Rise of DeepSeek AI Models Boosts Chinese Chipmakers' Competitiveness
The emergence of DeepSeek's AI models is providing Chinese chipmakers like Huawei with enhanced opportunities to compete against formidable US processors in the domestic market.
DeepSeek's Focus on Inference
DeepSeek's models emphasize "inference," the stage where AI models draw conclusions. This approach prioritizes computational efficiency over raw processing power. Analysts believe this strategy could partially bridge the performance gap between Chinese-made AI processors and their US counterparts.
Chinese Chipmakers Embrace DeepSeek
Huawei, Hygon, EnFlame, Tsingmicro, and Moore Threads have announced support for DeepSeek models. These chipmakers aim to leverage DeepSeek's open-source nature and low costs to accelerate AI adoption and real-life applications.
Overcoming US Export Restrictions
Industry experts predict that DeepSeek's characteristics could aid Chinese firms in overcoming US export restrictions on their most potent chips. Products like Huawei's Ascend 910B demonstrate the suitability of Chinese AI processors for less computationally intensive "inference" tasks.
Nvidia's Dominance Persists
Despite DeepSeek's advances, Bernstein analyst Lin Qingyuan emphasizes that Nvidia chips remain superior even for inference tasks. While US export restrictions prohibit the entry of Nvidia's most advanced AI training chips into China, the company can still sell less powerful training chips for inference purposes.
Nvidia's CUDA Platform
In addition to hardware, Nvidia's CUDA parallel computing platform plays a pivotal role in its dominance. Many Chinese AI chip companies have attempted to develop CUDA-compatible products, but persuasion has proven challenging.
Software Performance Challenges
Omdia's Lian Jye Su highlights software performance as another area where Chinese AI chip firms face limitations. CUDA's comprehensive library and software capabilities demand substantial long-term investment.