JD Cloud, Moore Threads Plan 100,000-GPU Chinese-Silicon Cluster
TL;DR
- The cluster is a tenfold expansion of an existing JD Cloud/Moore Threads 10,000-GPU deployment, giving the announcement operational credibility beyond a paper commitment.
- Moore Threads' BIS Entity List designation removes customer optionality: JD Cloud is the only credible hyperscale buyer for PH100/MTT S5000 silicon at this scale.
- Global Times frames the deal as explicit fulfillment of MIIT's five-year plan for 10,000-plus-card intelligent computing clusters, converting private commercial action into a policy-accountability data point.
JD Cloud will build a 100,000-GPU intelligent computing cluster on Moore Threads' Universal GPUs, in what TechNode described as the first deployment of Chinese GPUs at that scale inside a major domestic AI cloud provider. The plan was unveiled Wednesday at JD's 2026 Global Technology Explorers Conference.
The two companies have worked together on earlier deployments, which the Global Times reported ranged from thousands to tens of thousands of GPUs. They will now "cooperate across the full technology stack, from chips and cloud platforms to model training," with the new cluster supporting large-model training and inference, embodied intelligence, and JoyAI, JD's own model family.
The announcement lands against a policy push. China's Ministry of Industry and Information Technology has issued guidance calling for intelligent-computing clusters at the 10,000- and 100,000-accelerator-card scale and this week set a 9,800 Eflops national target for 2030, a signal that has fed a steady run of stories on our China AI tracker this month.
Moore Threads only listed on Shanghai's STAR Market in December 2025, and this is the largest publicly disclosed cluster built on its silicon. Neither company published a delivery timeline or a spend figure.
What others are reporting
-
Global Times Read →
State media frames the deal as fulfillment of MIIT's five-year plan mandate for 10,000-plus-card clusters, anchoring a private commercial announcement to official policy standing.
It would be the first time domestically developed GPUs are used in a 100,000-card core computing cluster at a leading Chinese AI cloud provider.
-
Data Center Dynamics Read →
Infrastructure trade press covers the buildout's physical-layer implications; DCD readership treats a 100K-GPU deployment as a proof-of-scale milestone for domestic silicon at hyperscale.
-
TechTimes Read →
Only source to flag the verification gap on Moore Threads' 95% linear-scaling claim, surfacing the open question of whether the performance assertion can withstand independent audit.
-
Aroged Read →
Names the specific hardware stack (PH100 chip, MTT S5000 accelerator, PingHu architecture), confirms Entity List status, and establishes the tenfold scale-up from the prior 10K-GPU collaboration.
The cluster will be composed of 'full-featured' Moore Threads artificial intelligence accelerators designed to train large-scale artificial intelligence models.
Originally reported by technode.com
Read the original article →Original headline: JD Cloud to Build China's First Domestic 100,000-GPU Cluster on Moore Threads Silicon