瑞银分析师熊瑋发布对中国AI产业的展望，指出中国AI模型在成本控制方面形成了显著竞争优势[1]。部分领先模型的训练成本仅为海外龙头的约1成，API定价也仅为海外同类模型的10%至20%[1]。尽管价格更低，中国模型业者在API业务上仍能维持20%至40%的毛利率[1]。DeepSeek、智譜AI及Kimi等中国领先模型均采取开源策略[1]。
然而，算力供给不足成为制约中国AI发展的主要瓶颈[1]。各家模型业者的收入前景主要受限于能够支持推理需求的算力规模[1]。算力需求明显大于供给，几乎所有模型业者都反映算力短缺问题[1]，这将影响模型能力提升和行业营收前景[1]。
UBS analyst Xiong Wei released an outlook on China's artificial intelligence sector, highlighting how Chinese AI models have built competitive advantages through significantly lower training and inference costs compared to overseas competitors.[1] According to the analysis, some leading Chinese models achieve training costs at merely 10 percent of those incurred by global tech leaders, while API pricing averages between 10 and 20 percent of comparable international models.[1] Despite these cost efficiencies, Chinese AI companies maintain healthy profit margins, sustaining gross margins of 20 to 40 percent on API services.[1]
However, the analyst identified insufficient computing power as the primary bottleneck constraining China's AI development trajectory.[1] Demand for computing capacity substantially exceeds available supply, with virtually all model developers reporting inadequate computing resources.[1] This shortage directly impacts both the capability advancement of AI models and the revenue prospects of industry players.[1] Meanwhile, leading Chinese models including DeepSeek, Zhipu AI, and Kimi have adopted open-source strategies.[1]