|The recent and rapid expansion of artificial intelligence (AI), data centers, and other digitalization technologies has accelerated global electricity consumption, creating a new paradigm in energy and growth. Still, no comprehensive framework exists to evaluate the role of low-carbon innovations across AI's complex sociotechnical ecosystem. This review addresses three questions: What low- and zero-carbon technologies can help mitigate the energy and carbon footprint of AI and digitalization? What barriers prevent their adoption? Which policy interventions can overcome these barriers? Using a sociotechnical systems approach, we conducted a systematic literature search and screened 364 articles published from 2000 to 2025 to analyze impacts and opportunities across four critical dimensions of AI, data centers, and digitalization provisioning: natural resources, facilities and components, applications, and users and institutions. We identify over 70 mitigation technologies, with reported energy reductions ranging from 13% to 94% across individual studies, alongside projections in high-growth scenarios where data center electricity demand could grow by 13–15% per year to 2030. Three barrier categories emerged: technological constraints, institutional and political limitations, and behavioral resistance. Policy measures such as carbon pricing and mandatory energy reporting, and operational strategies, such as geographic load balancing, are frequently highlighted as high-leverage options for overcoming these barriers. This holistic STS framework provides a foundation for future interdisciplinary research and policy development, identifying critical research gaps including demand forecasting, Global South equity, and organizational change. • A sociotechnical systems approach is used to understand and assess digitalization-energy interactions. • Innovations to achieve a sustainable AI-driven digital future are reviewed. • Institutional constraints, technology gaps, and resistance to behavioral change are major barriers. • Categorized policy options are suggested to overcome identified barriers. • Nine future research agendas are identified.
Kim et al. (2026) studied this question.