The emergence of DeepSeek's artificial intelligence (AI) models is expected to enhance the competitiveness of Chinese chipmakers like Huawei against more dominant U.S. processors in the domestic market. Contrary to struggling to match Nvidia in developing high-end chips for model training, DeepSeek's approach focuses on optimizing computational efficiency over raw processing power, bridging the gap between Chinese and U.S. AI processors.
Several Chinese companies, including Huawei, Hygon, EnFlame (backed by Tencent), Tsingmicro, and Moore Threads, have hinted at incorporating DeepSeek models into their products, aiming to capitalize on its open-source nature, low fees, and potential to circumvent U.S. export restrictions on high-powered chips. Notably, Huawei's Ascend 910B has gained traction for less computationally demanding "inference" tasks, reflecting a shift towards Chinese AI chipset vendors' strengths in industry-specific applications.
Despite Chinese AI chips being competitive cost-wise for inference tasks domestically, challenges persist in terms of software performance and compatibility with Nvidia's CUDA platform, a pivotal element of Nvidia's market dominance. While efforts like Huawei's Compute Architecture for Neural Networks (CANN) aim to break away from CUDA, the long-term investment needed to match its capabilities could pose obstacles for Chinese chip firms.