1月30日晚间，国产AI芯片龙头企业寒武纪发布2025年年度业绩预告，交出了一份亮眼的成绩单。公司不仅实现营业收入同比超4倍增 长，更迎来发展史上的重要里程碑——首度实现年度归母净利润、扣非净利润双盈利，彻底扭转此前的亏损局面。 早在2025年第一季度，公司就已展现出强劲的增长势头，营收同比激增4230.22%，归母净利润与扣非净利润同步实现转正，为全年业 绩奠定了良好基础。进入第二季度，增长趋势依旧延续，半年报显示公司上半年营收同比增长4347.82%，净利润同样扭亏为盈。第三 季度公司营收与利润继续保持高增，至此2025年前三季度公司累计营收达46.07亿元，同比增长2386.38%；归母净利润16.05亿元，已接 近全年业绩预告的下限。以此测算，寒武纪2025年第四季度营收区间为13.93亿元至23.93亿元，净利润区间为2.45亿元至5.45亿元，依旧 保持着稳健的经营态势。 对于业绩大幅增长并实现扭亏的原因，寒武纪在公告中表示，核心得益于人工智能行业算力需求的持续攀升，同时公司始终深耕人工智 能芯片产品的研发与技术创新领域，凭借产品的优异竞争力持续拓展市场，积极推动人工智能应用场景落地，最终带动营业 ...

今日晚间，寒武纪发布公告，预计2025年实现营业收入60亿元-70亿元，同比增长410.87%到496.02%； 归属于母公司所有者的净利润为18.5亿元-21.5亿元，与上年同期亏损4.52亿元相比，实现扭亏为盈。 面对每天上千份上市公司公告该看哪些？重大事项公告动辄几十页几百页重点是啥？公告里 一堆专业术语不知道算利好还是利空？请看智通财经公司新闻部《速读公告》栏目，我们派 驻全国的记者们将于公告当晚为您带来准确、快速、专业的解读。 智通财经1月29日讯（记者 王碧微）曾长期身处亏损泥潭的国产AI芯片龙头寒武纪-U（688256.SH）， 交出了一份营收与净利润大幅增长的成绩单。 值得注意的是，若2025年年度报告审计结果与预告一致，寒武纪将符合科创板"上市时未盈利公司首次 实现盈利"的相关规定，证券简称中的特别标识"U"有望摘除。这将是公司自2020年7月上市以来的一个 关键节点。 自2024年以来，受限于美国出口管制政策，英伟达等国际芯片巨头的高端算力产品无法向中国市场正常 供应。 尽管英伟达后续推出了针对中国市场的特供版芯片H20，并放开了H200入华，但市场对于供应链安全及 长期技术支持的担忧始终 ...

特朗普一语，美股崩盘 芯片是国家科技实力的底子，在全球竞争里谁也绕不开。美国一直想牢牢抓住这个优势，从2022年开始 就对中国芯片产业使劲压着。拜登那时候签了芯片法案，总额2800亿美元，里面有五百多亿直接补贴半 导体厂，还有两百多亿税收优惠，鼓励企业在美国本土搞研发和生产，另外两千亿扔给科研。 结果，美国商务部直接禁了向中国出口高端芯片设备，还拉着荷兰日本这些盟友一块儿玩，目的就是卡 住中国技术链条。荷兰的阿斯麦公司没法卖极紫外光刻机给中国，日本东京电子也跟着调整出口。中国 企业只好自己想办法，转向本土供应链，加速自主研发。 特朗普2024年大选赢了，重回白宫，他的思路跟拜登不一样，更爱用关税当武器。2025年3月4日，他在 国会演讲里把芯片法案批得一文不值，说这是浪费钱，最好废掉它，用那些资金减债。 中国天才，逆境腾飞 陈云霁1983年出生在江西南昌，家里条件一般，但他从小数学就牛。小学时候就搞定二元一次方程，还 拿了全国奥数一等奖。1997年，14岁考上中科大少年班，19岁去中科院计算所读硕博连读，24岁博士毕 业，25岁就当了中科院8核龙芯的主架构师。 特朗普觉得补贴芯片公司没用，那些企业拿了钱也不见 ...

