In-Depth, Objective, Insightful

Boundless Energy, Serendipitous Encounters!

The Right Moment, Yet Time Slips Away!

Editor: Junyi

Reviewer: Kele

Source: Shoucai - Shoutiao Finance Research Institute

A successful listing marks just the start of a value marathon.

Time races on like a swift steed, and before we realize, 2026 is halfway over. Zhipu and MiniMax, the two large-scale model giants listed in January, have nearly simultaneously faced their first wave of lifted bans. Despite being in the same sector and timeframe, the market's reaction has been vastly different.

Let's start with Zhipu: On July 8th, 25.68 million restricted shares were released, representing a market value exceeding HK$40 billion. The stock opened lower but quickly rebounded, surging 19% at one point during the session and closing with a 13.35% gain, pushing the total market value past HK$800 billion. The next day, it rose another 11.3%. JPMorgan Chase also raised its target price to HK$2,000, reaffirming an overweight rating.

Now, let's examine MiniMax: On July 9th, the day the ban was lifted, the stock plummeted over 17.98%, followed by additional declines of 9.68% and 17.13% on the 10th and 13th, respectively. Fortunately, it recovered over the next three days, but the closing price of HK$256.00 on July 16th was still down over 30% from HK$362.6 on July 8th.

It's worth noting that both companies went public on the Hong Kong Stock Exchange nearly simultaneously earlier in the year. At the time, the market seemed more optimistic about MiniMax's multimodal capabilities. On its debut, the stock surged 109%, with a market value exceeding HK$100 billion, while Zhipu nearly fell below its issue price, with a market value of around HK$57.9 billion. Now, just over six months later, Zhipu has surpassed HK$720 billion, while MiniMax remains below HK$100 billion.

From leading by a wide margin to a significant gap, questions arise: Both are in the large model sector, both follow an open-source approach, and both discuss agents and coding. With so many similarities, why is capital so selective? Is Zhipu overvalued, or is MiniMax undervalued? What is the market hesitating about?

1 Two Dimensions and Two Faces: No Need for Excessive Worry

Intuitively, the answer may lie in the two dimensions of "quantity" and "people" related to the lifted bans.

First, let's discuss quantity. Zhipu's lifted ban involved 25.68 million shares, accounting for nearly 6% of its total share capital, a relatively limited proportion with manageable impact. In contrast, MiniMax's lifted ban involved a staggering 146 million shares, approximately 45% of its issued shares, far exceeding Zhipu in both absolute numbers and proportion.

Further analysis from brokerage perspectives reveals even greater pressure. CICC points out that MiniMax's lifted shares on July 9th accounted for about 63% of its Hong Kong-listed share capital, with financial investors holding over one-third. Meanwhile, Zhipu's lifted shares on the same day accounted for only about 11.6%.

Now, let's talk about people. The market not only focuses on the scale of lifted bans but also on the nature of the shareholders involved. Many of Zhipu's shareholders involved in the lifted ban have state-owned backgrounds, industrial capital attributes, or long-term strategic intentions. These shareholders tend to focus on long-term investments, making large-scale short-term sales less likely even after the lock-up period ends, alleviating market concerns about "immediate selling upon ban lifting." Before the lifted ban, several institutions, including the Beijing AI Industry Investment Fund, stated their intention to continue holding shares. According to Securities Daily, these institutions collectively hold about 70% of the shares being lifted, providing reassurance to the market.

In contrast, MiniMax's lifted ban list includes not only industrial or strategic shareholders like Alibaba, Tencent, Chia Tai Group, and CCTV Media Convergence Industrial Fund but also a large number of VC, PE, fund, and asset management institutions, including Miheng under Hillhouse, Himalia and HSG under Sequoia and Sequoia China, as well as MPC, Future Capital, China Life funds, Planetree, Vitalbridge, CICC-related funds, and Janchor. Additionally, IPO cornerstone investors holding about 16.5 million shares, accounting for about 5.34% of the total post-listing share capital, also had their lock-up periods lifted after July 8th.

The difference in shareholder structures directly affects market expectations. While these institutions may not necessarily sell their shares, the market often prices in risks based on the "possibility of selling." Especially for VC and PE funds, which have defined lifespans, lifting the ban is a core exit channel for realizing investment returns. Moreover, despite the significant decline, MiniMax's current stock price is still notably higher than its issue price of HK$165, likely resulting in substantial paper gains for early investors. Thus, it's not surprising that the market shows more concerns.

