前不久，机器人初创公司Dyna Robotics（达纳灵动）宣布完成1.2亿美元（约合8.5亿元人民币）A轮融 资，此轮融资后该公司估值突破6亿美元（约合42.7亿元人民币）。 值得一提的是，英伟达、亚马逊、三星和LG等行业巨头参与了此次融资，投资者具体包括专注于机器 人领域的投资基金Robostrategy、CRV、First Round Capital领投，Salesforce Venture、英伟达、亚马逊、 三星和LG Technology Ventures参投。 Dyna Robotics成立于2024年9月，总部位于美国德克萨斯州休斯顿，在上海设有硬件研发中心。在今年3 月，该公司完成了超2000万美元的种子轮融资，由CRV、First Round Capital领投，真格基金参投。彼时 该公司估值约1亿美元。 不过与一些人形机器人公司不同，Dyna Robotics在硬件端的重点并非人形产品。目前，公司的固定式机 械臂已经应用于多个行业，包括为餐厅折叠餐巾、为健身中心整理毛巾以及为自助洗衣店处理衣物。这 些看似简单的任务，实际上需要高度的精确性和适应性，是机器人在现实世界中运作的重要一步。 Li ...

【导读】华为轮值董事长徐直军：未来3年，昇腾芯片将以"几乎一年一代算力翻倍"的速度演 进 中国基金报记者 王建蔷 9月18日，华为全联接大会2025在上海召开，华为轮值董事长徐直军分享了昇腾芯片未来三 年的发展规划路线，并推出全球最强超节点和集群。 他透露，未来3年， 华为开发和 规划了三个系列，分别是Ascend（昇腾）950系列以及 Ascend 960、Ascend 970系列，更多具体芯片还在规划中。此外，以昇腾芯片为基础，华 为推出多款超节点和集群产品，到2027年昇腾超节点集群规模将达到百万卡级。 未来3年 昇腾芯片将以"几乎一年一代算力翻倍"的速度演进 徐直军认为，尽管DeepSeek模式减少了算力需求，但走向AGI和物理AI，算力仍是关键，尤 其是对中国人工智能而言。 他明确表示，未来3年至2028年，华为规划了Ascend 950、960、970三个系列昇腾芯片， 将以"几乎一年一代算力翻倍"的速度演进，同时，围绕易用性、数据格式、带宽等方向优 化，满足AI算力增长需求。 徐直军介绍，Ascend 950系列（含Ascend 950PR和Ascend 950DT两颗芯片）共用 Ascend ...

英伟达预计为此项目投入30亿美元，未来将在美国落地开城。作为对比，行业领导者Waymo为达成当前运营规模累计投入约120亿美元。 英伟达发力Robotaxi的逻辑并非简单业务扩张，而是推出Robotaxi技术样本，验证其从GPU芯片到物理AI大模型的全链路工程能力，进而定义下一代"物理 AI"的基础设施与生态标准。 2025年美国Robotaxi市场正加速发展。Waymo已在美国7座城市运营，每周提供超过25万次付费出行。特斯拉于今年9月在奥斯汀和湾区开放Robotaxi服务， 首日下载量超过Uber 40%，比Waymo历史最高下载量高出6倍。 英伟达做 Robotaxi, 马斯克 你怎么看? it T 赤 示 不 止 大 字母榜 据36氪报导，英伟达内部正在孵化全新的Robotaxi项目，由资深高级总监Ruchi Bhargava负责。英伟达采用了一段式端到端技术路线，核心在于通过仿真技 术构建的世界模型对神经网络进行强化训练。这与特斯拉FSD采用的路线高度相似，但英伟达拥有更强大的技术基础。 该技术路线的关键支撑是英伟达今年1月发布的Cosmos世界基础模型。这一平台整合文本、图像、视频及传感器数据，生 ...

