技术和商业化双轮驱动，实现两年从0到3.37亿营收的火箭式增长。 作者 | 木盒 编辑 | 小白 2026年1月2日，壁仞科技（06082.HK）正式登陆港交所，上市当天大涨76%，并创下两项重要里程 碑：不仅成为港股第一家GPU上市公司，更创下香港上市规则18C章节（新兴科技公司专属上市政 策）自2023年3月实施以来的最大募资规模。 招股资料显示，壁仞科技IPO发行阶段就已经获得资本市场火热追捧，其公开发售超额认购倍数飙升 至2347倍，市场打新热情空前高涨。最终，公司将H股发行价锁定19.6港元，合计净募资53.75亿港 元。 值得关注的是，此次募资获得重量级资本加持，近半数资金来自基石投资——启明创投、平安人寿、 瑞银集团、泰康人寿等知名机构赫然在列，为公司上市后的发展注入强劲信心。 风云君认为，壁仞科技IPO的火爆行情绝非偶然，其背后折射的是资本市场对公司在国产GPU领域核 心地位的高度认可，以及对其未来发展潜力的坚定看好。 最热门的赛道，最核心的公司 AI（人工智能）已经成为全球经济和社会发展的重要驱动力，尤其是大模型语言将大幅推动AI向AGI （通用人工智能）的演进，AGI有望在生产力和创造力方 ...

Core Viewpoint - Wallen Technology (06082.HK) successfully listed on the Hong Kong Stock Exchange, marking the first IPO of 2026 in Hong Kong, with a strong response from investors, particularly retail investors [2][4] Group 1: IPO Details - The IPO attracted 471,000 subscriptions in the public offering segment, the highest for a new stock in the past year in the Hong Kong market [2] - This IPO is the largest fundraising project since the implementation of Chapter 18C of the Hong Kong listing rules [2] - The net proceeds from the IPO will primarily fund R&D, with approximately 85% allocated for technological advancements and product iterations [2][14] Group 2: Product Development - The next-generation flagship chip, BR20X, is planned for commercialization in 2026, featuring significant upgrades in computing power, memory capacity, and interconnect bandwidth [2] - Initial R&D for the BR30X and BR31X products has begun, with expected launches in 2028, indicating a continuous product pipeline for growth [2] Group 3: Business Model and Technology - Wallen Technology develops General-Purpose Graphics Processing Unit (GPGPU) chips and intelligent computing solutions, providing essential computing power for AI applications [4][5] - The company integrates its proprietary BIRENSUPA software platform with GPGPU hardware to support a wide range of AI model training and inference applications [4][7] - The GPGPU architecture allows for high performance, energy efficiency, and flexibility, reducing the total cost of ownership for clients [8] Group 4: Market Position and Financials - The Chinese intelligent computing chip market is highly concentrated, with the top two players holding 94.4% market share as of 2024, presenting opportunities for growth for emerging players like Wallen Technology [13] - Revenue figures for Wallen Technology from 2022 to 2024 show a significant increase, with revenues of 0.5 million, 62 million, and 337 million RMB respectively [13][14] - The company has a strong focus on R&D, with total R&D expenditures of 1.018 billion, 886 million, and 827 million RMB for the years 2022, 2023, and 2024 respectively, representing a high percentage of total operating expenses [9]

