Chinese AI startup DeepSeek has postponed the release of its anticipated R2 model after training efforts using Huawei's Ascend chips encountered persistent technical issues, the Financial Times reported on August 14, 2025. The company ultimately resorted to using Nvidia chips for training while employing Huawei's Ascend chips only for inference, highlighting the challenges Chinese developers face as they attempt to reduce their dependence on American technology.
The R2 model was originally scheduled for release in May, but problems with Huawei chips pushed the launch back by weeks as DeepSeek falls behind competitors' high-profile releases. The situation underscores the dilemma facing Chinese AI developers who are encouraged by Beijing to use Huawei technology, particularly as chip exports between the United States and China became a major point of contention this year. While the U.S. permitted Nvidia to resume sales of its H20 chips in China in July, Beijing recently raised security concerns about the use of American-made chips in critical Chinese AI development.
DeepSeek's previous R1 model, which shook up the AI sector earlier this year, was developed largely using H20 chips, as are those of other major Chinese AI developers including ByteDance, Tencent, and Alibaba. China asked local developers this week to justify their Nvidia chip orders, signaling the growing tension between technological independence and development capabilities. Chinese media reports indicate the R2 model will be released in the coming weeks, though experts caution that Chinese chip manufacturers are rapidly catching up with their American counterparts, making it just a matter of time.
