The Energy Economics of Artificial Intelligence Data Centers: Demand Uncertainty, System Costs, Flexibility, and Credible Decarbonization

Artificial intelligence is accelerating data center electricity demand, but annual energy totals alone do not reveal when, where, or at whose cost new loads enter power systems. This review asks how uncertain projects should be translated into demand, infrastructure, price, flexibility, and environmental consequences. We conducted a structured critical review of 44 peer-reviewed studies, working papers, and official, regulatory, statistical, system-operator, and institutional reports identified through 31 August 2026, with explicit source selection and claim-level reliability appraisal. The central finding is that the economic effect of AI data centers depends on project-realization probability, coincident demand, and location, not on announced capacity alone. Concentrated growth can require generation and network investment and raise local prices. Enforceable temporal or spatial flexibility can reduce these costs, but unverified flexibility warrants no planning credit. Annual renewable matching supports an accounting claim but not necessarily consequential decarbonization, which also requires additional supply, temporal matching, and geographic deliverability. Policy should therefore condition connection on project credibility, cost responsibility, measured flexibility, and credible procurement. This review’s incremental contribution is to move beyond consumption inventories by integrating uncertain project realization, short-run operation, medium-run connection and investment, long-run expansion, distributional incidence, and environmental counterfactuals in one energy-economics framework.