通信行业周报：ASIC需求指引强劲，光模块有望筑底

Investment Rating - The report maintains an "Overweight" rating for the communication industry [4][28]. Core Insights - Strong demand for ASICs is highlighted, with Broadcom's AI revenue expected to grow by 220% to reach $12.2 billion in FY2024, and the AI chip market projected to be between $15 billion and $20 billion next year. Marvell's revenue is also expected to increase by 7% year-on-year to $1.516 billion in Q3 FY2025, driven by sales of custom AI chips to Amazon and other data center companies [5][20]. - The optical module market remains robust, with a high probability of a second bottoming out. The demand for high-speed optical modules is expected to remain strong, with 800G demand anticipated to grow significantly by 2025, and 1.6T expected to gradually increase. The silicon photonics solution is expected to maintain profitability due to its cost advantages [5][20]. Summary by Sections 1. Market Review - In the past week (December 9-13, 2024), the Shanghai Composite Index and Shenzhen Component Index decreased by 0.36% and 0.73%, respectively, while the CITIC Communication Index increased by 0.95%, ranking 11th among 30 first-level industries. The communication sector showed an upward trend, with sub-sectors like communication equipment and telecom operations II changing by 0.37% and 1.50%, respectively [4][14]. 2. Key Industry Insights - ASIC demand is strong, with a focus on scaling at the edge. ASICs are designed for specific applications, offering higher performance and lower power consumption compared to general-purpose GPUs. Major companies are actively investing in this area [5][20]. - The optical module sector is expected to perform well, with a low valuation level and a potential bottoming out after recent corrections. The report suggests monitoring companies involved in computing/ASIC, AEC, and optical modules [5][20]. 3. Industry News - TSMC's 2nm chip production yield has reportedly reached over 60%, with expectations for large-scale production next year. Apple plans to use TSMC's 3nm process for its iPhone 17 series, while the first product using the 2nm chip is expected to be the iPad Pro M5 in late 2025 [21][28]. - IBM has announced a new optical-electrical co-packaging technology that could enhance AI model training speeds by five times and significantly reduce energy consumption [28].