Updated
Updated · Global Times · Sep 9
JD Cloud, Moore Threads to Build 100,000-GPU AI Cluster as China Pushes Domestic Chips
Updated
Updated · Global Times · Sep 9

JD Cloud, Moore Threads to Build 100,000-GPU AI Cluster as China Pushes Domestic Chips

3 articles · Updated · Global Times · Sep 9

Summary

  • Moore Threads said the project will mark the first use of domestically developed GPUs in a 100,000-card core cluster at a leading Chinese AI cloud provider, moving homegrown AI computing toward broader commercial deployment.
  • The cluster will use Moore Threads' Universal GPUs and support large-model training, inference and embodied-intelligence workloads, with JD Cloud opening the computing capacity to companies across industries.
  • JD Cloud said the partnership will span chips, cloud platforms and model training, aiming to speed development of its JoyAI models and build a pipeline from data and training to simulation and deployment.
  • China's policy backdrop has turned more supportive: MIIT this week called for orderly deployment of intelligent-computing clusters with 10,000, 100,000 or more accelerator cards and better adaptation to domestic chips.

Insights

Can China's new 100,000-GPU cluster truly rival foreign tech, or will software bottlenecks hinder its massive domestic hardware ambitions?
Will brute-force scaling of 100,000 GPUs overshadow the urgent need for energy-efficient, brain-inspired AI architectures in edge computing?
How will the colossal energy demands of these massive domestic AI clusters be met without straining regional power grids?