每经AI快讯，"全球大模型第一股"智谱(02513.HK)在港交所主板挂牌上市，发行价为每股116.20港元， 以120港元/股的价格开盘后，最高冲至121港元/股，但是之后震荡下行，最终在9：50分跌破发行价， 跌至116.10港元/股，后有所回升。 ...

Summary of Conference Call Records Companies and Industry - **Companies Involved**: Zhipu, MiniMax, and SenseTime - **Industry Focus**: AI and large model technology, targeting both B-end (business) and C-end (consumer) markets Key Points and Arguments Zhipu - **Business Model**: Operates an integrated MaaS (Model as a Service) platform that supports both local and cloud deployments, with local deployment revenue accounting for approximately 85% of total revenue [1][5] - **Financial Performance**: - Revenue compound annual growth rate (CAGR) from 2022 to 2024 is 130% - 2024 revenue reached 300 million RMB, with a net loss of 2.5 billion RMB in 2024 and 1.75 billion RMB in the first half of 2025, indicating a 70% year-over-year increase in losses [1][12] - Maintains a gross margin above 50%, with 2024 gross margin at 56% and 2025 first half at 50% [12] - **Customer Growth**: Customer count for local deployment increased from 48 in 2022 to 123 in 2024, with annual customer spending rising from 1.14 million RMB to 2.15 million RMB [5] MiniMax - **Business Model**: Focuses on AI-native applications and an open platform for enterprise services, with over 70% of revenue coming from AI-native applications [1][6] - **Financial Performance**: - 2024 revenue of 30 million USD (approximately 200 million RMB), an increase of nearly 8 times year-over-year [1][13] - First three quarters of 2025 revenue reached 50 million USD (approximately 360 million RMB), a 175% year-over-year increase [13] - Gross margin improved from -12% in 2023 to 23% in the first three quarters of 2025 [14] - **User Engagement**: Monthly active users for Talkie reached 20 million, contributing significantly to revenue [6] SenseTime - **Business Model**: Combines software and hardware, focusing on B-end large model applications, with a strong computational infrastructure [2][10] - **Financial Performance**: - Generated 1.8 billion RMB in revenue from generative AI in the first half of 2025, a year-over-year increase of over 70%, accounting for 77% of total revenue [4][15] - Gross margin around 40%, with significant improvements in trade receivables [15] - **Computational Infrastructure**: Operates a substantial computational center with over 25,000 units, primarily using NVIDIA cards [10] Market Dynamics - **B-end vs. C-end Performance**: B-end commercialization is progressing faster than C-end, with Zhipu's revenue tripling in the first half of 2025. MiniMax's C-end product penetration is only 0.9%, significantly below the global average of 3% [16] - **Global C-end AI Product Potential**: Approximately 1.7 to 1.8 billion people have interacted with AI tools, but the overall payment penetration rate is only about 3%, compared to over 20% for other consumer products [17][18] Valuation and Future Outlook - **Valuation Estimates**: - MiniMax's IPO valuation is estimated between 46 billion to 50 billion HKD, while Zhipu is around 51 billion HKD [20] - SenseTime's market value is approximately 95 billion HKD, with a projected revenue of 5.5 billion RMB from generative AI in 2026 [20] - **Investment Potential**: If these companies maintain high revenue growth, their future prospects appear promising, with SenseTime being considered undervalued [22] Additional Important Information - **Training Costs**: MiniMax's training costs for computational power reached 142 million USD in the first three quarters of 2025, indicating a strong focus on maintaining technological leadership [19] - **Commercialization Challenges**: The C-end market remains in an exploratory phase, with unclear monetization paths for AI products [18]

