Adam Mosseri, the head of Instagram, has indicated that technology companies, including Meta, may soon implement limits on how much individual engineers can spend on artificial intelligence (AI) tools. Speaking on Lenny's Podcast, Mosseri suggested that within a year or two, the 'burn rate' from an engineer's AI token usage could equal their annual salary, making caps an inevitable necessity to manage burgeoning operational costs.
AI token spend refers to the computational cost incurred when processing AI prompts and generating responses. This expense has become a significant concern across the tech industry, with Meta itself on track for billions of pounds in AI-related costs for 2026. The company has already taken steps to curb spending, including discontinuing an internal leaderboard that tracked AI token usage among employees.
Mosseri emphasised that managing AI token costs would become as critical as overseeing other core business expenditures, such as payroll, operational expenses (OpEx), and the deployment of hardware resources like GPUs and CPUs. He drew parallels to how companies allocate budgets for staffing or specific projects, suggesting that AI token budgets would similarly need to be proportional to an engineer's ability to deliver a positive return on investment (ROI).
Meta currently does not impose specific token caps on its employees. However, Mosseri views such measures as a potentially healthy development for future financial management. He also expressed a long-term expectation that AI token costs might decrease as competition intensifies among AI model providers, potentially leading to a 'pricing war' that benefits users.
The challenges faced by Meta are not isolated. Other major tech firms are also grappling with the financial implications of extensive AI experimentation. Uber, for instance, reportedly exhausted its 2026 AI coding budget by April, while Microsoft cancelled certain Claude Code licenses, opting to consolidate its engineers around its proprietary Copilot CLI tool to manage soaring token costs. These instances highlight a broader industry trend towards a more disciplined approach to AI resource allocation.