MiniMax: A young man from a county town in Henan and his 300 billion yuan

Author: Lin Wanwan, Rhythm

In 2014, a PhD student from the Institute of Automation, Chinese Academy of Sciences, came to Baidu Research Institute. He came from a small county in Henan Province. He had calculated his future: his ideal job after graduation was at IBM, where he would write Java code and earn an annual salary of 280,000 yuan.

During the 2026 Spring Festival, an agent tool called OpenClaw became a global sensation. Developers needed a large underlying model to support their Lobster-themed projects. One model was both fast and cheap, absorbing 1.44 trillion tokens on OpenRouter in a single week, topping the entire platform.

This model is called M2.5, and the company is called MiniMax.

Within two months of its listing, the stock price surged from HK$165 to HK$1,300, and its market capitalization exceeded HK$300 billion, even though it was a company with annual revenue of less than US$80 million.

The person who created the MiniMax was Yan Junjie, an intern from twelve years ago.

A bet made more than a year in advance

During the 2021 Spring Festival, Yan Junjie returned to his hometown in Henan to celebrate the New Year and visited his maternal grandfather.

His grandfather told him that he wanted to write a memoir to record his 80 years of life. But he couldn't type and couldn't organize the story properly, so after a few talks, he gave up.

Yan Junjie has worked in the AI ​​industry for over a decade. At that moment, he suddenly realized that all the things he had done, even though they had been implemented in the industry and helped countless companies, were of no use to an old man who wanted to write his memoirs.

This detail was later repeatedly cited, taking on a somewhat inspirational tone. But it does explain one thing: his motivation for developing AI was simple—to make it truly usable for ordinary people. This conviction later drove a series of counterintuitive decisions.

He left SenseTime at the end of 2021.

Timing is crucial. SenseTime was preparing for its Hong Kong IPO at the time. He was the vice president, deputy director of the research institute, and CTO of the smart city business group. He left when the company was at one of its most valuable periods. He didn't live to see the IPO, didn't live to see his wealth realized, and left.

ChatGPT was only released in November 2022.

MiniMax was founded in December 2021.

This time lag was the foundation for everything that followed. Yan Junjie later said that if they hadn't started early, MiniMax would have been no match for others in the later funding environment where "star researchers and AI backgrounds from big companies were more popular".

His parents are ordinary people. He attended high school in the county town, was admitted to the Department of Mathematics at Southeast University, later pursued his doctorate at the Institute of Automation, Chinese Academy of Sciences, completed his postdoctoral studies at Tsinghua University, and then joined SenseTime. He rose through the ranks step by step without any overseas background or prominent connections.

During his internship at Baidu, he crossed paths with Yu Kai of Horizon Robotics. Yu Kai later said that academic ability can be trained, but those who can engineer and implement AI technology are extremely rare. Yan Junjie is one of them.

After joining SenseTime, he rose from intern to vice president in seven years. In 2018, despite a shortage of manpower, he led a team to develop an "All for One" model algorithm, surpassing Megvii and Yitu in a bidding process to win first place in the industry. Some have commented that he "reads papers incredibly fast, ignoring clichés and focusing only on the essential points." This efficiency later became part of MiniMax's company culture.

He named the company MiniMax, which comes from von Neumann's minimax algorithm in game theory.

His explanation is that when making decisions, one should first guard against the worst risks, and then choose the relatively optimal solution.

A peculiar shareholder list

In December 2021, MiniMax completed its angel round, raising $31 million, with a pre-money valuation of $170 million. Investors included miHoYo, IDG, Hillhouse Capital, and Yunqi Capital.

Mihoyo's funding was a bit unusual. Yan Junjie had a good personal relationship with Mihoyo's chairman, Liu Wei, and invested in the company during its angel round. Liu Wei is still listed as a non-executive director on MiniMax's board of directors.

miHoYo is a MiniMax client; their models are used for NPC dialogues and story generation in the game.

After the angel round, the story encountered a small episode.

