China's Deepseek triggered the selling of technology stocks, and the Nvida transaction fell 12% before the market
American technology companies plummeted during the trading before trading on Monday, which is part of the global selling, because Chinese startup Deepseek has triggered people's concerns about artificial intelligence competitiveness and the leading position of the United States in the field.
Chip designer's shares NivineThe main beneficiaries of artificial intelligence speculation fell more than 11%before the opening. This defeat has also extended to other artificial intelligence industries, such as Micron and Moumou HoldingsAnd derivatives related to power construction, such as Constellation energy and viewEssence
The international market has also felt the impact. Dutch chip company Asley Company and ASM International In European transactions, the stock fell more than 7% and 11% respectively. In Asia, Japanese chip -related stocks include Edwan Test and Tokyo Electronics Generally low.
Deepseek launched a free, open source large language model at the end of December, claiming that its development time was only two months, and the cost was less than $ 6 million -which was much less than the cost required by Western peers. Last week, the company released a reasoning model that reported that the model was also better than the latest model of OpenAI in many third -party tests.
The development of these situations has caused people's concerns, and people are worried about whether large technology companies invest in artificial intelligence models and data centers reasonable. They also warned that the United States did not lead the industry as before.
Raymond James semiconductor analyst Srini Pajjuri said in a report on Monday: “Deepseek obviously cannot obtain as much computing power as large -scale companies in the United States, but trying to develop a model that looks very competitive.”
Pajjuri said that DeepSeek may “make American oversized companies more urgent”. These large -scale computing infrastructure participants include including Amazon and MicrosoftEssence Specifically, analysts said that these companies can use their advantages to obtain graphical processing units to make themselves stand out in cheaper choices.
The GPU is the key part of the infrastructure required for training large artificial intelligence models. NVIDIA is the leader of the GPU market.
Citi analysts said that DeepSeek's big language model also “prompts investors to ask for calculation costs.”
Nevertheless, although they point out that American companies' dominant positions on the most advanced artificial intelligence model may be threatened, a key obstacle to competitors is to get the best chips. Therefore, they expressed that leading artificial intelligence companies may not give up more advanced GPUs.
They added that President Donald Trump recently announced the $ 500 billion Stargate AI project is “recognition of advanced chips.”
To be sure, Bernstein analysts have doubted whether the actual development cost of the DeepSeek tool is less than $ 6 million. They believe that the number ignores other costs of previous research and experiments so that the technology can reach today's level.
They emphasized that the model of DeepSeek “looks great”, but should not be regarded as “miracle”. They said that the panic of “the funeral of the artificial intelligence infrastructure complex we know” was “exaggerated.”
—CNBC's Lee Ying Shan, Alex Harring and Michael Bloom contributed to this article.