行业研究 半导体行业周报 "大而美"法案落地，半导体在美建厂税惠力度升级 证券研究报告 2025 年 7 月 6 日 湘财证券研究所 相关研究： | 核心要点： | | --- | "大而美"法案规定，在美国本土新建工厂的芯片制造商，其享有的税 收抵免比例将从 25%提升至 35%。符合条件的新建项目需在 2026 年底前 动工。本法案将大幅降低厂商扩产成本，或激励台积电、Intel、美光等 加快美国落地项目。法案强调继续提高进口关税，包括对半导体制造设 备和高端芯片进口。 ❑ 费城半导体指数上涨态势延续 高资本开支改善市场对长期科技需求的预期，为科技板块提供支撑。此 外，联储官员近期表态明显松动，推动费指延续上涨态势。根据wind数 据显示，截至 2025 年 7 月 3 日费城半导体指数收于 5647.12 点，年初至 今涨幅为 12.46%，周涨幅为 1.8%。 ❑ 投资建议 AI 大模型的破圈及 AI 智能眼镜、AI 手机等设备市占率的提升，带动端 侧 AI 算力的需求上行，驱动高性能以太网交换机、先进存储产品、 GPU 及边缘计算/端侧算力芯片等多种半导体硬件的市场需求稳步增 长。传统消费电子领域， ...

Core Viewpoint - In 2024, as AI data centers thrive, cloud giants are heavily investing in AI chips, with Nvidia emerging as the biggest winner in this landscape [1][2]. Group 1: Arista Networks' Emergence - Arista Networks, founded in 2004 by three Silicon Valley figures, aims to revolutionize network systems with a focus on simplicity and cloud readiness [5][6]. - The company transitioned from Arastra to Arista Networks under the leadership of Jayshree Ullal in 2008, emphasizing software-defined networking (SDN) integrated into hardware design [12][15]. - Arista's innovative approach, including the use of "white box" switches and the EOS operating system, has positioned it favorably in the data center market, particularly among cloud service providers [13][15]. Group 2: Growth and Challenges - Arista's revenue surged from $584 million in 2014 to $2.32 billion in 2020, reflecting a compound annual growth rate of 25% [16]. - The company faced legal challenges from Cisco, which initiated multiple patent lawsuits against Arista, but managed to prevail in most cases [18]. - Despite competition from traditional players and new entrants, Arista has maintained its focus on innovation and customer service, expanding its product line to include network analytics and security solutions [18]. Group 3: AI-Driven Opportunities - The rise of AI applications has created a significant demand for AI data centers, leading to a second wave of growth for Arista [19][20]. - Arista has launched products optimized for AI workloads, including high-speed switches and network architecture solutions designed for large AI clusters [20][21]. - The company has secured pilot contracts with major clients in the AI sector, indicating strong market demand and potential for future growth [21][23]. Group 4: Competitive Landscape - Arista leads the data center Ethernet switch market with $1.48 billion in sales, closely followed by Nvidia at $1.46 billion [26][28]. - Nvidia's rapid ascent in the market is attributed to its Spectrum-X Ethernet solution, which is gaining traction among AI customers [26][28]. - The competition between Arista and Nvidia is intensifying, with both companies vying for dominance in the evolving AI infrastructure landscape [29]. Group 5: Future Outlook - As AI technology continues to evolve, the demand for high-performance network infrastructure is expected to grow, positioning Arista as a key player in this space [31]. - The outcome of the competition between Arista and Nvidia will significantly influence the future of AI infrastructure [29][31].

证券研究报告：电子 | 公司点评报告 股票投资评级 买入 |维持 公司基本情况 | 最新收盘价（元） | 92.30 | | --- | --- | | 总股本/流通股本（亿股）0.80 | / 0.50 | | 总市值/流通市值（亿元）74 | / 46 | | 52 周内最高/最低价 | 120.75 / 49.90 | | 资产负债率(%) | 7.8% | | 市盈率 | -36.48 | | 第一大股东 | 史清 | 研究所 分析师:万玮 SAC 登记编号:S1340525030001 Email:wanwei@cnpsec.com 分析师:吴文吉 SAC 登记编号:S1340523050004 Email:wuwenji@cnpsec.com 裕太微(688515) 七大产品线形成框架，研发投入步入收获期 l 投资要点 市场持续复苏，下游客户需求有所增长。随着半导体行业周期性 下行收尾，市场需求逐步复苏及客户库存逐步优化，下游客户需求有 所增长；同时公司 2.5G 网通以太网物理层芯片、多口网通以太网交 换机芯片、千兆网通以太网网卡芯片、车载芯片等持续放量以及其他 高速有线通信新品的逐年推出，公司 ...

