Water use has become a benchmark metric for data center operational efficiency and sustainability, as well as a frequent target of community criticism. Operators should track and optimize their water use to protect the environment and secure the facility cooling’s autonomy when its capacity relies on water. However, water use does not fit into a standard template — each data center location has a distinct water signature. Despite this, studies often generalize and flatten several important factors that contribute to the variation in daily and annual water use.

Legacy facilities, except those in arid climates or constrained urban environments, often consume large quantities of water by using open evaporative cooling towers to produce low-temperature chilled water. Conversely, new data centers can use little to no water for cooling by deploying state-of-the-art water-efficient, waterless and hybrid heat rejection systems, as detailed in this report. An accurate assessment of a data center's water use — and its effects on communities and the environment — must examine local restrictions on water use, competition for water, the watershed’s safe withdrawal rate, the cooling system type, and climatic conditions at a given location.

Water use and distribution

The equipment a data center uses to reject heat into the environment has the greatest influence on its water consumption. The facility’s water loop (the chilled water closed loop in Figure 1), which collects and transports heat from the IT space, typically interfaces with the heat rejection loop at chillers or heat exchangers. This setup transmits heat energy while keeping the water loops physically isolated.

An IT side water loop can absorb IT heat from the air through computer room air handlers (CRAHs) or rear door heat exchangers. In contrast, direct liquid cooled (DLC) IT transfers heat to facility water through a coolant distribution unit (CDU). Computer room air conditioners (CRACs) absorb heat from the air similarly to a CRAH, but they use a refrigerant rather than water to transport it away. If the heat collection system uses water or a water-glycol mixture, the system is closed and consumes little (if any) water — even when serving DLC equipment.

Figure 1 Heat rejection with open cooling tower and water-side economizer

Water is consumed only in the heat rejection loop (left in Figure 1). In open cooling towers, as well as evaporative and adiabatic coolers, water absorbs heat from the data center as it changes to vapor. This vapor, together with the waste heat, dissipates into the environment, while the heat rejection loop is continuously replenished from the main water supply.

More efficient evaporative and adiabatic cooling systems installed in the past five years have significantly reduced water use, in part due to the widespread adoption of higher supply temperature set points. When outside temperatures are low enough, these systems cool without evaporation (i.e., free cooling) to reject heat — and many can operate this way 50% to 80% or more of the time.

Where full free cooling is not practical, advanced heat rejection systems start to use water with precision. For example, spray or trickle media adiabatic systems cool down ambient air as it enters the system, helping to meet set point and capacity demand, or the heat exchanger is wetted for evaporative effect to deliver the same result. These processes can reduce water usage by half or more compared with conventional open systems for the same cooling capacity. In cooler climates, data center heat rejection may require water only on the hottest days — sometimes for just 4 to 12 hours per day, over 5 to 20 days per year.

The water use profile of a data center is determined by local climatic conditions, the amount of available and captured free cooling, and the operating characteristics of its heat rejection system. In water-rich locations, a water-cooled heat rejection system typically offers the most energy-efficient cooling system. However, in areas with water stress or scarcity, data center designs should prioritize dry free cooling and mechanical cooling modes, using minimal (e.g., peak shaving) to no water.

Water quantity in context

Organizations evaluating the impact of a data center’s water consumption must measure the facility’s projected water use against the safe withdrawal rate of the watershed or groundwater basin supplying it. In temperate or tropical climates with abundant water, a watershed may be able to support many high-demand industrial and residential users. However, in arid and semi-arid regions with many water supply demands, there may not be enough available water to support cooling systems with open towers or adiabatic cooling. The environmental consequences of a data center’s water use at a location will depend on the percentage of the available supply it consumes and the needs of other nearby water consumers.

When data center cooling consumes water, the (often large) quantities require context. A typical US household uses 110,000 US gallons (420 m³) of water annually. One million US gallons (3,800 m³) of water can meet the needs of 10 US and 20 EU households for a year (see Table 1). A small city in the EU with 50,000 people (assuming four people per household) uses 700 million US gallons (2.7 million m³) of water annually.

