Does Liquid Cooling Use Water? Closed Loops, Dry Coolers and Cooling Towers

Sometimes. Liquid cooling moves heat into a sealed loop, filled once and recirculated. Water use depends on outdoor heat rejection: dry coolers evaporate none during normal operation; cooling towers evaporate continuously; adiabatic coolers use water only above a trigger temperature.

That distinction matters because “liquid cooled” describes how heat is captured from IT equipment. It does not describe how the facility finally releases that heat outdoors. A data center can have liquid-cooled racks and still consume substantial water at a cooling tower. It can also circulate water inside the building while using no recurring process water for cooling.

This article follows the heat from the rack to ambient air, explains what “closed loop” does and does not prove, and gives a practical test for dry, adiabatic and evaporative proposals. For a broader comparison of direct-to-chip, rear-door and immersion systems, start with our liquid-cooling design guide.

TL;DR: does liquid cooling use water?

• Liquid cooling describes heat capture at the IT equipment; it does not determine the site's water consumption.
• A closed technology loop is filled and recirculated, but it may still reject heat through a water-consuming cooling tower.
• Dry coolers use no recurring process water for normal cooling. Adiabatic coolers use water only when the wet stage runs. Cooling towers consume make-up water by design.
• Check both the liquid load and the residual room-air load; the wetter heat-rejection path sets the site answer.
• Ask for the outdoor device, temperatures, wet-mode hours and complete water balance before accepting a “waterless” claim.

Liquid cooled and water consuming can both be true

The easiest way to resolve the question is to separate three boundaries.

The heat path is usually: chip or server air → technology cooling system → CDU → facility water system → outdoor heat-rejection equipment → ambient air. A CDU transfers heat between systems; it does not make the heat disappear. The last device in that chain decides whether recurring cooling water is consumed.

Which liquid-cooling configurations consume water?

The heat-capture method and the heat-rejection method are separate design choices. The table below keeps them separate.

The final row is easy to miss. Cold plates may remove most processor heat while memory, networking, storage, power equipment and support spaces continue to reject heat to air. If that air system uses chilled water from a water-cooled chiller and tower, the site still consumes water. Ask for both heat paths, not only the GPU loop.

What does “closed-loop liquid cooling” actually mean?

A closed loop recirculates its working fluid. Depending on the equipment and exposure conditions, a technology or facility loop may contain treated water, aqueous propylene glycol, aqueous ethylene glycol or another approved coolant. Glycol lowers the freezing point for systems exposed to sub-zero conditions, but it also changes heat-transfer properties, viscosity, pump duty and materials compatibility. The fluid and concentration must therefore follow the rack, CDU and heat-rejection vendors' requirements.

The loop is charged during commissioning and then topped up after leaks, venting or service. That is inventory, not the same thing as continuous evaporative consumption.

But there may be more than one loop. A liquid-to-liquid CDU normally separates the IT-side coolant from the facility water system. That facility loop may then connect to a dry cooler, an adiabatic cooler, a cooling tower, a chiller or a heat-reuse interface. Saying “the loop is closed” is incomplete until the supplier says which loop and names the outdoor device.

The US Department of Energy’s data-center cooling-water guidance illustrates this directly: heat can pass from IT equipment through closed water loops and still reach an open cooling tower as the final stage. At that tower, make-up water replaces evaporation, blowdown and drift.

Dry coolers, adiabatic coolers and cooling towers

Dry cooler: no recurring evaporation

A dry cooler passes outdoor air over a sealed coil. The facility fluid stays inside the coil, so normal operation consumes no process water for cooling. Initial fill and maintenance top-up still exist. Other site uses—such as domestic water, cleaning or fire-system testing—also remain outside this claim.

The constraint is temperature. A dry cooler cannot cool the fluid below outdoor dry-bulb temperature plus the heat exchanger’s approach. It is therefore the warmest option, not automatically the easiest one.

Adiabatic or hybrid cooler: dry until the wet stage starts

An adiabatic cooler adds a wetted pad or spray when ambient conditions exceed a control threshold. Below that trigger it behaves like a dry cooler. Above it, evaporation lowers the air temperature entering the coil and water use begins.

