MiniMax Financing Story: 4 Years, 7 Rounds, Who is Driving China's AI First Capital Feast

Author: Cheng Manqi

For two consecutive days, large model startup companies Zhipu and MiniMax have IPO’d in Hong Kong. Compared to the several IPOs during the peak of mobile internet, IPOs in the large model field are not happening after the major battles have subsided. They are not rewards for winners but the drumbeats for the next race.

After Zhipu and MiniMax landed in the secondary market one after another, they will initiate larger-scale private placements. This is a field where commercialization remains uncertain, but continuous R&D investment is very certain. The substantive meaning of an IPO is to more efficiently acquire more resources.

On the eve of MiniMax’s listing, we interviewed the MiniMax team and several of their investors to reconstruct the various perspectives on the market’s opportunities for large models over the past three years, as well as the company’s characteristics.

In the seven funding rounds before the listing, 30 institutions invested a total of 1.5 billion USD in MiniMax. Alibaba invested the most; Hillhouse was the lead investor in the first round, with a share second only to Alibaba, making it the second-largest external shareholder; MingShi participated in the most rounds.

Before the bell rang at the Hong Kong Stock Exchange this morning (January 9), MiniMax founder Yan Junjie shared his thoughts with LatePost:

We hope to have the opportunity to make a greater contribution to the overall improvement of industry intelligence levels in the future. We initially explored a purely grassroots AI startup path. Although it remains very challenging, if we can inspire the development of the AI innovation and startup ecosystem, we will feel very honored.

By the close of trading today, MiniMax’s stock price rose more than 78% from the issue price of HKD 165 to HKD 294, with a market value reaching HKD 89.8 billion.

Starting Before the Boom

Hillhouse: Earliest investor, largest external financial investor

MiniMax was founded in early 2022, just before the ChatGPT craze, and Hillhouse was its first investor.

During the startup preparation period, at Hillhouse partner Li Liang’s visit to MiniMax’s Beijing office, he and Yan Junjie and Qu Yeyi talked for three hours, after which they issued a TS (Term Sheet) with a blank valuation: you can specify your desired valuation and investment amount. Yan Junjie provided the planned figures: $30 million in funding, with a valuation of $200 million.

Before founding, Yan Junjie and Qu Yeyi were colleagues at SenseTime. Yan Junjie was born in 1989, grew up in a small town in Henan, and earned his PhD from the Institute of Automation, Chinese Academy of Sciences. During his seven years at SenseTime, he rose from researcher to the company’s youngest vice president, serving as executive director of the research institute and overseeing smart city, gaming, and other businesses; Qu Yeyi graduated from Johns Hopkins University and previously served as strategic head of SenseTime’s CEO office.

The first Hillhouse investor to contact MiniMax, Xue Zizhao, told LatePost that a few days before meeting Li Liang, Yan Junjie gave Hillhouse a nine-hour “technical lesson”: scaling laws of Transformer architecture models, progress of GPT-3, reinforcement learning by DeepMind, image generation with diffusion models, how CLIP combines images and language… “At that time, few people could piece these technologies together. I personally found it hard to believe it could be done, but in hindsight, IO (Yan Junjie) was right in all his judgments.”

In the early days of MiniMax, the team dissected technological changes and market opportunities on a whiteboard.

MiniMax’s approach is to develop text, speech, and image models simultaneously, using large models and multimodal technology to create AI applications accessible to ordinary people. This was the vision established when MiniMax was founded: “Intelligence with everyone.”

Hillhouse judged at the time that this was a systematic project; the team needed not only algorithm expertise but also comprehensive capabilities in hardware, data, engineering, and applications. Yan Junjie, who led over 1,000 people at SenseTime, had experience in algorithms, engineering, organization, and AI commercialization.

In less than two weeks, Hillhouse completed the TS review and investment decision, leading the angel round for MiniMax. About a week later, Sequoia, which had also engaged with MiniMax, did not participate in this round. One and a half years later, in July 2023, during the fourth round of financing with a pre-money valuation of $1.55 billion, Sequoia became a shareholder of MiniMax.

