Direct evaporative cooling passes outdoor air through wetted media, reducing the dry-bulb temperature by converting sensible heat to latent heat through water evaporation. The cooling effectiveness is governed by the psychrometric saturation efficiency: ε = (T_dbin - T_dbout) / (T_dbin - T_wbin), where T_dbin and T_dbout are the entering and leaving dry-bulb temperatures, and T_wbin is the entering wet-bulb temperature. A high-quality Celdek media pad achieves ε = 0.85-0.90, meaning the leaving air temperature approaches 85-90% of the wet-bulb depression (the difference between dry-bulb and wet-bulb temperatures). In a Phoenix, AZ environment with T_db = 43°C and T_wb = 18°C (a typical monsoon-condition wet-bulb depression of 25°C), the leaving temperature is 43 - 0.85 × 25 = 21.75°C—well within the ASHRAE Class A1 server inlet range of 15-32°C. However, the leaving air is nearly saturated: relative humidity approaches 95-98%, which exceeds the ASHRAE Class A1 upper humidity limit of 80% RH at 21.75°C. The condensation risk on server components and the potential for corrosion from water-saturated air require careful monitoring and, in many cases, a post-cooling dehumidification step that erodes the energy advantage of evaporative cooling.

The water consumption of direct evaporative cooling is substantial: roughly 0.5-1.0 gallons per ton-hour of cooling, or approximately 0.004-0.008 GPM per kW of IT load. For a 500 kW edge AI pod, the peak water consumption is 2-4 GPM, or 2,880-5,760 gallons per day. In water-stressed regions where WUE (Water Usage Effectiveness) is increasingly regulated under LEED v4 and local watershed management plans, this water consumption may be untenable. Adiabatic cooling—where water is sprayed onto the intake of a cooling coil or radiator rather than into the airstream—consumes 60-80% less water because evaporation occurs on the coil surface rather than in the supply airstream, and the water that does not evaporate is recirculated rather than exhausted. The trade-off is a higher leaving air temperature: the coil-based adiabatic approach achieves T_dbout ≈ T_wbin + 3-5°C compared to T_dbout ≈ T_wbin + 1-2°C for direct evaporative media. For a 100 kW GPU cluster at 30 kW/rack density, this 2-3°C temperature difference translates to an additional 5-8% fan power at the server level to maintain the target GPU inlet temperature.

The hybrid evaporative-plus-chiller configuration is the most common deployment for AI-optimized data centers seeking to balance PUE, WUE, and geographic deployment flexibility. In this architecture, the evaporative cooling stages are used during ambient conditions where the wet-bulb temperature is below the chilled water supply setpoint (typically 10-15°C for a 22°C server inlet), providing "free cooling" with chiller bypass. When the wet-bulb temperature exceeds the supply setpoint, the chiller operates in series with the evaporative stage, using the evaporative cooler to reduce the chiller's condensing temperature. The chiller's COP improves by 15-25% when its condensing temperature drops from 40°C (air-cooled condenser) to 24°C (with evaporative pre-cooling). The annual energy savings depend on the climate zone: a facility in Northern Virginia (ASHRAE Climate Zone 4) achieves 6,500-7,000 hours of partial or full evaporative cooling per year, reducing the annualized PUE by 0.08-0.12 compared to a chiller-only design, at a water consumption of 200-400 gallons/MWh of IT energy. In contrast, a facility in Singapore (Climate Zone 0, tropical) achieves fewer than 500 hours of free cooling per year, and the evaporative stages operate primarily to boost chiller COP at a higher water cost per MWh of IT energy saved.

The water treatment and scaling prevention engineering for evaporative cooling is an often-underestimated operational cost center. The recirculating water in evaporative cooling systems concentrates dissolved minerals as pure water evaporates, leaving calcium carbonate (CaCO₃), silica (SiO₂), and magnesium silicate scale behind. A typical 500-ton cooling tower recirculates 150 GPM, and if the make-up water has 200 ppm of total dissolved solids (TDS)—typical of municipal water in the Southwestern US—the concentration cycles in the recirculating water will reach 5-8 cycles (1,000-1,600 ppm TDS) before blowdown is required. At 8 cycles of concentration, the scaling potential for CaCO₃ is approximately 12× the saturation limit at a pH of 8.5, requiring either chemical scale inhibitors (phosphonates at 2-5 mg/L, costing approximately $1,500-3,000/year per 500 tons) or a side-stream water softening system (ion exchange at $10,000-15,000 capital cost plus $500-1,000/year in salt regeneration). The PUE benefit of evaporative cooling must be weighed against this water chemistry management cost: for a 10 MW facility, the annual savings from 0.10 PUE improvement ($175,000 at $0.08/kWh) must be reduced by the water and chemical costs ($15,000-25,000/year for a 500-ton system), yielding a net benefit of $150,000-160,000/year—still positive but requiring the operational discipline of chemical water management that many IT teams lack.