制图:黄亚岚(元宝AI、豆包AI生成) 2025年，沉淀已久的国产AI芯片迎来关键性转折。在DeepSeek助力下，AI算力需求引爆，国产替代被 按下"加速键"。寒武纪股价狂飙，屡次登上科创板"股王"之位。国产替代年底迎来资本"盛宴"——摩尔 线程和沐曦股份先后顶着"国产GPU第一股"和"第二股"的光环登陆科创板，壁仞科技和天数智芯也在港 交所接连传出好消息，将在2026年年初正式赴港上市。几乎同期，市场传出英伟达H200被批准对华出 售的消息，业内探讨这是否会挤压国产AI芯片市场空间。今年国产AI芯片行业的发展将去往何处? 趋势1 DeepSeek引爆国产替代 国产芯片在推理赛道有机会 2025年，国产AI芯片赛道可谓是热闹非凡。2025年年初，DeepSeek横空出世，随其"破圈"而来的是AI 应用侧的大爆发，AI在口口相传间真正走入了寻常百姓家，这也意味着算力需求逐步从训练转向推 理。之后，华为打"头阵"，一众国产AI芯片厂商宣布支持DeepSeek并快速部署，国产算力热情被"点 燃"。 2025年8月，DeepSeek又发布了新一代模型DeepSeekV3.1，并在公众号留言中强调，"UE8M0FP8 ...

Group 1 - The Ministry of Industry and Information Technology and eight other departments issued the "Implementation Opinions on the Special Action of 'Artificial Intelligence + Manufacturing'" which emphasizes strengthening the supply of artificial intelligence computing power [1] - The opinions promote the development of intelligent chips and support breakthroughs in key core technologies such as high-end training chips, edge inference chips, AI servers, high-speed interconnection, and intelligent cloud operating systems [1] - By 2027, China's key core technologies in artificial intelligence are expected to achieve safe and reliable supply, with the industry scale and empowerment level remaining among the world's top [1] Group 2 - The analysis indicates that the safe and reliable supply of key artificial intelligence technologies is crucial for achieving a fully autonomous and controllable industrial ecosystem, reducing dependence on external technologies [2] - The focus has shifted from "technology selection" to "strategic necessity," which will accelerate the replacement of domestic computing power, algorithms, and security technologies in the short term [2] - In the long term, this will reshape the global AI industry landscape and provide a "secure foundation" and "innovation engine" for the high-quality development of China's digital economy [2] Group 3 - Haiguang Information is a leading enterprise in the domestic high-end processor R&D field, with its CPU and DCU driving growth, and the upcoming release of the next-generation Haiguang 5 and Deep Computing 4 is expected to maintain rapid revenue growth [3] - Cambrian is a leading company in the domestic AI chip sector, focusing on AI chip R&D and technological innovation, with a product matrix covering cloud, edge, and terminal scenarios [3] - Cambrian's products include the Siyuan 590 and Siyuan 370 cloud training and inference chips, which meet the AI application needs across various industries such as internet, finance, transportation, and energy [3]

