DEPSEEK statement about the low cost of teaching AI and the use of only 2048 GPU NVIDIA is refuted by analysts

The Chinese startup Deepseek attracted general attention, announcing the creation of a competitive model of artificial intelligence (AI) with minimal costs. The company claimed that only $ 6 million was spent on chatbot training and 2048 graphic processors (GPU) were used. However, the Semianalysis report showed that at least $ 1.6 billion was invested and 50,000 GPU NVIDIA was used.

Image source: nvidia, Tomshardware.com

Recall that Deepseek appeared as a subsidiary of the Chinese Head Fund High-Flyer, which since 2023 allocated a startup in a separate direction, focused on AI technologies. Unlike most startups that rent capacities from cloud providers, the company has its own data centers, which gives full control in terms of optimization of II Models and allows you to quickly introduce innovations. In addition, since Deepseek remains a fully self -financial organization, any solutions are made faster and more flexible, making the company more effective compared to traditional market players.

Personal policy deserves special attention. The company hires experts exclusively from China, focusing on skills and the ability to solve complex problems, and not on formal diplomas. At the same time, the salaries of some researchers in Deepseek exceed $ 1.3 million per year, which allows you to attract talents from leading universities in the country.

Although DeepSeek positions itself as an innovator capable of challenging the industry leaders, Semianalysis analysts emphasize that the company’s success is based on multi-billion dollar investments, technical breakthroughs and a strong team, and the statements about the “revolutionary budget” spent on the AI ​​model can be exaggerated.