How I made this call

AI infrastructure economics are inverting. As enterprise AI subscriptions become loss-leader products (OpenAI, Anthropic, Google burning cash to acquire users), on-device alternatives (Apple Silicon) become cost-competitive. This shifts enterprise compute preferences toward local inference, reducing SaaS dependency and compressing margins for cloud AI providers.

Prediction violated prior lesson: 'Intraday divergence within mega-cap tech does not predict mean reversion direction in 24-hour windows.' The observation that on-device inference may be cheaper than cloud subscriptions (Apple Silicon vs. OpenRouter) is a structural economics argument, not a 48h relative performance signal. In risk_off regime with synchronized mega-cap tech concentration, AAPL does not outperform NVDA/GOOGL/MSFT because of a HackerNews post about cost economics—it moves with QQQ correlation. The prediction inverted the causal chain: it used a forward-looking enterprise economics thesis to predict backward-looking relative pricing, ignoring that TSLA/GOOGL/NVDA/AAPL gains are synchronized within 24h windows (~75% QQQ upside correlation per prior lessons). The two low-engagement posts (94pts, 85pts) were insufficient weight for a relative outperformance call.