Core Viewpoint - The ongoing AI boom may lead to Nvidia losing its current dominance as large data center operators shift towards custom chips (ASICs) to reduce costs [1] Group 1: Market Dynamics - Counterpoint's report indicates that Broadcom is expected to maintain its position as the top AI server computing ASIC design partner, with its market share projected to expand to 60% by 2027 [1] - TSMC, closely partnered with Broadcom, is anticipated to rapidly grow, capturing nearly 99% of the wafer manufacturing orders from the top ten global data centers and ASIC customers [1] Group 2: Competitive Landscape - The second phase of the AI chip boom is expected to see intense competition between ASICs and GPUs following Nvidia's dominance in the GPU market [1] - Broadcom and TSMC are positioned to emerge as the biggest winners in this evolving competitive landscape [1]

Group 1 - The A-share market indices continued to decline in the afternoon, with the Shanghai Composite Index down 1.3%, the Shenzhen Component Index down 1.91%, and the ChiNext Index down 2.34% [1] - The largest AI ETF in the Shanghai market (515070) fell by 2.55% during the session, while its constituent stocks such as Desay SV, Kewo Si, and Junsheng Electronics led the gains, whereas stocks like Heheta, Zhongke Xingtou, and Cambricon Technologies led the losses [1] - Muxi Co., Ltd. announced that its stock will be listed on the Shanghai Stock Exchange's Sci-Tech Innovation Board on December 17, 2025, and will be included in the Sci-Tech Growth Tier due to its current unprofitability [1] Group 2 - The AI chip market is currently characterized by "GPU leading and ASIC rising," with general-purpose GPUs maintaining dominance in AI training and inference due to their strong parallel computing capabilities [2] - ASICs are experiencing rapid development due to their higher computational efficiency and energy efficiency in specific scenarios, with significant growth in shipment volumes [2] - The focus of the industry is shifting from individual chip performance to system-level solutions like "super nodes" for larger-scale computing power integration and efficient utilization [2]