[Big read] The returnees: Inside China’s AI talent reversal

(Edited and refined by Grace Chong, with the assistance of AI translation.)

In 2023, computer science PhD graduate Pan Zizheng faced a choice: stay at Nvidia, which had just crossed a US$1 trillion valuation, or join a fledgling Chinese startup. He chose the latter without hesitation — Hangzhou-based DeepSeek.

Just over a year later, DeepSeek became the top free app on Apple’s US app store. Pan celebrated the milestone in a post reposted by Nvidia principal research scientist Yu Zhiding, who wrote: “Many of our best talents come from China, and these talents don’t have to succeed only in a US company.”

Graham Allison, founding dean of Harvard’s John F. Kennedy School of Government, posed a different question: “Why did Zizheng Pan… invest his talents in creating this remarkable advance in China rather than here in the US?”

Holding pivotal positions in China

For years, the US drew in large numbers of Chinese academics and engineers. Data from the Paulson Institute shows that around 38% of top AI researchers in the US completed their undergraduate studies at Chinese universities. In recent years, however, more individuals, as in Pan’s case, have chosen to return to China.

Last year, Wu Yonghui, former vice-president of research at Google DeepMind, joined ByteDance. Vinces Yao, who led several agent projects at OpenAI, joined Tencent as chief AI scientist.

In January this year, Zhou Hao, formerly head of reinforcement learning for Gemini, joined Alibaba.

Their trajectories are remarkably similar: undergraduate education in China, doctoral training in Europe or America, experience at major international tech firms, and then a return to China.

The number of returning AI scientists in academia is even larger. Since 2020, Chinese universities and emerging research institutes have recruited at least dozens of AI academics with overseas backgrounds.

An analysis by The Economist of authors at NeurIPS — the leading conference on neural information processing systems — found that in 2019, only 12% of Chinese AI researchers who earned graduate degrees overseas returned to China. By 2025, that figure had risen to 28%.

Several factors are driving this shift: escalating US technology restrictions on China, tighter immigration and visa policies, the rise of Chinese large language models (LLMs), the maturation of China’s AI industry, and increased investment by universities in scientific research, giving researchers more room to thrive.

Damien Ma, director of Carnegie China, told Lianhe Zaobao (LHZB): “This is a highly globally mobile talent pool. They go where the most competitive, exciting sectors are. Money matters, but so does meaningful work, and the best environment to do it in.”

Although the absolute number of returnees remains relatively small, many hold pivotal positions — senior executives, chief scientists and university professors.

A classic example is Moonshot AI Founder Yang Zhilin. After graduating from Tsinghua University, he pursued a doctorate at Carnegie Mellon University before conducting AI research at Google Brain and Meta. In 2023, Moonshot AI became one of China’s most closely watched LLM startups.

Tencent has also said that activity around its Hunyuan LLM increased significantly after Yao joined the company.

Ma noted that the vast majority of top tier AI researchers and talent are of Chinese origin, and they are now also influencing, and even leading, China’s AI development.

Who are you working for?

Dai Mingjie, an associate research professor at the South China University of Technology’s Institute of Public Policy, said cross-border talent flows have long played a major role in advancing technological progress in China, the US and the wider world. They also underpinned China’s late-mover advantage by shortening the learning curve for its technology industries.

Compared with the internet era, China has moved further ahead in AI, with talent spanning both foundational research and applications — effectively forming a de facto “G2” with the US.

Yet amid intensifying China–US technological rivalry, talent has become more strategic, and mobility is no longer purely a personal choice. Beyond salary and ideals, the question in great power competition is increasingly stark: which system are you choosing, and who are you working for?

As the US seeks to curb China’s technological development by tightening restrictions on capital flows, computing power exports and model services for Chinese tech firms, China has also stepped up scrutiny over the outflow of key technologies, data and talent.

The impact of the Manus case

The Manus case was a landmark episode. The China-founded AI company relocated to Singapore in July last year and shut down its operations in China. Late last year, reports emerged that Meta was considering an acquisition. Chinese authorities then launched an investigation into the deal, and the company’s CEO and chief scientist were reportedly barred from leaving the country.

