DeepSeek, a Chinese AI lab, released a model that reportedly matches frontier U.S. AI performance at a fraction of the training cost, sending Nvidia shares down roughly 16% in one of the largest single-day market-cap wipeouts in history. The fear is straightforward: if frontier AI can be achieved with far fewer GPUs, the hyperscaler capex supercycle that has driven Nvidia's 65% revenue growth and 71% gross margins may be structurally impaired — not merely deferred.

The countervailing case is that cheaper, more efficient AI inference historically expands total demand (Jevons paradox), and Nvidia's actual order book and revenue trajectory haven't changed overnight. The next critical data points are hyperscaler earnings calls — where Microsoft, Google, and Meta will either reaffirm or walk back GPU capex guidance — and Nvidia's own next quarterly print, which will be the first hard evidence of whether orders are softening.