Nicolaus Copernicus Superior School

Chinese AI shakes up Wall Street

Chinese company DeepSeek has introduced the DeepSeek V3 artificial intelligence model, which has shaken the tech industry and threatened the dominance of US tech giants (read more in our previous article). The model’s debut sent the Nasdaq 100 stock market plummeting by more than 2.5%, with shares of companies such as Nvidia and Microsoft posting significant losses. 

DeepSeek managed to train its AI model using about 2,000 Nvidia H800 GPUs in 55 days, at a cost of about $5.6 million. By comparison, US technology companies often invest hundreds of millions of dollars in similar projects. DeepSeek’s success challenges the notion that only the largest corporations with huge financial resources can dominate the field of artificial intelligence. 

Following the debut of DeepSeek V3, Nvidia’s stock fell 17-18%, and other technology companies such as Microsoft and Alphabet also reported losses. In total, about $1 trillion in value evaporated from the US market as a result of the event. 

It is worth recalling that DeepSeek has made its model open source, meaning that it is available for modification and analysis by the community. This approach contrasts with the practices of many Western companies, which keep their models as intellectual property and allows DeepSeek to reach profitability more quickly without relying on external funding. 

Despite its success, there are concerns about the potential use of DeepSeek to spread propaganda or collect user data. Some sources suggest that the chatbot could be used to spread disinformation in line with the Chinese government’s political line. 

Among other things, the model avoids answering questions about China’s policies, such as the events in Tiananmen Square in 1989 or the issue of Taiwan independence. In such cases, AI may initially generate an answer, but then deletes it, replacing it with the message: “Sorry, this is beyond my current scope. Let’s talk about something else.” This two-level censorship system involves the initial generation of a response, followed by a filter that evaluates and removes inappropriate content. However, the system is not foolproof and can be circumvented by asking questions in a different way or using a different language. 

Sources: theverge.com, thesun.ie, XTB, Elpais.com / Cover photos: Atlantic Council 

Scroll to Top
Skip to content