Why is NVIDIA investors scared by Chinese AI Upstart Deepseek
A Chinese company obtained a $ 5.6 million artificial intelligence breakthrough from the market value of NVIDIA on Monday that Wall Street has weakened Wall Street's confidence in AI's AI expenditure frenzy, and a significant blow to American technical leaders.
However, many people in Silicon Valley believe that extensive selling is an excessive response to the latest model of Deepseek. They believe that this can fundamentally reduce the cost of the technology and continue to stimulate the broader adoption and of AI's demand for NVIDIA chips. Practical.
Pat Gelsinger, who has recently been forced to serve as CEO of Intel, is one of the people who have purchased RBI Nvidia shares on Monday on Monday. He said in Linkedin's post: “Market response is wrong: reducing the cost of AI will expand the market.” Deepseek is an incredible project that will attract more AI use. ”
As investors bet on large technology companies' dissatisfaction with its strong AI processors, NVIDIA has become the most valuable company in the world. Jensen Huan, CEO of chip manufacturers, predicts that the AI data center worth 1 TN will be built in the next few years.
Confidence is the concept of AI's “expansion law”. The concept is promoted by senior leaders of AI startups such as AI and Anthropic, which indicates that the AI model has become smarter because they are provided with more data and computing resources.
DeepSeek has released its high -featured R1 model and research papers, explaining the public production method, and it seems to break the spell of the zoomed law, because its chat robot has crossed the top of the US App Store Chart on the weekend. The Philadelphia Semiconductor Index has been reduced by 9.2 %, which is the worst decline in daily since March 2020.
China Science and Technology Champion Huawei has become Nvidia's main competitor in China. The Financial Times has previously reported that it has been working with AI companies including Deepseek to adapt to the model of receiving NVIDIA GPU training to infer its ascending chip.
“Huawei is getting better and better. They have an openness because the government told large technology companies that they need to buy chips and use it for inferring.” A semiconductor investor in Beijing said.
Because the largest technology company in the United States is preparing to report its latest income, the largest technology company in the United States announced its leadership position on AI, which triggered the announcement of the stock market spasm on Monday. US President Donald Trump said Deepseek “should be the alarm of our industry, and we need to focus on the victory of laser competition.”
In December, Deepseek released the V3 model. The model claims that the model is equivalent to the model of Openai and Google, but trained a small part of the budget for $ 5.6 million. The Chinese company said it only uses 2,048 NVIDIA chips, and this can be obtained without violating US export control, which can access the latest products between China and American chip manufacturers.
Then, last week, it launched its latest R1 model, which is a “reasoning model” equivalent to OPENAI O1.
Further frightening investors, DeepSeek engineers can unlock greater performance by writing code without relying on NVIDIA's CUDA software platform. The platform is widely believed that the dominant position of Silicon Valley chip manufacturers for AI development.
“Deepseek has been upgraded,” Stephen Yiu, chief investor of Blue Whale Growth supported by Billionaire Peter Hargreaves, said that the fund reduced to the huge seven large groups of large American technology companies last month. Its concern of the huge expenditure of AI.
YIU said that the largest technology company in the United States “has monopolized the price of AI -admission tickets for billions of dollars, otherwise you have no chance to challenge the status quo.” And the penetration of very positive development. ”
Monday's very ecstatic seller bet on the high stock price of NVIDIA in recent weeks. According to the calculation of the data group S3 PARTNERS, NVIDIA's 17 % stock price fell to the short vending side to lead to $ 6.75 billion in profits.
“A Chinese entity issued an open source code before the income of all large technology companies,” a short -selling seller who has interest in many large AI companies. “They tell you that there is no value (in the AI models of these companies), this is commercialized.”
However, some analysts challenged the idea of the breakthrough AI of DeepSeek's breakthrough AI.
The Dylan Patel's half -analysis of Chip Consulting estimates that Deepseek and their sister companies The Hedge Fund Flyer can use tens of thousands of NVIDIA GPUs, which are used to train their predecessors.
Patt said: “Deepseek spent more than $ 500 million in GPUs in the history of the company.” “Although their training running is very effective, a large number of experiments and testing are needed to work.”
Techinsights's G Dan Hutcheson said that market response did not reflect who was the most breakthrough in Deepseek. He said: “I think this is not a big blow to NVIDIA. I think this is a bigger problem for companies like OpenAi trying to sell these services.”
NVIDIA argued on Monday that Deepseek's innovation will benefit rather than bombing its business.
NVIDIA said: “DeepSeek is an excellent AI progress, and it is also a perfect example of testing time expansion.” Refers to the AI system, when the user raises questions or responds through the steps of “reasoning” or adopting multiple links, the system is Consume more computing resources or setting tasks. “Reasoning requires a lot of NVIDIA GPU and high -performance network.”
The meaning of the NVIDIA statement is that by breaking through the possible boundaries of the “open source” AI model, Deepseek has actually increased demand for the needs of the chip used to run them.
Although NVIDIA is famous for providing chips for “training” or building new AI systems, it said that it now obtains the same “reasoning” income or use the finished model to handle user requests.
Huang argued in a recent podcast that because of the “reasoning” new AI model or spending time plan and provided answers to complex query, the demand for reasoning was “one billion yuan.”
AI Investor Radical Ventures co -founder Jordan Jacobs said: “NVIDIA chip, there are two uses for training and reasoning, we are at the beginning of reasoning stories.” The stock fell 17 %. “Just as the world we see to AI, a large number of chips need to be upgraded. Selling seems to be an excessive reaction and lack of understanding.”
AI search startup PLLEXITY's chief business officer Dmitry Shevelenko, the headquarters of San Francisco, said: “The market is a good choice for NVIDIA.” “In any case, Jensen won.”
Christina Criddle's other reports in San Francisco and Beijing's Eleanor Olcott (Eleanor Olcott)