智谱 领跑中国大模型，开拓进取AGI 2026.2 中金研究部 中金公司研究部 赵丽萍 分析师 SAC Reg. No.: S0080516060004; SFC CE Ref: BEH709 王之昊 分析师 SAC Reg. No.: S0080522050001; SFC CE Ref: BSS168 袁佳妮 分析师 SAC Reg. No.: S0080523050003; SFC CE Ref: BTM577 投资要点 2 ✓ 源自清华的国产大模型领军，扎实的技术底蕴和模型能力是基石。公司成立于2019年，专注GLM系 列基座模型深耕，打磨Coding、推理、Agentic等模型核心能力。最新一代基座模型GLM-5在HLE、 SWE等多个基准评分中达到SOTA水准，并获得海内外用户广泛好评。 ✓ 模型商业价值兑现，赋能千行百业。公司依托MaaS平台输出模型能力，我们预计2023-2027年收入 CAGR达130%+。API收入将成为主要增长引擎。我们估计公司至2026年初API相关ARR已接近6亿元， 较去年同期已有数十倍的增长。同时，公司覆盖互联网、软件、芯片等千行百业，聚焦释放模型价值 赋能客户提 ...

GLM系列模型受到全球开发者喜爱，市场需求持续强劲增长，用户规模不断扩大，调用量也快速攀 升。目前，GLM-5已完成与华为昇腾、摩尔线程、寒武纪、昆仑芯、沐曦、燧原、海光等国产算力平 台的深度推理适配。通过底层算子优化与硬件加速，GLM-5在国产芯片集群上已经实现高吞吐、低延 迟的稳定运行。 在国际权威榜单Artificial Analysis中，GLM-5的表现同样亮眼，位居全球第四、开源模型第一。这一成 绩不仅彰显了智谱的技术实力，也为国产大模型在国际舞台上赢得了更多关注。随着GLM-5的发布和 开源，智谱有望进一步推动国产大模型生态的发展，为行业创新注入新动力。 智谱今日(2月12日)凌晨正式开源发布新一代基座模型GLM-5，被誉为智能体工程时代最强开源模型。 它宣告大模型编程从"写代码片段"的Vibe Coding时代，正式迈入"完成系统工程"的智能体工程时代。据 悉，GLM-5在编程能力上实现了对齐Claude Opus 4.5，性能超过Gemini3 Pro。官方GLM Coding Plan上 线即售罄，一个国产AI编程模型的付费套餐被抢空，这在行业历史上尚属首次。 ...

Summary of Conference Call Records Industry Overview Macro Environment - Recent comments from Japanese politician Takashi Shimizu indicate a growing unfriendly sentiment among Japanese youth towards China [1] - The Shanghai and Shenzhen Stock Exchanges announced a package of measures to optimize refinancing, emphasizing support for high-quality and high-tech companies, streamlining the refinancing process, and enhancing regulatory oversight throughout the refinancing process [1] - Reports suggest that some major banks in the U.S. have been verbally requested to limit their purchases of U.S. Treasury bonds [1] - Federal Reserve official Bessenet indicated that the Fed will act cautiously regarding balance sheet adjustments, not expecting rapid reductions and requiring up to a year for decision-making [1] Artificial Intelligence - Seedance 2.0 has prohibited the upload of real human materials [1] - OpenAI plans to release an upgraded conversational model this week [1] - Oracle raised $25 billion, with subscriptions reaching $130 billion, indicating a fivefold oversubscription [1] - Altman boasted to employees about the accelerated growth of ChatGPT, with OpenAI set to secure $100 billion in financing [1] Robotics - Sellers report that the V3 model is focused on in-house use in North American factories, while the V4 model is a key focus for Musk, aiming to become a blockbuster product [2] Dye Industry - Dye prices continue to rise, supported by strong pricing power from leading companies due to integration and scale [3] Solid-State Batteries - Leading battery manufacturers have reported that mid-term review results show over 87% state of charge (SOC) cycle life exceeding 300 cycles, meeting standards that are double that of GXB [4] Stock Market Performance - From 2021 to the 2025 Spring Festival holiday, the Hang Seng Index showed an average absolute return of 3.32%, with a median increase of 2.93%. The Hang Seng Technology Index had an average absolute return of 5.93% and a median increase of 5.50% [5] Market Strategy Observations - Today's trading volume reached 22.495 trillion, with an increase of 1.038 trillion. The index rebounded well, but there is limited upward space for the Shanghai Composite Index, suggesting greater potential for small-cap indices [6] - Key sectors include communication (AI), media (applications), and electronics (primarily AI) [7] - Hardware is following overseas recovery trends, while software is benefiting from significant gains in ByteDance [8] - Despite efforts, low-position stocks have not outperformed new tech directions significantly. The market experienced a broad rally, particularly in AI and space photovoltaic sectors, although there was limited follow-up buying, leading to some pullbacks [9] - Core chemical sub-sectors are experiencing price increases, but demand for these niche products remains low [10] Geopolitical Events - Ongoing geopolitical events in Japan and Iran should be monitored, although there has been little market impact recently. The strategy before the holiday involves maintaining positions in logical investments rather than switching, focusing on familiar sectors [11]