“Near-zero water” is not a useful design claim on its own. Ask for the trigger condition, the predicted annual hours in wet mode using the site’s weather data, and the spray rate at design and part load. Without all three, annual consumption cannot be checked.

Cooling tower: recurring make-up by design

A cooling tower intentionally evaporates water to reject heat. It also discharges blowdown to control dissolved solids, while drift carries a much smaller amount away as droplets. The tower’s make-up-water requirement is therefore evaporation plus blowdown, drift and any other losses.

Cooling towers can provide lower fluid temperatures and reduce fan or compressor energy under the right conditions. Water use is not evidence of a poor design by itself; it is a trade-off that must fit the site’s water source, discharge conditions, climate and operating priorities.

When can a liquid-cooled data center run dry?

Dry rejection works when the IT equipment can accept facility water warm enough to stay above the site’s design dry-bulb plus approach. ASHRAE’s official 2021 thermal-guidelines reference card lists liquid-cooling classes W17, W27, W32, W40, W45 and W+, based on maximum facility-water supply temperature. The class name is not a guarantee of dry operation; the site and equipment still have to match.

Consider a simplified check. At a 30°C design dry-bulb and a 6 K approach, the dry cooler could supply fluid at about 36°C, which calls for equipment compatible with at least a W40-type supply limit. At 45°C ambient with the same approach, leaving fluid rises to roughly 51°C—above W45 and into W+ territory. An adiabatic stage or mechanical cooling may then be necessary.

This calculation must be run twice: once for the liquid loop and once for the residual-air system. The site’s answer is the wetter of the two. Hot-climate projects should also examine the broader energy-water choice in our guide to data-center cooling in extreme heat.

How much water does a liquid-cooling data center use?

A calculation needs a heat-rejection boundary, operating hours, latent fraction and cycles of concentration. It should not start with the common “3 gpm per ton” condenser-water convention unless the stated heat load includes chiller compressor work.

For an illustrative 1 MW IT load operating at full load for 8,760 hours, assume the outdoor loop rejects 1.05 MW of heat, including a 5% allowance for pump work and other heat on the same path. With a latent fraction of 0.85 and latent heat of vaporization of 2,430 kJ/kg, annual evaporation is about 11,580 m³. At four cycles of concentration, adding the combined blowdown-and-drift allowance implicit in the cycles formula brings tower make-up to about 15,440 m³ per year, or a cooling-only WUE of about 1.76 L/kWh. The selected tower's drift rating is needed to split the non-evaporative volume.

That is a derived example, not a project forecast. The physical ceiling is 1.481 litres of evaporation per kWh of heat rejected if every joule leaves through evaporation. A published annual evaporation figure above that ceiling has the wrong boundary or arithmetic. Load factor, economizer hours, climate, water chemistry, drift, treatment and the actual heat-rejection multiplier change the site result.

For annual volume, peak make-up and storage calculations, use the AI data-center water-budget guide. This guide explains where water is consumed; the budget guide sizes the site volumes and supply resilience.

Does zero cooling water mean zero site water?

No. A fully dry cooling system can still use water for humidification, reverse-osmosis reject and softener regeneration, cleaning, domestic facilities, fire-system fill and testing, landscaping, and construction flushing. “Zero water for cooling” is a bounded claim. “Zero-water data center” is usually not.

Microsoft’s December 2024 description of its next-generation zero-water-evaporation design makes this distinction: it says the cooling water circulates after initial fill, while water remains in use for restrooms and kitchens. The announcement also notes a nominal energy increase relative to evaporative designs and projected the first named sites to come online from late 2027. It documents a design direction and dated plan, not a completed-site result.

In the EU, Commission Delegated Regulation (EU) 2024/1364 applies to operators of data centers with installed IT power demand of at least 500 kW. The regulation requires total water input and potable water input to be reported separately, and calculates WUE as total water input divided by IT energy. The regulation establishes reporting; it does not set a universal WUE limit. The wider reporting framework belongs in our EU data-center regulations guide.