Core Insights - Zhipu AI, known as the "first global AI large model stock," debuted on the Hong Kong stock market with an issue price of HKD 116.2 per share, resulting in a market capitalization exceeding HKD 51.1 billion, leading to approximately 89% paper profit for recent state-owned investors [1] - The company has raised over CNY 8.3 billion since its establishment in 2019, with over 90% of this funding occurring after the release of ChatGPT [1] Investment Background - Zhipu AI was founded in June 2019, emerging from Tsinghua University's Knowledge Engineering Laboratory, with a core team primarily composed of Tsinghua alumni [1] - Initial funding came from Zhongke Chuangxing and Dacheng Capital, who recognized the potential in natural language processing and knowledge graphs, leading to an investment of CNY 40 million in 2019 [1][2] Strategic Positioning - In May 2020, following the release of GPT-3, Zhipu AI decided to enter the large model sector, leveraging its founder's connections to top research resources [2] - The company completed three rounds of financing before the launch of ChatGPT, with significant contributions from various investors, including a CNY 20.37 million investment in intellectual property [2][3] Funding and Valuation Growth - Zhipu AI's valuation skyrocketed from CNY 2.11 billion in the B1 round to CNY 24.38 billion before its IPO, with a total of CNY 8.36 billion raised, primarily between 2023 and 2025 [5] - Major investments in later rounds came from industry capital and state-owned enterprises, with Ant Group, Meituan, and Tencent among the key investors [5][6] Operational Challenges - The company's operational costs, particularly for computing resources, have surged from CNY 15 million to CNY 1.55 billion projected for 2024, raising concerns about cash flow sustainability [6] - Despite a projected revenue of CNY 312 million for 2024, the high costs associated with computing resources pose significant challenges for the company [6][8] Market Position and Future Outlook - Zhipu AI's revenue is heavily reliant on government and large enterprise clients, with over 80% of its income coming from localized AI solutions [8] - The company faces the challenge of converting its resource advantages from state and industry capital into a sustainable business model while navigating competition from major tech firms [8]

作者 | 褚杏娟 今天，北京智谱华章科技股份有限公司（以下简称"智谱"，股票代码 02513）在港交所正式挂牌上市。发行价为 116.2 港元 / 每股，募资规模预计将达 43 亿港元，预估市值 511 亿港元。开盘后，智谱股价涨超3%，但随后有所回落。截至发文，股票价格为116.5 港元 / 每股。 此时上市，会太早吗？ "全球大模型上市第一股"的头衔无疑让智谱获得更多的出圈效应，可以获得更多的潜在用户。而选择此时冲击上市，郭炜认为，主要还是因为智谱需要 更多的现金流与确定性。 他分析称，大模型不是互联网产品那种"先增长后盈利"的轻资产模式，它更像"研发 + 算力 + 工程化"三台发动机一起烧油才能在大模型军备竞赛里获得 生存空间。私募融资能解决一段时间的研发成本，但每一轮都要讲更大的故事、给更高估值，整体公司和股东可获得收益的确定性越来越少。所以，智 谱上市代表着在估值没有触达最高的时候，把确定性先确定下来。从资本运作来讲，这不失为一步好棋。 不过，郭炜认为，500 亿左右人民币预估市值，对于未来中国大模型头部企业来讲，还是过低了。 | 营销 | | | --- | --- | | 02513 交易中 0 ...

Core Viewpoint - The Hong Kong stock market experienced a short-term pullback, particularly affecting leading tech stocks, with the Hong Kong Internet ETF (513770) declining by 2.54% while seeing a net inflow of 272 million yuan during the dip. The fund's size reached a historical high of 13.305 billion yuan [1][3]. Group 1: ETF Performance and Composition - The Hong Kong Internet ETF (513770) and its linked funds passively track the CSI Hong Kong Internet Index, with major holdings including Alibaba-W, Tencent Holdings, and Xiaomi Group, which collectively account for over 78% of the top ten weighted stocks [3][4]. - The top ten weighted stocks in the ETF include Tencent Holdings (15.42%), Alibaba-W (14.50%), and Xiaomi Group (13.11%), indicating a strong concentration in leading tech companies [4]. Group 2: Market Trends and Institutional Insights - Southbound capital showed a net inflow of 9.178 billion yuan, with Tencent Holdings and Xiaomi Group receiving significant increases in investment, amounting to 1.955 billion and 1.633 billion HKD respectively [4]. - Recent reports highlight the surge in revenue for Kuaishou's AI mobile platform, with daily income increasing by 102% compared to December 2025, positioning Kuaishou as a leader in generative AI [5]. - Citigroup and JPMorgan have identified Tencent Holdings and Alibaba-W as core AI concept stocks, with positive outlooks on their AI development and cloud revenue growth [5]. - Industry analysts suggest that leading internet companies in China are poised for a resurgence, driven by the deepening of AI applications and potential upward revisions in profit expectations and valuations [5][7].

