# 研究：AI实际使用已改变投资者对企业价值判断

- 来源：Rohan Paul (@rohanpaul_ai)
- 发布时间：2026-07-09 12:30
- AIHOT 分数：63
- AIHOT 链接：https://aihot.virxact.com/items/cmrd1ecg4004biha1atai1q82
- 原文链接：https://x.com/rohanpaul_ai/status/2075075214946492454

## AI 摘要

一项研究利用OpenRouter上380万亿模型token数据，将token、支出和用户增长整合为AI需求信号，衡量公司股价对该信号的敏感度（AI beta）。AI beta更高的公司后续获得更高股票回报，差异在闭源模型、付费用户、经验用户和长提示场景下最显著。剔除科技股、半导体、AI ETF等因素后效应仍存。市场对沟通、指令等交互类工作的AI暴露评价积极，而对分析、科学和运营控制类工作评价消极。研究表明实际AI消费已体现在市场定价中。

## 正文

Real AI usage is already changing which companies investors think will gain or lose value.

The study uses 380 trillion tokens from OpenRouter to estimate where real AI demand is growing across firms, countries, and tasks.

Then they checked which companies’ stocks moved when AI usage went up.

Those companies later earned higher stock returns.

It turns token growth, spending growth, and user growth into one AI demand signal.

Then it checks which companies’ stock prices move with that signal, calling that sensitivity AI beta.

Firms with higher AI beta later earn higher returns, with the strongest gap coming from closed models, paid users, experienced users, and long prompts.

The effect is not just a tech-stock rally, because it remains after controlling for tech, semiconductors, AI exchange-traded funds, and AI search interest.

The market sees AI exposure most positively in interactive work like communication and instruction, and less positively in analytical, scientific, and operations-control work.

The study basically concluded that actual AI consumption already seems to carry a market price, so AI is being treated as an economic force that shifts value across firms and jobs.

---

– arxiv. org/abs/2606.30583v2

Title: "AI Premium"