Is it worth evaporating water to cool data centers in Spain? A reading of the water, thermal and electricity evidence.

Key findings

• The strongest quantitative case for evaporation is about energy: available estimates place the reduction in peak cooling power in a 10%–35% range versus fully dry systems, although the upper end comes from a synthesis that has not been peer-reviewed (Han et al., 2026; Google, 2026).
• That saving can require large-scale water infrastructure: an order-of-magnitude example for 100 MW of IT load works out to 30 MW less peak power in exchange for 1.9–9.5 million liters per day of water capacity (Han et al., 2026).
• Spain's exposure is not theoretical: the average annual temperature has risen 1.75 °C since 1961, and June 2026 combined a +3.2 °C anomaly with precipitation equal to 39% of normal (AEMET, 2026; AEMET, 2026).
• The climate argument for saving every kilowatt is less conclusive in Spain than on fossil-heavy grids: in 2025, 75.5% of generation was free of CO₂-equivalent emissions, and solar supplied 29.4% in June 2026 (Red Eléctrica, 2026; Red Eléctrica, 2026).
• Regulatory pressure already puts water and electricity in the same assessment: the Spanish draft requires demonstrating performance within the top 15% on both indicators, while the EU moves toward a rating scheme and minimum standards (Comisión Europea, 2026; MITECO, 2025; Unión Europea, 2024).

Why is evaporative cooling technically attractive?

Evaporation harnesses the latent heat of the phase change: as it evaporates, water absorbs a large amount of energy and allows the rejection temperature to approach the wet-bulb temperature, reducing the thermal gap the mechanical system has to overcome (Han et al., 2026; Microsoft, 2026). Under favorable humidity and temperature conditions, that smaller thermal lift reduces the work done by compressors or fans and explains the continued presence of evaporative systems in efficiency-oriented designs (Han et al., 2026; Microsoft, 2026).

The case in favor grows stronger when water is abundant, competition between uses is low and marginal electricity comes from fossil fuels. The IEA estimates that coal and gas together supply around 56% of the electricity consumed by data centers globally; in that context, reducing cooling can avoid emissions if it displaces marginal fossil generation (IEA, 2025).

Hybrid systems limit evaporation to the most demanding hours and can preserve part of the electrical benefit with lower annual consumption (Amazon, 2026; Microsoft, 2026). Amazon states that it uses outside air roughly 90% of the time, and Microsoft describes evaporative assistance above 29.5 °C in certain climates; both figures come from the companies themselves and do not constitute independent verification (Amazon, 2026; Microsoft, 2026).

How much electricity can it save, and when does that saving matter?

The sources examined do not support a universal figure. Google publishes an approximate 10% reduction versus air cooling, while Han and co-authors synthesize disclosures placing the peak-power reduction between 10% and 35% versus equivalent dry solutions; the first source is an interested party and the second is a preprint (Google, 2026; Han et al., 2026).

The benefit can matter more at the critical hour than in the annual average, because grid connections, transformers and backup are sized to withstand the peak. In a reference example of 100 MW of IT load, a PUE improvement of 0.15 cuts the facility's estimated maximum demand by 30 MW (Han et al., 2026).

That example assumes a 50% utilization factor, a specific PUE reduction and US infrastructure costs; it is neither a Spanish measurement nor a rule applicable to any design (Han et al., 2026). Its purpose is to show the order of magnitude of the trade-off, because climate, density, supply temperatures and architecture can shift the result substantially (Han et al., 2026).

How much water turns that saving into an operational dependency?

In the same example, sustaining the 30 MW saving for a 100 MW IT load requires between 0.5 and 2.5 million gallons per day, equivalent to roughly 1.9–9.5 million liters a day (Han et al., 2026). The range represents peak supply capacity, not average annual consumption, and excludes extreme weather conditions (Han et al., 2026).

The difference between annual volume and peak capacity is material because water networks are sized to meet maximums, not just averages. Evaporative demand grows with temperature, so the greatest cooling need can coincide with the greatest strain on urban supply; Han and co-authors describe this asymmetry as a "big pipe" problem even when average consumption seems moderate (Han et al., 2026).

