At 27, Tsinghua’s Youngest AI Professor Raises Nearly $50M for Self-Evolving Models
When Yongchao Chen stood at a career fork in early 2026, the options looked familiar to anyone tracking frontier AI talent. On one side sat a job offer from Google DeepMind and co-founder invitations from overseas startups chasing automated science. On the other was a bet he had been waiting two years to make: return to China, build foundation models that improve themselves, and test whether research—not chat—can become the engine of the next intelligence curve.
He chose the latter. Chen, 27, founded Apex Intelligence in June 2026 and simultaneously joined Tsinghua University’s School of Artificial Intelligence as an assistant professor—described in Chinese business press as the youngest in the school’s history. By mid-September, less than three months after the company’s founding, Apex had closed angel and angel-plus rounds totaling nearly US$50 million (reported in Chinese coverage as nearly RMB 400 million).
A news peg built on recursive self-improvement
The financing was announced around Sept. 16, 2026. According to a company statement carried by PR Newswire, the angel round was co-led by IDG Capital, LinkX Capital and XtalPi, with participation from Decent Capital, SEE Fund, Monad Ventures and Winsoul Capital. The angel-plus round was co-led by Zhongguancun Science City Fund, SCGC (Shenzhen Capital Group) and Shanghai Engine Fund.
TechNode Global and FinSMEs independently reported the same roughly $50 million total. 36Kr’s English edition, citing reporting under the “Rise of Intelligence” brand, framed the raise as two successive 100-million-yuan-level rounds completed within about two months of Apex’s July public establishment announcement.
Proceeds, the company said, are earmarked for foundation-model construction, computing infrastructure, recursive self-improvement (RSI) research and core-team expansion—standard language for a research-heavy lab still early in product commercialization.
From Harvard–MIT training to a Beijing lab
Chen’s résumé is unusually dense for a first-time founder-CEO. His undergraduate path ran through the University of Science and Technology of China; graduate work included a Harvard–MIT joint doctoral program. Company materials and 36Kr report research stints at Google DeepMind, Microsoft and the MIT–IBM Watson AI Lab. Apex’s core team, according to the firm, draws from Tsinghua, Peking University, Harvard and MIT, as well as large-model groups at firms such as ByteDance, Kimi and Zhipu AI.
That pedigree matters because Apex is not pitching another consumer chatbot wrapper. Its thesis is that today’s large language models are optimized for stability and human preference—good at pleasing people, weaker at open-ended scientific exploration. Chen has argued in interviews that innovation-oriented models must tolerate failed hypotheses in bulk, then retain and reuse the rare successes. In a founder statement released with the funding news, he said models taught only by human preference ultimately lock intelligence at a human ceiling, and that “the limits of true intelligence can only be defined by the objective world.”
Early demos, still in beta
Apex says its self-evolving system has already produced research artifacts suitable for leading AI conference review and claimed advances in AI-for-math problems, including work it describes as a complete proof of a long-open majorization conjecture. 36Kr reported that in May 2026 the company let its system independently produce 34 papers submitted to ACL Rolling Review, with 11 scoring above 3 and two scoring higher than 99% of human researchers—company-reported results that have not been independently audited by outside newsrooms.
Commercially, Chen told 36Kr the product “has been launched recently and is in a small-scale beta test.” That restraint is useful context: the September funding is angel-stage capital for a June 2026 company, not a late-stage validation of revenue scale.
Why the story travels beyond China
RSI has become a crowded narrative in 2026, with high-profile overseas efforts and large checks elsewhere in the sector. Apex’s angle is to attack self-evolution at the foundation-model layer rather than only at an agent or “AI scientist” wrapper—and to treat scientific research trajectories as both product output and training fuel. Whether that converts into durable models remains an open technical question. What is already clear is the leadership peg: a 27-year-old founder-CEO who left a DeepMind-shaped path, took a Tsinghua faculty post, and raised nearly $50 million before his company turned one quarter old.