推荐北美及国产算力板块 20250624 摘要 AI 算力基础设施发展应关注 TOKEN 增长，若 TOKEN 增长强劲，表明 前置投资消耗良好，需维持甚至加速投资。市场对 AI 算力板块景气度持 续性的担忧可能是不必要的，核心在于 TOKEN 增长而非单季度业绩提 速。 AI 大模型通过提升听说读写看能力，赋能新终端硬件，长期发展乐观。 其潜移默化提升全社会效率和智能水平，全面提升生产效率，可能颠覆 现有公司和业务模式，胜者通吃，并最终促进收入和利润增长。 投资方面，优先选择北美算力链，其次是国产算力链。北美链不受国际 环境影响，更具全球化优势，经营环境和竞争格局更好。国产链虽业绩 爆发力较弱，但在各自赛道仍有增长潜力，关注资本开支带来的业绩机 会。 光模块领域，关注 Meta 和谷歌新供应商带来的市场份额变化，以及业 绩弹性。持续推荐易中天等公司，重点关注边际变化明显且弹性较大的 公司，如联特科技、剑桥科技、东山精密以及源杰科技。 Q&A 对于北美和国产算力链的投资，您有何最新观点？ 我们对整个算力链的总体观点是乐观的。近期我们发布了通信行业的中期策略 报告，其中提到"精选算力"。我们认为 2023 年和 ...

在过去的美好时光里，网络只是连接少量本地计算机。但时代变了。在人工智能主导的世界里，关键 在于协调数万台服务器的活动，以训练大型语言模型，并且不造成任何通信延迟。现在，有一个经过 优化的架构可以实现这一点。Cornelis Networks表示，其 CN500 网络结构可最大限度地提高人工智 能性能，支持部署多达 50 万台计算机或处理器（比现在高出一个数量级），并且不会增加延迟。 继以太网和InfiniBand之后，这项新技术为网络领域带来了第三款重要产品。它旨在使人工智能和高 性能计算机（HPC，即超级计算机）能够以更高的效率实现更快、更可预测的完成时间。Cornelis 声 称，在 HPC 方面，其技术优于2022 年推出的InfiniBand NDR版本，每秒传递的消息数量是后者的 两倍，延迟降低了 35%。在人工智能应用方面，与基于以太网的协议相比，其通信速度提高了六 倍。 以太网长期以来一直是局域网（LAN）的代名词。软件补丁使其通信协议经受住了时间的考验。 InfiniBand 的发明是一项改进，但它的设计目标仍然是相同的：连接少量本地设备。"这些技术发明 之初，与并行计算毫无关联，"宾夕法尼亚 ...