分析指出，GLM系列模型近年来在代码生成和智能体能力上的进步有目共睹，这与Pony Alpha的主打方 向完全一致；智谱首席科学家唐杰发微博明确表示最近智谱将发布GLM-5。目前，智谱等疑似关联方 尚未对Pony Alpha的身份作出官方回应。 消息面上，2月6日，全球模型服务平台OpenRouter悄然上线一款代号为"Pony Alpha"的匿名模型，因其 强大的编码能力、超长上下文窗口及针对智能体工作流的深度优化，迅速引发开发者社区关注。市场猜 测该模型的身份可能是智谱即将发布的新一代模型GLM-5。 智通财经APP获悉，智谱(02513)午前涨超22%，高见338.2港元，较招股价116.2港元已涨近200%，总市 值逼近1500亿港元。截至发稿，涨21.17%，报335.4港元，成交额6.65亿港元。 ...

多家国内大模型企业都计划于春节前后发布新一代模型。 本文为IPO早知道原创 作者｜ Stone Jin 微信公众号｜ipozaozhidao 据IPO早知道消息，"全球大模型第一股"智谱（2513.HK）今日股价再创新高，市值突破1200亿港 元。 事实上，包括 DeepSeek、智谱在内的多家国内大模型企业都计划于春节前后发布新一代模型。 日前 ， 全球模型服务平台 OpenRouter 就 上架了一个名为「 Pony Alpha」的神秘模型，并 迅速 在 24小时内登顶平台热度榜首，全球开发者 均积极参与测试与讨论。 当然，之所以受到如此大的关注度，还是由于这款模型具有 强大的编码能力、超长上下文窗口及针 对智能体工作流的深度优化 。 OpenRouter官方 则 将 Pony Alpha描述为"前沿基础模型"，在编 另据 OpenRouter合作方Kilo Code在其博客中留下了一个隐晦的线索，称Pony Alpha是"某个全 球实验室最受欢迎的开源模型的专项进化版"。 这意味着， Pony Alpha 更有 可能是智谱即将发布 的新一代模型 GLM-5。 一方面， GLM系列模型近年来在代码生成和智 ...

Core Viewpoint - The article discusses the rapid valuation increase of AI large model companies in China, particularly focusing on MoonshotAI, which has seen its valuation rise significantly in a short period, reflecting the evolving landscape of the AI industry and the implications for investment strategies [4][5][6][20]. Group 1: Valuation Changes - In just 20 days, MoonshotAI's valuation increased by approximately 3.5 billion RMB, reaching around 48 billion RMB after a recent funding round [7][14]. - The company completed a $500 million Series C funding round on December 31, 2025, with a post-money valuation of about $4.3 billion [14]. - The valuation adjustments in the primary market are influenced by the recent public listings of AI companies like Zhiyuan AI and MiniMax, which have established a new pricing framework for the industry [20][22]. Group 2: Company Strategy - MoonshotAI has adopted a restrained approach to its IPO timeline, focusing on model iteration and product refinement rather than rushing to market [8][9]. - The company was founded in early 2023 during a challenging market period, emphasizing a technology-driven approach without immediate product offerings [12]. - The initial funding rounds were primarily aimed at building a core team and computational resources, with a focus on developing foundational model capabilities [13]. Group 3: Product Development - The AI assistant Kimi, launched in the second half of 2023, serves as a testing ground for MoonshotAI's model capabilities, focusing on long text processing and complex data understanding [16][17]. - The product strategy for Kimi is designed to validate model performance in real-world applications, creating a feedback loop for continuous improvement [17][19]. - Unlike competitors, MoonshotAI has chosen to concentrate resources on a single product, allowing for deeper development and refinement of its model capabilities [19]. Group 4: Market Dynamics - The recent public listings have shifted market focus from computational power to model and application capabilities, affecting investment strategies in the AI sector [23][24]. - Companies that have successfully integrated AI capabilities into existing products, such as Kingsoft Office and Yonyou Network, are gaining investor attention due to their stable customer bases [24]. - The transition of AI companies to public markets signifies a move from a "technology explosion phase" to a "mid-stage industrialization," where long-term pricing power is determined by model iteration and product efficiency [24][26].

