DeepSeek shock: Nvidia loses 17% in one session
2025-01-27 · Publisher : Nvidia · Domain : Artificial Intelligence
A week after DeepSeek R1's release, markets question the need for massive chip spending: Nvidia falls about 17% and loses nearly $589 billion in market value, a record for a single session.
What changes
- Nvidia falls about 17% on January 27, 2025, losing nearly $589 billion in market value.
- It is the largest single-session loss of market value ever recorded by a listed company.
- The Nasdaq drops about 3%; chipmakers and power suppliers also fall.
- Trigger: an open Chinese model, R1, close to the best US models at a far lower announced training cost.
- The market realizes that algorithmic efficiency can challenge the "ever more compute" assumption.
On Monday, January 27, 2025, a week after DeepSeek R1 was released (see the "DeepSeek R1" entry of January 20, 2025), markets take the measure of the announcement. The Chinese reasoning model, released with its weights under the MIT license, rivals OpenAI's o1, and DeepSeek claims it trained its V3 base model for about $5.6 million of compute on H800 chips, which are restricted for export. At the same time, the DeepSeek app tops the US App Store. ### What happened - Nvidia closes down about 17% and loses nearly $589 billion in market value, the largest single-session loss in US stock market history. - The Nasdaq falls about 3%. Broadcom, ASML, data center and electricity suppliers follow. - Nvidia responds by calling DeepSeek "an excellent AI advancement" and stresses that inference for reasoning models will still require many chips. ### Why it matters The session exposes the fragility of a central market assumption: that model performance mostly grows with the amount of compute purchased. DeepSeek shows that a lab subject to export restrictions can approach the state of the art through algorithmic efficiency (mixture of experts, reinforcement learning). The Jevons paradox debate begins: cheaper AI could instead make demand soar. ### Caveats The $5.6 million figure only covers the compute of V3's final training run, not research, experiments or the company's GPU fleet. Several analysts called it misleading when compared with the full budgets of US labs. The market shock was partly erased in the following days. ### Key takeaways The "DeepSeek moment" marks the arrival of Chinese open models in the leading pack and is a reminder that the value of AI does not come down to the volume of chips purchased.
Links
- CNBC — Nvidia sheds almost $600 billion in market cap, biggest one-day loss in U.S. history
- NBC News — Nvidia loses nearly $600 billion in market value after Chinese AI startup bursts onto scene
- Yahoo Finance — Nvidia stock plummets, loses record $589 billion as DeepSeek prompts questions over AI spending
Tags : IA, Calcul, Marchés, DeepSeek, NVIDIA, Open Source