Nvidia(US:NVDA) Zhong Guo Jing Ying Bao·2026-01-09 02:08Core Insights - The main focus of the article is on NVIDIA's introduction of the Vera Rubin AI computing platform at CES 2026, highlighting its transition from a chip manufacturer to an AI infrastructure company [2][9]. Group 1: Rubin Platform Overview - The Rubin platform is NVIDIA's first AI platform that integrates six chips, including Vera CPU and Rubin GPU, and has fully entered production [4][5]. - The Rubin GPU shows significant performance improvements over the previous Blackwell GPU, with NVFP4 inference performance increasing to 50 PFLOPS (5 times), training performance to 35 PFLOPS (3.5 times), and HBM4 memory bandwidth to 22TB/s (2.8 times) [5]. - The design of the Vera Rubin NVL72 system allows for faster assembly of computing nodes, reducing assembly time from 2 hours to 5 minutes, while the system operates at 100% liquid cooling [7]. Group 2: AI Infrastructure and Storage Solutions - The introduction of the BlueField-4 DPU supports a new AI storage infrastructure, addressing the growing memory requirements for complex AI tasks [6]. - NVIDIA aims to become a major player in the storage market, not by building storage systems but through partnerships with companies like HP and Dell [6]. Group 3: Open Source Models and AI Applications - Open source models are gaining traction, with 25% of tokens generated from these models, and NVIDIA is leading the open source model ecosystem [2][8]. - The company has expanded its open source model ecosystem across six domains, including biomedical AI and robotics, showcasing the rapid advancement of open source models [8][10]. - The concept of "Physical AI" is emphasized, with NVIDIA's Cosmos model enabling AI to understand physical laws and perform reasoning tasks [9][10].