In March 2023, Silicon Valley Bank declared bankruptcy. MiniMax had all its funds held in that bank at the time. This was the most dangerous period for the startup in its early stages; they lost their money, and the fundraising environment was chaotic. But they survived, and two months later they secured $257 million in Series A funding, valuing the company at $1.157 billion.

The list that followed became increasingly extravagant. Alibaba came in, Tencent came in, and Sequoia Capital followed suit. Prior to the IPO, the company had completed seven rounds of financing, raising nearly $1.5 billion in total, reaching a valuation of $4.2 billion. After the IPO, Alibaba held 12.52% of the shares, becoming the largest external shareholder.

Yan Junjie had a habit during his early fundraising rounds: he only talked to the top executives of investment institutions. He met with Shen Nanpeng of Sequoia Capital and Zhang Lei of Hillhouse Capital.

But there is one person on this shareholder list who deserves special mention: Yun Yeyi.

Born in 1994, she holds a bachelor's degree in Electrical Engineering from Johns Hopkins University, with minors in Economics and Mathematics. Immediately after graduating in 2017, she joined SenseTime, working in financing and strategic investment. A year later, she was promoted to Executive Assistant and Director of Strategy for CEO Xu Li. She was deeply involved in SenseTime's entire process, from its early stages to its Hong Kong IPO.

In 2021, she and Yan Junjie started their own business.

One investor described her as "capable, charismatic, and with strong execution skills, possessing a maturity beyond her years." Her division of labor with Yan Junjie is clear: one defines the technological vision, the other transforms that vision into money and resources. Yan Junjie is so engrossed in technology that he doesn't care if he shaves his head, but the market, capital, and globalization are Yun Yeyi's battleground.

On the day of the IPO, the two stood on the same platform. Yun Yeyi was 31 years old and had a net worth of over HK$4 billion.

385 people and 1% of the money

When MiniMax went public, the company had 385 employees with an average age of 29.

From its founding until September 2025, the company has spent approximately $500 million. OpenAI spent between $40 billion and $55 billion during the same period.

This comparison is somewhat absurd. They built a globally leading company across all modalities with less than 1% of their competitors' money. Saving money is just a result. The real reason is that they've pushed AI to its limits. 80% of the company's code is written by AI, which internally calls "interns." These interns have high privileges—they can directly access the codebase, modify the production environment, chat with it on Lark, and deploy it immediately after a review.

This efficiency makes MiniMax's per capita output abnormally high.

At the product level, they adopted a multimodal approach from the very beginning: simultaneously focusing on language, video, voice, and music. While others were learning from ChatGPT to develop dialogue systems, Yan Junjie bet on multimodal fusion. His judgment was that multimodality is a fundamental prerequisite for continuously improving intelligence; without a multimodal approach, there would be no chance for the next generation of models.

In the summer of 2023, he made an even more radical decision.

We've allocated 80% of our computing power and R&D resources to MoE (Hybrid Expert System).

At that time, the mainstream technology in China was still iterating on dense models, and MoE was considered a "cutting-edge but immature" technology. Yan Junjie's logic was simple: if you want to serve tens of millions or hundreds of millions of users, the cost and latency of generating tokens simply cannot be handled by dense models. Without MoE, the scale cannot be increased, and everything else is in vain.

In early 2024, MiniMax released the first MoE large model in China.

In terms of products, they didn't try to dominate the domestic market. For consumers, they developed Hoshino and Talkie, one in China and the other overseas, focusing on AI companionship; Conch AI focuses on video generation, and in the second half of 2024, it ranked first in monthly active users globally for six consecutive months.

Current figures: 236 million users, covering 200 countries and regions, with overseas revenue accounting for 73%. 214,000 enterprise customers and developers on the B-end. MiniMax models have been deployed by Google Vertex AI, Microsoft Azure, and AWS. MiniMax was also Notion's first open-source model choice.

In February, ARR exceeded $150 million, and the daily token consumption of the M2 series was 6 times that of December last year, with the programming category growing more than 10 times.

This is why the market is willing to give it a price-to-sales ratio of 200.

