Tech Desk 
Aug 15, 2025
In what appears to be the last-minute escape for the global tech shares' collapse, Chinese AI startup DeepSeek's launch of new AI model, DeepSeek R2, has been delayed after technical difficulties with Huawei Technologies’ chips.
The Huawei chip mishap reportedly forced the company to rely on US-made Nvidia Corp. chips instead.
DeepSeek experienced challenges while training the R2 model with Huawei's Ascend processors, Financial Times noted. As a result, the company shifted to using Nvidia chips for training, utilising Huawei hardware solely for inference tasks.
This delay, originally set for May, has hindered DeepSeek's progress in the competitive AI landscape. This setback highlights the limitations of China's efforts to reduce reliance on U.S. technology.
Despite Beijing's push for domestic AI chip usage and scrutiny of Nvidia's H20 processors, Chinese chips still struggle with stability, software support, and inter-chip connectivity compared to Nvidia's offerings. Huawei even sent engineers to assist DeepSeek, but the team could not successfully run the Ascend chips.
DeepSeek continues to collaborate with Huawei to ensure the model's compatibility for inference. Amid rising US-China tensions, Nvidia and Advanced Micro Devices have agreed to remit 15% of their China chip sales to the U.S. government for export licences.
Despite the current challenges, Liang Wenfeng, DeepSeek's founder, expressed optimism about the launch of the R2 model in the coming weeks.
