MiniMax Goes Public: The $1.5 Billion Fundraising Blitz, 15 Reasons Investors Bet on It, and How Long Each One Lasts

By Songdao Tu | Produced by AI Nao

Intro

On January 9, four-year-old MiniMax filed for its Hong Kong IPO.

Over those four years, MiniMax raised more than $1.5 billion, fighting its way out of China's brutal large-model wars.

Behind that $1.5 billion stood miHoYo, Hillhouse, HSG, IDG Capital, Matrix Partners China, Yunqi Capital, Future Capital, Oasis Capital, Shunwei Capital, Xiaohongshu, Tencent, Alibaba, Kingsoft Office, CP Group, and more. The roster included China's top VC and PE firms, its most inscrutable tech giants, and its most forward-looking and active corporate investors.

Different money meant different faces and different judgments.

Betting on the same company — MiniMax and Junjie Yan — people saw and thought very different things at different points along the wave.

AI Nao has pieced together investors' public and private remarks about this bet over the past four years. It's a fascinating retrospective — some former non-consensus views quickly became consensus; some former certainties have long since scattered to the wind.

With pre-IPO oversubscription exceeding 1,200x, MiniMax's expedition gained fresh momentum. In today's explosion of China tech narratives, this company's future weight is hardly in question.

From today's vantage, backing MiniMax already counts as a success for any of these investors. Yet set against the buried verdicts of the past four years, the formula for success in primary markets comes partly from foresight validated by time, and equally from misjudgments forgotten by time. It is an alloy of necessity and contingency, of clarity and luck.

2021

$20 million raised

$200 million valuation

Led by Hillhouse, with miHoYo, IDG Capital, and Yunqi Capital co-investing

"You could tell Yan was brilliant the moment you saw him... He told me you could use foundation models to solve more general and generalized problems." — Yu Chen, Partner at Yunqi Capital (LatePost interview, November 24, 2025)

Chen was among the earliest investors to meet Junjie Yan in 2021.

Because he had met Yeyi Yuan four years earlier — Yuan was Chen's junior at Johns Hopkins, and after joining SenseTime became Yan's co-founder-to-be and soon-to-be co-founder of MiniMax.

The bald Yan made a strong first impression on Chen. At the time, the entire industry was talking about using small models to solve complex problems; Yan's contrarian approach was refreshing. Chen quickly decided to invest.

"First met IO [Yan] in September 2021. It was more than a year before ChatGPT-3.5 launched. The concept of large models barely existed in the market. We invested soon after the meeting, because this team had deep technical conviction... and unusual technical foresight." — Liang Li, Founding Partner at Hillhouse (MiniMax Partner Day video, August 31, 2024)

At this point Yan had just left SenseTime and was beginning to assemble his team; MiniMax was still three months from formal incorporation. He shared his ideas about artificial general intelligence with Lei Zhang and Liang Li, emphasizing that interaction was the prerequisite for AGI.

Before committing, Hillhouse displayed the characteristic caution of a major fund, conducting intensive due diligence on MiniMax's technical approach with CTOs at leading cloud providers. The feedback was almost devastating — this group of technical experts collectively judged Yan's vision of going straight for AGI interaction as completely unreliable, something no single company should attempt.

But Zhang and Li ultimately rejected their judgment. Despite the extreme uncertainty, Hillhouse displaced Yunqi as the lead investor in MiniMax's angel round, pricing the company at a steep $200 million.

May 2022

$50 million raised

$500 million valuation

Future Capital enters; Hillhouse, Yunqi Capital, and other existing investors follow on

"There's someone you absolutely need to meet. Two years ago he was talking about AGI and large models — I didn't understand it at the time. Then my partner listened and didn't understand either. But he really believed." — Wei Liu, Co-founder and Chairman of miHoYo (conversation with Oasis Capital's Jinjian Zhang, late 2022)

"I first met Yan at the end of 2021. The introduction came from miHoYo co-founder Wei Liu. Three of us from Future Capital went to that first meeting — frankly, I didn't understand what he was saying. Fortunately, one of the three did: Ling Xia. Xia later went back to Yan and insisted we had to invest. That gave us a ticket to the new world." — Mingming Huang, Partner at Future Capital (36kr Waves Conference, June 2024)

From 2021 to 2022, "not understanding" MiniMax was a widespread and genuine condition.

ChatGPT's global detonation was still months away, and the previous round of AI investment in China had left more disappointment than returns. The industry was in the rational darkness before dawn. Yan's narrative about AGI and interaction exceeded the cognitive maps of most investors, which were built on past experience.

But "not understanding" didn't prevent this from becoming the prelude to a great investment. The key perhaps lay not in understanding, but in whether you were willing to believe. Liu was exactly this way — not understanding didn't stop him from becoming the story's earliest evangelist. He pushed MiniMax onto one critical investor table after another.

