Deepseek challenges ChatGPT and Co.
Deepseek is shaking up the market with its large language models. The AI models developed by this Chinese startup are said to be at least as powerful as those behind ChatGPT and similar services, yet require significantly fewer resources.
When a company launches its own large-scale language model, this is usually no longer big news these days. It's a different story with Deepseek, whose AI products Deepseek V3 and Deepseek R1 have been making headlines for days now - and not just for that: the share prices of well-known AI companies such as Nvidia have also taken a tumble in recent days, such as Euronews reports.
The main reason for the uproar is that the Deepseek AI models appear to perform better in various performance tests than large language models from well-known manufacturers. At the same time, Deepseek's AI can be executed with significantly fewer resources, meaning that less powerful processors are required.
Deepseek also makes its fully trained AI models available as open source. This means that they can all be run on their own hardware and further developed for specific use cases.
Turning chip necessity into a virtue
Added to all this is the fact that the start-up Deepseek is based in China. The company was founded in 2023 by Liang Wenfeng, as can be seen in a portrait by Wired can be taken from the portal. He started Deepseek for research and out of curiosity, not to make a short-term profit, the portal quotes the founder from an interview.
If the US had its way, Chinese companies would not be able to make any great leaps in AI - at least not with Western technology. The United States is making sure of this by restricting chip exports to China. But how Wired it seems that this is exactly what helped Deepseek to succeed: According to the article, the company had 10,000 A100 chips from Nvidia at its disposal for its developments - which is little compared to the many resources of major Western players. The start-up made a virtue of necessity and then endeavored to develop its AI as efficiently as possible in order to compensate for the disadvantages of the comparatively weaker processors.
The fact that the player from China has succeeded in doing this is a shock to many Western companies, according to a Analysis by Ben Thompson. Deepseek not only shows what China is capable of. In general, the company has proven that the most expensive hardware is not necessary to increase the performance of an AI model. (René Jaun/dda)
