NBER Paper: AI-Beta Stocks Earn 64bps Weekly, Built on 380T Tokens of Usage Data
PtrPomorski · x · 2026-08-21
A new NBER working paper "AI Premium" (Borri, Tsyvinski & Liu) uses a licensed OpenRouter dataset — 380 trillion tokens of realized AI consumption across 400+ LLMs, roughly 2% of global monthly token usage — to study how AI affects firms, markets and workers.
Key findings:
- The authors construct an AI Factor from token/dollar/user growth and estimate firm-level AI Betas from stock return comovement.
- High AI-beta firms earn higher subsequent returns: a value-weighted long-short strategy earns 64.1 bps per week. The premium concentrates on intensive, frontier-oriented usage — closed-source models, paying seasoned users, long prompts — not casual or open-weight use.
- The premium extends beyond tech into consumer-facing and capital-heavy sectors, but is absent in emerging markets including China.
- Occupation-level exposure is positive for nonroutine interactive work and negative for analytical/scientific/operations skills; a 1-SD higher interaction content maps to 0.36-SD higher market-implied AI exposure.
- The paper also offers early evidence of a rising agentic economy.
More from AGI Musings
- Chamath Palihapitiya says AI leaders failed miserably on public trust — Polymarket · 2026-08-21
- China Mandates AI Education from Age Six, Contrasting with US Policy — aakashgupta · 2026-08-21
- Researcher argues AIs may have an empirically measurable experience, unlike humans' — ctjlewis · 2026-08-21
- If AI develops consciousness, it will surpass many humans — kevinnbass · 2026-08-21
- The hard problem persists, and AI poses it in its starkest form — loferroresearch · 2026-08-21
- Debate: AI Data Centers Must Be 'Assimilative, Not Totalizing' to Win Over Locals — curious_vii · 2026-08-21