Nvidia loses $589 billion in a day after DeepSeek’s R1

Nvidia’s shares fell roughly 17 percent on Monday 27 January 2025, taking about $589 billion of market value with them. Bloomberg recorded it as the largest one-day dollar loss for any company in the history of the United States stock market. The Nasdaq Composite closed down more than 3 percent, and the selling reached chip-equipment makers in Europe as well.

The cause was a piece of software published a week earlier. On 20 January the Chinese laboratory DeepSeek had released R1, an open-weight model trained with reinforcement learning to reason step by step, which matched OpenAI’s o1 on several benchmarks at a reported cost far below what American laboratories were spending. The chatbot app built on it reached the top of the United States iOS App Store over the following weekend.

Nvidia’s valuation rested on an assumption the market had not been asked to examine closely: that each advance in model capability would require more of its accelerators than the last. R1 offered a counter-example. If a laboratory outside the United States could reach comparable results for a fraction of the money, the demand that justified the buildout might be smaller than it looked.

The timing sharpened the point. Six days before the selloff, OpenAI, SoftBank, and Oracle had stood at the White House to announce Stargate, a venture to spend up to $500 billion on American data centers by 2029. The question analysts began asking on 27 January was what return that spending would earn if the frontier turned out to be cheaper to reach than assumed.

The rout settled nothing about whether the spending was justified. What it established was that a single model release, from a laboratory most investors had not heard of a month earlier, could move hundreds of billions of dollars of American market value in an afternoon, and that the case for the buildout would from then on have to be argued rather than assumed.