The U.S. Better Treat Chinese Scientists Well, or Its AI Potential Will Collapse
According to a report by Nikkei Asia, an analysis from MacroPolo, a think tank under the Paulson Institute in Chicago, highlights that nearly 40% of the top AI talent in U.S. companies and research institutions comes from Chinese universities. The percentage of graduates from Chinese universities even surpasses those from American universities, including esteemed individuals like Stanford University professor and World Labs co-founder Fei-Fei Li, as well as Microsoft’s Chief Scientist in AI, Li Deng.
At the press conference, Elon Musk gave center stage to two Chinese scientists who are primarily responsible for the model research. One of them is Wu Yuhai, a post-95 academic prodigy from Hangzhou. Wu studied machine learning at the University of Toronto, was a student of Geoffrey Hinton, the “father of deep learning,” and interned […]
According to the report, in 2019, 27% of top AI talent in the U.S. originated from Chinese universities. By 2022, this proportion rose to 38%, surpassing the 37% from American universities. These talents complete their undergraduate studies in China before pursuing graduate programs in the U.S. and eventually finding employment there.
MacroPolo conducted background research on authors of papers presented at the prestigious NeurIPS AI conference in 2022, discovering that 7 out of the top 10 institutions employing these AI talents are in the U.S., including Google and Stanford University, with China closely following. NeurIPS has been increasingly accepting papers from China, with Tsinghua University and Peking University making it to the top 10.
If a significant number of Chinese AI experts in the U.S. were to flow back to China, it could dramatically shift the balance in AI expertise between the two countries. However, MacroPolo suggests that the leading IT companies and research institutions in the U.S. continue to attract Chinese AI talent. Currently, about 80% of foreign doctoral graduates choose to remain in the U.S. for work.
Interestingly, according to the 2024 Nature Index report by Springer Nature, which ranks research institutions, Chinese institutions tend to favor independent research over international collaborations. Despite the rapid growth of China’s independent research output, it remains relatively weak in global cooperation.
This observation contrasts sharply with the perception that much of China’s research accomplishments stem from collaborations with foreign institutions.
Combining data from both aspects reveals intriguing insights.
It’s long understood that the role of Chinese talent is pivotal in the U.S. tech sector, but the 40% figure exceeds expectations. Indeed, MacroPolo’s data show only 38%, but considering historical trends, it’s a conservative estimate that this figure has likely reached 40% by now.
Losing this 40% would significantly impact the U.S.’s AI advantages (if any remain). Conversely, gaining this 40% would render China’s AI dominance unshakable. These are just the short-term impacts. In the long term, as AI rapidly develops, the current AI academia might amass the richest practical experiences, which will be crucial for training the next generation of AI talent. This flow of 40% determines the future of AI expertise.
It’s worth noting that among the 37% from American institutions, a significant portion likely includes second and third-generation Chinese-Americans, as well as international students from China pursuing undergraduate studies.
In the film A World Without Thieves, Uncle Li remarks: “What is most valuable in the 21st century? Talent!”
For AI, talent really is the key.
This underscores the counterproductive nature of initiatives like Trump’s China Initiative 1.0, which effectively harms U.S. interests more than China’s.
This generation of Chinese expatriates must also connect with the broader era they inhabit, and their stay or departure intertwines with U.S. policies toward China. They left China during the era of reform and economic boom, driven by academic and career aspirations rather than political motivations.
Naturally, they maintain emotional and academic ties with classmates, teachers, and family back home. As the U.S. attempts to contain China’s AI progress, these 40% constitute a natural “loophole.” The rapid economic and scientific development in China provides fertile ground for research, further motivating them to sustain and even reinforce ties with China. This is not an act of betrayal. Academic pursuits know no boundaries, and politicizing academia is the true betrayal. In the event of a U.S.-China hot war, their loyalty would inevitably be questioned. If anti-China sentiment in the U.S. escalates, they have no choice but to return to China.
In fact, returning to China during their career peak isn’t a “career downgrade.” The U.S. “Big Leap Forward” AI route, proven inefficient and unsustainable by DeepSeek, necessitates a new path. Chinese-style AI, characterized by open-source innovation, application-orientation, low-cost training, and lightweight deployment, is both pragmatic and expansive in its potential, with no better place to embrace it than China.
Furthermore, as AI developments rise and fall, the current 40% is vital, but the future even more so. Chinese university graduates form a “renewable resource.” The U.S. has already tightened restrictions on Chinese students, with Campbell advocating for Chinese students to study more Shakespeare and less STEM, creating policies to support this stance, including stricter visas or entry denials.
This raises another issue: Can these fresh “Chinese academic seedlings” flourish in China? Examples from DeepSeek indicate they can. Data from Springer Nature also shows that much of China’s scientific achievements are blooming independently from within its robust and healthy research ecosystem, bolstered further by barriers set by Western governments.
This suggests that, in the context of AI talent, the U.S. might experience a double loss, while China stands to gain doubly. This double loss or win truly indicates losing or winning twice.
If someone examines the broader STEM talent pool in the U.S. beyond AI, similar issues would likely arise.
Some might point out that Indian Americans also constitute a significant portion of the U.S. STEM talent pool. The U.S. is encouraging the replacement of Chinese students with Indian students. However, there are two issues:
1. While Indians excel at leadership, how many remain in STEM? Observe university STEM faculties, and you’ll find many older Indian academics, with younger generations largely replaced by Chinese.
2. Indians lack a unique academic ecosystem of domestic interaction, so their success in the U.S. is often confined to the local talent framework.
Chinese students have historically achieved great success in STEM, even in the era of pioneering figures like Qian Xuesen and Yang Zhenning, albeit as rare stars. In the past two decades, Chinese students have achieved significant success, closely linked to China’s rise, warranting in-depth analysis. In comparison, Indian students are yet to reach a similar stage, despite not having their equivalents to Qian Xuesen or Yang Zhenning.
Editor: Zhongxiaowen
