Tsinghua Yao Class—after 20 years, it has trained only 700 people, yet it has precisely occupied half of the world’s AI?

Chen Lijie works at OpenAI on mathematical reasoning, Ding Liyu is on the founding team at xAI, and DeepSeek’s core R&D also traces back to the Yao Class.
In China, it’s even more absurd: the founders of Pony.ai and Megvii are all classmates from the same program.
The senior brother leads the charge in Silicon Valley to tackle foundational theories, while the junior brother carries the banner in Shenzhen to push the implementation of complex systems—this kind of “cross-ocean resonance” is basically a nightmare-level scenario.
2005: Academician Yao Qizhi returned to China to found the program, building a top-tier local academic ecosystem.
2015: Unicorns such as Pony.ai and Megvii were born from the Yao Class. 2025: top-tier talent accelerates its return to China, creating a talent closed loop to repay and strengthen domestic development.
2026: a rumored internship “special offer” with a daily wage of 5,500 yuan reflects an intense demand for foundational-level talent.
I think “half of the world’s AI” is a bit exaggerated, but the Yao Class’s self-sustaining capacity built on foundational algorithms, mathematical theories, and system architecture is precisely the solid backbone behind China’s technological breakthroughs—unafraid of being held back by hardware bottlenecks.