DeepSeek R1 Rocks Wall Street: Nvidia Sheds a Record ~$589 Billion in One Day

DeepSeek released and open-sourced its R1 reasoning model (MIT license) on January 20, 2025 — matching OpenAI's o1 at a fraction of the training and inference cost — and its app promptly topped the U.S. App Store free chart.

A week later, on January 27, the 'R1 shockwave' triggered a historic U.S. selloff: per Reuters and Yahoo Finance, Nvidia plunged about 17% in a single day, erasing roughly $589 billion in market value — the largest one-day loss for any U.S. stock — while the Nasdaq fell over 3%. Marc Andreessen called it 'AI's Sputnik moment.'

Why R1 Was the Detonator#

Technically, R1 stood on the V3 base released a month earlier and used reinforcement learning to reproduce the reasoning paradigm o1 had pioneered (see our o1 coverage). The real explosive was three variables stacking: MIT-licensed open weights anyone could download and self-host; training and inference costs low enough to challenge the "compute buys intelligence" narrative; and a Chinese team operating under export controls. Any one alone would not have moved markets — together they shook the foundation of the Nvidia story.

How the Shock Propagated#

The selloff logic was blunt: if frontier models can be trained this cheaply, massive compute capex needs rethinking — Nvidia first, with power and data-center supply chains dragged along. Subsequent facts argued back: Nvidia rebounded the very next day, and the year delivered Stargate's $500 billion and the Nvidia-OpenAI $100 billion LOI — cheaper intelligence stimulated more compute demand, not less. Jevons paradox, confirmed again in AI.

What It Changed#

R1's lasting impact ran deeper than the stock chart: it pulled open reasoning models up to frontier level and forced an industry price war; even OpenAI returned to open weights with gpt-oss that August (see our coverage); Chinese open models (Qwen, Kimi, GLM) became routine choices for developers worldwide. The true meaning of the 'Sputnik moment' was not who overtook whom — both sides of the race were forced to shift up a gear.

Our Take#

In hindsight the crash reads as an expectations correction, not a trend reversal: compute demand never vanished, but efficiency joined scale as the industry's co-equal storyline — confirmed repeatedly by R2's low-cost reasoning and the small-model wave that followed. For investors and builders, that week in January is the watershed: no AI asset can be valued anymore without asking, "what if someone does this at a tenth of the cost?"