Case Study: Evaporative Cooling for a Data Center Intake
High-pressure misting precools data center intake air: droplet sizing, nozzle density, water quality and published cooling performance ranges.
When a data center hits its heat ceiling, the options are expensive. Every kilowatt of IT load becomes a kilowatt of heat that must be carried away by airflow. Published design references put typical rack airflow at roughly 160 CFM per kilowatt. Compressor cooling draws its heaviest power exactly when the ambient temperature peaks. Evaporative cooling removes the same heat with a fraction of the energy, because the evaporation of water does the work. But a misting system sized wrong wets the filters or carries drift into the hall. This case is about a colocation facility that made that tradeoff deliberately, and the sizing logic that kept the water out of the electronics.
Case Snapshot
The reference configuration below is the generic shape of the installation. It is the starting point the sizing was built from. Every value is a realistic working figure for a facility of this class.
| Item | Value |
|---|---|
| Industry | Data center HVAC, colocation and edge hosting, anonymized |
| Duty | Precooling intake air for air handling units, adiabatic cooling |
| Facility class | 1 MW IT load class, single hall |
| Original setup | Mechanical DX cooling only, no economizer stage |
| System chosen | High-pressure misting, 70 bar class |
| Operating point | 10 to 20 µm droplets, 1 to 3 nozzles per m² of intake, air velocity 1.5 to 3 m/s |
| Water quality | RO or softened water to control scaling |
| Compliance | Water use reporting, PUE targets, ASHRAE recommended envelope |
The facility class matters more than the exact numbers. A 1 MW hall with a single air handling plant is the most common shape in colocation. The logic below transfers to most halls in the 0.5 to 5 MW range. The customer stays anonymous here because water use is a sensitive metric in their market. What matters is the sizing path, and that transfers completely.
The Challenge
Every watt of server power ends up as heat inside the hall. Published cooling references treat IT load and heat load as equal, watt for watt. The airflow must be large enough to lift that heat from the racks to the cooling plant. Typical published values run 130 to 160 CFM per kilowatt of IT load at a 10 to 15 °C supply-to-return split. Halve the airflow and you double the temperature rise. Double the temperature rise and the servers throttle.
The compressor plant carried the whole load in summer. Mechanical DX cooling loses efficiency as the outdoor air warms, because the condenser rejects heat against a hotter sink. Published data center references put the coefficient of performance of DX systems well below that of evaporative options on hot afternoons. The compressor power draw at 35 °C ambient is the number that drove this project. Peak demand pricing made those afternoon hours the most expensive in the day.
The facility ran under a PUE target set by the parent group. Published industry reporting shows typical colocation PUE values between 1.3 and 1.6, with efficient designs lower. Every kilowatt of cooling power adds directly to PUE, because PUE is total facility power divided by IT power. Shaving cooling power at peak ambient was the fastest lever available. The plant did not need a new chiller. It needed the chillers to run less.
Evaporative cooling trades water for electricity. That tradeoff is normal in published data center practice, where water use effectiveness, or WUE, is reported alongside PUE. The facility had a water budget and a reporting obligation, so the water side had to be designed and metered, not ignored. The point of the project was to spend a little water and save a lot of electricity. The system had to run only when the weather made that trade worthwhile.
The risk that shaped the design was water where it does not belong. Direct spray into an intake airstream can wet the filters if the droplets are too large. Wet filters load faster, grow biology, and add pressure drop. Drift can also carry mineral residue onto coils and into the hall. Published references on direct evaporative cooling for data centers treat filter wetting and humidity control as the two main failure modes. The design exists to avoid both.
Evaporative cooling for data centers is well documented in published case references. Direct and indirect evaporative cooling are both established approaches. Indirect systems cool without adding moisture to the supply air, at the cost of more hardware. Direct systems spray or pad the intake airstream directly and are simpler to retrofit. Published case references report evaporative economizers operating in hot-dry climates with meaningful reductions in mechanical runtime. This project chose the direct route because the intake plenum made it a small retrofit.
