军事知识服务平台
开源闭源模式选择与中美大模型产业竞争
Open versus closed:Model strategy choices and the China-U.S.competition in large models
摘要 开源与闭源的模式选择,是大模型产业发展中最具战略意义的问题之一,也是理解中美人工智能竞争的关键.本文在厘清大模型语境下"开放权重"与传统开源软件本质差异的基础上,系统比较两种模式的优势与局限,从技术位势、利润池位置、可占有性与制度环境四个维度阐释模式选择的内在逻辑,进而考察中美大模型产业的发展态势与模式分布,重点分析模式选择对中美大模型产业竞争、上下游相关产业竞争、中国追赶进程以及智能化扩散与治理格局的影响.研究表明:美国形成前沿闭源为主导的金字塔格局,中国形成开放权重模式主导的雁阵格局.开源使我国实现了以扩散换时间、以生态换位势的非对称追赶,但也面临前沿模型差距再度拉大、价值向算力环节集中而算力受制于人、生态优势具备可逆性等约束.立足"十五五"规划纲要部署,应将人工智能作为培育未来产业和新质生产力的主攻方向,健全多元化科技投入机制,统筹算力底座建设与国产化替代,巩固开源生态优势,以智能化赋能实体经济和中小企业,深化国际开放合作,在中美人工智能竞争和整体科技产业竞争中赢得战略主动.
Abstract:
The choice between open and closed model strategies is one of the most strategically significant issues in the development of the large model industry,and a key factor in understanding the China-U.S.competition in artificial intelligence.Building on a clarification of the fundamental differences between"open weights"in the large model context and traditional open-source software,this paper systematically compares the advantages and limitations of the two strategies,and explains the underlying logic of strategy choice along four dimensions:technological position,profit-pool location,appropriability,and institutional environment.It then examines the development trends and strategy distribution of the Chinese and U.S.large model industries,with a focus on how strategy choice affects competition in the large model industry itself,competition in related upstream and downstream industries,China's catch-up process,and the pattern of intelligent-technology diffusion and governance in both countries.The study finds that the United States has formed a pyramid structure dominated by frontier closed-source models,while China has formed a flying-geese structure dominated by open-weight models.Open-sourcing has enabled China to achieve asymmetric catch-up by trading diffusion for time and ecosystem for position,yet it also faces constraints such as a renewed widening of the frontier-model gap,the concentration of value in the compute segment where China remains dependent on others,and the reversibility of its ecosystem advantage.In line with the deployment of the 15th Five-Year Plan,China should make artificial intelligence a primary focus for cultivating future industries and new quality productive forces:maintaining diversified mechanisms for technological investment,coordinating the build-out of compute infrastructure with domestic substitution,consolidating its open-source ecosystem advantage,empowering the real economy and small and medium-sized enterprises through intelligent technologies,and deepening international cooperation and openness,so as to gain strategic initiative in both the China-U.S.AI competition and the broader science and technology industrial competition.
刊 名 中国软科学 ISTICPKU
年,卷(期): 2026, (9)
分类号: F124
机标分类号: F127 F270 F812.45
在线出版日期: 2026年11月3日
基金项目: 26ZDA011:国家社会科学基金DF2023YS24:中国社会科学院学科建设登峰战略优势学科(产业经济学)(DF2023YS24),