韩国科学技术信息通信部长官裴庆勋日前在就任50天的首次新闻发布会上，阐释了韩国未来五年科技发 展重点方向，包括到2030年跻身世界AI强国前三名，建立人人都能使用AI的社会等。 裴庆勋表示，韩国计划成立新的科学技术与人工智能部长级理事会。他说，韩国2026年AI领域预算将 超过10万亿韩元，因此需要消除各部委之间的预算重叠提高效率。科学技术与人工智能部长级理事会将 作为政府层面的机构，负责监督人工智能政策。已升任副总理的科学技术信息通信部部长将担任该理事 会主席，预计理事会将于10月在国会全体会议和内阁会议后正式成立。 裴庆勋透露，韩国还将发展物理AI。他说，凭借在制造业和软件开发方面的优势，韩国可在医疗保 健、造船、国防和内容领域实现AI转型，从而在物理AI领域获得全球竞争力。 （文章来源：科技日报） 韩国还将建设AI包容型社会。裴庆勋表示，尽管韩国目前在ChatGPT使用率方面位居世界第一，但必须 把最佳替代方案开发成韩国自己的AI基础模型，并将其开源，以便全世界任何人都可以使用。"包容型 AI"意味着任何人都可使用和接触它，无论年龄、性别或经济状况如何。到2030年，韩国将确保20万台 用于AI学习和数 ...

机构调研要点 问：公司年初发布的物理 AI 产品包含哪些功能？ 答：索辰开物系列涵盖(1) 物理 I 开发及应用平台的全场景解决方案，使开发者和工业用户能够轻松开 发和部署物理 I应用。这个是与英伟达的 omnivorous 对标的一款产品，严格意义来说是一种生态，公司 物理 I开发平台具备非常多的功能模块，包含生成式建模仿真一体化技术、实时环境渲染、物理 I模拟 引擎、实时环境感知、实时数据库等核心功能。(2) 基于物理 I 的自动设计训练一体化系统，实现成千 上万的设计样本智能衍生、验证与训练；生成式物理 I计算引擎结合自动 3D 实景渲染技术，能够高效 构建高保真的虚拟验证环境，精准复现和预测装备与环境之间的实时、多维互动，并为具身智能、低空 经济、深海经济、智能工业等场景提供高精度 3D仿真训练场。 截至2025年9月12日收盘，索辰科技（688507）报收于96.89元，较上周的94.73元上涨2.28%。本周，索 辰科技9月9日盘中最高价报99.99元。9月8日盘中最低价报94.0元。索辰科技当前最新总市值86.74亿 元，在软件开发板块市值排名58/133，在两市A股市值排名2140/5153。 ...

例如，在演示机器人叠衣服的操作场景时，由于现实世界中衣服的纹理、颜色、款式种类繁多且不断更 新，相关数据根本无法完全采集。 朱政认为，世界模型有望解决这一困境——通过生成式数据填补真实数据的不足，从而让物理世界的数 据在体量与泛化性上，达到与数字世界互联网数据相媲美的效果。 王业全则认为物理AI发展的最大难点在于建模方法，包括数据、训练流程等，目前尚未有明确的方 向。他建议第一步可利用大规模自监督数据、高质量数据得到一个基础模型，这需要耗费大量资源，第 二步通过对齐、强化等工作，将模型能力释放到具身智能等相关物理世界相关场景中，"这条路线或许 是可行的，但客观来讲难度依然很大，因为我们目前还未找到合适的建模方法与合理的数据组织形 式"。 智源研究院FLM团队负责人、"新一代人工智能"国家科技重大专项负责人王业全则通过对比ChatGPT的 发展历程，给出了更长期的预估。他指出，从GPT-1到GPT-2，再到GPT-3乃至ChatGPT，期间经历了多 年的技术积累与迭代。目前，具身智能AGI发展水平大致处于潜在的GPT-1.6、1.7阶段，按照线性时间 估计，在一两年内达到ChatGPT级别难度还是很大。王业全认 ...

