AI Token Budgets: Will Meta Soon Cap Spending Per Engineer?

AI Costs Are Alarming Big Tech

The cost of using AI models has become a growing burden for major tech companies. No longer a marginal line item, it has evolved into a figure that rivals hiring and infrastructure expenses. Against this backdrop, Adam Mosseri, CEO of Instagram and a senior executive at Meta, revealed that the near future may require imposing strict caps on AI token consumption for each individual employee.

What Exactly Did Mosseri Say?

Speaking on the Lenny's Podcast, Mosseri noted that a high-performing engineer's AI spending could, within one to two years, reach a level equivalent to their full salary or total compensation cost. He made clear that this reality will inevitably require the introduction of structured limits on that spending.

Mosseri compared token budgets to other operational resources the company manages on a daily basis, noting that he had long been accustomed to allocating compute resources—processors, storage, RAM—across different teams, alongside headcount and compensation budgets. In his view, AI tokens will be governed by precisely the same logic.

AI Tokens: What Are They and What Do They Cost?

A token is the unit of measurement by which AI models process text, both as input and output. The more employees use AI tools for coding, analysis, and content generation, the higher the token bill climbs. The scale of these costs at Meta came into sharp relief when the company shut down an internal leaderboard that tracked employee token spending—the numbers were pointing to overruns in the billions of dollars for 2026 alone.

Meta Is Not Alone in This Predicament

Meta is far from the only company to have absorbed the shock of AI costs. Uber burned through its entire 2026 AI coding budget by April alone. Microsoft made the notable decision to revoke Claude Code licenses for its engineers, switching instead to its own Copilot CLI tool in a bid to rein in expenses. Taken together, these developments reveal that uncontrolled AI spending has become a widespread concern at the leadership level of major companies.

How Is Meta Handling the Problem Now?

Mosseri explained that Meta does not currently impose any token usage caps on employees, but has managed to curb some excess by shutting down what he described as "silly practices"—including the leaderboard that had turned token consumption into a pointless competitive sport. He captured the sentiment with a memorable phrase, saying that building a "token bonfire" is trivially easy, but produces virtually no real value.

Will Costs Come Down in the Future?

Over the longer term, Mosseri expects token prices to gradually decline as competition intensifies among AI model providers vying to attract more users. However, that anticipated decline does not eliminate the need for sound spending policies. The proposed caps will be set—in his conception—based on the company's confidence in an employee's ability to deploy those resources in a way that generates a genuine return on investment.

What Is the Broader Significance of This Shift?

What Mosseri is proposing is more than an internal administrative measure—it signals a fundamental shift in how companies think about AI. After a period of excitement and open-door adoption, large organizations are entering a maturity phase in which AI is measured by standards of efficiency and return, not novelty and experimentation. The key takeaways are:

• Token spending will be treated like any operational budget, subject to oversight and planning.
• Caps will be proportional to an employee's competency and their ability to generate real value.
• Companies that fail to establish clear AI spending frameworks will find themselves facing bills that far exceed their projections.
• Competition among model providers may help drive costs down over the medium term.

Ultimately, the era of open-ended, unlimited AI use in enterprise environments appears to be waning. What we are witnessing today is the beginning of a new phase: the responsible governance of this technology.

✦ بقلم فريق دروب أيديا