In comparison, most data centers use significantly less water. In the 2024 Uptime Institute Cooling System survey, only 14% of respondents with water-cooled data centers used more than 16 million US gallons (60,000 m³) per year (see Figure 2). This suggests that few, if any, data centers using water for heat rejection consume amounts comparable to a city of 50,000 people.

Table 1 Water consumption in terms of average household consumption

Water use also varies by season. Depending on the system design and the climate type, economizers can often provide 30% to 80% of the cooling. One data center in Iceland has run for three years using only dry coolers, with an adiabatic spray system available for higher outside temperatures. In temperate climates, cooling systems will use large quantities of water on hot summer days but much less during cooler periods. Data center designers should assess the ability of the water supply system to meet demand during the hottest anticipated weather, as periods of high water use can coincide with lower water availability during the summer months.

Selecting heat rejection for new builds

Uptime Intelligence’s survey also indicates that water use is weakly correlated with data center size (see Figure 2), which is due to vast differences in water needs across heat rejection modes. In cooler climates, a large 25 MW data center using a closed-loop adiabatic evaporative system may rely on free cooling for 90% to 95% of the year (neither water use nor mechanical refrigeration), consuming only a small amount of water per megawatt of IT capacity. In contrast, a smaller 8 MW data center with an open evaporative cooling system, which can use free cooling for only 30% of the year, will have a higher water consumption per megawatt of IT capacity.

Figure 2 Water consumption for water-cooled systems

Data center organizations should design new facilities to reduce and minimize water use. At the same time, they need to consider the trade-offs, such as energy efficiency, and select a cooling system that is appropriate for local concerns. In water-rich areas, designers should prioritize water-efficient adiabatic spray and trickle media systems with integrated economizers to maximize energy efficiency while minimizing water use per megawatt.

New, improved designs in dry or waterless cooling technologies enable operators to economically reject heat with minimal water and low energy use in arid, semi-arid and temperate climates. Pumped refrigerant systems can cool a data center with a 10-year total cost of ownership (TCO) that is comparable to an evaporative-based cooling system. Historically, dry or direct expansion refrigerant-based systems had a much higher TCO due to their greater energy consumption.

Hybrid systems can also provide energy-efficient cooling with minimal water use, consuming water for trim cooling for 100 to 300 hours per year (depending on climate). In normal operation, these systems typically use direct expansion cooling with an integrated economizer. Adiabatic and/or pumped refrigerant systems provide trim cooling at ambient temperatures of 20°C (68°F) or higher.

More energy-efficient dry cooling systems have prompted Microsoft and several major colocation providers (e.g., Compass and Novva) to commit to waterless cooling for their data centers. This approach is likely to become predominant for new data centers in many locations, although operators will likely prefer adiabatic water-based heat rejection systems in areas with abundant water availability.

Communicating consumption clearly

Interpreting data center water use and its effects requires more detail than a simple consumption figure. The impact depends on the data center’s location, the cooling system type and operational practices. Raw survey data shows wide variation in water consumption, even among data centers of comparable size — underscoring the importance of considering these variables.

Data center operators should track and report their water consumption, consider water availability and use in their siting and design decisions, and then communicate and provide context for their annual water use. Uptime Institute recommends that operators publish a yearly report showing how the data center uses water, the total consumption, and the average, minimum and maximum daily use.

Minimizing water use should be the highest priority in water-stressed regions, where the risk of adversely affecting supply is greater. Data centers that do not stress their watershed’s capacity or affect other users can make greater use of evaporative methods for energy-efficient heat rejection. However, designers should still use the most efficient adiabatic designs where appropriate.

Studies that report a general water consumption figure for AI training, AI inference, or a standard Google search are not meaningful. This is because the actual water consumption will depend on factors such as the cooling system type, climatic conditions, and time of day and year at the data center where the operation is being executed. Data center operators can offer better insights on the water impact of their operations by publicly reporting the water signature of their individual data centers.

Other related reports published by Uptime Institute include:
Cooling systems: balancing cost, energy and water use
Water is a local issue: site selection and facility design