DeepSeek, a Chinese AI lab, released a model that reportedly achieves frontier-level performance at a fraction of the compute cost of comparable U.S. models, sparking fears that the massive GPU buildout underpinning Nvidia's 65% YoY revenue surge could slow if AI can be done cheaper. Nvidia fell ~16% in a single session, wiping out hundreds of billions in market cap, and dragged the broader AI infrastructure complex — data center suppliers, power plays, memory makers — sharply lower alongside it.

The tension now is whether this is a Jevons Paradox setup — cheaper AI inference drives more AI adoption and ultimately more GPU demand — or a genuine inflection where hyperscaler capex plans get trimmed. Nvidia's 71% gross margins and $4.90 diluted EPS show the current business is exceptionally profitable, but forward estimates were priced for continued exponential demand; watch hyperscaler capex guidance in upcoming earnings for the real signal.