Atlantic Investigation: AI Data Center Water Debate Distorted by Missing Data and Bad-Faith Arguments on Both Sides
Summary
- • U.S. data centers used ~17 billion gallons of water for cooling in 2023 — less than 0.1% of farm usage nationally, but this masks severe local-level impacts
- • New York Governor Hochul signed the nation's first statewide data-center moratorium this week, citing potential massive water demands from hyper-scale facilities
- • A Meta campus under construction in Lebanon, Indiana could demand 8 million gallons per day — more than double the host town's peak demand
- • Researchers say no reliable ground-truth data exists on data-center water use, enabling both alarmism and tech-industry denial to thrive unchecked
Details
U.S. data centers used ~17B gallons of water for cooling in 2023
Equivalent to roughly 500,000 Americans' annual use; less than one-tenth of one percent of total U.S. farm water that year — alarming in isolation, negligible in agricultural context
NY Governor Hochul signs nation's first statewide data-center moratorium
Hochul cited 'massive amounts' of water hyper-scale data centers could require; first such statewide restriction in U.S. history, signed this week
Meta Indiana campus: up to 8M gallons/day at peak
Meta's Lebanon, Indiana campus would more than double the town's peak daily water demand; city officials say planned infrastructure upgrades will handle the load
Newton County, GA official: 'We just don't have the water'
Local official says county must 'race' to improve water-recycling facilities to handle incoming data center demand; county lacks current capacity
No ground-truth data on data-center water use, says UC Santa Barbara researcher
Eric Masanet: 'We don't have the real ground-truth numbers on data-center water use to put everyone on the same page' — operators and utilities provide little public data
Water impact is hyper-local — no single national answer applies
Impact depends on local climate, regional water supply, electrical grid mix, and data center design; national aggregates are misleading for any specific community decision
Sam Altman called water concerns 'totally fake'; activists claim data centers will drain the Great Lakes
Both extreme positions exploit the lack of public data; The Atlantic finds the reality far more conditional and case-specific
Permitting decisions made without mandatory disclosure or measurement standards
County-level approvals and statewide moratoria alike are being made without reliable ground-truth data, as neither operators nor utilities face disclosure requirements
Covers the nuanced debate over AI data center water consumption, with national vs. local context, key case studies, and the transparency gap driving policy dysfunction
What This Means
The data center water debate is fundamentally a transparency problem — without mandatory disclosure from operators and utilities, neither regulators nor communities can make evidence-based decisions about new facilities. New York's first-in-the-nation statewide moratorium demonstrates that when reliable data is absent, policymakers default to precautionary restriction, which could slow AI infrastructure development even in water-rich regions. As AI model training and inference demand continues to grow, local water availability will become an increasingly critical factor in data center siting — and communities that lack the data to evaluate these risks are structurally disadvantaged when negotiating with large tech companies. The transparency gap means every community, drought-stressed or water-rich, is effectively flying blind when approving or blocking data center permits.