Core Viewpoint - The transition of data centers to Co-Packaged Optics (CPO) switches is inevitable, primarily driven by the power savings offered by CPO technology [1][2]. Group 1: CPO Technology and Market Trends - CPO technology is gaining traction as it significantly reduces power consumption, with potential savings of 30-50% compared to traditional optical devices [1][2]. - The industry has made substantial progress in CPO reliability over the past two years, making it a viable option for future high-speed data rates [2]. - The upcoming 400G SerDes generation may see CPO as the only feasible choice due to excessive insertion loss from traditional PCB traces and cables [2]. Group 2: Technical Integration of CPO - CPO solutions typically integrate electronic integrated circuits (EIC) and photonic integrated circuits (PIC) within the same package [3]. - Two main integration methods for optical engines within ASIC packages are the silicon interposer approach and the organic substrate approach [4][5]. - The silicon interposer method allows for high-density connections but complicates thermal management due to the proximity of high-power EICs [6]. - The organic substrate method offers better thermal isolation and modularity, allowing for independent testing of optical engines before assembly [7][8]. Group 3: Bandwidth Density and Performance - Bandwidth density, a critical metric for CPO solutions, measures the amount of data transmitted per millimeter along optical interfaces, typically expressed in Tbps/mm [9]. - Higher bandwidth density is essential to meet the explosive growth in bandwidth demand in data centers and high-performance computing systems [9]. Group 4: Competitive Landscape - Broadcom's Bailly CPO switch integrates eight 6.4 Tbps optical engines, achieving a total external bandwidth of 51.2 Tbps [12]. - NVIDIA's Quantum-X InfiniBand switch aims for higher scalability, targeting over 100 Tbps with advanced optical engine integration [17][18]. - Broadcom's next-generation switches are expected to reach 102.4 Tbps, while NVIDIA's architecture is designed for future demands of 200G SerDes and beyond [16][19]. Group 5: Power Efficiency and Thermal Management - Both Broadcom and NVIDIA report significant reductions in power consumption per bit with their CPO solutions, with Broadcom achieving approximately 5.5W per 800 Gb/s port compared to 15W for traditional modules [35]. - Effective cooling solutions, such as liquid cooling, are necessary to manage the heat generated by high-density ASIC packages [35][36]. Group 6: Future Directions and Challenges - The industry is exploring advanced coupling methods, such as vertical coupling and multi-core fibers, to enhance optical connectivity and bandwidth density [38][40]. - Challenges in deploying CPO include ecosystem disruption, operational complexity, and the need for robust reliability validation [42][43]. - CPO's prospects appear brighter in vertical scaling applications, where integrated solutions from single vendors can simplify procurement and deployment [45].

Core Viewpoint - Unisoc Co., Ltd. is preparing for a secondary listing on the Hong Kong Stock Exchange, driven by the increasing demand for computing power from its DeepSeek large model, positioning itself as a leading provider of digital and AI solutions in the ICT sector [1][2]. Company Overview - Unisoc is a subsidiary of Tsinghua Unigroup, originally listed on the Shenzhen Stock Exchange in November 1999, and is part of a larger group that includes multiple listed companies in both A-shares and H-shares [1]. - The company ranks third in China's digital infrastructure market and second in both the networking and computing/storage infrastructure markets, according to Frost & Sullivan [1]. Business Model and Revenue - Unisoc offers a comprehensive range of digital solutions, integrating cloud computing, big data, AI, IoT, cybersecurity, and edge computing, which supports various industries in their digital transformation [2]. - The company has four major subsidiaries, with H3C contributing the most to its revenue, recognized as a leading manufacturer of AI servers and Ethernet switches [2]. Financial Performance - Unisoc's revenue has shown steady growth, with reported revenues of approximately 737.52 billion RMB, 775.38 billion RMB, and 790.24 billion RMB for the years 2022, 2023, and 2024, respectively [3]. - The digital solutions segment has become the main revenue driver, accounting for 62.7%, 68.4%, and 70.5% of total revenue in the same years [3][4]. Profitability - The company's net profits from continuing operations were approximately 37.42 billion RMB, 36.85 billion RMB, and 19.82 billion RMB for 2022, 2023, and 2024, respectively, with a declining gross margin from 19.8% to 16.0% over the same period [5]. - Despite a drop in profit for 2024 due to increased costs and reduced margins, the company maintains a strong cash position with 73.17 billion RMB in cash and cash equivalents by the end of 2024 [5]. Market Trends - The global digital solutions market has been growing steadily, projected to increase from 1.5 trillion USD in 2020 to 2.6 trillion USD by 2024, with a compound annual growth rate (CAGR) of 14.1% [7]. - The market is expected to reach 4.8 trillion USD by 2029, with a CAGR of 12.7% from 2024 to 2029, driven by advancements in cloud computing, AI, and other technologies [7]. Competitive Landscape - The company faces increasing competition from major players like Huawei and ZTE, particularly in the telecommunications sector, and must navigate challenges from rising self-developed hardware by cloud service providers [11]. - Unisoc's IPO proceeds are intended for R&D, strategic investments, and global market expansion to strengthen its competitive position [11]. Future Outlook - The successful listing on the Hong Kong Stock Exchange is anticipated to elevate the company's market presence, with potential for strong long-term value driven by technological barriers and favorable policies [12].

