Core Viewpoint - The year 2026 is expected to be "extremely tight" for the industry, with a significant surge in demand for high bandwidth memory (HBM) and advanced packaging, making computing power a key term in the capital market [1][16]. Group 1: Industry Insights - Huang Renxun, CEO of NVIDIA, highlighted the challenges in producing the new generation AI chip architecture, Blackwell, in 2025, and indicated that the supply chain will be "extremely tight" in 2026 [3]. - The dinner gathering included key executives from major AI supply chain companies, with a combined market value exceeding $5 trillion, indicating the scale and importance of the AI industry [2]. - TSMC is expected to significantly increase its CoWoS (Chip on Wafer on Substrate) capacity from 35,000 pieces per month in 2024 to over 65,000 pieces in 2025, with NVIDIA accounting for over 60% of the demand [5]. Group 2: Company Performance - Victory Technology (胜宏科技) anticipates a substantial increase in net profit for 2025, projected between 4.16 billion to 4.56 billion RMB, representing a year-on-year growth of 260.35% to 295.00% [7]. - Industrial Fulian (工业富联) expects a net profit of 35.1 billion to 35.7 billion RMB for 2025, an increase of 119 million to 125 million RMB from the previous year, reflecting a growth of 51% to 54% [8]. - Cambrian (寒武纪) forecasts a revenue of 6 billion to 7 billion RMB for 2025, marking a year-on-year growth of 410.87% to 496.02%, indicating its first annual profit since establishment [14]. Group 3: Market Dynamics - The demand for high-density interconnect (HDI) boards is surging due to the increased data throughput requirements of AI servers, with Victory Technology positioned to capitalize on this trend [8]. - The emergence of domestic chip companies like Cambrian and the recent IPOs of several GPU firms indicate a shift towards domestic alternatives in the computing power market, driven by the limitations of NVIDIA's high-end chips [15]. - The rapid growth of companies in the optical module sector, such as Zhongji Xuchuang, reflects the increasing demand for high-speed optical modules driven by GPU performance improvements [11].
黄仁勋台北“夜宴”:2026年仍将是AI供应链极度吃紧的一年?