计算机｜物理AI：下一代AI形态，军事仿真新引擎

Core Viewpoint - Physical AI is the next generation of AI that can understand physical rules and generate vast training datasets, with potential applications in intelligent driving, robotics, and military sectors [1][2][8] Summary by Sections Definition of Physical AI - Physical AI is defined as the ability to understand and comply with various physical laws, enabling dynamic interaction and autonomous operation between virtual and real worlds [2] - It can accelerate CAE simulation, support digital twins, and synthesize physically compliant scene data for training models in humanoid robots, autonomous driving, and military applications [2] NVIDIA's End-to-End Physical AI Practice - NVIDIA's Omniverse serves as the next-generation "soft core," evolving into a comprehensive physical simulation platform by integrating the PhysX5 physics engine and the Cosmos simulation engine [3] - This platform provides end-to-end support for industrial applications, allowing developers to train models, create virtual environments, and deploy applications [3] - NVIDIA announced collaborations with General Motors for future autonomous vehicle fleets and released the open-source humanoid robot model Groot N1 [3] Military Simulation and Domestic Developments - Physical AI is accelerating the intelligentization of military simulations, with applications across the entire lifecycle of military operations [4] - Domestic company Suochen Technology has launched the "Tiangong·Kaiwu" physical AI platform, mirroring NVIDIA's approach and offering comprehensive solutions for various industries [5] Military AI Market Outlook - China's defense budget for 2025 is approximately 1,784.7 billion yuan, with a consistent annual growth rate of around 7.1% to 7.2% from 2022 to 2024 [6] - The military AI sector is expected to experience structural reversals as the country enhances its military informationization and AI capabilities [6] Investment Strategy - Investment opportunities are identified in companies with algorithm advantages and engineering experience in the simulation field, as well as those with core data resources and application experience in military AI [8]