证券研究报告 2025 年 7 月 3 日 星期四 【A 股市场大势研判】 市场窄幅震荡，沪指微跌 市场表现： | 指数名称 | 收盘点位 | 涨跌幅 | 涨跌 | 上证指数分时图 | | --- | --- | --- | --- | --- | | 上证指数 | 3454.79 | -0.09% | -2.96 | | | 深证成指 | 10412.63 | -0.61% | -63.67 | | | 沪深 300 | 3943.68 | 0.02% | 0.92 | | | 创业板 | 2123.72 | -1.13% | -24.20 | | | 科创 50 | 982.64 | -1.22% | -12.16 | | | 北证 50 | 1439.15 | -1.23% | -17.92 | | 资料来源：东莞证券研究所，iFinD 数据 板块排名： | 申万行业表现前五 | | | 申万行业表现后五 | 概 | 念板块表现前五 | 概念板块表现后五 | | | --- | --- | --- | --- | --- | --- | --- | --- | | 钢铁 | 3.37% | 电子 | ...

Core Viewpoint - The "Silicon Photonics Revolution" led by major chip companies like Nvidia, TSMC, and Broadcom is expected to transform the AI computing industry, accelerating the adoption of Co-Packaged Optics (CPO) and Optical I/O technologies from experimental labs to global applications [1][5][19]. Industry Overview - The global optical communication data center interconnect market, including CPO and Optical I/O technologies, is projected to grow at a compound annual growth rate (CAGR) of over 80% from 2028 to 2033, reaching nearly $2.5 billion by 2033 [5][19]. - The demand for Optical I/O modules in large data centers is expected to be 2-7 times that of CPO modules by 2033, indicating a significant market expansion for Optical I/O [5][11]. Company Insights - Nvidia is integrating silicon photonics technology into its high-performance networking products, which is anticipated to enhance the scalability and efficiency of its AI GPU clusters, solidifying its leadership in the AI infrastructure market [20][21]. - TSMC is expected to lead in silicon photonics chip manufacturing, leveraging its advanced processes and packaging capabilities to attract major clients like Nvidia and AMD, thereby strengthening its dominant position in the high-performance computing and optical communication chip manufacturing sectors [21][24]. - Broadcom is developing its own CPO high-performance switching chip solutions and is expected to integrate silicon photonics interfaces into its next-generation switching platforms, which will significantly enhance its competitiveness in the cloud data center market [24][27].

Core Viewpoint - The article discusses the emergence of supernodes in high-performance computing, emphasizing their role in enhancing the efficiency of large-scale model training and inference through optical technology [1][2][21]. Group 1: Supernode Architecture and Performance - Supernodes provide a new solution for large-scale model training and inference, significantly improving efficiency by optimizing resource allocation and data transmission [1]. - The architecture of supernodes can be categorized into single-layer and two-layer designs, with single-layer architecture being the ultimate goal due to its lower latency and higher reliability [4][6]. - The demand for GPU power has surged with the exponential growth of model sizes, necessitating thousands of GPUs to work in tandem, which supernodes can facilitate [1][2]. Group 2: Challenges in Domestic Ecosystem - Domestic GPUs face significant performance gaps compared to international counterparts, requiring hundreds of domestic GPUs to match the power of a few high-end international GPUs [6][8]. - The implementation of supernodes in the domestic market is hindered by limitations in manufacturing processes, such as the 7nm technology [6]. Group 3: Development Paths for Supernodes - Two main development paths are proposed: increasing the power capacity of individual cabinets to accommodate more GPUs or increasing the number of cabinets while ensuring efficient interconnection [8][10]. - Optical interconnect technology is crucial for multi-cabinet scenarios, offering significant advantages over traditional copper cables in terms of transmission distance and flexibility [10][12]. Group 4: Optical Technology Advancements - The transition to higher integration optical products, such as Co-Packaged Optics (CPO), enhances system performance by reducing complexity and improving reliability [14][16]. - CPO technology can save 1/3 to 2/3 of power consumption, which is significant even though communication power is a smaller fraction of total GPU power [16][17]. Group 5: Reliability and Flexibility - The use of distributed optical switching technology enhances the flexibility and reliability of supernodes, allowing for dynamic topology adjustments in case of node failures [18][19]. - Optical interconnect technology simplifies the supply chain, making it more controllable compared to advanced process-dependent components [19][21]. Group 6: Future Outlook - With advancements in domestic GPU performance and the maturation of optical interconnect technology, the supernode ecosystem is expected to achieve significant breakthroughs, supporting the rapid development of artificial intelligence [21].

据新华社5月30日消息，工业和信息化部日前印发《算力互联互通行动计划》，根据行动计划，到2026 年，我国将建立较为完备的算力互联互通标准、标识和规则体系；推广新型高性能传输协议，提升算力 节点间网络互联互通水平；建成国家、区域、行业算力互联互通平台；推动算、存、网多种业务互通， 实现跨主体、跨架构、跨地域算力供需调度；开展算力互联网试验网试点，赋能产业普惠用算。到2028 年，基本实现全国公共算力标准化互联，逐步形成具备智能感知、实时发现、随需获取的算力互联网。 根据IDC发布的《中国人工智能计算力发展评估报告》预测，2025年中国智能算力规模将达到1037.3 EFLOPS，同比增长43%；到2026年，这一数字将增至1460.3 EFLOPS，较2024年翻番。 长江证券指出，智能算力的爆发式扩张加速推动算间与算内网络的技术重构：智算体系由单点部署向多 中心协同架构演进，智算中心内部算力集群规模跃迁，对组网规模、吞吐性能与传输质量提出更高要 求。 新华财经北京6月5日电（王媛媛）算力产业链6月5日集体走强。共封装光学（CPO）、元件、电力物联 网、MLOps（机器学习运维）概念、AI PC等方向涨幅居前 ...

