China’s AI talent competition is brutal, and most founders lose their best people to bigger wallets. Yang Zhilin, the founder of Beijing-based Moonshot AI, appears to be winning that fight anyway.
Moonshot AI has held together what observers describe as China’s strongest AI research team, a notable feat given that well-funded rivals are constantly circling with better offers. For a company that launched in March 2023 and employs around 80 people, the retention story alone is worth paying attention to.
Who is Yang Zhilin, and why does it matter
Yang Zhilin holds a PhD from Carnegie Mellon University and co-authored Transformer-XL and XLNet, two models that shaped how the industry thinks about language understanding.
He co-founded Moonshot alongside Zhou Xinyu and Wu Yuxin, and the three have built the company on a focused bet: consumer-facing AI products, with the Kimi personal assistant as the flagship.
The company raised $60 million shortly after its founding, a relatively lean war chest compared to the billions flowing into frontier AI labs in the US.
Kimi K3 and the race for open-weight dominance
The next concrete test of Moonshot’s research strength is the Kimi K3 model, expected to launch in mid-July 2026. The company is positioning it as one of the largest publicly available large language models, a category where open-weight releases have become a geopolitical signal as much as a technical one.
Open-weight models are AI systems where the underlying parameters are released publicly, allowing developers worldwide to build on top of them, which creates ecosystem momentum that proprietary models struggle to match.
If Kimi K3 delivers on its ambitions, it would put Moonshot in direct competition with leading US AI systems, a significant statement from an 80-person team operating under export restrictions and geopolitical headwinds.
US officials have raised scrutiny over Moonshot’s technology acquisition practices, including allegations that the company distilled Anthropic’s models and improperly accessed restricted Nvidia chips. Moonshot has not publicly addressed those allegations in detail.
Nvidia’s most powerful semiconductors have been subject to US export controls targeting China, meaning Chinese AI labs must either work with older hardware, source chips through alternative channels, or architect their models differently. The allegation that Moonshot accessed restricted Nvidia hardware, if substantiated, would have regulatory consequences that go well beyond reputational risk.
A $30 billion Hong Kong listing on the horizon
Beyond model releases, Moonshot is reportedly exploring a listing on the Hong Kong Stock Exchange in 2026. Early estimates put the potential valuation at around $30 billion, a number that would place it among the more significant tech debuts in the Hong Kong market in recent years.
A $30 billion valuation for a company with roughly 80 employees and no disclosed revenue figures is essentially a bet on the Kimi model family’s trajectory and on Yang’s ability to keep his research team together long enough to capitalize on it.
For investors watching the AI sector, the Moonshot listing, if it proceeds, would offer a rare public entry point into the Chinese generative AI space. Competitors like Baichuan and Minimax are building in the same space, but neither has moved closer to a public offering at this scale.