The local burden also depends on the size of the host system. Akinade and co-authors estimate that ten US projects would represent between 0.2% and 134% of their local networks' capacity, a spread that cannot be extrapolated directly to Spain, but which does show that a national percentage can hide a municipal risk (Akinade et al., 2026).

What do PUE and WUE measure, and what do they leave out?

PUE relates the facility's total energy to the energy delivered to IT equipment, while WUE relates operational water use to that same IT energy (Comisión Europea, 2026; Unión Europea, 2024). European regulation requires reporting both types of indicator for centers with significant demand, but neither metric on its own describes the full impact of a facility (Comisión Europea, 2026; Unión Europea, 2024).

A lower PUE can coexist with higher water consumption when electrical efficiency is obtained through evaporation (Privette, Barros y Cai, 2026; Amazon, 2026). A low annual WUE can also hide a peak concentrated in a few weeks: Amazon publishes a WUE of 0.12 L/kWh for 2025 and notes that it uses water mainly during the hottest days, so average and maximum answer different questions (Amazon, 2026).

A useful comparison must include annual energy, peak power, annual water, maximum daily capacity, water source and basin stress. Privette, Barros and Cai conclude that the lack of local disclosure prevents rigorous planning and that an impact can be significant even when its aggregate percentage looks small; that lack of transparency is the main limit of current comparisons (Privette, Barros y Cai, 2026).

Why do place and season change the outcome?

The impact of a liter changes with the basin and the month. Talukder and co-authors analyze 4,147 US sites and estimate that nearly 75% experience at least one month of water stress per year, while 12% remain under stress all year round (Talukder et al., 2026).

The same research warns that optimizing volume alone can misrank the alternatives, because shifting load to a region with a lower WUE can increase impact when the water or electricity comes from more stressed systems (Talukder et al., 2026). Their framework reduces stress-adjusted consumption by up to 25% through spatial and temporal scheduling, attributing to location and timing a weight comparable to that of the technology (Talukder et al., 2026).

Reclaimed water can reduce competition with drinking water, but it requires a stable flow, treatment, storage and distribution (Talukder et al., 2026; Amazon, 2026). In arid areas, rainwater harvesting has diminishing returns and reclaimed water retains alternative uses; its opportunity cost therefore remains local and does not disappear because the source carries the "reclaimed" label (Talukder et al., 2026; Amazon, 2026).

What do Spain's recent heat and drought show?

AEMET puts the increase in Spain's average annual temperature at 1.75 °C since 1961 and ranks 2025 as the third-warmest year in the series (AEMET, 2026). The twelve warmest years all belong to the 21st century, so a facility's thermal assessment must respond to an operational trend, not an isolated anomaly (AEMET, 2026).

June 2026 combined the two conditions that most stress an evaporative system: it was the second-warmest and third-driest June since 1961, with an anomaly of +3.2 °C and peninsular precipitation of 12.4 mm, equal to 39% of the normal value (AEMET, 2026).

Scarcity is neither uniform nor permanent across the country, but the EEA identifies Spain among the states that suffer scarcity especially in spring and summer (EEA, 2025). MITECO publishes monthly maps and special plans by river-basin district, confirming that risk must be assessed by basin and season; that temporal overlap with cooling demand is the variable that matters for design (MITECO, 2026).

Does Spain's electricity weaken the climate case for evaporating?

In 2025, renewables generated 55.5% of Spain's electricity and 75.5% of output was free of CO₂-equivalent emissions; solar photovoltaic capacity approached 50 GW installed and accounted for 18.4% of annual generation (Red Eléctrica, 2026). This profile does not eliminate the value of saving electricity, but it does distinguish it from that of a grid dominated by coal or gas (Red Eléctrica, 2026; IEA, 2025).

The hourly overlap reinforces that difference. In June 2026, when heat drove up cooling demand, photovoltaics supplied 29.4% of generation and reached a daily maximum of 34.9%, while combined-cycle plants accounted for 15.65% of the monthly mix (Red Eléctrica, 2026). Not every additional kilowatt of dry cooling would therefore have come from fossil fuels (Red Eléctrica, 2026).

