Energy infrastructure for AI data centers: A critical review of grid integration, techno-economic challenges, and future pathways

Artificial intelligence (AI) workloads are rapidly changing the energy landscape of today's data centers, both in terms of total electricity usage and power density at the facility level. The ongoing deployment of large-scale AI training and inference infrastructure is further driving and increasing the demand for electrical reliability, cooling capacity, power quality, grid interconnection, and low-carbon electricity supply. These interdependent issues require coordinated consideration of energy-infrastructure planning that takes into consideration renewable energy generation and storage, electrical architecture, operational flexibility, economic performance, and system resilience. The critical review combines 155 sources, mostly published between 2015 and 2026, identified and screened following the PRISMA 2020 guidelines, which includes peer-reviewed journal articles, conference papers, technical standards, agency and institutional reports, market analyses, and corporate disclosure. The resulting evidence is presented in a seven-component decision framework that includes site and resource assessment, technology selection, electrical architecture, control and dispatch, workload flexibility, economic evaluation, and resilience and sustainability. The reviewed evidence indicates that renewable generation, energy storage, grid-support technologies, and workload flexibility can support lower-carbon AI data-center operations; however, the technical and economic performance of these approaches remains highly dependent on project-specific resource conditions, market structures, reliability requirements, and system boundaries. Reliability assessment requires a chronological, site-specific evaluation of load, resource, and infrastructure conditions. Wind–solar complementarity and workload flexibility may improve renewable-energy utilization; however, their benefits remain strongly dependent on operational and geographic constraints. The proposed framework is intended as a screening and decision-support tool rather than a prescriptive design methodology or site-specific optimization model. Important research needs include reproducible techno-economic assessment methods, independently verifiable corporate metrics, improved workload-flexibility characterization, and integrated evaluation of energy, water, reliability, and sustainability impacts. The resulting framework provides a structured basis for evaluating renewable-powered AI data center infrastructure across technical, economic, operational, and sustainability dimensions.