中国资产集体大涨。 今日晚间，美股开盘后，中国资产表现亮眼，纳斯达克中国金龙指数一度大涨超2%，两倍做多中国互联网股票ETF一度 大涨超5%。热门中概股全线走强，百度盘初一度暴涨超8%。今日亚洲交易时段，恒生科技指数大涨超4%，据高盛最新 数据，有迹象表明今日港股大涨是外资驱动的。 目前美股市场静待美联储利率决议，三大指数盘初窄幅震荡，纳指小幅下挫，美股芯片股多数走低。据悉，美国联邦公 开市场委员会（FOMC）政策会议已于北京时间17日21:00开始，将于18日02:00公布9月利率决议。华尔街一致预期，美 联储将在此次会议上降息25个基点。 中国资产大涨 北京时间9月17日晚间，美股开盘后，中国资产全线走强， 截至发稿 ，纳斯达克中国金龙指数大涨2.4%，两倍做多中国 互联网股票ETF大涨超5%，三倍做多富时中国ETF大涨超4%，中国科技ETF大涨超3%，中国海外互联网ETF涨超2%。 热门中概股普涨，百度大涨超7%，百度H股17日大涨超15%，多家机构大幅上调其目标价；金山云大涨超6%，万国数 据、世纪互联、老虎证券涨超5%，拼多多、腾讯音乐、贝壳涨超4%，蔚来、极氪、新东方涨超3%，阿里巴巴、京东、 理 ...

明日主题前瞻 1、国产芯片 | 央视《新闻联播》最新披露了阿里平头哥最新研发的面向人工智能的PPU芯片，其各项主要参数指标均超越了英伟达A800，与H20相当。与其 他国产AI芯片相比，平头哥PPU在相关指标方面也大多处于领先地位。 点评：分析认为，阿里平头哥半导体凭借 RISC-V 架构的先发优势、阿里系生态的协同效应及国家政策的强力支持，已在边缘计算、AI 推理等领域形成竞 争力，其玄铁系列处理器通过免费开源IP降低行业门槛，结合阿里云的生态优势，形成"芯片+云计算+应用"的全栈能力。当前市场对国产AI芯片需求迫切， 平头哥的技术路线与本土化战略契合政策导向，未来发展潜力巨大。机构报告预计到2027年，平头哥RISC-V芯片出货量将突破100亿颗，AI芯片全球市场份 额有望达到15%。 2、智能驾驶 | 9月17日，工业和信息化部就《智能网联汽车组合驾驶辅助系统安全要求》强制性国家标准公开征求意见。标准从"提升产品能力表现""强化 安全保障要求""规范系统使用方式"维度构建"三重安全保障"，多维度规范组合驾驶辅助系统安全能力，支撑道路交通安全水平提升。该标准填补了我国组 合驾驶辅助系统产品安全基线空白，将 ...

Investment Rating - The report does not explicitly state an investment rating for the industry or specific companies involved Core Insights - The Ascend ecosystem is shifting from a hardware-centric approach to a software-hardware co-design with full open-sourcing, enhancing its ecosystem and industry opportunities [1][8] - The focus in AI acceleration has transitioned from peak chip performance to end-to-end usability, emphasizing the importance of operator libraries and efficient communication-computation overlap [2][9] - The report highlights the need for high-quality documentation and support to build a robust open-source ecosystem that can compete with established frameworks like CUDA [3][11] Summary by Sections Event Highlights - At the AI Chip Summit, Huawei's Ascend chip manager presented a strategic shift towards a fully open-source and integrated software-hardware ecosystem, covering major AI frameworks and tools [1][8] - The collaboration with DeepSeek showcased advancements in operator fusion and system-level performance improvements for complex AI tasks [1][8] Commentary - The "software-first" strategy is essential for competing with CUDA, with a clear roadmap requiring substantial R&D investment and strong engineering governance [2][9] - Data movement is identified as a core bottleneck in AI computing efficiency, necessitating optimizations that can generalize across various model families [10] - A differentiated market strategy aims to lower entry barriers for developers while emphasizing the importance of comprehensive support and examples to enhance user experience [3][11] Industry Perspective - The industry is at a critical development juncture, with increased investment in AI infrastructure and a focus on model innovation [4][12] - Developers are encouraged to conduct proof-of-concept tests to evaluate the efficiency and stability of the Ascend ecosystem [4][12]

Key Points - Multiple listed companies disclosed their latest developments in low-altitude economy and chip sectors through various channels on September 17 [1] Low Altitude Economy - Yipuli has made certain technological achievements in the field of drone blasting [1] - Zhonghaidah has established some cooperation with DJI in the aviation measurement business [1] Chip Sector - Daoshi Technology's invested company, Chipenson, has achieved sales of its APU chips [1] - Xingchen Technology has shipped chips suitable for AI glasses [1] CPO (Chip-on-Board) - Huagong Technology plans to launch new products such as 1.6/3.2T CPO optical engines in 2025 [1] - Ruide Intelligent's smart controller products have not yet entered the CPO field [1] Battery Sector - Aoni Electronics has not yet achieved large-scale shipments of liquid-cooled servers in collaboration with AMD [1] - Aohai Technology's entire product line supports PD3.1/3.2 fast charging protocols and AVS protocol, compatible with the iPhone 17 series [1] - CATL's sodium-ion passenger vehicle power batteries are expected to be supplied in bulk next year [1] - Hailiang Co. has developed new nickel-plated copper foil products suitable for solid-state batteries and is capable of mass production [1]

