the AI Token Study
The research includes several important limitations.
First, it is a working paper and has not yet passed through the peer-review process. Its methods and conclusions could change after further examination.
Second, OpenRouter users may not represent the average AI customer.
The platform is likely to attract experienced developers and heavy AI users who actively compare models and prices. Their behavior may differ from that of consumers who use a single service through a monthly subscription.
The dataset may therefore provide a stronger picture of professional and technical AI use than everyday consumer activity.
The study also measures how stock prices respond to overall AI usage. It does not reveal how many tokens each company uses internally or whether AI has directly improved its revenue, productivity or profits.
Finally, investors may be wrong about which companies will gain from artificial intelligence.
Even when market expectations are reasonable, much of the expected benefit may already be included in a company’s share price.
A New Way to Measure the AI Economy
The most important contribution of the research may be its method rather than its stock-market rankings.
Token data could give economists a new way to observe how artificial intelligence spreads through businesses and industries.
Future studies could compare AI usage with employment levels, productivity growth, business investment and changes in consumer behavior.
Researchers may also be able to examine how companies respond when model prices fall or when more capable systems become available.
Unlike earlier industrial revolutions, the rise of AI is producing detailed digital records of how the technology is used. That could allow economists to study its development with much greater precision.
The AI Token Economy Is Still Emerging
Tokens remain unfamiliar to many ordinary AI users, but their importance is growing.
They determine how much many businesses pay for artificial intelligence, influence which models companies choose and provide valuable evidence about the pace of AI adoption.
As artificial intelligence becomes integrated into more products and workplaces, tokens could become a widely recognized measurement of digital activity.
The AI token economy may eventually help businesses manage costs, allow providers to price their services and give researchers a clearer understanding of how artificial intelligence is reshaping the global economy.
