• AI hardware energy efficiency is evaluated holistically. • Behavioral fuzzy modeling improves decision realism. • Long-term sustainability is the key evaluation driver. • 3D stacked GPU architectures rank highest. • Results support strategic AI infrastructure planning. The rapid expansion of artificial intelligence applications has led to a substantial increase in energy consumption at the hardware level, raising critical concerns regarding long term sustainability, scalability, and environmental impact. While existing studies predominantly emphasize short term performance improvements, comparatively little attention has been given to systematically identifying which factors and hardware strategies are most decisive for energy efficient AI infrastructure development. Addressing this gap, this study aims to evaluate and prioritize artificial intelligence hardware technologies from an energy efficiency perspective by considering both technical performance and long-term sustainability requirements. This study employs a fuzzy multi criteria decision making framework based on the logarithmic percentage change driven objective weighting method (LOPCOW) to determine criterion weights, ranking alternatives by the trace of geometric scores method (RATGOS) to prioritize alternatives, and the weighted sum product (WISP) for comparative validation, all integrated within a behavioral fuzzy environment. The analysis identifies sustainability and dissemination potential toward 2035 is the most influential criterion, indicating that long term viability and large-scale adoption are perceived as more critical than short term efficiency gains alone. Among the evaluated technologies, 3D stacked GPU and chiplet based architectures emerge as the most favorable option due to their superior ability to reduce data movement energy and support scalable deployment, followed by data processing unit and SmartNIC based solutions. These findings provide strategic insights for policymakers, data center operators, and technology developers by highlighting which hardware pathways offer the greatest potential for balancing energy efficiency, scalability, and future oriented sustainability in artificial intelligence systems.
Eti et al. (Wed,) studied this question.