Meta’s Mosseri signals impending AI token spending caps for engineers

Adam Mosseri, head of Instagram at Meta, has indicated that the social media giant may soon impose strict spending limits on AI tokens for its engineering staff. Speaking on Lenny’s Podcast, Mosseri predicted that within the next year or two, the cost of processing AI prompts and responses for a high-performing engineer could equal their salary or total cost of employment. In such a scenario, he argued, implementing caps would become a necessary operational discipline.

The executive drew parallels between managing AI token budgets and traditional resource allocation, such as payroll or hardware capacity. Mosseri explained that just as the company must decide how to deploy limited GPUs, CPUs, storage, and RAM across teams, it must similarly manage operating expenditure for AI tools. He noted that individual token caps would likely be calibrated based on the company’s trust in an engineer’s ability to utilise the budget in an ROI-positive manner.

This potential policy shift follows significant cost escalations that have already forced Meta to alter its internal practices. The company recently shut down an internal AI token leaderboard, a move Mosseri described as removing "silly things" that drove up costs without creating value. The leaderboard had previously placed Meta on track for billions of dollars in spending by 2026, highlighting the urgency of the current cost management strategy.

Meta’s experience mirrors a broader industry trend as other major technology firms grapple with soaring AI expenses. Uber faced an "AI reckoning" after exceeding its 2026 AI coding budget by April, while Microsoft cancelled Claude Code licenses due to rising token costs, subsequently consolidating its engineers around its own Copilot CLI tool. These moves suggest that unchecked AI experimentation is becoming financially unsustainable for many large organisations.

Despite the push for tighter controls, Mosseri clarified that Meta does not currently have token caps for any employee and believes their introduction is a future necessity rather than a present reality. Looking further ahead, he predicted that token costs would eventually decrease as AI model providers enter a pricing war to attract users. For now, however, the focus remains on curbing wasteful usage and aligning AI spend with broader operational efficiency goals.