DeepSeek plans to deploy at least 160,000 Huawei Technologies AI accelerators at a data center it is building in Inner Mongolia, according to Bloomberg. If completed at that scale, Bloomberg says the project could become one of the largest known clusters of Huawei AI chips. The planned chips are Huawei’s next-generation Ascend 950DT accelerators, Bloomberg reports, citing people familiar with the matter. DeepSeek intends to use them for operating its models, while the installation schedule will depend on Huawei’s production capabilities. The distinction between operating and training matters. Bloomberg reports that DeepSeek does not currently plan to use the Ascend 950DT chips for training, even though Huawei designed and marketed the processors for that more demanding workload. The report says DeepSeek tried to train earlier models on Huawei chips, but has so far relied on Nvidia accelerators for that step. That makes the planned Inner Mongolia buildout a major test of Huawei’s role in China’s artificial intelligence infrastructure. The reported order would not, based on Bloomberg’s account, mean DeepSeek has fully replaced Nvidia for its most compute-intensive model development work. It would, however, put a large Huawei-based cluster behind model operations if Huawei can supply the chips. Bloomberg frames the project as part of China’s broader effort to replace Nvidia in AI compute. The current evidence supports that strategic read, but the reported deployment remains prospective: the chips have not been described as installed, and Bloomberg says timing depends on Huawei’s production capacity. Who benefits: Huawei benefits if it can supply the Ascend 950DT chips at the reported scale. DeepSeek could gain a large domestic compute base for running models if the deployment proceeds. Who's exposed: Nvidia is exposed to China’s push to replace its accelerators in AI infrastructure, though Bloomberg’s report says DeepSeek has still relied on Nvidia for training. Huawei is also exposed operationally because the deployment timeline depends on its production capabilities.