Understand this story
Shanghang Technology has deployed a dual-node computing power layout in Huailai and Ulanqab. According to an October 28, 2026 report by China.com Technology, the Huailai node has a fiber-optic latency of 1 to 2 milliseconds from Beijing and handles latency-sensitive workloads such as online inference; the Ulanqab node has a one-way latency of 2.1 milliseconds and an electricity price of 0.32 to 0.35 yuan/kWh, and carries large-scale pre-training and batch tasks. The report proposes data-affinity scheduling, combined with three synchronization paths—cross-region object storage replication, rsync incremental synchronization, and OTN+RDMA dedicated lines—to achieve primary-secondary coordination, and states that moving training corpora and checkpoints to Ulanqab can save 350,000 to 550,000 yuan in storage costs per year. No further progress has been disclosed so far.
AI-generated from public reports
Developments
1 confirmed developments
Shanghang Technology Deploys Computing Power Across Dual Nodes in Huailai and Ulanqab
Source:China.com Technology
Reports and discussion timeline
Browse reports and discussion by time. Their evidence levels remain distinct.
Oct 28
China.com Technology
Shanghang Technology's Huailai + Ulanqab dual nodes: a cross-regional scheduling solution for core inference and batch training
Shanghang Technology has deployed dual computing nodes in Huailai and Ulanqab. Huailai, with a fiber-optic latency of 1 to 2 milliseconds from Beijing, handles latency-sensitive workloads such as online inference, while Ulanqab, with a one-way latency of 2.1 milliseconds and an electricity price of 0.32 to 0.35 yuan/kWh, carries large-scale pre-training and batch tasks. The article proposes achieving primary-secondary collaboration through data-affinity scheduling and three types of synchronization paths: cross-regional replication of object storage, rsync incremental synchronization, and OTN+RDMA dedicated lines, and states that migrating training corpora and checkpoints to Ulanqab can save 350,000 to 550,000 yuan in annual storage costs.