CPO网络
Search documents
2026年通信行业投资策略:AI主线多维强化,网络国产纵深推进
Shenwan Hongyuan Securities· 2025-12-31 11:49
Group 1 - The core viewpoint of the report emphasizes "AI mainline multi-dimensional reinforcement and deepening of domestic trends" for the 2026 communication industry investment strategy [3][89]. - The report identifies three main areas of focus for AI communication: overseas mapping, optical communication prosperity, and changes in optoelectronic technology, all expected to evolve in 2026 [4][90]. Group 2 - The overseas AI computing network is shifting from a "Scale out" to a "Scale up" approach, with various standards and solutions emerging domestically, indicating a significant growth in Ethernet-based and open ecosystems [6][92]. - The optical communication industry is projected to experience a boom in silicon photonics solutions and the introduction of CPO networks in 2026, with the prosperity of the optical module market extending to upstream and downstream components [17][90]. - The trend towards optoelectronic integration is expected to reshape the communication industry’s division of labor and value distribution, with a high certainty of mixed optoelectronic trends [19][90]. Group 3 - The domestic trend in the communication sector is accelerating across four key areas: chips, cloud and infrastructure, networks, and end-side AI [91][92]. - In the chip sector, there is significant growth potential for domestic optical and Ethernet chips due to the current global supply chain dynamics [45][92]. - The cloud and infrastructure segment is witnessing a marked improvement in supply-demand dynamics, with the domestic AIDC industry poised for a "second inflection point" [52][92]. Group 4 - The network segment is expected to see a surge in domestic Scale up and supernode solutions, with the satellite internet industry entering a critical phase of large-scale launches and commercialization [74][92]. - End-side AI applications are anticipated to experience explosive growth in 2026, driven by the optimization of domestic large models and hardware costs [82][92].