A source close to the company told LHZB that the move to Singapore was partly driven by scrutiny from Washington following investment from US venture capital firm Benchmark, and partly by disruptions to its reliance on Anthropic’s Claude model after the company stopped serving users in China.

From the company’s perspective, this was a commercial response to US restrictions on funding and technology. But Chinese regulators saw it as Manus taking with it 40 core technical staff and related data, thereby harming national interests.

The impact spread to other AI firms. In March, MiroMind, another AI company that had relocated technical staff overseas, also received regulatory warnings.

According to Bloomberg, Shanda founder and MiroMind backer Chen Tianqiao established a “firewall” between the company’s China and global operations, banning cross-border code sharing and reducing the movement of staff and data across borders.

Chen said he had once believed it was possible to bring together Chinese and global talent to contribute to humanity’s future. But after the Manus incident, that vision of globalisation had become unsustainable. Companies now have little choice but to “pick a side”.

To access overseas computing power, MiroMind had previously arranged work visas in Japan and Singapore for some of its Chinese AI researchers. After the Manus incident, such arrangements became politically sensitive. 

It is understood that Dai Jifeng, co-founder of MiroMind and an associate professor at Tsinghua University’s Department of Electronic Engineering, left the company after opposing plans to relocate staff.

The Wall Street Journal also reported that, since early 2025, Chinese authorities have advised leading AI experts not to travel to the US, citing national security concerns and the risk of information leakage.

Talent will always be mobile

These cases have raised concerns over restrictions on talent mobility. South China University of Technology’s Dai argued that the simultaneous push to attract returnees while restricting outward flows reflects an intensifying global tech race and the rising weight of national security concerns. He noted that scrutiny of key talent and technology flows has increasingly become standard practice among major powers.

However, she stressed that China’s controls are “defensive and limited”, aimed at protecting its own critical technologies and talent rather than using extraterritorial measures to seize resources from other countries.

Carnegie China’s Ma, meanwhile, warned that attempts to control talent mobility carry risks. To retain talent and foster innovation, “carrots” are much better than “sticks”.

In his view, talent will always remain mobile. Despite the intense China-US AI competition, the talent ecosystem should not — and will not — become fully decoupled. “If there’s a full stoppage of the flow of human capital, that would be a big downside for both countries,” he noted.

He added that the Chinese and American AI ecosystems offer different things and are not directly competitive in that sense. Exchanges and mobility should continue.

“No one has figured out a better way to produce and generate sustainable creativity, good ideas and breakthrough innovations, other than having a high concentration of diverse, talented people in one place that are open to different ideas and thoughts,” he opined.

China’s 5 million AI talent gap

Behind China’s efforts to curb talent outflows lies a parallel concern over a shortage of talent. According to a report published by People’s Daily in May last year, estimates from China’s Ministry of Human Resources and Social Security suggest the country faces a shortfall of more than five million AI professionals. A survey by Beijing News of more than 100 companies last year found that nearly half of the respondents lacked multidisciplinary AI talent.

To address the shortage, local governments, companies and universities have all stepped up. Local authorities have begun targeting AI talent specifically, offering subsidies to AI startups. At the same time, leading tech firms have expanded overseas hiring efforts — at last year’s NeurIPS conference in San Diego, companies including ByteDance and Kuaishou set up recruitment booths.

Chinese universities are leveraging national programmes such as the “Excellent Young Scientists Fund Program (Overseas)”, global recruitment fairs and international young scholars’ forums to attract overseas academics back to China.

Identity and the pull of national narrative

A postdoctoral researcher at a university in Singapore told LHZB that he had already received offers from China before graduating. Leading tech firms were discussing technical management roles with annual salaries of 2 to 3 million RMB (US$300,000 – US$450,000), while universities were offering associate professorships, with the ability to supervise graduate students and build a research group.

Beyond material incentives and career opportunities, some Chinese scientists are also influenced by a less tangible pull: the force of a historical narrative.

In 2020, AI scientist Zhu Song-Chun returned to China after 28 years in the US, surprising many. Invited by the Beijing municipal government, he founded the Beijing Institute for General Artificial Intelligence while simultaneously teaching at Peking University and Tsinghua University.