Yet occasionally one or two people did understand.

"The three of us met Yan... I was the only one who understood. There was a reason behind it. At Future Capital, I was responsible for the smart electric vehicle track. Anyone following autonomous driving in 2021 knows the landmark shift that year: Tesla's transformer-based BEV and the end-to-end, data-driven approach to perception. This was no mystery to an investor focused on vehicles and autonomous driving — it was essentially asking whether the end-to-end, data-driven paradigm from autonomous driving could migrate to NLP. ...When I pushed this internally at Future Capital, I never framed MiniMax as 'China's OpenAI.' The 'large' in large models wasn't the essence. The essence was the new paradigm of end-to-end, data-driven approaches... My internal logic was consistent: how I viewed autonomous driving, how I viewed NLP, and starting in mid-2022 how I viewed end-to-end, data-driven robotics — the logic behind these was coherent and continuous. We were betting on the next-generation technical paradigm of end-to-end, data-driven approaches." — Ling Xia, Partner at Future Capital (Anyong interview, June 2024)

A year later, as ChatGPT swept the globe, Yan himself and all MiniMax investors would insist they were personally building China's OpenAI. Looking back today, Xia's original verdict about "autonomous driving paradigm migration" reads as hardcore with a touch of misalignment.

But this is a classic VC move — a lucky-hit victory. In 2021, at that desolate moment when no one yet felt AGI in their bones, investors had to find a rational anchor for an inexplicable fervor. If language models sounded too esoteric, explain them as autonomous driving without wheels.**

It turned out that the "large" in large models was indeed essential (Scaling Law ultimately ruled everything), and the end-to-end logic of NLP and autonomous driving weren't fully congruent. Yet at that moment, Xia won MiniMax a critical ticket by using an intelligible old paradigm, and brought Future Capital onto the fast clipper to the new continent.

This logical detour became a corrective compensation that hit the target. Great investments often don't come from precise rehearsals of the future, but from believing — in the fog — in an intuition that wasn't entirely correct but was strong enough.

May 2023

$200 million raised

Valuation exceeding $1.2 billion

Oasis Capital, Tencent, Xiaohongshu, Shunwei Capital, and HSG become new investors Hillhouse and other existing investors continue to add

"From day one, many of MiniMax's decisions were deeply non-consensus. Last year they bet on MoE [Mixture-of-Experts] architecture. At that time [June 2023], MoE wasn't even consensus in Silicon Valley. Only OpenAI was going all-in on MoE; Google was fully committed to Dense models. Even the originators of MoE weren't fully convinced about the path." — Mingming Huang, Founder of Future Capital (36kr Waves Conference, June 2024)

"Faced with the Dense and MoE technical paths, Yan chose MoE without hesitation — the option almost no one in China was taking. MoE had higher ceilings but also greater risk. In the middle, MiniMax went six months without releasing a model, pouring all resources and energy into MoE. The pressure on them must have been enormous, but Yan and the team carried through." — Liang Li, Founding Partner at Hillhouse (MiniMax Partner Day video, August 2024)

This was MiniMax's most precarious strategic silence.

In mid-2023, Huiwen Wang had rapidly attracted $50 million in financing through personal heroism, and Zhilin Yang's fledgling Kimi was raising its second round, trying to tear open a gap in the long-context race. Everyone feared being eliminated from the finals if they slowed for even a second; the prevailing playbook was rapid model releases, rapid benchmark climbs, rapid fundraising.

At this moment, MiniMax chose to go into seclusion. It halted all short-term updates that could excite investors, committing 80% of its compute to the MoE route that even Silicon Valley hadn't fully converged on, resulting in "six months without a new model release."

This choice appeared wildly ill-timed. In the US, beyond OpenAI's secretive MoE push, giants like Google and Meta were sticking with Dense models; domestically, beyond ByteDance and DeepSeek, all major tech companies and startups including Tencent and Alibaba had chosen Dense or Dense-dominant hybrid architectures.

Yan's bet on MoE was a bet on technology, but also a bet on capital's tolerance for a blank period.

Six months of seemingly suicidal falling behind bought MiniMax absolute generational cost advantages in inference after 2024. Post-2024, MoE gradually shifted from non-consensus to standard answer, becoming the default architecture for ultra-large parameter models.

MiniMax's wager demonstrated Yan's formidable strategic foresight. And in this critical battle, investors proved their most valuable capability: believing the founder's judgment. Then enduring. Then maintaining silence.