In November 2022, Zhang Lei organized a small-scale trip to the Middle East. Among the entrepreneurs he traveled with were BYD founder Wang Chuanfu and Horizon Robotics founder Yu Kai, as well as Yan Junjie, who was then relatively unknown and had just started his entrepreneurial journey less than a year earlier.

In the VIP lounge of the Qatar World Cup, Yan Junjie explained what AGI (Artificial General Intelligence) is in somewhat unskilled English to Middle Eastern friends. Half a month later, this relatively niche term spread worldwide with the release of ChatGPT.

The earliest Hillhouse investor to contact MiniMax, Xue Zizhao, officially joined MiniMax in 2023: “I invested myself.”

Hillhouse made multiple investments in MiniMax after the angel round. Before the IPO, Hillhouse held 7.14% of MiniMax, making it the second-largest external investor after Alibaba and the largest financial investor.

MiHoYo’s Liu Wei: “Super smart is overestimated, resilience is underestimated”

In early 2021, MiHoYo founder Cai Haoyu once said in a speech: “To build a virtual world with 1 billion people living in it by 2030.” MiniMax’s initial envisioned application aligns with this: creating multimodal AI agents that interact with ordinary people (Note: “AI agent” here refers to “AI character,” not the now common “AI Agent”). Liu Wei (“Da Wei Ge”) and Luo Yuhao (“Luo Ye”) from the three founders of MiHoYo have known the MiniMax team for a long time. They learned about Yan Junjie’s startup almost simultaneously with Hillhouse, and their investments came naturally.

MiHoYo is the type of investor founders prefer most. Qu Yeyi told LatePost: “They don’t particularly inquire about operational details. When we meet, we talk more about ‘life principles.’ That’s based on trusting the team.”

Yan Junjie experienced the ups and downs of the previous AI boom. “I went through a year and a half of losing, and also experienced winning after doing the right thing.” In a recent interview with Luo Yonghao, Yan Junjie reflected on the setbacks during Face Recognition at SenseTime. After facing enormous pressure and catching up against the wind, he gained stronger technical confidence.

He also witnessed the entire computer vision era, where the industry experienced a high start and low decline. A founder of one of the “Four Little Dragons” of AI once commented on Yan Junjie: “He has tasted the苦 of AI 1.0.”

Last year, when discussing large models, Liu Wei told us: “Among large model entrepreneurs, the ‘Super smart’ part is always overestimated, and resilience is underestimated. But entrepreneurship is a marathon, and resilience is extremely valuable.”

Yunqi, IDG join, ending the angel round

During Qu Yeyi’s time at SenseTime, he became familiar with many investors, including Johns Hopkins alumnus and Yunqi managing partner Chen Yu, and IDG partner Niu Kuiguang, who had invested in SenseTime multiple times.

These two institutions quickly joined MiniMax’s angel round. Chen Yu had earlier discussed technological trends with Yan Junjie and Qu Yeyi. After Yan Junjie officially decided to start the business, a dinner in Shanghai led Chen Yu to express his desire to invest: “I want to bet on a route that could potentially overturn existing technology solutions. Previously, it was all small models. Yan Junjie saw the value of foundational large models early on.”

Niu Kuiguang contacted Qu Yeyi shortly after MiHoYo and Hillhouse issued TS. They had a phone call in the morning and flew to Shanghai in the evening to meet Yan Junjie and the core team, quickly finalizing the investment.

Throughout the entire angel round, MiniMax raised a total of $31 million, with a post-money valuation of $200 million. This was very close to the team’s initial plan—to raise $30 million with a post-investment valuation of $200 million. MiniMax declined proposals to raise more money and achieve higher valuations.

MingShi: The only new shareholder in MiniMax’s second round, the most invested institution

Before ChatGPT’s release, MiniMax had two rounds of financing. The only new shareholder in the second round was MingShi Capital. Founding partner Huang Mingming and partner Xia Ling first met Yan Junjie at a hotel lobby in Beijing, where Yan Junjie was reading papers on an iPad. To this day, he still maintains the habit of spending about an hour daily browsing new papers.

Their first meeting lasted over two hours. Yan Junjie started by discussing technological changes. Xia Ling first heard about AGI from Yan Junjie, and he was searching for information on the spot. Now, many people’s first reaction in similar situations is no longer just Baidu search, but asking ChatGPT or Doubao.

