According to the research report released by Soochow Securities on the Smart Finance app, the top players in the industry chain are all laying out, and AI+ Ethernet is a confirmed trend. The penetration of RoCE will effectively stimulate the demand growth of the AI computing interconnection industry chain and product technology iteration. The leading manufacturers in the industry are expected to maintain their shipment growth with more stable market shares, while domestic chip makers are expected to achieve breakthroughs in technology and customers. Zhongji Innolight (300308.SZ) and Suzhou TFC Optical Communication (300394.SZ) are recommended for attention, and Eoptolink Technology Inc. (300502.SZ), Yuanjie Technology (688498.SH), and Shengke Communication (688702.SH) are worth paying attention to at various stages of the industry chain.

The top players in the industry chain are all laying out, and AI+ Ethernet is a confirmed trend. IB networks have dominated in the early stage of AI computing construction, but the industry chain has been exploring the possibility of Ethernet adapting to AI computing. The Ultra Ethernet Consortium (UEC) has emerged, and various network, system, and cloud computing companies in the industry chain such as Broadcom, Nvidia, and Meta have continued to invest and make progress. RoCE is expected to gradually replace IB's mainstream position.

Why use Ethernet & RoCE? Compared with traditional TCP / IP technology, RDMA is more in line with the requirements of high concurrency and low latency of AI computing, so it is a better choice for network technology. IB and Ethernet both support RDMA, but IB naturally supports RDMA and is the local optimal solution for short-term rapid, high-quality, and high-volume implementation of computing power on AI computing construction in the early stage. Ethernet has a deep application foundation and low cost, and is expected to be the best solution in the future.

How does Ethernet achieve high requirements for AI interconnection? AI interconnection mainly faces two major problems: 1) the long-tail effect brought about by the significant increase in "elephant flows" can be solved through RoCE's adaptive routing function; 2) congestion caused by multiple data receivers sharing the same receiving end in different computing processes can be solved through RoCE's switch congestion control algorithm + buffer pool.

What industrial changes does Ethernet bring? 1) Switch capacity is increased, and RoCE supporting functions such as adaptive routing and congestion control are added. At the same time, more abundant software and hardware also provide greater space for white-box switches, which is conducive to their further penetration; 2) The growth in inferring demand has opened up the implementation of RoCE. Cloud factories have accelerated self-built inference computing power to bring about 800G optical module additions, while Nvidia's customer structure continues to optimize, based on the advantages of training and inference on one platform leading the iteration and application of cutting-edge products such as 1.6T; 3) Silicon photonics has a three-layer industrial logic of guaranteeing optical module supply, undertaking the trend of silicon-based co-packaging, and reducing costs. It is expected to accelerate penetration.