Meta is evaluating the implementation of limits on AI token expenditures per engineer, as indicated by Instagram head Adam Mosseri. In a recent discussion, Mosseri projected that within the next year or two, the cost associated with AI token usage by a proficient engineer could equate to their salary or overall employment cost. This scenario would necessitate the introduction of spending caps to manage expenses effectively.
AI tokens are the computational units required to process prompts and generate responses in AI systems. The escalating costs associated with these tokens have become a significant concern for tech companies. Meta, for instance, recently discontinued an internal leaderboard that tracked AI token spending after projections indicated potential expenditures reaching billions of dollars in 2026.
Other tech giants are facing similar challenges. Uber exceeded its AI coding budget for 2026 by April, prompting a reassessment of its AI strategies. Microsoft also took measures by canceling licenses for certain AI tools, opting instead to consolidate its engineers around its proprietary Copilot CLI tool to manage costs.
Mosseri emphasized the importance of managing AI token costs akin to other business resources, such as payroll and operational expenditures. He highlighted the necessity of allocating resources judiciously across teams, considering factors like GPU and CPU capacity, storage, and RAM. Introducing token budgets would require proportional caps based on the company’s confidence in an engineer’s ability to utilize the budget effectively and generate a positive return on investment.
Currently, Meta does not impose token caps on its employees. However, Mosseri believes that implementing such measures could be beneficial in the future. He also anticipates that AI token costs will decrease over time as competition among AI model providers intensifies, leading to more attractive pricing to capture market share.
To address immediate concerns, Meta has taken steps to curtail unnecessary token expenditures by eliminating non-essential activities, such as the aforementioned token spend leaderboard. Mosseri noted that it’s relatively easy to create systems that consume tokens without adding significant value, underscoring the need for strategic management of AI resources.
As AI technologies continue to evolve and integrate into various business operations, companies must balance innovation with cost management. Establishing clear guidelines and budgets for AI token usage will be crucial in ensuring sustainable growth and maximizing the return on investment in AI initiatives.
