Nvidia(US:NVDA) Tai Mei Ti A P P·2026-01-07 03:35Core Insights - NVIDIA's CEO Jensen Huang emphasized the accelerating demand for advanced processors in AI model training and operation, highlighting the semiconductor industry's need to adapt quickly to increasing computational complexity [1] - The introduction of the DeepSeek R1 and the mention of Chinese open-source models Kimi K2 and Qwen were noted as significant developments in the AI landscape [1] Group 1: New Chip Architecture - NVIDIA unveiled the Rubin platform, consisting of six components including Rubin and Rubin Ultra GPUs and CPUs, designed for handling massive computational loads required for AI model training [2] - The Rubin GPU's NVFP4 inference performance is 50 PFLOPS, five times that of the Blackwell platform, while its training performance is 35 PFLOPS, 3.5 times higher than Blackwell [2] - The Rubin platform also features a memory bandwidth of 22 TB/s and 336 billion transistors, representing significant improvements over the previous generation [2][3] Group 2: Focus on Agentic AI - NVIDIA is working to lower the development costs for Agentic AI, introducing Nemotron-CC, a multilingual pre-training corpus covering over 140 languages with a total of 1.4 trillion tokens [4] - The company also launched the "Granary" instruction dataset aimed at making models ready for enterprise-level tasks [4][5] - The ease of building functional personal assistants using NVIDIA's hardware and frameworks marks a significant shift in AI development accessibility [5] Group 3: Emergence of Physical AI - Huang highlighted Physical AI as the next major focus, with applications in autonomous driving, robotics, and industrial manufacturing [6][7] - NVIDIA has been developing Physical AI for eight years, with simulation at the core of its efforts, utilizing the Omniverse platform to create a digital twin environment for safe and efficient AI training [7] - The introduction of the open-source inference VLA model Alpamayo aims to accelerate the development of safe autonomous vehicles [8] Group 4: Industrial Applications - NVIDIA announced a deepened collaboration with Siemens to integrate its Physical AI models and Omniverse simulation platform into Siemens' industrial software, covering the entire lifecycle from chip design to production operations [9] - Huang described this collaboration as the beginning of a new industrial revolution, with Physical AI enabling automated production lines and digital twin systems [9] Group 5: Strategic Vision - The event served as Huang's declaration on the future of AI and computing over the next decade, with NVIDIA aiming to define the technological standards and infrastructure for the next AI era [10] - The company's strategy continues to focus on an open-source and integrated hardware-software approach, ensuring a strong presence across all computing nodes from data centers to smart devices [10]