AI Premium
创建于 更新于
摘要
Using 380 trillion tokens of realized AI consumption across more than four hundred large language models from the licensed OpenRouter dataset, covering roughly 2% of global monthly AI token consumption, this paper analyzes how AI affects firms, markets, and workers [page::0]. We construct the AI Factor from growth in tokens, dollars, and users, estimate firm-level AI Betas from stock-return comovement, and characterize the AI Premium [page::1]. First, we build a high-frequency AI factor and decompose it into salient components: frontier closed-source vs open-weight models, core vs new users, seasoned vs casual users, long vs short prompts, prompt-content categories, and agentic vs ordinary requests [page::1]. The AI factor shows uneven diffusion with token consumption concentrated in programming, technology, and sciences, and an agentic share rising from negligible to more than half of total AI consumption by 2026 [page::1][page::2][page::4]. Second, firms with higher AI betas earn higher subsequent returns; the value-weighted long-short AI-beta portfolio yields 64.1 basis points per week, with robust alphas after FF5 and momentum adjustments; robustness checks indicate the premium prices exposure to realized AI consumption rather than to technology stocks or AI-themed attention [page::3][page::11]. Third, the AI premium extends beyond technology into consumer-facing and capital-heavy sectors but is absent in emerging markets including China; developed markets near the frontier show larger premia [page::16]. Fourth, AI exposure is higher for nonroutine interactive work and social skills and lower for analytical/scientific/operations-control skills (0.36 standard deviations for interaction and communication), with early evidence of the rise of the agentic economy [page::4][page::26]. These findings collectively imply a sizeable, heterogeneous AI premium with meaningful implications for asset pricing and the allocation of economic gains from AI adoption [page::0][page::16].