目前，瑞声科技位列全球乘用车音响系统市场前十，是其中唯一的中国企业。公司每年将营收的7%至 10%投入研发，在全球19个国家和地区设有研发与生产中心，累计专利申请量超过18,000件。持续的技 术投入为其技术发展和业务拓展提供了支持，公司也曾入选"全球百强创新机构"榜单。 来源：环球网 2025年9月9日，瑞声科技（AAC Technologies）以"The Future of Sound in Motion"为主题，在德国慕尼 黑国际车展亮相，与普瑞姆（PSS）、初光汽车（First Light Auto Parts）及先锋高科（PSG）等企业共 同展示其在车载声学、触觉反馈、光学影像、传感交互与电机系统等领域的技术成果。 自2018年进入车载声学领域以来，瑞声科技通过多次战略收购实现了关键技术的整合，目前已构建 从"感知-处理-输出"的声学生态链。基于在声学方面的技术积累，瑞声科技成为能够提供硬件、软件、 算法与调音等全链路服务的车载声学解决方案供应商之一。 展会期间，瑞声科技展出了包括全栈声学解决方案在内的五大技术板块，吸引了众多观众参观。多家全 球知名车企代表，如奥迪、宝马、保时捷、长安汽车、大众、 ...

Core Insights - The evolution of automation, particularly through physical AI, is transforming industrial operations, enhancing flexibility and risk resilience in manufacturing environments [1][6][11] - The current challenges faced by manufacturers, including rising costs, labor shortages, and fluctuating customer demands, necessitate the adoption of advanced robotic technologies [6][11] Group 1: Evolution of Industrial Robots - The application of physical AI represents the next step in the long-term evolution of industrial robots, which date back to the 1960s [2] - Early industrial robots were rule-based, designed for high-precision, repetitive tasks, but lacked flexibility [2][3] - Training-based robots are now emerging, capable of learning from experiences and handling variable tasks, making them suitable for medium-volume and non-repetitive production [3][5] Group 2: Importance of Physical AI and Intelligent Robots - The integration of robotic technology is timely for manufacturers, addressing vulnerabilities in supply chains exacerbated by geopolitical tensions and material shortages [6][7] - Intelligent robots enhance operational flexibility, enabling manufacturers to meet evolving customer demands for customization, faster delivery, and sustainability [6][7] Group 3: Workforce Development for Automation - A skilled workforce is essential for the successful integration of intelligent robots, with a focus on reskilling and continuous learning [8] - The transition to automation will not eliminate jobs but will transform them, creating new high-skill positions in robotics and AI [8] Group 4: Real-World Applications of Physical AI - Early adopters like Amazon have demonstrated the value of intelligent robots, achieving a 25% increase in efficiency and a 10% improvement in operational efficiency through the deployment of over 1 million robots [10] - Foxconn is transitioning to an AI-driven robotic workforce, utilizing digital twin technology to automate complex tasks, resulting in a 20-30% reduction in production cycles and a 15% decrease in operational costs [10] Group 5: Leveraging the Value of Physical AI - The momentum of intelligent robots is expected to grow, with manufacturers urged to adopt a layered automation strategy that integrates various robotic technologies [11] - Companies should prioritize a human-centered approach to ensure sustainable and inclusive integration of robotic systems [11]

Core Insights - The era of robotics is emerging, with the Robotics ETF (562500) deeply embedded in this transformation [1] - Tesla's "Master Plan" emphasizes that 80% of its future value will come from the Optimus robot, indicating a significant shift towards automation [1][4] - Nvidia's launch of the Jetson AGX Thor developer kit marks a pivotal moment in the advancement of physical AI and robotics [1][5] - China's robotics industry is transitioning from a follower to a disruptor, showcasing significant advancements and applications in various sectors [1][6] Tesla's Robotics Strategy - Tesla's "Master Plan" highlights a growing focus on AI and robotics, aiming for large-scale production of robots to replace human labor [4] - Elon Musk projects that by 2029, the production of Optimus robots could reach 1 million units annually [4] - The Optimus 3 robot is expected to be officially launched by the end of this year, marking a significant milestone in Tesla's robotics ambitions [4] Nvidia's Role in Robotics - Nvidia's Jetson AGX Thor is described as a comprehensive technology stack for robotics, integrating advanced hardware, software tools, and AI models [5] - The company aims to build a complete ecosystem for robotics, covering all necessary components from foundational models to specialized hardware [5] China's Robotics Industry - China's share of the global robotics market is projected to reach approximately 40% by 2024, with the market size expected to grow from $47 billion in 2024 to $108 billion by 2028 [8][9] - China leads the world in industrial robot installations, accounting for over half of the global total [9] - By 2030, it is estimated that China will have 252,000 humanoid robots, increasing to 302 million by 2050, surpassing the U.S. market significantly [9] Robotics ETF Insights - The Robotics ETF (562500) has seen substantial growth, with a current size of 17.032 billion yuan, reflecting strong market interest in the robotics sector [13] - The ETF includes a significant proportion of humanoid robot stocks, with 63.67% of its components related to humanoid robotics [13] - The index tracks 73 stocks, with the top ten accounting for 48.86%, providing exposure to leading companies in the robotics field [14]

