Moonshot AI files for Hong Kong IPO at reported $50 billion valuation

Yang built Moonshot AI in Beijing after Carnegie Mellon PhD

Moonshot AI, the Beijing-based artificial intelligence lab whose Kimi family of models has gained worldwide notice among software developers, has confidentially applied to list on the Hong Kong stock exchange, The Wall Street Journal has reported. The startup was most recently valued at $50 billion in private funding, according to the Journal.

The company’s co-founder and CEO is Yang Zhilin, 33, who earned his PhD at Carnegie Mellon University before returning to China in 2023 to start the company. Yang declined to comment for the Journal’s story.

His Carnegie Mellon doctoral adviser, Ruslan Salakhutdinov, told the Journal he thought returning to China was a bad idea. “That’s not what I would have done in his shoes,” Salakhutdinov said, noting that Apple had been recruiting Yang. “He was on the trajectory upwards. That’s just not what I would have done.” Salakhutdinov, a renowned computer scientist, said that of all his Chinese national students, only two had returned to China, Yang being one of them. “He never even did a proper job search,” Salakhutdinov said.

During his doctoral work, Yang co-authored two machine-learning papers with Google Brain founding member Quoc V. Le that have accumulated more than 20,000 citations. “It’s highly unlikely for Ph.D. students to publish papers like this,” Salakhutdinov said. A fellow PhD candidate at the time, Bhuwan Dhingra, who now teaches at Duke University, recalled Yang solving problems that other researchers spent weeks working through. “He’d just go on a blackboard and write out a solution,” Dhingra said. “I still think of that a lot.”

Yang had told Dhingra during his internship that “if somebody gives him the money and the compute, he could build something that was going to beat all the models from OpenAI and Google,” Dhingra recalled. “He was very confident in his abilities.”

Yang began intensive programming training at 17 in his home province of Guangdong, where he competed in China’s National Olympiad in Informatics. He went on to Tsinghua University, often called the “MIT of China,” where he studied under Tang Jie. Tang has since founded Z.AI, which competes with Moonshot. While at Tsinghua, Yang formed a rock band called Splay — named after a computer science data structure — and co-authored machine-learning research with Tang that was presented at international conferences.

William Cohen, a Carnegie Mellon professor who co-advised Yang with Salakhutdinov, described the work as unusual for an undergraduate. “He was already pretty well-known in the community,” Cohen said. “He was sort of like this hotshot from the top school in China and was clearly a catch for whoever was able to recruit him.”

Moonshot released Kimi K3, the latest version of its flagship Kimi model, in July. Some AI researchers have described the new model as nearly as capable as U.S.-born models while costing far less. Kimi is among a wave of open-weight models from Chinese labs — meaning the model parameters are released publicly — and is generally cheaper and more customizable than the closed-source systems produced by Anthropic and OpenAI.

The growing popularity of such models has weighed on the valuations of Anthropic and OpenAI as the two U.S. firms prepare for public listings as early as this fall. Some executives and U.S. government officials have proposed bans on Chinese open-weight models, citing the practice of “distillation” — training new models on massive volumes of outputs from advanced U.S. AI systems.

Founded in early 2023, Moonshot eschews titles and hierarchy, a practice Yang sees as essential to cutting red tape. The company’s Chinese name translates to “The Dark Side of the Moon,” a reference to Yang’s Pink Floyd fandom. The startup’s roughly 300 employees sometimes wear slippers in the office and step into conference rooms named after bands and musicians including Led Zeppelin, Metallica, Queen, Prince and Bob Dylan. A leather jacket and a pair of black cowboy boots sit on an office wall display.

“I think a lot of us who did grad school in the U.S. would share this kind of openness and defiance to hierarchy and authority,” said Peng Qi, an executive at U.S. tech company Uniphore who worked with Yang on research projects during an internship at Meta Platforms.

The cultural blending goes in both directions. Many Chinese nationals are among the small group of core employees building AI systems at U.S. companies, including at OpenAI and Meta, the Journal reported. Some American startups, racing to build the next generation of trillion-dollar companies, have also embraced the “9-9-6” schedule — a grueling timetable of working from 9 a.m. to 9 p.m., six days a week — that has its origins in China.

Salakhutdinov called Yang one of the top students he has encountered. His other mentees include Jimmy Ba, a co-founder of Elon Musk’s artificial intelligence venture xAI, and Devendra Chaplot, a founding team member of Mistral AI and Thinking Machines Lab.

Yang and Moonshot declined to comment further for the Journal’s story. A Moonshot spokeswoman pointed to Yang’s comments in recent years in Chinese publications.