Group 1 - The core viewpoint of the article highlights Beijing's rapid development as a leading hub for artificial intelligence (AI), showcasing significant advancements in AI technology, particularly in chip development and large model applications [1][2][3] - As of the first half of 2025, Beijing's AI core industry is projected to reach a scale of 215.2 billion yuan, with an expected annual growth rate of 25.3%, potentially reaching 450 billion yuan by the end of the year [1] - Beijing has established a "chip matrix" in the AI chip sector, featuring leading domestic products such as Kunlun, Cambricon, and Moore Threads, which are now integral to real business applications [3][8] Group 2 - The article emphasizes the importance of a cohesive ecosystem that integrates algorithms, chips, and data, enabling AI technologies to transition from laboratories to everyday applications [2] - The launch of the FlagOS 1.6 system software stack by Zhiyuan Research Institute aims to unify various AI chips and frameworks, addressing the compatibility challenges faced by large models [7][8] - Beijing is home to 209 registered large models, accounting for nearly 30% of the national total, with notable models like Doubao and Kimi being developed locally [9][10] Group 3 - The article discusses the innovative capabilities of Beijing's AI models, such as the GLM-4.7-flash, which enhances training efficiency while reducing resource consumption [10][11] - The MiniCPM-o 4.5 model, developed by Mianbi Intelligent, is highlighted for its unique features, including full-duplex and multimodal capabilities, showcasing advancements in human-computer interaction [12][13] - The "density law" of large models, which suggests that their capability density doubles approximately every 3.5 months, has been validated by recent research and is a key aspect of Beijing's AI development [13] Group 4 - Beijing plans to implement nine major actions to further establish itself as a global AI innovation hub, focusing on technology innovation, data quality, and application empowerment [14][19] - The RoboFinals platform, introduced by Guanglun Intelligent, represents a significant advancement in industrial-grade evaluation for embodied intelligence, creating a closed-loop system for data generation, model training, and capability assessment [18][19] - The article concludes by emphasizing the collaborative efforts of various stakeholders in Beijing's AI ecosystem, which have contributed to its sustainable growth and resilience [20]

12月24日，一则消息引发广泛关注：英伟达已告知中国客户，计划于2026年2月中旬（农历春节前）交 付AI芯片H200。 回顾过去三年，英伟达对中国市场的供应策略可谓步步退让。 2022年9月，美国首次禁止A100/H100对华出口，英伟达迅速推出"阉割版"A800/H800以维持中国市场营 收；2023年10月，美方再度收紧管制，连A800/H800也被列入禁售清单。同年12月，英伟达推出进一步 降标准的L20和H20芯片，但因性能缩水严重（H20算力约为H100的1/3）、价格却高达一半，被一些业 内人士怒斥为"智商税"。 彼时，中国市场几乎陷入"无高端GPU可用"的困境。尽管非官方渠道可能仍有少量H100/A100流入，但 价格翻倍、供应不稳，难以支撑大规模智算中心建设。与此同时，国家互联网信息办公室于2025年7月 底就H20芯片"漏洞后门安全风险"约谈英伟达，释放出对"特供版"安全性的高度警惕。 在此背景下，此时H200的回归显得颇为微妙。据蓉和半导体咨询CEO吴梓豪透露，基于性能密度 （TPP）来看，H200的算力值为15832，大约是H20的6.7倍。一方面，它确实填补了市场空白，可满足 大模型 ...

Core Viewpoint - The domestic GPU industry has experienced a significant surge in capital interest, with companies like Moore Threads and Muxi Co. achieving remarkable stock performance upon their listings, indicating a strong market belief in the potential of domestic GPU technology in the AI sector [1][3][4]. Group 1: Market Performance and Trends - In August, AI chip company Cambricon briefly surpassed Kweichow Moutai to become the "new king" of A-shares, while Moore Threads debuted on the STAR Market with a first-day increase of 468.78%, closing at 600.50 CNY per share, a rise of 425.46% [1]. - Muxi Co., the "second domestic GPU stock," listed on December 17, closing at 829.90 CNY per share, marking a 692.95% increase and a market capitalization of 332.04 billion CNY [1]. - The capital frenzy is expected to continue, with other companies like Birran Technology and Suiruan Technology also advancing their listing processes, and Kunlun Chip Technology, a subsidiary of Baidu, preparing for a potential Hong Kong listing in the first half of 2026 [3][12][16]. Group 2: Investment Dynamics - The market has seen a rush for shares of Moore Threads and Muxi Co., with Moore Threads attracting 267 institutional investors managing 7,555 allocation objects, and Muxi Co. drawing 269 investors managing 7,719 allocation objects, resulting in a total effective subscription of 57.169 billion shares [5]. - The final subscription rates for both companies were extremely low, with Moore Threads at 0.0364% and Muxi Co. at 0.0335%, highlighting the intense competition for shares [6]. Group 3: Early Investor Returns - Early investors in Moore Threads, such as the "Peixian Qianyao" fund, saw returns exceeding 6200 times their initial investment of 1.9048 million CNY, holding 16.9987 million shares by the time of the IPO [9]. - In Muxi Co., the lead investor, Heli Guoxin Zhixin Fund, invested 500 million CNY in its first round and held 14.3789 million shares, resulting in a market value of nearly 12 billion CNY at the IPO [11]. Group 4: Industry Context and Challenges - The GPU industry has gained attention since the rise of AI applications, with domestic companies benefiting from the exit of NVIDIA from the Chinese market due to U.S. export controls, creating a market vacuum worth hundreds of billions [24][26]. - Despite the growth potential, most domestic GPU companies are currently operating at a loss due to high R&D expenditures, and there are concerns about the sustainability of their valuations amid market speculation [27][29]. - The competitive landscape is evolving, with NVIDIA showing signs of re-entering the Chinese market, which could intensify competition for domestic GPU firms [29].

