黄仁勋罕见提前宣布：新一代GPU全面投产

Core Insights - NVIDIA has accelerated its product release schedule by unveiling the next-generation AI chip platform "Rubin" earlier than usual at CES 2026, breaking its traditional March GTC announcement timeline [2][3] - The Rubin platform is designed to meet the increasing computational demands of AI for both training and inference, featuring a collaborative design of six new chips [5][6] Group 1: Rubin Platform Details - The Rubin platform integrates six chips: NVIDIA Vera CPU, Rubin GPU, NVLink 6 switch chip, ConnectX-9 SuperNIC, BlueField-4 DPU, and Spectrum-6 Ethernet switch chip, covering multiple layers from computing to networking, storage, and security [5] - Compared to the previous Blackwell architecture, Rubin accelerators improve AI training performance by 3.5 times and operational performance by 5 times, with a new CPU featuring 88 cores [5] - The Rubin platform can reduce inference token costs by up to 10 times and decrease the number of GPUs required for training MoE (Mixture of Experts) models by four times compared to the Blackwell platform [5] Group 2: Ecosystem and Market Response - The NVL72 system, which includes 72 GPU packaging units, was also announced, with each unit containing two Rubin dies, totaling 144 Rubin dies in the system [6] - Major cloud providers and model companies, including AWS, Microsoft, Google, OpenAI, and Meta, have responded positively to the Rubin platform and are among the first adopters [6][7] Group 3: Strategic Shift in AI Focus - NVIDIA's presentation at CES included a broader AI ecosystem strategy, shifting focus from "training scale" to "inference systems," with the introduction of an Inference Context Memory Storage Platform designed for efficient management of KV Cache [9] - The company is also advancing its long-term vision of physical AI, releasing open-source models and frameworks aimed at extending AI capabilities into robotics, autonomous driving, and industrial edge scenarios [9][10] - The introduction of the Alpamayo open-source model family for autonomous driving, along with a high-fidelity simulation framework, indicates NVIDIA's commitment to enhancing inference-based autonomous driving systems [13]