Training, not just inference

DeepSeek has already used Huawei AI chips for inference workloads. The new development is its effort to extend Huawei silicon to training — the more computationally demanding stage of building AI models, which has remained firmly in the domain of Nvidia GPUs.

Context and caveats

The assessment was based on a review of 12 primary sources. While the shift toward Huawei hardware for training is substantiated, the claim carries uncertainty regarding the scale of deployment and whether training on Huawei chips can match the performance of Nvidia-based systems. Details on production readiness were not confirmed.

Reducing dependence on Nvidia carries strategic weight for Chinese AI firms, as US export controls restrict access to Nvidia's most advanced GPUs. A successful move to domestic training hardware would mark a significant milestone for China's semiconductor and AI ecosystem.

The claim that DeepSeek is challenging training workloads on Huawei AI chips, going beyond inference in a bid to reduce Nvidia reliance, is largely accurate. Verdict: Mostly True