Chatbot from China AI expert: "In three weeks nobody talks about Deepseek"






Deepseek put the AI ​​scene and especially the stock exchange in turmoil. AI expert Fabian Westerheide considers this to be exaggerated-but warns of influence from China

It was the news that the Wallstreet shocked: The AI ​​model of the Chinese startup Deepseek was roughly as good as that of the American competitors around chatt and grok, but was developed significantly cheaper. The view that artificial intelligence may get out with fewer resources than expected let Tech shares collapse at the beginning of the week-especially Nvidia. The company’s course fell by 17 percent in one day, almost $ 600 billion IPO value were gone. A US company had never lost so much value in one day.

Deepseek’s promise: The voice model delivers similarly robust results such as chatt. However, according to the company, the open source model works with cheaper chips and needs less computing power. US President Donald Trump gave the news a “wake-up call” that it was good if cheaper and faster AI applications were possible. But is it really the revolution for the AI ​​industry or just a small pivot?

“Good” artificial intelligence in China also

If it goes to Fabian Westerheide, panic is exaggerated. “In my opinion there is nothing noise. The entrepreneur and investor has been working in the field of artificial intelligence for years and organizes the “Rise of AI” conference annually. “The greatest realization is that in China there is also good work in the development of voice models.”

The entrepreneur Fabian Westerheide has been investing in artificial intelligence for years and organizing an annual industry meeting
© Thomas Tiefsea diver

According to Deepseek, the training of the model cost around $ 5.6 million – which would be significantly less than the dozens of billions that have needed other models so far. It took much fewer chips than the American competitors – and these chips were even out of date. However, the information can hardly be checked.

From a technological point of view, the model is quite interesting, says Westerheide. “There are a few new approaches in Deepseek’s architecture, but it’s not a revolution. I think Deepseek is overestimated.” There have always been new models that are more efficient than everything earlier. “The technology did not surprise us, the effects on the stock exchange are more surprising. This shows us how nervous the AI ​​bubble is and that it does not know how to react.”

Doubts about the presentation of Deepseek

The panic of the stock markets comes above all because investors fear that high -performance chips will be asked less in the future. This primarily affects Nvidia, whose chips have so far been the first choice for AI. Westerheide thinks for a song on Nvidia: Deepseek was also trained with Nvidia chips-although their export to China is severely restricted by the USA. “Therefore, the numbers that are now circulating should not be honest because many of the chips used are illegal in China,” says Westerheide. In the United States, too, some doubts about the official presentation of Deepseek. The industry has been speculating in the industry for some time that the company has more chips available than officially admitted. “The need for Nvidia chips will remain great,” says Fabian Westerheide.

In Germany, the news of the progress in China also ensured positive reactions. “Germany and Europe have not lost the race at AI, but we finally have to end the warm -up and start with the race,” said Susanne Dehmel from the Digitalverband Bitkom of the Reuters news agency. Digital Minister Volker Wissing also spoke of a good perspective for the German economy in an interview with Deutschlandfunk. In research and development, Germany is “particularly good in the race”.

There are also security concerns

Westerheide also sees the application fields for the deepseek model there. “The innovations are interesting for researchers, but I don’t expect a large -scale effort,” says the entrepreneur. “So far, hardly any German companies have really used artificial intelligence or have the technical infrastructure to incorporate such models. Deepseek does not change the fact that the German middle class is not digital.”

There are also security concerns. The Deepseek servers are stored in China and so far the model seems to be on a party line for the Chinese government. If you ask, for example, about the suppression of the Uiguren or the massacre at Tiananmen-Platz in 1989, information is embezzled. “You shouldn’t underestimate the Chinese influence by the model – Deepseek is also censored,” says Westerheide. Therefore, the model should never be installed in critical infrastructure. “It’s not just about the technology competition, but about geopolitics. Artificial intelligence is a figure more on the chess board between China and the USA.”

Westerheide therefore does not believe that Deepseek is now revolutionizing the AI ​​industry. “The next big model will come in two or three weeks and then nobody talks about Deepseek.”

Capital

Capital is a partner brand of the star. You can use your selected content with your star+ See the subscription. You can find more from Capital at www.stern.de/capital.

  • Artificial intelligence

  • China

  • USA

  • Nvidia

  • Chatt

  • Donald Trump

  • Chip industry