Group 1 - The core viewpoint is that despite the challenges in the US and European markets, Chinese brands are demonstrating resilience and adaptability in their export strategies [1][5] - China's goods exports increased by 6.9% in the first eight months of the year, with August's export growth reaching 4.8%, driven by non-US exports and private enterprise exports [1][4] - The Trade Desk's insights indicate that Chinese companies are increasing investments in marketing and brand building, showcasing strong craftsmanship and technological innovation in AI and robotics [1][3] Group 2 - The automotive sector is enhancing the global recognition of Chinese chips, with companies like Naxin Micro seeing over 10% of their revenue coming from overseas by 2024 [3][5] - Chinese brands are diversifying their export markets, with significant growth in exports to the EU and ASEAN, indicating a shift away from reliance on the US market [4][5] - The trend of early preparation for the holiday shopping season reflects a more rational consumer mindset, with half of US consumers planning to complete their purchases before Black Friday [2][5] Group 3 - Chinese brands are transitioning from merely exporting products to establishing themselves as international brands, focusing on brand value and consumer trust [7][8] - The global expansion of Chinese manufacturing is evolving from low-end to high-end products, enhancing brand internationalization [8] - There is a notable difference in consumer perceptions between Chinese fast-moving consumer goods and international brands, highlighting the need for greater cultural understanding and marketing efforts [8]

2025.09. 17 本文字数：1877，阅读时长大约3分钟 作者 | 第一财 经 王媛丽 名下公司被溢价超过5倍收购,创始人直接、间接持有的超过60%股权，所获对价却接近于0。这样的事情，发生在晶晨股份（688099）最近的一桩收购 当中。 晶晨股份9月15日披露，将以3.16亿元的价格，收购芯迈微半导体（嘉兴）有限公司（下称"芯迈微"）100%股权。相较于芯迈微经评估的净资产，这一 收购价的溢价率，已经超过5倍。但该公司芯迈微创始人及实控人孙滇，持有的60%以上股权，收购价仅有112万元。而绝大多数收购款，都被创投机 构"笑纳"。 晶晨股份高溢价收购芯迈微，是买下"一个未来"，还是当"接盘侠"？ 创始人近乎"0元"退出 晶晨股份9月15日公告，拟使用自有资金3.16亿元收购芯迈微100%股权。 公开资料显示，芯迈微成立于2021年，目前已经完成五轮融资，此次收购以2024年9月A+轮融资的4.3亿元估值为基础，基于市场化交易原则折价至 3.16亿元，与股东实缴资本总额持平。 芯迈微目前的股东共13家，包括创始人孙滇、两家员工持股平台，以及多家投资机构。由于入股时间、价格都不同，此次收购采取差异化定价方案 ...

证券日报网讯星宸科技9月17日在互动平台回答投资者提问时表示，公司智能眼镜芯片已出货至终端客 户，也在与手机品牌、初创潮牌、ODM、方案商等多类客户持续接洽与合作中。 （文章来源：证券日报） ...

Group 1 - The company, Runxin Technology, has involvement in sensor chip products primarily used in smart terminals and the Internet of Things (IoT) sector [2] - The company indicates potential applications of its products in robotic components [2]

Core Insights - The rapid growth of computing power is essential for the digital economy, with Huawei predicting a 100,000-fold increase in total computing power by 2035, enabling real-time high-definition video calls from Earth to Mars [1][4] - The demand for computing power is experiencing non-linear growth, particularly in AI training and inference, with a potential 100-fold increase in computing needs when the token volume processed by AI increases tenfold [3][4] Computing Power Revolution - The exponential evolution of AI models, such as GPT-5 and Claude 3, requires significant computational resources, consuming 25,000 H100 GPUs for three months for a single training session [3] - The future of computing will involve a "computing revolution," driven by advancements in architecture, materials, and engineering processes, including photon computing and quantum bits [9] Connectivity and Data Processing - By 2035, the number of connected devices is expected to rise from 9 billion to 900 billion, necessitating advanced edge computing capabilities to handle massive data streams, such as 1TB of sensor data per second from a single L4 autonomous vehicle [4] - Traditional computing architectures face challenges, such as the "memory wall," where data transfer times consume over 70% of processing time, hindering performance improvements [7] Energy and Efficiency Challenges - Global data centers currently consume 3% of total electricity, with a single AI server's power consumption equivalent to that of 30 household air conditioners, highlighting the need for energy-efficient solutions [8] - The International Energy Agency (IEA) predicts that global data center electricity consumption will reach 945 TWh by 2030, necessitating innovations to reduce carbon intensity by 80% [21] Innovations in Healthcare and Industry - AI is projected to assist in preventing 80% of chronic diseases by 2035 through real-time monitoring and predictive modeling, enabling early warnings for conditions like cancer and diabetes [11] - In industrial manufacturing, predictive maintenance driven by computing power can reduce equipment downtime by 70% and improve product quality significantly [14] National Strategies and Technological Breakthroughs - The "East Data West Computing" initiative aims to deploy 1 million standard racks across eight hubs, optimizing the match between high real-time computing demands in the east and green energy resources in the west [16] - Huawei's Ascend 910B chip, with a computing density of 320 TFLOPS, represents a significant advancement in AI chip technology, doubling performance compared to previous generations [17]

证券日报网讯四维图新9月17日在互动平台回答投资者提问时表示，各家公司有各自不同的产品矩阵和 产品线规划，各有不同的生态位。杰发科技后续会继续在集成智能座舱与L2+级辅助驾驶功能方面，在 动力底盘、域控制器、新能源三电系统等场景，推进中高阶车规级芯片的量产与国产化替代。 （文章来源：证券日报） ...