It started as a workplace trend. Then it became a competition. Now it has turned into a financial headache that some of the world’s biggest technology companies are scrambling to control.
The phenomenon is called token maxxing. The name comes from the way computing tools charge for their services. Every time a person uses one of these tools to write something, answer a question or carry out a task, the system processes a certain amount of data. That data is measured in units called tokens. A token is roughly three quarters of a word. Sending a question and receiving an answer might use a few hundred tokens. Running a complex, multi-step task can burn through millions.
Companies pay for tokens. The more their employees use, the larger the bill. For most of last year and into early 2026, many large technology companies actively encouraged their staff to use as many tokens as possible. The reasoning seemed logical at the time. The more people used these tools, the faster they would learn how to use them well. The more skilled workers became, the more productive they would be. Token usage became a measure of how seriously employees were taking the technology.
Some companies began tracking usage internally. Leaderboards appeared. At Meta, an engineer built an internal ranking system called Claudeonomics that showed which of the company’s 85,000 participating employees was using the most computing credits. The top user in a single month may have cost the company more than $1.4 million in charges alone. Amazon created its own internal leaderboards. Managers began using token consumption figures when assessing how engaged their teams were with the technology.
Nvidia’s chief executive, Jensen Huang, added fuel to the fire. He said publicly that an engineer earning $500,000 a year should be spending at least $250,000 of their own productivity in computing credits annually. Anything less, he suggested, would leave him deeply alarmed about how well the technology was actually being put to use.
The results were predictable in hindsight. Staff at some companies began using the tools not because they had genuine tasks to complete but because they wanted to maintain high usage numbers. At Amazon, some employees reportedly ran automated tasks that served no real business purpose. The goal was not to get work done. The goal was to appear active.
The bills began arriving. Uber was among the first large companies to speak publicly about the damage. Its chief technology officer said in April that the company had already spent its entire computing budget for the year 2026. The full year was gone in four months. The company’s chief operating officer later described the situation in plain terms. “All of a sudden you get the bill and ask, why are we spending all this money? What are we even doing with it?” he said. One unnamed company, according to a report by the American news site Axios, burned through half a billion dollars worth of computing credits in a single month after failing to set any limits on staff usage.
As reported, the response from companies has been swift. Meta removed the Claudeonomics leaderboard. Amazon scrapped its internal rankings. A senior manager told staff directly not to use computing tools for the sake of using them. Microsoft cancelled subscriptions to a popular coding tool for employees in several of its product divisions. Walmart and Starbucks both pulled back plans to expand their use of automated task management systems. The era of encouraging limitless usage was quietly brought to an end.
Critics had been warning about this for some time. One technology expert compared token maxxing to an environment from twenty years ago when software engineers were assessed by how many lines of code they wrote. Quantity is not the same as quality. A developer who writes a thousand lines of unnecessary code is not more valuable than one who solves a problem in ten. The same logic applies to computing usage.
See Also: Elon Musk’s X money plan has just taken its biggest step yet toward becoming a bank
A chief operating officer at a software firm described the trap clearly. Ranking engineers by how much they spend on computing tools is like ranking a marketing team by how much money they have spent. Spending is not achievement.
The shift has consequences beyond the companies directly involved. The providers of these computing services had been growing rapidly, in part because large employers were telling their staff to use as many credits as possible. If those same employers now restrict usage, the growth figures that those providers have been reporting to investors may slow. Both OpenAI and Anthropic are preparing to list their shares on public markets. Their valuations depend on continued rapid growth. A pullback in corporate usage arrives at an uncomfortable moment.
The new direction has a name too. Companies are now talking about efficiency maxxing instead. The goal is to get the most useful output for the fewest credits spent. It is a more sensible approach. It is also a sign that the initial excitement has given way to the harder question of what this technology is actually worth paying for.
