Chinese technology firm DeepSeek (DEEPSEEK) is using some of Huawei’s processors to train smaller artificial intelligence models, as it looks to reduce its reliance on Nvidia (NASDAQ:NVDA), The Information reported.
DeepSeek, which stunned the world in January, had tested AI accelerators from Huawei, Baidu (BIDU) and Cambricon to train its models, the news outlet added, citing sources familiar with the matter. It eventually chose Huawei and is working with the Chinese telecom giant to use its Ascend processors for training and refining smaller versions of its R2 model, which has not yet been released.
DeepSeek is still using Nvidia’s processors for its more powerful R2 models, the news outlet added. The R2 model has not yet been publicly released and has been reportedly delayed amid a setback with some of Huawei’s chips.
DeepSeek, Huawei and Nvidia did not immediately respond to a request for comment from Seeking Alpha.
Cambricon — which is seen as a Chinese alternative to Nvidia — reported a 4,000% surge in revenue and swung to profit in the first half of the year earlier this week.
