Meta’s Adam Mosseri says AI token budgets could soon be capped per engineer

In a recent conversation, Adam Mosseri, head of Instagram, shared insights about the potential future of AI token spending within Meta. He indicated that within the next year or two, the company may need to impose limits on the AI token expenditures of its engineers. Mosseri highlighted a concerning trend where the costs associated with AI token usage could match the salaries of top engineers. "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," he stated during an appearance on Lenny’s Podcast. This surge in AI token costs has become a hot topic, particularly as Meta recently halted an internal leaderboard tracking AI token spending due to projections that could see expenses reach billions by 2026. Meta's situation is not unique, as other tech giants like Uber have also faced challenges with their AI budgets, exhausting their resources months ahead of schedule. Microsoft has taken a different route, opting to cancel certain licenses in response to rising token costs, which led to a consolidation of its engineering efforts around its Copilot CLI tool. Mosseri explained that managing AI token expenses will need to be treated like any other resource in the company. He drew parallels to payroll and operating expenses, emphasizing the need for strategic allocation of resources. "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.," he elaborated. He believes that future token budgets will need to be proportional to the company's trust in an engineer's ability to generate a positive return on investment with their spending. Currently, Meta does not enforce spending caps on its employees, but Mosseri sees potential benefits in introducing such measures. He also anticipates that as competition among AI model developers heats up, token costs may decrease over time. For now, the company has managed to control its expenses by eliminating less valuable initiatives, such as the token spend leaderboard, which Mosseri described as "silly." "It’s not that hard to build a token incinerator, and that doesn’t create a lot of value," he concluded, indicating a need for more efficient management of AI resources moving forward.