Industry analyst Sun Yewen states that the stark differences in the scale of lifted bans and shareholder structures justify the market's divergent pricing. According to HKEX rules, MiniMax can only welcome southbound funds in August at the earliest, leaving the market without stable incremental funds to absorb the lifted shares. Combined with the over-30% proportion of financial investors, real exit pressure exists, making it easier for the stock price to fall than rise. However, investors need not worry too much, as the stock price is likely to stabilize and rebound once emotions subside. Nevertheless, MiniMax's weakness wasn't triggered solely by the lifted ban; its stock price has been oscillating downward since its March peak, with the lifted ban merely amplifying the decline.

2 BC-End Challenges: The Pros and Cons of Multiple Storylines

The pendulum effect reminds us that stock price fluctuations are normal. To assess a stock's true value, we must ultimately look at its fundamentals.

In terms of path, MiniMax stands out among domestic large model companies. Unlike DeepSeek, which relies on community feedback for model iteration, or Zhipu, which depends on B-end enterprise data for optimization, MiniMax chooses to self-develop a full-modal large model as its foundation, handling language, video, voice, and music independently, a relatively heavy approach. On this basis, the company pursues two businesses: launching a series of AI-native apps primarily for overseas markets targeting individual users and building an open platform to output model capabilities through APIs based on usage volume for enterprises and developers.

From a developmental perspective, MiniMax has undergone a strategic shift from starting with B-end services to a comprehensive C-end globalization approach. This shift significantly drove rapid revenue growth but also made revenue more dependent on the C-end. According to the 2025 financial report, the company's AI-native product (C-end) and open platform (B-end) revenues were RMB 362 million and RMB 177 million, respectively, accounting for 67.2% and 32.8%. The underlying operational logic is: models produce products, products acquire user data globally, and data feeds back into model iteration, forming a virtuous cycle.

The direction is sound, but specific implementation hurdles remain.

For instance, at the model layer, while MiniMax has notable strengths, it still lacks irreplaceability. MiniMax has made achievements in vertical capabilities, such as its Speech-02 voice synthesis model topping the Artificial Analysis global TTS rankings and its MiniMax-01 series text model known for its ultra-long 4-million-token context window. However, the effectiveness of these parameters in enhancing user stickiness needs improvement.

After all, the sector is evolving rapidly, with intense technological competition among leading large models. "Leading this month and being surpassed next month" is the norm. Single-point advantages are insufficient to build a moat; sustained leading engineering capabilities coupled with long-term user stickiness are key. From market feedback, MiniMax's stability in entering the first tier in terms of model recognition and user mindshare needs enhancement.

Fortunately, at the product layer, running four parallel lines helps meet more user needs and retain a larger market. However, whether the data links can truly be connected remains to be seen over time. For example, Hailuo AI focuses on video creation, while Talkie and Xingye specialize in AI role-playing and emotional companionship, with significant differences in user behavior. Talkie's data can feed back into the emotional companionship model itself but has limited impact on enhancing core capabilities like programming and reasoning. Cross-product and cross-modal data flow remains uncertain. While the unified multimodal direction is correct, the dividing line lies in whether it can achieve a large-scale integration of engineering, productization, and commercialization.

In the fourth quarter of 2025, MiniMax's AI-native product revenue growth rate dropped sharply from 181% in the first three quarters to 82%. Meanwhile, the lowest C-end package tier increased from RMB 29 to RMB 49 per month, sparking accusations of "disguised price hikes." Caution is needed regarding price-sensitive users voting with their feet, affecting retention and paid conversion rates, and potentially loosening the C-end foundation.

Notably, Talkie's monthly active users declined significantly in the fourth quarter of 2025. Hailuo AI also lost its leading position on Artificial Analysis's video model rankings, surpassed by Alibaba's HappyHorse, ByteDance's Seedance, and Kuaishou's Kling, prompting the market to reevaluate MiniMax.

Facing C-end pressures, the urgency of B-end breakthroughs becomes apparent. MiniMax desperately needs a benchmark product that technically rivals international giants while offering cost advantages for domestic substitution. The M3 model released on June 1st is thus highly anticipated. It is the first domestic open-source model with cutting-edge programming capabilities, a 1-million-token ultra-long context window, and native multimodal support. In SWE-Bench Pro evaluations, M3 surpassed GPT-5.5 and Gemini 3.1 Pro, approaching Opus 4.7. It also led in OmniDocBench, providing a narrative foundation for competing with top overseas models.

