六年时间，在更难走的路上把事情真正推进了一步。 文 丨 江思远 2026 年 1 月 2 日早晨，GPU 制造商壁仞科技成为今年第一家在港股上市的公司。这时距离壁仞科技 创始人张文点亮中国 GPU 创业潮正好六年。 壁仞科技创始人张文（右一）敲钟。来源：壁仞科技。 这是一个并不偶然的时间点。按照张文和壁仞科技的多位早期投资人估算，一家初创 GPU 公司站稳 脚跟至少要这么久——从组建团队到点亮第一颗 GPU，往往已经过去 3 年，再调试、优化产品，把一 片片 GPU 集成服务器装到客户的机房里，又要 3 年。 这条路英伟达在二十多年前也走过，英伟达从成立到上市也是六年。但同样是六年，壁仞科技、摩尔 线程等公司所经历的产业变化，远比当年的英伟达更为剧烈。 壁仞科技成立时，GPT 还停留在 2.0 时代。AI 算力的需求主要集中在科研和小规模商用场景，GPU 仍然被视为图形芯片和高性能计算的工具。 六年之后，生成式 AI 爆发，GPU 成为全球科技产业最核心、最稀缺的基础设施之一。英伟达市值站 上全球科技公司之巅，寒武纪、沐曦等国产芯片公司也在资本市场创下新纪录。 局势的变化，为壁仞科技这样的国产 GPU 公司打 ...

Group 1 - Wallan Technology (06082.HK) has officially listed on the Hong Kong Stock Exchange, becoming the first general-purpose AI chip company to go public in 2025, with an opening price increase of over 100% and a market capitalization exceeding HKD 94 billion [1] - Founded in 2019 by a team of former AMD and NVIDIA chip architects, Wallan Technology has launched several high-performance GPUs, including BR100, BR104, BR106, and BR166, and has become a key player in China's AI infrastructure [1] - The investment from Shanghai State-owned Capital, which holds a 3.7853% stake in Wallan through Shanghai Lingang Economic Development Group, signifies a strategic move in the AI computing power sector [2] Group 2 - The Shanghai Guotou Pioneer AI Industry Fund made a direct investment in Wallan Technology, marking its first direct investment project, which breaks from the traditional indirect investment model [3] - This investment not only provides financial support but also enhances Wallan's capabilities in technology development, industry connections, and ecosystem collaboration [4] - Wallan's R&D expenses reached CNY 572 million in the first half of 2025, accounting for 79.1% of total operating expenses, highlighting the ongoing need for funding in its expansion and development efforts [4] Group 3 - The collaboration between Wallan Technology and Shanghai State-owned Capital has opened pathways for market entry, customer integration, and supply chain collaboration, enhancing the company's market recognition and stability [5] - Wallan's success is part of a broader ecosystem in Shanghai, which includes multiple GPU and computing power chip companies, forming a dense entrepreneurial matrix [8] - Shanghai has emerged as a capital hub for the domestic GPU sector, with leading companies collectively raising over CNY 20 billion, indicating strong investor interest and confidence in the industry [9] Group 4 - The Shanghai government has set ambitious goals for the AI computing cloud industry, aiming for a scale of CNY 200 billion by 2027, with a focus on building large-scale computing clusters and a unified scheduling platform [12] - Wallan's IPO is seen as a significant milestone for the domestic computing power ecosystem, marking the beginning of a new phase in the development of China's AI infrastructure [12] - The strategic investment and support from Shanghai State-owned Capital are crucial for fostering a sustainable and competitive environment for hard-tech companies like Wallan Technology [11]

