The Sustainable Future of Knowledge: How Green Libraries Embrace Energy-Intensive AI

Abstract

Modern libraries increasingly institutionalize sustainability through green building certifications, yet frequently overlook an invisible emissions giant: generative AI (GenAI). Recent data indicates that a large language model like GPT-3 used 700,000 liters of water during its pre-training period (Li et al., 2023). Thus, the sustainability that libraries try to maintain by reducing the physical plant is being rapidly offset by the increasing cost of resources. Drawing on the ESG and Green Information Systems (Green IS) literature, this paper addresses the contradiction between environmental management and unchecked technology adoption. To alleviate this, we propose a conceptual framework of “digital vegetarianism” within information management. We believe that libraries should strategically pivot toward Small Language Models (SLMs), which is more efficient. Furthermore, we formalize a mathematical model for the ontological measurement of digital service carbon to reveal the quantifiable physical weight of digital knowledge labor. By confronting the systemic ecological blind spots of algorithmic expansion, this study advances Green IS theory and redefines the boundaries of sustainable digital research infrastructures, shifting the academic discourse from physical dematerialization to systemic infrastructure accountability.