However, to establish a foothold in the B-end market, parameters alone are insufficient; customers prioritize the performance-to-price ratio. While M3's performance improved, its price also increased: compared to M2.7, M3 (≤512k) prices doubled after the 7-day discount period, and long-context scenarios saw a threefold increase. Horizontally, M3 adopts a strategy of "benchmarking overseas models while offering domestic premiums," with performance close to Claude Opus 4.7 but pricing well below Anthropic and OpenAI, making it a rare "high-performance alternative" for enterprises facing compliance or data export restrictions. However, its sustained attractiveness in domestic competition remains to be seen over time.

In 2025, MiniMax's revenue surged 158.95%, but net profit plummeted 302.29%. In terms of scale, its RMB 790.38 million in revenue was less than one-ninth of Zhipu's RMB 724 million, while its over RMB 1.8 billion in losses exceeded one-third of Zhipu's.

Industry analyst Wang Tingyan states that while MiniMax's diverse storylines demonstrate vitality, market confidence, and growth potential, they also risk lacking clarity, affecting focus on the core business. Simultaneously narrating large models, multimodality, video, audio, AI companionship, overseas C-end products, agent platforms, coding, and API commercialization creates too many anchors, effectively meaning none. The high valuation MiniMax received at listing was largely bet on "C-end globalization." The decline in C-end data is naturally not a plus, compounded by the vulnerability of overseas revenue amid tightening regulations. Which business line can support the valuation? Can C-end growth sustainably subsidize model training costs? These questions lack clear answers.

3 ARR Surpasses $150 Million: Ambitions for the Largest Open-Source Weight Model

Of course, few businesses navigate smoothly. Continuously identifying and resolving issues, rising through challenges, is the norm. Looking at MiniMax again, its ability to emerge from the crowd and join the top tier of large model companies indicates underlying value potential and core strengths.

If the complexity of its business narrative leaves the market uncertain about MiniMax's long-term path, its sustained financial improvements and accelerated technological iterations offer an alternative narrative: a growth story climbing out of losses, with thickening technological barriers and long-term value.

Financially, MiniMax has shown clear signs of improvement: total revenue reached $79.038 million in 2025, up over 158% year-on-year. By February 2026, annual recurring revenue (ARR) surpassed $150 million, nearly doubling its 2025 full-year revenue, showcasing visible growth acceleration.

It's important to note that revenue represents historical income, while ARR estimates future twelve-month revenue based on current contracts and subscriptions, better reflecting business momentum and sustainability. At this growth rate, MiniMax's 2026 revenue could very likely double again.

Cost-side changes are equally encouraging: overall gross margin jumped from -24.68% in 2023 to +25.4% in 2025, a significant improvement. The enterprise-focused open platform and services segment maintained a high gross margin around 70%, while the AI-native product line achieved a historic turnaround from -380.2% to +4.7%, marking its first profit, indicating that the consumer business is gradually reducing reliance on external financing and stabilizing its operational foundation.

On the expenditure front, the sales expense ratio witnessed a dramatic decline, dropping from 659.74% in 2023 to 65.66% in 2025. This shift reflects a growing reliance on word-of-mouth and organic traffic for customer acquisition, indicating a reduced inclination towards investing in paid traffic. Excluding non-cash factors such as financial liability losses arising from fair value changes of preferred shares, the company's adjusted net loss stood at a mere $282 million in 2025, marking a significant reduction from $465 million in 2024. Moreover, the quality of revenue is on the rise, with an increasingly positive correlation observed between business operations, technological advancements, and financial performance.

Technologically speaking, new horizons are unfolding. According to sources like Guandian Network and The Information, MiniMax Group is in the process of developing a colossal 2.7-trillion-parameter large language model, internally dubbed "M3 Pro," with a potential launch as early as the third quarter of this year. If successful, this model could emerge as the largest open-source weight model released by a Chinese company and rank among the largest globally.

Early training feedback has been promising, with model convergence exceeding expectations and several key metrics surpassing internal benchmarks. This progress is attributed to MiniMax's continuous refinement of its proprietary MSA architecture. The MSA 2.0 version is currently being rapidly implemented, with a focus on enhancing long-sequence task processing capabilities, reducing computational load in attention mechanisms, and improving hardware utilization during large-scale training.

In terms of capital channels, new opportunities have also arisen. In June 2026, the Shanghai Stock Exchange revised its relevant regulations, paving the way for large model companies to apply for listing under the fifth standard of the Science and Technology Innovation Board. In May, MiniMax signed a coaching agreement with CITIC Securities, officially embarking on the A-share listing process. Achieving an A+H dual listing will further broaden the company's capital acquisition channels, providing more robust financial support for long-term R&D investment and market expansion endeavors.

