LessWrong Model: AI Pause Would Crash S&P 500 by 27.8% and AI Stocks by 69%

PeterMcCluskey, writing on LessWrong, has released a detailed financial model exploring the near-term economic impact of a hypothetical global AI pause. The scenario assumes governments treat AI as slightly more dangerous than nuclear weapons, enforcing drastic limits on development from 2028 to 2030. The median estimate shows a 27.8% drop in the S&P 500 and a 34-69% decline in AI stocks (hyperscalers like Microsoft, Google, Amazon, Meta, and Oracle). The model is sensitive to two key variables: how central AI is to economic growth (low 0.4 to high 0.95) and how much near-term value depends on new frontier training runs (0.25 to 0.8). When both are high, the S&P 500 could fall 35.9% and hyperscalers 40.4%. The analysis also considers post-pause regulation strength, compute growth rates, and pause duration, with results detailed in interactive Python code.

The model builds on the assumption that both markets and voters will be surprised by AI’s power, mainly in 2027, leading to a political push for a pause. The full framework includes sensitivity tables for each variable: for example, a low AI centrality assumption reduces the S&P 500 drop to 22.5%, while a high training dependence pushes it to 31.2%. Post-pause regulation that allows only 30% of unregulated progress (aggressive regulation) would see a 30.3% S&P drop. Pause duration matters less: one year vs. four years shifts the impact from -27.5% to -29.6%. The author emphasizes the numbers are based on fairly arbitrary assumptions and encourages readers to explore the source code and Claude-expanded reasoning. The key takeaway for professionals is the stark trade-off: a pause aimed at safety comes with substantial immediate financial pain, particularly for tech-heavy portfolios.