JD Cloud Plans to Co-Build a 100,000-Card Domestic GPU Cluster with Moore Threads

At the 2026 JD Global Tech Explorer Conference today, JD Cloud announced plans to partner with Moore Threads to build a 100,000-card full-featured GPU cluster, creating a large-scale domestic intelligent computing infrastructure. The cluster will focus on large language model (LLM) training, inference, and embodied intelligence, and will open its computing resources to the entire industry. The two parties will conduct full-stack collaboration around chips, cloud platforms, and model training, supporting the iteration of JD’s JoyAI large model, accelerating model training, simulation, and high-quality data generation, with a key focus on serving supply chain AI, embodied intelligence, integrated LLM training and inference, and industrial applications. Previously, JD Cloud had built a domestic 10,000-card cluster with partners including Moore Threads; the latest plan for a 100,000-card cluster marks an expansion of their collaboration from model R&D and inference platforms to large-scale intelligent computing infrastructure. Moore Threads stated that the 100,000-card AI factory will cover scenarios such as LLM training, token generation, and agent training, and serve various industries. Moore Threads has made continuous progress in commercial operations recently: the company reported revenue of 1.736 billion yuan in the first half of the year, a year-on-year increase of 147.42%, with cloud product revenue reaching 1.693 billion yuan, accounting for approximately 97.5% of total revenue. The company had previously completed the deployment of a 10,000-card intelligent computing cluster, and is advancing the construction of GPUs, interconnection networks, software stacks, and computing power scheduling platforms required for the 100,000-card cluster.