打造10万卡国产算力集群推出JoyAI世界模型,京东发布物理AI建设最新成果
TL;DR - JD.com announced a major physical-AI expansion spanning domestic compute infrastructure, embodied-data collection, world models, robotics, and consumer services. Its strategy integrates AI with retail and supply-chain operations to accelerate deployment in real-world environments.
- JD Cloud plans to scale from a domestic 10,000-accelerator cluster to 100,000 accelerators, supporting large-model training and low-latency physical-AI applications.
- The company released the interactive JoyAI-Echo WM world model, reporting a leading score of 81.6 on WBench Navigation.
- JD.com is building an embodied-data pipeline targeting over 10 million hours of real-world human video; its first open dataset, EgoLive, is already available.
- Its deployment plans include 80-plus RoboBase sites and, over five years, procurement of 3 million robots, 1 million autonomous vehicles, and 100,000 drones for logistics.