Inside China’s 'genius' pipeline powering its AI surge: FT
An inside look at how intense competition, early selection, and massive state investment are producing China’s next wave of scientific and AI leaders.
-
The icon for the smartphone app DeepSeek is seen on a smartphone screen in Beijing, Jan.28, 2025 (AP)
China’s strategy for cultivating elite scientific talent is already delivering results, the Financial Times argued in an in-depth feature on the country's scientific talent pipeline.
A vast, highly competitive system embedded in top high schools has become a critical engine behind the country’s rise in advanced technology, particularly artificial intelligence.
In late 2022, amid Beijing’s strict COVID-19 lockdowns, Stacey Tang, a pharmaceutical company manager, received an unexpected phone call from an unidentified landline. She was instructed to send her 15-year-old son to sit an entrance test for a so-called “genius class” at one of the capital’s most prestigious high schools.
The testing arrangement was unusual even by China’s exacting standards: the exam would take place inside a moving van that would drive through the city for an hour while the teenager tackled university-level mathematics problems.
'His golden ticket'
While some parents might have hesitated, Tang did not. “In any other country, you would immediately suspect an abduction plot or simple lunacy,” she said, smiling over a latte.
“Instead, I was weeping with joy, and sent my boy right away. I understood this for what it was: his golden ticket to the best educational resources in China,” she said.
Her son was one of roughly 100,000 gifted students selected annually to join science-focused elite tracks, commonly known as genius, experiment, or competition classes, within China’s top secondary schools.
An assembly line for scientific elites
These programs are designed to prepare students for international competitions in mathematics, physics, chemistry, biology, and computer science. Tang herself followed this path nearly three decades ago in Chengdu, which helped her secure a place at Peking University and later a well-paid career in Beijing.
Over the years, these classes have produced a disproportionate share of China’s scientific and technological leadership. Their influence is deeply embedded in the firms now challenging US dominance in AI, robotics, and advanced manufacturing.
From Olympiads to tech titans
Graduates of the genius system include the founder of ByteDance, the architects of TikTok’s recommendation algorithm, the leaders of Taobao and PDD, and the billionaire behind Meituan. The brothers who founded Cambricon, a major Chinese rival to Nvidia, also emerged from this stream, as did engineers behind large language models at DeepSeek and Alibaba’s Qwen, and Tencent’s new chief scientist, recruited from OpenAI.
China’s approach differs sharply from Western talent pipelines. It operates at a massive scale and is driven by the state, as per the feature.
According to state media outlet Xinhua, China produces around five million STEM graduates each year, ten times the number in the US.
Tens of thousands of these students are drawn from regular classes between the ages of 16 and 18 for intensive training. Unlike their peers preparing for the grueling gaokao university entrance exam, some are able to bypass it entirely by securing early admission through exceptional competition results.
'World-Class' talent
When Nvidia CEO Jensen Huang described Chinese AI researchers as “world-class", he was likely referring to this cohort. “You walk up and down the aisles of Anthropic or OpenAI or [Google] DeepMind,” Huang said last May, “there’s a whole bunch of AI researchers there, and they are from China . . . They are extraordinary and so the fact that they do extraordinary work is not surprising to me.”
Western observers were stunned in early 2025 when DeepSeek released its R1 language model, which matched leading global systems at a fraction of the cost. Many questioned how a relatively small Chinese team could challenge American AI supremacy. The genius-class system offers a large part of the explanation.
Inside DeepSeek
Wang Zihan joined DeepSeek as an intern in 2024 at the age of 21, unaware that the company was on the brink of reshaping global AI perceptions. At the time, prevailing views in Washington and Silicon Valley held that US export controls had stalled China’s AI progress and that Chinese firms were merely copying Western models.
Wang worked on DeepSeek’s V2 model, which later underpinned R1. Unlike closed systems developed by OpenAI, DeepSeek published its entire development process and made the model freely available.
DeepSeek stood out for its largely domestic workforce. Its founder, Liang Wenfeng, emphasized this point in a rare 2024 interview: “We want to grow our own top talents, otherwise China will always be a follower.”
