MOZOM-analyse

Data centers choose water cooling to cut costs

RIVM, Rijkswaterstaat, European Commission, Berkeley Lab, IEA, ASHRAE and scientific research Technology 6 augustus 2026 om 12:48

Type: Analyse Author: Paul Giezen Published: 6 augustus 2026 om 12:48 Updated: 6 Aug 2026, 16:49 Report a correction

Rapid action needed to prevent a drinking-water shortage by 2030 The policy looks like a double standard: citizens are visibly urged to use less water, while water-cooled data centers face far less prominent pressure over their consumption. Data centers are not the country's largest drinking-water users, but a single facility can be a major continuous local user. Where scarce drinking water is evaporated even though workable dry or circular alternatives exist, calling it waste is defensible.

Realistic editorial caricature of a data center diverting a lot of water to cooling towers while dry coolers are available and households receive little water

Source: RIVM, Rijkswaterstaat, European Commission, Berkeley Lab, IEA, ASHRAE and scientific research
MOZOM headline: Data centers choose water cooling to cut costs
Original headline: Water is often not a technical necessity for data centers, but an economic choice with local consequences
Author: Paul Giezen
Date: 6 augustus 2026 om 12:48
Subject: MOZOM investigates worldwide why data centers choose evaporative cooling, which dry and circular alternatives exist and who bears the water, energy and environmental risks of that choice.

Summary of the original report

The statement that data centers need water to exist is too absolute. Servers need cooling; the heat does not necessarily have to be released to the outside air through evaporation of drinking water. Conventional air-cooled chillers and dry coolers use little or no cooling water on site, but often require more electricity, more cooling surface and a higher investment during warm hours. Evaporative cooling can be more energy efficient and cheaper, especially in hot and dry climates, but it uses water because some evaporates and some is discharged as mineral and chemical-rich drain water. New hot water direct-to-chip systems can dissipate heat at a higher temperature and therefore work with dry coolers. A recent life cycle analysis based on Microsoft data calculated 31 to 52 percent less blue water for cold plates and immersion cooling than for conventional air cooling, in addition to lower energy and greenhouse gas impacts. That is promising, but not a universal outcome: climate, power mix, server density, construction costs, fluids used and the way in which heat is removed outside the building remain decisive. Even a closed liquid loop in the server room can still lead to an evaporative cooling tower outside.

Global evidence: what do we know, and what remains local?

The units, years and definitions differ. Water extraction is not the same as consumption, direct cooling water is not the same as indirect water for power, and company claims are not independent measurements.

Case or source	Verified finding	Meaning for the cooling choice	Limit of the conclusion
United States, Berkeley Lab	For 2023, approximately 66 billion liters of direct water and almost 800 billion liters of indirect electricity water were estimated. Hyperscale and colocation centers accounted for 84 percent of direct usage.	Less cooling water on site helps, but extra electricity for dry cooling can shift water use to power stations and other power sources.	They are model estimates. The outcome depends on power mix, location, cooling system and definitions.
The Netherlands, RIVM and Rijkswaterstaat	Without additional measures, the Netherlands will have a drinking water shortage by 2030; According to national figures, data centers used a maximum of approximately 1 million cubic meters of drinking water per year.	A small national percentage may still conflict locally with security of supply. According to Rijkswaterstaat, Vitens no longer wants to allow new data center connections.	The national CBS category includes more than just data centers and says little about individual locations or summer peaks.
Water stress in the United States	A 2021 peer-reviewed study estimated that one-fifth of direct water footprint volume came from areas with medium or high water stress.	Location choice is just as important as an average economy score. An efficient installation can still cause extra risk in the wrong catchment area.	The study used 2018 data and modeled infrastructure; the fleet and cooling systems are changing rapidly.
The Dalles, Oregon	Publicly released figures put Google's 2024 consumption at 461.1 million gallons per year, about 30 percent of the city's average daily demand.	Here, data center water is not an abstract world total but a substantial item within a municipal system.	The percentage applies to this city and this year. It should not be generalized to every Google data center.
Chile and Uruguay	In Chile, a Google design was changed from cooling towers to air-cooled condensers after water and climate concerns. The project in Uruguay uses air cooling and closed loops, according to the environmental file.	Dry solutions exist on an industrial scale and can virtually avoid ground or drinking water for cooling.	Air cooling can require more electricity, space and investment. Projects and climates are not comparable one to one.
Singapore	The government directs major water users towards measurement and efficiency; Much data center cooling uses NEWater, purified wastewater, instead of regular fresh drinking water.	Non-potable sources can reduce competition with domestic water without directly depreciating cooling towers.	Reused water also requires treatment, pipes and energy and remains part of a local water system.
New liquid cooling	A Nature life cycle analysis calculated 31 to 52 percent less blue water and 15 to 20 percent less energy for cold plates and immersion than conventional air cooling.	Higher cooling water temperatures can enable dry heat removal and reduce the perceived fixed choice between water and power.	The model used primary data from Microsoft and additional assumptions; fluids, hardware and local conditions determine the actual outcome.
Microsoft, AWS and Google	Microsoft says it will build new designs without evaporative water; AWS reports increased recycled water and Google reports annual withdrawal, consumption and water replenishment.	The sector recognizes the risk and at the same time shows that other cooling paths are technically and commercially feasible.	These are our own company reports. Without location data, uniform control and the same definitions, mutual comparisons remain limited.

