供应链
采购
采购
业务
利润(经济学)
双头垄断
产业组织
计算机科学
风险分析(工程)
竞争优势
炸薯条
战略式采购
新产品开发
供求关系
供应链管理
质量(理念)
博弈论
降低成本
供应商关系管理
实证研究
竞赛(生物学)
商业模式
作者
Tang Guoying,Yali Zhang,Liaoliao Li,Li Wenjing
出处
期刊:Conference proceedings
[Institute of Electrical and Electronics Engineers]
日期:2025-10-05
卷期号:: 6733-6738
标识
DOI:10.1109/smc58881.2025.11343643
摘要
The rapid development of large language models (LLM) has driven a significant surge in demand for computing power, but the high concentration in the supply market has exacerbated the risk of "computing power dependency." Some LLM manufacturers have attempted to address this challenge through research and development (R&D) of artificial intelligence chips. This study constructs a duopoly game model to analyze the chip acquisition strategies of LLM manufacturers under supply chain risks. We compare two scenarios—where both LLM manufacturers procure chips (NN model) and where one manufacturer relies on R&D (NR model). The findings reveal that: (1) When the manufacturer’s competitor opts for R&D chips, if the performance of the competitor’s chips is at a low level, the manufacturer can experience a significant increase in price and profit by choosing to purchase chips; (2) When the competitor consistently opts for procurement, the LLM manufacturer should choose R&D over purchasing if their internally developed chips achieve moderate performance levels; (3) Customer sensitivity to supply disruptions increases LLM manufacturers’ stability preference, thereby enhancing the competitive advantages of R&D strategy; computational efficiency improvements can substitute part of the value brought by R&D strategy. These findings offer theoretical insights into how LLM manufacturers can balance cost, performance, and risk in chip strategy decisions.
科研通智能强力驱动
Strongly Powered by AbleSci AI