报告导读： 2026 年，英伟达在具身智能领域的进展已经从技术探索逐步进入工程落地与 生态化阶段，核心围绕软件模型、仿真基础设施、硬件部署与合作生态四个方向展开。 机器人模型： 发布 GROOT N1.7 ，预告 GROOT N2 。与 1 月份 CES 的 N1.5 相比，机器人"操作能力升级"，并引入了推理模块。机器人从"会做动 作"升级到了"会理解任务并规划动作"。 N2 目标是 : 让机器人具备"世界模型级的物理理解能力。 机器人数据： Cosmos3 结合了合成世界生成、视觉推理和动作模拟的新模型，能够自动完成从原始数据到最终训练数据集的整个流程。 机器人训练策略： 发布 Isaac Lab 3.0 抢先体验版，基于全新 Newton 物理引擎 1.0 与 NVIDIA PhysX® 软件开发套件，新增了多物理场仿真功能，并强化 了对复杂灵巧操作场景的支持。 算力基础设施： DGX™-AI 训练服务器平台，特点是集成多块高端 GPU 、 NVLink 高速互联、专为 AI 训练优化、可以组成 GPU 集群。 生态体系： 与 1X 、智元机器人、 Agility 、 Agile Robots 、 Bo ...

Investment Rating - The report assigns an "Overweight" rating for the industry, indicating a projected performance that exceeds the Shanghai and Shenzhen 300 Index by more than 15% [11]. Core Insights - In 2026, NVIDIA's advancements in embodied intelligence have transitioned from technical exploration to engineering implementation and ecosystem development, focusing on software models, simulation infrastructure, hardware deployment, and collaborative ecosystems [2][4]. - The humanoid robot industry is entering the "engineering implementation" phase, with significant investment opportunities identified in both beta (software and embodied models) and alpha (hardware) categories [4]. Summary by Relevant Sections Investment Recommendations - The report suggests focusing on: 1. Beta opportunities in embodied models and software, recommending companies such as Hangcha Group, UBTECH, and Woan Robotics. 2. Alpha opportunities in hardware, including: - Force/Torque sensors: Anpeilong - Encoders: Yapu Co., Ltd., Shuo Beid - Visual sensors: Aoptical, Orbbec - IMU: Joyson Electronics, Huayi Technology - Power devices: Innodisk - MCU and hollow cup: Fengcai Technology - Dexterous hands: Zhaowei Electric - Linear joints: Hengli Hydraulic, Zhejiang Rongtai, Zhenyu Technology - Rotary joints: Lide Harmonic, Shuanghuan Transmission, Minshi Group, Keda Li [4][5]. NVIDIA's Ecosystem Development - NVIDIA is building a "CUDA ecosystem" for robotics, utilizing Cosmos (data generation), Isaac (simulation training), and GROOT (robotic foundational models) to attract developers and accelerate commercialization [4]. - The GROOT N1.7 model has been released, enhancing robotic operational capabilities and introducing reasoning modules, with the upcoming GROOT N2 aimed at achieving world-model-level physical understanding [4]. Robotics Training and Infrastructure - The report highlights the release of Isaac Lab 3.0, which features a new Newton physics engine and enhanced support for complex dexterous operation scenarios [4]. - The DGX™-AI training server platform integrates multiple high-end GPUs and is optimized for AI training, capable of forming GPU clusters to enhance training scale [4].