Meta limits internal AI token spending as costs near $1 billion by 2026

Meta has announced a new cap on its internal spending of AI tokens, citing that projected costs for the year 2026 are expected to reach billions of dollars. The company’s internal AI token system, used to pay for compute resources across its data centers, has seen rapid growth as the firm expands its generative‑AI initiatives. By limiting the total token allocation, Meta aims to curb runaway expenses while maintaining the flexibility needed for research and product development.

The decision follows a detailed cost‑analysis report that highlighted a steep rise in token consumption, driven by larger language models and increased inference workloads. The cap will be implemented gradually, with a phased reduction in token budgets for high‑cost projects and a reallocation of resources toward more cost‑efficient architectures. Meta’s leadership stated that the move is intended to preserve long‑term sustainability without compromising innovation in its AI pipeline.

Industry observers note that Meta’s token‑spending cap mirrors similar measures taken by other tech giants grappling with the economics of large‑scale AI. While the restriction may slow the pace of some experimental projects, it also signals a broader shift toward tighter cost controls in the rapidly evolving AI ecosystem. The company will monitor token usage closely and adjust the cap as needed to balance fiscal responsibility with continued advancement in artificial intelligence.