A 36-year-old former computer-vision researcher rang the bell at the Hong Kong Stock Exchange on January 9, 2026. His company, MiniMax, had been founded less than four years earlier in a cramped Shanghai office. Few outside China had heard the name — yet the debut valued the firm above HK$100 billion.
The mathematician who left SenseTime
Born in 1989 in a county town in Henan province, Yan Junjie (闫俊杰) took a bachelor's degree in mathematics from Southeast University in 2010, then earned a doctorate in artificial intelligence from the Institute of Automation at the Chinese Academy of Sciences in 2015, followed by postdoctoral research at Tsinghua University. For more than six years he built computer-vision systems at SenseTime (商汤), the Hong Kong-listed AI firm, rising to vice president and deputy head of its research institute.
What nagged at him, he later told interviewers, was the distance between headline-grabbing AI demos — face recognition, a program that beat humans at Go — and tools an ordinary person would actually open every day. In early 2022 he left to found MiniMax (稀宇科技) in Shanghai, with a written mission of "co-creating intelligence with everyone."
Two bets that defined the company
Yan made two calls that looked odd at the time. The first was technical: he steered the team toward Mixture-of-Experts (MoE) sparse architectures and very long context windows years before they were fashionable, using a custom "Lightning Attention" layer to keep long sequences affordable. The second was strategic: build for end users directly, and go international from the first day.
That second choice gave MiniMax a different shape from most Chinese labs. Instead of leaning on enterprise sales at home, it shipped consumer products — the character companion Talkie, the video tool Hailuo AI (海螺AI), and an open platform for developers — and watched non-Chinese users accumulate. By September 2025 the company reported 212 million individual users across more than 200 countries and regions, with overseas markets supplying over 70% of revenue.
Inside the model stack
MiniMax's claim to attention is breadth. Its large models (大模型) span text, speech, video and music in a single research program. The MiniMax-01 family, released in January 2025, pairs a 456-billion-parameter backbone with 45.9 billion active parameters per token, and the company says the design lets it process extremely long inputs cheaply. Later came the MiniMax M1 reasoning model and the Hailuo-02 video model, the latter aimed squarely at creators who need generated footage rather than text.
Holding that range together is unusual: most rivals concentrate on one modality. Yan has argued that in an AGI era the product is the model, whether it emits a sentence, a clip or a voice.
Early believers and a young team
MiniMax's first outside money came from MiHoYo, the studio behind the game Genshin Impact, which also became an early customer and later put a co-founder on the board. The company's R&D team of roughly 300 includes engineers formerly at Microsoft, Google, Meta, Alibaba, ByteDance and DeepSeek. One local report described the workforce as mostly born after 1995 — a deliberately young group for a firm racing global incumbents.
The listing that reset the clock
MiniMax filed confidentially for a Hong Kong IPO in July 2025 and cleared its listing hearing in December. On January 9, 2026 it began trading on HKEX under code 0100 at HK$165 per share — the top of its range — and closed at HK$345, up about 109%. The offering raised roughly HK$4.8 billion (around US$620 million), and the debut valued the company above HK$106 billion (about US$13.7 billion).
Forbes Asia estimated the first day made the 36-year-old Yan worth approximately US$3.2 billion. The public tranche drew about 1,837 times oversubscription, and cornerstone backers included Alibaba, the Abu Dhabi Investment Authority, Mirae Asset and E Fund. "This is only the beginning," Yan said on the floor.
Why list a loss-making lab
The numbers carried a caveat. MiniMax's filing showed revenue up more than 170% in the first nine months of 2025, but also an adjusted loss near US$186 million for the period. Like peers, it is buying scale with cash. Hong Kong's Chapter 18C route, built for specialist technology firms, has become the default exit for AI companies that do not yet clear mainland profitability tests; MiniMax listed just one day after its domestic rival Zhipu AI opened on the same exchange.
What readers can do now
- If you build on AI, benchmark MiniMax's open models and the Hailuo video tool against OpenAI, Google and other domestic options before choosing a vendor.
- Treat the prospectus's user and revenue figures as engagement, not profit — the company remains loss-making.
- Watch whether the overseas share of revenue stays above 70% as non-Chinese markets get more crowded.
Honest limitations
This profile draws on the company's IPO filings, HKEX disclosures and Reuters/Bloomberg reporting; it is not an independent audit of user counts. Yan's net worth is a Forbes Asia estimate, not a disclosed number. The claim that MiniMax's multimodal models all sit in the world's first tier is the company's own positioning and was not benchmarked here against every competitor. Sources disagree slightly on the founding date (late 2021 versus early 2022); we followed the exchange documents' "early 2022" framing.