U.S. equities sold off sharply Monday led by the Nasdaq, after Chinese AI lab DeepSeek released a model that reportedly matches frontier U.S. performance at a fraction of the training cost. Nvidia fell roughly 16% in what would rank among the largest single-day market cap wipes in stock market history, dragging the broader AI infrastructure complex — chipmakers, data center plays, power suppliers — into double-digit declines.
The DeepSeek shock cuts to the core of the 'picks and shovels' bull thesis: if frontier AI can be trained cheaply on fewer GPUs, the multi-hundred-billion-dollar capex buildout priced into Nvidia's $215.9B revenue run-rate (up 65% YoY) and 71% gross margins may be overstated. Hyperscalers that have been guiding aggressive data center spend face immediate analyst scrutiny on whether that spend gets trimmed.
The bull case rests on Jevons' paradox: cheaper, more efficient AI drives exponentially more demand for inference compute, not less. Nvidia's data center business has consistently beat expectations, margins are elite, and a 16% gap-down historically attracts institutional re-entry on the dip. Revenue trajectory and EPS of $4.90 diluted still point to a business with few peers.
The bear case is structural: if DeepSeek's efficiency breakthrough is real and reproducible, it compresses the addressable market for high-end training clusters and raises the risk that hyperscaler GPU orders get rationalized. Nvidia trades at a premium multiple that embeds years of outsized growth — any guidance cut or capex softening from Microsoft, Google, or Amazon would compound the damage.
The key things to watch: hyperscaler capex commentary on upcoming earnings calls, Nvidia's own February earnings print, and whether DeepSeek's cost claims hold up under independent scrutiny. If the efficiency story is validated, the overhang is long-tailed; if it proves overstated, the selloff may well be the year's best entry.