The average person born after 1995 has over one billion dollars in their account, and MiniMax knocks on the door of the Hong Kong stock market.

Source: Machine Heart

I didn't expect the pace of IPOs for large model startups to change so quickly.

On Sunday evening, it was reported that the well-known domestic AI startup MiniMax (Xiyu Technology) published its prospectus information for the Post Hearing Information Pack (PHIP) on the Hong Kong Stock Exchange, marking a crucial sprint phase in its listing process.

MiniMax was founded in December 2021, headquartered in Shanghai, by former SenseTime executives including Yan Junjie, focusing on the research and development of general artificial intelligence technology (AGI) that integrates text, speech, and visual modalities. The company is well-known for its MiniMax M1 and M2 foundational large models as well as AI-native products such as MiniMax Voice, Conch AI, and Xingye.

The company is expected to break records, becoming the AI company with the shortest time from establishment to IPO.

As a global leader in full-modal large model companies, MiniMax has for the first time systematically disclosed its business layout, user scale, financial data, and future strategy in its prospectus, showcasing its strong momentum and unique positioning in the global AI track.

Over 200 million users, with overseas revenue accounting for more than 70%.

MiniMax is a dual-driven company focused on “large models + AI native applications.” The company states that its open platform has become “one of the largest enterprise-level and developer open platforms in the world,” supporting rapid deployment across multiple industries such as smart terminals, healthcare, cultural tourism, and finance.

The prospectus shows that as of September 30, 2025, MiniMax has over 212 million individual users across more than 200 countries and regions, as well as over 100,000 enterprises and developers from more than 100 countries. Among them, the average monthly active users of AI-native products have rapidly increased from 3.144 million in 2023 to 27.622 million in the first nine months of 2025, demonstrating strong user engagement and growth momentum.

A user base of such a large scale is distributed across various corners of the globe: MiniMax disclosed in its prospectus that its overseas market revenue contribution exceeds 70%, demonstrating its significant achievements in international expansion and cross-market commercialization.

Financial data shows that MiniMax achieved a revenue of $53.437 million in the first nine months of 2025, representing a year-on-year growth of approximately 174.7%, demonstrating a rapid growth trend. The revenue mainly comes from two major segments: AI-native product subscriptions and in-app purchases, as well as open platform and enterprise services.

MiniMax adopts the “model as a product” concept, and its efficient operations have led to a healthy financial cycle: its accounts receivable turnover days are only 38 days, far lower than the average of 60-90 days in the AI or SaaS industry, demonstrating excellent cash flow capabilities and business operational efficiency.

In terms of To C, MiniMax directly serves global consumers through its native products such as Hai Luo AI, Xing Ye / Talkie (AI social), and MiniMax Voice. As of September 30, To C revenue has increased by 181% year-on-year, and the number of paying users has skyrocketed 15 times in less than two years.

In the To B aspect, MiniMax empowers multiple industries through an open platform, providing leading model capabilities to enterprises and developers via API. Its open platform processes over one trillion Token requests daily, achieving a 160% high growth in To B revenue during the same period, with a gross profit margin of 69.4%, demonstrating strong profitability.

MiniMax has also established a collaborative ecosystem with leading technology companies both domestically and internationally. Overseas, its models are launched on the global AI platforms of the three major cloud providers: Amazon, Google, and Microsoft; providing video and voice technology for LinkedIn, Monks (a digital creative company), and others. Domestically, MiniMax's technology serves the core products of giants like Alibaba, Tencent, ByteDance, Xiaomi, and Kingsoft Office, and provides humanoid interaction capabilities for humanoid robots like Zhiyuan Robot. This extensive and in-depth industrial integration proves the practicality and reliability of its technology.

The prospectus shows that MiniMax's adjusted net loss is nearly flat in 2025 compared to the same period last year, achieving a significant reduction in losses amidst rapid growth. This is attributed to a diversified revenue model and efficient expense inputs — as of September 2025, the company's R&D spending increased by 30% year-on-year, while sales and marketing expenses decreased by 26%.

Omni-modal AI technology layout

The success of MiniMax is built on a clear and efficient path: its forward-looking “multi-modal” technology matrix has established a solid underlying competitive advantage.

MiniMax is one of the few large model companies that have focused on the research and development of full-modal models since its establishment. Since its inception, MiniMax has undergone intensive technological iterations, achieving breakthrough advancements in models related to speech, video, and text, frequently attracting attention from the tech industry.

Starting in 2023, MiniMax launched the country's first speech large model Speech 01 based on the Transformer architecture, with an upgraded version Speech 02 released in 2024, ranking first in overall performance. To date, the MiniMax speech model has helped users generate over 220 million hours of speech.

In August 2024, MiniMax released the video generation model Video 01 and the video generation product Hailuo AI. In June 2025, the upgraded version Hailuo 02 was launched, ranking second in the AA video arena. As of now, the MiniMax video model has helped creators generate over 590 million videos.

In October 2025, MiniMax released and open-sourced the next-generation text large model MiniMax M2, ranking among the top five globally in Artificial Analysis and first in open-source, marking the first time a Chinese open-source large model has made it into the global top five on this list.

It can be seen that its research and development is not a breakthrough at a single point, but rather a systematic advancement across four key modalities: text, video, voice, and music, forming a complete and mutually collaborative technological matrix.

Among them, the newly released and open-sourced next-generation text large model MiniMax-M2 by MiniMax on October 27 this year is particularly noteworthy.