“Honestly, I wasn’t very sensitive to AGI at the time, but he quickly mentioned that GPT is an end-to-end data-driven model.” MingShi had previously invested in Ideal Auto for seven rounds. Since 2021, a major trend in intelligent driving has been the significant improvement brought by end-to-end models.

Yan Junjie also discussed how technological changes are transforming business logic: the long-term difficulty for the previous AI companies was that their models were not sufficiently general. For different scenarios and tasks, retraining models was often necessary. Large models are “One Model for all”—a single model serving multiple scenarios and tasks—meaning AI commercialization can break out of the old path of custom development for B2B or G.

In early 2022, Xia Ling met Yan Junjie twice more. Not long before that, at MingShi’s year-end summary meeting, Xia Ling made a projection of AI technology trends for the next five years. When he shared ideas like “multimodal technology worth redoing Adobe,” “Agency,” and “more intelligent robots” with Yan Junjie, the latter put down his chopsticks and shared specific application directions for MiniMax.

The next time they met was on February 14. Xia Ling wanted to verify whether MiniMax was more optimistic about to-C or to-B. “To C,” Yan Junjie replied. He said they would no longer follow the old path of customized B2 projects, which was also what Xia Ling had in mind. Both were very engaged, and they hardly ate their food. That holiday, no flowers, but Xia Ling packed a plate of squid flowers to go home.

Like Hillhouse, MingShi also values Yan Junjie’s comprehensive experience from algorithms and engineering to business. This was validated during the investment: in the second half of 2022, MiniMax, which had trained several versions of text models, began searching for GPUs nationwide. When some autonomous driving companies exited, many GPU resources were leased back, and MiniMax obtained computing power at half the market price amid the large model boom.

MingShi founding partner Huang Mingming described Yan Junjie’s entrepreneurial determination: “At that time, ChatGPT hadn’t been released yet, OpenAI was still lukewarm, and Yan Junjie, who had already reached mid-to-senior level at SenseTime, decided to jump into entrepreneurship.” MingShi is one of the most active investors in MiniMax, with investments in 8 rounds, including the cornerstone round and the 7 rounds before the IPO.

ChatGPT arrived, and everything changed

Rapid consensus in investment frenzy

In October 2022, MiniMax launched its first product, Glow. Without much marketing, it accumulated over one million secondary anime users in two months. For a startup with less than a year of formal operation and exploring a new AI-to-C path, this was a good start. But soon, in November, ChatGPT’s wave swept across, and Glow became a small ripple.

Under rapid consensus, MiniMax benefited directly: in early 2023, it quickly launched its third round, raising a total of $260 million—more than three times the previous two rounds combined—with a post-money valuation of $1.157 billion. Strategic investors like Tencent, Xiaomi, Xiaohongshu, as well as new shareholders like Shunwei and Oasis joined, and all existing shareholders continued to invest.

On the other hand, MiniMax was no longer one of the few options in the market. The “Hundred Models” battle began, with a wave of startups each with their own features and advantages emerging: Wang Huiwen’s Guangnian Beyond, founded with a $50 million investment, created super apps like Sogou Input Method; Wang Xiaochuan founded Baichuan Intelligence; Li Kaifu founded Zero One Wanyou. Among the emerging tech forces, there are also companies like Zhipu, founded in 2019; and Moonshade, founded by Yang Zhilin, who developed XLNet and Transformer-XL, with academic background directly related to large language models. These companies quickly secured funding. Some investors invested in multiple companies simultaneously, such as Alibaba, Tencent, and Shunwei.

MiniMax’s strategy is to retain more control and avoid rapid dilution. Tencent initially wanted to invest more in MiniMax’s third round, but ultimately, MiniMax accepted Tencent’s $50 million offer.

ByteDance and Sequoia enter

In May 2023, Wang Huiwen ended Guangnian Beyond due to health reasons. Later, another key player influencing China’s large model landscape made a choice: ByteDance.

ByteDance had already assembled a large model team but was also considering external investments. Similar to Google investing in Anthropic; Tencent and Alibaba pursuing both self-research and investments. By around June 2023, ByteDance had issued investment intentions to two large model companies: MiniMax and the then newly formed Jie Yue Xing Chen.

