今天分享的是：华为：2025年Atlas 800T A3超节点技术白皮书 报告共计：77页 Atlas 800T A3超节点技术白皮书核心内容总结 Atlas 800T A3超节点是华为面向互联网、运营商、金融等行业大模型训练场景推出的AI智算产品，以高性能、高可靠、易部署 为核心优势，适配通用风冷机房部署，为数据中心基础设施提供强劲硬件支撑。 该产品由超节点服务器和灵衢总线设备两大部分构成。超节点服务器采用10U机框设计，支持标准19英寸机柜安装，通过正交盲 插架构集成CPU抽屉、NPU抽屉、灵衢总线板、IO框等核心部件，集成密度高且便于维护。核心配置上，搭载4个鲲鹏920处理 器与8个昇腾910 AI模组，整机峰值算力可达6.016PFLOPS@FP16、12.032POPS@INT8，超节点内任意两个NPU模组间双向互联 带宽达784GB/s，满足大规模AI训练的算力需求。 供电与散热系统保障设备稳定运行。供电方面，支持220VAC或336HVDC/240HVDC双输入，最大输入功耗16.2kW，电源转换效 率最高可达96%，并具备5+1备份及多重保护功能。散热采用风冷结合华为自研LAAC液冷模组的方案 ...

科士达近日在投资者关系活动中表示，公司数据中心产品品类齐全，覆盖UPS电源、高压直流电源 （HVDC）、温控设备、铅酸/锂电池、微模块、电力模块、动环监控等全系列产品及系统。顺应中大 型高功率数据中心的建设浪潮，设备端正朝着高功率、高密度方向升级。针对AI智算应用场景，公司 已构建完备的核心产品体系，涵盖适配AI服务器集群的PowerFort一体化电力模块、兆瓦级UPS、 YMK3300高频模块化UPS、IDM风液融合微模块数据中心，以及可满足高算力散热需求的LiquiX AI智 算液冷CDU等关键产品。 ...

12月5日，摩尔线程正式登陆科创板。上市首日，摩尔线程竞价高开超468%，总市值突破3000亿。随后 持续上扬，一度涨超540%。截至发稿，报590元/股，涨幅为416.29%，总市值为2769亿元。 若按当前价格计算，投资者中一签浮盈近24万元。 由于GPU行业属于典型的资本与技术双密集型领域，导致其研发投入居高不下。2022年至2025年上半 年，公司累计研发投入高达43.66亿元，是同期累计营收的6倍以上。 在此背景下，摩尔线程也尚未实现盈利。2022年至2024年，其归母净利润分别为亏损18.94亿元、17.03 亿元及16.18亿元。今年上半年，归母净利润亏损收窄至2.71亿元。 据摩尔线程披露，目前公司在手订单充足，AI智算领域正在洽谈的项目合同金额超过17亿元，预计最 早可于2027年实现合并报表盈利。 摩尔线程成立于2020年，核心壁垒在于国产全功能GPU路线。其自研的MUSA架构已具备与由英伟达 主导的国际主流GPU生态的兼容性，目前已推出四代GPU架构芯片。上市后，将保持每年一代的迭代 速度。 股权结构上，创始人张建中直接、间接持有30.94%股份，为摩尔线程实控人。值得注意的是，摩尔线 ...

证券代码：600633 证券简称：浙数文化 公告编号：2025-068 ■ 一、合作投资基本概述情况 2025年11月25日，公司全资子公司杭州富春云科技有限公司（以下简称"富春云科技"）与杭数管理咨询 （杭州）有限公司（以下简称"杭数管理咨询"）、浙江甬元财通富浙高端装备产业股权投资合伙企业 （有限合伙）（以下简称"浙江甬元财通富浙基金"）共同签订《杭州杭数智能壹号创业投资基金合伙企 业（有限合伙）合伙协议》，富春云科技将作为有限合伙人参与设立杭州杭数智能壹号创业投资基金合 伙企业（有限合伙）（以下简称"基金""合伙企业"）。本次合伙企业总认缴出资金额为8,500万元，富 春云科技拟认缴出资2,100万元，占合伙企业本次总认缴出资金额的24.71%。具体内容详见公司2025年 11月27日披露于《上海证券报》和上海证券交易所官网（www.sse.com.cn）的《浙数文化关于参与设立 杭州杭数智能壹号创业投资基金合伙企业（有限合伙）的公告》（公告编号：临2025-066）。 二、本次对外投资进展情况 2025年12月3日，富春云科技与杭数管理咨询、浙江甬元财通富浙基金共同签订新的《杭州杭数智能壹 号创业投资 ...

