Data Center Cooling Towers: Installation, Redundancy and Repair

Data center cooling towers throw the heat collected from servers, chillers and liquid cooling loops into the outdoor air by evaporating a small share of the condenser water passing through them, and Paul Industries installs, rigs, pipes, starts up, cleans, maintains and repairs them nationwide. This page is written for the operators, developers and engineers who own that equipment. It explains how the towers fit the condenser water loop, how cell redundancy keeps cooling online during service, what makeup water and winter weather demand, and what a mechanical contractor should hand over when the work is finished.

What is a data center cooling tower?

A data center cooling tower is an evaporative heat exchanger that cools the condenser water leaving the chillers, or the water feeding a plate heat exchanger, by letting outdoor air pull heat out of it as a small fraction evaporates. Warm water is spread over a large wetted surface called fill, fans move air through the fill, and the cooled water collects in a basin and returns to the plant. On most campuses the towers are rows of identical rectangular cells set on the roof or in the mechanical yard, each with its own fan, fill, drift eliminators and basin section, piped to shared supply and return headers. Large sites use field-erected towers; smaller and modular sites use factory-assembled units shipped in sections and set by crane.

How do data center cooling towers work?

They work by evaporation: every pound of water that turns to vapor carries away roughly a thousand Btu, so a tower can drop the condenser water temperature to within a few degrees of the outdoor wet-bulb temperature. The heat arrives along a chain. IT equipment warms the air or the liquid cooling loop, that heat moves into chilled water at the cooling coils or the coolant distribution units, the chillers lift it into the condenser water, and the condenser water carries it to the towers. The compressor’s own work is added on the way, which FEMP puts at up to about 20 percent of the heat the towers finally reject. That is why a tower ton is commonly taken as 15,000 Btu per hour rather than the 12,000 of a refrigeration ton.

Evaporation is the mechanism, not a loss to be engineered away. The EPA estimates evaporation at roughly 1 percent of the recirculating flow for every 10 degrees Fahrenheit of cooling the tower achieves, so a plant circulating 10,000 gallons per minute through a 10 degree range evaporates on the order of 100 gallons per minute at full load. The system-level picture, including where the water ends up, is on our data center cooling systems page.

Cooling tower range and approach

Range is the temperature drop across the tower, hot water in minus cold water out. Approach is how close the cold water gets to the entering wet-bulb temperature. The industry reference point is 95 degrees in, 85 degrees out at a 78 degree wet bulb, a 10 degree range and 7 degree approach at about 3 gallons per minute per ton, and CTI STD-201 certifies that a manufacturer’s published ratings hold at conditions like these. A real selection uses the site’s own design wet bulb, the chiller’s condenser flow and the warmest water the plant can accept. Tighter approaches buy colder water but need a much larger tower, so the approach is a cost decision as much as a thermal one.

Towers, chillers and free cooling

In mechanical cooling mode the towers serve the chiller condensers. When the wet bulb falls far enough, a waterside economizer lets the same towers cool the chilled water directly through a plate heat exchanger with the chillers off, and in partial mode the exchanger pre-cools return water ahead of the chillers. Every hour of economizer operation shifts load away from compressors, which is the largest single lever on data center cooling tower efficiency. The other levers are fan energy per ton, clean fill and distribution, and holding the design flow through each cell.

Do data centers use cooling towers?

Many do, especially large campuses with water-cooled centrifugal chillers, because evaporative heat rejection uses far less fan and compressor energy than air-cooled equipment in most climates. Why do data centers need cooling towers when other options exist? Every watt delivered to servers becomes heat that must leave around the clock, and a tower cools water toward the wet-bulb temperature, which sits below the dry-bulb air temperature, so chillers work less and economizers run more hours. Others choose dry coolers or air-cooled chillers where water is scarce, permits are hard to obtain or the operator has committed to near-zero water use. Where a campus runs evaporative cooling towers, data center water use rises and falls with the weather and the load, which makes makeup supply and storage a design item rather than a utility afterthought. Our data center water usage guide works through the numbers.

