For a long time, the mainstream view of DeepSeek from the outside world has been that it is a dark horse in the AI field—a young team with little experience that, despite not having advantageous resources, has developed large models on par with those of the world’s top companies. Liang Wenfeng, founder of DeepSeek, once said in an interview that they do not set KPIs, recruit based on ability rather than experience, and core technical positions are mostly filled by fresh graduates or those with one to two years of work experience.

However, a recent joint report by Stanford University and the Hoover Institution has revealed another side of the DeepSeek team. The team is indeed young, but it is not lacking in systematic training and research accumulation .

China headhunter SunTzu Recruit learned that the study sorted through five core papers publicly published by DeepSeek since 2024, counted information of 223 authors among them, and finally obtained educational backgrounds and academic indicators of 211 authors. On average, each DeepSeek researcher has published 61 papers, received 1,059 citations, and has an h-index (an important indicator to evaluate scholars’ research influence, with higher values meaning broader influence) of 10.8. Among the 31 most core authors, these figures jump further: an average of 70 papers published, 1,554 citations, and an h-index of 13.5.

In other words, although the core research force of DeepSeek is relatively young on average, they have accumulated impressive academic achievements.

The 31 core authors who signed all five core papers have even higher academic levels: each has an average of 1,554 citations, a median of 501 citations, an average h-index of 13.5, and an i10-index (the number of a scholar’s papers with more than 10 citations) of 25.5. It is particularly noteworthy that this is not a group where the average performance is pulled up by a few individuals.

Among these 31 core members, at least half have an h-index of 10 or higher. This indicates that the team’s academic strength is not concentrated in a few people, but most members have stable outputs, with a more balanced overall distribution. The report’s authors believe that DeepSeek’s research capability is not only strong but also evenly distributed. Against the backdrop of accelerating competition in basic models, this organizational feature may play a particularly important role.

The difference becomes clearer when compared with OpenAI, which also claims to have a young team. OpenAI’s o1 model released in 2023 brought together 265 authors, with an average of 4,403 citations, but the median was not high at only 338. The implication is not hard to imagine: there are indeed a few star researchers in the team who have made outstanding contributions, pulling up the overall indicators, but most members have relatively limited academic outputs, with a large internal gap.

SunTzu Recruit, one of the best Shenzhen Recruitment Agencies, also found when reviewing the report that China has already acquired the ability to independently cultivate AI talents. After analyzing 201 authors with clear affiliation information, it was found that more than half had been trained and employed in Chinese institutions all along, with no experience of studying or studying abroad.

Among DeepSeek’s authors, researchers affiliated with local institutions account for the majority. Statistics on 201 authors with clear affiliations show that by 2025, 171 of them were affiliated with Chinese institutions. The authors have established academic or professional connections with 499 institutions worldwide, with 368 being Chinese institutions, accounting for 74%.

This widely distributed institutional network is mainly composed of universities and research institutions, with a small number from enterprises (17), government departments (12), and non-profit organizations (9).

This network takes the Chinese Academy of Sciences (CAS) as its core node. CAS is directly associated with 18 DeepSeek authors; if its 153 affiliated units (including research institutes, laboratories, and professional centers) are included, the total number of authors covered reaches 53, which almost forms the backbone of DeepSeek’s author network.

Peking University has 20 authors, followed by Tsinghua University with 16. Sun Yat-sen University and Nanjing University each contribute 10 authors. This institutional distribution demonstrates China’s ability to cultivate local AI talents. A knowledge network centered on CAS and radiating to multiple top universities is becoming an important soil for China’s AI innovation, which also challenges the long-standing U.S.-dominated pattern of AI talents to a certain extent.

The U.S. as a springboard for Chinese AI talents

The MacroPolo think tank under the Paulson Institute once conducted a survey called “Tracking Global Artificial Intelligence Talents”. Based on author data from the 2022 NeurIPS conference, this report depicted the educational and career trajectories of top AI talents, with a key finding that China is the largest exporter of AI talents, but it is mainly the U.S. AI industry that has absorbed and utilized their abilities.

