NVIDIA, which was rising to the peak of the stock market by selling computer chips that powered world artificial intelligence, a small Chinese company was modified by a severe control of reality that showed that it could do more with what NVIDIA is doing.
On Monday, NVIDIA shares rushed by more than 16 percent after the Deepseek company showed that it can train a top AI system with a fraction of the NVIDIA chips, which in the past uses Openai, the Chatgpt manufacturer.
Deepseek issued a consensus in the technology industry that companies would have to spend billions and billions of dollars for new data centers to build larger and better AI systems. In the middle of these data centers, there would be one thing that could not have any AI project without: the huge cache of NVIDIA chips.
Silicon Valley Company, according to some estimates, controls 90 percent of the market for specialized chips used to create AI systems. Since Openi released Chatgpt at the end of 2022, a remarkable run has been running. In the last two calendar years, NVIDIA revenues have jumped by more than 200 percent to $ 126 billion, while the total value of the company has closed by 700 percent since Friday. , the top of $ 3.62 trillion in November.
However, the apparent breakthrough of Deepseek has shown that NVIDIA chips may not be as unlimited as some imagined just a week ago. While NVIDIA is still in the enviable position – there is little competition for its AI chips – companies that buy their technology could slow their expenses.
“Before, AI was bigger, better, faster.” Larger chips are equal to greater AI abilities, ”said Patrick Moorhead, CEO of Moor Insights & Strategy, research company Tech and semiconductors. “But it was so fast that it raises questions about how long it has been valid for Nvidia and whether people will need so many of their chips in the future.”
Deepseek also went down sharing other semiconductor companies, including Broadcom, Micron Technology and Tai -Wan semiconductors Manufacturing Inc.
Deepseek’s arrival crystallized concerns that had already thrown the shadow above NVIDIA. At the end of last year, AI leaders began to warn that the improvement of chatbots was slowing down. Previously, they relied on a simple formula to provide progress: spending as many data from the Internet as possible and pushing into large language models-technology that powered Chatboty-on Ever Bigger Computers.
But the concept that is known in the field as the scaling lawsHe began to fall out of kindness because data companies run out of data. This has led companies to start experimenting with new techniques to improve their systems. It also raised questions from NVIDIA investors about the consequences for their business.
Jensen Huang, CEO of NVIDIA, said at the CES technology exhibition in January, that the new techniques “control the huge demand for nvidia chips”. He said that companies use NVIDIA chips to power models that make more justification in data centers, which means it will be about its technology, no less demand.
In Monday’s statement, Mylene Mangalindan, a spokesman for NVIDIA, said Deepseek shows that these new techniques work. She added that “Deepseek is excellent AI progress.”
However, investors watched whether the changes in the way AI are made would strangle the business of NVIDIA and the success of Deepseek talks about the uncertainty of whether to build and supply AI systems.
“That was something we were working on,” said Daniel Newman, CEO of Futurum Group, a technology research company. “Everybody watched this goal because the cost of AI training was too high.”
Not everyone is convinced that the rise of NVIDIA is stopping. Its supplies were volatile and in August fell by more than 10 percent Despite the news, the company would delay the transport of its latest artificial intelligence chip. Later it bounced.
In the note for investors on Monday, Stacy Rasgon, semiconductor analysts in Bernstein Research, he said Deepseek spent more money on the construction of his system than he claimed. He added that more efficiently to build multiple AI systems should mean more demand, as more companies can afford to invest in them.
Panic at the weekend, Mr. Rasgon said, “It seems to be exaggerated.”
Nvidia changes are the latest evidence of how AI continues to support the stock market. Last year, Microsoft replaced Apple as the most valuable company in the world for its early pressure in AI in June, and NVIDIA demanded sharply a crown.
Now Apple is back to the lead after released its own AI system called Apple Intelligence for iPhone. However, there are indications that they do not have to hold this position for a long time. This month, IT deactivated one of the abilities of the signature – Aggregation and Summary of Report Notification – after customers and companies have complained that its software is distorting messages.
“There was no winner,” said Mr. Newman. “The tools were fine.” However, if everyone can earn better models at lower costs, people can start accepting AI ”