On a fossil grid, saving electricity can dominate the environmental balance; on a high-solar grid, the same saving keeps its economic and capacity value, but its carbon benefit can be smaller during the hours of peak irradiation (IEA, 2025; Red Eléctrica, 2026; Red Eléctrica, 2026). That change of context is the finding that organizes the Spanish case: water is consumed locally and consumptively, while the climate cost of marginal electricity varies by the hour.

Do reclaimed water and hybrid operation change the balance?

Hybrid operation reduces annual volume when it reserves evaporation for the hottest hours (Amazon, 2026). Amazon states that its centers use outside air roughly 90% of the time and water only during the hottest days; the figure comes from the company itself and does not constitute independent verification (Amazon, 2026).

Microsoft describes a similar logic in existing facilities, with evaporative assistance above 29.5 °C in certain climates and dry modes the rest of the time (Microsoft, 2026). The same company is moving new Spanish projects to closed loops without evaporation, a sign that the electrical benefit does not make water an unavoidable technical requirement (Microsoft, 2025; Microsoft, 2026).

Reclaimed water reduces pressure on drinking-water resources, but it requires treatment, storage and distribution, and the evaporated fraction does not return immediately to the local network (Privette, Barros y Cai, 2026; Talukder et al., 2026; Amazon, 2026). Its use can be defensible when there is a stable surplus and no environmental or urban destination of higher value; that condition must be demonstrated with basin data, not presumed from an administrative designation.

Where is European and Spanish regulation heading?

The European Union already requires data centers above the applicable thresholds to report energy and water-footprint indicators (Unión Europea, 2024; Comisión Europea, 2026). The 2024 Delegated Regulation harmonizes the reporting, and the Commission is using that basis to develop a comparative rating and study minimum performance standards (Unión Europea, 2024; Comisión Europea, 2026).

The 2025 Spanish draft requires reporting total and potable water and proposes that new projects demonstrate performance within the top 15% on electricity and water indicators (MITECO, 2025). The text remains a proposal under public consultation and does not constitute a ban on evaporation currently in force (MITECO, 2025).

European water-resilience monitoring maintains an aspirational target of improving efficiency by 10% by 2030 and applies the "Water Efficiency First" principle (Consejo de la Unión Europea, 2026). The regulatory sources examined contain no general Spanish ban on evaporative cooling, but they do describe a direction in which source, consumption, performance and water continuity will become increasingly comparable and enforceable (Comisión Europea, 2026; Consejo de la Unión Europea, 2026; MITECO, 2025; Unión Europea, 2024).

What does a non-evaporative architecture in Madrid demonstrate?

Microsoft has announced new centers for Spain with direct-to-chip cooling and closed loops without evaporation, and states that they will maintain thermal control with no operational water consumption for cooling (Microsoft, 2025; Microsoft, 2026). This is a corporate statement, not an independent measurement, but it shows that a major operator considers that architecture viable in the Spanish market.

MAD-NE uses Carrier chillers and direct expansion with N+1 redundancy and rejects heat without evaporative or adiabatic systems, according to operational information from ipcore (ipcore, 2026). The case demonstrates that a commercial data center can operate in Madrid without turning water into a thermal consumable, although it does not by itself determine the optimal PUE for every density or scale (ipcore, 2026).

Non-evaporative cooling may require more investment, more heat-exchange surface or more electricity during hot hours (Han et al., 2026). In Spain, those costs must be weighed against a +1.75 °C trend, seasonal scarcity, high solar output and regulation that is starting to assess water and electricity together; under that set of conditions, the dry alternative stops being a defensive exception and becomes the design baseline (AEMET, 2026; Red Eléctrica, 2026; Comisión Europea, 2026; MITECO, 2025).

Methodology and limitations

The analysis uses 20 sources: fifteen published or updated in 2026, four documents from 2024–2025 and one operational statement from ipcore. Official records, peer-reviewed research and recent academic work were prioritized; the figures from Google, Amazon, Microsoft and ipcore are identified as publications by interested parties.

