今日，半导体芯片持续走高，消息面上，研究机构Omdia数据显示，半导体行业2025年第三季度的营收 达到2163亿美元，首次突破单季2000亿美元大关，环比实现14.5%增长，按目前情况预测，2025全年半 导体营收将站上8000亿美元。 科创50增强ETF紧密跟踪上证科创板50成份指数，上证科创板50成份指数由上海证券交易所科创板中市 值大、流动性好的50只证券组成，反映最具市场代表性的一批科创企业的整体表现。 数据显示，截至2025年11月28日，上证科创板50成份指数(000688)前十大权重股分别为寒武纪 (688256)、海光信息(688041)、中芯国际(688981)、澜起科技(688008)、中微公司(688012)、金山办公 (688111)、联影医疗(688271)、芯原股份(688521)、拓荆科技(688072)、佰维存储(688525)，前十大权重 股合计占比57.27%。 科创50ETF指数(588040)，科创50增强ETF（588460），科创100ETF基金(588220)，科创200ETF指数 (588240)，科创综指ETF(589680) 截至2025年12月22日 1 ...

转自：新华财经 新华财经北京12月22日电（王媛媛）市场早盘集体走强，沪指收复3900点关口，创业板指一度涨超 2%。截至午间收盘，沪指报3915.20点，涨0.64%，成交5031亿元；深证成指报13318.80点，涨1.36%， 成交6828亿元；创业板指报3178.51点，涨1.80%，成交3087亿元。 板块方面，海南自贸区、贵金属、算力硬件等板块涨幅居前，医药商业、影视院线、银行等板块跌幅居 前。 盘面热点 盘面上，海南自贸概念集体爆发，海汽集团、海南瑞泽等近20股涨停。商业航天概念反复走强，神剑股 份、北斗星通涨停。算力硬件概念表现活跃，"易中天"光模块三巨头集体上涨。半导体设备股拉升，拓 荆科技等多股创历史新高。下跌方面，AI医疗概念震荡回落，华人健康大跌。 机构观点 排排网财富：展望后市，A股整体仍将运行在震荡消化的格局中，但结构性行情仍具备演绎空间。一方 面，中央经济工作会议相关政策逐步落地，扩内需、稳增长取向明确，对市场中枢形成支撑；另一方 面，盈利修复节奏仍偏缓，叠加海外货币政策分化，市场风险偏好短期难以显著抬升。 广州"十五五"规划建议：深化穗港马产业合作 构建赛马产业经济圈 中共广州 ...

Core Viewpoint - The domestic computing power sector is experiencing a new wave of development, with companies like Biren Technology, Tensun Zhixin, and Suiruan Technology advancing their IPO processes and capitalizing on the growing demand for AI computing power [1][4]. Group 1: Company Developments - Biren Technology and Tensun Zhixin have successfully passed the Hong Kong Stock Exchange listing hearings, indicating strong confidence in their technological capabilities and market opportunities [1]. - Suiruan Technology is also progressing with its IPO, focusing on AI cloud computing solutions and has received significant orders for its products [4][5]. Group 2: Financial Performance - Biren Technology's revenue is projected to grow from 0.499 million in 2022 to 33.7 million in 2024, with a 50% year-on-year increase expected in the first half of 2025 [2]. - Tensun Zhixin's revenue has shown rapid growth, increasing from 189 million in 2022 to 540 million in 2024, with a compound annual growth rate of 68.8% [3]. Group 3: Product Development - Biren Technology is developing next-generation GPGPU chips and enhancing its software platform to support a wide range of AI applications [2]. - Tensun Zhixin is focusing on optimizing its new generation of products for large language models, aiming to support low-precision data types and mixed-precision computing [3]. - Suiruan Technology's "S60" product has already shipped over 100,000 units and supports more than 300 application scenarios [4].

