DeepSeek V4 Matches GPT-5 Performance on Huawei Chips, Bypassing US Export Restrictions

In a significant breakthrough for AI hardware independence, DeepSeek V4 Pro Max has matched the reasoning benchmarks of GPT-5 using Huawei Ascend silicon instead of the dominant Nvidia GPUs. The achievement highlights architectural efficiency over raw compute power, as the Mixture of Experts model delivers a 1-million token context window while circumventing US export bans. The results suggest that the so-called compute moat is eroding, as alternative hardware proves capable of supporting cutting-edge AI models.