However, after a senior management meeting mid-year, ByteDance decided to cease external investments in large model companies. Zhang Yiming’s attitude was: “Why don’t we do large models ourselves? We should do it ourselves, and we can do it well.”

Meanwhile, Sequoia China led the Series A+ investment in MiniMax. By this time, Hillhouse’s three rounds of Super Pro-rata (priority investment rights) had concluded.

In this round, MiniMax raised $50 million, with a post-money valuation of $1.6 billion. Sequoia continued to support in multiple rounds afterward. This remains one of Sequoia China’s largest investments in the large model field to date. Before the IPO, Sequoia China held 3.81% of MiniMax, making it the third-largest financial investor. Sequoia also invested in companies like Guangnian Beyond, Moonshade, and Jie Yue Xing Chen.

According to LatePost, Sequoia and Hillhouse discussed their respective shareholdings in this round down to three decimal places.

Alibaba’s Big Move, the Spring Festival that Changed Many Lives

In the previous AI boom, Alibaba was both an important supporter and had subtle competition with startups, reflecting the current relationship between big companies and AI startups.

Alibaba had served on the boards of SenseTime and Megvii. In 2017, both companies aimed to acquire China’s third-largest security company, Uniview, to find hardware carriers for visual AI, through acquiring Uniview’s parent company, Qianfang Technology. But ultimately, Alibaba bought Qianfang for 3.7 billion RMB, bringing Uniview into its fold, to help Alibaba Cloud expand into government and enterprise markets.

Time has passed, and Alibaba Cloud has gradually phased out its government and enterprise business focused on private deployment. After the large model craze, Alibaba’s No. 6 employee Wu Yongming returned in 2023 as Group CEO and Cloud CEO, proposing a new strategy: AI-driven, public cloud first.

Alibaba began broadly investing in large model companies, which are the main clients for cloud AI computing power. In the second half of 2023, Alibaba successively invested in Zhipu, Baichuan Intelligence, and Zero One Wanyou.

By the end of 2023, Alibaba started engaging with both MiniMax and Moonshade.

This was a key round for Moonshade’s subsequent rise. At that time, Moonshade was seeking a pre-money valuation of $900 million to attract investments from Xiaohongshu and others. Before the Spring Festival, Alibaba intervened, causing the pre-money valuation to jump to $1.5 billion, and Alibaba invested nearly $800 million.

Early-stage companies usually hesitate to accept such large investments due to high equity dilution, but Alibaba’s influence was immediate: the $800 million investment became a headline in the AI industry; combined with Kimi’s product launches and growth in the first half of 2024, Moonshade’s visibility peaked.

Initially, Alibaba also wanted to hold about 30-40% of MiniMax’s shares, but the final agreement was to accept Alibaba’s $400 million investment. This was MiniMax’s fifth round of financing, completed in March 2024, with a total raise of $654 million, and a post-money valuation of $2.55 billion. New shareholders included GGV Capital and China Life.

Before the IPO, Alibaba held over 13% of MiniMax, making it the largest external shareholder.

More Institutions Invest in Large Models: Risk Capital, Manufacturing, and Family Offices

After Alibaba’s large investment in early 2024, the frequency of financing in foundational models significantly decreased from 2024 to 2025. Tech giants like ByteDance and Alibaba increased their comprehensive investments in AI models and products, with several times the manpower and computing power of startups, and controlling traffic and advertising platforms. Temporarily, companies like DeepSeek, focusing solely on R&D without commercial pressure, pushed out a batch of typical startups needing funding, pushing them into the shadows. Fewer companies could continue raising funds, and fewer investors were willing to make large bets.

Beyond venture capital, more types of investors became shareholders of MiniMax, including China’s earliest equity investors—China Life Insurance; Yingke Expansion led by Li Zehua; and Bo Rui Capital, founded by Ningde Times co-founder and vice chairman Li Ping, who invested alone. They offer different perspectives on large models.

China Life: Seeing a trustworthy team

“Young, highly determined, always smiling, speaking at a steady pace.” This was Gu Yechi’s first impression of Yan Junjie. Gu Yechi is head of equity business at China Life Investment Insurance Asset Management, who worked in regulatory agencies for ten years before doing ten more in equity investment.

