DeepSeek AI Model Delayed Due to Huawei Chip Issues

Chinese AI startup DeepSeek faced significant technical challenges using Huawei’s Ascend chips to train its upcoming R2 model, leading to a delayed release. Despite encouragement from Beijing to adopt domestic technology, DeepSeek encountered persistent issues during the training phase, forcing the company to rely on Nvidia chips instead.
The delay, initially scheduled for May, underscores the limitations of Chinese AI chips compared to their US counterparts. Industry insiders have highlighted stability issues, slower inter-chip connectivity, and inferior software in Huawei’s Ascend chips, which lag behind Nvidia’s products.
Huawei dispatched engineers to assist DeepSeek in integrating Ascend chips for the R2 model, but even with on-site support, the company could not successfully complete the training process. DeepSeek remains committed to making the model compatible with Ascend chips for inference, though the training phase continues to rely on Nvidia technology.
The startup’s founder, Liang Wenfeng, has expressed frustration over the delays and is pushing for a more advanced model to maintain the company’s competitive edge. Additionally, longer-than-expected data labeling contributed to the postponement, though reports suggest the R2 model could be released within weeks.
AI researcher Ritwik Gupta noted that while Huawei faces “growing pains” in using Ascend chips for training, the company is expected to adapt over time. However, for now, Chinese AI companies are gravitating toward alternatives like Alibaba’s Qwen3, which has gained popularity for its efficiency and flexibility.
The situation highlights the broader challenges Beijing faces in achieving technological self-sufficiency, as developers increasingly favor proven US-based solutions.

Published: 8/14/2025