Net sustainability assessment of AI-driven data centers in the GCC: an MCDA approach framework

Abstract

The Gulf Cooperation Council (GCC) is witnessing a rapid growth in artificial intelligence (AI)-enabled data center development, thanks to national digital transformation strategies and the region’s strategic location bridging global markets. Yet, no evaluation framework holistically compares the direct impacts of data center operations on the environment with the dispersed sustainability impacts of AI applications across major business sectors. To address this, this study proposes and applies a Multi-Criteria Decision Analysis (MCDA) framework to assess carbon emissions, water use, energy consumption, social impacts and governance structures across three scenarios (business-as-usual (BAU), moderate transition (MOD), and aggressive decarbonization (OPT). Using secondary structured data from peer-reviewed sources, institutional databases (IEA, UNFCCC, Uptime Institute) and independently verified case studies, the analysis applies clearly defined boundary rules and attribution principles. The findings suggest that under the BAU high-growth scenario, modelled emissions increase substantially by 2035, while AI-enabled sustainability applications in buildings, industry, and utilities provide only partial offsets under optimistic adoption assumptions. The NIS indicates that net-positive outcomes are achievable only when substantial renewable-energy procurement, best-practice cooling, and transparent verification of AI benefits occur together. The results extend socio-technical systems theory and ecological modernization theory by addressing the boundary problem in sustainability assessment of digital infrastructure, providing a scalable approach for policymakers and investors in high-carbon, water-scarce regions. The Net Impact Score (NIS) is not to be construed as an absolute causal prediction, but rather as a scenario-conditional decision-support indicator because AI-benefit attribution and scaling are both dependent on the mentioned assumptions and data limits.