4 Oversubscribed Financing + Collective Alignment: How Close Are We to a Turnaround?

Investing is all about having faith and betting on the future. Amidst market concerns over the potential impact of share lock-up expirations, a new round of oversubscribed financing and collective alignment from core shareholders undoubtedly serve as a significant confidence booster.

On July 10, MiniMax announced the successful completion of a new financing round, raising a total of HK$16 billion. This round attracted participation from over a hundred global institutions and achieved approximately seven times oversubscription. According to the announcement, MiniMax plans to issue 35.6 million new shares at a placement price of HK$268 per share, representing a discount of roughly 9.89% from the previous day's closing price. Additionally, the company will issue HK$6.5 billion in zero-coupon convertible bonds maturing in 2027.

Despite the discounted issuance, the sevenfold subscription rate genuinely reflects the prevailing supply and demand sentiment, with buyers far outnumbering sellers. Participating institutions include multiple international sovereign funds, long-term funds, top-tier Chinese institutions, and elite multi-strategy funds, with over 20 long-term and sovereign funds among them. The high caliber of participants offers the market a valuable window to reassess core AI assets in China.

The collective alignment of existing shareholders also instills confidence. As the lock-up expiration window approaches, institutional investors such as Aspex, Boyu, IDG, Janchor, and Martis Fund, along with state-owned entities like China Life Investment and Xuhui Capital, as well as early shareholders including Alibaba, miHoYo, Yunqi Capital, and Future Capital, have collectively chosen to maintain their stakes.

Two strategic shareholders, Alibaba and miHoYo, directly stated their commitment to continuing their long-term cooperation with MiniMax and emphasized that short-term stock price fluctuations would not influence their holding stance. Goldman Sachs, Bank of America, and Citigroup, three prominent international investment banks, also consecutively issued "Buy" ratings. Against the backdrop of deep differentiation in the current AI sector, such coordinated moves are indeed rare.

The collective morale boost from both new and existing capital essentially represents a re-evaluation of the AI sector by capital. As the industry transitions from a phase of intense competition among numerous models to a knockout stage, the market no longer rewards mere PowerPoint presentations and impressive parameter counts. Instead, it will only cast votes of confidence for companies that demonstrate high-quality R&D investment, stable team structures, and actionable commercialization paths. From this perspective, regardless of the temporary decline during the lock-up expiration period or its market capitalization of less than RMB 100 billion, MiniMax is clearly undervalued.

On the same day as the financing announcement, Yan Junjie, the founder and CEO of MiniMax, sent an internal letter to all employees. The content was concise, focusing on two long-term commitments: from that day until the company achieves Artificial General Intelligence (AGI), he would forgo any salary from the company; over the next four years, he would allocate shares equivalent to 4% of the total share capital under his name to incentivize the team, along with establishing a 1% special fund to support the open-source community.

Generally, founders forgoing salaries is a practice typically observed during the startup phase or in times of crisis. However, with the company having just secured HK$16 billion in capital and being a leader in the industry, this choice carries far more strategic than financial significance. It serves as a declaration to the market that MiniMax will not deviate from its long-term technological roadmap for the sake of short-term market capitalization management. Instead, all resources and attention will be fully dedicated to the long-term race. While it may appear to be a gamble, it is actually a display of confidence and strength.

There was a time when the mere mention of lock-up expirations would send the market into a panic. However, at ShouCai, we believe that lock-up expirations serve as a litmus test, revealing a company's true quality. Those with strong fundamentals often weather these periods with ease and may even buck the trend. Even if there is a temporary downturn, it serves to weed out short-term profit-seekers or those lacking resolve.

From this perspective, lock-up expirations should be viewed as a long-term observation window, and it might be prudent to let events unfold further. For instance, MiniMax has witnessed an upward trend over the past three trading days, while Zhipu has experienced volatile declines. Fundamentally, the divergent stock price movements of the two companies appear to stem from differences in their lock-up structures but are actually a reflection of the market's vote on their respective capital narratives. Once short-term selling pressure dissipates, what will ultimately determine the heights these two companies can reach are the solidity of their fundamentals, the irreplaceability of their core barriers, and the precision and foresight of their strategic layouts.

They say that true gold fears no fire and that prosperity often follows adversity. While a high proportion of lock-up shares may exert a strong short-term impact, it may not necessarily be detrimental in the long run. A one-time emotional clear-out can facilitate the company's subsequent steady and sustained growth. Hopefully, after this ordeal, MiniMax can travel light and embark on a trajectory of matching market capitalization growth.

This article is original content from ShouCai.

Please leave a message if you need to reprint it.