Chinese artificial intelligence start-up DeepSeek has postponed the launch of its new AI model due to technical difficulties with Huawei’s chips, highlighting the ongoing reliance of Chinese firms on US technology, the Financial Times reports.
The company had been encouraged by Chinese authorities to use Huawei’s Ascend processor rather than Nvidia’s systems following the release of its R1 model in January. However, DeepSeek faced persistent problems training its R2 model on Ascend chips. As a result, the start-up resorted to using Nvidia hardware for training while relying on Huawei’s chips for inference tasks, according to sources familiar with the matter.
Training — the process where a model learns from large datasets — proved particularly challenging, and the issues were cited as the primary reason for the model’s delayed release, originally scheduled for May. Data labelling for the updated model also took longer than expected, further pushing back the launch. Chinese media have suggested the R2 model could be released in the coming weeks.
The difficulties underscore the gaps between domestic Chinese AI chips and US alternatives, especially in stability, inter-chip connectivity, and software support. Huawei reportedly sent engineers to assist DeepSeek on-site, but even with this support, a successful training run on the Ascend chip remained elusive.
Industry experts say the challenges reflect broader hurdles for Beijing’s drive toward technological self-sufficiency. Ritwik Gupta, an AI researcher at the University of California, Berkeley, told the Financial Times that while Huawei is facing “growing pains” in training models, the company is likely to adapt over time.
Meanwhile, Nvidia — at the heart of a geopolitical battle between Washington and Beijing — has resumed sales of its H20 chips to Chinese firms after agreeing to share a portion of its revenue with the US government. “Developers will play a crucial role in building the winning AI ecosystem,” the company said, adding that ceding entire markets could harm US economic and national security.
DeepSeek founder Liang Wenfeng reportedly remains dissatisfied with the R2’s progress, pushing internally to invest more time in developing an advanced model to maintain the company’s AI lead. Both DeepSeek and Huawei declined to comment.
The Financial Times coverage highlights how China’s AI ambitions continue to be shaped by the capabilities of foreign technology, despite Beijing’s push for homegrown alternatives.