Evaporative cooling towers for data centers are also gaining ground for a second reason: high-density liquid cooled racks accept much warmer facility water than air-cooled halls. Under the ASHRAE liquid cooling classes, W32 and W40 allow facility supply water up to 32 and 40 degrees Celsius, and towers working through a heat exchanger can make water that warm without compressors for most of the year across much of the country. The facility and technical cooling loops behind that arrangement are covered on our data center liquid cooling page.

Open towers, closed-circuit fluid coolers and hybrid towers

An open, or direct, tower sprays the condenser water itself over the fill, so the circulating water is exposed to air, dust and biology and must be filtered and treated. A closed-circuit fluid cooler keeps the process water sealed inside a coil; a separate spray water system wets the outside of the coil and does the evaporating. A hybrid unit adds a dry, finned coil section so part of the heat leaves without evaporation.

Fluid cooler vs cooling tower

The practical difference is cleanliness against cost. A fluid cooler protects the loop it serves, so a chiller condenser or plate exchanger downstream stays free of the airborne dirt that an open tower collects, and its coil can be drained or run on glycol for winter. An open tower of the same capacity is lighter, cheaper and reaches a closer approach because there is no coil wall between water and air. Many data center plants use open towers with a plate heat exchanger to get both: the tower water stays on one side and the clean building loop on the other.

What is a hybrid cooling tower?

A hybrid cooling tower combines a wet evaporative section with a dry coil section in one unit, so it can run wet in hot weather, dry in cold weather and in a mixed mode between. The dry section also heats and dries the exhaust air before it leaves, which suppresses the visible plume that forms on cold, humid days. A hybrid cooling tower for data center use earns its higher first cost where makeup water is expensive or restricted, or where the plume would sit over a neighbor, an airport approach or the site’s own air intakes.

Cooling tower redundancy: N+1 cells and concurrent maintenance

Redundancy in a tower plant means installing more cells than the design load needs, usually at least one extra cell (N+1), and piping them so any single cell can be isolated, drained and repaired while the others carry the full load on the design day. That second half is the hard part. Uptime Institute’s Tier III definition, concurrent maintainability, asks that every component supporting the IT load be removable for maintenance without affecting it, and on a tower plant that reaches past the cells to valves, headers, basins, pumps and the makeup supply.

Good cooling tower data center design therefore starts with the basin arrangement. A single shared basin under all cells cannot be drained without stopping the plant, so redundant plants use separate basins, partitioned basins or individual cells joined by equalizer lines that each carry their own isolation valve. The supply and return headers are looped or split so a valve can be replaced without losing a path, and every cell gets its own isolation valves on both the hot water inlet and the cold water outlet.

The same principle covers the chiller side, which is described on our chilled water systems page.

How a multi-cell tower plant is sequenced

A cooling tower sequence of operation sets how many cells run, at what fan speed, and how water flow is shared as load and weather change. For efficiency, energy codes built on ASHRAE 90.1 require speed control on larger tower fans and, on multi-cell towers with variable-speed drives, running as many fans as the manufacturer allows at the same speed. Because fan power falls steeply as speed drops, spreading air and water across all available cells at low speed rejects the same heat for less energy than running a few cells hard.

Water distribution sets the limit on how far that can go. The cooling tower turndown ratio, the span between the most and least water a cell can distribute evenly across its fill, tells the controls when to stop adding cells; below the manufacturer’s minimum flow, gravity basins or spray nozzles leave dry patches, scale forms at the wet edges and capacity falls. The CDC adds a health reason to share the work: it recommends balancing run time across towers and cells, and circulating water through cells on wet standby at least three times a week when standby lasts under five days. A standby cell that never runs is stagnant water, and stagnant water is where Legionella grows.

Makeup water supply and on-site storage

Makeup water replaces everything the towers lose to evaporation, drift, blowdown and leaks, so a data center that rejects heat evaporatively is only as reliable as its water supply. Uptime Institute’s design papers treat loss of utility water as a condition to plan for, and on-site makeup storage sized for a set number of hours at design load is the usual answer; 12 hours is a widely used basis. The storage tank, transfer pumps and the controls that switch between utility and storage all need the same redundancy as the towers themselves.

