英伟达财报引爆AI硬件需求！CPO、PCB、液冷，哪些赛道最受益？

Core Insights - Nvidia reported Q4 FY2026 earnings with revenue of $68.1 billion and net income of $42.96 billion, both reaching historical highs and exceeding analyst expectations [1] - The data center business was particularly strong, generating $62.3 billion in revenue, accounting for over 91% of total revenue, with a year-on-year growth rate of 75% [1] - Nvidia's gross margin reached 75%, significantly higher than the industry average [1] - The Blackwell architecture's Grace Blackwell platform has become central to inference business, reducing the cost per token by an order of magnitude [1] - The next-generation Vera Rubin architecture has delivered initial samples to customers, with full production expected in the second half of 2026, promising a 10x improvement in performance-to-power ratio [1] - CEO Jensen Huang emphasized that the inflection point for AI has arrived, with computing power directly translating into revenue, indicating a vast long-term market opportunity in AI infrastructure [1] Industry Chain Analysis CPO Industry Chain - Zhongji Xuchuang specializes in high-speed optical modules and related products, participating in the R&D and mass production of CPO technology-related high-speed optical modules to meet AI computing cluster interconnection needs [2] - Tianfu Communication focuses on manufacturing precision optical components required for CPO, supporting the application of co-packaged optical technology [3] - New Yisheng is engaged in the R&D and production of optical modules and devices, developing high-speed optical module products related to CPO to meet high bandwidth transmission needs for AI computing networks [3] AI Server Industry Chain - Inspur Information specializes in servers and computing infrastructure products, offering customized AI servers compatible with Nvidia GPUs to support AI training and inference deployments [4] - Unisoc provides AI servers and computing infrastructure, integrating software and hardware resources to build computing support platforms for AI applications [5] - Foxconn Industrial Internet manufactures AI server systems and components, providing AI computing hardware manufacturing services to global customers, compatible with high-performance computing chip assembly needs [5] PCB Industry Chain - Shennan Circuit focuses on printed circuit boards, developing and producing high-value PCB products such as high-layer and high-density interconnect (HDI) boards required for AI servers [6] - Huitian Technology specializes in high-end PCB R&D and manufacturing, supplying high-speed PCBs for AI servers to ensure stable signal transmission for computing hardware [7] - Shenghong Technology produces high-density and high-performance PCB products, covering AI servers and computing chip substrates, adapting to performance enhancement needs during AI hardware upgrades [7] Liquid Cooling Industry Chain - Gaolan Co. specializes in liquid cooling solutions, launching immersion and cold plate liquid cooling products suitable for AI computing infrastructure to enhance operational efficiency [9] - Jialitu provides liquid cooling systems for data centers, addressing cooling needs for high-density AI server deployments while reducing energy consumption levels [9] - Yingweike develops precision temperature control equipment, creating liquid cooling solutions specifically for AI computing scenarios to support stable operation of high-power AI hardware [9] Domestic Computing Chip Industry Chain - Cambricon focuses on AI chip R&D and design, offering multi-scenario AI chip products for cloud and edge computing, providing domestic AI computing hardware solutions [10] - Haiguang Information specializes in domestic x86 server chips and AI chip R&D, supporting AI computing scenarios and promoting the localization of computing power [10] - Moore Threads develops GPUs and AI computing chips, adapting to AI training and inference applications, enriching the domestic computing hardware product matrix [11]