证券研究报告|行业点评报告 [Table_Date] 2025 年 06 月 22 日 [Table_Title] Marvell 上调预期，拉动海外算力链持续上涨 [Table_Title2] 通信行业 [Table_Summary] 1、Marvell 上调数据中心 Capex 与 ASIC 预期指引，引发海外算 力链持续上涨 美东时间 6 月 18 日，Marvell 举办 AI 投资者日，预计 2028 年 数据中心资本开支将超过一万亿美元。同时进一步上调数据中 心市场规模 2028 年预期至 940 亿美元，较 2024 年 4 月的预期 750 亿美元上调 26%，其中上调定制计算芯片（XPU 和 XPU 配套 芯片）市场规模指引 37%。 2、投资建议 算力需求在过去两年受益于算力芯片催生一波跳跃式需求增 长，受数据源限制、密度功耗、Scaling Law 效率递减等多重 因素影响下开源与闭源模型效用逐渐达到同一等级，在 DeepSeek 影响下训练算力成本大幅下降，对于主权算力以及算 力资本开支下滑的担忧不止。 尽管如此，OpenAI 其营收规模与商用付费客户也持续呈现快速 增长，包括 Ora ...

摘要 风险提示 芯片制程发展与良率不及预期 中美科技领域政策恶化 智能手机销量不及预期 敬请参阅最后一页特别声明 1 - 聊天助手应用活跃度持续上升，海外 Claude 及国内通义、豆包增速显著。阿里巴巴开源专为苹果芯片优化的 32 款 Qwen3 MLX 量化模型，被视为其为国行 Apple Intelligence 所做的准备。在数据供应商领域，因 Meta 投资引 发数据中立性担忧，OpenAI 与谷歌相继计划停止与数据标注公司 Scale AI 的合作，对 Scale AI 的业务模式构 成重大挑战。 - Marvell 将数据中心 TAM 预期上调至 940 亿美元，其中定制计算市场达 550 亿美元，年增速显著。公司凭借 3/2nm 工艺、SerDes、HBM 等技术优势，结合端到端交付能力，定制芯片市占率已升至 13%，并锁定 18 个插槽、 追踪 750 亿美元潜在收入。尽管增长动能强劲，亚马逊 Trainium 项目仍待实质落地，客户集中度亦构成阶段性 风险。定制业务有望成为公司未来的关键增长引擎。 - 美光在 FY25Q2 实现 HBM 与数据中心高价值产品线的强劲增长，为 FY25Q3 ...

◼ AI ASIC 趋势已明朗，关注数据中心铜连接市场。 从目前已获知的 CSP ASIC 服务器方案来看，使用铜缆进行短距互 连已成趋势。根据 Fibermall 报告，AWS 今年有望采购 150 万颗自研芯 片，大部分使用AEC互连，尽管Trainium2的计算能力低于英伟达H100， 但可以使用 400G AEC，随着 Trainium3 于年底推出，800G AEC 的需求 有望得到增加；虽然微软在使用 AEC 铜缆方面较 AWS 缓慢，但目前已 经开始使用 AEC 构建 AI 网络；谷歌 TPU 互连目前主要使用无源铜缆 DAC，随着速度的提升，可能会转向 AEC。尽管目前还没有 Meta 使用 AEC 铜缆的明确信息，但在其 Minerva 机柜系统中，使用了 2 个可拆卸 电缆背板盒来连接计算托盘以及网络托盘。该电缆背板盒使用了 112G PAM4 连接器和电缆，每个盒子包括 4 个电缆组，每个电缆组包含 384 对不同电缆线，在 8 个 MTIA 托盘，6 个网络托盘和 1 个机柜管理系统 托盘间形成了 1536 对电缆网络。头部 CSP 厂商之外，X.AI 也对 AEC 表现出了大量需求， ...

行 业 研 究 通信 2025 年 06 月 22 日 投资评级：看好（维持） 行业走势图 数据来源：聚源 -14% 0% 14% 29% 43% 58% 2024-06 2024-10 2025-02 通信 沪深300 相关研究报告 《政策奖励游戏企业应用 AI，或提升 AIDC 算力需求—行业点评报告》- 2025.6.20 《字节多模态模型加速，Oracle 大投 AI，看好全球 AIDC 产业链—行业周 报》-2025.6.15 《字节 Force 原动力大会亮点：豆包 大模型 1.6 发布、多模态模型升级、 Tokens 加速增长、多行业落地—行业 点评报告》-2025.6.11 首批 IDC REITs 获批，Marvell 上调 2028 年 AIDC 规模预期，全球 AIDC 迎共振 ——行业周报 | 蒋颖（分析师） | 雷星宇（联系人） | 陈光毅（联系人） | | --- | --- | --- | | jiangying@kysec.cn | leixingyu@kysec.cn | chenguangyi@kysec.cn | | 证书编号：S0790523120003 | 证书编号： ...

Core Insights - Marvell Technology has experienced a significant stock price surge, driven by a strong market reaction to its AI strategy announcements and increased trading volume [1][2][6] Group 1: Company Developments - Marvell's recent event highlighted its AI strategy, confirming two new large-scale design wins for custom silicon chips, specifically Application-Specific Integrated Circuits (ASICs) [2][3] - The company now has a total of 13 custom programs in its pipeline, positioning itself as a core development partner for major technology firms [4] - Management indicated that these new contracts represent multi-billion-dollar revenue opportunities, with substantial revenue expected to begin in the 2026-2027 timeframe [5] Group 2: Market Reaction - On June 18, Marvell's trading volume surged to 55.28 million shares, significantly higher than the average daily volume of around 15 million shares, indicating strong institutional buying [7][8] - The stock opened at $74.53, up from a previous close of $69.99, reflecting a gap-up phenomenon typical of institutional investment activity [8][9] Group 3: Analyst Sentiment - Following the announcements, analysts raised their price targets for Marvell, with a consensus target of $96.33, indicating a potential upside of 31.05% [10][11] - Bank of America and B. Riley Financial both increased their price targets to $90.00 and $115.00 respectively, reflecting a more optimistic outlook on Marvell's growth potential [11][12] - The broad consensus among analysts reinforces a stronger growth narrative for Marvell, enhancing investor confidence [12][13]

