Heat Rejection: Chillers, Dry Coolers, Towers, Adiabatic & Economizers

Everything upstream of this chapter (the cold plate, the CDU, the secondary loop, the facility water loop) moves heat from the die to a pipe of warm water on the roof or in the yard. Heat rejection is the last meter: dumping that heat into the atmosphere so the loop can come back cold enough to do it again. It is the least glamorous subsystem in the building and the one that most directly sets your annualized PUE, your annual water bill, and whether the local utility and watershed will tolerate you. The physics fits in one sentence: you can reject heat to dry air, to evaporating water, or to a refrigeration cycle that pumps it uphill. Which one you lean on, and for how many hours a year, determines the plant's electricity use, its cooling makeup-water demand and the operating cost of both.

One variable dominates: the temperature of the water you push to the rack, which you set upstream. Warm-water DLC (W32–W45 in ASHRAE's classes) is what makes the cheap rejection paths reachable at all, and a lower supply temperature cuts the hours they can cover. The chapter therefore runs as a cascade: loop temperature, rejection technology, climate-bin economics, the water/energy operating point, then the redundancy and ride-through the plant must carry, with a cost attached to each fork. Loop temperature and the W-classes are set in Chapter 5.7; the metrics we score against are canonical in Chapter 15.1.

There are four families of heat-rejection plant, and the honest way to order them is by the resource each spends to move a watt of heat into the sky. Below they run from evaporating water to spending compressor energy; compare any two at the same supply temperature, heat duty and design state.

Open cooling towers (evaporative). The thermodynamic champion: they reject heat by evaporating a fraction of the circulating water, so their floor is the ambient wet-bulb temperature, which varies by site and by hour, so work from your site's coincident design bins, not a rule-of-thumb hot-afternoon depression. That low approach lets a tower-cooled plant hold a low PUE in climates where a dry cooler would have surrendered to a compressor. The price is liters: evaporation, blowdown and drift lift site WUE above a dry system's near-zero process-water floor. The plant also needs biocide treatment and an ASHRAE-188 program. Open towers also expose whatever loop they serve to airborne fouling, so a DLC facility loop is isolated from them by a plate heat exchanger or served by a closed-circuit fluid cooler; chiller condenser water usually runs directly through the tower.

Adiabatic / evaporatively-assisted dry coolers. The hybrid middle. A dry cooler (a finned air-to-water coil with fans) runs bone-dry most of the year. On the hot, high-dry-bulb hours it cannot meet alone, a pre-cooling stage wets pads or sprays a mist into the inlet air to pull it toward its wet-bulb, recovering capacity without year-round evaporation. You spend water only in the bins where the dry stage falls short and you have chosen to assist. At a cool site those bins can be rare; at a hotter one you buy many more wet hours, each traded directly against buying or running a chiller. This is the canonical water-vs-energy compromise instrument.

Dry coolers (sensible only). Zero process water by construction: they reject heat to dry air across a coil, like a giant car radiator. Their floor is the ambient dry-bulb, so their fortunes depend entirely on how warm your loop is and how cool your air gets. A W40+ loop in a temperate or cold climate can run dry-cooler-only for most or all of the year, with near-zero WUE and low PUE together, the best-case outcome. The same dry cooler on an 18 °C chilled-water loop in Phoenix is a fan that cannot keep up, and a chiller has to carry the day.

Mechanical chillers (vapor-compression). The brute-force backstop: a refrigeration cycle that can deliver loop water colder than the ambient wet-bulb, at the cost of a compressor whose electrical draw is the single largest non-IT load in a chiller-based plant. Chillers cover the bins the free-cooling and adiabatic paths cannot, and serve any loop too cold for them. Every compressor hour is a direct, measurable hit to PUE. In a 2026 warm-water DLC facility, the design goal is to make the chiller a rarely used insurance policy, or to delete it.

