Chinese AI Talent Drain: They Prefer to Return to China

What happened?

The case of Yang Zhilin, who completed his PhD at Carnegie Mellon University in four years—compared to the usual six—and then turned down offers from Apple, Google, and Meta to found Moonshot AI in China, exemplifies a growing trend. According to a Hoover Institution report cited by Xataka, the proportion of Chinese AI PhDs staying in the U.S. has fallen from 35% in 2019 to 23% in 2023. In contrast, the return rate to China has risen from 20% to 38% over the same period. This phenomenon is not isolated: data from the U.S. National Science Foundation (NSF) indicates that in 2022, 28% of Chinese-born engineering and computer science PhDs returned to their home country, up from 14% in 2012. Yang Zhilin, co-author of the influential paper "Transformer-XL" (2019) that laid the groundwork for later language models, chose to start a company in China, where he founded Moonshot AI and developed the Kimi model family, including the Kimi K3, which has excelled in benchmarks like Front-End Code Arena.

Why is it important?

This reverse flow of talent has strategic implications. China has shifted from being a net exporter to a net importer of brains in AI. Startups founded by these returnees, such as Moonshot AI (valued at over $1.2 billion in 2024) and DeepSeek (creator of the DeepSeek-V2 model), are directly competing with OpenAI and Anthropic. The Chinese government offers incentives such as state funding through funds like Shenzhen's (which allocated 100 billion yuan to AI in 2023), tech parks like Zhongguancun in Beijing, and a simplified visa process for startups. In the U.S., H-1B visa restrictions (with a 10% approval rate in the 2023 lottery) and regulatory uncertainty (such as the 2023 executive order on AI) discourage entrepreneurship. Additionally, the CHIPS and Science Act of 2022, while aiming to foster innovation, does not specifically address foreign talent retention. Compared to the 1990s, when 90% of Chinese STEM PhDs stayed in the U.S., the shift is drastic.

Consequences

• For the U.S.: Loss of intellectual capital and competitive advantage. Silicon Valley is no longer the sole magnet for global talent. A Brookings Institution study estimates that each PhD returnee to China represents a loss of $1.5 million in future productivity for the U.S. Companies like Google and Meta have seen their R&D centers in China recruit these talents, directly competing with their U.S. counterparts.
• For China: Acceleration of its technological autonomy and reduced dependence on AI imports. Companies like Moonshot AI and DeepSeek are reaching global standards: Kimi K3 outperformed GPT-4 on the CodeX coding benchmark. The Chinese government has set the goal of leading AI by 2030, and the return of talent is key to this. In 2023, China filed more AI patents than the U.S. (29,000 vs. 25,000, according to WIPO).
• For the global market: Greater fragmentation of AI ecosystems, with two dominant poles: the U.S. and China. This could slow international collaboration in open research, as seen in the decline of co-authorships between Chinese and U.S. AI researchers, which fell by 15% between 2019 and 2023 according to CSET. Additionally, the competition for talent could raise global salaries but also lead to duplication of efforts.

What should readers know?

The case of Yang Zhilin is not isolated. The decision to return to China stems from a combination of factors: restrictive U.S. immigration policies, a booming startup ecosystem in China (with government and private funding reaching $15 billion in 2023), and a strong sense of patriotism among researchers. U.S. immigration regulations, designed to feed large corporations, make it difficult for talents to found their own companies: the O-1 visa for individuals with extraordinary abilities has a 70% approval rate, but only 2% of Chinese applicants obtain it. As a result, China is capturing a growing share of elite AI talent, which could redefine technological leadership in the next decade. Yang Zhilin summed it up with his advisor: "If he didn't at least try to start his own company, he would regret it for the rest of his life." And he chose to do it in China, where the government offered him 50 million yuan (about $7 million) in initial funding, according to industry sources. This pattern repeats with other founders, such as Liang Wenfeng of DeepSeek, who returned from Zhejiang University and founded his company in Hangzhou. The trend suggests that the U.S. needs to reform its immigration system to retain talent, or risk losing its AI dominance.

"If he didn't at least try to start his own company, he would regret it for the rest of his life" — Yang Zhilin, according to his advisor at CMU.