2024

$600 million raised

Valuation reaching $2.5 billion

Alibaba, Matrix Partners China, ADIA Kingsoft Office and others become new investors HSG and other existing investors continue to add

"If you believe in AGI, and MiniMax is a good company, then don't overfocus on its valuation today." — Wei Liu, Co-founder of miHoYo (conversation with Oasis Capital's Jinjian Zhang, late 2022)

"Today, whether it's MiniMax or Moonshot in China, valuations are definitely undervalued... If you compare with European and American peers, given these two companies' technical levels, their valuations should absolutely be at least 2-3x higher." — Ling Xia, Partner at Future Capital (Anyong interview, June 2024)

By 2024, China's large models had entered their most valuation-ambiguous summer. At that same roundtable, Kimi investor Yusen Dai offered a blunter footnote to the valuation anxiety of the moment: "(If they) make it, $3 billion is cheap; if they don't, $300 million is wasted."

No one could calculate these companies' true value. Any assertion that "valuation should double" was more like a necessary act of self-persuasion in the face of massive uncertainty.

This wasn't actuarial science; it was a collective bet with manic-depressive temperament — everyone had to believe their chips weren't expensive enough yet, just to keep sitting at the table.

"From the very beginning, the business model and company positioning Yan described to us was co-governance and co-construction with users, walking together on the road to AGI." — Qingsheng Zheng, Partner at HSG (MiniMax Partner Day video, August 2024)

"Among the AGI companies I've engaged with, MiniMax is the most resolute about its goals... What IO [Yan] talked about earliest were two major tracks — having your own business in hand, while using that business to drive model development." — Tian Ran, Kingsoft President's Office (MiniMax Partner Day video, August 2024)

"First time meeting Yan... He kept emphasizing that large models aren't atomic bombs, he didn't say rockets either — it's not some highly sophisticated technology. It's user in the loop. It ultimately has to be a product serving tens of millions of users." — Jinjian Zhang, Founder of Oasis Capital (Oasis Capital WeChat account, September 2024)

"The profile of tech founders we like is very clear: focused, resilient, low-profile. All the founders we've backed are [like this]." — Ling Xia, Partner at Future Capital (Anyong interview, June 2024)

If before 2023 Yan's external narrative focused more on descriptions of artificial general intelligence, by 2024, Intelligence with Everyone had fully occupied MiniMax's official communications and Yan's various personal expressions, becoming the company's publicly declared core vision.

This reveals some essential difference in底色 [fundamental character] between Yan and Zhilin Yang. Yang's self-positioning is as a haloed genius — younger (born 1993), likes playing in bands and names his company after a rock album, his personal image highly aligned with Kimi's aesthetic, pursuing a premium entrepreneurial elite route.

By contrast, Yan — born in a Henan county town in the late 1980s — from appearance to register couldn't and didn't intend to cater to世俗 [secular] notions of "cool."

This difference in faces naturally led them down divergent paths. For Yan, the foundation model was merely a hammer; he already had the hammer. His greater ambition was to find the most suitable and best nails, and nails meant products used by more people, whether B2B or B2C.

2025

Cornerstone investment of HK$2.7 billion

Valuation of HK$48 billion

**

Alibaba and ADIA become key cornerstone investors

Other cornerstone investors include IDG Capital, Boyu Capital,

Aspex, E Fund, China Universal Asset Management, etc.

"Made a few investments when I was young, threw money around blindly, pure sentimentality, barely managed them after."

——Wei Liu, Co-founder and Chairman of miHoYo (Shanghai Jiao Tong University School of Artificial Intelligence AI Theme Week, April 2025)

"A year ago, many overseas LPs questioned whether Chinese model companies could actually build models on par with OpenAI. Today everyone is full of confidence.

Stretching the timeline, if the future global landscape consists of 5-6 major model companies, we believe at least 2-3 will be Chinese — but which ones, that格局 [landscape] is far from settled."

——Ling Xia, Partner at Future Capital (Anyong interview, June 2025)

It is now 2025.

The discourse around Chinese AI has completely shifted. The hesitation and self-persuasion of a year ago have been replaced by a more relaxed narrative confidence. Competitors at the table have survived a brutal round of washing out, and the names that remain — whether MiniMax, Zhipu AI, or Kimi — no longer need to justify their right to exist.

The gold investors paid out is no longer a question of whether it returns, but of which step on the staircase each person ultimately stands on.

In January 2026, with the sequential IPOs of Zhipu AI and MiniMax, China's large models will have fully completed their adolescence of self-proof.

A new consensus is forming: the war is far from over, but it is now a war of survivors about distant horizons. And today, as another Chinese large model company enters the public markets, everyone understands — the era of grassland chaos and shared delusions has officially ended.

"AI is an industry that changes in the blink of an eye. One year equals three, five, even ten years in other industries."

——Yu Chen, Partner at Yunqi Capital (Anyong interview, June 2025)