Core Viewpoint - The domestic GPU company Muxi Co., Ltd. is set to go public on the Sci-Tech Innovation Board with an issue price of 104.66 yuan per share, lower than the 114.28 yuan per share pricing of its competitor, Moer Thread, which will debut as the "first domestic GPU stock" on December 5 [2] Group 1: Product Development and Performance - Muxi's main GPU chip, the Xiyun C600 series, is positioned between NVIDIA's A100 and H100 in terms of performance, with risk mass production expected by the end of 2025 and formal production in the first half of 2026 [3] - The Xiyun C700 series, which is currently in the design and development phase, aims to support low-precision calculations and expand its application in various fields beyond the AI training and inference scenarios [4] - Muxi claims that its Xiyun C series products can be used for model pre-training, having successfully supported large models with billions of parameters [5] Group 2: Financial Performance and Projections - Muxi's revenue is projected to grow significantly from 424,000 yuan in 2022 to 743 million yuan in 2024, with a compound annual growth rate of 4074.52%, and 915 million yuan in the first half of 2025, reflecting a year-on-year growth of 404.51% [6] - The company expects to achieve annual revenue between 1.5 billion to 1.98 billion yuan in 2025, representing an increase of 101.86% to 166.46% compared to 2024, while net losses are projected to decrease significantly [7] - Muxi's management anticipates reaching breakeven as early as 2026, driven by stable gross margins and increased sales from new product developments [8]

报告导读： AI智算是人工智能与高性能计算融合产物，以"算力+算法+数据"支撑AI模型训练推理，是数字经济关键基础设施。当前，全球AI市场进入"应用红利期"， 算力架构协同高效，多模态与具身智能提升模型实用性，科技巨头与终端厂商构建全栈生态，AI应用广泛渗透，行业竞争焦点转变。数据显示，2024年全 球AI市场规模达2340亿美元，预计2032年突破2.75万亿美元。全球算力格局加速，智能算力成核心驱动力，但需求攀升致"需求—建设—供给"三重错配。中 国2024年算力总规模280 EFLOPS，资源紧张，2025年智能算力增量有望突破100EFLOPS。2024年我国智能算力规模90 EFLOPS，智算中心成地区级AI新型 基础设施核心。企业竞争格局方面，全产业链竞争形成，基础设施层华为主导，算力服务层运营商与云厂商主导，模型应用层企业推动研发与落地。未来， 行业将"协同进化"，算力架构协同调度，国产化生态与国际并行，重心转向与实体经济深度融合。 基于此，依托智研咨询旗下AI智算行业研究团队深厚的市场洞察力，并结合多年调研数据与一线实战需求，智研咨询推出《2026-2032年中国AI智算行业市 场分析研究及投 ...

