Sustainability in large language model supply chains-insights and recommendations using analysis of utility for affecting factors

Abstract

Abstract The increasing adoption of Large Language Models (LLMs) has intensified concerns regarding the sustainability of their supply chains, particularly concerning energy consumption, resource utilization, and carbon emissions. To address these concerns, this study proposes a two-step approach. First, a Delphi method is employed to systematically identify the critical factors affecting the sustainability of LLM supply chains. Expert consensus through four rounds of feedback highlights key factors such as Environmental Impact, Computational Efficiency & Resource Optimization, Data Quality & Ethical Considerations, and Social Responsibility & Governance. In the second step, the identified factor’s relative importance was calculated using Conjoint Analysis, a statistical technique used to determine how respondents value different factors of a supply chain of LLMs. This prioritization helps formulate strategies to make LLM’s supply chain sustainable. The low score for environmental impact suggests a lack of awareness about the sustainability of LLMs’ supply chain. The study finds Data Quality and Ethical considerations to be the most important considerations for the respondents. Thus, it provides a framework for implementing sustainable practices in LLMs’ supply chains in resource-constrained settings. The results demonstrate the effectiveness of this combined Delphi-Conjoint Analysis approach, providing actionable insights for AI organizations aiming to enhance the sustainability of their LLM operations.