Meters on makeup and on blowdown are the cheapest diagnostic a tower plant can have. The EPA notes that cycles of concentration can be calculated two ways, from the conductivity ratio of blowdown to makeup and from the volume ratio of makeup to blowdown, and when the volume figure runs more than 10 percent above the conductivity figure the system is probably losing water through a leak, an overflow or excess drift. The makeup itself is often softened, filtered or run through reverse osmosis so the towers can hold more cycles; Paul installs that equipment, covered on our industrial water treatment page, while the program that sets cycles, blowdown and biological control lives on the cooling tower water treatment page.

Siting, drift, plume and recirculation

Drift is liquid water carried out of the tower as droplets, and it matters because those droplets contain whatever is in the circulating water. Older towers lost 0.05 to 0.2 percent of their flow as drift; modern drift eliminators cut that to below 0.005 percent, but only while they are intact, seated and clean. Plume is different: it is water vapor condensing into visible fog as the warm, saturated exhaust meets cold air, and it carries no treatment chemicals.

Where should a cooling tower be installed on a data center site?

Away from building air intakes, with open air on every inlet face, and downwind of anything that would foul it. The CDC recommends at least 25 feet between a tower and building air intakes so drift is not drawn into ventilation, and a data center adds its own intakes to protect: air handlers, generator combustion air and battery room ventilation. Cooling tower recirculation happens when warm, saturated discharge air is drawn back into the air inlets, which raises the effective wet bulb and steals capacity on exactly the hot, still days the plant needs it. Screen walls, closely packed rows and nearby taller buildings are the usual causes, and the manufacturer’s spacing guidance for multiple cells should be followed rather than squeezed.

Cooling tower winter operation and freeze protection

Cold weather is a design condition at data centers rather than a seasonal shutdown, because the IT load never stops and the towers often run through the coldest nights to feed the waterside economizer. The aim is to keep the basin, the fill and every exposed pipe above freezing while the plant still rejects heat. Typical measures are electric basin heaters, which protect only the basin, heat trace and insulation on makeup lines, drains and any pipe that can stand still, and fan control that cycles or slows fans as the water temperature falls.

Can cooling towers freeze?

Yes, and ice usually forms first where water moves slowly or air is coldest: on the air inlet louvers, at the edges of the fill, in idle cells and in stagnant sections of pipe. A cooling tower winter bypass sends the returning condenser water straight into the cold water basin instead of over the fill, keeping the basin warm when load is light, and periodic fan reversal can melt ice that builds up at the inlets. Cells that are taken out of service for the winter should be drained, including their risers, distribution and any low points, rather than left full and still. Closed-circuit coils need glycol or a reliable drain-down, since a frozen coil splits.

Legionella water management for tower plants

Any building with a cooling tower should have a written water management program, and ASHRAE Standard 188 is the consensus standard that describes one, with ASHRAE Guideline 12 giving implementation detail and the CDC Legionella toolkit translating both into worksheets. Legionella grows best between 77 and 113 degrees Fahrenheit, which is where condenser water lives for much of the year. The CDC’s cooling tower module recommends high-efficiency drift eliminators, piping without dead legs, automated disinfectant and blowdown control, weekly flushing of low-flow runs, and cleaning and disinfection before commissioning, before startup after a shutdown and at least once a year offline.

For a data center, the redundancy design is part of the Legionella plan. Spare cells, bypass lines and rarely used headers are exactly the stagnant water the CDC warns about, so the program should name every one of them and say how it is kept moving or kept dry. Cleaning and disinfection are scheduled one cell at a time; our cooling tower cleaning and maintenance page covers that procedure in full.

Mechanical maintenance and repair schedule

Mechanical maintenance on data center cooling towers is planned around redundancy: each task is done on one isolated cell while the others carry the load, and the cell is proven back in service before the next one opens. Treat the intervals below as a baseline, then adjust them once the manufacturer’s manual and a year of the site’s own records are in hand.

Paul carries that work on preventive maintenance schedules and answers breakdowns with emergency repair crews, so a cell lost to a gearbox or a basin leak is back in the redundancy count quickly instead of leaving the plant at N.

Retrofit and capacity expansion for AI loads

AI data center cooling towers see heat loads that rise in steps as GPU halls are fitted out, so the original plant is often too small within a few refresh cycles. The options run from least to most disruptive. Higher-efficiency fill and clean distribution recover lost capacity in existing cells. A cooling tower fan retrofit replaces fans, gearboxes or motors with assemblies that move more air for the same power, often alongside speed control. Adding cells on the existing header gains capacity where the yard, structure and header size allow it. Re-piping the headers and adding pumps comes next, and a parallel tower plant with its own headers is the last step.