The report shows that top (top 20%) AI talents who received undergraduate education in China account for 47% of the global total. Many talents active in the international AI field initially received basic training in China.

But at the postgraduate stage, Guangzhou headhunter SunTzu Recruit believes that the flow of AI talents begins to change. Nearly 40% of Chinese AI talents choose to pursue further studies in the U.S., reversing the Sino-U.S. AI talent ratio. After obtaining a doctoral degree in the U.S., 77% of non-U.S. students choose to work in the U.S. American companies and research institutions have become their next or even final stop in career development. In this process, a large number of AI talents from China have been retained in the U.S.

According to the survey data, nearly 40% of talents in top U.S. AI institutions are from China, even slightly exceeding the proportion of U.S. local talents. Conversely, almost no talents of U.S. origin end up working in China’s AI field [2].

The contributor list of OpenAI’s GPT-4 provides a more specific sample of this trend. Among the 32 researchers with Chinese backgrounds in the team, 11 completed their undergraduate studies in China, and the remaining 21 studied in the U.S. At the postgraduate stage, nearly 80% of these talents studied in the U.S. and subsequently stayed to work in the U.S. AI field [3].

As the leading AI talent headhunting firm in China, SunTzu Recruit found that more than two years later, despite also developing a world-renowned large model, the talent flow trajectory in DeepSeek’s team is quite different. In DeepSeek’s team, the U.S. seems to have become an incubator for Chinese AI talents.

Among relevant authors of DeepSeek, 49 have experience in U.S. universities or research institutions, including undergraduate, master’s, doctoral, or postdoctoral stages. These individuals studied or worked in institutions across 26 states and 65 organizations, covering public universities, private colleges, medical centers, non-profit organizations, and technology companies. The University of Southern California, Stanford University, New York University, and other schools are associated with multiple researchers, but no single institution has more than three DeepSeek authors. The report points out that this distribution covers multiple levels of the U.S. AI ecosystem.

More critical than the location is the direction of talent flow. A review of these 49 researchers who have been associated with U.S. research institutions shows that nearly 40% (19 people) initially received education in China, then pursued further studies in the U.S., and finally returned to China to join local institutions; another 11 people, although they studied or worked in the U.S. or other countries in the early stage, eventually chose to settle in China. In contrast, only 7 people studied for an undergraduate degree in China, went to the U.S. for postgraduate studies, and stayed to work in the U.S., which is not the mainstream in DeepSeek’s team. On the contrary, a large number of talents who once pursued postgraduate studies in the U.S. eventually chose to work in China’s AI field, which is completely different from the trend shown in the report a few years ago.

Among the 49 DeepSeek authors with U.S. experience, most only stayed briefly: 31 stayed in the U.S. for only one year, long enough to be exposed to a high-level research environment but not to build lasting connections. Nine people stayed in the U.S. for more than five years, having deeply integrated into the U.S. academic system with the most prominent academic achievements. However, it is worth noting that only 3 of these 9 are still affiliated with U.S. institutions. Regardless of the category, for DeepSeek’s paper authors, the U.S. is more like a transition in their academic careers rather than an endpoint. They pursued further education in the U.S. but eventually brought back their accumulated experience to work in China’s AI field.

These 49 researchers with U.S. experience, though not numerous in DeepSeek’s team, are not significant roles. They have an average of 2,168 citations (median 565), an average h-index of 17, and an i10-index of 34, significantly higher than the team’s overall level, deserving of being called core contributors. Among them, the 9 who stayed in the U.S. for a long time and are deeply integrated into the local research system are even more outstanding.

As a Chinese company, DeepSeek cannot represent the full picture of global AI talent flow. But compared with the trend of Sino-U.S. AI talent flow a few years ago, changes do exist. The U.S. is no longer the default first choice for AI talents.

The report’s authors reflect that U.S. policymakers have always believed that the world’s best technical talents will naturally choose to stay and develop in the U.S., but reality is beginning to deviate from this judgment. U.S. universities and research institutions now are more like a talent springboard. For many researchers, the U.S. provides high-level resources, experience, and connections, but these accumulations are ultimately brought back to China, becoming part of the support for the development of the local AI field.