Core Insights - The article discusses the advancements in the domestic GPU industry, highlighting the launch of the "Huagang" architecture by Moore Threads and the "scaleX" supercluster system by Inspur, indicating a shift in focus from individual GPU performance to building scalable systems capable of handling massive computational tasks [2][6]. Group 1: Moore Threads Developments - Moore Threads unveiled its latest "Huagang" architecture, which boasts a 50% increase in computing density and a 10-fold improvement in efficiency compared to the previous generation [3]. - The "Huagang" architecture supports full precision calculations from FP4 to FP64 and introduces new support for MTFP6, MTFP4, and mixed low precision [3]. - Future chip plans include "Huashan," aimed at AI training and inference, and "Lushan," focused on high-performance graphics rendering, with "Lushan" showing a 64-fold increase in AI computing performance and a 50% improvement in ray tracing performance [4]. Group 2: Inspur Developments - Inspur's "scaleX" supercluster system, which publicly debuted, consists of 16 scaleX640 supernodes interconnected via the scaleFabric high-speed network, capable of deploying 10,240 AI accelerator cards [10]. - The scaleX system employs immersion phase change liquid cooling technology to address heat dissipation challenges, achieving a 20-fold increase in computing density per rack and a PUE (Power Usage Effectiveness) of 1.04 [11][12]. - The system supports multi-brand accelerator cards and has optimized compatibility with over 400 mainstream large models, reflecting a strategy to provide a versatile platform for various domestic computing resources [14]. Group 3: Industry Challenges and Solutions - The industry faces challenges in scaling up computational power, particularly in managing heat, power supply, and physical space limitations when deploying thousands of high-power chips in data centers [8][9]. - Both companies are addressing communication delays in distributed computing, with Moore Threads integrating a new asynchronous programming model and self-developed MTLink technology to support clusters exceeding 100,000 cards, while Inspur's scaleFabric network achieves 400 Gb/s bandwidth and sub-microsecond communication latency [12][13]. Group 4: Software Ecosystem and Compatibility - As the hardware specifications approach international standards, the focus is shifting towards optimizing the software stack, with Moore Threads announcing an upgrade to its MUSA unified architecture and achieving over 98% efficiency in core computing libraries [13]. - Inspur emphasizes the compatibility of its systems with various brands of accelerator cards, promoting an open architecture strategy that allows for coexistence of multiple chips [14].

Core Viewpoint - The launch of the scaleX 10,000-card AI supernode by Zhongke Shuguang marks a significant milestone in China's AI computing power history, entering the era of 10,000-card supernodes [1][3]. Group 1: Development of AI Computing Power - The establishment of the scaleX 10,000-card supernode represents a new answer to the development path of China's AI computing infrastructure [3]. - Three years ago, China's AI computing power system heavily relied on NVIDIA for GPU acceleration, NVLink technology, and CUDA software, creating a dependency on a single supplier [4]. - The turning point came with export restrictions on NVIDIA chips, prompting domestic manufacturers to explore alternative computing power systems [4]. Group 2: Competitive Landscape - Major players like Huawei, Inspur, and Alibaba are entering the AI supernode market, each adopting different technological routes [5]. - Huawei has taken a "fully self-developed" approach, while Inspur and Alibaba focus on "open architecture" to build a domestic AI computing foundation [6]. - The scaleX 10,000-card supernode consists of 16 scaleX640 supernodes, totaling 10,240 AI accelerator cards and exceeding 5 EFlops in computing power [7]. Group 3: Technological Innovations - The scaleX640 supernode features a self-developed scaleFabric high-speed network with a bandwidth of 400 Gb/s and an end-to-end latency of less than 1 microsecond [7]. - The system supports multiple brands of accelerator cards, indicating a shift towards a diversified computing power ecosystem in China [7]. Group 4: Industry Trends - The trend of "de-NVIDIA" is driven by the need for computing power security and independent innovation in China, especially following U.S. export restrictions on high-performance GPUs [8]. - The domestic AI industry is not merely replicating NVIDIA but aims to establish a complete, replaceable computing power ecosystem [8]. - The development paths of closed-stack integration, represented by Huawei, and open collaboration, represented by Shuguang, Inspur, and Alibaba, are emerging as two significant trends in the industry [8]. Group 5: Application and Impact - Various products have already been deployed, with Huawei's CM384 and Inspur's SD200 being used in operational data centers [9]. - The open architecture approach has facilitated the large-scale application of domestic chips, moving away from reliance on NVIDIA's ecosystem [9]. - The year 2025 is seen as a turning point for China's AI computing power system, emphasizing the importance of both performance and collaborative ecosystems [11].