人类已经走上了创造 AGI（通用人工智能）的道路，而其中一个关键方面是持续学习，即 AI 能通过与环境互动而不断学习新的知识和能力。 想象一下你生命中的第一次机器学习讲座：你或许记不清教授开口说的第一个单词，但那场讲座留给你的直觉和逻辑，此刻正潜移默化地帮助你理解这篇复杂的 论文。这种能力的本质在于 压缩 。 近日，Astera 研究所、英伟达、斯坦福大学、加州大学伯克利分校、加州大学圣地亚哥分校的一个联合团队提出的 TTT-E2E（端到端测试时训练） 沿着这条 AGI 的必经之路迈出了重要一步。它彻底打破了传统模型在推理时静态不变的局限，让长上下文建模从一种「架构设计」进化为一种「学习问题」。 为此，研究社区已经在探索多种不同的道路，比如开发能够实时更新状态的循环神经网络（RNN），或者试图通过极大的缓存空间来容纳海量历史。然而，真正 的 AGI 或许不应仅仅被动地「存储」信息，而应像人类一样在阅读中「进化」。 该方法可以在测试阶段通过给定上下文的下一个 token 预测持续学习， 将读取的上下文信息压缩至权重参数中 。 编辑｜Panda 论文标题：End-to-End Test-Time Training ...

" 欧米伽未来研究所 " 关注科技未来发展趋势，研究人类向欧米伽点演化过程中面临的重大机遇与挑战。将不定期推荐 和发布世界范围重要科技研究进展和未来趋势研究。 （ 点击这里查看广义智能体理论 ） 罗马圣贤马可·奥勒留曾说过，我们不应该让未来扰乱我们的心绪。但他显然从未与未来学家尼克·博斯特罗姆探讨 过2050年的世界状况。 博斯特罗姆在一封电子邮件中表示："到2050年，所有科学研究很可能都将由超级人工智能而非人类研究人员完 成。有些人可能会把科学作为爱好，但他们不会做出任何有用的贡献。" 布拉瑟及其同事在2025年的一项研究¹ 中指出，地缘政治紧张局势意味着，对2050年未来气候的预测必须越来越多地考 虑大气物理学以外的因素 。十年前，气候科学家们还在庆祝《巴黎协定》促成了全球对气候变化的认可。而如 今，美国却被迫从政府报告和网站中删除"气候变化"一词。与此同时，其他优先事项也占据了主导地位。 "如果我们谈论气候科学，人们根本不想听，因为他们更害怕其他问题。他们想要食物，他们想要和平，"他说。所 有这些都表明，到2050年，世界将面临本世纪末气温升高3摄氏度甚至更高的局面。 位于冰岛的一座由瑞士公司Clim ...

Group 1 - MiniMax, a general artificial intelligence company, is set to launch its IPO on December 31, 2025, with an issuance of 25.3892 million shares priced between HKD 151 and HKD 165 per share, leading to a valuation between HKD 46.123 billion and HKD 50.399 billion [1] - The company has secured 14 cornerstone investors, including Aspex, Eastspring, Mirae Asset, Alibaba, and E Fund, with a total subscription amount of approximately HKD 2.723 billion, representing a diverse range of investor types [1] - Founded in early 2022, MiniMax focuses on the development of multimodal models and has launched the first domestic speech model based on the Transformer architecture, Speech01, in 2023, followed by an upgraded version, Speech02 [1] Group 2 - As of September 30, 2025, MiniMax has over 200 countries and regions with more than 212 million individual users, indicating a significant global reach [2] - The company's revenue for the first nine months of 2025 has increased by over 170% year-on-year, with more than 70% of revenue coming from overseas markets [2]

将投中网设为"星标⭐"，第一时间收获最新推送 冲击IPO的最年轻AI独角兽。 作者丨 簪竹 来源丨 投中网 值得一提的是， MiniMax在这次IPO发行中引入了一个堪称豪华的基石阵容，包括Aspex、Eastspring、Mirae Asset、阿里巴巴及易方达在内的14家基石投资者，认购总额约27.23亿港元。 投资者类型横跨国际长线、头部科技、 中资长线及产业战略等多个维度，在近期港股市场，如此多元的顶级资本罕见地协同一致，形成了强有力的价值背书。 MiniMax开启发售引发市场热议，不少投资人认为很可能成为又一个超认购的明星标的，这也引申出了一个被忽略的关 键问题：这家拥有全球2.12亿用户、技术位列第一梯队的公司，该如何用估值模型进行合理定价。 如果说过去三年，一级市场在争论"谁能做出模型"，那么今天的问题已经变成，当模型已经成为基础设施，市场是否还 在用上一代"单点能力公司"的逻辑去定义真正的平台型AI公司？ 成立四年，冲击IPO的最年轻AI独角兽 关于MiniMax的创业故事，已经无需赘述。2022年初，前商汤副总裁闫俊杰在商汤上市前夕，毅然放弃手中的期权， 辞职创业，他从上海起步，以研发通用人 ...