But there is one set of numbers that needs to be broken down.

The annual report shows a gross profit margin of 4.7% for consumers (C-end) and 69.4% for businesses (B-end). While 67% of the company's revenue comes from consumers, these consumers contribute almost no gross profit. A rough calculation for the fourth quarter shows that the gross profit margin for consumers (C-end) has fallen to approximately 2.1%. The overall gross profit margin increased from 12.2% to 25.4%, primarily due to a rapid increase in the proportion of revenue from businesses (B-end) in the fourth quarter, which boosted the overall figure.

This is an unsolved problem.

Mountains are not insurmountable

In June 2025, MiniMax released the M1 model.

Yan Junjie posted a sentence on his WeChat Moments:

"For the first time, I realized that mountains are not insurmountable."

The reality behind this statement is that while the technological capabilities of leading Chinese and American companies in model development may only differ by 5%, this 5% allows overseas companies to occupy scenarios with 10 times the value and charge 10 times the price, ultimately resulting in a nearly 100-fold commercialization gap. OpenAI's latest valuation exceeds $700 billion. MiniMax's IPO market capitalization was HK$80 billion, less than $10 billion.

He made a prediction that there will be five top AGI companies in the world in the future, at least two of which will be from China, and one of them may even become number one.

Following his IPO on January 9, he appeared at a symposium with experts and entrepreneurs chaired by the Premier on January 19, becoming the second AI big model founder to attend the meeting after Liang Wenfeng of DeepSeek.

Then on March 2, the first annual report was released, and Hong Kong stocks surged that day.

During the earnings call, Yan Junjie spent a lot of time talking about one thing: MiniMax wants to transform from a "large model company" into a "platform company in the AI ​​era".

He defined platform value with a formula: intelligence density × token throughput. In the internet era, platforms were traffic gateways; in the AI ​​era, platforms are companies that define the boundaries of intelligence and simultaneously reap commercial benefits. Google is doing it, OpenAI is doing it, and they want to do it too.

His opponent was dozens of times larger than him.

The Hong Kong listing simply pushed him onto a different battlefield. Quarterly reports, analysts, market capitalization pressure—these are all completely different things from writing code. The secondary market doesn't believe in sentiment; it only looks at numbers. Whether the C-end story can be converted into gross profit, whether the B-end growth rate can be maintained, and when M3 will be released—these are questions he will have to answer every quarter going forward.

But looking at it from a broader perspective, the story of MiniMax is not just about one company.

The US has been tightening its grip on chip supplies in recent years. Sales of the A100, H100, and H800 are all restricted. The logic is straightforward: controlling computing power controls AI.

China was forced to take a completely different path.

DeepSeek achieved near-H100 performance with the H800. MiniMax accomplished with $500 million what OpenAI spent tens of billions to achieve. Yan Junjie bet on MoE in 2023 because his limited GPU resources couldn't support the inference volume of hundreds of millions of users. The M2.5 costs $1 per hour of continuous operation, one-twentieth the cost of GPT-5. Hybrid attention architecture, linear attention, and the CISPO algorithm—innovation is often born out of necessity.

The original intention of the chip blockade was to widen the gap, but the actual effect was to force Chinese AI companies onto an evolutionary path of low computing power and high efficiency.

Limited funds, fewer cards, and fewer people have ironically forced out the development of exceptional engineering capabilities and architectural innovation.

This is the same logic Huawei uses in chip manufacturing: if you block one of my capabilities, I will make up for it in other dimensions. In the process of making up for it, I may develop something that you don't have.

OpenAI currently has over 4,000 employees and burned through $8 billion in cash by 2025, with plans to invest $600 billion in computing power by 2030. MiniMax has 385 employees and has spent a total of $500 million.

It's still too early to tell who will win. But at least for now, fewer and fewer people are betting on the Mini Max going to die.

The Henan PhD student who interned at Baidu in 2014 probably never imagined that twelve years later, the position he stood in would be linked to a technological competition of national significance.

He chose to keep running.