CITIC Securities: Super nodes enhance AI training and deployment efficiency, focusing on three major growth segments

Mars Finance reports that Citic Securities’ research states that supernodes connect multiple accelerator cards through a high-bandwidth, low-latency Scale-up network, providing memory pooling and memory direct connection mechanisms, greatly enhancing AI training and inference efficiency, which has become a clear industry trend. With manufacturers like NVIDIA promoting the efficiency and expansion of supernodes, the chip exchange between GPUs, liquid cooling, and cabinet power supply segments will benefit from the pure incremental usage of Scale-up, the penetration rate and ASP increases driven by high power in cabinets, and the ASP increase driven by high power in cabinets, achieving growth; it is expected that by 2028, the incremental space will reach $100 billion, $13 billion, and $24 billion respectively. Among them, the domestic substitution opportunities for switch chips are worth looking forward to; the good commercial attributes of switch chips shape a stable oligopoly pattern, with a large and clear incremental space for GPU interconnections, and the domestic space is expected to reach $5 billion by 2028. Currently, Ethernet solutions are becoming its main technical direction, and with the trend of domestic production of computing power chips, supporting domestic solutions have already been implemented. Currently, the leading revenue scale is about 1 billion yuan, with significant growth potential; it is recommended to pay attention to domestic Ethernet switch chip players; in addition, the domestic substitution of optical interconnect CPO/NPO links based on switch chips is also worth noting.