For insurance capital, avoiding mistakes is more important than achieving excess returns. After meeting nearly all top large model founders, Gu Yechi and the China Life investment team chose MiniMax, investing in two rounds in early and late 2024.

Gu Yechi meets Yan Junjie roughly every two months. He believes Yan Junjie is a “genuine, forward-thinking, deeply committed to technology, and consistent” founder: “In 2023, Junjie started talking about MoE (Mixture of Experts), and then trained MoE. This is now the mainstream architecture in the industry; more than a year ago, he told me that large model companies should mainly rely on technology rather than flow, and they later did exactly that. Now, this is the industry’s mainstream narrative.”

“This makes us very confident,” Gu Yechi said.

Bo Rui Capital: Looking for scientists who can become entrepreneurs

“If it weren’t for the MiniMax project, we might not have invested in large models,” Wang Limin, managing partner of Bo Rui Capital, told LatePost.

Bo Rui Capital is a venture capital firm supported solely by Li Ping, vice chairman of Ningde Times. Li Ping co-founded Ningde Times in 2010 and is now its vice chairman.

After ChatGPT, Bo Rui Capital, which initially focused on advanced manufacturing and hard tech, also began studying the generative AI revolution driven by large language models, but they were not in a rush to act.

In November 2023, after a three-hour discussion with Yan Junjie at Jinjiang Hotel in Shanghai, Li Ping and the Bo Rui team reached a preliminary investment intention. Bo Rui later participated in MiniMax’s fourth-round financing in early 2024, marking its first investment in software and information technology.

Yan Junjie’s cost control awareness, MiniMax’s early deployment of computing power, and batch exploration of applications, as well as their early efforts to generate revenue for R&D, resonated with Bo Rui. Ningde Times also experienced a similar process in early automotive power battery development: first forming a commercial loop with bus and commercial vehicle businesses, then reinvesting the revenue into subsequent R&D to improve battery performance and sharply reduce costs.

“Yan Junjie is very clear that today’s large model startups, especially in China, don’t have that much money to burn, nor do they have the most advanced clusters of computing power. Chinese large model companies must forge their own path, developing their foundational models under cost and compute constraints.”

“What I learned from Robin (Yuan Yuchun) is that a top scientist also needs top business thinking to build a successful enterprise,” Wang Limin said.

“Stay in the tide”

Over the past three years, consensus has formed rapidly and been overturned even faster: 2023 was a chase for GPT-4; 2024 started with Alibaba’s huge bet and ended with Doubao’s later rise; by 2025, DeepSeek open-sourced world-class inference models at extremely low cost, and the valuation of top global startups reached hundreds of billions of dollars. The question of who will become China’s OpenAI has lost its meaning.

MiniMax’s survival strategy is not to maximize a single advantage but to continuously adjust its footing, approaching a direction that makes AI serve ordinary people.

It develops large language models while not abandoning multimodal generation, because Yan Junjie believes serving ordinary people with AI requires intelligence and multimodal interactions involving vision and speech. They build models and applications; Yan Junjie once said, “Without a product to connect, even if you have technological progress, it ultimately isn’t yours.” They target both domestic and overseas markets.

Applications like Xingye/Talkie, Heluo AI for generating videos and voice content, MiniMax’s voice models, and open platform APIs each contribute about 30% of revenue, maintaining a balanced 1:1:1 ratio.

In the technical domain directly overseen by Yan Junjie, he is willing to take risks. In the second half of 2023, MiniMax almost dedicated all R&D resources to building MoE (Mixture of Experts) models, which failed twice during training; in 2024, 80% of resources were allocated to a new model based on linear attention architecture, later released as M1 in early 2025, all in pursuit of the highest technical ceiling he saw at the time.

Exploring commercialization possibilities and focusing on technical R&D are two sides of the same coin. Like the name MiniMax, in great uncertainty, it seeks the smallest chance of success with limited resources.

In early 2025, Yan Junjie said he hopes to stay in the tide: “One is to participate in driving the tide to continue; the other is for our company to keep developing.”