One of the most dramatic talent moves in AI is unfolding: Nathan Chen (Chen Guangyu), a 17-year-old high-school graduate who served as co-first author on core architecture research for the Kimi family of large models, confirmed on October 15 that he has left Beijing-based Moonshot AI. Multiple reports point to his next destination: Anthropic, the company behind Claude, in Silicon Valley. At the height of the US-China AI race, a teenager who has not yet started college has become a focal point of contention between the two camps — and the latest footnote in the “AI talent war.”
A Farewell Post: “The Best Time of My Life So Far”
“The past year working at Kimi was the best time of my life so far,” Chen wrote in his departure post, thanking the team for trusting and hiring him when he was just 16. He did not reveal his destination, saying only that he hopes to keep exploring new directions. His short career began at a middle-school AI hackathon hosted by Moonshot in February 2025, where his technical posts on social media earned him a Silicon Valley startup internship; that November, he formally joined Moonshot and dove into the foundational architecture of Kimi K3.
Rewriting Model Architecture at 16: Attention Residuals
At Moonshot, Chen worked alongside senior researchers Su Jianlin and Zhang Yu as one of three co-first authors of the “Attention Residuals” paper. The work restructures the residual-connection design that deep learning has relied on for nearly a decade: experiments showed that on the Kimi Linear 48B model, the new architecture delivers gains equivalent to roughly 25% more compute, while inference latency increases by less than 2% — a near-free boost in capability.
Twice Praised by Musk: Kimi K3’s 2.8 Trillion Parameters
When the paper dropped in March, Tesla CEO Elon Musk publicly called it “Impressive work from Kimi” on X. In July, Kimi K3 — built on that architecture — launched with 2.8 trillion total parameters, the world’s first open-weight model above the 3T scale, drawing another “Impressive” from Musk under a related benchmark post. A high-school researcher’s name was thus written into the foundation of one of the world’s most-watched open-source models.
Next Stop Silicon Valley? A Personal Invite from Sholto Douglas
According to screenshots circulating online, Sholto Douglas, Anthropic’s reinforcement-learning lead, invited Chen on X to join his team in Silicon Valley; Chen reportedly replied confirming the move, in the direction of Scaling RL. Neither Anthropic nor Chen has publicly confirmed the rumor, but multiple sources point the same way. Born in Shenzhen in 2009 and fresh out of high school, Chen would begin the next chapter of his AI career in the San Francisco Bay Area at 17.
The Ironic Footnote: The White House Just Accused Moonshot of “Stealing”
The timing is delicate. In July, White House tech policy adviser Michael Kratsios publicly accused Moonshot of “stealing” Anthropic’s capabilities while developing Kimi K3. Three months later, Anthropic is reportedly recruiting a core young researcher from Moonshot itself. Commentators note that model weights can be copied and architectures re-implemented, but the flow of top talent is the true scoreboard of the AI race. Researcher Nathan Lambert recently argued that Chinese AI labs are now talent-equal with OpenAI and Anthropic — the gap is compute, and it is shrinking fast.
Conclusion: Credential Barriers Are Collapsing
What makes Chen’s story striking is not which company he left or joined, but the proof it offers that credential barriers are eroding in the AI era: a teenager with no university degree can become the object of a bidding war between the world’s top labs on the strength of a single paper. Moonshot’s newly launched “Wild Card” hiring program is designed precisely for such unconventional talent. For the next generation of AI practitioners, portfolios now speak louder than diplomas; for the US and China, whoever retains the next Nathan Chen gains an edge. The talent war ignited by a 17-year-old has only just begun.
