Photo by suntorn somtong on Pexels

DeepSeek, an AI company, has reportedly abandoned its efforts to train its R2 AI model using Huawei’s Ascend chips due to persistent technical difficulties. This setback underscores the challenges facing China’s ambition to achieve self-reliance in advanced AI hardware.

Sources indicate that DeepSeek experienced significant issues during the training phase of its R2 model, forcing a return to NVIDIA’s established and powerful GPU infrastructure. The problems encountered with Huawei’s chips reportedly stalled the project, exposing a gap in capabilities between AI model training and inference.

While Huawei’s chips have shown promise for inference tasks, they apparently proved inadequate for the computationally intensive demands of large-scale AI model training. The situation highlights the existing performance differential between NVIDIA’s leading-edge technology and domestically produced alternatives, raising broader questions about the viability of relying solely on local hardware for developing cutting-edge AI applications.