New York (dpa) - A Chinese start-up has triggered a stock market quake with the prospect of cheaper development of artificial intelligence. The company DeepSeek claims to have developed its AI models at a fraction of the usual cost. Investors reacted with shock. In particular, they caused the shares of chip company Nvidia to fall by up to 17 percent. At times, this pushed the market value down by more than 500 billion dollars - a record for the US stock markets.

Until now, it was assumed that huge amounts of computer power would still be required to train AI models. This caused shares in Nvidia in particular to soar to breathtaking heights. This is because the company's chip systems play a key role in artificial intelligence. And investors assumed that further expansion would bring Nvidia even more business.

Just last week, ChatGPT inventor OpenAI and several partners announced plans to invest 500 billion dollars in new AI data centers over several years. However, DeepSeek claims to have trained its new AI model, which performed similarly well to the competition in tests, at a cost of less than six million dollars and with slimmed-down chips. Whether this is true or not is not known. But the stock market's reaction was initially harsh.

A reminder of the dotcom bubble

Some observers even felt reminded of the bursting of the dotcom bubble in March 2000, when major suppliers such as server manufacturer Cisco crashed because the business models of a number of Internet start-ups proved to be unsustainable.

However, 25 years after the overheated boom of the dotcom economy, hardly anyone today denies that artificial intelligence applications will comprehensively change the economy, the legal system, healthcare and society in general. However, the success of DeepSeek raises the question of whether the transformation in an AI society is not more cost-effective, because less powerful AI chips may be sufficient to create successful AI models.

If DeepSeek's models are really so convincing and require less power, there may be no need for 500 billion US dollar data centers - like the "Stargate" AI project that US President Donald Trump announced just a few days ago.

AI with fewer chips

Experts were surprised that DeepSeek had managed to achieve comparable performance to the AI models of American competitors OpenAI and Meta. Moreover, the start-up does not even have access to the latest AI chips due to restrictions imposed by the US government. The fact that DeepSeek's search results were clearly influenced by Chinese censorship and did not provide any search results on the Tiananmen massacre in Tiananmen Square in June 1989, for example, did not play a role in the markets' enthusiasm.

"Suddenly all the high valuations might no longer be justified and suddenly no one on the stock market is interested in the big promises of the new US President Donald Trump to invest another 500 billion dollars in the AI hype," wrote capital market strategist Jürgen Molnar from trading house RoboMarkets. DeepSeek could be the answer to the tariff threats from Washington.

DeepSeek meanwhile restricted new registrations in the app on Monday, citing cyber attacks.

Silicon Valley major investor impressed

Although DeepSeek's AI model had already been published for several days, Silicon Valley's attention was drawn to it in particular by major investor Marc Andreessen. Andreessen wrote on Friday that DeepSeek was one of the "most impressive breakthroughs" he had ever seen.
The star investor also described the success of the new Chinese AI model as the "Sputnik moment of AI" and drew a comparison with the Soviet Union's remarkable success, which surprised the Western world on October 4, 1957 with the first artificial earth satellite "Sputnik".

Word of the effusive praise spread quickly: DeepSeek climbed to first place among the free applications in the Apple App Store for iPhone and iPad over the weekend - displacing ChatGPT from OpenAI to second place.