Core Viewpoint - Qingyun Technology continues its strategic partnership with Moore Threads, focusing on product adaptation and certification, as well as deep integration in AI computing management platforms [1] Company Collaboration - Qingyun Technology has engaged in product adaptation and mutual product certification with Moore Threads [1] - The collaboration includes deep integration and cooperation in AI computing management platforms, leveraging Qingyun's expertise in KubeSphere and intelligent computing operations [1] - As of November 24, 2025, the strategic partnership between Qingyun Technology and Moore Threads remains active [1]

11月24日新股申购 (共1J | | | 摩尔线程(科创板,688 | | | --- | --- | --- | --- | | 申购代码 | 787795 | | | | 发行价格 | 114.28元 | | 顶格 | | 总市值 | 457.2亿 | | | | 发行市盈率 | 27.06 | | 行业 | | 同业可比公司 | 海光信息 同业公司 动态市盈率 | | | | | 寒武纪 | | | | | 风神股份 | | | | 主营构成 | AI智算: 76.54%; | | | | | 专业图形加速：19.30%； | | | | | 桌面级图形加速：2.45%; | | | | | 其他: 1.71% | | | | | 1、公司是国内极少数具备全功能GPU研发能力β | | | | | 位，公司已成功推出四代GPU架构，并拓展出覆; | | | | 业务及亮点 | 域的计算加速产品矩阵。 | | | --- | --- | --- | | | 2、公司MTT S80显卡的单精度浮点算力性能接近 | | | | S5000产品构建的千卡GPU智算集群效率超过同 | | | | 3、基于自主研 ...

Core Insights - The company, Moer Thread Intelligent Technology (Beijing) Co., Ltd., is primarily engaged in the research, design, and sales of GPU chips and related products, focusing on full-function GPUs [2] - Moer Thread has launched four generations of GPU architecture and intelligent SoCs, covering a comprehensive range of computing acceleration products in AI computing, high-performance computing, and graphics rendering [2] - The company has established a full-stack AI product line that spans cloud, edge, and end devices [2] Company Development - During the IPO guidance period, the rapid development of Moer Thread was observed, highlighting the company's commitment to an innovative development culture [2] - The management team's perseverance and dedication were notably admired, indicating strong leadership within the company [2] - The company aims to enhance corporate governance and information disclosure while striving for profitability to deliver excellent performance to investors [2]

Core Insights - The AI computing industry is transitioning into an "application dividend period," driven by the integration of artificial intelligence and high-performance computing, with a projected global market size of $234 billion in 2024 and expected to exceed $2.75 trillion by 2032, reflecting a compound annual growth rate of 36.8% [5][6][10] AI Computing Industry Overview - AI computing (Intelligent Computing) combines AI with high-performance computing, utilizing dedicated hardware and distributed architecture to provide efficient and scalable computing power for AI tasks [2][3] - The industry is characterized by a three-pronged model of "computing power + algorithms + data," which supports complex AI model training and inference, making it a critical infrastructure for the digital economy [2][3] Global and Chinese AI Computing Development - The global AI market is evolving from a "model dividend period" to an "application dividend period," becoming a core engine for the digital economy, with significant advancements in computing power architecture [5][6] - In 2024, the total global computing power is expected to reach 2207 EFLOPS, with intelligent computing power contributing 1610 EFLOPS, marking a year-on-year growth of 63.8% [5][6] - China's total computing power is projected to reach 280 EFLOPS in 2024, with intelligent computing power at 90 EFLOPS, indicating a significant gap between supply and future demand [6][10] AI Computing Industry Chain in China - The AI computing industry chain in China consists of upstream hardware (GPUs, NPUs), midstream service providers (telecom operators, cloud service providers), and downstream application sectors (internet, finance, manufacturing) [8][10] - The market for AI acceleration chips is expected to grow from 17.56 billion yuan in 2020 to 142.54 billion yuan in 2024, with a compound annual growth rate of 68.8% [9][10] Competitive Landscape - Major players in the infrastructure layer include Huawei, which leads in domestic chip production, and Inspur, which dominates the AI server market [10] - Telecom operators and cloud service providers, such as China Telecom, Alibaba Cloud, and Tencent Cloud, are key players in the computing service layer [10] Future Trends in AI Computing Industry - The industry is expected to evolve towards collaborative evolution, with a focus on efficient resource utilization and deep integration with the real economy [11][12] - The emphasis will shift from model performance to creating measurable business value through large-scale applications in key sectors such as manufacturing, finance, and healthcare [13]