黄仁勋点赞三款中国大模型，英伟达押宝物理AI

Core Insights - The CES 2026 showcased NVIDIA's strategic focus on next-generation computing platforms and advancements in physical AI, marking the first time in five years that NVIDIA did not release a new GPU at the event [2][3]. Group 1: Open Source Ecosystem - NVIDIA's CEO highlighted the significant investment of approximately $10 trillion in computing resources over the past decade, emphasizing a shift in software paradigms rather than just hardware upgrades [3]. - The presentation acknowledged the rapid development of Chinese open-source models, specifically naming Kimi K2, DeepSeek V3.2, and Qwen, which are leading the open-source ecosystem alongside OpenAI's GPT-OSS [5]. - Despite being approximately six months behind the top models, the open-source models are expected to see new iterations every six months, attracting interest from startups, giants, and researchers alike [5]. Group 2: Next-Generation Computing Platform - NVIDIA introduced the Vera Rubin computing platform, designed to accelerate AI training speeds and facilitate the development of next-generation models [7]. - The platform features a complete redesign of six chips, including Vera CPU and Rubin GPU, with the Rubin GPU achieving a performance of 50 PFLOPS, five times that of its predecessor [7][8]. - The engineering design of Vera Rubin simplifies assembly, reducing the number of cables from 43 to just six liquid cooling pipes, allowing for quicker setup times [8]. Group 3: Advancements in Physical AI - NVIDIA's CEO announced the launch of the Alpamayo open-source AI model aimed at enhancing autonomous driving capabilities, addressing complex driving scenarios through a new reasoning model [10][11]. - The Alpamayo series incorporates a "thinking chain" reasoning model, improving decision-making processes in autonomous vehicles and enhancing user trust in the technology [11]. - The first vehicles utilizing NVIDIA's technology are expected to hit the roads in the U.S. in Q1, Europe in Q2, and Asia later in the year, with interest from companies like Jaguar Land Rover and Uber [11]. Group 4: Robotics Development - NVIDIA unveiled two new open-source models for robotics, NVIDIA Cosmos and GR00T, along with a performance evaluation tool, Isaac Lab-Arena, aimed at simplifying robot training processes [12]. - The collaboration with Hugging Face integrates NVIDIA's Isaac open-source models into the LeRobot project, accelerating the development of the open-source robotics community [12]. - Companies such as Boston Dynamics and Caterpillar are developing new robots and autonomous devices based on NVIDIA's technology, indicating a significant advancement in the robotics sector [13].