MiniMax secured HKD 16 billion in funding from long-term investors and sovereign funds, with YAN Junjie committing to "taking no salary before achieving AGI"

According to IPO Early Knowledge, MiniMax (0100.HK) has completed a new round of financing of HK$16 billion. For large model companies, continuous capital support is required for training, inference, computing power locking, and globalization investment. This financing will further supplement MiniMax's capital reserves and support its subsequent AI infrastructure investment, model R&D, and global commercial expansion.

It is worth mentioning that this transaction has attracted many top international sovereign funds, long-term funds, first-tier Chinese-funded institutions, and top multi-strategy funds to participate, covering the Asia-Pacific, European, and U.S. markets. Among them, there are more than 20 long-term and sovereign funds, and many Pre-IPO and cornerstone investors have further increased their positions in this transaction to express their support for MiniMax's long-term development.

It is reported that this transaction has attracted more than 100 institutions to participate, achieving 7 times of subscription coverage. The initial issuance size was about 1.8 billion U.S. dollars, and it was finally expanded to more than 2 billion U.S. dollars driven by institutional demand. Against the backdrop of recent capital market fluctuations, the subscription situation of this transaction shows that existing shareholders, cornerstone investors, and international long-term funds continue to pay attention to MiniMax's medium and long-term fundamentals and industry position.

Yan Junjie's all-hands letter says "will continue to devote all efforts to the AGI cause"

Over 80% of Pre-IPO and cornerstone shareholders express their intention to continue holding

On the day the financing announcement was released, Yan Junjie, founder and CEO of MiniMax, sent an all-hands letter.

In the all-hands letter, Yan Junjie stated that from today until the day MiniMax achieves AGI, he will no longer receive any salary from the company. In addition, Yan Junjie will set aside 4% of the company's total share capital under his personal name in the next four years to incentivize team members who have fought side by side with the company for a long time and created value together; at the same time, he will set aside 1% of the shares to establish a special fund to continuously support the development of related open source communities.

"The market will fluctuate, and there will be noises from the outside world, but the direction of progress will not change." Yan Junjie said in the all-hands letter, "As people on the front line of the industry, we are more aware of the real speed of technological evolution than anyone else, and we are more aware of the long-term value we are creating and accumulating. I will devote all my time, energy, and resources to this cause. This is my long-term commitment as a founder."

In fact, over the past period of time, both MiniMax's founding team and its core shareholders have demonstrated their renewed recognition of MiniMax's long-term value.

On the one hand, MiniMax's founding team has set a 12-month voluntary lock-up period; on the other hand, Alibaba and miHoYo, two early strategic shareholders of MiniMax, have both stated that they "will not reduce their holdings", and more than 80% of shareholders, including many cornerstone investors such as Aspex, Boyu Capital, IDG Capital, Janchor Partners, and Martis Fund, have expressed their intention to continue holding.

Alibaba pointed out that as a strategic investor, Alibaba highly recognizes the technical strength and strategic determination of the MiniMax team. At present, the AI industry is moving towards large-scale commercial implementation, and Chinese AI enterprises are facing globalization opportunities. Alibaba will continue to deepen ecological collaboration with MiniMax in cloud computing, enterprise services, and other fields, help its technology iteration and commercialization speed up, and jointly build the global competitiveness of Chinese AI.

miHoYo emphasized that MiniMax is one of the very few companies in the world that have reached the cutting-edge level in both language text and multimodal capabilities. Its full-stack technical capabilities and depth of cutting-edge exploration are highly consistent with miHoYo's long-term investment direction in general artificial intelligence. As a strategic investor, miHoYo will continue to deepen comprehensive collaboration with MiniMax from basic technology to industrial ecology, and jointly accelerate the arrival of the general artificial intelligence era.

M3 Pro reaches 2.7 trillion parameters, exceeding known Chinese AI models

Underlying Infra capabilities continue to be strengthened, and computing power supply leads the industry

From a fundamental perspective, The Information reported a few days ago that MiniMax is advancing the new generation of large language model M3 Pro, with a parameter scale of 2.7 trillion, exceeding the currently known Chinese AI models, and plans to make it open source. Compared with the existing flagship model M3, M3 Pro means that MiniMax is still maintaining high-intensity investment in the basic model level, rather than only relying on application products and commercialization of existing models.

The significance of this news to the market is that the competition for large models is not over. For leading model companies, basic model capabilities are still the most fundamental competitive threshold. Only with continuous model iteration can application products, API calls, enterprise services, and developer ecology have sustained growth support.

