Global old-money bets on MiniMax! Why does a team of 385 people challenge OpenAI?

MiniMax listed on the Hong Kong Stock Exchange on January 9th under stock code 0100, with a valuation ranging from HKD 46.1 billion to HKD 50.4 billion. The team of 385 members has an average age of 29, attracting 14 cornerstone investors who subscribed for HKD 2.723 billion. Having burned only USD 500 million, it has achieved industry-leading results across all modalities, with the text model M2 ranking in the top five globally, and speech models outperforming OpenAI.

Why do seasoned investors choose young MiniMax?

In the stock market, “old money” refers to institutional funds with long investment horizons that emphasize fundamentals, including public funds, sovereign wealth funds, and strategic holdings companies. If venture capital is about betting that one in ten projects will become a unicorn, then old money is adept at giving enough patience,坚持复利和长期主义。他们关注股息、获利增长以及企业长期价值，投资门槛极高，更关注企业健康发展，相信技术壁垒能带来数十年的垄断收益。

In 1988, Buffett’s Berkshire Hathaway made a large position in Coca-Cola, peaking at about 9%, and never traded based on short-term fluctuations. Later, major U.S. pension funds and index-based long-term capital followed suit. They focus on strong, healthy cash flow, predictable market demand, and clear pricing power. Long-term funds are seeking reliability that can be repeatedly validated over time, especially companies that can embed technology as an irreplaceable production factor. At the next wave of production factor replacement, these long-term funds have identified AI, and within this domain, they have chosen MiniMax.

MiniMax’s financial data shows a unique growth curve: revenue increased by 174.7%, R&D expenditure grew only 30%. This asymmetric growth conceals its core survival logic and is also key to attracting old money. The prospectus shows that USD 1.1 billion in cash reserves can support 53 months of operations, accounts receivable turnover days are 38 days (below the industry average of 60-90 days), B2B gross margin is 69.4%, and revenue sources include subscription and API models that are sustainable. These data points form a perfect match with the investment logic of long-term funds: high moat, sustainable cash flow, and cycle resistance.

Self-developed multimodal robustness verification

“A truly powerful system does not stay unchanged under impact, but benefits from chaos and becomes stronger under pressure.” This is Nassim Taleb’s concept of antifragility. MiniMax embodies this philosophy.

Three years ago, MiniMax chose to self-develop all modalities as a bold move. At that time, industry consensus was to focus on a single track because multimodality implied extended frontlines and dispersed resources. But MiniMax persisted in self-developing all modalities, now becoming one of only four companies worldwide capable of leading in multimodality, achieving comprehensive breakthroughs in text, video, speech, and music.

Technological breakthroughs in four major modalities:

Text models ranked in the top five globally: M1 is the world’s first open-source large-scale hybrid attention inference model, and after the second-generation M2 was open-sourced, it secured a top five global ranking and became the leading open-source model. On OpenRouter, domestic token usage peaked, programming scenarios entered the top three worldwide, and major cloud providers like Google, Amazon, and Microsoft integrated it immediately. The WebDev leaderboard from LMArena AI shows M2.1 Preview reaching fourth place globally and first domestically, only behind Anthropic, OpenAI, and Google.

Speech models outperform OpenAI: The newly released Speech 2.6 directly surpasses OpenAI and ElevenLabs in performance, serving as the underlying engine for high-end ChatGPT voice modes with LiveKit. It smoothly switches among more than 40 languages, reducing response time in Voice Agent scenarios to 250 milliseconds.

Video model commercialization leadership: Hailuo supports text-to-video, image-to-video, subject reference, and scene transition features. The latest Hailuo 2.3 excels in image quality and coherence. Hailuo AI has generated over 590 million videos, with partners including Cannes Film Festival and NYU’s Tisch School of the Arts.

Dual-drive B+C model: With 214 million users, the platform handles trillions of token requests daily. Quark’s audiobooks, Xiao Ai speakers, Tencent video content generation, and WPS AI features are all backed by its technology, with overseas revenue accounting for over 70%.

Long-term commitment: 90% of fundraising allocated to R&D

MiniMax plans to issue 25,389,220 shares in this IPO, with a price range of HKD 151 to HKD 165 per share. Regarding the use of funds, MiniMax explicitly states that 90% of the proceeds will be invested in R&D, with 70% dedicated to large model development and the remaining 20% for product iteration and optimization. This resource allocation marks the industry’s transition from the era of traffic dominance to a mature long-term focus.

The lineup of 14 cornerstone investors reveals the market’s recognition of MiniMax. These long-term institutional investors are not interested in short-term stock price fluctuations but see the potential for MiniMax to become a core player in AI infrastructure over the next 5 to 10 years. Time itself is the best growth leverage, and with its antifragile nature, MiniMax stands at a critical starting point for market long-term testing.