Within a 48-hour window, two &#34;unicorns&#34; in China's large model field submitted their prospectuses to the Hong Kong Stock Exchange.
On December 19 and December 21, Beijing Zhipu Huazhang Technology Co., Ltd. (hereinafter referred to as &#34;Zhipu&#34;) and Shanghai Xiyu Technology Co., Ltd. (hereinafter referred to as &#34;MiniMax&#34;) successively disclosed their Hong Kong stock prospectuses.
These two documents, totaling nearly a thousand pages, showcase the true financials of Chinese AI large model companies with unprecedented granularity, reflecting the fierce collision of two distinctly different commercialization paths: ToC and ToB. Under the spotlight of the capital market, although both companies are in the large model track, they exhibit completely different worlds in terms of technological foundation, revenue structure, customer profiles, and cost control.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/ba3ec415-25f6-405d-8458-52401d9cca90.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/ba3ec415-25f6-405d-8458-52401d9cca90.png"/>
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/2ff53878-55b5-4ed6-a503-2c671ebf821c.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/2ff53878-55b5-4ed6-a503-2c671ebf821c.png"/>
<h2>Debate on the Endgame of Business Models: ToC VS ToB</h2>
MiniMax: Driven by C-end, Betting on &#34;Super Applications&#34;
MiniMax's business logic is closer to that of an internet product company, clearly promoting the banner of &#34;AI Native Products,&#34; with core products including the multimodal interaction platform Talkie (overseas version) / Xingye (domestic version).
According to MiniMax's prospectus, &#34;AI Native Products&#34; are its absolute pillar. In 2023, the revenue from this segment was only $758,000, but it skyrocketed to $21.8 million in 2024, reaching $38.02 million in the first nine months of 2025, accounting for 71.1% of total revenue.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/08cbf9f7-c38c-4ea3-9849-c11fab8942bf.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/08cbf9f7-c38c-4ea3-9849-c11fab8942bf.png"/>
Behind this growth is an explosive increase in user scale. The average monthly active users (MAU) of MiniMax's products like Talkie/Xingye grew from 3.1 million in 2023 to 27.6 million by September 2025. More critically, the number of itspaid usersreached1.7716 millionin the first nine months of 2025, with the average revenue per paying user (ARPPU) increasing from $6 in 2023 to$15**. This indicates that MiniMax has successfully established a C-end closed loop of &#34;user growth-retention-payment.&#34;**
<ul>
<li>
User Data: As of September 30, 2025, its AI native products have cumulatively served over 200 million users, with an average monthly active user (MAU) of 27.64 million.
</li>
<li>
Monetization Model: Mainly monetized through subscriptions, in-app purchases of virtual goods, and online marketing services. Among them, Talkie/Xingye has built strong user stickiness through deep interaction between users and virtual characters
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/1fe10a74-d820-4eb1-a821-c070e97bfe63.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/1fe10a74-d820-4eb1-a821-c070e97bfe63.png"/>
</li>
</ul>
It is worth noting that MiniMax's &#34;specialization&#34; is not only in the C-end App, but its B-end output in the audio modality has become an invisible giant.
The prospectus reveals that its Speech-02 voice model is recognized as one of the top in the world. Data shows that its largest client in 2023 (contributing 37.2% of revenue) is described as a &#34;digital reading and IP development giant headquartered in Shanghai&#34; (Note: corresponding to industry characteristics such as China Literature Group). This means that the AI reading/voice-over functions of a large number of audiobooks, podcasts, and content platforms on the market are fundamentally provided by MiniMax. This model of &#34;C-end products generating traffic and B-end audio generating technical premiums&#34; explains why its open platform can maintain an ultra-high gross margin of nearly 70%.
Zhipu AI: B-end Dominance, Deepening &#34;Local Deployment&#34;
Zhipu AI has taken a typical enterprise service path, with the company acting more like a technology enabler.
As the first company in China to release a trillion-parameter large model GLM-130B, Zhipu AI is building its business landscape around the MaaS (Model as a Service) platform.
As of June 30, 2025, its local deployment (private deployment) revenue reached 162 million RMB, accounting for as much as 84.8% of total revenue.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/3caff5ca-374a-41fa-942f-09bfb6825760.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/3caff5ca-374a-41fa-942f-09bfb6825760.png"/>
This data reflects Zhipu's current customer structure: mainly serving central state-owned enterprises and large institutions with rigid demands for data security and private deployment. As of June 30, 2025, Zhipu has provided services to over 8,000 institutional clients, covering multiple industries such as technology, finance, and public services; its GLM series models have been downloaded over 45 million times in the open-source community.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/e2ba3455-d4e6-44aa-a8d8-f19ad8c07255.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/e2ba3455-d4e6-44aa-a8d8-f19ad8c07255.png"/>
<h2>Technical Background: Full Modality vs Pre-training Framework</h2>
In terms of technical routes, while both ultimately aim for AGI, their focuses have their own merits.
