The head of Instagram at Meta, Adam Mosseri, has stated that the tech giant may have to impose AI token spending caps on individual engineers within the next one to two years. Speaking on Lenny's Podcast, Mosseri emphasized that managing the cost of AI computing and processing is becoming an urgent resource allocation issue as budgets for large language models continue to spiral out of control.
Background & Causes
According to TechCrunch, the cost of AI token consumption (the expense of processing prompts and responses) is becoming a major financial burden in Silicon Valley. Previously, Meta had to shut down an internal leaderboard tracking employees' token spend after these costs threatened to drive the company's AI expenditures into billions of dollars in 2026. This situation is not unique to Meta; Uber also had to tighten spending after blowing through its entire 2026 AI coding budget in just the first four months, while Microsoft was forced to cancel Claude Code licenses to consolidate engineers around its own Copilot CLI tool.
Technical Analysis & Technology
Tokens are the basic units used by large language models (LLMs) to process text, with each token equivalent to a few characters or a short word. Engineers continuously experimenting, optimizing source code, and running automated test cycles via AI coding assistants consume millions of tokens daily. The lack of control over API calls to advanced AI models not only overloads specialized hardware like GPUs and CPUs, but also creates valueless "token incinerators" if engineers do not optimize prompts or repeat redundant tasks.
Expert Opinions & Insights
Adam Mosseri compared AI token limits to any other operating expenditure (OpEx) resource, such as headcount, RAM, or storage capacity. He explained that allocating token quotas to each engineer in the future would have to be proportional to the company's trust in their ability to optimize costs to deliver a positive return on investment (ROI). Currently, Meta has not implemented any hard caps for employees, but leadership believes establishing this financial discipline is healthy and necessary for sustainable growth.
Impact & Future
Despite current high costs, Mosseri forecasts that token expenses will gradually decrease in the long term as AI model creators enter a price war to attract users. For the tech community, this move by giants like Meta signals that the era of unrestricted AI experimentation is ending, giving way to a phase of strict financial performance management, where engineers must not only excel technically but also know how to optimize algorithmic operational costs.