Each step changes hydraulics, so the condenser water flow balance, pump curves and minimum cell flows are rechecked before the new cells are tied in. Tie-ins to a live plant are planned around isolation valves installed in advance, or hot taps where none exist, and every new section is pressure tested through our hydrostatic testing service before it is opened to the system.

Cooling towers vs dry coolers: choosing heat rejection for a data center

The choice trades water against electricity and temperature. A tower delivers water close to the wet bulb with low fan power but consumes water continuously; a dry cooler consumes none in normal operation but can only cool toward the higher dry-bulb temperature, so it needs more fan power, more coil area or chiller help on hot days. Hybrid towers and adiabatic coolers sit between the two. Selecting a cooling tower for data center duty makes the most sense where water is available and affordable, the climate has a meaningful gap between wet and dry bulb, and the cooling loads are dense enough that the energy saving pays for the water. The other side of that decision is set out in our dry cooler guide.

Installing data center cooling towers: rigging to startup

Installation is a sequence where each step is recorded before the next begins, because the commissioning team will ask for every record later.

Heavy picks are planned by our millwright and rigging crews, the piping is built to ASME B31.9 or the project specification, and the records feed straight into data center commissioning.

How to choose a contractor for data center cooling towers

Choose a contractor who treats the tower plant as part of a live, redundant system rather than as a piece of equipment. Ask how they would isolate one cell without touching the others, and how they plan tie-ins to an operating plant. Ask to see a startup report and a cell-by-cell flow balance from previous work, how they hold welding and pressure test records, and how mechanical repairs are staged so the plant never falls below N+1. Ask whether the same crew can rig, pipe, start up, clean and repair, since every handoff between firms is a place where records and responsibility get lost. The wider mechanical scope on a campus is laid out on our data center mechanical contractors page, and the hub for all of it is data centers.

Paul Industries on data center cooling towers

For cooling towers for data centers, Paul Industries receives, rigs and sets new cells; pipes condenser water headers, cell connections, equalizers, makeup, drain and overflow lines; installs isolation valves and the plate heat exchangers that tie towers to economizer and liquid cooling loops; pressure tests and flushes; installs water treatment equipment including controllers, meters, side-stream filtration and makeup softening or reverse osmosis; disinfects and starts up the plant with a cell-by-cell flow balance; and then cleans, disinfects, maintains and repairs it, covering fans, gearboxes, motors, fill, drift eliminators, distribution, leaks, basins, casing, piping and pumps. Capacity additions and retrofits are planned around the site’s redundancy so cooling stays online throughout. Request a quote for a new tower plant, an added cell or a repair.

Standards referenced: ASHRAE Standard 188 · ASHRAE Guideline 12 · CDC Legionella Control Toolkit · CTI STD-201 · CTI ATC-105 · ASHRAE 90.1 · ASME B31.9 · Uptime Institute Tier Standard

Frequently asked questions

Is a cooling tower plant cheaper to run than an air-cooled chiller plant?

In most climates, yes, on energy: water-cooled chillers serving towers typically use noticeably less power per ton than air-cooled machines, because the condenser sees water near the wet bulb instead of hot outdoor air. The offset is the cost of makeup water, sewer charges on blowdown, water treatment and the extra maintenance an evaporative plant needs.

What does a data center cooling tower look like?

Usually a line of boxy cells, often the size of a shipping container or larger, with a large fan in a stack on top, louvered or open air inlets on the sides and a basin underneath. Rooftop units on urban sites are smaller; hyperscale campuses may have long rows of field-erected cells in the mechanical yard.

How does a cooling tower and chiller work together in a data center?

The chiller removes heat from the chilled water and rejects it, plus its compressor heat, into the condenser water. The condenser water flows to the tower, gives up that heat by evaporation and returns cooler to the chiller condenser. The cooler the tower water, the less work the chiller compressor has to do.

What is approach in cooling tower design, and why does it matter for a data center?

Approach is the difference between the water leaving the tower and the outdoor wet-bulb temperature. A smaller approach gives colder water, which improves chiller efficiency and extends economizer hours, but each degree closer requires noticeably more fill, airflow and footprint, so it is set by comparing tower cost with energy saved.