Group 1 - Google's AI models are performing exceptionally well, accelerating the evolution of self-developed chips and cluster solutions, which is expected to increase demand for liquid cooling systems [1][2] - The compatibility of TPU with PyTorch is being enhanced, significantly reducing migration costs for enterprises and expanding the market [2] - Google has released TPUv7, achieving 4614 TeraFlops per chip, with a memory capacity of 192GB and a bandwidth of 7.2 Tbps, allowing for cluster scalability up to 9216 chips using liquid cooling solutions [2] Group 2 - Companies like Anthropic and Meta are expressing strong interest in renting TPUs, which will continue to drive demand for ASICs [3] - Anthropic plans to deploy up to 1 million Google TPUs for training its AI model Claude, with a project value in the hundreds of millions of dollars, aiming for a computational capacity of 1GW by 2026 [3] - Meta is negotiating with Google to rent TPUs starting in 2026, with potential investments worth billions of dollars [3] Group 3 - The demand for optical fibers and cables is expected to rise due to the ongoing iteration of AI large model training and the acceleration of AI application deployment [4] - The growth in external data center interconnect (DCI) and metropolitan area network cable demand may lead to a recovery in optical fiber and cable prices [4] Group 4 - The company maintains a positive outlook on three core themes: "light, liquid cooling, and domestic computing power," while also emphasizing the importance of satellite and edge AI [5]

Core Insights - The article discusses the shift in AI from training to inference, highlighting that inference is now surpassing training in terms of power consumption and importance in the industry [22][24]. - The focus is on the evolution of AI technology, particularly in China, where companies like Yuntian Lifei are building their own AI technology systems by investing in both algorithms and chips [5][6]. Group 1: AI Industry Evolution - The AI industry has undergone significant changes since 2014, with a notable acceleration in the pace of technological development, particularly with the advent of large models [18][20]. - The demand for inference capabilities has increased dramatically, with a reported growth of nearly 100 times from last year to this year [8][28]. - By the end of 2024, it is expected that domestic AI chips will account for over 50% of the AI chip market in China, surpassing non-domestic high-end GPUs [28][24]. Group 2: Yuntian Lifei's Strategy - Yuntian Lifei has adopted a dual approach of focusing on both algorithms and chips, which has allowed the company to navigate the complexities of the AI landscape effectively [5][6]. - The company emphasizes the importance of integrating into existing ecosystems, particularly the CUDA ecosystem, to reduce adaptation costs for clients [8][9]. - Yuntian Lifei aims to enhance its core capabilities in inference, ensuring that its technology is both reusable and deliverable, thereby providing clear value to customers [13][31]. Group 3: Challenges and Opportunities - The primary challenge facing AI inference is the cost, as companies strive to make AI more precise while managing expenses [11][12]. - The article highlights the need for a robust ecosystem that supports the integration of various technologies, including the development of standards and protocols for AI chips [12][30]. - The future of AI infrastructure is expected to move towards heterogeneity and high cost-effectiveness, addressing the performance-cost-accuracy trade-off [39][41].

