Meta Caps Internal AI Token Spending After Costs Approach Billions in 2026
Meta has restricted internal spending on AI tokens after projected costs for the technology neared billions of dollars by 2026. The cap aims to control rapidly growing expenses related to the company's extensive AI development efforts.
Background
- Meta (Facebook, Instagram, WhatsApp's parent) has been racing to build its own large language models (LLMs) and generative AI tools, competing with OpenAI, Google, and Anthropic.
- "Token spending" refers to the computational cost of running AI inference (generating responses) internally — every query to an AI model costs money based on the number of text "tokens" processed.
- Meta's internal AI usage (employees using AI coding assistants, content generation tools, etc.) has grown so fast that its projected 2026 inference costs are approaching billions of dollars.
- The company is now capping how many AI tokens employees can use, a sign that even deep-pocketed tech giants are wrestling with the enormous operational expenses of running AI at scale.
- This is an early, concrete example of the "inference cost crisis" many analysts predicted: training AI is expensive, but running it for millions of daily users can be even more costly over time.