A chatbot from China shocked the Wall Street: Panic investors broke the IPO of the Nvidia chip group in one day with almost $ 600 billion. The trigger was the realization that AI software can be trained with much less computing power than before, after a message after a message from the Chinese start-up Deepseek.
Deepseek, a reportedly uncomplicated and cheap AI model of a Chinese start-up, causes excitement in the industry. If it is true that the costs of training artificial intelligence (AI) are only six instead of the usual more than $ 100 million, this would be a breakthrough and dignity of the spread of this technology, according to portfolio manager Jon Withaar of the Active manager Pictet.
“We don’t know all the details yet and something 100 % has been attached so far,” added with Aar. But in the United States, Deepseek Ai Chatgpt has already overtaken as the most popular app in Apple’s App Store. The risk capital investor Marc Andreessen therefore described the presentation of the “R1” of the Chinese provider as a “sputnik moment”. He referred to the first satellite that the then Soviet Union was shot into a job in 1957. The United States and other Western states then strengthened their space research.
“Deepseek R1 is one of the most amazing and impressive performance I have ever seen and a great gift to the world as an open source,” Andreessen wrote on the short message service X. With open source software, the program code is freely accessible and can of Everyone is freely accessible to be changed. The best known examples are the operating system “Linux” and the internet browser “Firefox”. Andreessen was considerably involved in the development of the “Firefox” Fumitor “Netscape Navigator”.
Lower database than the competition
According to Deepseek, the new AI is based on a considerably lower database than the competition. The NVIDIAS KI chips company from the “H800” model series used for training the “R1” predecessor “V3” that was published a few weeks ago. This is a stripped version of the processors that are available in the West who are not covered by the American embargo for technology exports to China. Nevertheless, both “V3” and the successor “R1”, who was only a few days old, were able to compete with rivals such as Chatgpt from OpenAAI or “Lama” from the Facebook Mother Meta.
Western technology values under pressure
According to official Chinese documents, Deepseek is checked by Liang Wenfeng, the co-founder of the HEDSE Fund High Flyer. In March 2023, the fund announced that they wanted to make a new attempt to develop artificial super information. Such an artificial general intelligence (AGI) can perform complex tasks without human intervention.
You don’t know if this is all true. This is how speculation in the industry was that Deepseek may have access to more Nvidia chips than allowed in view of the American export restrictions. Investors who have driven the share of Nvidia in the expectation of a future mega company in recent months, however, received cold feet on Monday. US President Donald Trump spoke about a wake-up call for American companies and at the same time thought it was “positive” that AI could be cheaper.
Nvidia’s share fell by almost 17 percent and thus $ 589 billion. An American company had never lost so much value in one day, but none was as expensive as Nvidia with nearly 3.5 trillion dollars.
What can the US do with better chips?
But what does that mean for the future? American companies have recently surpassed with announcements how much money they want to place in the AI infrastructure. The Chatgpt Inventor OpenAAI and various partners alone promised to invest $ 500 billion in data centers in the coming years. On the Facebook Group Meta, founder Mark Zuckerberg is only a prospect of $ 60 billion this year. Deepseek now shows that you are doing it with considerably less computing power?
American experts prefer to see it the other way around. The question is not whether Deepseek can overtake the current market leader in the United States, emphasized X. Eyé of the AI consultancy Malo Santo. It is much more about how quickly you can implement Chinese research tracks. “If Deepseek can develop this with old hardware, what can we do with recent hardware?” She asked in the American broadcaster CNBC.
Doubts about the costs of Deepseek
The technical industry will still “need a lot of chips”, was also convinced by industrial analyst Stacy Rasgon. “It is clearly a panic today,” he emphasized with a view to the stock market reaction. At the same time, Rasgon doubted the information from China: “You did not train the model for $ five million, that did not happen.”
Last week the head of the AI Company Scale AI, Alexander Wang, that Depseek had access to 50,000 H100 chip systems from Nvidia after his information, but could not talk about it because of the export restrictions. The US government enables Nvidia to only sell chips to China that are less powerful.
Not only affected nvidia
The Deepseek -shock also influenced other shares. The newspaper of the Chip Company Broadcom also lost around 17 percent. The energy companies found even harder, where investors have been speculating for months about good companies with the range of data centers. The share of Constellation Energy decreased by a fifth and the Vistra course even collapsed by 28 percent. The shares of Apple and the Facebook Group Meta, on the other hand, gave the prospect of cheaper AI.
Source: Krone
I’m Ben Stock, a journalist and author at Today Times Live. I specialize in economic news and have been working in the news industry for over five years. My experience spans from local journalism to international business reporting. In my career I’ve had the opportunity to interview some of the world’s leading economists and financial experts, giving me an insight into global trends that is unique among journalists.
