黄仁勋CES扔AI核弹！“六芯”Rubin量产，英伟达大杀器来了

Core Insights - NVIDIA's CEO Jensen Huang announced the new Rubin platform at CES 2026, emphasizing a shift towards "physical AI" [2][3] - The Rubin platform is set to outperform the previous Blackwell generation, with inference performance five times and training performance 3.5 times better, while reducing inference token generation costs by up to ten times [4][10] - The platform is expected to be delivered to initial customers in the second half of 2026 [5] Group 1: Rubin Platform Details - The Rubin platform is the next generation of AI computing architecture, following Hopper and Blackwell, named after astronomer Vera Rubin [8] - It consists of six independent chips, including the core Rubin GPU, designed to significantly lower inference costs [9] - Innovations in the Rubin platform include the latest NVIDIA NVLink interconnect technology, a Transformer engine, confidential computing, and RAS engine [9] Group 2: AI Development and Open Source Models - Huang introduced four open-source models targeting different verticals: Nemotron for agent AI, Cosmos for physical AI, Alpamayo for autonomous driving, and Clara for healthcare [14] - The automotive sector is identified as the first large-scale application for physical AI [15] - Huang highlighted the rise of open-source models as a catalyst for global innovation, noting that their performance is increasingly approaching that of leading proprietary models [17]