JD Cloud said on Sept. 9 at the 2026 JD Global Tech Explorers Conference that it plans to work with Moore Threads to build a 100,000-card full-function GPU cluster, aiming to create a large-scale domestic intelligent computing infrastructure platform.
The companies said the cluster will focus on large model training, inference and embodied intelligence, and that its computing power will be opened to the broader market.
Stack-wide coordination across chips, cloud and model training
According to JD Cloud, the partnership will span chips, cloud platforms and model training. The effort is intended to support iteration of the JD JoyAI large model, while accelerating model training, simulation and high-quality data generation.
JD Cloud said the planned infrastructure will mainly serve supply chain AI, embodied intelligence, integrated large-model training and inference, and industrial use cases.
Expansion from a 10,000-card cluster to a 100,000-card plan
JD Cloud said it had previously worked with Moore Threads and other partners to build a domestic 10,000-card cluster. With the new 100,000-card plan, the cooperation is extending from model development and inference platforms to larger-scale computing infrastructure.
Moore Threads said the planned 100,000-card AI factory will cover scenarios including large model training, token production and agent training, and will serve multiple industries.
Moore Threads discloses recent commercialization progress
Moore Threads said its commercialization progress has continued to advance in recent months. The company reported first-half revenue of 1.736 billion yuan, up 147.42% from a year earlier. Revenue from cloud products reached 1.693 billion yuan, accounting for about 97.5% of the total.
The company also said it had already completed deployment of a 10,000-card intelligent computing cluster and is continuing work on the GPU hardware, interconnect network, software stack and computing scheduling platform needed for a 100,000-card cluster.