Core Viewpoint - The introduction of Rubin CPX by Nvidia is expected to significantly innovate in PCB, connectors, and cooling architectures, potentially expanding the hardware market size, with the AI PCB market projected to reach $6.96 billion in 2027, a 142% increase from 2026 [2][5][32]. Group 1: Rubin CPX Innovations - Rubin CPX is designed for ultra-long context AI inference tasks, featuring a decoupled inference architecture that separates computation-intensive context processing from memory bandwidth-intensive generation stages, enhancing efficiency and performance [4][7]. - The GPU provides 30 Peta FLOPS of computing performance at NV FP4 precision and is equipped with 128GB GDDR7 memory, achieving a memory bandwidth of 2 TB/s, which offers a cost-effective solution compared to HBM [11][20]. - The hardware changes include a modular design for the tray architecture, liquid cooling upgrades, and a wireless cable structure, which collectively improve thermal management and connectivity [20][23]. Group 2: Market Impact and Projections - The estimated value of a single VR200 NVL144 cabinet PCB is approximately 456,000 yuan, with a single GPU corresponding to a PCB value of 6,333 yuan ($880), reflecting a 113% increase compared to the GB300 [5][28]. - The projected shipment volumes for 2027 are 10,000 racks across various models, leading to a total PCB market size of $6.96 billion, indicating a substantial growth trajectory for the AI PCB market [5][32]. - The anticipated PCB market sizes for 2025 and 2026 are $1.53 billion and $2.64 billion, respectively, highlighting the significant growth expected with the introduction of Rubin CPX [32].
中金 | AI进化论(17):解耦式推理创新,Rubin CPX驱动PCB市场迭代升级