How to test a “waterless” cooling proposal

1. Name the outdoor device. Is it a dry cooler, air-cooled chiller, adiabatic cooler, open tower, water-cooled chiller with tower, or something else?
2. Define the claim boundary. Does “zero water” mean the IT loop, process cooling or every use across the site meter?
3. Check both heat paths. What captures the liquid load, and what serves the residual air load?
4. State the temperatures. Record the IT-side requirement, facility supply and return, design dry-bulb, coincident wet-bulb and heat-exchanger approach.
5. Quantify wet operation. For an adiabatic system, obtain the trigger, annual wet hours and spray rate. For a tower, obtain heat rejected, cycles of concentration, blowdown, drift and water quality.
6. Separate stocks from flows. Initial loop fill, storage volume, hourly make-up and annual consumption are different numbers.
7. Meter the streams. Cooling make-up, adiabatic feed, humidification, domestic and fire-system water should not disappear into one site total.

These questions also prevent a common procurement error: comparing a dry design and a wet design at different temperatures, loads or climate assumptions. A fair comparison holds the IT and service requirements constant, then changes the heat-rejection method.

How ModulEdge applies the distinction

ModulEdge treats the IT cooling interface, facility loop and outdoor rejection plant as separate parts of one thermal design. The MDC-2000 reference, for example, specifies a 41/51°C primary facility-water pair, direct liquid cooling plus a separate residual-air load, and dry heat rejection as the primary strategy. Tank-fed adiabatic assistance is enabled when ambient conditions require it.

That architecture supports operation without a permanent utility-water connection for the described cooling strategy. It does not justify one universal annual-water figure. The product reference's 100 m³ storage equivalent is storage, not yearly consumption.

Separately, ModulEdge's current planning estimate for one roughly 2 MW IT-load configuration in temperate European or US conditions is about 100 m³ of cooling water per year. It is not a guaranteed WUE or a value that transfers to every site. Very hot climates, such as Dubai, can require more adiabatic operating hours and higher annual water use. Final performance still requires site-specific climate, load and control simulation.

For a project review, share the proposed IT equipment, rack schedule, liquid and air heat split, site design conditions and water constraints. ModulEdge can then define the facility interfaces and rejection strategy around the actual configuration rather than an unbounded “closed-loop” label.

Frequently asked questions

Does direct-to-chip liquid cooling use water?

The IT-side loop is normally filled and recirculated rather than consumed. The site may still consume water if the facility rejects heat through a cooling tower or an adiabatic stage. Direct-to-chip describes heat capture, not the outdoor plant.

Do dry coolers use water?

A fully dry cooler consumes no recurring process water for cooling. Its closed loop still needs initial fill and occasional service top-up. A product sold as a dry cooler may include optional adiabatic spray, so confirm whether and when that wet stage operates.

Does a closed-loop data center need make-up water?

A closed IT or facility loop may need small top-ups after leaks, venting or maintenance. Continuous make-up belongs to an evaporative stage such as a cooling tower. Ask which loop is closed and what equipment rejects heat outdoors.

Can liquid cooling reduce both PUE and WUE?

It can, but neither result is automatic. PUE depends on facility energy, while WUE depends on the reporting boundary and water input. Site climate, temperatures, pumps, fans, compressors and the heat-rejection method determine the outcome. See the PUE guide and AI water-use guide for the two accounting views.

What should I ask when a supplier claims “zero water”?

Ask for the boundary, the final heat-rejection device, the residual-air system, and the site’s complete water balance. If the design is adiabatic, also ask for the trigger temperature, annual wet-mode hours and spray rate.

Does a liquid-to-liquid CDU consume water?

A CDU normally recirculates coolant on both sides of its heat exchanger and does not intentionally consume it. Small top-ups may follow service or leakage. Recurring process-water consumption appears downstream if the facility loop rejects heat through an adiabatic stage or cooling tower.

Is adiabatic cooling considered waterless?

No. An adiabatic cooler may run dry for much of the year, but it consumes water whenever pads or sprays operate. A meaningful proposal states the trigger condition, expected annual wet hours, peak spray rate, water quality and storage or supply requirement.

How much cooling water does ModulEdge estimate for a roughly 2 MW configuration?

For early planning, ModulEdge currently estimates about 100 m³/year in temperate European or US conditions for one roughly 2 MW IT-load configuration. Hotter climates can require more water. This estimate is not a universal product rating and is separate from the MDC-2000 reference's 100 m³ water-storage equivalent.