AI
Meta's Mosseri Sees Token Budgets as Future Cost Control for AI-Heavy Engineering
The Instagram head predicts AI spending caps per employee may arrive within a year or two as inference costs threaten to rival engineer salaries.
TechCrunch — AI
Meta is grappling with the spiraling expense of AI token consumption—the cost of running prompts through large language models—and its leadership increasingly views caps as an inevitable business tool. During a recent podcast appearance, Instagram head Adam Mosseri suggested that within the next year or two, constraining each engineer's AI spending could become as routine as managing any other operational resource.
Why Token Costs Are Becoming a Problem#
The scale of the issue is stark. Meta's AI token expenses were on track to reach billions of dollars in 2026, prompting the company to shut down an internal leaderboard that had inadvertently gamified spending. The problem extends across the industry: Uber exhausted its entire 2026 AI engineering budget by April, and Microsoft has begun discontinuing Claude Code licenses, consolidating teams around its own tools instead.
How Budget Caps Could Work#
Mosseri framed token budgets as no different from other constrained resources—compute, storage, or headcount. Rather than a blanket ceiling, he envisions caps calibrated to each engineer's track record of generating positive returns on AI-assisted work. The premise is straightforward: if a company can quantify which engineers produce outsized value from their token spending, it can allocate budgets proportionally.
"It's not that hard to build a token incinerator, and that doesn't create a lot of value," Mosseri acknowledged, alluding to the company's earlier decision to dismantle its spend leaderboard, which had encouraged wasteful experimentation without guardrails.
Looking Ahead#
Meta has not yet imposed per-engineer token limits. However, Mosseri predicted that stricter management will eventually give way to pricing competition as AI providers vie for market share, driving inference costs downward. Until then, organizations will likely resort to tighter oversight and behavioral nudges—eliminating the "silly things," as Mosseri put it—to keep spending under control.