How long do cooling towers last at a data center?

ASHRAE's median service life figures put galvanized steel towers at about two decades, with concrete and ceramic construction lasting considerably longer. Fill, drift eliminators, gearboxes and fans are replaced several times within that life, and water quality, basin upkeep and redundancy that allows proper maintenance have the largest effect on how long the structure survives.

How many spare tower cells should a data center have?

Enough to carry the design load at the local design wet bulb with at least one cell out of service. A plant needing four cells for the load is built with five for N+1; some operators specify N+2 so that one cell can be in planned maintenance while another fails.

Can one cooling tower cell be serviced while the rest keep running?

Yes, when the plant is built for it: each cell needs isolation valves on its inlet and outlet, a basin that can be separated from the others and a drain of its own. Without those, cleaning or repairing one cell means reducing flow or shutting the whole tower down.

How much makeup water storage should a data center keep for its towers?

Enough to ride through the longest credible loss of utility water at design load. Twelve hours is a commonly used basis for redundant designs, and the tank, transfer pumps and switchover controls should be as redundant as the towers they supply.

Should idle standby tower cells be kept wet or drained?

For short standby, keep them wet and circulate water through them regularly; the CDC suggests at least three times a week when standby lasts under five days. For longer outages, drain and dry the cell, then clean and disinfect it before it returns to service.

What does a tower fan failure do to a redundant plant?

The cell loses most of its capacity, and the plant controls should stage up a standby cell or speed up the remaining fans. A vibration switch normally trips a fan with a damaged blade or failing gearbox before it destroys itself, which is why that switch is tested on a schedule.

Why do cooling towers make a visible plume in winter?

The exhaust air leaves the tower warm and saturated, and when it meets cold outdoor air the vapor condenses into fog. The plume is condensed water vapor rather than drift, but where it would drift across roads, air intakes or neighbors, a hybrid tower with a dry section can suppress it.

How does a cooling tower connect to a liquid cooled data hall?

Through a plate heat exchanger or a closed-circuit cooler, so the open tower water stays on one side and the clean, inhibited facility loop that feeds the coolant distribution units stays on the other. That keeps airborne dirt and biology away from the fine passages in cold plates and CDU exchangers.

What is the difference between counterflow and crossflow cooling towers?

In a crossflow tower, hot water falls from open gravity distribution basins through the fill while air moves horizontally across it, which makes distribution easy to inspect. In a counterflow tower, air rises against water sprayed from pressurized nozzles, which usually gives a smaller footprint per unit of capacity.

What happens at startup of a new tower plant?

After the piping is pressure tested and flushed and the basins are clean, the system is filled, disinfected and circulated, flow is balanced cell by cell, fan rotation, vibration and level controls are checked, and alarms are proven. The startup report becomes the baseline for later maintenance.

Are factory-assembled or field-erected towers better for data centers?

Factory-assembled cells ship in sections, set quickly by crane and suit modular and phased builds. Field-erected towers are built on site, reach larger capacities per cell and suit large, fixed plants. Many campuses use factory-assembled cells because capacity can be added in matching increments as halls come online.

How is cooling tower capacity checked after installation?

By a field thermal performance test, commonly to CTI ATC-105, measuring water flow, hot and cold water temperatures, wet bulb and fan power, then comparing the results with the manufacturer's rated curve. CTI STD-201 certification of the model line gives added confidence that the published ratings are realistic.

What gets inspected on a tower cell during a data center maintenance window?

With the cell isolated and drained: fan blades and hub, gearbox oil and seals, drive shaft and couplings, motor bearings, fill condition, drift eliminator seating, distribution nozzles, basin sediment and corrosion, makeup valve and level controls, and the isolation valves themselves.

Can capacity be added to an existing tower plant without shutting it down?

Usually, if isolation valves were placed for future cells or can be added by hot tapping. New cells are set, piped and pressure tested separately, then tied in during a planned window while the existing cells keep the load, with flow rebalanced afterward.

Related: Data Centers · Data Center Cooling Systems · Cooling Tower Water Treatment · Cooling Tower Cleaning · Chilled Water Systems · Data Center Water Usage · Request a quote