Core Viewpoint - The rise of domestic general GPU companies is highlighted as they support the underlying computing power for large model companies, marking a shift in the Chinese large model industry from "concept validation" to "engineering implementation" [2][17]. Group 1: Industry Overview - The competition among large models fundamentally revolves around computing infrastructure, with domestic general GPUs emerging as a critical industrial base as reliance on imported GPUs and cloud rentals is deemed unsustainable [2]. - The Chinese general GPU market is projected to grow at a compound annual growth rate (CAGR) of 72.8% from 2022 to 2024, and 33.0% from 2025 to 2029, with domestic GPU market share expected to rise from 8.3% in 2022 to over 50% by 2029 [5]. Group 2: Company Positioning - The company, Tensu Zhixin, aims to be a key provider of domestic computing power for large models, with a diverse client base including cloud service providers, AI developers, and various industries [3]. - Tensu Zhixin's training general GPU products hold the third-largest market share in China, while its inference GPUs rank second, positioning the company to benefit from a growing market [6]. Group 3: Competitive Advantages - The company has established a "three-fold moat" consisting of deep hardware-software collaboration, comprehensive full-stack solution capabilities, and a forward-looking ecological layout [6][10]. - The company employs over 480 R&D professionals, with a significant portion having over ten years of industry experience, and holds numerous patents covering key technologies [7]. Group 4: Financial Performance - From 2022 to 2024, the company's revenue is expected to grow from RMB 189.4 million to RMB 539.5 million, reflecting a CAGR of 68.8%, with a revenue of RMB 324.3 million recorded in the first half of 2025, a 64.2% year-on-year increase [12]. - The proportion of R&D expenditure relative to revenue has decreased significantly, indicating improved efficiency in R&D activities [12]. Group 5: Future Growth Prospects - The company is actively developing next-generation products, with plans for multiple product launches between 2025 and 2027, ensuring a continuous growth trajectory [15]. - The customer base has expanded from 22 in 2022 to 106 by mid-2025, demonstrating strong market penetration and diversification of revenue sources [13][14].

Group 1 - The Hong Kong stock market experienced a short-term pullback, with leading tech stocks declining, while the Hong Kong Internet ETF (513770) saw a price drop of 2.54% but recorded a net inflow of 272 million yuan during the dip, reaching a historical high of 13.305 billion yuan in fund size [1][3] - The Hong Kong Internet ETF (513770) passively tracks the CSI Hong Kong Internet Index, with major holdings including Alibaba-W, Tencent Holdings, and Xiaomi Group-W, where the top ten weighted stocks account for over 78% of the total [3][4] - Notably, southbound funds showed a contrary net inflow of 9.178 billion yuan, with Tencent Holdings and Xiaomi Group-W receiving significant increases in positions of 1.955 billion and 1.633 billion HKD, respectively [4] Group 2 - The AI video featuring dancing pets has gained significant traction in overseas markets, leading to a 102% increase in daily revenue for Kuaishou's AI mobile platform compared to December 2025. JPMorgan highlighted Kuaishou's leading position in generative AI, labeling it as "one of the cheapest AI stocks globally" [5] - Citigroup's report favored Tencent Holdings and Alibaba-W as core AI concept stocks, optimistic about Tencent's AI development prospects in enterprise and user applications, and Alibaba's growth in cloud revenue and efficiency [5] - Industrial Securities indicated that internet leaders are poised to be the frontrunners in China's AI sector, expecting a resonance of buying from both domestic and foreign investors, which could lead to an upward adjustment in long-term profit growth expectations and valuation improvements [5]

Core Viewpoint - Zhiyuan AI, known as the "first global AI large model stock," debuted on the Hong Kong stock market on January 8, with an issue price of HKD 116.2 per share, corresponding to a market value exceeding HKD 51.1 billion [1] Group 1: Company Background and Funding - Founded in June 2019, Zhiyuan AI originated from Tsinghua University's Knowledge Engineering Laboratory, focusing initially on the development and commercialization of the academic technology intelligence mining system "AMiner" [4] - The company has raised over RMB 8.3 billion in funding, with over 90% occurring after the release of ChatGPT [2] - The strategic foresight to enter the large model sector was pivotal, with the decision made in May 2020 following the release of GPT-3 [4] Group 2: Investment Rounds and Valuation - Zhiyuan AI completed three rounds of financing before the release of ChatGPT, with the first two rounds in April 2021, raising a total of RMB 1.52 billion [5] - The valuation surged from RMB 2.11 billion in the B1 round to RMB 24.38 billion before the IPO, with a total of RMB 8.36 billion raised, primarily in 2023-2025 [12] - Notable investors include Ant Group, Tencent, and various state-owned enterprises, which have significantly contributed to the funding rounds [15] Group 3: Business Model and Challenges - The company aims to build a sustainable business model, facing competition from major players like Tencent, Alibaba, and ByteDance, which are increasing their investments in AI [16] - The cost of computing services has risen sharply, from RMB 15 million to RMB 1.55 billion by 2024, constituting 70.7% of R&D expenses, while revenue is projected at only RMB 312 million for 2024 [15] - The reliance on government contracts and large clients poses risks, as the top five clients from 2022 to 2025 do not overlap, indicating a lack of long-term partnerships [17]

