19 May 2025 16:24:07 ET │ 11 pages NVIDIA Corp (NVDA.O) Computex Day 1 Takeaways CITI'S TAKE Flash | 更多资料加入知识星球：水木调研纪要 关注公众号：水木Alpha On Sunday, NVIDIA's CEO Jensen Huang, via a keynote, reiterated the company's goal to accelerate Gen AI adoption through GPU-powered data centers at Computex in Taipei. While various announcements were made, we highlight the following three: 1) NVLink Fusion announcement (see body for details), 2) new developments in physical AI with NVIDIA Isaac GR00T N1.5 foundation model fo ...

Core Insights - NVIDIA has introduced the NVIDIA Isaac GR00T N1.5, an updated foundation model for humanoid reasoning and skills, along with the GR00T-Dreams blueprint for generating synthetic motion data and Blackwell systems to enhance humanoid robot development [1][3][10] Group 1: Product Announcements - The NVIDIA Isaac GR00T N1.5 model improves adaptability to new environments and workspace configurations, enhancing its success rate in material handling and manufacturing tasks [7][8] - The GR00T-Dreams blueprint generates vast amounts of synthetic motion data, allowing robots to learn new behaviors and adapt to changing environments [3][4] - The GR00T-Dreams blueprint complements the previously released GR00T-Mimic blueprint, which augments existing data rather than generating new data [5] Group 2: Industry Adoption - Various humanoid and robotics developers, including Agility Robotics, Boston Dynamics, and XPENG Robotics, are adopting NVIDIA's Isaac platform technologies to advance their robot development [2][9] - Early adopters of the GR00T N models are utilizing them for tasks such as natural language understanding and complex workflows in industrial settings [8] Group 3: Technological Advancements - NVIDIA's simulation technologies, including Isaac Sim and Isaac Lab, are being used by companies to close the data and testing gap in humanoid robot development [9] - The Blackwell systems, including NVIDIA RTX PRO 6000 workstations, provide a unified architecture for running various robot development workloads, achieving up to 18 times greater performance for data processing [10][11]