小马智行与摩尔线程达成战略合作，以国产AI算力加速中国自动驾驶规模化落地

Core Viewpoint - The strategic collaboration between Pony.ai and Moore Threads represents a significant demonstration of collaborative innovation in China's artificial intelligence industry, focusing on the application and optimization of L4 autonomous driving technology [2][3]. Group 1: Strategic Collaboration - Pony.ai and Moore Threads will focus on the implementation and scaling of L4 autonomous driving technology, leveraging Pony.ai's core technology, including world models and virtual driver systems [2]. - This partnership marks the first large-scale application of domestic AI computing power in critical training and simulation processes for autonomous driving [2]. - The collaboration aims to create a comprehensive ecosystem for autonomous driving capabilities, enhancing the synergy across algorithms, data, computing power, and applications [2]. Group 2: Technological Advancements - Pony.ai's world model, based on reinforcement learning, generates over 10 billion kilometers of testing data weekly, allowing the virtual driver to evolve through repeated training [3]. - The collaboration will utilize Moore Threads' MTT S5000 training and inference integrated computing card and the Kuangya computing cluster to support the training and validation of Pony.ai's models [2][4]. - Moore Threads has rapidly iterated four generations of GPU architecture, achieving international advanced levels in training large models with hundreds of billions to trillions of parameters [4]. Group 3: Market Position and Goals - Pony.ai has become the first company in China to operate fully autonomous Robotaxi services in four major cities, with a fleet size of 1,159 vehicles as of December 31 last year, exceeding its annual target [4]. - The seventh-generation Robotaxi system has achieved a positive unit economic model in Guangzhou, indicating significant progress in commercial viability [4]. - The goal is to scale the Robotaxi fleet to over 3,000 vehicles by the end of 2026, with the collaboration expected to enhance training efficiency and overall system performance [4].