AI Startup DeepSeek Turns to Huawei Chips Amid Nvidia Curbs

Chinese artificial-intelligence startup DeepSeek has begun using Huawei’s Ascend processors to train some of its large language models, according to The Information. The company had previously relied mainly on Nvidia’s A100 and H100 graphics-processing units accessed through cloud providers, but will now shift a portion of its workloads to Huawei hardware. The decision underscores the increasing substitution of domestic chips in China’s generative-AI sector after U.S. export controls restricted access to Nvidia’s most advanced GPUs. Huawei, which introduced its upgraded Ascend 910B accelerator last year, has been courting local model developers with a combination of competitive pricing and guaranteed supply. While DeepSeek is expected to continue using Nvidia products where they remain available, its move adds to pressure on the U.S. company’s China sales and signals growing confidence in Huawei’s silicon for high-end AI training tasks.