According to The Guardian, Zhu once told a friend that he did not return to help China win, but because China was “giving me resources that I could never get in the United States”.

Yet his decision could hardly be separated from the wider context of national technological competition. He once wept publicly while watching a documentary about China aerospace engineer Qian Xuesen, and, as a member of the Chinese People’s Political Consultative Conference, he advocated elevating general artificial intelligence (AI) to the same strategic importance as the “Two Bombs, One Satellite” programme. In an interview with CCTV, he said that he “could never forgive myself” if the country needed him and he turned his back and said no.

Lin Shen, who completed a PhD in AI at Singapore’s Nanyang Technological University (NTU), told LHZB that appeals to overseas talent’s sense of belonging to China — and the framing of scientific achievement within a broader historical narrative — can also be a draw. He argued that this reflects a distinctive approach China takes toward Chinese scientists, whereas in the US, there is comparatively “much less emphasis on a sense of national or cultural attachment.”

In addition, China’s broader social environment also contributes to its appeal. Lin said that, under government-led guidance, Chinese society is highly receptive to AI, with all industries engaging with AI and offering substantial scope for real-world applications. By contrast, in the US, there is considerably more public debate over AI and its implications for the public interest. 

“In the US, government policy plays a relatively smaller role; the market’s appeal lies in its free-market system, access to capital, and innovation environment,” Lin noted.

Can returning talent stay for good?

Whether returning talent can remain in China over the long term is still shaped by personal, environmental and cultural factors.

In 2020, Wu Yi, a former OpenAI researcher involved in the development of several major projects, returned to China and joined Tsinghua University’s Institute for Interdisciplinary Information Sciences as an assistant professor. Together with three other young scientists from the University of California, Berkeley specialising in deep reinforcement learning and embodied AI, they became known as the “Berkeley Four”.

In mid-April this year, Wu left for Meta’s Superintelligence Lab, and Tsinghua University removed his faculty profile from its website.

It was understood that Wu’s departure from Tsinghua University was far from smooth. Discussions between the two sides reportedly went on for several months, as Tsinghua was reluctant to see him leave. During the handover process, negative feedback regarding Wu’s teaching also emerged, further fuelling the university’s dissatisfaction.

Sources said his decision was driven not only by Meta’s attractive offer, but also by difficulties adapting to the environment at Tsinghua.

Silicon Valley continues to attract top talent with its high salaries and cutting-edge research opportunities, and Wu’s adaptation difficulties are commonplace. Returnees who come back to China with working habits shaped by overseas research environments often find themselves navigating a very different system of evaluation and culture of collaboration.

NTU’s Lin said that Chinese universities invest substantial resources in recruiting outstanding overseas talent, and therefore place correspondingly higher demands on publishing papers and producing results, making the environment “far more competitive”.

Common difficulties, he said, include intense involution, complex working relationships and the challenge of finding like-minded collaborators. “For example, if a returning professor wants to pursue the most cutting-edge research, but their graduate students only want a degree, collaboration becomes difficult,” Lin noted.

South China University of Technology’s Dai noted that although China has made substantial efforts to attract talent, it still lags behind the US in areas such as commercialising research outcomes and fostering an innovative environment.

She said that, compared with the ecosystem in Silicon Valley, the “revolving door” between Chinese universities, research institutes and companies remains far less fluid. There is still a clear divide between organisations inside and outside the state system, making talent mobility and the commercialisation of research outcomes more costly. 

In addition, the emphasis on “formalism” and KPI-driven assessments can, over time, erode researchers’ creativity and willingness to take risks.

However, Dai also stressed that China is attempting reforms at the policy level. Some universities have piloted efforts to “break the five-only criteria”, moving away from an assessment system overly focused on “papers, titles, academic qualifications, awards and ‘hats’ (prestigious labels)”. National funding programmes have also increased support for young talent and disruptive innovation. “However, fully changing longstanding institutional habits will still take time,” she said.

This article was first published in Lianhe Zaobao as “中美科技竞争白热化 AI精英携西经东归 旋转门恐难再流转”.

Related: How Chinese talent powers US AI dominance | Who steers AI: China’s industrial state vs America’s frontier builders?