The estimates of electrical savings and water capacity come in part from US preprints, not from controlled tests in Spain (Han et al., 2026; Akinade et al., 2026). They are used to establish mechanisms and orders of magnitude, not to attribute to any specific Spanish project a WUE, a PUE or a consumption that has not been measured.

The report does not build a full hourly model of a Spanish facility, has no marginal emissions by grid node and does not compare equivalent capital costs. The conclusion is therefore framed as a design presumption for new facilities, not as a verdict on every existing center or a substitute for a basin assessment.

Frequently asked questions

Sources

1. 01Google, 2026. “Google announces water stewardship commitments and initiatives.” Dato publicado por la propia compañía; la fuente es parte interesada en la afirmación.
2. 02Han, Y.; Li, P.; Wierman, A.; Ren, S., 2026. “Small Bottle, Big Pipe: Quantifying and Addressing the Impact of Data Centers on Public Water Systems.” arXiv:2603.02705.
3. 03Agencia Estatal de Meteorología, 2026. “El año 2025 volvió a batir récords de temperatura en España.” Informe sobre el estado del clima de España 2025.
4. 04Agencia Estatal de Meteorología, 2026. “Junio de 2026 fue el segundo más cálido y el tercero más seco de la serie histórica.”
5. 05Red Eléctrica, 2026. “La demanda de energía eléctrica de España desciende un 0,9% en junio.”
6. 06Red Eléctrica, 2026. “El sistema eléctrico español en 2025: aumenta la demanda, la generación y la potencia instalada.”
7. 07Comisión Europea, 2026. “Energy performance of data centres.” Base europea de reporte, esquema de calificación y trabajo sobre estándares mínimos.
8. 08Ministerio para la Transición Ecológica y el Reto Demográfico, 2025. “Proyecto de Real Decreto por el que se regula la eficiencia energética y la sostenibilidad para los centros de datos.” Texto sometido a audiencia pública.
9. 09Unión Europea, 2024. “Reglamento Delegado (UE) 2024/1364 relativo a la primera fase del establecimiento de un sistema común de la Unión para calificar centros de datos.”
10. 10Microsoft, 2026. “El recorrido de Microsoft durante dos décadas para reducir el uso de agua mientras impulsa su crecimiento.” Dato publicado por la propia compañía; la fuente es parte interesada en la afirmación.
11. 11International Energy Agency, 2025. “Energy and AI: Energy demand from AI.”
12. 12Amazon, 2026. “Amazon’s data centers are 7x more water-efficient than the industry average. Here’s how we do it.” Dato publicado por la propia compañía; la fuente es parte interesada en la afirmación.
13. 13Akinade, B. A.; Amanambu, A. C.; Frame, J. M.; Ren, S., 2026. “AI Data Centers and the Water Use Feedback Loop.” arXiv:2606.21760.
14. 14Privette, A. P.; Barros, A.; Cai, X., 2026. “Data Centers Water Footprint: The Need for More Transparency.” AGU Advances 7(2), e2025AV002140. DOI: 10.1029/2025AV002140.
15. 15Talukder, Z.; Rahim, I. B.; Sen Gupta, P.; Ren, S.; Islam, M. A., 2026. “Balancing Bits and Drops: Stress-Adjusted Water Management for Data Centers.” Proceedings of ACM e-Energy ’26. DOI: 10.1145/3744255.3811726.
16. 16European Environment Agency, 2025. “Water scarcity conditions in Europe.” Water Exploitation Index Plus indicator.
17. 17Ministerio para la Transición Ecológica y el Reto Demográfico, 2026. “Informes y mapas mensuales de sequía y escasez. Año 2026.”
18. 18Microsoft, 2025. “Los nuevos centros de datos de Microsoft en España no consumirán agua para su refrigeración.” Dato publicado por la propia compañía; la fuente es parte interesada en la afirmación.
19. 19Consejo de la Unión Europea, 2026. “Follow-up report on the implementation of the European Water Resilience Strategy.” Documento ST 10288/1/26 REV 1.
20. 20ipcore, 2026. “Arquitectura térmica pública de MAD-NE: chillers Carrier, expansión directa y redundancia N+1.” Información operativa revisada por ipcore; la organización es parte interesada en la descripción.