Zhi Tong Cai Jing·2025-12-22 23:45Core Viewpoint - The rapid upgrade of GPU and ASIC technologies is driving significant demand for higher bandwidth network hardware in AI data centers, with a notable increase in the demand for 800G optical modules and the initiation of 3.2T optical module research and development [1][10]. Group 1: Market Performance - The optical device/module sector is projected to achieve a revenue of 662 billion yuan in the first three quarters of 2025, reflecting a year-on-year growth of 65% driven by AI [2]. - The net profit growth for the optical device/module sector is expected to reach 169.4 billion yuan, with a remarkable year-on-year increase of 123% [5]. - In Q3 2025, the optical device/module sector's revenue is anticipated to be 255 billion yuan, showing a year-on-year growth of 62% [7]. Group 2: Technological Advancements - The demand for 800G optical modules is expected to continue its rapid growth into 2026, with significant increases in the shipment scale of 1.6T modules and the commencement of 3.2T module development [1][10]. - NVIDIA's roadmap indicates a compression of product upgrade cycles to one year, with increasing rates and bandwidths for their platforms, enhancing the overall performance of network hardware [14]. - Google's TPU product evolution shows a consistent increase in bandwidth, with the latest V7 Ironwood chip achieving an ICI bandwidth of 1200GB/s [15][17]. Group 3: Industry Trends - The industry is witnessing a shift towards Scale-up networks, which require higher bandwidth compared to Scale-out networks, with the average Scale-up bandwidth reaching 10Tbps [19]. - The market potential for Scale-up networks is estimated to be 5-10 times larger than that of Scale-out networks, indicating a significant opportunity for network hardware, including switches and optical modules [19][23]. - Major companies like Google, Meta, and Huawei are already utilizing optical modules to build Scale-up networks, highlighting the growing importance of optical interconnects in addressing bandwidth and distance limitations [22].