JD Cloud Partners with Moore Threads to Plan 100,000-GPU Domestic AI Factory
At the 2026 JD Global Tech Explorers Conference, JD Cloud announced plans to build a 100,000-card domestic AI cluster in partnership with Chinese GPU maker Moore Threads. Building upon their existing 10,000-card cluster collaboration, this facility is designed to train large-scale foundation models, support token production, and power embodied AI applications. This partnership highlights China's aggressive push toward self-reliance in large-scale AI infrastructure amid Western chip export restrictions. Establishing a 100,000-card cluster on domestic GPU hardware represents a significant effort to run massive frontier model training and physical AI workloads without relying on Nvidia chips. Alongside the compute expansion, JD Cloud highlighted progress on its 10-million-hour human data collection initiative and its JoyAI model matrix covering multimodal, world, and embodied AI models. Moore Threads CEO Zhang Jianzhong noted that the 100,000-card factory aims to serve industry developers, drive advancements in embodied intelligence, and support JD's physical operations.
## BACKGROUND
Moore Threads is a Chinese semiconductor enterprise developing domestic GPUs built on its proprietary MUSA architecture as an alternative to Nvidia hardware. JD Cloud is the cloud computing branch of Chinese e-commerce titan JD.com, which powers logistics, customer services, and robotics through its JoyAI model ecosystem. Faced with external hardware export limits, Chinese tech firms are increasingly building massive computing clusters powered entirely by domestic accelerator chips.