Meta’s Mosseri sees AI usage caps as token costs climb

Instagram head Adam Mosseri said Meta may eventually set individual limits on how much engineers can spend using AI tools. For investors, the comment points to a growing cost line inside big tech: AI usage can raise productivity, but every prompt also carries a bill.

Mosseri, speaking on Lenny’s Podcast, said he could imagine a future one or two years from now when a highly productive engineer’s AI “burn rate” could match that employee’s salary or total employment cost. In that kind of setup, he said, companies would probably need spending caps.

AI token spend refers to the cost of running prompts and outputs through an AI model. A token is a small unit of text, such as part of a word, that models process when they read a prompt or generate an answer. More usage means more tokens, and more tokens mean more computing expense.

Mosseri framed those costs as another company resource that managers have to allocate. He compared AI token budgets with payroll, graphics processing units, central processing units, storage, memory and labeling budgets. Operating expenditure, often called OpEx, means the everyday costs of running a business.

“I think of it like any other resource,” Mosseri said on the podcast, according to TechCrunch.

Why AI bills are getting more attention

Meta currently has no AI token caps for employees, Mosseri said. He added that caps could become useful later, with limits tied to how much confidence the company has that an engineer can use the budget in a way that creates a positive return on investment.

The issue has already moved beyond theory at large tech companies. TechCrunch reported that Meta shut down an internal leaderboard that tracked AI token spending. The Information reported on Meta’s efforts to curb employee AI use, and MLQ.ai reported that the company’s AI costs were approaching billions of dollars in 2026.

Mosseri said Meta has reduced some spending by ending what he called “silly things,” including the leaderboard. He said it is easy to create what he described as a “token incinerator,” meaning heavy AI usage that consumes money without producing much business value.

Other companies have also tightened AI spending, according to TechCrunch. Uber had to reassess its AI coding budget after using up its 2026 allocation by April. Microsoft canceled Claude Code licenses and moved engineers toward its own Copilot CLI tool, according to The Verge.

The investor angle

The discussion matters for shareholders because AI is becoming both a product strategy and an expense-management problem. Companies are racing to give employees AI tools for coding, design, support and analysis. Those tools may save time, but the costs can rise quickly when thousands of workers use them throughout the day.

Mosseri also said he expects token prices to fall over time if AI model providers compete more aggressively on pricing. That would help customers using the tools, though his comments were a view of what could happen rather than a company forecast.

For now, his message was that AI spending will have to be managed with the same discipline as other operating costs. Meta has not introduced employee token caps, but one of its top executives is already talking about them as a possible part of how large companies control the economics of AI work.

This story draws on original reporting from TechCrunch.