The inference speed of this large model is nearly twice that of Claude Sonnet 4.5, with performance comparable and a price only 8% of it. At the AWS re:Invent 2025 held in December, Amazon announced that its fully managed generative AI cloud service Amazon Bedrock introduced MiniMax-M2 as a representative of domestic models.

MiniMax-M2 not only excels at planning and executing long chain tool invocation tasks stably but can also coordinate calls to Shell, browsers, Python code interpreters, and various MCP tools.

The key technology adopted by MiniMax M2, “Interleaved Thinking,” has been widely discussed in the overseas AI research community. This technology allows the model to continuously accumulate contextual understanding in a closed loop of “thinking - acting - reflecting,” and to adjust strategies in real-time based on feedback. This mode of operation, which is closer to that of real engineers, significantly enhances the execution capabilities of MiniMax M2's Agent, resulting in stronger planning in complex tasks, higher execution robustness, and more reliable self-correction abilities, thereby forming its most recognizable core advantages.

During the release period of the M2 model, it quickly climbed to the first place in domestic model token usage on the global model aggregation platform OpenRouter and ranked third in global token usage in programming scenarios.

The investment in research and development is enormous, and the pursuit of efficiency is extreme.

Similar to most AI large model companies, MiniMax is currently still in a stage of continuous high R&D investment, rapidly iterating on large models, and has not yet achieved profitability.

In the first nine months of 2025, the company's net loss reached $186 million (adjusted net loss, a Non-IFRS metric). Despite achieving over 170% explosive growth in revenue, the adjusted net loss only increased slightly by 8.6%. The losses were primarily due to R&D expenses, changes in the fair value of financial liabilities, and expenses related to the listing.

In the first nine months of 2025, R&D spending reached $180 million, accounting for 337.4% of total revenue; the R&D expense ratio has decreased from over 2000% in 2023 to 337.4%. This change reflects the operating leverage brought about by the rapid expansion of revenue scale on the one hand, and on the other hand, it also indicates the company's ongoing investments in key areas such as model iteration and infrastructure.

It is worth noting that, alongside significant R&D investments, the organizational and business efficiency demonstrated by MiniMax is exceptionally prominent.

The company's total employee size is approximately 385 people, of which the R&D team accounts for about 300 people, nearly 80% of the total staff, with core members coming from top global tech companies such as Microsoft, Google, Meta, and Alibaba.

Its research and product teams are overall very young, primarily consisting of those born after 1995, and have absorbed a large number of engineers born after 2000. This structure aligns closely with the current paradigm of AI research—having a natural understanding of new tools, automated processes, and agent-based, AI-native work methods, which significantly amplifies the output of individual labor.

In the R&D model, there are no more than 3 layers of administrative command below the CEO, and the project-oriented approach, along with “minimalism,” greatly shortens the implementation path, allowing the company to maintain a high cost-effectiveness ratio in a competition that is highly intensive in computing power and financial investment.

Since its establishment, MiniMax is expected to spend a total of 500 million USD by September 2025, achieving:

A leap from leading in single modality to leading in all modalities.

Three modalities in the international first tier, and there are breakout products.

Every year there are breakthroughs, continuously stepping up.

In the first nine months of this year, the loss of 186 million is still less than the quarterly investment costs of some major internet companies' large models.

This efficiency advantage is also reflected in recent financial trends.

In the first nine months of 2025, the company's R&D expenses grew by only about 30% year-on-year, while revenue increased by 174.7% in the same period. Sales and marketing expenses decreased by 26%, which is uncommon for an AI company still pursuing high-speed growth.

This “scissors gap” indicates that the core driver of growth is not simply “burning money for scale” — more comes from improvements in model capabilities, the spread of product reputation, and the release of organizational efficiency, rather than relying on large-scale capital inflows and market subsidies.

“Post-95s” at the helm

According to the information disclosed in the prospectus, the composition of MiniMax's board of directors is highly consistent with the company's characteristics of being “youthful and highly technology-driven.”

The average age of the four executive directors is only 32 years old (“post-95s”), which is extremely rare in the history of Hong Kong listed companies.

They are deeply embedded in the company's front-line research and development as well as business advancement, and are the relevant direct responsible persons, including:

The company's founder & CEO Yan Junjie (36 years old), COO Yun Yeyi (31 years old), head of research and engineering for large language models Zhao Pengyu (29 years old), and head of research and engineering for visual models Zhou Yucong (32 years old).

Non-executive directors and independent non-executive directors take on more roles in governance, supervision, and institutional checks and balances.

This board composition is highly aligned with a young AI company that is still in the rapid technological evolution stage, led by founders and engineering systems, and is consistent with its overall strategy emphasizing long-term technological investment and high organizational efficiency.

MiniMax stated in the prospectus that the funds raised from this listing are intended to be used mainly for the following aspects:

About 70% will be used for R&D in the next five years, including large model upgrades and AI native product development;

Approximately 30% for working capital and general corporate purposes.

The company stated that it will continue to promote the vision of “intelligent levels continuously improving and everyone benefiting,” enhancing social productivity and personal quality of life through technological inclusiveness.

The story of AGI should be young.

MiniMax's sprint into the Hong Kong stock market is significant not just as the capitalization of another AI unicorn. It is telling a younger story about AGI.

In this young team, a group of people is creating a new type of business with AI-native thinking, making their story bold enough - firmly betting on all modalities and globalization. This reckless sprint is bound to bring both challenges and opportunities, but perhaps it is also the right way to AGI.

As reflected in the MiniMax prospectus, with the continuous development of technology and business, this company will continue to break through and realize its long-term vision: Intelligence with Everyone.