This dissertation investigates the intersection of artificial intelligence (AI) and sustainability, revealing how recent and anticipated AI expansion creates new environmental pressures while also enabling novel mitigation strategies. It first reviews how rapid growth in AI, particularly generative AI, is intensifying electricity demand and straining water and carbon budgets, with implications for grid reliability and sustainability targets. This review synthesizes uncertainties in quantifying AI electricity use, evaluates risks at both data center and power system scales, and organizes mitigation strategies into four pillars: efficient AI development, data center efficiency, clean energy adoption, and offset mechanisms. It concludes by outlining key research and policy priorities for aligning AI growth with sustainability goals.The foundation of this dissertation is a quantitative framework for characterizing the energy–water–carbon impacts of future AI facilities, with explicit attention to spatial–temporal dynamics and magnitudes. Chapter 2 develops a pioneering methodology to systematically estimate sustainability impact and net-zero roadmaps for AI servers. It quantifies the energy consumption, water footprint, and carbon emissions of U.S. AI servers from 2024 to 2030 and explores how key drivers, including technology efficiency improvements, siting decisions, and grid decarbonization trajectories, shape net-zero pathways for AI expansion. This work establishes the baseline understanding needed for the subsequent chapters. The second part of the dissertation, encompassing Chapters 3–5, addresses specific system-level challenges created by AI growth and proposes potential solutions. One study develops an integrated framework that couples deep reinforcement learning (DRL) control with cost-effectiveness optimization to enhance cooling efficiency and enable economically viable integration of renewable energy in AI data centers. A second study presents an integrated assessment of how concurrent pressures from AI expansion and climate-driven increases in cooling demand may create electricity bottlenecks in U.S. power systems and we may solve it through multi-level solutions. A third study introduces a comprehensive modeling and assessment framework for U.S. grid regions’ hourly dynamics, revealing how generative AI computing profiles, on-site renewable generation, and alternative AI capacity growth trajectories jointly reshape the power system and how battery storage and AI workload scheduling could be used to mitigate the outlined challenges. Taken together, this dissertation addresses the coupled evolution of AI development, energy system planning, and sustainability. It demonstrates how AI can both hinder and advance grid performance and sustainability targets, further providing actionable recommendations and policy insights for industry, regulators, and other stakeholders seeking to guide AI infrastructure onto climate-aligned, resource-conscious pathways.