Meta’s head of Instagram, Adam Mosseri, has indicated that the company is considering implementing per-engineer caps on AI token usage, suggesting a fundamental shift in how large technology firms manage escalating generative AI costs, according to TechCrunch AI.
The comments, made during a recent industry discussion, signal that Meta may soon treat AI token budgets similarly to traditional payroll allocations—a move that could establish new precedents for enterprise AI cost control as organisations grapple with the financial implications of widespread large language model adoption.
“We’re looking at token budgets the same way we look at headcount,” Mosseri reportedly said, indicating that individual engineers may soon face monthly or quarterly limits on their AI assistant usage. The approach would represent a marked departure from the current model at many technology companies, where AI coding assistants and other generative tools remain largely uncapped for engineering staff.
The potential policy shift comes as enterprises across sectors confront the reality that generative AI tools, whilst productivity-enhancing, carry substantial and often unpredictable operational costs. Unlike traditional software licences with fixed per-seat pricing, token-based AI services charge based on usage volume, creating budget uncertainty that finance departments have struggled to manage.
Meta’s consideration of hard caps suggests the company’s AI spending has reached levels that warrant formal rationing mechanisms. Whilst Mosseri did not disclose specific figures, industry estimates suggest that providing unlimited access to AI coding assistants for large engineering teams can cost enterprises millions of pounds annually, with individual heavy users potentially consuming hundreds of pounds worth of tokens monthly.
The business implications extend well beyond Meta. If implemented, per-engineer token caps could establish an industry template for AI cost management, potentially affecting how AI service providers structure their enterprise offerings. Companies selling token-based AI services may face pressure to develop more predictable pricing models, whilst enterprises may need to build sophisticated internal systems to monitor and allocate AI usage across departments.
For engineering teams, the shift could create new friction points. Developers have rapidly adopted AI coding assistants as productivity tools, with some reporting that AI-generated code now comprises significant portions of their output. Imposing caps could force engineers to ration their AI usage, potentially selecting which tasks warrant AI assistance—a calculation that could affect both productivity and job satisfaction.
The move also raises questions about equity and resource allocation within organisations. High-performing engineers who leverage AI tools most effectively might exhaust their budgets quickly, whilst others may barely use their allocations. Companies will need to determine whether caps should be uniform, role-based, or performance-linked—decisions with implications for talent management and organisational culture.
From a market perspective, per-engineer caps could benefit AI providers offering more efficient models or flat-rate pricing structures. Anthropic, OpenAI, and other foundation model companies may need to compete not just on capability but on cost-effectiveness, potentially accelerating development of smaller, more efficient models suitable for routine tasks.
The timing of Mosseri’s comments is notable. Meta has invested heavily in AI infrastructure, spending billions on graphics processing units and model development. The company offers its Llama models as open-source alternatives, yet still incurs substantial costs running AI services internally. If Meta—with its technical sophistication and scale advantages—finds AI costs challenging to manage, smaller enterprises likely face even greater difficulties.
Industry observers should monitor whether Meta actually implements such caps and how other major technology employers respond. The emergence of formal AI rationing policies at leading firms would confirm that generative AI’s operational costs have become material enough to require active management, potentially tempering expectations about near-term AI adoption rates across the enterprise sector.
Meta’s consideration of per-engineer token caps marks a maturation point for enterprise AI adoption, where initial enthusiasm meets budgetary reality. How the industry resolves this tension between AI-enabled productivity and cost control will likely shape enterprise technology strategies for years to come.
