Data centers in the age of AI: A tutorial survey on infrastructure, sustainability, and emerging challenges

Abstract

Data centers have become the backbone of the digital economy, supporting cloud services, digital platforms, and the growing demands of artificial intelligence (AI). The emergence of large language models (LLMs) has amplified their energy use, cooling intensity, and environmental impact, raising urgent concerns around sustainability, grid stability, and operational efficiency. This tutorial surveys the evolution of modern data centers, beginning with their classification and architectures, and examining key dimensions of power consumption, cooling technologies, and water utilization. It underscores how performance indicators such as Power Usage Effectiveness (PUE) and Water Usage Effectiveness (WUE), and carbon-aware operations are shaping sustainable practices. Particular emphasis is placed on AI workloads, where training consumes gigawatt-hours of energy and inference, scaled across billions of queries, drives long-term infrastructure demand despite uncertain profitability models. The survey also explores latency-critical operations, edge deployments, and the transformation of optical networking and fabrics to support AI-driven bandwidth and efficiency requirements. Emerging paradigms including digital twins, AIOps-driven orchestration, and hybrid optical/electrical fabrics are identified as enablers of more resilient and adaptive infrastructures. Looking ahead, the integration of advanced cooling, renewable alignment, demand-responsive scheduling, and net-zero commitments will be pivotal in reconciling scalability with sustainability. By consolidating insights from research and practice, this work provides both a foundational overview and a forward-looking roadmap for designing AI-ready data centers that balance performance, resilience, and environmental responsibility at the core of the digital economy.