The Solution
The solution was adiabatic precooling of the intake airstream. High-pressure misting nozzles spray a fine fog into the plenum ahead of the filters. Each droplet evaporates and pulls its heat of vaporization from the air, so the dry-bulb temperature falls. The wet-bulb temperature does not change. That is the floor of the process, and it is fixed by the weather.
Evaporation physics sets the numbers. The latent heat of vaporization of water is about 2257 kJ per kilogram. Every kilogram of water that fully evaporates removes roughly 2257 kJ from the airstream. A cooling load of 1 kW therefore needs about 1.6 liters of evaporated water per hour, before losses. Published references quote similar figures for direct evaporative cooling water consumption.
Media pads were the obvious alternative, and they lost on three counts. First, retrofit. A pad bank needs a wet section, a sump, and a recirculating pump, which is a structural change to the plenum. Second, pressure drop. Pads add resistance across the intake, so the fans work harder and eat part of the energy saving. Published references put media pad pressure drop in the range of 25 to 125 Pa at typical face velocities. Third, maintenance. Pads hold moisture and biology, and they are consumable. A high-pressure spray bank is stainless steel and cleans itself.
Droplet size decided everything downstream. An evaporation cooling nozzle that makes droplets of 10 to 20 µm produces fog that evaporates in well under a second in dry air. A droplet of 100 µm survives far longer and can cross the plenum wet. The rule is simple: the droplet must finish evaporating before it reaches anything it could wet. Small droplets also fall slowly, so they ride the airstream instead of dropping to the floor.
The layout followed the airflow. Nozzles were arranged in rows across the plenum, spraying with the flow. The plenum length ahead of the filters set the residence time, and the residence time set the maximum droplet size. Nozzle spacing came from the spray angle and the overlap needed for even coverage. The first filter bank stayed dry, and that was the design target.
The Engineering Behind the Choice
Psychrometrics explains why the process works. Air holds a limited amount of water vapor, and the wet-bulb temperature marks the limit it can reach by evaporation alone. The difference between dry-bulb and wet-bulb temperature is the depression available for cooling. Published references put the evaporation efficiency of well-designed direct systems at 80 to 95 percent. That range means the air typically reaches within a few degrees of its wet-bulb temperature.
Residence time is the plenum length divided by the air velocity. At 2 m/s, a 4 m plenum gives two seconds of residence. Published droplet evaporation data show that 10 to 20 µm droplets evaporate in a small fraction of a second in hot-dry air. The margin is generous, which is why this class of duty works. The evaporation distance is the droplet’s travel length over that time. Keep the plenum long enough and the spray finishes its work before the filters.
Nozzle density follows a rule of thumb. Published sizing practice for high-pressure fogging uses roughly 1 to 3 nozzles per square meter of intake area. The count depends on nozzle flow and the target temperature drop. The count is set by the water flow needed for the cooling load, then checked against even coverage. Each nozzle in the 70 bar class delivers a small flow at very fine droplet size. More nozzles at lower flow give better coverage than fewer nozzles at higher flow.
Water consumption falls out of the physics. One liter of water removes about 2257 kJ when it evaporates. A cooling effect of 1 kW therefore needs about 1.6 L/h of evaporated water. Add spray inefficiency and bleed, and the working figure lands in the 1.5 to 2 L/h per kW range. For example, a 500 kW precooling effect at peak draws roughly 750 to 1000 L/h. That is the number the water report carries, and it is why the system only runs when the depression is worth it.
Water quality protects the fine orifices. The 70 bar class nozzles use orifices in the 0.1 to 0.2 mm range, and hard water scales them shut. Published fogging guidance recommends RO or softened water for 10 to 20 µm duty. The facility’s RO supply feeds the misting skid, with a filter train ahead of the pump. Scaling on the nozzle face also grows droplets, which is the failure mode that wets filters.
Drift eliminators guard the coils and the hall. A chevron eliminator bank downstream of the spray catches the few droplets that survive. Published references put drift from well-designed eliminators at a small fraction of a percent of water flow. The pressure drop of the eliminator is small compared to the pad it replaces. Where the layout allows, the eliminators also give the fog extra distance to finish evaporating.