What a compressor hour costs. Chiller efficiency follows lift, the gap between the refrigerant's evaporating and condensing temperatures, and compressor work rises with it. The US code floor shows the scale at AHRI 550/590 standard rating conditions: 6.7 °C leaving chilled water, with 29.4 °C entering condenser water or 35 °C air for an air-cooled unit. ASHRAE 90.1 Table 6.8.1-3 allows a large water-cooled centrifugal chiller at most 0.560 kW/ton at full load, a COP of about 6.3, and asks an air-cooled chiller for at least 10.1 EER, a COP of about 3.0. Warm water changes that arithmetic. Hold condensing at 45 °C, close to an air-cooled condenser's hot-day temperature, let the refrigerant evaporate about 5 K below the leaving water, and raise that water from 7 °C to 30 °C: the ideal (Carnot) COP climbs from about 6 to about 15. Compressor work per kW of heat falls by more than half. Oil-free magnetic-bearing centrifugal machines are built for that low-lift trim duty, but check the leaving-water and minimum-lift limits of the unit you buy: a standard comfort-cooling chiller is not necessarily rated to make 30 °C water.

Refrigerant choice is now a schedule risk. In the EU, new stationary chillers above 12 kW running on F-gases with a GWP of 750 or more fall under the placing-on-the-market ban in point 7(d) of Annex IV to the F-gas Regulation (EU) 2024/573. The ban starts on 1 January 2027, unless safety requirements at the site require such a fluid. R-134a (GWP 1,430) is out for new units; installed machines are not affected. In the US, EPA's Technology Transitions rule (40 CFR Part 84, Subpart B) caps comfort-cooling chillers at GWP 700 for systems installed from 1 January 2025, and data-center and IT-equipment cooling systems from 1 January 2027. EPA's May 2026 revision kept both dates. The rule's summary table never assigns a data-center chiller plant to either row, but the choice moves only the date: under both, any such chiller going in on or after 1 January 2027 must run below GWP 700, and the comfort-cooling row has held that cap since 2025. The low-GWP candidates trade one risk for another. HFOs such as R-1234ze(E) (mildly flammable, A2L) and R-1233zd(E), and blends such as R-514A (nonflammable but B1, the higher-toxicity class) and R-515B, fall within the EU's universal PFAS restriction proposal. ECHA's risk committee adopted its opinion in March 2026; its socio-economic committee had not adopted a final opinion by mid-October 2026. Propane and ammonia avoid the PFAS question but bring flammability or toxicity, and the safety-code and machinery-room requirements that go with them. A chiller ordered now on a long lead time can land after these dates, so fix the refrigerant in the purchase specification, not at delivery.

"Free cooling" is a misleading name, because the fans and pumps still draw power. The term of art means rejecting heat without running a compressor. There are two economizer families, and the fact that you are cooling liquid, not air, mostly settles the choice.

Air-side economization comes in two forms. Direct systems duct filtered outside air to the IT and need humidity and contaminant controls, because they import humidity, particulates and, in some regions, corrosive gases. Indirect systems exchange heat between separate indoor and outdoor airstreams through a plate exchanger or heat wheel, which keeps outside air out of the IT space; check the wheel you buy for leakage and carryover. Air-side economization is the workhorse of legacy air-cooled hyperscale halls and can deliver near-1.1 PUE in cool climates, but it scales with airflow, not with a liquid loop. In a dense liquid-cooled AI hall with L2L CDUs, it cools the residual room-air duty, not the heat the cold plates hand to facility water. An L2A exchanger dumps its cold-plate heat into that room air, so count each rack's actual liquid/air split and where each share ends up. Water-side economization is the relevant lever for liquid plants: when the ambient wet-bulb (tower) or dry-bulb (dry cooler) is low enough, a plate heat exchanger lets the rejection loop cool the facility loop directly and the chiller idles. It is what turns a chiller plant into a part-time chiller plant.