当地时间6月3日，美股的AI总龙头英伟达（NVDA.US）大涨2.8%，报141.22美元/股，公司市值升至约 3.45万亿美元，超越微软重夺全球市值第一头衔。 值得一提的是，4月上旬至今，英伟达累计反弹幅度接近50%，可谓"大象起舞"。 作为"卖铲人"，英伟达在AI淘金时代赚得盆满钵满，旗下GPU产品颇受各大科技巨头的追捧，包括亚马 逊（AMZN.US）、谷歌（GOOGL.US）、微软、OpenAI、甲骨文（ORCL.US）、特斯拉（TSLA.US） 均是英伟达的客户，上周英伟达公布的财报显示，第一财季营收同比增长69%，达440.62亿美元，远超 市场预期。 此前，曾有投资者担忧AI的"泡沫"还能持续多久，英伟达在GPU领域的护城河如何，能否维持其估值溢 价的合理性，以及贸易的冲击是否会产生负面影响等等，资本市场众说纷纭。 此外，近期Yole发布2025年数据中心共封装光学市场预测报告。报告指出，英伟达的硅光子共封装光学 （CPO）技术正推动AI数据中心变革，预计CPO市场规模将从2024年的4600万美元飙升至2030年的81亿 美元，年复合增长率高达137%。 除了CPO之外，国内其他的算力概念股也集 ...

一、【早盘盘面回顾】 智通财经6月4日讯，市场早盘延续反弹，创业板指领涨。沪深两市半日成交额7270亿，较上个交易日缩量234亿。盘面上热点快速轮动，个股涨多跌少，全 市场超3900只个股上涨。从板块来看，大消费股反复活跃，食品饮料等方向领涨，乐惠国际等多股涨停。固态电池概念股一度冲高，龙蟠科技等涨停。算力 概念股展开反弹，太辰光涨超10%。截至收盘，沪指涨0.43%，深成指涨0.91%，创业板指涨1.22%。 个股来看，今日早盘涨停数量为49家（不包括ST及未开板新股），炸板率为34%，连板股数量为15家。御银股份5连板，海南海药、中衡设计、永悦科技、 共创草坪3连板，九芝堂、萃华珠宝、金时科技、金陵体育、广晟有色、乐惠国际、南华期货、龙蟠科技、合兴股份、山东玻纤2连板。 板块上，CPO等算力硬件股全线走强，太辰光、斯瑞新材、德科立涨超10%，联特科技、生益电子、新易盛、景旺电子、中际旭创等涨幅居前。消息面上， 美东时间6月3日，英伟达股价上涨近3%，达141.22美元，超越微软再次成为全球市值最高的上市公司。 此外，近期Yole发布2025年数据中心共封装光学（Co-Packaged Optics for ...

2. Q2 指引： a. 营收：预计 20 亿美元左右，±5%。 b. 利润：GAAP 毛利率 50%-51%，Non-GAAP 毛利率 59%-60%；GAAP 摊薄每股收益 0.16-0.26 美元，Non-GAAP 摊薄每股收益 0.62-0.72 美元。 c. 费用：GAAP 运营费用约 7.35 亿美元，Non-GAAP 运营费用约 4.95 亿美元；其他收入和费用约 4900 万美元，非 GAAP 税率 10% 。 迈威尔科技 Marvell(MRVL.O) 北京时间 5 月 30 日凌晨，美股盘后发布 2026 财年第一季度财报（截至 2025 年 4 月）： 以下为 Marvell 2026 财年第一季度的财报电话会纪要，财报解读请移步《Marvell：AI 环增 "失速"，下一张 "王牌" 在哪？》 一、 $迈威尔科技.US 财报核心信息回顾 | Marvell (MRVL.O) 1QFY26 Financial Performance | | | | | | | | | | --- | --- | --- | --- | --- | --- | --- | --- | --- | | (i ...

作者 |赵雨荷 编辑 | 硬 AI 当地时间3月18日周二，英伟达CEO黄仁勋在加州圣何塞举行的英伟达AI盛会GTC 2025上发表主题演 讲。美国知名半导体咨询机构Semianalysis深度解读黄仁勋GTC演讲，详细阐述英伟达在推动AI推理性能 提升方面的最新进展。 市场担心的是，DeepSeek式的软件优化以及英伟达主导的硬件进步带来的巨大成本节省，可能导致对AI 硬件的需求下降。然而，价格会影响需求，当AI成本降低时，AI能力的边界不断被突破，而需求随之增 加。 随着英伟达在硬件和软件方面的推理效率提升，使得模型推理和智能代理的部署成本大幅降低，从而实现 成本效益的扩散效应，实际的消费量反而会增加，正如英伟达的口号所说的那样："买越多、省越多"。 点击 上方 硬AI 关注我们 Semianalysis表示，在GTC2025大会上，英伟达推出的推理Token扩展、推理堆栈与Dynamo技术、共封装光学 （CPO）技术等创新将显著降低AI总拥有成本，使得高效推理系统的部署成本大幅下降，并巩固了英伟达在全球AI生态 系统中的领先地位。 硬·AI 以下为文章的核心观点： 推理Token扩展：预训练、后训练与推 ...