Season operation keeps the water out of winter. The system runs only when the wet-bulb depression justifies it, typically in hot-dry months. Below a set ambient threshold the dampers open and the plant runs on outdoor air alone. The water lines drain down on a schedule, and the nozzles purge with air after each run. Published practice for evaporative cooling in data centers treats the seasonal shutdown as part of the design, not an afterthought.
The Results
No site audit numbers appear in this table. The figures come from published reference ranges for direct evaporative cooling and for evaporative precooling in data center HVAC. They are the band this class of system delivers, not a promise for a specific hall.
| Metric | Typical published result | Why |
|---|---|---|
| Intake air temperature reduction | 5 to 10 °C in hot-dry climates | Evaporation efficiency of 80 to 95 percent converts the wet-bulb depression into a dry-bulb drop |
| Evaporative efficiency | 80 to 95 percent | Well-designed direct systems approach the wet-bulb limit within a few degrees |
| Compressor energy reduction | 20 to 30 percent in published precooling case references | Precooled intake air cuts compressor runtime and peak power draw |
| Water use | 1.5 to 2 L/h per kW of cooling | Latent heat of vaporization, about 2257 kJ per liter of evaporated water |
| PUE improvement | 0.1 to 0.3 in published case studies | Mechanical cooling power falls at peak ambient, where PUE is worst |
These are published ranges, not site audit results. The actual outcome depends on the local climate, the wet-bulb depression, and the hours of operation. A humid climate gives a small depression and a small benefit. A hot-dry climate gives the full 5 to 10 °C class of drop. That is why the sizing step starts with the design weather, not with the catalogue.
The facility used the published ranges as the acceptance band for the trial. Intake temperature, filter pressure drop, humidity at the filters, and compressor runtime were compared against the old baseline before the change was made permanent. That is the honest way to run a retrofit. It is also the reason the numbers above are framed as ranges rather than as a single promise.
Why This Case Matters
The sizing logic transfers. Start with the design weather and the wet-bulb floor, then the airflow and plenum length, then the droplet size and nozzle count. That order rules out most of the catalogue before flow is discussed. The cooling nozzle selection guide walks through the same steps for any cooling duty, and it ends with a shortlist instead of a guess.
If your space is tight, check the droplet technology before the flow. The tradeoff between ultrasonic atomizers and high-pressure fogging is covered in the ultrasonic evaporative cooling guide. It compares the two technologies on droplet size, energy input, and maintenance.
The application context matters too. Evaporative cooling and humidification share the same physics and much of the same hardware. The cooling and humidification page explains where this duty sits in the wider range.
The product range for this duty lives on the misting nozzle page. The 70 bar class nozzles, the droplet sizes, and the sizing path are the same ones that produced this result.
What to send for a sizing, in one list:
- Intake airflow in m³/s or CFM and the plenum cross-section, because nozzle density comes from area and velocity.
- Design weather for the hottest month, dry bulb and wet bulb, because the wet-bulb floor sets the possible drop.
- Plenum length from the spray bank to the first filter or coil, because residence time sets the maximum droplet size.
- Target cooling load in kW, or the supply temperature you need, because that sets the water flow.
- Water supply quality and pressure, because RO or softened water protects the fine orifices.
- Seasonal operating profile and freeze risk, because drain-down and bypass change the skid design.
- Filter and coil arrangement, because the spray bank must not wet them.
Each item changes the answer, so a complete list returns a sized configuration in one round.
Start with the design weather and the plenum drawing. Send both through the contact page. BoreJet comes back with the model, the count, and the water flow, sized to your intake rather than guessed from a catalogue. The sizing path above is the same one that produced this result.
Your Duty May Differ
Send the Vessel Drawing and the Pump Curve.
We size the head class, the count and the operating point for your tank, not a catalogue guess. A coverage test on the actual vessel beats any estimate.