这不是显卡，是一座2吨重的AI工厂

Core Insights - NVIDIA has announced the full production of the Vera Rubin platform, which weighs nearly 2 tons and integrates six new chips, significantly enhancing inference cost and training efficiency, achieving AI computations at a trillion operations per second, marking it as a true AI factory [2] - The company has also open-sourced its first inference VLA (Vision-Language-Action) model, Alpamayo 1, designed for vehicles to "think" and solve problems in unexpected situations, utilizing a 10 billion parameter architecture [3] - The first vehicles equipped with NVIDIA technology are set to hit the roads in the US in Q1, Europe in Q2, and Asia later in the year [4][17] Production and Performance - The new Rubin platform has a performance increase of 5 times compared to the previous Blackwell version, with training performance being 3.5 times better [5][8] - The Rubin platform can reduce inference token generation costs by up to 10 times and decrease the number of GPUs required for training mixture of experts models by 4 times [8] - The Vera CPU in the Rubin platform features 88 cores, providing double the performance of its predecessor, and is designed for agent inference, making it the most energy-efficient processor in large-scale AI factories [8] Ecosystem and Deployment - Major cloud providers, including Microsoft, are expected to be among the first to deploy the new hardware in the second half of the year, with Microsoft’s next-generation Fairwater AI super factory set to utilize NVIDIA's Vera Rubin NVL72 systems [6] - NVIDIA maintains a long-term bullish outlook, predicting the total market size could reach several trillion dollars despite concerns about increasing competition and sustainability of AI spending [7] Innovations and Technologies - The Rubin platform incorporates five innovative technologies, including the sixth-generation NVLink interconnect technology and a third-generation transformer engine, which provides 50 petaflops of NVFP4 computing power for AI inference [9] - The platform's modular design allows for faster assembly and maintenance, with an 18 times quicker process compared to Blackwell [9] Open Source Initiatives - NVIDIA has released the Alpamayo model as part of a complete open ecosystem for autonomous driving development, which includes simulation frameworks and datasets [14][21] - The Alpamayo model is designed for the autonomous driving research community, allowing developers to adapt it for vehicle development and as a foundational tool for autonomous driving technology [15][18] - NVIDIA has also launched various open-source models and tools across different sectors, including the Nemotron family for agent AI and the Cosmos platform for physical AI [26][27]