If the 2.7-trillion-parameter model is successfully released, it will further strengthen MiniMax's technical narrative in China's open source large model camp —— a larger-scale model usually means a higher upper limit of capabilities, especially in complex reasoning, multi-step instructions, code generation, Agent tasks, and enterprise-level applications, where differences are more easily reflected. For MiniMax, the core highlight of M3 Pro is not just the parameter scale, but whether it can connect model capabilities, inference costs, and open source ecology. If the model achieves a balance between capability and cost after its release, MiniMax is expected to turn a technological progress into a common improvement in developer adoption, enterprise deployment, and capital market recognition.

Of course, the continuous expansion of model scale cannot be separated from the support of computing power and engineering systems. Industry information shows that MiniMax has legally obtained high-quality overseas computing power, leading most peers in the progress of computing power supply. For MiniMax, which is advancing the training of a new generation of large models, computing power is not a simple resource reserve, but a basic condition that determines whether the model can be continuously trained, iterated quickly, and controlled in cost.

Judging from MiniMax's recent computing power layout, it has made rapid progress in the construction of training-side clusters and the collaboration of system departments, achieving stability and performance improvement within a few months, and the cluster capability has reached the level of large cloud vendors. At the same time, MiniMax's independent networking has obvious effects in saving costs and time, and its stability has also exceeded expectations. This means that MiniMax's competitiveness is not only reflected in model releases, but also in the strengthening of underlying Infra capabilities.

In addition, domestic computing power has also been included in MiniMax's long-term layout. Relevant information shows that MiniMax expects to build its first domestic computing power cluster by the end of the third quarter of 2026, with reasoning first and then training in terms of rhythm. This arrangement conforms to the current stage of the domestic computing power ecology and will also help reduce its dependence on a single computing power source in the future. With new progress of 2.7T on the model side and two-way advancement of overseas resources and domestic clusters on the computing power side, the two together form the foundation for MiniMax's subsequent continuous iteration.

International major banks simultaneously give "buy" ratings

The policy window for large model enterprises returning to the A-share market is becoming clearer

In the first week of July, three major international banks, Goldman Sachs, Bank of America, and Citigroup, simultaneously gave MiniMax a "buy" rating, which is not common in the current market environment where the differentiation of the AI sector is intensifying.

Among them, Goldman Sachs gave the latest target price as high as HK$860. Goldman Sachs pointed out that DeepSeek V4 is about to introduce differentiated pricing during peak hours, which is an early signal of the rationalization of the industry's aggressive price war since the end of April; in this context, MiniMax M3 model's hybrid pricing is $0.22 per million tokens. With a higher proportion of self-built optimized computing power and an efficient architecture with smaller activation parameters, its gross profit margin is significantly higher than that of its peers.

Bank of America has noticed positive changes in MiniMax's revenue structure —— revenue has shifted from about 70% from consumer-side products in 2025 to an increased proportion of enterprise-side and cloud API businesses; the previous-generation model M2.7 has achieved an inference profit margin of over 40%, and by continuously improving infrastructure efficiency, the long-term profit margin will remain stable. Bank of America also predicts that MiniMax is expected to be included in the Shanghai-Hong Kong Stock Connect on August 6, bringing liquidity support from southbound funds.

Citigroup believes that M3 ranks first among global open source models on the Artificial Analysis Intelligence Index and the GDPval-AA benchmark. It is expected that its revenue growth rate will remain high, and the upcoming new-generation video model is expected to become a key catalyst to reverse market sentiment.

According to the target prices given by the three institutions, MiniMax still has at least 50% and even nearly double the upside potential.

Of course, at the capital market level, the policy window for large model enterprises to return to the A-share market is becoming clearer. In June this year, the Shanghai Stock Exchange issued the review guidelines for AI large model enterprises applying the fifth set of listing standards of the Sci-Tech Innovation Board, clearly requiring enterprises to explain their technical advantages and phased achievements in combination with model R&D, business models, user or deployment scale, model call volume, real business scenario implementation, industry status, and commercial arrangements.

This means that the policy does not support general "AI concept stocks", but leading large model enterprises that already have model products, large-scale applications, and industrial status. In other words, not all large models can return to the A-share market, nor can they gain recognition from the capital market just by telling an AI story. What the regulation really encourages are companies with hard technology attributes, technological accumulation, and industrial demonstration effects.

MiniMax's model capabilities, commercial progress, application scale, and other dimensions are undoubtedly the "outstanding" among domestic AI large model enterprises.

This article is from the WeChat official account "IPO Early Knowledge" (ID: ipozaozhidao), author: Stone Jin, and 36Kr is authorized to release it.