Core Viewpoint - Nvidia's CEO Jensen Huang visited Taiwan to meet with TSMC, highlighting the ongoing tensions between Washington and Beijing regarding AI chip exports and the company's upcoming financial report [1][8]. Group 1: Nvidia's AI Chip Developments - Nvidia has completed the initial tape-out of six new AI chips, including a new AI GPU based on the Rubin architecture and a silicon photonic processor [3]. - The Rubin architecture is set to succeed the Blackwell architecture, with production aimed for 2026, featuring significant upgrades such as HBM4 memory and enhanced NVLink bandwidth [3]. - The silicon photonic processor is expected to be used for AI data center networking and high-speed interconnects, supporting larger-scale AI GPU interconnections [4][5]. Group 2: Market Dynamics and Future Products - Nvidia is reportedly developing a new AI chip, tentatively named "B30A," specifically for the Chinese market, based on the latest Blackwell architecture [7]. - The H20 AI chip, designed for the Chinese market, has been noted for its unique advantages in AI inference workloads, despite its lower performance in AI training compared to the H100 [7][8]. - Huang indicated that the decision regarding the future of the H20 AI chip successor lies with the U.S. government, emphasizing the complexities of international trade regulations [8]. Group 3: Supply Chain and Production Adjustments - Nvidia has instructed suppliers, including Foxconn and Amkor Technology, to halt production related to the H20 AI chip amid concerns over safety risks and to manage existing inventory [9]. - The company has a substantial stock of H20 chips and is awaiting orders from Chinese customers before proceeding with further production [9].
黄仁勋闪现台北 透露英伟达(NVDA.US)下一个“AI蓝图”关键词:Rubin、硅光子与中国市场