💼 DeepSeek preparing a giant cluster of 160,000 Huawei chips
According to Bloomberg, DeepSeek is preparing an order for at least 160,000 Huawei Ascend 950DT accelerators for a new data center with a capacity of about 1 GW in Ulanqab (Inner Mongolia). The chips are intended for inference — running DeepSeek models, while the company continues to train on Nvidia. The stated specifications of the 950DT are 144 GB of HBM (HiZQ 2.0) and 4 TB/s of memory bandwidth, which is close to the Nvidia H200 (141 GB, 4.8 TB/s) in terms of memory capacity.
🌍 If the plans are realized, this will become the largest known cluster on Huawei chips and the largest known inference cluster on Chinese hardware. Huawei is covering large-scale production workloads for the first time, not just pilots, and the launch timelines depend not on demand, but on memory supply: CXMT has only just begun producing small batches of HBM3E and is 3–5 years behind Samsung, SK Hynix, and Micron.
👤 For the first time, the Chinese AI stack will be able to be measured in real combat inference workloads of this scale, not in benchmarks. An unusual detail: Huawei positioned the 950DT as a training chip, and DeepSeek is reserving it for inference, leaving training workloads on Nvidia. Part of the capacity is planned to be launched by the end of 2027 — the beginning of 2028.
Source 1: https://www.bloomberg.com/news/articles/2026-09-04/deepseek-plans-big-huawei-ai-chip-order-to-power-new-data-center Source 2: https://the-decoder.com/deepseek-plans-the-largest-known-huawei-chip-cluster-with-160000-processors-in-inner-mongolia/