如果您希望可以时常见面，欢迎标星收藏哦~ 来源：内容 编译自 nextplatform 。 尽管随着以太网路线图上的每一次减速，更扁平的网络和更快的网络都是可能的，但网络规模仍然 保持着足够快的增长速度，以至于交换机 ASIC 制造商和交换机制造商能够通过产量来弥补这一 不足，并保持交换机业务的增长。 随着 GenAI 的爆发式增长，所有大型 AI 厂商都一致希望摆脱英伟达控制的专有 InfiniBand 技 术，将 InfiniBand 的所有功能移植到全新升级的以太网上，使其能够进一步扩展，并在更扁平的 网络中实现扩展，从而创建规模更大的 AI 集群。超级以太网联盟 (UltraEthernet Consortium) 的 宏伟目标是实现 100 万个 GPU 端点，而要实现这一目标，需要容量更大的交换机 ASIC。 如今，商用芯片市场的行业领导者博通 (Broadcom) 在以太网领域面临着来自思科系统和 Nvidia 的激烈竞争，该公司正在向市场推出其"Tomahawk 6" StrataXGS 以太网交换机 ASIC，该市场将 以 102.4 Tb/秒 ASIC 为主导，并展望 204.8 Tb/秒 ...

如果您希望可以时常见面，欢迎标星收藏哦~ 来源：内容来自source：编译自theregister 。 AMD 本周宣布收购光子芯片初创公司Enosemi，正式加入共封装光学竞赛。 House of Zen 旨在将该技术融入其下一代机架式系统，以便在人工智能领域更好地与竞争对手 Nvidia 竞争。 与铜互连或走线相比，共封装光学器件具有许多优势，包括更高的带宽、更低的延迟和更低的功 耗。 顾名思义，这些改进通常是通过将光子芯片或中介层与计算芯片一起封装来实现的，通过光纤而不 是铜线传输信号。 在人工智能蓬勃发展的背景下，人们对这项技术的兴趣激增，因为芯片设计师和系统制造商一直在 努力解决传统铜缆的有限覆盖范围和带宽以及高性能可插拔光学器件不断增长的功率需求。 AMD 在共封装光学器件领域稍晚了一步。英特尔和博通多年来一直在探索这项技术，而在今年春 季的 GTC 大会上，Nvidia发布了两款将在今年晚些时候开始采用该技术的网络交换机。 照亮未来之路 AMD 可能计划在未来的机架级设计中使用 Enosemi 的 IP。然而，我们尚不清楚该光子技术将如 何以及在何处集成。 但AMD的高管此前曾讨论过将光子芯片集 ...

其实早在2024年，我国在用算力中心机架总规模超过830万标准机架，算力总规模达246EFLOPS，位居世 界前列。 明确告诉各位大家！ 面对这一历史性机遇，全球资本正疯狂涌入算力赛道，小编也梳理出5家龙头企业凭借技术壁垒和业绩暴 增，成为机构抢筹的核心标的！ 无论市场如何波动，算力都始终是最核心的主线之一！ 作为AI时代最坚固的的基石和源动力，算力需求已呈爆发式增长。 据权威机构预测，2025年中国智能算力规模将突破1037.3EFLOPS，同比增长43%，2026年更将飙升至 1460.3EFLOPS，较2024年翻倍！ 第三家：海兰信 公司不仅在成本控制、安全性能上优势显著，还得益于率先布局和政策支持，再加上传统业务的协同效 应，构建起独特的市场竞争力。 第四家：浙文互联 国内领先的数字营销企业，与阿里、华为、百度、腾讯等巨头深度合作。公司成立浙文智算、浙文蓝耘 智算等专业算力公司，并参与浙江国资项目投资，构建算力产业链布局。 最后一家：国资委控股，算力赛道最强黑马！ 作为AI服务器核心网络设备商，其以太网交换机国内市占率第三，Wi-Fi 6设备全球销量五连冠。更值得 关注的是，想知晓莱肱纵浩牛板，一季 ...