<ul>
<li>
MiniMax: Emphasizes full modality capabilities. The prospectus shows that its model matrix covers text (MiniMax M series), voice (Speech-02), video (Hailuo-02), and music. In particular, its video generation model Hailuo-02 ranked among the top globally in independent evaluation rankings after its release. The company also highlights its proprietary MoE (Mixture of Experts) architecture and linear attention mechanism, aimed at reducing inference costs, which is crucial for large-scale applications on the C-end
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/c88d6a25-9446-4522-9bdf-99a249d0ad23.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/c88d6a25-9446-4522-9bdf-99a249d0ad23.png"/>
</li>
<li>
Zhipu AI: Emphasizes academic foundation and GLM framework. Originating from Tsinghua University's Knowledge Engineering Group (KEG), Zhipu AI began developing the GLM pre-training framework as early as 2020. Its technical highlights include the GLM-4.5 base model and the AutoGLM agent, which can simulate human operations on mobile phones and computers, achieving a leap from &#34;dialogue&#34; to &#34;execution.&#34; Zhipu AI repeatedly mentions &#34;safety&#34; and &#34;controllability&#34; in its prospectus, which aligns closely with its positioning to serve large enterprises and the public sector.
</li>
</ul>
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/e5e54ba9-c2cd-48a8-b3e6-b5bcbe40b2c0.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/e5e54ba9-c2cd-48a8-b3e6-b5bcbe40b2c0.png"/>
<h2>Financial Deconstruction: The Touchstone of Profitability—Gross Margin</h2>
Gross margin is the core indicator of an AI company's ability to command a technology premium. In this disclosure, MiniMax's performance in the API business surprised the market, while Zhipu AI showed signs of fatigue in its cloud business.
MiniMax: Accelerating C-end Monetization, Gross Margin Turns Positive
MiniMax's overall gross margin has experienced a rollercoaster recovery: -24.7% in 2023, turning positive to 12.2% in 2024, and further improving to 23.3% in the first nine months of 2025.
Notably, its B-end business—&#34;Open Platform and Other Enterprise Services.&#34; Although the revenue share of this business has declined, its gross margin reached 69.4% in the first nine months of 2025. This figure far exceeds the industry average, suggesting that MiniMax has achieved significant efficiency advantages in model inference cost control, API pricing power, or technical architecture (such as MoE architecture and linear attention mechanisms), enabling it to support commercial services at a lower computational cost.
The prospectus shows that its open platform has deeply served giants in &#34;digital reading&#34; and &#34;internet services,&#34; indicating that MiniMax not only creates super apps but also acts as the &#34;AI sound engineer&#34; behind content giants like Ximalaya and Yuedu.
At the same time, according to MiniMax's prospectus, the company's revenue has grown rapidly. In 2023, revenue was $34.6 million, and by the first three quarters of 2025 (as of September 30), revenue had reached $53.44 million, a significant year-on-year increase.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/81cf328d-eac0-411d-a3e2-5cb74e426178.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/81cf328d-eac0-411d-a3e2-5cb74e426178.png"/>
Zhipu AI: Steady Growth in B-end, Gross Margin Reaches 50%
The prospectus shows that from 2022 to 2024, Zhipu AI's revenue was 57.41 million yuan, 125 million yuan, and 312 million yuan, with a compound annual growth rate exceeding 130%
As of June 30, 2025, Zhipu AI achieved revenue of 191 million yuan.
Zhipu AI's overall gross margin remains at a high level of 50%, mainly due to the high gross margin of 59.1% for localized deployment business (2025 H1 data). Localized deployment typically includes software licensing and technical services, which have a high premium space.
However, the company's cloud deployment (MaaS) business has raised a red flag. The prospectus shows that the gross margin for this business has dropped from 76.1% in 2022, to -0.4% in the first half of 2025, resulting in a gross loss. Zhipu admits in the prospectus that this is due to &#34;lowering service prices in response to market trends.&#34; This reflects the extremely fierce price war in the domestic large model API market, making it difficult for the pure MaaS model to be profitable at present.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/cdc016b8-ec6f-4b69-8d8e-6adf3074c1b5.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/cdc016b8-ec6f-4b69-8d8e-6adf3074c1b5.png"/>
<h2>The &#34;Arms Race&#34; in R&amp;D Investment: Who is Spending Recklessly?</h2>
The large model industry remains a &#34;money-eating beast.&#34; Both companies have high R&amp;D investment intensity, but Zhipu AI's investment appears to be more aggressive.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/aa10f269-0aab-4c48-a5da-efd122ff2000.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/aa10f269-0aab-4c48-a5da-efd122ff2000.png"/>
Zhipu AI: Oversaturated Investment
In the first half of 2025, Zhipu AI's revenue was 191 million yuan, but during the same period, R&amp;D expenses reached 1.595 billion yuan. This means that for every 1 yuan earned, over 8 yuan is invested in R&amp;D, with an R&amp;D expense ratio of 835.4%. This &#34;oversaturated&#34; investment is mainly used for computing power service fees (1.145 billion yuan) and R&amp;D personnel salaries, demonstrating Zhipu's firm determination to catch up with SOTA (state-of-the-art) model capabilities.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/d62077c5-5628-4612-a119-cf4cfc13181c.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/d62077c5-5628-4612-a119-cf4cfc13181c.png"/>
MiniMax: Efficiency Improvement
Although MiniMax is also burning cash, thanks to the rapid increase in revenue, its R&amp;D expense ratio is quickly converging. In 2023, its R&amp;D expense ratio exceeded 2000%, but by the first nine months of 2025, it had dropped to 337.4% (R&amp;D expenditure of 180 million USD vs revenue of 53 million USD). This indicates that as C-end products ramp up, MiniMax is gradually diluting its massive R&amp;D costs, and the operating leverage effect is beginning to show.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/072951fd-f3be-4c95-bbea-dcc7fea6ad89.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/072951fd-f3be-4c95-bbea-dcc7fea6ad89.png"/>
<h2>Globalization and Regional Risks: Different Battlefields</h2>
In terms of regional distribution, the two companies have chosen different battlefields.
