NVIDIA, whose stock price fell with the emergence of DeepSeek, claims that 'DeepSeek inference also requires a significant amount of NVIDIA chips' and appeals to its value
Chinese AI company DeepSeek announced 'R1,' which is said to be superior to AI models completed by large companies at great cost and time, and it was found that it was trained at a much lower cost than existing models. This raised doubts among investors about the value of the NVIDIA chips used for training, and NVIDIA's stock price fell by about 17% in one day. In response, NVIDIA hastily announced that 'DeepSeek's inference execution requires a considerable amount of NVIDIA chips,' in an attempt to restore trust.
Nvidia calls China's DeepSeek R1 model 'an excellent AI advancement'
Nvidia says DeepSeek advances prove need for more of its chips By Reuters
https://www.investing.com/news/stock-market-news/nvidia-says-deepseek-advances-prove-need-for-more-of-its-chips-3832771
Nvidia drops nearly 17% as China's cheaper AI model DeepSeek sparks global tech sell-off
https://www.cnbc.com/2025/01/27/nvidia-falls-10percent-in-premarket-trading-as-chinas-deepseek-triggers-global-tech-sell-off.html
On January 20, 2025, DeepSeek released the inference models 'DeepSeek-R1-Zero' and 'DeepSeek-R1' as open source under the MIT license. While other companies such as OpenAI are said to be spending billions of dollars (several hundred billion yen) on training, this model was developed at a cost of less than $6 million (about 930 million yen) and the training period was announced to be only two months. In addition, it has many advantages over other AI models, such as being open source. The AI industry's expectations for DeepSeek have risen sharply.
Chinese AI development company 'DeepSeek' is rapidly emerging as a hot topic in the technology industry, and has also ranked first in the App Store's free app rankings - GIGAZINE
On the other hand, the fact that it was developed at low cost has cast a dark cloud over NVIDIA's corporate value.
DeepSeek is a Chinese company, but China is one of the countries known for being subject to strict chip export restrictions from the U.S. In particular, high-performance chips used for training and inference of AI models are completely shut out, and although NVIDIA and others have developed and exported chips with reduced performance to China, it is said that the performance is significantly different from the chips used by OpenAI.
However, a chip with reduced performance was developed that exceeded the OpenAI model, raising doubts about the usefulness of the chips that many companies had invested heavily in, and NVIDIA's stock price fell.
According to Reuters and other sources, on January 27, 2025, US technology stocks in general fell, with NVIDIA in particular recording a decline of about 17%, the worst decline since 2020. NVIDIA's market capitalization fell by about $588.8 billion (about 91 trillion yen).
Other semiconductor companies include Micron, which fell more than 11%, Arm, which fell more than 10%, Broadcom, which fell more than 17%, and AMD, which fell more than 6%. Energy companies also fell 28% and Constellation Energy, which fell 21%, and the emergence of DeepSeek sent shock waves through the AI industry.
In response to this situation, NVIDIA released a statement, claiming that 'DeepSeek's inference requires a large number of NVIDIA chips and high-performance networking.' In addition, since it is known that DeepSeek trained using about 2,000 NVIDIA chips for China, 'H800,' NVIDIA added, 'This shows the usefulness of our chips in the Chinese market, and many more chips will be needed in the future to meet the demand for DeepSeek's services.'
In addition, the H800 is a chip specially designed by NVIDIA to comply with US export regulations, but was subject to name restrictions in October 2023. For this reason, China cannot obtain the H800 at the time of writing.
NVIDIA welcomed DeepSeek's work, saying, 'DeepSeek has demonstrated the ability to create new models using widely available models and fully export-compliant equipment.'
Related Posts:
in Software, Posted by log1p_kr