Tokens are the units that AI models use to process a prompt or generate a response. Every time an engineer asks an AI tool for help writing code or analyzing data, it burns tokens, and those tokens cost money.
Speaking on Lenny's Podcast, Mosseri said: "I think that you can imagine, at least in a year or two … that the burn rate of a strong engineer might be the same as their salary, or their cost of employment. And in that world, you're going to probably need to put in some caps."
Meta's internal AI spending had already been rising. The company shut down an internal leaderboard that tracked token use after spending projections indicated Meta could face billions in costs by 2026. Mosseri noted that the company had managed to rein in costs by stopping "silly things" like that leaderboard. "It's not that hard to build a token incinerator, and that doesn't create a lot of value," he said.
Get the market news that matters in a five-minute read with Market Briefs, our free daily newsletter
Mosseri compared token budgets to other resources a company manages. "I think of it like any other resource," he said.
"I have to decide how to deploy capacity to my different teams because I have a limited number of GPUs and CPUs and storage and RAM etc. I have to decide how to deploy OpEx for labeling budgets across my teams. I have to decide how to deploy payroll for headcount across my teams." Token budgets, he added, would be the same, with limits based on how much the company believes an engineer can spend productively toward positive returns. Mosseri noted that Meta has not yet enforced any token limits on its staff, but he thinks such caps might be beneficial down the line.
Other companies have also faced AI cost overruns. Uber exhausted its planned AI coding spending for 2026 as early as April, according to the TechCrunch report. Microsoft dropped its Claude Code subscriptions from Anthropic and moved its engineers to focus solely on its own Copilot CLI tool.
Mosseri also predicted that token prices will fall over time as companies creating AI models compete on price to draw in users.
The rising cost of AI tokens is not unique to Meta. As more companies integrate large language models/) into their daily operations, the expense of running inference at scale has become a significant line item. Engineers, who rely on AI for code generation, debugging, and data analysis, can easily consume thousands of tokens per session.
Without proper oversight, these costs can spiral. Mosseri's proposal to treat token budgets like any other resource - such as GPU capacity or headcount - signals a shift toward treating AI as a resource to be managed similarly to payroll or operating expenses. This approach could become standard across the tech industry as firms grapple with the financial realities of AI adoption.
Join Market Briefs, our free daily newsletter, for a quick daily rundown of the markets