Core Viewpoint - The emergence of the "Four Little Dragons" of domestic GPUs—Mole Thread, Muxi Technology, Bilan Technology, and Tensu Zhixin—marks a significant moment in the capital market, raising questions about their sustainable growth amidst international competition [1][8]. Group 1: Company Profiles and Strategies - Mole Thread, with a team rooted in Nvidia, emphasizes ecosystem compatibility and parallel development of graphics rendering and AI computing, utilizing its self-developed MUSA architecture [2][3]. - Muxi Technology and Tensu Zhixin adopt a pragmatic approach similar to AMD, focusing on high-performance parallel computing while sacrificing complex graphics rendering capabilities [2][3]. - Bilan Technology stands out with its original architecture and Chiplet technology, aiming for extreme computing performance and flexibility in product offerings [3][4]. Group 2: Market Position and Performance - The "Four Little Dragons" have begun to establish their presence in various sectors, with Bilan achieving commercial success in heterogeneous training solutions, Mole Thread securing large-scale delivery contracts, Muxi reporting significant orders, and Tensu being the first to achieve mass production of 7nm GPGPU [7][8]. - The competitive landscape is characterized by overlapping targets among the four companies, particularly in securing contracts from major internet firms and local government data centers [15][16]. Group 3: Financial Insights and Challenges - All four companies exhibit high growth accompanied by significant losses, with R&D expenses consuming a large portion of their revenues, particularly for Muxi and Tensu [19][20]. - The R&D expenditure for 2024 shows that Mole Thread has 886 R&D personnel with a rate of 309.88%, while Muxi has 706 personnel with a rate of 121.24% [21]. - The timeline for achieving profitability is projected around 2026-2027, mirroring Nvidia's early growth trajectory [22]. Group 4: Ecosystem and Market Dynamics - The competition extends beyond hardware performance to include software ecosystems, with each company developing its own software stack to attract developers and enhance product value [23][24]. - The ability to expand customer bases and maintain high repurchase rates will be crucial for validating the true value of their products in a market increasingly driven by performance and cost [24][25]. - The stability of supply chains is critical, as the GPU industry is interconnected with various sectors, and any disruptions could significantly impact production and market presence [25].

Core Viewpoint - The approval of Shanghai Birun Technology Co., Ltd.'s overseas listing and the full circulation of unlisted shares marks a significant milestone, positioning the company as a potential leader in the Hong Kong GPU market and providing new investment opportunities in China's AI industry [2][7]. Group 1: Company Overview - Founded in September 2019, Birun Technology focuses on the development of original general-purpose computing systems and aims to achieve breakthroughs in high-end domestic intelligent computing chips [3]. - The company has achieved several milestones, including the launch of its first general-purpose GPU in March 2022 and the release of the BR100 GPU chip in August 2022, which broke international computing power records [3]. Group 2: Financial and Investment Highlights - Birun Technology has attracted significant capital, raising nearly 20 billion yuan in total funding since its inception, with a valuation of approximately 155 billion yuan (around 21.77 billion USD) [4]. - The company has a diverse investor base, including prominent firms such as Qiming Venture Partners and IDG Capital [4]. Group 3: Technological Development and Collaborations - The company has established partnerships with leading universities and institutions to enhance its AI hardware and software ecosystem, including collaborations with Tsinghua University and Shanghai AI Laboratory [5]. - In October 2023, Birun Technology was placed on the U.S. Entity List, prompting the company to accelerate its independent innovation efforts [5]. Group 4: Leadership and Strategic Vision - The founder, Zhang Wen, has a strong background in law and investment, which has contributed to the company's rapid growth and strategic direction [6]. - Zhang emphasizes the importance of talent, management, and large-scale applications in the semiconductor industry, aiming to create versatile products that meet market demands [7].