12月15日晚间，AI芯片厂商寒武纪发布公告称，公司拟使用母公司资本公积27.78亿元用于弥补母公司累计亏损。本次公积金弥补亏损以公司2024年末母公 司未分配利润负数弥补至零为限。 热门ETF方面，科创人工智能ETF（589520）标的指数均衡配置应用软件、终端应用、终端芯片、云端芯片四大环节，AI产业链正由云端向边缘侧发展，从 依赖海外技术向自主可控发展，科创人工智能方向契合AI产业链当下现状，或具备更大潜力。 盘面上，12月16日早盘，重仓端侧AI的——科创人工智能ETF（589520）随市回调，场内价格现跌1.77%，该ETF场内宽幅溢价，显示买盘资金更为强势， 或有资金逢跌进场抢筹！事实上，该ETF昨天单日吸金511万元。 此前，寒武纪业绩表现亮眼，东海证券指出，2025年前三季度该公司营收与归母净利润同比大幅增长，主要系以思元590为代表的云端产品大幅放量所致， 后续思元690也有望受益于量价齐升的逻辑。在中美科技摩擦的背景下，国产AI芯片迎来加速替代期。 值得关注的是，端侧AI领域不断涌现新成果，字节跳动旗下豆包手机助手技术预览版12月初发布；阿里巴巴正式开售自研的夸克AI眼镜。政策面上，国务 ...

Group 1 - The core point of the article is that AI chip manufacturer Cambricon plans to use 2.778 billion yuan from its capital reserve to offset accumulated losses, which will not affect its net assets or cash flow, and will allow for potential dividend distribution after the adjustment [1][6] - Analysts noted that the capital reserve adjustment is an internal accounting measure that increases retained earnings without actual cash flow impact, thus posing no operational risks [1][6] - Cambricon's performance has been strong, with significant revenue and net profit growth expected in the first three quarters of 2025, driven by the increased sales of its cloud products [1][6] Group 2 - The AI application sector is seen as having greater potential compared to computing power, with the investment logic shifting from hardware to actual applications and commercialization [2][7] - The Sci-Tech Innovation AI ETF (589520) is strategically positioned to capture the AI industry chain, focusing on software applications, terminal applications, and chips, indicating a shift towards domestic technology [2][7] - On December 16, the ETF experienced a price drop of 1.77%, but still showed strong buying interest, indicating potential for future investment [2][8] Group 3 - The importance of information and industrial security has been highlighted in the context of technological friction, emphasizing the need for self-reliance in core technologies like AI [4] - The ETF and its associated funds are heavily invested in the domestic AI industry, with over 70% of the top ten holdings focused on semiconductor sectors, indicating a high concentration and aggressive investment strategy [4]