PaperlayerView exact revision on arXiv

Source-linked AI summary

Water-Energy-Data nexus reveals sustainability gaps and priorities for AI infrastructure

Jincheng Huang, Guoan Xiao, Xin Lan, Caichun Yin, Zhimeng Jiang, Yue Cui, Jianguo Liu

arXiv:2610.23614v1cs.CY

TL;DR

AI’s resource demands span interconnected infrastructure and regions, but assessments often examine them separately. The paper defines a WED nexus and maps 12,825 studies across four AI infrastructure activities, finding fragmented evidence coverage and a pronounced focus on direct operational demands.

Abstract

from arXiv · show

The rapid growth of artificial intelligence (AI) depends on interconnected water, energy, and data systems spanning infrastructure life cycles and diverse geographies, with the potential to intensify water and energy insecurity, exacerbate social inequities and lead to cascading sustainability issues globally. Yet assessments commonly examine them separately, limiting our understanding of where resource pressures arise, how they are transferred across space, and who bears their consequences. To address these challenges, we established a water-energy-data (WED) nexus framework that captures the spatial and life-cycle dependencies of AI infrastructure, and use large language models to synthesize 12,825 water- or energy-related studies across four resource-intensive AI infrastructure activities: mineral extraction, semiconductor manufacturing, data center operation and electronic-waste processing. The evidence map identifies a mismatch between the interconnected activities supporting AI infrastructure and boundaries used to assess them. Researchers assess operational energy more than water use and other indirect demands, while interactions across locations and life-cycle activities remain understudied. These gaps obscure how resource demands shift among infrastructure nodes and regions, and how data-enabled benefits compare with their underlying costs. The WED nexus provides a foundation for integrated assessment of spatially metacoupled systems, while the evidence map can inform priorities for managing the sustainability implications of the AI boom.

Introduction

AI’s sustainability pressures extend beyond data-center operations to linked water, energy, and data systems across infrastructure life cycles and regions. This study frames those dependencies through the WED nexus and maps evidence across four AI infrastructure activities.

Introduction

  • AI infrastructure connects direct facility demands with upstream production, downstream waste, and resource pressures across regions.Data-center cooling-water reductions can be offset by electricity-related water use, while infrastructure capacity is clustered and supply chains extend beyond end-users.Source: Introduction · p. 2
  • The WED nexus treats water, energy, and data systems as interacting through dependencies, interactions, and feedbacks across activities supporting AI services.It extends the water–energy nexus by adding data and computing infrastructure, linking resource use at computing facilities to upstream production, downstream waste, and regional constraints.Source: Introduction · pp. 2–3
  • The metacoupling framework organizes WED relationships within places, between adjacent systems, and across distant systems.A data center is the focal system: computing and cooling occur within it, while utilities and communities connect it to adjacent systems and hardware production and end-of-life activities connect it to distant locations.Source: Introduction · p. 3
  • The study synthesizes 12,825 water- or energy-related studies across mineral extraction, semiconductor manufacturing, data-center operation, and e-waste processing.It maps coverage of direct and indirect resource use, water–energy relationships, nearby resource pressures, links across infrastructure activities, impacts, and feedbacks.Source: Introduction · p. 3

Results

Across four AI infrastructure activities, the evidence base is growing but uneven across resources, life-cycle stages, locations, and social and environmental interactions. Research more often measures direct energy use than indirect demands or pressures beyond facilities.

Rapid growth and geographical unevenness of the evidence base

  • Research has accelerated since 2023, but remains concentrated in data-center energy studies, upstream water pressures, and a limited set of countries.Mineral extraction contributes substantially to recent water-related growth, while semiconductor manufacturing remains weakly represented across the three corpora.Source: Rapid growth and geographical… · pp. 5–6, Figure 2 · p. 6

Limited attention to indirect water and energy consumption

  • At data centers, 70.60% of energy-data studies quantify direct energy use, versus 24.53% of water-data studies quantifying direct water use.Indirect water use is rarely quantified: 2.45% of water-focused data-center studies measure water use for energy, and 0.58% measure other life-cycle water use.Source: Limited attention to indirect water… · p. 7, Figure 3 · p. 8

Little assessment of nearby and distant resource pressures

  • Facility-level resource use is assessed far more often than nearby pressures, while cross-node assessment is selective rather than balanced across the life cycle.Nearby water pressures appear in 1.75% of data-center studies; no eligible semiconductor-manufacturing or e-waste studies assess nearby energy pressures. Mining-to-e-waste is the most frequently examined cross-node link, while no water-data data-center study quantifies a link to another node.Source: Little assessment of nearby and… · pp. 9–10, Figure 4 · p. 10

Gaps in environmental and social impacts and feedback

  • Environmental interactions receive approximately 7.5 times more attention than social interactions, and distant social impacts and feedbacks are especially rarely assessed.Feedbacks are generally less studied than outward impacts; the highest bidirectional coverage is 0.96% in energy-data data-center research, and no study assesses bidirectional social interactions.Source: Gaps in environmental and social… · pp. 11–12, Figure 5 · p. 13

Discussion

The evidence map finds a mismatch between interconnected AI infrastructure and fragmented assessments, which often fail to connect life-cycle accounting with pressures on host resource systems. The WED nexus links these demands across infrastructure activities, places, and resource systems.

Fragmented evidence establishes the need for the WED nexus

  • Life-cycle accounting and assessment of nearby resource capacity answer different questions, so both are needed to reveal availability, competing demands, and feedbacks.The evidence map finds that broader life-cycle coverage does not consistently assess pressures on nearby resource systems.Source: Fragmented evidence establishes the… · pp. 13–14
  • The WED nexus connects facility operations to water–energy dependencies and traces demands across the infrastructure life cycle, making cross-sectoral and cross-regional transfers visible.It links, for example, water consumed in electricity generation and energy required for ultrapure-water production.Source: Fragmented evidence establishes the… · p. 14

Priorities for the WED nexus assessment and governance

  • WED assessments should account for trade-offs shaped by regional conditions, such as cooling choices that shift burdens between water and electricity systems.Greater material recovery may also reduce resource losses while requiring additional water and energy.Source: Priorities for the WED nexus… · pp. 15–16
  • Spatially explicit analysis should trace how infrastructure choices and constraints propagate through host regions and distant supply chains.Improving performance at one facility may shift environmental footprints elsewhere.Source: Priorities for the WED nexus… · p. 16
  • Governance should connect direct and indirect resource use with local and transboundary burdens, including who bears impacts and can participate in decisions.The passage calls for coordination across resource systems and supply chains, informed by disclosure of life-cycle resource demands.Source: Priorities for the WED nexus… · p. 16
  • Concurrent climate, geopolitical, pandemic, and supply-chain shocks can disrupt interdependent water, energy, and data systems and amplify one another.Heat waves and droughts can jointly increase cooling demand while constraining water and electricity supply.Source: Priorities for the WED nexus… · p. 16

Limitations and Outlook

  • The evidence map prioritizes broad coverage over extracting every study result, and title-and-abstract coding may omit assessments reported only in full texts.Web of Science selection, possible underrepresentation of grey literature and non-English publications, four non-exhaustive infrastructure nodes, and difficulty attributing shared sectoral resources exclusively to AI bound the synthesis.Source: Limitations and Outlook · p. 17
  • Fragmented assessments focus on direct facility use and miss how resource pressures and social and environmental interactions shift across host regions and distant supply chains.Research also underrepresents resource-vulnerable regions where many indirect impacts occur.Source: Limitations and Outlook · pp. 17–18
Loading 2610.23614v1…