JD Cloud and Moore Threads plan a 100,000-GPU domestic deployment

Data Center Dynamics reports that JD Cloud and Moore Threads are set to collaborate on a 100,000-GPU domestic deployment—a tenfold step up from the 10,000-GPU system the two companies have previously partnered on—and JD Cloud says it would be the first such capacity hosted by a major domestic AI cloud provider; the announcement at JD's 2026 Global Technology Explorers Conference reads as a declaration of intent more than a construction schedule.

Securities Times reports that the new capacity will run on Moore Threads' "full-featured GPUs" for large-scale model training, inference, and embodied intelligence workloads, though the announcement does not say which part of a portfolio spanning the MTT S5000 (built on the PingHu architecture with full-precision support from FP8 to FP64), the MTT S4000, and the MTT S3000 would fill the order; JD Cloud operates data centers across China with core regions in Beijing, Guangzhou, Shanghai, and Suqian, but the reporting does not specify a location or timeline.

Moore Threads' history frames the project as more than procurement: founded in 2020 by James Zhang Jianzhong, a former Nvidia executive, the company has attracted funding from Shenzhen Capital Group, GGV Capital, Sequoia Capital China, ByteDance, and Tencent, listed on the Shanghai Stock Exchange in December 2025, and sits on the US Bureau of Industry and Security's Entity List—which is why a domestic cloud provider and a domestic chipmaker publicly promising 100,000 domestic GPUs is a statement about China's AI stack as much as it is about capacity.

The pattern has a recent precedent: in August, Starcloud raised $250 million to scale on an unbuilt chip. JD Cloud's project differs in that Moore Threads has public product specifications, but the distance between a press-conference announcement and an operating deployment is the same, and the reporting as it stands does not identify the site, the power arrangement, a delivery date, or a customer for the compute. Contract-backed capacity earns infrastructure pricing; merchant compute without a named buyer is a merchant shell until proven otherwise, and with no customer named, this project sits on the merchant side of that line—a statement of intent, important, but not yet an asset. The first named customer would change that.