DeepSeek plans to install at least 160,000 Huawei Ascend 950DT processors in an Inner Mongolia data center, according to reporting cited by The Decoder. If completed, it would be the largest publicly known cluster built from Huawei AI chips.
The company intends to use the system for inference—the work of running trained models and answering requests—rather than the more demanding process of training them. DeepSeek reportedly continues to rely on Nvidia hardware for training. Even with that division, a cluster of this size would represent a significant attempt to shift large-scale Chinese AI operations onto domestic processors.
Delivery is the main constraint. Huawei may need more than a year to fill the order because of processor production limits and shortages of high-bandwidth memory, which feeds data to AI chips at high speed. Chinese memory maker CXMT has begun producing small batches of HBM3E, but remains behind suppliers that are already mass-producing the newer HBM4 generation.
The project is therefore a plan rather than an operating system, and its timetable depends on a supply chain still catching up with demand. It also shows where domestic hardware may gain traction first: serving models can be separated from Nvidia-dependent training while Chinese chip and memory capacity develops.