Why US sanctions can’t stop China’s AI rise
US tech sanctions have failed to curb China’s AI rise. With shrinking gaps in model performance, open-source dominance and industry-driven innovation, China is now at parity — and poised to lead. The AI race has changed. Technology expert Yin Ruizhi tells us more.
On 9 April, US artificial intelligence (AI) chip giant Nvidia stated that the US government now requires the company to obtain a licence to export its H20 chips to China and several other countries. This restriction, effective “indefinitely”, also applies to companies with parent entities in those countries, effectively banning even lower-performance chips from the Chinese market.
However, it seems that Chinese AI companies are not significantly affected. In fact, the US has increasingly fewer restrictions that can effectively limit China’s progress in the AI field. China is now on the verge of catching up with the US in AI and even surpassing it in certain areas.
In 2023, Chinese models lagged behind the best US models by nearly 20 percentage points in the Massive Multitask Language Understanding benchmark commonly used for large language models (LLMs). By the end of 2024, this figure had shrunk to just 0.3 percentage points.
Closing the AI gap
China’s AI capabilities are now very close to the US in three important indicators.
One, the gap between Chinese and US models. In 2023, Chinese models lagged behind the best US models by nearly 20 percentage points in the Massive Multitask Language Understanding benchmark commonly used for large language models (LLMs). By the end of 2024, this figure had shrunk to just 0.3 percentage points.
Two, the narrowing gap between open-source and closed-source models. Initially, OpenAI (closed-source) had a relatively clear advantage over other models. But at the start of 2024, the performance gap between the top contenders in both categories narrowed to 8%, and less than 1% in many aspects as of 2025, according to industry assessments.
Unlike in the US, where there is an ongoing debate between open-source and closed-source models, China’s AI sector fully embraces open-source. As a result, the gap between Chinese models and their US counterparts is expected to narrow as open-source continues to flourish in China.
... once the leadership in AI shifts from academia to the corporate sector, China quickly catches up or even surpasses the US.
Three, the role of academia in the AI sector is rapidly diminishing. This indicator has long been neglected by the media. Prior to 2006, the development of LLMs was primarily driven by academia, with models developed by industry accounting for less than 20%. This proportion jumped to 60% in 2023, and then to a whopping 90% in 2024.
Gap in academia
Apart from AI chips, another shortcoming in the China-US AI battle lies in academia. Due to historical reasons, China has yet to develop a powerful academic system, and Chinese universities contribute far less to the AI field compared to American ones. Thus, at the start of each AI wave, China tends to lag behind the US. However, as the sector matures and develops, and once the leadership in AI shifts from academia to the corporate sector, China quickly catches up or even surpasses the US.
Today, China is already leading in terms of the number of AI papers and patents. AI is also becoming more efficient, compact and cost-effective. As various infrastructures mature, we will see more LLM companies and teams consisting of just a few dozen people, or even just a handful of individuals, in the future. This signifies that a large number of small and medium-sized enterprises, as well as innovative companies, can independently enter this field.
In the previous internet wave, while the US maintained its lead in cutting-edge technology, it was ultimately surpassed by China in terms of internet application scenarios. The same thing could happen again in the AI sector.
The scale of China’s industrial and internet services sectors far exceeds that of the US. In the previous internet wave, while the US maintained its lead in cutting-edge technology, it was ultimately surpassed by China in terms of internet application scenarios. The same thing could happen again in the AI sector.
