Instagram head Adam Mosseri said in a recent interview that there will come a time in the future, probably within a year or two, when it will be necessary to put limits on Meta employees’ spending on AI tokens.
“I think you can imagine that in at least a year or two, the burn rate for a good engineer might be the same as their salary or their payroll costs. And in that world, you probably need to put some caps on them,” the Mehta executive said on Lenny’s podcast.
AI token spending, or the cost of processing AI prompts and responses, has been a hot topic in recent days. Meta has shut down its internal AI token spending leaderboard after AI costs could help the company reach its multi-billion dollar goal in 2026.
Meta is not alone in rethinking its approach to AI experimentation. Uber was also calculating AI after exhausting its 2026 AI coding budget by April. Due to rising token costs, Microsoft canceled Claude Code licenses and instead consolidated its engineers around its Copilot CLI tool.
Mosseri believes that the cost of AI tokens needs to be managed like any other resource, likening it to the day-to-day costs of running a company, such as salaries and operating expenses (OpEx).
“I look at it like any other resource,” Mosseri said. “You have a limited number of GPUs, CPUs, storage, RAM, etc., so you have to decide how to deploy capacity to different teams. You have to decide how to deploy OpEx to label the budget across teams. You have to decide how to deploy payroll according to headcount across teams.”
He added that the token budget will be the same, noting that the per-engineer cap needs to be proportionate to the company’s confidence in being able to use the budget in an “ROI-positive” way.
Mosseri said Meta currently does not have a token cap for any employee, but believes token usage could become healthy in the future. Furthermore, in the future, he expects the cost of the token to fall as AI model makers enter into a price war to entice people to use their tools over competitors.
So far, Mosseri noted that the company has managed to keep token costs down a bit by shutting down some “stupid things” it was doing, such as token spending leaderboards.
“It’s not that difficult to build a token incinerator, but it doesn’t create much value,” he said.
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
