DeepSeek CEO Liang Wenfeng informed investors the startup will increasingly use Huawei hardware for AI model training. This strategic shift responds to U.S. export controls that restrict access to high-performance Nvidia chips.
The company expects to receive a new batch of Huawei training processors between Q4 2026 and Q1 2027. DeepSeek is currently adapting its models to run on Huawei’s Ascend architecture.
While the startup previously utilized Nvidia hardware for intensive training, it is now prioritizing a domestic supply chain. This move reflects a broader trend among Chinese tech firms to secure critical AI components locally.