Four years after a group of former SenseTime researchers opened a Shanghai office, their company rang the bell at Hong Kong's exchange. The stock didn't just list — it more than doubled on day one. The open question now is whether a model-maker can stay profitable once the debut hype cools.
The company is MiniMax (稀宇科技), and its January 9, 2026 listing on the Hong Kong Stock Exchange turned a quietly global AI lab into a public one. It followed Zhipu (智谱) to the market by a single day, part of a sudden cluster of Chinese large-model companies seeking public capital.
From SenseTime spin-off to HKEX
MiniMax was founded in 2021 by Yan Junjie (闫俊杰), a former SenseTime vice president born in 1989. The early backer was MiHoYo (米哈游), the studio behind Genshin Impact, which shaped MiniMax's consumer instinct. Later investors stacked up: Alibaba, Tencent, Xiaohongshu (小红书), Hillhouse (高瓴), IDG, Sequoia (红杉), Matrix (经纬) and others.
The firm committed early to a full-modality strategy — text, video, speech and music — rather than betting on one. Its product line is unusually broad for a lab its age:
- MiniMax M2, the large language model.
- Hailuo (海螺), a text-to-video generator that became one of the more capable consumer video models out of China (upgraded as Hailuo-02 / Hailuo 2.3).
- Speech-02, a voice model the company says has helped users generate more than 220 million hours of audio.
- Music 2.0, a music-generation model.
- Talkie / 星野, an AI-companion app, and MiniMax Agent, a general assistant.
The numbers that landed
The listing itself was the story. Key figures, drawn from the prospectus and exchange filings:
- Stock code 0100.HK, priced at the top of a HK$151–165 range, at HK$165 per share (≈ US$21).
- Base offering of about 25.4 million shares, expanded with a 15% over-allotment to 29.2 million; with the full green shoe, about 33.6 million shares. Gross proceeds: HK$4.82 billion (≈ US$617 million) base, up to HK$5.54 billion (≈ US$709 million) with the green shoe.
- Debut: opened up 42%, then closed the first day up about 109% at HK$345, implying a market capitalization near HK$107 billion (≈ US$13.7 billion).
- Retail oversubscription of 1,837 times; more than 420,000 individual applicants.
- 14 cornerstone investors committing about HK$2.723 billion (≈ US$349 million), including ADIA (Abu Dhabi Investment Authority), Aspex, Eastspring, Mirae Asset, Alibaba and E Fund (易方达).
Revenue, while small, is growing fast off a low base: US$3.46 million in 2023, US$30.5 million in 2024, and US$53.4 million in the first nine months of 2025 — increases the prospectus puts at 782% year on year for 2024 and 175% for the 2025 period. The company is not yet reporting net profit in its public materials.
Why Hong Kong, and why now
As one capital-markets observer cited by Chinese financial media put it, AI firms chase the Hong Kong window partly because they don't yet meet A-share profitability thresholds, and partly to lock in funding while valuations are hot. MiniMax said about 90% of proceeds go to R&D over five years — roughly 70% to model development and 20% to AI-native products — with the rest for working capital.
The "model-is-product" philosophy is the through-line. A MiniMax public-affairs director told Jiemian (界面) the company deliberately avoids betting everything on a single "Super App," arguing its edge is the model capability itself: "in the honors class of large models, everyone needs a specialty." That framing — ship frontier-adjacent open weights at a fraction of US pricing — is what separates MiniMax from app-first peers.
A crowded IPO window
MiniMax was not alone. Zhipu (智谱, code 02513) listed on HKEX on January 8, 2026 at HK$116.20, closing its first two days around HK$158.6 for a market cap near HK$70 billion (≈ US$9 billion), and billed itself as the "first large-model stock." Moonshot (Kimi) has also been preparing a Hong Kong listing. The cluster signals a structural shift: China's leading model labs are maturing past venture funding and into public markets, with Hong Kong as the preferred venue.
MiniMax's global posture helps the narrative. It built overseas consumer hits (Hailuo, Talkie) early and pitches a "born global" strategy — useful when courting international cornerstone money like ADIA and Mirae Asset.
Honest limitations
- Valuation rests on growth, not profit. A ~US$14 billion market cap on sub-US$100 million trailing revenue is a bet on future model leadership, not current earnings. The 109% first-day pop says as much about supply scarcity as about fundamentals.
- Revenue base is small and concentrated. Triple-digit percentage growth looks dramatic off a US$3.5 million starting line; sustainability depends on enterprise and overseas adoption that is not yet demonstrated in filings.
- Founding date varies by source. Some filings cite early 2022; the company and several reports date it to 2021. The "four years to IPO" framing holds either way, but the exact date is unsettled.
- Competition is fierce on price. MiniMax's low-cost open-weight model is also DeepSeek's, Qwen's and Kimi's strategy; differentiation on price alone is hard to defend.
What readers can do now
- Read the prospectus, not the headlines. MiniMax's HKEX filings (code 0100) show revenue trajectory, investor roster and use-of-proceeds — the real test of the "model-is-product" claim is whether R&D spend converts to retained users.
- Benchmark the products yourself. Run the same prompt through Hailuo video, Talkie and the MiniMax API, and compare cost-per-token against DeepSeek and Qwen before assuming MiniMax's "cheap frontier" edge holds.
- Track the HK AI-lab cluster as one signal. With Zhipu, MiniMax and (expected) Moonshot all listing, watch whether post-listing disclosures show a path to profit or just a longer runway — that tells you more than any debut pop.