Core Insights - Beijing is recognized as the leading city in China's AI development, holding significant advantages in industry scale, talent, technology breakthroughs, and ecosystem [1][2] - The successful IPO of Zhizhu AI, a company derived from Tsinghua University, marks a milestone as the "first global AI large model stock," highlighting Beijing's leadership in AI research and development [1] - The AI industry in Beijing is supported by a dense ecosystem of top universities and research institutions, contributing to its dominance in talent and innovation [2] Industry Overview - The AI core industry in Beijing has a scale of nearly 350 billion yuan, with around 2,400 related companies, accounting for half of the national figures [2] - Beijing is home to 148 top AI scholars, representing 40% of the national total, and has launched 162 large models, nearly 30% of the national total [2] - Key players in the AI core industry chain, including chip companies and leading large model products, are concentrated in Beijing, showcasing its commercial viability [3] Future Developments - Beijing's AI strategy is evolving towards integrating AI into the physical world, moving beyond mere computational power and model iteration [3][5] - The concept of "embodied intelligence" is gaining traction, aiming to create AI systems that can interact seamlessly with the real world [5][9] - Companies in Beijing are developing advanced robotics and simulation technologies to enhance the capabilities of AI in practical applications [12][13] Innovation Ecosystem - The establishment of four "AI innovation districts" in Beijing signifies a shift towards real-world applications of AI technologies [13] - The focus on embodied intelligence represents an opportunity for China to redefine the rules of the game in AI, moving from application innovation to foundational technology development [13][14] - Beijing's comprehensive approach to AI, from brain to body, positions it as a potential leader in defining the future of the industry [14]

Group 1 - The software sector experienced a significant surge, with the Software 50 ETF (159590) rising over 5% and a substantial trading volume, indicating strong market interest [1] - Major stocks within the Software 50 ETF saw positive performance, including Zhongke Xingtou up over 14%, Tuorisi up nearly 12%, and Keda Xunfei up over 6% [1] - The net subscription amount for the Software 50 ETF exceeded 22 million yuan shortly after the market opened [1] Group 2 - The Software 50 ETF tracks the CSI Software Index, which includes 50 constituent stocks, with approximately 67% weight in application software and over 15% in AI-related fields [6] - The index aims to provide comprehensive exposure to the entire AI software industry chain, making it a strategic investment option [6] Group 3 - The upcoming release of DeepSeek V4 is anticipated to enhance programming capabilities and understanding of data patterns, potentially reshaping the global AI competition landscape [3] - Fire Mountain Engine has become the exclusive AI cloud partner for the Spring Festival Gala, highlighting the growing importance of multi-modal capabilities in AI applications [3] Group 4 - Companies like Zhiyu and MiniMax have recently listed on the Hong Kong stock market, which is expected to accelerate their growth with capital support [4] - The shift from traditional SEO to GEO (Generative Engine Optimization) is seen as a long-term growth opportunity in AI marketing, as it enhances content visibility and brand authority [5]