Core Viewpoint - The company, Zhihui Huazhang Technology Co., Ltd., is set to launch its IPO on the Hong Kong Stock Exchange, marking a significant milestone as the first publicly listed company focused on AGI (Artificial General Intelligence) in China, indicating a new phase of integration between technology leadership and capital empowerment in the AI large model industry [1] Group 1: IPO and Financial Overview - Zhihui plans to issue 37.42 million H-shares globally, with 1.87 million shares for public offering in Hong Kong and 35.55 million shares for international offering [1] - The company has completed 8 rounds of financing prior to the IPO, raising over 8.3 billion yuan [1] - Revenue from large models is the primary source of income for the company, with a significant increase in the proportion of revenue from its MaaS (Model as a Service) platform [2] Group 2: Revenue Growth and Business Model - The company has experienced exponential growth in its cloud MaaS and subscription services, with a revenue compound annual growth rate (CAGR) of 130% from 2022 to 2024 [2] - Revenue figures for 2022, 2023, and 2024 are reported as 57.4 million yuan, 124.5 million yuan, and 312.4 million yuan respectively, with a 325% year-on-year growth in the first half of 2025 [2] - The subscription product for AI programming has contributed significantly to revenue growth, with over 150,000 paid developer users within two months of launch [2] Group 3: Research and Development - The company has a high R&D personnel ratio of 74%, with substantial investments in R&D amounting to approximately 4.4 billion yuan cumulatively [3] - The company has released its flagship model GLM-4.7, which has achieved top rankings in various global AI model assessments [3] - Continuous R&D investment is planned, with 70% of the funds raised from the IPO allocated to further development of general AI large models [5] Group 4: Market Context and Future Outlook - The global AI market is projected to exceed $500 billion by 2027, with generative AI and related services expected to account for over 30% of this growth [4] - The company is positioned as a key player in transitioning the industry from a focus on technological competition to realizing commercial value [4][5]

Core Viewpoint - The company Zhihua Technology is set to launch its IPO on the Hong Kong Stock Exchange, marking a significant milestone as the first publicly listed company focused on AGI (Artificial General Intelligence) in China, indicating a new phase of integration between technology leadership and capital empowerment in the AI large model industry [1] Group 1: IPO and Financial Overview - Zhihua plans to issue 37.42 million H-shares globally, with 1.87 million shares for public offering in Hong Kong and 35.55 million shares for international offering [1] - The company has completed 8 rounds of financing prior to the IPO, raising over 8.3 billion yuan [1] - Revenue from large models is the primary source of income for Zhihua, with a compound annual growth rate (CAGR) of 130% from 2022 to 2024, and revenue figures of 57.4 million yuan, 124.5 million yuan, and 312.4 million yuan for those years respectively [2] Group 2: Business Model and Growth - Zhihua employs a MaaS (Model as a Service) model, providing general intelligence capabilities through API calls to developers and enterprises, which has led to exponential growth in cloud-based MaaS and subscription services [2] - The company has seen a 325% year-on-year revenue increase in the first half of 2025, with a significant contribution from its AI programming subscription products, which have attracted over 150,000 paid developer users globally [2] Group 3: Research and Development - Established in 2019, Zhihua is recognized as "China's OpenAI," with a research and development team comprising 74% of its workforce [3] - The company has significantly increased its R&D investment, totaling approximately 4.4 billion yuan from 2022 to mid-2025, with the latest flagship model GLM-4.7 achieving top rankings in both open-source and domestic model evaluations [3] Group 4: Market Context and Future Outlook - According to IDC, the global AI market is projected to exceed $500 billion by 2027, with generative AI and related services expected to account for over 30% of this growth [4] - Zhihua aims to allocate 70% of the net proceeds from its IPO to further R&D in general AI models, enhancing its competitive edge in this sector [5] - The IPO is seen as a pivotal opportunity for Zhihua to drive its R&D and business expansion, shifting the narrative in the AI large model field from "technical stories" to "commercial value realization" [5]