What matters is the bin-hour count, not the nameplate. Take your site's TMY (typical meteorological year) wet-bulb and dry-bulb distribution, lay your loop's approach temperature over it, and read off how many of the 8,760 hours each mode covers. A warm-water loop moves the crossover so far that in temperate and cold climates compressor hours fall to a rounding error. In one study of a building in temperate Suwon, Korea, water-side economizing on cooling towers cut heat-source energy (chillers, towers and pumps) by 52.8% in a model of the five economizer months, January to March and November to December, and by 42.0% in a separate measured 42-day winter trial (Kim, Lee & Kim, 2024). Neither is a full-year saving. The saving grows with warmer loop temperatures and shrinks in hot, humid climates. The same overlay at a hot-humid site shows the compressor carrying real load, and tells you quantitatively what you are buying when you site there.

A modern AI heat-rejection plant is rarely one machine. It is a hybrid that runs three modes against changeover setpoints, and the control logic that sequences them decides annualized PUE and WUE far more than the equipment selection does. The canonical sequence, coldest ambient to hottest:

• Free-cooling (dry) mode. Modulate the dry-cooler fans alone while the dry cooler, at the actual load, flow, ambient and fouling, still delivers the loop supply setpoint. Ambient below the return temperature gets you partial economization, but it does not on its own prove the cooler can carry full dry mode. Full dry mode is a capacity you have to size the cooler for. It uses no evaporative water, though fans and pumps still draw power. A warm loop keeps the plant here for the bulk of the year in most of the populated temperate world.
• Adiabatic-assist mode. As dry-bulb rises and the dry coil loses its approach, stage on pre-cooling (wet the pads or pulse the spray) to drag entering-air temperature toward the wet-bulb and recover capacity. You begin spending water, but only on these bins, and you have still not started a compressor.
• Mechanical (chiller / compressor) mode. Commit the compressor when even the adiabatic-assisted approach cannot hold the loop setpoint, when water is unavailable, or when the water-versus-power crossover favors it, as the worked example below shows. In a well-sequenced warm-water plant in a temperate climate this can be a few hundred hours a year or fewer; in some Nordic and high-altitude sites, effectively never.

The decision embedded in the sequence is the changeover wet-bulb (and dry-bulb) setpoints and the hysteresis band around them. Set them too aggressive, chasing the last fraction of free cooling, and the plant hunts: fans, water and compressors cycle on the boundary, churning energy and water and stressing equipment. Set them too conservative and you start the compressor on hours the adiabatic stage could have covered, paying PUE you did not owe. Set each changeover by weighing total plant power, the water you are permitted and what it costs, and the capacity available at that bin. This is a control-tuning problem, the thermal sibling of the loop-side setpoint-stability problem. → control-loop tuning, anti-hunting, and the no-inertia transient in Chapter 5.12.

Cold-bin dry supply screen = 15 + 5 = 20 °C: modulate to 30 °C. In the middle bin, dry gives 33 °C, while assist gives 28 − 0.750(28 − 20) + 5 = 27 °C: use assist or mechanical cooling. In the hot bin, assist gives 35 − 0.750(35 − 27) + 5 = 34 °C, so mechanical cooling is required. A temperature screen is no substitute for the capacity curve.

Hybrid liquid-plant energy = 6,000 × 20.0 + 2,000 × 25.0 + 760 × 100 = 246 MWh; dry/mechanical = 6,000 × 20.0 + 2,760 × 100 = 396 MWh. IT energy is 3,504 MWh. Common overhead is (16.0 + 20.0) × 8,760 = 315.36 MWh. Facility totals round to 4.07 and 4.22 GWh, so PUE is 1.16 and 1.20. Hybrid input water is 2,000 × 0.500 = 1,000 m³, which gives 0.285 L/kWh IT on a total-input boundary. If the plant discharges water, not all of that input evaporates.