1. Report Industry Investment Rating - No relevant information provided. 2. Core Viewpoints - The company focuses on the "storage, computing, and connectivity" integrated technology layout, aiming to become a leading domestic memory chip design enterprise, and is committed to providing multi - product line memory chip products. The investment in Shanghai Lisuan is based on its development potential, which may increase short - term profit pressure but is beneficial for long - term strategy. The 7G100 series products of Shanghai Lisuan are self - designed, and the domestic GPU chip market has large development space. [27][34][35] 3. Summary by Directory 3.1. Research Basic Situation - The research object is Dongxin Co., Ltd., belonging to the semiconductor industry. The reception time was on September 4, 2025. The company's reception staff included the chairman, general manager, deputy general manager, board secretary, financial director, and independent director. [16] 3.2. Detailed Research Institutions - The institutions are investors who participated in the company's 2025 semi - annual performance briefing online. [19] 3.3. Research Institution Proportion - No relevant information provided. 3.4. Main Content Data - **Pre - solicited Questions**: Shanghai Lisuan's GPU products are mainly used in scenarios such as personal computers, professional design, AIPC, cloud gaming, cloud rendering, and digital twins, and do not involve chiplet technology for now. The company will consider business and cooperation according to strategic planning and disclose information as required. The management pays attention to Lisuan's development, and the 7G100 series graphics cards can be used in multiple scenarios. [23][25][25] - **Formal Questions**: - **Lisuan's Business Plan**: It is a GPU chip design enterprise focusing on graphic rendering, aiming to meet various graphic rendering and general computing needs. It adheres to self - developed architecture, products are for mainstream graphic rendering and AI acceleration at the edge, cloud, and terminal, and strives to solve the problem of independent and controllable domestic mainstream GPU architecture. [25] - **Information Disclosure**: The company will disclose major impact matters of Lisuan as required, and specific information should be based on official announcements. [26][29][36] - **Dongxin's Work Focus**: It will adhere to the strategy, provide diversified products for key applications and customers, explore new application fields, and promote technology innovation and product process iteration. It also conducts technology layout in the "storage, computing, and connectivity" integrated field. [27] - **Investment in Lisuan**: The investment is based on development potential. Although it may increase short - term profit pressure, it helps the long - term strategy. The impact on performance depends on Lisuan's operation. The company has a low asset - liability ratio and good financial stability. [34] - **Response to Physical AI**: The company will continue to pay attention to the development of cutting - edge technologies and conduct strategic layout according to its actual situation. [29] - **Lisuan's Product Development**: The planning and R & D design of the new product 7G200 are in progress. As of now, Lisuan has sent samples to some customers and is continuously optimizing, and will continue to promote production and sales. [29][31][37] - **Cooperation with Giants**: The company and Lisuan actively seek cooperation with mainstream manufacturers, and will disclose major cooperation matters that meet the disclosure standards. [32] - **Competition and Strategy**: The company will drive through technological innovation, ensure quality systems, collaborate in the supply chain, and support with talent strategies to enhance global market competitiveness. [35] - **Technical Threshold and Commercialization**: Shanghai Lisuan's 7G100 series products have independent GPU architecture, and the domestic GPU chip market has large development space. The profit cycle of large - scale chips is difficult to accurately estimate. [35][36]