The Hidden Water Geography of U.S. Hyperscale Data Centers in the AI Era

Technical Summary: The Hidden Water Geography of U.S. Hyperscale Data Centers in the AI Era

Problem Statement
As hyperscale data centers expand to support cloud services and artificial intelligence workloads, their water consumption has become a critical concern. However, existing assessments routinely report water use as a single, aggregated footprint. This approach obscures the physical reality that data center water consumption occurs through two distinct pathways:

1. Scope 1 (Direct Cooling): Water consumed on-site for thermal management, primarily via evaporative cooling.
2. Scope 2 (Electricity-Related): Water consumed off-site within the power generation system (e.g., thermoelectric cooling at fossil-fuel and nuclear plants, and reservoir evaporation associated with hydropower) to generate the electricity the facility uses.

Collapsing these pathways into a single metric hides the geographic divergence of risks. A facility may appear water-efficient locally but rely on a water-intensive grid, or conversely, impose direct cooling burdens in a stressed basin while drawing from a low-water grid. Current national or state-level totals are insufficient for guiding specific governance responses because they fail to identify where water is consumed and which institutions hold the authority to reduce it.

Methodology
The authors developed a facility-resolved national framework to estimate annual operational water consumption for 472 U.S. hyperscale data centers, separating the two pathways.

• Data Integration: Facility locations were linked to three spatial layers via geospatial joins:
  • Balancing Authorities (BAs): Operational units of the U.S. electricity grid (from eGRID) to attribute Scope 2 water.
  • Hydrologic Basins: Level-6 polygons (from HydroBASINS) to attribute Scope 1 water.
  • Water Stress: Aqueduct Water Risk Atlas 4.0 baseline stress indicators (0–5 scale).
• Scope 1 Estimation: Calculated using Water Usage Effectiveness (WUE), defined as liters of site water per kWh of IT electricity (ISO/IEC 30134-9). Since facility-specific WUE is rarely disclosed, the study applied scenario-based WUE values derived from literature, adjusted for facility Power Usage Effectiveness (PUE) and utilization factors.
• Scope 2 Estimation: Calculated by multiplying facility electricity demand by the water-consumption intensity of the hosting balancing authority. This intensity was derived from the region's generation mix (fuel shares) and technology-specific operational water factors (e.g., coal, gas, nuclear, hydro, wind, solar).
• Scenarios: The analysis included a baseline scenario, an efficiency scenario, a high-load scenario, and a "no-hydro" sensitivity (setting hydropower reservoir evaporation factors to zero) to test the robustness of the attribution method.

Key Results

• Total Consumption and Pathway Split: Under baseline assumptions, the 472 facilities consume approximately 300 GL yr⁻¹ (range 205–451 GL yr⁻¹ across scenarios).
  • Scope 2 (Electricity-related) accounts for roughly 75% of the total (~226 GL yr⁻¹).
  • Scope 1 (Direct cooling) accounts for roughly 25% (~74 GL yr⁻¹).
  • Even in a conservative "no-hydro" sensitivity where hydropower attribution is removed, Scope 2 remains the dominant component (63% of the total).
• Divergent Geographies of Burden:
  • Scope 1 Hotspots: Direct cooling burdens concentrate in hydrologic basins in the western and south-central U.S. where cooling demand overlaps with high water stress (Aqueduct score ≥3).
  • Scope 2 Hotspots: Electricity-related burdens concentrate in a few eastern grid regions characterized by fossil-heavy supply mixes. Specifically, just 3 of 24 hosting balancing authorities account for 59% of the total Scope 2 water.
• Concentration Differences: Electricity-related water is significantly more spatially concentrated than direct cooling water. The top 3 balancing authorities hold 59% of Scope 2 water, whereas the top 16 hydrologic basins hold only 51% of Scope 1 water.
• Actionable Leverage: The regions where the greatest potential reductions can be achieved differ by pathway.
  • Scope 1 reductions depend on local cooling design, water sourcing, and drought planning in stressed basins.
  • Scope 2 reductions depend on regional grid planning, generation mix changes, and procurement in specific balancing authorities (e.g., PJM, BPA, PACW).

Significance and Claims
The paper argues that separating water pathways is essential for effective governance because the two problems require different decision-makers and solutions:

1. Governance Mismatch: Current permitting and environmental reviews often focus solely on on-site (Scope 1) water use. However, for many facilities, the larger operational burden lies in the electricity system (Scope 2). For example, Virginia, the largest data center market, faces a massive Scope 2 burden tied to its grid supply, which local cooling permits cannot address. Conversely, western states with Scope 1-dominant profiles require locally tailored cooling strategies.
2. Targeted Intervention: Because Scope 2 water is highly concentrated in a small number of grid regions, decisions regarding generation mix and transmission in those specific areas can disproportionately reduce the national data center water footprint.
3. Methodological Clarity: The study provides a framework to distinguish between local hydrologic risks and regional grid risks, preventing the conflation of problems that require distinct policy responses.

The authors note limitations, including the use of annual averages (ignoring seasonal droughts), the reliance on scenario-based WUE rather than measured facility data, and the contested nature of hydropower reservoir evaporation attribution. However, they assert that the core geographic finding—that cooling-side and electricity-side water concerns concentrate in different places—remains robust across all tested scenarios.