英伟达在过去三年里一直是人工智能算力基础设施领域以及席卷全球的这股史无前例AI投资浪潮中的 最大亮点，但纵观整个2025年的美国股票市场，这五只聚焦于为全球范围正在如火如荼建设的超大规模 AI数据中心提供最核心算力系统的AI算力类科技股票的涨幅更为惊人——即Lumentum(LITE.US)、西部 数据(WDC.US)、美光科技(MU.US)、希捷(STX.US)以及天弘科技(CLS.US)。 除了英伟达与博通这两大AI芯片领军者所主导的AI GPU以及AI ASIC算力集群，完全投入运营的超大规 模AI数据中心(比如"星际之门"项目)还迫切需要存储芯片/组件、光纤电缆、高性能以太网交换机、光互 连技术设备、定制化的电源/高功率类芯片以及数据中心级别的高性能中央处理器等最核心的AI算力组 件。 "谷歌AI链"的最核心参与者，同时也是"英伟达AI算力链"不可或缺光模块组件供应商的Lumentum堪称 2025年全球AI算力产业链中最大赢家，该公司股价今年迄今股价涨幅接近400%，Celestica、西部数 据、希捷以及美光科技的股价在2025年迄今的涨幅则全都超过200%。 英伟达(NVDA.US)，近年来可谓 ...

自ChatGPT在2022年底爆发以来，光模块作为算力基础设施中负责数据高速传输的核心器件，也悄然催 生出"易中天"这样的光模块领头羊，它们分别取自新易盛（300502.SZ）的"易"、中际旭创 （300308.SZ）的"中"，以及天孚通信（300394.SZ）的"天"。 当算力焦虑从"地球烧到了太空"，市场早已开始寻找另一种缓解AI算力瓶颈的方式。 近日，片间光互连（OIO）创业公司光联芯科接连获得多家机构投资，成为国内光互连芯片赛道规模最 大的早期融资之一。光互连技术被业内视为突破当下算力瓶颈的关键路径，通过将芯片间短距互连从电 转向光，在传输能耗、带宽密度、延迟等核心指标上实现数量级提升。 中国算力底层技术架构正在发生深刻变革的同时，海外公司也在加快融资速度。据公开资料显示，美国 光互连公司Ayar Labs和光芯片创业公司Lightmatter也在去年年底完成了新一轮融资，其中不乏英伟达、 AMD、英特尔这样的投资者。 "在AI的发展中，买更多的GPU造更大的集群、追求更大的模型是一条路径，但边际效应是递减的。"光 联芯科CEO陈超在一次行业论坛中表示，英伟达从五年前就开始做光互连，头部企业很早就看到 ...

Core Insights - The market is actively seeking solutions to alleviate the AI computing power bottleneck, with optical interconnect technology being viewed as a key path to breakthrough [1] - Domestic companies like Guanglian Chip and international firms such as Ayar Labs and Lightmatter are rapidly securing funding to advance optical interconnect technology [1][5] - The shift from electrical to optical interconnects is driven by the need for higher efficiency in data transmission, as traditional copper interconnects face significant limitations [3][4] Industry Developments - The optical module market has seen significant growth, with leading companies emerging in the sector, such as New Yisheng, Zhongji Xuchuang, and Tianfu Communication [2] - A report from China Mobile Research Institute highlights that over 90% of energy consumption in large model computing clusters is used for data movement between chips, emphasizing the need for more efficient interconnect solutions [3] - The transition to optical interconnects is expected to enable data centers to achieve better performance and lower overall costs compared to traditional methods [4] Competitive Landscape - Companies like Ayar Labs have introduced innovative optical interconnect solutions, enabling efficient communication over long distances, supported by major chip manufacturers [5] - Intel and NVIDIA are also developing their optical interconnect capabilities, with Intel showcasing a fully integrated optical computing interconnect chip [5][6] - Huawei's "Lingqu Interconnect" protocol is pushing the boundaries of optical interconnect technology, with new products designed to exceed existing performance requirements [6] Future Outlook - The CEO of Guanglian Chip envisions a collaborative ecosystem akin to "Android" for optical interconnects, contrasting with the closed ecosystems of companies like Apple [7] - Despite a 2-3 year technological gap, domestic companies are expected to leverage faster iteration speeds and cost advantages to potentially surpass international competitors in the optical interconnect space [7]