Investment Rating - The report maintains a rating of "Outperform" for the industry, indicating an expected performance that exceeds the market average [2]. Core Insights - The report emphasizes that AI inference will lead innovation, focusing on demand-driven Opex-related areas, specifically storage, power, ASIC, and supernodes [2][28]. - It predicts that 2025 will be a significant year for the expansion of storage capacity, driven by the increasing demand for SSDs and HBM products [2][28]. - The report highlights the importance of domestic supply chains in the storage industry, particularly in NAND and DRAM sectors, and suggests monitoring key players in these areas [2][28]. - The introduction of the GPT-5.2 model by OpenAI is noted as a significant development in the AI sector, showcasing advancements in professional knowledge work [25][26]. Summary by Sections Market Review - The Shanghai Composite Index decreased by 0.34%, while the Shenzhen Component Index increased by 0.84%, and the ChiNext Index rose by 2.74% [11]. - The Shenwan Electronics Index increased by 2.63%, ranking third among 31 Shenwan industries, with a year-to-date increase of 45.9% [11]. Weekly Insights - The report identifies key areas of investment: storage, power, ASIC, and supernodes, with specific companies highlighted for each category [2][28]. - It notes the expected growth in the power industry, particularly in new technologies for both generation and consumption [28]. Related Research - The report references several previous studies focusing on AI inference and domestic semiconductor trends, indicating a consistent theme of growth and innovation in these sectors [4][5][28]. Company Performance - The report details the performance of specific companies within the electronics sector, noting significant gains for firms like Fuxin Technology and Dongtian Micro, while others like ST Huilun faced declines [19][22].

根据招股说明书，沐曦股份本次募集资金将重点投向新型高性能通用GPU研发及产业化等项目，目前公 司的产品应用部署于10余个智算集群，算力网络覆盖国家人工智能公共算力平台、运营商智算平台和商 业化智算中心。 转自：北京日报客户端 今天，国产GPU芯片龙头沐曦股份在上交所科创板上市，发行价格为104.66元/股，截至收盘，公司股 价报829.9元，上涨692.95%，市值达3320亿元。沐曦股份的上市对国内人工智能行业发展将带来哪些影 响？ 国产GPU芯片龙头沐曦股份今日上市 总台央视记者 平凡：今天，沐曦股份登陆上交所科创板，它也是国内GPU芯片和计算平台的提供商， 它的上市将加快国内人工智能基础设施建设。截至目前，国产算力公司包括寒武纪、摩尔线程和沐曦股 份，总市值已经突破万亿元。 沐曦股份高级副总裁 孙国梁：我们是作为国产算力底座的核心供应商这个角色进入到资本市场的。我 们还是要认真花好每一分钱，用在底层的核心技术、核心产品、核心生态的打造。 国产算力进入市场之后，也将带动相关的上下游生态体系，包括芯片制造、操作系统、通用和行业模 型、智能体等核心领域。 专家表示，国内通用大模型和各个行业的垂直模型加快升级和 ...

在上海电影制片厂，国内首部超百分钟生成式人工智能拟真人剧集近日上线，制作周期算短了三分之二，成本大幅降低。项目背后的支撑就是 影视专属算力池，满足实时渲染、文生视频、虚拟拍摄等多样化需求。 上海电影技术厂有限公司副总经理郗岳介绍，预计到2026年，可以实现百部以上的AIGC（生成式人工智能）项目规模化制作。对国产算力， 以及国产算力相关产业生态提出更高需求。 央视网消息：记者了解到，人工智能在制造业、金融、能源、医疗健康等领域的应用加速落地，这对国产算力相关产业链又将带来哪些机会？ 在复旦大学医学院，基于医生、患者和科研等场景的垂直模型和智能体正在开发落地当中，实现医生有AI智能体助手，患者有AI健康顾问。 复旦大学医学院副院长朱同玉介绍，对医生来说是他的一个"大脑"，也是他的"助手"。随时可以给患者提供一些健康服务。而每一个疾病在不 同人群也需要一个小模型，它有无限的想象空间。 在TCL的工厂，显示面板生产线部署垂域大模型，通过智能体实现全域自主监控、分析与决策，相关AI应用创造综合效益超过10亿元。 TCL创始人、董事长李东生表示，把AI的应用在先进制造业，能够实实在在创造出效果。他们一直在积极推动和国内 ...