Core Viewpoint - The company Zhiyu Huazhang Technology Co., Ltd. is set to become the first publicly listed company focused on large AI models globally, with its IPO expected to raise approximately HKD 4.3 billion and achieve a market valuation exceeding HKD 51.1 billion [1][2]. Group 1: IPO Details - Zhiyu plans to issue 37.42 million H-shares, with 1.87 million shares available for public offering in Hong Kong and 35.55 million shares for international offering, priced at HKD 116.20 per share [1]. - The company has raised over 8 rounds of financing, accumulating more than RMB 8.3 billion prior to the IPO [1]. Group 2: Market Position and Investor Interest - According to Frost & Sullivan, Zhiyu ranks first among independent general-purpose large model developers in China and second globally based on projected revenue for 2024 [2]. - The IPO has attracted significant interest from international long-term funds and notable industry capital, with cornerstone investors committing approximately HKD 2.98 billion, representing nearly 70% of the total offering [3]. Group 3: Use of Proceeds - Approximately 70% of the net proceeds from the IPO (around HKD 2.9 billion) will be allocated to research and development of general AI models, while about 10% (approximately HKD 420 million) will be used to optimize the company's MaaS platform [3]. Group 4: Revenue Growth and Business Model - Zhiyu's revenue primarily comes from its MaaS (Model as a Service) model, which has shown exponential growth, with revenues of RMB 57.4 million, RMB 124.5 million, and RMB 312.4 million for 2022, 2023, and 2024 respectively, reflecting a compound annual growth rate of 130% [4]. - The company has empowered over 12,000 enterprise clients and more than 45 million developers, making it the largest independent general-purpose large model provider in China [4]. Group 5: R&D Investment and Market Outlook - The company faces cost pressures related to its foundational models, with R&D investments of RMB 844 million, RMB 5.289 billion, and RMB 21.954 billion for 2022, 2023, and 2024 respectively, totaling approximately RMB 4.4 billion [5]. - The global AI market is projected to exceed USD 500 billion by 2027, with generative AI and related services expected to account for over 30% of this growth [5].

2025年12月30日，被誉为"全球大模型第一股"的北京智谱华章科技股份有限公司（以下简称"智谱AI"） 正式启动在香港的上市招股程序，计划于2026年1月8日挂牌交易。这标志着，自ChatGPT掀起全球AI浪 潮三年后，中国大模型产业的头部玩家正式步入接受公开市场审视的新阶段。 智谱AI的上市之路，与其主要竞争对手MiniMax（上海稀宇科技）几乎同步，两者在12月相继通过港交 所聆讯，一场关于商业模式、估值逻辑和未来话语权的终极较量已然展开。然而，招股书所揭示的，远 不止"第一股"的光环，更是一幅高增长、高投入与高亏损并存的行业真实图景：三年半累计亏损超62亿 元，账面现金一度被推算仅能支撑数月。 与此同时，其收入结构高度集中，截至2025上半年，84.8%来自私有化部署项目，主要客户为政府与大 型企业，商业模式仍深陷定制化交付的重资产逻辑。 01 营收飞速增长，暗藏结构性隐忧 在此背景下，智谱AI成立于2019年，至今仅六年时间，却已在中国大模型创业浪潮中占据先发优势。 在业内广为流传的"大模型六小龙"中，智谱AI是唯一一家成立于2020年之前的公司。其余五家，包括月 之暗面、百川智能、零一万物、阶跃星 ...