Assist saves 150 MWh for 1,000 m³, or 150 kWh avoided per m³. Choose assist when water is permitted and a cubic meter, all burdens counted, costs you less than the electricity it saves. Above that crossover, or with no water allocation, choose the dry/mechanical alternative if it meets capacity. No tariff assumption is needed to expose the trade. The crossover moves with the chiller. The 100 kW mechanical-bin draw is 80 kW above the 20 kW fan-and-pump baseline, a compressor COP of about 4.0 on 323 kW. A low-lift machine at a COP of 6 cuts that draw to about 74 kW; assist then saves about 98 MWh, and the water price at which it stops paying falls by about a third.

Both fail the outage case: 323 − 300 = 23 kW uncovered. Close the gap with more surviving capacity, or prove an accepted load reduction of at least 23 kW that keeps the air path running. The data-center chapter of ASHRAE's 2023 HVAC Applications handbook supports the method. Chapter 15.1 defines the metrics, 15.4 the water accounting and 13.5 the acceptance tests on the installed plant.

Strip away the equipment and heat rejection is a single trade: liters of water against kilowatt-hours of energy. Evaporative rejection cools water toward the ambient wet-bulb temperature, with a finite approach, so your compressor runs less and your PUE improves, at the cost of WUE. Dry rejection pushes cooling process-water input toward zero. Whether PUE rises, and by how much, takes a comparison on matched annual load, weather and equipment curves; in hot climates a compressor draws the difference. There is no costless option, only the operating point that minimizes your blended cost of water and power under your permit constraints.

The numbers anchoring the frontier are stark. LBNL's December 2024 US fleet model puts annual-average site WUE at ~0.36 L/kWh in 2023 and projects 0.45–0.48 L/kWh by 2028 across all cooling types; its June 2026 update revised the energy estimates but published no new water figure. Designs with dry, non-evaporative heat rejection approach zero site process water. Microsoft’s FY2025 cooling-and-humidification WUE was ~0.27 L/kWh for its fully owned and controlled data centers operating for that entire fiscal year. Its separate December 9, 2024 design announcement estimated avoiding more than 125 million liters per data center annually by eliminating cooling evaporation. Microsoft's first Fairwater site in Mount Pleasant, Wisconsin, fully operational since June 2026, runs over 90% of its capacity on that kind of fill-once closed loop. The remaining air-cooled servers use outside air, with water only on the hottest days. Microsoft's September 2025 description has the loop's heat leaving through finned coolers with 172 fans, each 20 feet across, on a site that also carries a large chiller plant. The water saving is bought with fan and chiller energy, the trade this section is about, and its annual electrical cost still depends on local weather and plant curves.

The fork is genuine and it is regional. Cold Nordic warm-water sites are candidates for dry or economizer rejection, with the actual hours set by the loop setpoint, the weather and the equipment curves. In hot-arid, water-constrained sites such as the US Southwest or the Gulf, price dry or mechanical rejection against whatever evaporative option the permit allows, using the same annual energy and water accounting. The social license and the discharge permit push those sites toward dry rejection even though the climate punishes it.

Your water-safety obligations follow how the equipment evaporates water, not what the brochure calls it. Open towers and evaporative condensers need water treatment and Legionella risk management. Adiabatic systems that recirculate spray water need a risk assessment for their specific design. Once-through wetted media, with no stored or recirculated water and no aerosol, take a lower-risk control path. Cooling towers can support Legionella pneumophila growth and release contaminated aerosol beyond the facility: they warm water into the bacterium's growth range and then spray it into the air. Both open- and closed-circuit evaporative towers need the applicable controls. ANSI/ASHRAE Standard 188 (current edition 2021) is the consensus standard that governs the response. Wherever its building-water-system requirements apply, it calls for a Water Management Program. Whether the standard binds you in law depends on the jurisdiction. The program is a documented analysis of the water system, defined control limits at critical control points (disinfectant residual, temperature), monitoring, corrective actions and verification records. In the US, CDC guidance reinforces it. Skipping it is a liability and, in several jurisdictions, an operating-permit violation, not an O&M shortcut.

