开源证券：重视硅光与CPO链投资机会

Core Viewpoint - NVIDIA's release of the Vera Rubin POD AI supercomputer and Spectrum-X Ethernet co-packaged optical devices is expected to accelerate demand for CPO, optical modules, and liquid cooling, indicating a clear trend in the silicon photonics industry [1][2]. Group 1: Product Launch and Specifications - NVIDIA CEO Jensen Huang announced the Vera Rubin POD AI supercomputer, featuring the NVL72 cabinet with 20 trillion transistors, achieving 3.6 EFLOPS for inference and 2.5 EFLOPS for training [1]. - The NVL72 cabinet has a memory capacity of 54 TB for LPDDR5X, which is 2.5 times that of the previous generation, and a total HBM4 memory of 20.7 TB, which is 1.5 times the previous generation [1]. - The HBM4 bandwidth reaches 1.6 PB/s, which is 2.8 times that of the previous generation, with a total vertical expansion bandwidth of 260 TB/s [1]. Group 2: Design and Efficiency - The Rubin NVL72 is based on the third-generation MGX rack design, featuring a modular, hostless, cable-free, and fanless design that reduces assembly time from 2 hours to approximately 5 minutes, enhancing standardization [2]. - All computing nodes, switching nodes, and CPO nodes utilize 100% full cold plate liquid cooling to meet the thermal management needs of CPUs, GPUs, switching chips, network cards, and optical modules, further increasing the value of liquid cooling [2]. Group 3: Industry Trends and Opportunities - The silicon photonics wave is accelerating the evolution of the CPO industry, which is expected to drive demand for silicon photonic engines and supporting processes, CW light sources, optical fibers, FAU, and MPO/MTP [3]. - The silicon photonic engine, as the core component of CPO, is aligned with silicon photonic module technology, presenting significant opportunities for traditional optical module manufacturers [3]. - The optical interconnect sector, including ELS/CW light sources, TEC, FAU, optical fibers, and connectors, is expected to see growth, particularly with the increasing application of silicon photonics technology [3].