Amazon employees are now joining the ranks of those “tokenmaxxing” at their boss’ request, the Financial Times reported Tuesday. Only these Amazon employees are more resistant—they’ve reportedly been running the company’s internal AI tool on trivial tasks to inflate their token counts and climb the leaderboard measuring their usage.
“Tokenmaxxing” is a burgeoning trend at the hyperscalers where employers are rewarding employees for using AI the most, quantified by using tokens. While it isn’t clear that the usage determines much more than brownie points at Amazon, similar behavior was reported other big hyperscalers, like Microsoft and Meta. Notably, all three of these companies are heavily invested in the very tech that they’re encouraging their employees to use. Amazon even reported in their recent earnings that Anthropic’s increased valuation made up nearly half of the company’s profits.
Gil Luria, head of technology research at brokerage D.A. Davidson, said the dynamic concerned him.
“That doesn’t sound very healthy,” Luria told Fortune. “You get the behavior that you create the incentive for. So if you tell people they’ll succeed if they use a resource more, of course they’ll use it more.”
Luria clarified that, for him, there isn’t a question that AI tools are very powerful and have the opportunity to make everyone more productive. But the “hurdle,” so-to-speak, is in diffusion. “Humans are rigid in how they do things,” Luria said. “So if you don’t create an incentive for humans to change their behavior, try something new, most of us won’t.”
The question is how to incentivize that change without producing gaming, a problem formalized in Goodhart’s Law: “when a measure becomes a target, it ceases to be a good measure.” While Amazon evidently told employees that their “tokenmaxxing” would not be a factor in their performance reviews, multiple employees told the FT that they worried managers watched it anyway. One said there was “so much pressure” to use the tools, and at that, the most.
This trend doesn’t seem to be confined at just Amazon. At Meta, an employee built an internal leaderboard called “Claudeonomics” that ranked the company’s roughly 85,000 workers by token consumption. In a 30-day window, total usage on the dashboard exceeded 60 trillion tokens, though neither CEO Mark Zuckerberg nor CTO Andrew Bosworth ranked in the top 250.
The dashboard was taken down after The Information’s reporting, but Meta CTO Andrew Boswort has publicly endorsed the underlying logic. He said his best engineer was spending the equivalent of his salary in tokens but, as a result, was “5x to 10x more productive.”
“It’s like, this is easy money,” Bosworth told Forbes. “Keep doing it. No limit.”
The stakes for the hyperscalers are huge. Combined 2026 capital expenditure from Amazon, Microsoft, Alphabet, and Meta is already pushing $700 billion, with some Wall Street projections exceeding $1 trillion for 2027, up significantly from just under $400 billion in 2025. The companies are telling investors that their inference chips are consumed as fast they are deployed, while also engaging in what Luria called “circular activity”: the same companies invest in their suppliers and customers. He added that dynamic was “part of the overhang around all of the large technology companies, especially Amazon, Google, Microsoft, Meta, Nvidia.”
Demand for AI is the highest it’s ever been. OpenAI and Anthropic are at a combined run rateof more than $70 billion, he noted, up from roughly zero two years ago. “Those companies actually represent real economic activity,” he said. “That is consumers and businesses paying for access to their model.” He didn’t believe the hyperscalers themselves were a disproportionate source of that revenue, any more than they’re full of programmers, and programmers are using the software. “But that’s true for any company.”
Amazon did not immediately respond to Fortune’s request for comment.
The post ‘That doesn’t sound very healthy’: Amazon’s reported tokenmaxxing might gamify AI usage, analyst warns appeared first on Fortune.
