Why Water Demand Is No Longer Enough: Rethinking Sustainability in AI Data Centers

W ater has shaped the trajectory of human civilization and technological progress for millennia.As Giulio Boccaletti argues, the relationship between societies and water has often determined the course of civilization. 1 The rapid advancement of artificial intelligence (AI), particularly its scalability, follows a similar trajectory.Although AI is portrayed as digital, and often virtual, infrastructure, its operation depends on physical thermodynamic systems that must dissipate heat generated during computation.Consequently, AI is becoming closely coupled to hydrologic systems.The global expansion of AI both depends on and influences water and the environment, but its needs and impacts remain local.As a result, water demand has emerged as an important consideration in AI infrastructure planning and development. 2,3However, water demand alone may no longer be sufficient for evaluating the sustainability of AI infrastructure.As cooling systems increasingly rely on recirculation and reuse, the challenge extends beyond water quantity to water compatibility.The water demand of AI infrastructure originates from one issue: heat.AI data centers operate continuously and at high computational intensity, requiring uninterrupted cooling to