About

The unprecedented demand for data centers, fueled by the rapid growth of artificial intelligence (AI), has led to the construction of many gigawatt-scale data centers. While the AI boom is a remarkable technological milestone, it also raises urgent concerns about sustainability. The inherently resource-intensive nature of AI is projected to double the energy footprint of data centers in the coming years, placing stress on vulnerable power infrastructures while exacerbating environmental and public health challenges. The operation of data centers has been linked to worsening water stress as well as health impacts from toxic byproducts resulting from the energy they consume. Unfortunately, the water footprint and health impact of data centers have not received as much attention as their carbon emissions. Moreover, we lack well-established and well-understood methodologies for quantifying this water footprint and health impacts, leading to misleading comparisons and claims.

The goal of this tutorial is to educate the research community on methodologies to quantify the water footprint and health impacts of AI data centers. The tutorial will enable the research community to move beyond carbon-centric perspectives toward a more holistic view of sustainable AI infrastructure. We will offer hands-on exercises on tracking the carbon-water-health footprints of the AI model lifecycle (training and inference). We will also discuss approaches for addressing these concerns, using examples and exercises designed to highlight the technical and policy trade-offs.

Contents

  • Introduction and Motivation
    • State of AI data centers
    • Why look beyond carbon: introducing water & public health
  • Module 1: Background on Data Center Infrastructure
    • Power and cooling
    • Workloads
    • Data center management
  • Module 2: Water
    • Water consumption in AI data centers
    • Quantifying AI's water footprint
    • Case study: dataset walk-through
    • Hands-on exercise: water footprint quantification
  • Module 3: Public Health
    • Public health impacts of AI data centers
    • Quantifying AI's public health impacts
    • Case study: community exposure example
    • Hands-on exercise: quantifying public health impacts
  • Module 4: Addressing the Impacts
    • Strategies to reduce water footprint
    • Strategies to mitigate health risks
    • Hands-on exercise: policy/technical trade-off scenarios
  • Wrap-up with Discussion on Key Takeaways

Logistics

Duration: 4 hours (approx.) Format: In-person Audience: Systems, sustainability, AI researchers and practitioners