The cooling method determines who gets the bill

An evaporative cooling tower uses the heat of vaporization of water. This means it can remove a lot of heat with less compressor work. The US Department of Energy explicitly calls this a trade-off: a dry cooler minimizes on-site water use, but may have an energy penalty; evaporation reduces energy use at the expense of water. Not only the visible vapor disappears. Minerals concentrate in the remaining water, requiring periodic flushing and the use of chemicals against limescale, corrosion and microbiological growth. Better management helps: according to the ministry, increasing from three to six concentration cycles can reduce the required replenishment by approximately 20 percent and the discharge water by 50 percent. Free outdoor air is cheap and economical in cool climates, but requires filtration, moisture control and a reserve for warm hours. Dry chillers work everywhere but become less efficient on hot days. Hybrid installations use air most of the year and only use a limited amount of water during peak heat. Direct-to-chip and immersion release heat more efficiently from densely packed AI hardware. ASHRAE describes how a hot water loop around 45 degrees Celsius with dry coolers can operate virtually without cooling water and sometimes even make chillers unnecessary. Underground thermal storage can shift heat between seasons; Under favorable geological conditions, an NREL model resulted in lower costs and emissions than a reference system. That is not yet a universal construction recipe, but it debunks the claim that large-scale digital computing power automatically requires large-scale drinking water evaporation.

A small percentage can hide a large pipe

The sector can provide a favorable but incomplete picture with three metrics at once: a small share of national annual consumption, a declining global WUE and a large water replenishment program. Any number can be correct. None of the three automatically tells you how much drinking water a specific location uses on a hot day, which river basin it comes from, how much disappears as vapor and how much water the extra electricity costs elsewhere. The opposite framing is equally misleading: presenting trillions of liters worldwide as if they all pass through cooling towers as local drinking water. Many large estimates include indirect water in electricity production. Nevertheless, transparency remains limited. The European reporting regulation does require large data centers to provide data on total and drinking water, but public EU results are aggregated and location data is in principle confidential. Investigate Europe and Le Monde documented that Microsoft and trade association DigitalEurope requested protection of individual data as business confidential during the preparation. The final legal text contains such a confidentiality rule. This does not prove conspiracy or illegal influence, but it does prove that information asymmetry was not accidentally left out of the policy debate.

Water, power, health and ecosystems

The environmental impact starts at the source. Groundwater extraction can put pressure on nature, agriculture and drinking water reserves during dry periods; surface water extraction can have a heavy ecological impact, especially during low discharge. Evaporative consumption does not immediately return to the same local water system. Drainage water contains concentrated salts and treatment agents and must be disposed of appropriately in accordance with permit requirements. Poorly managed evaporative cooling towers can also spread Legionella; that is a known management risk of cooling towers in general and not a unique data center scandal. Dry cooling prevents many of these local water problems, but can cause more electricity and therefore more emissions and indirect water use during hot hours. Therefore, a waterless installation is not automatically the greenest installation. The best solution combines a suitable location, economical hardware, warmer liquid cooling, dry or hybrid heat dissipation, reused water where appropriate, clean power and utilization of residual heat. At the same time, data centers deliver real benefits: they support cloud software, payments, healthcare, research, government services and AI. A modern hyperscale center can also calculate more efficiently than many old server rooms. The social question is therefore not whether digital infrastructure is allowed to exist, but whether the private cost benefit is fairly weighed against the local water and grid space that society makes available.

Possible message behind the news

The uncomfortable message is not that every data center owner is emptying the population's tap. It is that without firm public conditions the market will not automatically choose the socially best cooling method. A cubic meter is charged a rate on the company balance sheet, but drought risk, loss of nature, extra extraction and loss of public trust are not automatically included. Where these external costs are absent, water-intensive cooling can be rational for the company and yet unfavorable for the environment.

Neutral conclusion

Data centers do not have to be cooled with scarce drinking water as standard. The technology for dry, hybrid, closed and circular alternatives exists and is already being built worldwide. At the same time, there is no universally free alternative: those who save water may need more power, space or investment, unless hot water liquid cooling and a smart overall design reduce that trade-off. Good policy should therefore not be based on slogans but on location data. Publish source, withdrawal, consumption and peak consumption per location; price water according to scarcity and infrastructure; demand a drought plan and independent monitoring; only release drinking water for cooling when reused water or dry technology is demonstrably unsuitable; and assess water, energy, carbon, chemicals and ecosystems together. Then water will not be banned, but also no longer treated as the cheap invisible raw material through which private savings can end up on the public account.