Hong Kong Economic Times (HKET), the leading financial daily in Hong Kong, recently reported that Chinese AI vendor DeepSeek delayed its new AI model release originally planned for this May. DeepSeek failed its model training plan due to the use of Huawei Ascend AI chips.
After DeepSeek released its R1 model this January, which caused a major setback in the U.S. stock market, the company was encouraged by the Chinese authorities to use Huawei’s Ascend AI processors instead of NVidia chips to train its AI models.
However, DeepSeek encountered persistent technical problems while training R2 on Huawei Ascend chips. Huawei sent a dedicated team of engineers to DeepSeek’s office to help develop the R2 model using its AI chip. However, even with the team present, DeepSeek was unable to successfully train on the Ascend chip. DeepSeek has now abandoned Huawei AI chips and switched to Nvidia for AI model training. In the meantime, DeepSeek is still working with Huawei to make the R2 model compatible with Ascend chips for inference tasks only.
DeepSeek’s latest woes show that Chinese AI chips still lag behind their American competitors in key tasks, highlighting the challenges China faces in achieving technological self-sufficiency.
Source: HKET, August 14, 2025
https://inews.hket.com/article/3993355