MiniMax: Going Global is Key
MiniMax is a highly globalized company. The prospectus shows that in the first nine months of 2025, its revenue from mainland China accounted for only 26.9%, while revenue from the U.S. market accounted for 20.4%, with other overseas regions also having a high proportion. Its blockbuster product Talkie is aimed at overseas markets.
In contrast, Zhipu AI mainly focuses on the domestic market. Although its prospectus shows some overseas revenue (11.1% from Southeast Asia in H1 2025), its core remains serving government and enterprise customers in mainland China.
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/02c61427-3193-419a-afac-eb96af7d360d.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/02c61427-3193-419a-afac-eb96af7d360d.png"/>
<h2>Capital Landscape: A Shareholder List Full of Giants</h2>
Behind these two unicorns is a significant portion of China's top tech capital.
<ul>
<li>
MiniMax: Founder Dr. Yan Junjie holds shares and controls the company through a voting rights structure. The shareholder list includes Alibaba (Alisoft China) with approximately 13.66% stake, and Tencent with about 2.58%. Additionally, the well-known gaming company miHoYo is also an important shareholder, holding over 5%.
</li>
<li>
Zhipu AI: The shareholder structure is more diverse, combining market-oriented and state-owned backgrounds. In addition to internet giants like Ant Group, Tencent, Meituan, Xiaomi, it has also brought in state-owned platforms such as the Social Security Fund Zhongguancun Innovation Fund and the Beijing Artificial Intelligence Industry Investment Fund, as well as established VCs like Hillhouse Capital and Junlian Capital.
</li>
</ul>
<h2>Summary and Outlook</h2>
Currently, MiniMax has ample cash and financial assets (approximately $1 billion in total), while Zhipu AI's cash and cash equivalents have reached 2.55 billion RMB. In the short term, both companies may have enough &#34;ammunition&#34; to continue this technological marathon.
This submission to the Hong Kong stock market effectively presents two business narratives to investors:
<ul>
<li>
MiniMax Narrative: This is a story about &#34;AI-native super applications.&#34; Investors need to assess whether its C-end products can maintain high retention rates and whether its model reasoning efficiency can continue to lead, ultimately realizing the internet logic of &#34;traffic monetization covering computing costs.&#34;
</li>
<li>
Zhipu AI Narrative: This is a story about &#34;AI new infrastructure.&#34; Investors need to consider how deep its moat is in the B-end market and whether the high margins from localized deployment can offset the substantial investments in basic model R&amp;D in the future.
</li>
</ul>
Who will become the &#34;first stock of global large models&#34; remains uncertain, but these two prospectuses undoubtedly announce that China's AI industry has moved from the technical clamor of the &#34;hundred model battle&#34; into the deep waters of commercialization
<img src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/8d346201-98a8-4b4b-b2df-405b44f02d90.png?x-oss-process=image/auto-orient,1/interlace,1/resize,w_1440,h_1440/quality,q_95/format,jpg" alt="" original-src="https://imageproxy.pbkrs.com/https://wpimg-wscn.awtmt.com/8d346201-98a8-4b4b-b2df-405b44f02d90.png"/>

MiniMax bets on C-end super applications, with paid users exceeding 1.77 million and gross margin turning positive to 23%; Zhipu focuses on B-end local deployment, serving 8,000 institutions with a gross margin as high as 50%. Both companies have R&D investments several times their revenue, backed by top-tier capital such as Alibaba and Tencent, yet they have taken distinctly different commercialization paths for ToC and ToB. In addition, giants like Alibaba, Tencent, and miHoYo are listed as shareholders of both companies

In-depth analysis of the prospectus of AI large model unicorns: MiniMax to C, Zhipu to B