Core Viewpoint - The article emphasizes the rapid development and potential of optical chips as a solution to the limitations of traditional electronic chips, particularly in the context of increasing AI computing demands and the need for efficient data transmission [4][5][6]. Group 1: Optical Chip Introduction - Optical chips are gaining traction due to their high-speed and low-power characteristics, which are essential for enhancing computing capabilities and data transmission efficiency in the AI era [4]. - The global AI server market is projected to grow from $12.2 billion in 2020 to $125.1 billion by 2024, indicating a nearly tenfold increase over five years [5]. Group 2: Market Dynamics and Investment Trends - Major tech companies like Microsoft and Amazon are significantly increasing their capital expenditures for AI-related projects, with Microsoft planning to invest $80 billion in AI chip procurement and data center expansion [6]. - The shift in computing power is structural, moving from training to inference, with companies like Broadcom and Marvell reporting substantial increases in semiconductor revenue from inference applications [7]. Group 3: Challenges and Advantages of Optical Computing - Traditional electronic chips face challenges such as signal delay and energy consumption, which hinder performance as they approach physical limits [7][8]. - Optical computing offers significant advantages, including ultra-low energy consumption (as low as 10^-18 joules per bit) and reduced latency, making it a promising solution for overcoming computing bottlenecks [8]. Group 4: Current State and Future of Optical Chips - The optical chip industry is in its early stages, focusing on technology development and performance optimization, with expectations for small-scale production to begin soon [10][11]. - The global optical chip market is expected to exceed $30 billion by 2027, with a compound annual growth rate of around 25% from 2022 to 2027 [17]. Group 5: Domestic Developments in Optical Chips - The domestic optical chip market is characterized by leading enterprises, emerging startups, and accelerated domestic substitution, with major players like Intel, IBM, and NVIDIA holding a significant share of patents [15]. - Recent breakthroughs in optical chip technology in China, such as the successful production of 6-inch thin-film lithium niobate optical chips, indicate a shift from "technology following" to "industry leading" [19]. Group 6: Policy and Ecosystem Support - The Chinese government is actively supporting the development of the integrated circuit industry, including optical chips, through various policies and funding initiatives [20]. - The establishment of a collaborative ecosystem around optical chips is gradually taking shape, enhancing the synergy between design, manufacturing, and application [20]. Group 7: Investment Activity in Optical Chips - The optical chip sector has seen increased investment activity, with major tech companies and venture capital firms actively investing in startups to enhance their technological capabilities [22]. - Significant funding events have occurred in 2025, indicating a long-term commitment from capital markets to the optical chip industry [23].

Key Takeaways MRVL posted record $1.44B data center revenues in Q1 FY26, up 76% YoY on custom AI silicon demand. NVIDIA NVLink Fusion support boosts MRVL's custom chip flexibility for next-gen AI infrastructure. MRVL launched UALink and multi-die packaging to enhance performance and scale in rack-level AI systems.Marvell Technology (MRVL) is enhancing its role in artificial intelligence (AI) infrastructure by expanding its custom chip capabilities. Marvell continues to integrate its custom compute platfor ...

Marvell Technology (MRVL) is currently in Phase 7 of its 18-phase Adhishthana Cycle on the monthly chart, and the structure is setting up in a way that could lead to a significant breakout in Phase 9. Here's how the stock is positioned through the lens of the Adhishthana Principles.Current Structure and AlignmentSo far, MRVL has shown an 83.33% alignment with the Adhishthana Principles, our proprietary cyclical framework that integrates quantitative signals with behavioral archetypes.Fig.1 Marvell’s Cakra F ...

如果您希望可以时常见面，欢迎标星收藏哦~ 来源：内容来自 Chosun BIZ 。 Marvell 正在快速崛起，成为应用特定集成电路（ASIC）市场的一支强劲力量。该公司近日宣布 计划与台积公司（TSMC）在3nm以下的先进制程上展开合作。据报道，Marvell 还计划采用台积 电的硅光子技术，这项技术正作为下一代光学半导体方案受到业界关注。当前，台积电在晶圆代工 市场的份额超过60%，几乎在从传统节点到先进制程的各个层面确立了主导地位，如今正瞄准新兴 的应用特定半导体制造市场——这一市场正随着人工智能（AI）的迅猛发展，成为晶圆代工行业 的新战场。 所 谓 应 用 特 定 半 导 体 ， 是 为 特 定 功 能 定 制 设 计 的 芯 片 。 相 较 于 为 通 用 目 的 设 计 的 图 形 处 理 器 （GPU），这类芯片具有更低的成本、更小的功耗和更少的整体投入，因此在AI数据中心、汽 车、物联网（IoT）等多个市场中颇具吸引力。摩根大通数据显示，博通（Broadcom）以55%至 60%的市场份额位居该领域首位，Marvell 以约15%的份额紧随其后。 据产业界6月20日消息，Marvell 即 ...

A tried and tested way of making money in the stock market is by buying great companies that are tapping into a growing trend that also trade at attractive valuations and holding them for the long run. This strategy allows investors to benefit from secular growth opportunities and disruptive trends, as well as take advantage of the power of compounding.Artificial intelligence (AI) is turning out to be one such secular and disruptive trend that has the ability to supercharge the growth of many companies in t ...