Just three years ago, outside the realm of artificial intelligence research, very few knew about Yang Zhilin. Today, he is running a company valued in the hundreds of billions of dollars, has just introduced one of the most competitive AI models on the market, and has become one of the leading figures embodying China’s rise in the race for large language models.
What’s remarkable is that Yang is not a former tech executive turned entrepreneur. He belongs to a new generation of founders who initially helped develop the technology and then decided to create companies that compete directly with their former employers.
The key points about Yang Zhilin and Moonshot AI in 20 seconds
- Yang Zhilin worked in AI research before founding Moonshot AI in 2023.
- He contributed to scientific works like Transformer-XL and XLNet, landmarks in the evolution of large language models.
- His company just launched Kimi K3, one of the most competitive open models on the market.
- Moonshot AI is already valued over $20 billion and is preparing for a future IPO.
- His story reflects how China is shifting from following Silicon Valley to competing head-to-head with it.
For years, artificial intelligence was dominated by a small group of US labs. Google Brain, OpenAI, Meta AI, Microsoft Research, and DeepMind housed many of the world’s most influential researchers.
Now, that landscape is beginning to change.
From lab to market
Yang’s story echoes that of other names leading some of the industry’s biggest companies today.
Ilya Sutskever transitioned from heading OpenAI’s research to founding Safe Superintelligence (SSI).
Dario Amodei left OpenAI to create Anthropic.
Alexandr Wang built Scale AI at just twenty years old.
Yang Zhilin exemplifies this phenomenon from China.
After studying at Tsinghua University and earning his PhD from Carnegie Mellon University, he worked at both Google Brain and Meta AI, where he contributed to research on Transformer architectures.
His name appears as an author on two highly relevant publications for understanding the evolution of modern models: Transformer-XL and XLNet.
Although ChatGPT popularized large language models, many of the techniques used today were developed years earlier in academic labs like those where Yang worked.
Moonshot AI is growing at a pace that’s hard to ignore
Yang founded Moonshot AI in March 2023, just months after ChatGPT’s global debut.
While many startups focused on building applications using existing models, Moonshot chose to develop its own foundational models.
The first major success came with Kimi, an assistant notable for offering especially wide context windows when most competitors were still limited in this area.
The company has continued to grow rapidly.
In May, it closed a round close to $2 billion, boosting its valuation above $20 billion and establishing itself as one of China’s leading private AI labs.
Several reports also indicate that it’s preparing for a future IPO in Hong Kong.
Kimi K3 aims to compete where it matters most
The launch of Kimi K3 is not just about improving benchmarks.
The strategy appears much more ambitious.
Moonshot champions an open-weight model, enabling companies and researchers to deploy it on their own servers—a growing trend among organizations seeking to cut costs or keep control over their data.
They also offer highly aggressive pricing.
In an era where the cost of running large language models has become a major challenge for many companies, providing cheaper inference can be as crucial as gaining slight improvements on benchmarks.
It’s no coincidence that more and more companies are starting to evaluate Chinese models as alternatives to GPT, Claude, or Gemini for specific use cases.
The new generation of researchers no longer needs Silicon Valley
Perhaps the most significant shift isn’t technological but geographic.
A decade ago, the typical path for a Chinese researcher was to study in the US and build their career there.
Today, many still follow that route… but then return home.
China now offers funding, infrastructure, computational capacity, and a large enough market to compete with U.S. giants without leaving the country.
Moonshot AI, DeepSeek, MiniMax, Zhipu AI, and Baichuan represent this new wave of labs that are no longer just adapting Western technologies but developing their own models capable of competing on equal footing.
Yang Zhilin exemplifies this transformation better than anyone.
Not only because he leads one of the fastest-growing AI startups globally, but because he also helped author some of the research that made current models possible and now competes directly with his former colleagues.
The names shaping the future of AI
| Researcher | Current Company | Trajectory |
|---|---|---|
| Sam Altman | OpenAI | Entrepreneur and founder of OpenAI |
| Dario Amodei | Anthropic | Former OpenAI researcher |
| Ilya Sutskever | Safe Superintelligence | Co-founder and ex-chief scientist of OpenAI |
| Demis Hassabis | Google DeepMind | Co-founder of DeepMind, Nobel Prize 2024 |
| Alexandr Wang | Scale AI | Founder of Scale AI |
| Yang Zhilin | Moonshot AI | Ex-Google Brain, Meta AI researcher and author of Transformer-XL and XLNet |
Frequently Asked Questions
Who is Yang Zhilin?
He is the founder and CEO of Moonshot AI, a researcher specialized in artificial intelligence, and co-author of key scientific works such as Transformer-XL and XLNet.
What is Moonshot AI?
It is one of China’s leading AI laboratories, creator of the Kimi assistant and the Kimi K3 model.
What sets Kimi K3 apart?
It combines competitive performance with an open-weight strategy and lower prices than many Western commercial models.
Why is his story important?
Because it reflects a shift in trend: some of the researchers who helped build modern AI are no longer working for U.S tech giants but are establishing their own labs to compete with them.