Core Viewpoint - Zhiyu AI has officially gone public, marking a significant milestone as the first Chinese large model company to list on the Hong Kong Stock Exchange [1][50]. Group 1: Company Overview - Zhiyu AI, officially known as Beijing Zhiyu Huazhang Technology Co., Ltd., was established on June 11, 2019, and is rooted in Tsinghua University [6][10]. - The company has evolved from a research lab focused on knowledge graphs and NLP to a commercial entity, leveraging its academic foundation [9][10]. - CEO Zhang Peng and Chief Scientist Tang Jie are key figures in the company's development, transitioning from academic research to commercial applications [10][11]. Group 2: Business Positioning - Zhiyu AI is recognized as a leading independent large model company in China, with its GLM model serving as the core technology [16]. - The company has a market share of 6.6% in the large model sector, ranking second among Chinese competitors, with institutional clients contributing 4.7 billion yuan and individual clients 600 million yuan [16]. - By mid-2025, Zhiyu AI had served over 8,000 institutional clients, with its open-source models downloaded over 45 million times globally [17][18]. Group 3: Model Development - Zhiyu AI has released several significant models, including GLM-10B, CodeGeeX, and GLM-130B, contributing to the AI ecosystem [22][26]. - The company gained recognition in 2023 with the launch of ChatGLM-6B, a model that could be deployed on consumer-grade GPUs [27][30]. - The GLM series has seen substantial growth in usage, with GLM-4.7 processing approximately 122 billion tokens weekly by late 2025 [39]. Group 4: Financing Situation - Since its inception, Zhiyu AI has rapidly engaged in fundraising, with significant rounds starting from 2021, raising over 2.5 billion yuan in 2023 alone [45][46]. - By 2025, the company's valuation reached approximately 40 billion yuan, reflecting its strong market position and investor interest [46]. - The company has attracted investments from major players like Meituan, Ant Group, Alibaba, Tencent, and Xiaomi [46]. Group 5: Financial Performance - Zhiyu AI's revenue from 2022 to 2024 showed a compound annual growth rate of approximately 130%, with 2025's first half revenue nearing 1.91 billion yuan, a 325% year-over-year increase [53][56]. - The company primarily generates revenue from localized deployment solutions and cloud-based services, with localized deployment accounting for 84.5% of revenue in 2024 [58][59]. - Despite high revenue growth, the company has faced significant net losses, primarily due to substantial R&D expenditures, which reached 2.195 billion yuan in 2024 [73][74]. Group 6: Business Model - Zhiyu AI's business model focuses on providing a model-as-a-service (MaaS) platform, leveraging its GLM series for B2B solutions and AI-native applications [78][81]. - The company aims to expand its customer base and ecosystem penetration rather than prioritize short-term profits [84]. - Key clients include major players in IT services, telecommunications, and education, with a strong focus on retaining and expanding its customer relationships [85][86]. Group 7: Future Outlook - The large model market in China is projected to grow significantly, with expectations of reaching 101.1 billion yuan by 2030, driven primarily by enterprise demand [94]. - Zhiyu AI plans to focus on three key areas: technology, product development, and talent acquisition to sustain its growth trajectory [96][102]. - The company is positioned to capitalize on the growing demand for large models in various industries, aiming to replicate successful case studies across sectors [101][102].

今天，智谱AI终于要上市了。 | | | 真的蛮感慨的。 第一次去智谱参观，跟他们见面，还是2023年的4月。 公司技术根基来自清华大学计算机系知识工程实验室，也就是KEG Lab。 时间真的眨眼而过。 好梦幻的三年。 作为中国大模型第一股，作为曾经AI六小龙里，最闪耀的新星，我想，对智谱AI你也有一定有很多好奇。 在今天这个时刻，我想，也用这一篇文章，6个信息。 帮你了解关于智谱AI的一切。 话不多说，我们，开始。 一. 智谱AI是谁？ 智谱 AI，全称"北京智谱华章科技股份有限公司"。 2019年6月11日成立，出身非常标准的清华系。 这个实验室十几年一直在做知识图谱、NLP、多模态等底层研究，最早做出来的东西之一是学术搜索平台AMiner。 CEO张鹏算是学术出身+创业型的典型，原来就在KEG做AMiner，2019年出来把团队改造升级成智谱，把很多的科研成果，往商业公司方向重构。 而智谱的灵魂人物，首席科学家唐杰老师，之前就是实验室的领头人。 实验室里除了智谱的人之外，还有一个AI圈的非常有名的创业大佬，跟KEG有千丝万缕的关系，名字我就不提了，但是你们肯定都知道。 所以，说KEG Lab是中国AI的 ...