Zero Water Data Center Cooling: What It Really Means (and Where Liquid Cooling Fits)

TL;DR: A โ€œzero waterโ€ cooling claim usually means the site rejects heat without evaporating water. It leaves out the water in the cooling loop, the water the power plant uses, and the hottest days, when sites fall back on chillers or adiabatic spray. Liquid cooling helps because warmer coolant adds free-cooling hours. A narrow CDU approach temperature, such as the EcoCoreยฎ FCDโ€™s 2ยฐC at its lowest, lets warmer facility water do the work, so chillers run less.


Ever wondered how much of the conversation about data centers and water comes down to that one plume? A lot of it. When someone tells you their facility runs on โ€œzero waterโ€ cooling, they usually mean the plume is gone. No tower, no mist, no white cloud drifting over the parking lot. Sounds great, right? It mostly is. The phrase does get stretched, though, and once you know where the stretch marks are, youโ€™ll read every sustainability press release a little differently. (Sorry in advance.) Letโ€™s walk through it.

Why Do Data Centers Use So Much Water?

Evaporative cooling tower on a data center roof, the main source of data center water consumption

Every watt a server pulls ends up as heat. Every single one. A data center is, in a very real sense, an expensive space heater that also does math.

That heat has to leave the building somehow, and for decades the cheapest way to dump it was evaporation. Water soaks up a surprising amount of heat when it turns into vapor. You already know this from climbing out of a pool on a breezy day and shivering even though itโ€™s 85 degrees out. Cooling towers run the same trick at industrial scale. Warm water from the building trickles down through the tower, fans pull air across it, a slice of that water evaporates, and the rest comes out cooler.

It works beautifully. It also consumes water, permanently, by the truckload. In a dry region, or in a town thatโ€™s already arguing about its reservoir, that becomes a problem in a hurry.

What Does Zero Water Data Center Cooling Mean?

Aerial view of rooftop dry coolers and closed-loop cooling pipes used for zero water data center cooling

Most of the time, it means the site doesnโ€™t evaporate water to get rid of heat. Thatโ€™s the whole claim.

The industry even has a number for it. The Green Grid, an industry group, came up with a metric called water usage effectiveness, or WUE. Itโ€™s the liters of water a site uses for every kilowatt-hour of energy its IT equipment consumes. A site with cooling towers reports a WUE well above zero. A site that rejects heat with dry coolers (picture giant car radiators with fans on top) can report a number close to zero, because nothing evaporates.

Close to zero. Remember that phrase.

What Zero Water Cooling Claims Leave Out

Hereโ€™s where I get to be the annoying friend at the barbecue. A โ€œzero waterโ€ claim usually leaves out three things. Theyโ€™re all ordinary physics and accounting, and none of them should scare you.

  • The loop still has water in it. A closed cooling loop gets filled once and topped off now and then. Thatโ€™s a tiny amount next to a tower, and it still counts. Anyone whoโ€™s ever bled air out of a home boiler system knows loops get thirsty in small, irritating ways.
  • The power plant has a water bill too. Electricity comes from somewhere. Plenty of the grid still runs on thermal power plants that use water for their own cooling. A site that swaps its cooling tower for more hours of electric chillers can shrink its own water meter while quietly pushing water use upstream to the utility. WUE only counts what crosses the siteโ€™s fence.
  • Hot days are hard. A dry cooler can only bring water down to a few degrees above the outdoor air temperature. On a mild night, thatโ€™s fantastic. On a 104-degree afternoon in Phoenix, the outside air is hotter than a lot of the water youโ€™re trying to cool. So the site has two options. It can fire up mechanical chillers, which burn more electricity. Or it can mist water over the coils for a few hours. That second one is called adiabatic assist, and yes, it evaporates water.

Put those three together and you get the honest version. Zero water cooling usually trades water for electricity, especially during the hottest hours of the year. Sometimes that trade is absolutely worth making. Sometimes it moves the problem somewhere you canโ€™t see it. Think of it like diet soda. Fewer calories on the front of the can, and you still might want to read the back.

Does that make sense? Good, because the next part is where it gets interesting.

Water Consumption vs. Water Withdrawal: Cooling with River Water

Engineers who think about this for a living split water into two buckets. Thereโ€™s water you lose, which evaporates and leaves for good. Then thereโ€™s water you borrow, which you pull through a heat exchanger and send back a few degrees warmer.

Rivers, bays and the ocean are huge bodies of water that already run at useful temperatures for most of the year. If you can move heat into them without evaporating anything, the water stays in the river. It just goes back slightly warmer than it came in.

Nautilus did this for real at Stockton, a first-of-its-kind build: a liquid-cooled data center cooled by river water.

Borrowing water comes with its own homework. Youโ€™ve got discharge temperature limits, intake screens to protect fish, and permits that take a while. Nobody gets to treat a river like a free radiator. Done right, though, the water consumed for cooling can be tiny, and the hall still gets the cold water it needs.

How Liquid Cooling Reduces Data Center Water Use

Rows of server racks in a data hall, where liquid cooling removes heat from AI processors

On to the second half of the title.

Everything above is about the outside of the building, the last step where heat finally leaves the property. Liquid cooling handles the inside: pulling heat off the chips and out of the data hall. The two connect through one number, and I promise itโ€™s less boring than it sounds. Water temperature.

