英伟达推出新一代人工智能平台Vera Rubin

Core Viewpoint - NVIDIA has launched its next-generation super chip "Vera Rubin" at CES 2026, which integrates a Vera CPU and two Rubin GPUs, designed to support advanced AI models and mixed expert models, addressing the growing computational demands in AI training and inference [1][6]. Group 1: Product Features - The Rubin platform includes six chips, with the Vera CPU and Rubin GPUs being central to its architecture, aimed at enhancing AI capabilities [1][6]. - The platform is designed to achieve higher operational efficiency compared to the previous generation, with a 75% reduction in GPU requirements for training the same mixed expert models [3][8]. - The Rubin platform can lower token costs during inference by 90%, optimizing overall ownership costs for AI models [3][8]. Group 2: Market Position and Competition - NVIDIA's Rubin platform is currently in mass production and has been provided to partners for testing, reinforcing its leading position in the chip market with a valuation of approximately $46 billion [4][9]. - The company faces competition from AMD, which has launched its own computing systems, and from clients like Google and Amazon, who are expanding their use of custom chips [4][10]. - Despite increasing competition, NVIDIA's dominance in the AI chip sector is expected to remain strong if it continues its annual product iteration strategy [5][10]. Group 3: Client Engagement and Market Demand - Major cloud service providers like Microsoft, Google, and Amazon are investing billions in large-scale computing systems, including NVIDIA's NVL72 servers, which can integrate 72 GPUs [2][7]. - NVIDIA has introduced its AI storage solution, essential for managing and sharing large-scale AI models and data generated from multi-step inference processes [2][7].