Nvidia's chief scientist speaks out: the US AI chip ban on China has turned into a "gift", and China's overtaking in the AI field cannot be stopped.

The AI guru, who once led the development of the CUDA architecture and is the chief scientist of Nvidia, attended the second AI EXPO exhibition organized by the "Special Competition Study Program" think tank in Washington. He talked about the current situation of China's AI development at the exhibition.

First, the market push-back issue. Although Huawei's current level of AI technology development still falls short of Nvidia's, it can already provide quite reliable solutions. The US control ban has left Huawei without competitors in the Chinese market, which indirectly accelerates its growth.

Second, the talent issue. The US export control ban has instead attracted a large number of high-end talents in AI research to China. In 2019, China had approximately one-third of the world's high-end AI researchers, and now this proportion has approached half of the global total. Daly emphasized that in the past, these Chinese AI researchers mainly wrote programs for Nvidia, but now most of them have shifted to writing software for Huawei. Some media noted that Zhang Dixuan, president of Huawei's Ascend Computing Business Group, joked at an internal meeting, "Every time the US government issues a ban, it sends us a batch of top talents!"

Similarly, last month, Nvidia CEO Jensen Huang was not polite when he gave a keynote speech at the Taipei International Computer Show and accepted an interview with Ben Thompson, a famous American tech analyst. He bluntly stated that half of the world's AI researchers are from China, and the US cannot stop the advancement of AI development in mainland China. He frankly said that the US AI restriction measures strategy is "completely wrong," and if the goal of the restrictions is to ensure that the US remains ahead, then according to the current regulations, this will反而 lead to the US losing its leading position. He pointed out that if US companies do not compete in the Chinese market, then China's technology will spread across the world. "China is doing an excellent job, and 50% of the world's AI researchers are Chinese; you cannot stop them from advancing in AI development. Face reality, DeepSeek is a very impressive product."

Daly and Huang's views reflect the rapid development trend of China's AI field. The US ban is indeed a double-edged sword effect, forcing China to innovate independently.

The stricter the blockade, the stronger the breakthrough! The US AI chip export restrictions on China (such as the export controls on H100 and A100) have stimulated the development of domestic chips in China. For example, Huawei's Ascend 910B has a computing power of 376 TFLOPS, close to Nvidia's A100 level, while the Ascend 910C even surpasses the H100 in inference speed. SMIC has achieved 5-nanometer process verification through DUV multi-patterning technology, with a yield rate of 92%, surpassing TSMC's initial 5-nanometer level.

However, America's technological monopoly still leaves China with a gap in advanced processes.

As for talent, what Huang mentioned, "China has half of the world's AI talent," actually refers to the scale of practitioners. China accounts for nearly half of the total number of global AI researchers, especially in application development fields (such as computer vision and natural language processing) where there is a scale effect.

However, looking at the distribution of top AI talents, the US still has an advantage. The report on "The World's Most Influential AI Scholars" points out that the US accounts for 54% of the world's top AI scholars, while China only accounts for 14%. But it is worth noting that 42% of top AI talents choose to work in the US, of which 38% come from China, forming a "talent circulation" phenomenon.

The US has 90% of the world's high-end AI chip production capacity, and China's reliance on imports has not fundamentally changed.

Rationally speaking, the gap between China and the US in AI presents a pattern of "the US leading in basic research and China catching up in application implementation." In terms of computational hardware and core algorithms, the US still maintains a 3-5 year advantage; in data scale and application scenarios, China has formed unique competitiveness. For example, China accounts for 70% of global generative AI patents, but the performance of top models still lags behind the US by 1-2 generations.

As for the possibility of overtaking, China has opportunities for breakthroughs in specific fields (such as quantum computing and neuromorphic chips).

In summary, Daly and Huang's views reveal the resilience and potential of China's AI development, but the objective existence of the gap between China and the US cannot be ignored. China has advantages in talent scale, application innovation, and policy support, but still needs to make breakthroughs in basic research, core technologies, and high-end manufacturing. In the future, the competition between China and the US in AI will present a coexistence of "technical game and ecological competition." Whether China can achieve an overtaking will depend on the speed of technological breakthroughs, the degree of industrial chain autonomy, and the effectiveness of international cooperation strategies. In this race, openness and cooperation remain the optimal path for global AI development, while closure and confrontation will only delay the pace of technological progress.

Source: https://www.toutiao.com/article/1834006183191751/

Disclaimer: This article represents the author's personal opinions.