A Field Guide to AI Market Freakouts

What It Covers

A pattern analysis of recurring AI market freakouts since ChatGPT's 2022 launch, covering five distinct investor fear categories — from Chinese model distillation threats and circular financing concerns to CapEx escalation and performance plateaus — with context on why these cycles rarely signal a true bubble forming.

Key Questions Answered

• •AI Market Concentration: AI now represents 25% of US GDP growth and drove 75% of S&P 500 returns since ChatGPT's late 2022 launch, per Bloomberg and JPMorgan. Roughly 50% of S&P 500 stocks are AI-exposed, meaning passive index investors and 401(k) holders carry significant AI risk whether they track it or not.
• •Chinese Model Pricing Reality: Kimi K3 is priced at approximately one-third of Claude Sonnet or half of Opus — a meaningful discount, but not the pennies-on-the-dollar gap many analysts assume. Investors and enterprise buyers evaluating Chinese alternatives should verify actual pricing tiers rather than relying on headline narratives that consistently overstate the cost differential.
• •CapEx Threshold Psychology: Google reported 82% year-over-year cloud growth yet its stock fell 1.2% after announcing $200B in CapEx — a figure Wall Street treated as a psychological ceiling. Investors tracking hyperscaler earnings should monitor CapEx guidance trajectories, not just revenue growth, as the gap between the two drives market sentiment more than absolute numbers.
• •Summer Seasonality Pattern: Morgan Stanley data shows momentum stocks fell 40% in one recent month — the worst on record — consistent with a documented summer breakdown pattern that amplifies AI FUD cycles. Investors can use this seasonality framework to contextualize fear-driven selloffs in semiconductor and AI-adjacent stocks between June and September each year.
• •Compute Scarcity as a Chinese AI Limiter: Moonshot exhausted its compute capacity within the opening weekend of Kimi K3's launch. Evaluating Chinese AI competitive threats requires assessing inference capacity, not just model benchmarks — Chinese labs collectively may lack the infrastructure to serve even a fraction of the users that US labs currently handle at scale.