No one thought that the path of general artificial intelligence (AG) would be fluid for investors, but the emergence of Deepseek clearly launched a twist of the plot in the story of the AI.
Nvidia (Nvda -3.67%)) And other AI actions plunged on Monday, January 27 on Monday, January 27, while investors responded to the threat of Deepseek, the Chinese AI chatbot which competes with the best models like Chatgpt for a fraction of the cost.
NVIDIA lost 17% in a single session, annihilating $ 600 billion in market value, the largest loss of a day for a single actions in market history. Since then, Nvidia has recovered some of these losses, a sign that investors may believe that the sale could have been an excessive reaction.
Nevertheless, IA’s actions remain considerably lower, and Nvidia himself has tipped her hat to the Chinese start-up, with a spokesperson describing him as “excellent advancement of AI”.
Deepseek also seems to gain credibility, because MicrosoftWho is supposed to be the largest OPENAI investor, has already added the model to its Azure Cloud infrastructure service.
So what size is a deep threat to the AI ecosystem? To answer this question, let us first describe some facts on Deepseek.
Image source: NVIDIA.
What we know of Deepseek
Deepseek is a start-up of Chinese AI founded by the Head of Hell Funds Liang Wenfeng in May 2023. Unlike the Openai Chatppt or AlphabetGemini, Deepseek use an open source model model, which means that developers can update it and adapt it to their own needs.
Deepseek is significant because its R1 model competes with the O1 of Openai in categories such as mathematics, code and reasoning tasks, and it would do it allegedly with less advanced fleas and at a much lower cost.
According to an estimate, it costs the O1 model of OPENAI $ 60 to generate a million production tokens, while R1 Deepseek can provide the same amount for only $ 2.19.
Deepseek impressed the initiates of the industry with a 22 -page research document explaining the operation of its model, but the company was also accused by Openai of using a method called distillation to build its models, a profitable way to train An AI model using larger, more follower. This constitutes a violation of the conditions of use of OpenAi.
Distillation is commonly used in the AI, but if this accusation is true, it would seem to be able a lot of credibility of Deepseek, which gives the impression that the Chinese start-up has plagiarized at least part of its model.
There is also a debate on the amount that Deepseek really paid for its infrastructure, because it said it only cost $ 5.6 million to form its V3 model. The V3 was built on Nvidia H800s, which was designed to bypass the American export rules and operate in a similar way to H100S, the NVIDIA GPUs which have been widely used to build an IA infrastructure and models in states -Unis
Analysts have questioned the figure of $ 5.6 million, and that does not seem to include essential costs such as research, architecture or data, which makes it difficult to comparison with the AI models based In the United States that required billions of dollars in investments.
What it means for Nvidia
It is too early to know what are the implications of Deepseek for Nvidia and the wider AI sector, and there is still a lot of uncertainty around what Deepseek has achieved.
The company seems to have achieved real efficiency gains, but these seem less impressive if its model was partly built by borrowing from Openai. The real cost of the model is not clear either.
The Deepseek-R1 launch was called a “Sputnik moment” by the Honcho de la Silicon Valley Marc Andreessen and others, and the geopolitical implications of the new chatbot could be just as significant as technological technologies.
The United States could react by intensifying technological cold war with China, more tightening export rules and taking other measures. In addition, allowing Deepseek on American smartphones while prohibiting Tiktok seems incongruous, and American companies and governments are probably skeptical about the delivery of their data to a Chinese start-up.
For nvidia investors, it should also be remembered that this is only an episode of a technological development of several years, and is probably not as significant as a day sale of 600 billion Dollars makes it appear. Even if Deepseek moves the whole industry to a more effective open source architecture, this could be positive for Nvidia in the long term. According to Jews Paradox, the drop in price to execute AI models could increase demand, resulting in an increase in total consumption, which would lead to more purchases of AI chips from Nvidia, although probably at a cost lower.
It is also significant that Deepseek has been built on Nvidia chips. No one disputes his supremacy there.
Why am I not worried about Nvidia
Investors should not miss the trees forest here. You must remember that a competent cat-cat is not the goal here. The ultimate goal is Artificial general intelligenceIncluding applications such as autonomous vehicles and robotics, and it is not clear if Deepseek considerably modifies the calculation around this.
These technologies are powerful and precious enough for the race to AGE continues, and the technology giants participate will continue to pay billions in the infrastructure necessary to build it. Efficiency is important, but technological leadership is the real price here.
Nvidia’s actions have already been treated a Deepseek setback, and this could also be true for Nvidia’s affairs, but the company has proven to be agile before.
It has evolved its technology to pass video games mainly in service to the extraction of cryptocurrency, an AI, autonomous vehicles, a 3D rendering, etc. CEO Jensen Huang is rightly considered a visionary in industry, and he continues to innovate quickly with his new Development Rubin platform.
The AI border will continue to evolve and Nvidia will adapt to market conditions if necessary. Whatever the impact of Deepseek, the race for AG does not disappear, and Nvidia either.