DeepSeek: China's challenge to OpenAI and a new stage in AI development
10.02.2025 | 12:45 |The "AI Action Summit 2025" is taking place in Paris on February 10-11, where one of the central themes is the rapid rise of the Chinese company DeepSeek. Founded less than two years ago, the startup has made a breakthrough by introducing "DeepSeek R1", an AI model that, at a significantly lower cost, demonstrates a level of reasoning comparable to "OpenAI O1".
However, the main difference between DeepSeek is its focus on open source, which allows users to download and deploy models on local devices for free. This step has caused a stir in the tech world. Moreover, the announcement of DeepSeek R1 has even affected the stock market - NVIDIA shares have plummeted. This is especially remarkable considering that China faces US export restrictions, making it difficult to access advanced AI chips.
In a special series of articles "Catalyst DeepSeek", CGTN reviewer Yang Zhao analyzes how the company managed to achieve such impressive results.
Maximum efficiency with limited resources
DeepSeek develops models in conditions where Chinese companies are deprived of access to the most powerful NVIDIA chips (for example, H100). However, the startup uses innovative methods that minimize the loss of computing power. Among them:
MoE (Mixture of Experts) - a method in which only the necessary parts of the neural network are used, reducing the load.
DeepSeekMLA (Multi-head Latent Attention) - an approach that saves memory by focusing on key information.
FP8 instead of BF16 and FP32 - accuracy optimization, which allows reducing the amount of computation without significant losses.
Moreover, DeepSeek abandoned the standard method of working with GPU through CUDA and switched to PTX (Parallel Thread Execution), which made it possible to compensate for the limitations of the Chinese version of NVIDIA chips (H800).
More than just technology
DeepSeek not only creates advanced AI models - the company is changing the very approach to technology development. Its CEO, Liang Wenfeng, previously headed "High-Flyer", one of the largest quantitative hedge funds in China, where AI managed investments without the participation of traders. This experience shaped DeepSeek's philosophy: instead of closed ecosystems - open-source, instead of monetization at early stages - community development and attracting the best specialists.
"When revolutionary technologies appear, a closed model gives only a temporary competitive advantage. The real power is in the growth of the team and the culture of constant innovation," emphasizes Liang Wenfeng. This approach sets DeepSeek apart from companies like OpenAI, which abandoned open models in favor of closed solutions. That is why DeepSeek's success has become not only an engineering breakthrough, but also an important cultural shift in the AI industry.
What does DeepSeek's success mean for the future of AI?
DeepSeek has become a symbol of technological change in China. If earlier local companies were catching up with the West in the field of AI applications, now China is beginning to set trends in basic technologies. According to CGTN analysis, this has become possible thanks to systemic reforms in China aimed at innovation and integration of science with business. This approach allowed DeepSeek to compete even under sanctions pressure. In this context, DeepSeek's success can change the global market. The company has already proven that a technological breakthrough is possible even without access to top NVIDIA chips. Moreover, DeepSeek's open architecture reduces dependence on several technology giants and makes AI more accessible.
Based on CGTN's analysis, it is safe to say that the resonance around DeepSeek confirms that China is becoming not only a consumer, but also an active developer of advanced AI solutions. This underscores a larger trend - state support for innovation, the creation of technological infrastructure and the formation of an ecosystem of startups capable of competing with world leaders.
At the same time, the question of DeepSeek's global impact on the AI industry remains open. Will the company be able to maintain its openness when faced with commercial challenges? Will its model really be massively used outside of China? One thing is clear: DeepSeek technologies are becoming an important topic of discussion, and its open-source strategy is a serious challenge to traditional business models in the AI industry.