He expects a market correction before AGI, not because AI capabilities will plateau, but because markets are driven by hyped peak demo capabilities while real-economy adoption depends on reliability; the gap between peak capabilities and reliability is unlikely to be tolerated by markets for another three years.
Markets are not yet pricing the enormous inference compute demand that will be unlocked once humanoid robots cross reliability thresholds and wearable smart glasses have their iPhone moment; efficiency gains like DeepSeek and Turbo Quant will unlock more demand via Jevons paradox rather than leave chips idle, and even in a market correction chips will not sit idle and dark.