Wang described the company culture as unusually open. “No KPI [key performance indicators], no hierarchy, no one at your back, and endless resources for you to experiment new ideas,” he said. Most of his colleagues were graduates of genius classes, many from Tsinghua, Peking University, and Zhejiang University, and almost all had won medals in international competitions.
Wang entered a genius class at a top high school in Wuhan, one of China’s most competitive education environments. “The education I had growing up was extremely hard, but pressure and cut-throat competition makes one learn the best,” he stressed. “You feel like, after that, there’s no challenge in the world I can’t take on.”
Humanities in AI
Unlike many peers, Wang retained a strong interest in the humanities, which he believes later benefited his AI work. DeepSeek used human experts, known as Baixiaosheng, to train the model in culturally specific knowledge such as feng shui. Although the company has never officially confirmed this, observers believe it contributed to the model’s superior performance in these areas.
Wang later left DeepSeek to pursue a PhD at Northwestern University. He remains uncertain about staying in the US, citing visa rejections faced by other Chinese students.
“More Chinese students, who make up about half of science major PhDs here, are considering going back now because of the uncertainty,” he said. “If you have to live with the risk of getting kicked out any day, it’s too much pressure.”
“Plus, China is doing really well,” he added.
Roots in revolutionary ambition
China’s emphasis on science education dates back to the post-war era. By the 1980s, slogans such as “Produce talent quick and early” adorned education offices. Compulsory schooling expanded, while elite schools began grooming top students for international scientific competition.
China first entered the International Mathematical Olympiad in 1985, returning with a single bronze medal. The following year, a full team won three bronzes, generating national excitement and prompting schools to invest scarce resources in elite competition classes.
The strategy proved effective. By 2025, Chinese teams sent 23 students to international Olympiads, 22 returned with gold medals.
Reforming the path to university
University admissions reforms in the 2000s allowed competition winners to bypass the gaokao and gain direct entry to top institutions under the 985 Project. This incentive drew more students into the genius track, where focus narrowed to competition subjects at the expense of broader curricula.
As participation grew, parental backlash followed. Only about 3% of students ultimately secured direct admission, leaving others scrambling to prepare for the gaokao with little time left. In response, curricula were adjusted, and admission policies tightened.
Interest has increasingly shifted toward computer science. Informatics competitions have overtaken physics and mathematics, and AI has accelerated this trend. In 2017, China declared AI a “key national growth strategy,” and within a year, dozens of AI-branded classes were launched.
The Yao class
Among the most prestigious programs is Tsinghua University’s Special Pilot Class for Computer Science, known as the Yao Class after Andrew Yao, China’s only Turing Award winner. In 2004, Yao left Princeton to build a world-class talent hub in Beijing.
By 2018, he declared success, “My goal has been achieved . . . I think our students are actually better now [than those from the top US schools].”
Lou Tiancheng, co-founder and CTO of Pony.ai, was among the Yao Class’s earliest students. A former informatics Olympiad gold medalist, he said: “I had no hesitation, because of Professor Yao . . . I wanted to learn from the best and with the best.”
Lou credits the genius system for fostering independent thinking rather than rote learning. This mindset later shaped Pony.ai’s strategic pivot toward self-learning autonomous systems. “We were, and still are, determined that this is the right path to ultimate intelligence in autonomous driving,” he said.
The lasting trade-off
In the end, he chose journalism over science, diverging from many peers who went on to senior roles in technology, finance, healthcare, and academia.
China’s genius system has undeniably strengthened the country’s technological capacity. Yet, on a personal level, its value remains ambiguous. While much technical knowledge fades, what endures is “the curiosity to question, the discipline to reason and the courage to face the unknown.”
Talent as China’s edge
Dai Wenyuan, a former genius-class student and founder of AI firm Fourth Paradigm, sees talent as China’s decisive advantage. “There have been more than 1,000 registered generative AI models [in China], which is unthinkable elsewhere,” he said.
Reflecting on two decades of change, Dai added, “I’ve witnessed firsthand how China has grown from having zero AI talent 20 years ago to mass producing them… The real geniuses to change the world soon could well be among them.”