DeepSeek: Chinese AI start-up causes stock market earthquake, US tech stocks slide

manager magazin DeepSeek: Chinese AI start-up causes stock market earthquake, US tech stocks slide 9 hours • 3 minutes reading time DeepSeek from China caused massive losses on the stock market on Monday. The start-up's artificial intelligence is said to be significantly cheaper and to use less powerful chips than the competition - many tech stocks are losing significantly. The debate about the Chinese AI start-up DeepSeek is pulling tech stocks down significantly on Monday. On the German market, stocks such as Aixtron, Siltronic, Süss Microtec, Kontron and Infineon lost 5 to 10 percent of their value. Energy stocks could also suffer from the debate due to the power requirements of the computationally intensive systems: Siemens Energy had recently benefited significantly from this fantasy, with the share price slipping by more than 15 percent on Monday. The technology-heavy US index Nasdaq 100 also showed losses of more than two and a half percent. However, the index had also increased by around 25 percent in 2024, after an increase of more than 50 percent in the previous year, 2023. Over the weekend, discussions about DeepSeek's latest AI model continued to heat up, as it is said to be cost-efficient and may require less powerful AI chips than the large AI models of the established providers. Experts do not want to overvalue the latest developments. Nevertheless, the debate could trigger a consolidation of the sometimes high valuations in the tech sector, according to one stockbroker. The debate is raising concerns among investors with regard to the valuations of tech stocks such as the AI ​​chip specialists Nvidia, Broadcom and AMD or the software group Microsoft. On the Japanese stock exchange, the semiconductor industry suppliers Tokyo Electron and Advantest as well as the tech investor Softbank came under considerable pressure. According to experts, the panic about DeepSeek is exaggerated "The doomsday scenario that is currently being spread in the Twitter universe seems exaggerated," write the experts around Stacy Rasgon from the analysis house Bernstein Research. DeepSeek's AI models are good and offer good performance, but OpenAI definitely did not recreate them for $5 million. In addition, the efficiency of DeepSeek-V3 is not surprising given the model structure used. This so-called Mixture-of-Expert (MoE) architecture is designed to reduce the costs of training and operating AI models, since only a portion of the model parameters are active at any one time. According to Rasgon and colleagues, the fact that the current developments are making investors nervous is based on a misunderstanding regarding the costs of the latest DeepSeek model. Other reasons are that DeepSeek extracts smaller models from larger ones and the low prices it charges for using its programs. The first concern appears fundamentally wrong, as the company has not used any revolutionary or unknown technologies, say the experts. The second point is more interesting, but also nothing new, although the legitimacy of the approach has been substantiated. Investors' concerns about the prices that DeepSeek is charging cannot be dismissed, however. Although the company's profitability is unclear, the matter raises questions about the role and viability of proprietary AI models compared to open source approaches. But none of this is a reason to panic, because given the rapidly rising costs of global AI expansion, innovations such as those from DeepSeek are necessary. Moreover, these innovations are unlikely to go beyond what top AI developers are not aware of. And in the tech sector, increases in efficiency usually lead to growth in demand.