Todayโ€™s AI processors can run on surprisingly warm coolant. The water coming back from the servers is hot-tub warm, sometimes warmer. Thatโ€™s great news for the equipment outside. The warmer that return water is, the more hours of the year plain outdoor air (or a river) can cool it on its own, with no chiller running and nothing evaporating. Engineers call those free-cooling hours.

Thereโ€™s a catch, and it shows up in the handoffs.

Picture heat moving out of a data center like an old-fashioned bucket brigade. The chip hands heat to the coolant. The coolant hands it to a heat exchanger inside a cooling distribution unit, or CDU. The CDU hands it to the facilityโ€™s water loop. The facility loop hands it to the dry cooler, the tower or the river. Every handoff costs you a degree or two, because heat only flows from warmer to cooler, and every exchanger needs some gap to work with.

That gap at the CDU has a name: approach temperature. Itโ€™s the difference between the facility water coming in and the water you can deliver to the servers. A wide approach means you need colder facility water to hit the same server temperature, so the chillers come on more often. A narrow approach means warmer facility water still does the job. You rack up more free-cooling hours and lean on the chillers less.

Iโ€™ll put in a word here for the folks whose blog youโ€™re reading. The EcoCoreยฎ FCD, Nautilusโ€™s facility-scale CDU and an NVIDIA AI Factory Infrastructure Partner product, runs an approach temperature down to 2ยฐC. It works with almost any facility water source, river water and seawater included. Itโ€™s installed out in the mechanical gallery instead of on the data hall floor, so that floor space stays available for servers.

Two degrees doesnโ€™t sound like much. Spread it across every hour of a year, though, and it changes how often the chillers have to run. You still want chillers. On the worst afternoon of August, youโ€™ll be very glad theyโ€™re there. You just use them less.

So if less water is the goal, liquid cooling earns its place by giving the outdoor equipment an easier job. Warmer water means more free-cooling hours, which means less mechanical cooling and fewer reasons to spray water on anything.

Questions to Ask About a Zero Water Cooling Claim

The next time a vendor, a developer or a press release says โ€œzero water,โ€ try these:

  1. Zero measured where? At the site fence, or across the whole energy supply behind it?
  2. What happens on the hottest day of the year? Chillers, adiabatic spray or something else?
  3. What does the water saving cost in energy? Ask for kilowatt-hours, and skip the adjectives.
  4. Is any water borrowed from a river, lake or sea? If so, how warm does it go back, and under what permit?
  5. What coolant temperature do the servers accept? That number decides how many free-cooling hours are even possible.

A team with a solid design will answer every one of these without flinching. If somebody gets squirmy at question two, well, now you know.

Is Zero Water Cooling Worth It?

Zero water cooling is real, and itโ€™s a worthy goal. Itโ€™s also a claim with a boundary. It covers the water a site evaporates to reject heat. The gridโ€™s water use and the hottest afternoon of the year are separate questions, and they deserve their own answers.

The good news is that the physics works in our favor more than it used to. Chips run warmer. Coolant comes back warmer. Better heat exchangers give up fewer degrees at each handoff. Add all that up and you get more hours where outdoor air or a nearby river does the work and nothing has to evaporate.

What about your site? Are you up against water limits, power limits, or both at the same time? Tell us in the comments. We read every one. And if youโ€™re sizing cooling for an AI build and want to see what your facility water could do, talk to a Nautilus engineer. Bring your worst summer day. We like those.

FAQ

In most cases, it means the site doesnโ€™t evaporate water to reject heat. Instead of cooling towers, it uses dry coolers, chillers or a borrowed water source like a river. A small amount of water still sits in closed cooling loops and needs occasional top-offs.

WUE is a metric from The Green Grid that measures the liters of water a data center uses for every kilowatt-hour of energy its IT equipment consumes. Sites with cooling towers report a WUE well above zero, while sites using dry coolers can report a number close to zero. WUE only counts water used on site, not water used by the power plants supplying the electricity.

Often, yes, especially on the hottest days. When outdoor air is too warm for dry coolers to do the job alone, the site has to run mechanical chillers, which use more power. Thatโ€™s why itโ€™s worth asking any vendor what a water saving costs in kilowatt-hours.

Yes. Water can be pulled through a heat exchanger and returned a few degrees warmer, without being evaporated. Nautilus did this at its Stockton data center, a liquid-cooled facility cooled by river water. These systems come with discharge temperature limits, intake screens to protect fish and permitting requirements.

Liquid cooling lets servers run on warmer coolant, so warmer water returns to the outdoor equipment. Warmer return water means more free-cooling hours, when outdoor air or a river can handle the heat with no chillers running and nothing evaporating. That means less mechanical cooling and fewer reasons to spray water on coils.

Approach temperature is the difference between the facility water entering a cooling distribution unit (CDU) and the water it can deliver to the servers. A narrower approach lets warmer facility water do the job, adding free-cooling hours and reducing chiller use. The EcoCoreยฎ FCD from Nautilus runs an approach temperature down to 2ยฐC.

More Recent Posts

Facility-wide CDUs with red and blue coolant piping serving rows of liquid-cooled AI server racks in a data center

Facility-Wide CDUs for AI Data Centers

TL;DR – A facility-wide coolant distribution unit connects the liquid-cooling loop serving AI hardware to the data center’s heat-rejection infrastructure. It gives operators centralized control

Scroll to Top

Nautilus Data Technologies Named NVIDIA AI Factory Infrastructure Partner