U.S. equities suffered a broad AI-infrastructure rout after China's DeepSeek unveiled a large language model that reportedly matches leading U.S. models at dramatically lower training cost. Nvidia led the damage, falling roughly 16% in a single session — one of the largest single-day market-cap destructions in stock market history. The Nasdaq bore the brunt of the decline as the selloff swept across the AI supply chain.

The DeepSeek development hits Nvidia at its most sensitive nerve: the assumption that scaling laws require ever-more-expensive GPU clusters. If frontier AI can be trained efficiently with fewer chips, hyperscalers and enterprises may defer or shrink their infrastructure orders. Nvidia's fiscal-year revenue stands at $215.9B, up 65.5% YoY, with 71.1% gross margins — numbers that price in continued explosive demand. Any demand deceleration would re-rate the stock hard.

The bull case rests on the idea that cheaper inference typically expands total AI adoption (Jevons paradox), ultimately pulling through more compute, not less. Nvidia's CUDA moat, its data center backlog, and the sheer breadth of sovereign and enterprise AI spend are hard to displace in the near term. Bears counter that if training efficiency collapses the cost curve, the justification for $30-40B quarterly capex commitments by the hyperscalers weakens materially — and Nvidia's premium multiple (which embeds years of hypergrowth) is the first casualty.

The key watch items: hyperscaler capex guidance on upcoming earnings calls (Microsoft, Meta, Alphabet, Amazon), any Nvidia management response or updated revenue outlook, and whether DeepSeek's claimed efficiency gains hold up to independent scrutiny. A single capex cut signal from a tier-1 cloud customer could extend the selloff; a confirmation of unchanged spend plans could trigger a sharp reversal.