Group 1 - MiniMax, established only four years ago, had a remarkable debut on January 9, with a closing price increase of 109.09%, reaching a market capitalization of 105.4 billion HKD, surpassing the previous day's listing of Zhipu [1][3] - The contrasting market reactions to MiniMax and Zhipu highlight their different business models and growth narratives, with MiniMax focusing on C-end AI applications and over 70% of its revenue coming from overseas, while Zhipu targets the B-end enterprise market [3][4] - Financial data shows MiniMax's cash reserves at 2.57 billion RMB, more than double Zhipu's, with revenue growth from 3.46 million USD in 2023 to 30.52 million USD in 2024, a 782% increase [3][4] Group 2 - Zhipu, founded in 2019, has a strong academic background with a team primarily composed of Tsinghua University professors and PhDs, while MiniMax, founded in December 2021, is led by a younger team from SenseTime, with an average age of 29 [4][5] - MiniMax's product strategy focuses on "model as a product," directly targeting global C-end users with a diverse product matrix, while Zhipu emphasizes deep foundational models and MaaS services for enterprises [6][7] - MiniMax's impressive performance post-listing symbolizes a new entrepreneurial paradigm in the AI large model sector, showcasing its ability to rapidly iterate products and capture global user demand [7][9] Group 3 - MiniMax's global revenue distribution shows significant international reach, with 73.1% of its income coming from overseas markets, including China, Singapore, and the US [7][8] - Despite high growth, MiniMax reported cumulative losses of approximately 1.32 billion USD from 2022 to September 30, 2025, with R&D expenses reaching 180 million USD in the first three quarters of 2025, accounting for 337.4% of total revenue [8] - The differing paths of MiniMax and Zhipu reflect the diverse possibilities in AI commercialization, with Zhipu focusing on industry digitalization and MiniMax pursuing a global consumer approach [8][9]

Core Insights - The recent IPOs of Zhipu AI and MiniMax mark a significant milestone for the Hong Kong stock market, highlighting the entry of major AI players into the capital market [1] - Zhipu AI, as the first global large model company to go public, raised HKD 4.348 billion with a market capitalization of HKD 57.89 billion at IPO, while MiniMax saw a 109% increase in its stock price on the first day of trading [1] - Despite the excitement surrounding these IPOs, Zhipu AI faces challenges such as an imbalanced revenue structure, high computing costs, and a debt exceeding HKD 10 billion, raising questions about its future sustainability [1] IPO Performance - Zhipu AI's IPO was characterized by a strong demand, with a subscription rate of 910 times, indicating intense investor interest [6] - The stock opened at HKD 131.5, a 13.17% increase from its IPO price, and closed at HKD 158.6, with a market cap of HKD 698 billion [6] - Early investors in Zhipu AI, including major firms like Meituan and Ant Group, have seen substantial returns, with valuations skyrocketing from HKD 800 million in Series A to HKD 26 billion before the IPO [7][8] Financial Performance - Zhipu AI's revenue is projected to grow from HKD 57.41 million in 2022 to HKD 312 million in 2024, reflecting a compound annual growth rate of 130% [11] - However, the company is experiencing increasing adjusted net losses, projected to reach HKD 2.466 billion by 2024, primarily due to high R&D expenditures [11][12] - The company's gross margins are fluctuating, with 2022 at 54.6%, 2023 at 64.6%, and a drop to 56.3% in 2024, indicating potential profitability concerns [11] Revenue Structure - The majority of Zhipu AI's revenue comes from localized deployment, which accounted for 84.8% of total revenue in the first half of 2025, maintaining a gross margin of 59% [12] - In contrast, the cloud deployment segment is struggling with negative margins, attributed to aggressive pricing strategies to capture market share [13] - High customer concentration poses a risk, with the top five clients contributing 45.5% of revenue, indicating a need for diversification [13] Debt and Financial Health - Zhipu AI's debt has surged from HKD 542 million in 2022 to HKD 11.252 billion by mid-2025, resulting in a negative net asset position of HKD 6.151 billion [14] - The recent IPO proceeds of HKD 4.3 billion may alleviate short-term debt pressures, but long-term sustainability hinges on effective commercialization strategies [14] Industry Context - The IPO of Zhipu AI serves as a valuation benchmark for the large model industry, with a price-to-sales ratio of approximately 147 times based on 2024 revenue projections [15] - The Chinese large language model market is expected to grow significantly, from HKD 5.3 billion in 2024 to HKD 101.1 billion by 2030, with increasing market concentration favoring established players [15] - The future of Zhipu AI will be closely watched as it navigates the competitive landscape against global giants like OpenAI, emphasizing the importance of technological advancement and sustainable business models [16]