The program governs these operational mechanics:

• Biocide regime. Oxidizing (chlorine/bromine) and non-oxidizing biocides on a dosing schedule, plus dispersants to break up biofilm, the substrate Legionella actually hides in. Treatment chemistry interacts with corrosion and scale control, so this is one coupled water-chemistry problem, not three independent ones.
• Cycles of concentration and blowdown. As water evaporates, dissolved solids concentrate. "Cycles of concentration" (C) is how many times you let them concentrate before bleeding off (blowdown) and replacing with fresh makeup. This one ratio sets both your makeup-water draw and your discharge volume, and the balance follows from the heat. Every kWh the tower rejects as latent heat vaporizes about 1.5 L of water (3,600 kJ over water's latent heat of about 2,430 kJ/kg). A tower that rejects about 80% of its heat latently therefore evaporates about 1.2 L per kWh of heat (E). The bleed, blowdown B plus drift D, is E/(C − 1), so makeup is M = E + B + D = E·C/(C − 1). A 10 MW tower load at 80% latent evaporates about 12 m³/h; at five cycles it bleeds about 3 m³/h and draws about 15 m³/h of makeup. Going from three cycles to six cuts makeup from 1.5E to 1.2E, 20% less, and blowdown from half the evaporation to a fifth, at the price of higher scaling and corrosion risk and a more concentrated discharge. Chapter 15.4 turns the same balance into the consumed share, (C − 1)/C, for water accounting.
• Drift. Droplets carried out of the tower by the airflow. Drift eliminators minimize them, both to cut water loss and to limit aerosol dispersal of any contamination.

The discharge side is where arid siting collides with environmental permitting. Tower blowdown is a concentrated brine. A facility cannot always send it to a municipal sewer (volume, salinity, biocide residuals), and in a closed-basin arid region there may be no surface water to discharge to at all. At such a site the options are dry rejection, which makes no tower blowdown, or evaporative rejection with permitted treatment, reuse and disposal. Where no liquid-disposal route exists, zero-liquid-discharge (ZLD) is the price of admission to evaporative cooling. It is an on-site treatment train, typically reverse osmosis followed by thermal evaporation/crystallization, that recovers most of the water for reuse and reduces the waste to a solid salt cake for landfill, discharging essentially no liquid. ZLD is energy- and capex-intensive and is rarely chosen for its own sake: its bill is the toll for keeping the wet-bulb advantage. → discharge and withdrawal permitting as a siting critical-path item in Chapter 3.9; the water screen in siting in Chapter 3.1.

Heat-rejection plant must be sized not just for the worst climate bin but for the worst climate bin with a unit down for maintenance. Concurrent maintainability is the Tier-III-and-up property that any single component can be taken out of service for planned work without dropping below design load. It drives N+1 (or N+2) on towers, dry coolers, pumps and chillers, with isolation valves and headered piping so a cell can be drained and serviced while the rest carry the heat. Your maintenance philosophy and tier commitment set the redundancy count. The harder, AI-specific problem is how the product you buy actually responds, in test, to an unplanned loss.

Here dense DLC breaks the old playbook. A chilled-water plant rode a chiller trip or a utility blink on stored cold water while the generators picked up. A warm-water DLC plant often has no such reserve. The water has not lost its heat capacity, but the temperature margin and the liquid inventory that absorb the heat are smaller, so the thermal inertia you once counted on runs out sooner (Chapter 5.12 derives the flow-loss transient). Lose flow, and the junction response (throttle, controlled shutdown, emergency shutdown or trip) has to come from data and tests on your product, not from a comparison with a slow plant.

Choose the rejection system on annual energy and water, then check it against the required hottest state and the state with a unit out. Efficient weather bins do not make up for a capacity failure. Add rejection capacity or define an explicit heat-reduction state before you commit the hall’s usable IT load.