壁仞科技是"国产GPU四小龙"中唯一选择港股"全流通"上市的企业。而在此之前，该公司曾于2024年9月在上海证监局办理科创板IPO辅导备案。 12月15日，证监会国际合作司披露了关于上海壁仞科技股份有限公司(以下简称"壁仞科技")境外发行上市及境内未上市股份"全流通"备案通知书。 根据备案通知书，壁仞科技拟发行不超过372458000股境外上市普通股并在香港联合交易所上市。该公司57名股东拟将所持合计873272024股境内未上市股 份转为境外上市股份，并在香港联合交易所上市流通。 市场端同样对壁仞科技给予高度关注。2025年胡润百富独角兽排行榜显示，其估值已达160亿元人民币。 基于BR100及其衍生产品(如BR104)，壁仞科技打造了壁砺系列商用硬件产品线，已量产落地。该系列包含液冷/风冷OAM模组、推理AI加速卡等，依托原 创的GPGPU芯片架构以及BIRENSUPA软件开发平台，支持业内主流的深度学习框架与模型，为广泛的AI计算场景提供高能效比、高通用性的强大算力。 研发方面，截至2025年9月30日，壁仞科技在全球多个国家和地区累计公开专利1200余项，位列中国通用GPU公司第一，获得专利授权550 ...

Core Viewpoint - Wallen Technology's application for listing on the Hong Kong Stock Exchange signifies not only the capital advancement of this domestic GPU leader but also an important signal of China's breakthrough in autonomous computing power [1] Group 1: Technological Advancements - Wallen Technology's core strength lies in its technological breakthroughs, having developed an original architecture that disrupts the high-end GPU market's technological monopoly [3] - The flagship BR100 series, built on the "Bili Ran" architecture, utilizes 7nm process and Chiplet technology, achieving over three times the performance of mainstream products, with 770 billion transistors and INT8 computing power reaching 19.2 TOPS [3] - The proprietary BLink™ high-speed interconnect technology enables a single card to achieve 448GB/s bandwidth, supporting over 95% linear acceleration for large-scale clusters [3] - Wallen's heterogeneous mixed training technology allows four different GPUs to train the same large model, providing a viable solution to the computing power isolation problem [3] Group 2: Ecosystem Development - The progress of ecosystem construction directly influences the market viability of GPU products, with Wallen Technology striving to close the ecological gap with international giants through open collaboration [5] - The self-developed BIRENSUPA™ software toolchain has become the core of the ecosystem, supporting mainstream data precisions and achieving compatibility with major models [5] - Wallen has established a diverse cooperation network through technological openness, collaborating with various companies to enhance training performance and create a complete technology closed loop [5] Group 3: Commercialization and Market Strategy - The acceleration of commercialization has allowed Wallen Technology to convert technological value into market competitiveness, achieving comprehensive coverage of three major telecom operators [6] - Wallen has formed a clear product matrix targeting different application scenarios, with the BR100 focusing on AI training and the Bili 110E inference card offering significant performance density and energy savings [6] - The company has set a target of 20,000 chips for procurement by 2025 and is expanding its reach into government and enterprise markets through partnerships [7] Group 4: Financial Outlook and Growth Potential - Despite the promising technological advancements, Wallen Technology's revenue remains modest, projected at approximately 280 million yuan in 2025, highlighting the challenge of converting technical advantages into profitability [7] - The capital injection from the Hong Kong listing may accelerate investments in key areas such as packaging localization and ecosystem expansion [7] - The deep involvement of industrial capital, like Shanghai Lingang, not only shares in the company's growth but also strengthens the development foundation for hard-tech enterprises, fostering a beneficial cycle of "capital + technology" [7]

Group 1 - The article highlights the support from the Ministry of Industry and Information Technology for domestic GPU development, indicating a competitive landscape among local companies in the GPU sector [1] - Moores Threads is developing the MTT S80, a full-featured GPU capable of running major games, with a Pre-IPO valuation of 25.5 billion [1] - Jingjia Micro is positioned as a key player with its JM9 GPU nearing the performance of GTX 1050, and a significant market share in government and financial sectors [1] - Birran Technology's BR100 GPU boasts three times the computing power of international flagship products, positioning it as a leader in domestic alternatives despite being listed by the US [2] - Muxi Integration has successfully delivered the Xiyun C500 GPU, with plans for the C600 to support FP8 and CUDA compatibility, marking it as a major player in domestic scaling [3] - Lishuan Technology is set to launch a 6nm GPU in 2025 that will outperform RTX 4060, focusing on digital twins and out-of-order rendering [4] - Haiguang Information stands out as the only company with both CPU and GPU capabilities, forming a unique ecosystem with products widely used in various sectors [5] Group 2 - The article notes that while domestic GPUs may struggle to catch up with established players like NVIDIA, AMD, and Intel in the short term, there is potential for significant growth in the long term, similar to the evolution seen in the home appliance and automotive industries [5]

