The artificial intelligence (AI) industry was shaken this month by the arrival of DeepSeek R1a model developed by the Chinese laboratory DeepSeek. According to experts, this model rivals American leaders like o1 of OpenAI and Llama by Meta. With access to tens of thousands of NVIDIA GPUs, DeepSeek R1 is positioned as one of the most advanced models in the world.
A technological breakthrough supported by NVIDIA GPUs
GPUs are at the heart of advances in AI, and NVIDIAa world leader in this field, provides the most popular chips for model training. DeepSeek would have at its disposal approximately 50 000 GPU NVIDIA H100the current generation of Hopper chips, although this remains a sensitive subject due to export restrictions imposed by the United States. These chips, known for their computing power and memory capacity, are essential for training models like R1, which require massive resources.
However, China’s access to these technologies remains limited by strict controls. After banning the sale of the H100 and H200, the US administration also blocked alternatives like the H800 and A800 in October 2023. Despite these restrictions, Alexandr Wang, CEO of Scale AI, estimates that Chinese laboratories have accumulated much more Advanced GPUs than you think.
DeepSeek R1: High-level performance
DeepSeek R1 stood out during the evaluation test called “Humanity’s Last Examination”designed by Scale AI. This test is based on complex questions in mathematics, physics, biology and chemistry, developed by university experts. Result: the R1 model matched, or even surpassed, reference models like OpenAI’s o1. This performance confirms that China is quickly catching up in the field of AI.
-In an interview with CNBC, Alexandr Wang pointed out that the United States has historically dominated the AI industry, but Chinese players, like DeepSeek, are changing that. Wang called it symbolic that R1 was released on Christmas Day, a holiday period in many Western countries. This shows China’s determination to establish itself as a global leader, even in the face of supply constraints on critical technologies.
The future of AI: between innovation and geopolitics
The success of DeepSeek R1 highlights the growing tensions between technological innovation and geopolitical regulations. As the United States tightens export controls to limit China’s access to cutting-edge technology, Chinese labs continue to find ways to advance. This race for AI is not only about technological supremacy, but also economic, strategic and cultural issues.
DeepSeek R1 marks a major breakthrough for China in the field of artificial intelligence. With its remarkable performance and impressive GPU infrastructure, it stands out as a serious competitor to the American leaders. However, challenges related to material supply and international regulations could influence its future development. The competition between the United States and China for AI dominance is far from over, and it will continue to shape the world’s technological future.