Core Insights - The global computing power landscape is evolving into two parallel paths, with domestic and international routes differing, and opportunities for domestic chips in inference scenarios are expected to emerge this year [1][4] - Two AI chip companies, Muxi Integrated Circuit (Shanghai) Co., Ltd. and Moore Threads Intelligent Technology (Beijing) Co., Ltd., are nearing their IPOs, with Muxi's counseling work completed and Moore Threads' status changing to acceptance [1][2] - The demand for AI chips in China is increasing, with several companies adapting to the popular DeepSeek-R1 model, leading to a significant shift in market dynamics [3][4] Company Summaries - Muxi and Moore Threads were founded in 2020, while Suiruan Technology and Birun Technology were established in 2018 and 2019, respectively; the latter two companies have slower IPO progress compared to the former [2] - Moore Threads has the highest valuation among the four companies at 25.5 billion yuan, followed by Suiruan Technology at 16 billion yuan, Birun Technology at 15.5 billion yuan, and Muxi at 10 billion yuan [2] - Founders of these companies have backgrounds in major overseas chip firms, with Moore Threads' CEO previously serving as NVIDIA's China General Manager and Muxi's founders coming from AMD [2] Market Dynamics - The demand for domestic AI chips has surged, with IDC reporting that 34.6% of the Chinese data center accelerator market was comprised of domestic solutions last year, and this figure is expected to exceed 40% in the first half of this year [3] - The DeepSeek application has driven significant demand for domestic computing power, with many companies adapting their products to this model [3][4] - The global computing power foundation is shifting, with 98% of large model training still relying on NVIDIA, but there is potential for a portion of pre-training to transition to non-NVIDIA cards by the end of the year [4]

Core Viewpoint - The article discusses the performance of domestic GPUs, highlighting the competitive landscape and the impact of sanctions on production capabilities. It emphasizes that while the BR100 from Birun is the most powerful, it cannot be mass-produced due to sanctions, leaving Huawei's 910C as the strongest available option [1][2]. Performance Comparison - The performance ranking of GPUs indicates that the Huawei 910C leads, followed by the Haiguang BW100, with the latter achieving approximately 87% of its potential performance. The Cambrian 590's performance is uncertain, potentially reaching 80% [2]. - The Haiguang BW100 has a deep computing capability of over 400T, which is about 50% of the H800's performance, while the 910C reaches approximately 60% of the H800's performance [1][2]. Memory and Bandwidth - Most domestic GPUs utilize HBM2e memory due to sanctions preventing the use of HBM3e. Sufficient memory is crucial for initiating inference tasks, as insufficient memory can hinder model startup [3]. - Huawei leads significantly in memory bandwidth, which is a critical factor for performance [4]. Pricing Insights - Current market prices for GPUs are as follows: Haiguang BW100 at approximately 100,000 yuan, Huawei 910B at around 70,000 yuan, and 910C at about 180,000 yuan. The Cambrian 590 has seen a price drop from 80,000-85,000 yuan to between 60,000-70,000 yuan [2]. Technical Specifications - A table summarizes key specifications of various GPUs, including memory type, bandwidth, capacity, and arithmetic intensity. For instance, the Huawei 910C features HBM2e with a bandwidth of 3.2 TB/s and a memory capacity of 64 GB [5].