The Tool Desk
Outbyte PC Repair FREERepair Windows errors before they cause bigger problemsFix Now →Outbyte Driver Updater FREEScan for outdated or missing drivers - takes under a minuteDriver Scan →iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
There is no universal best cooling method for AI. Compare the complete thermal system: server-side cooling, coolant temperatures, coolant distribution unit (CDU), facility loop, heat-rejection plant, climate, water boundary, redundancy and operating controls. Air cooling can remain suitable at lower rack densities when airflow is well managed. Direct liquid cooling moves heat closer to high-power components and can support denser racks, but most installations still need some air cooling and a separate way to reject heat. The right choice is the one that meets your equipment limits and annual energy, water, resilience and expansion targets at your site.
What you are actually comparing
A label such as “liquid-cooled” describes only part of a data-center heat path. Start at the processor, memory and other heat-producing components, then follow heat through the rack, room, facility loop and outdoor heat-rejection equipment. A fair comparison uses the same IT load, climate, operating schedule, measurement boundary and reliability target for every option.
ASHRAE’s AI framework says that purpose-built AI facilities where compute densities routinely exceed 50–120 kW per rack should use a technology cooling system (TCS), with the potential for higher densities. That is framework guidance, not a universal changeover point. Your server design, rack layout, supply temperatures and allowable inlet conditions may justify a TCS at a different density. See the ASHRAE Energy and Thermal Efficiency guidance.
Recommended Free Tools
Component cooling versus heat rejection
Direct liquid cooling captures heat at the component and transfers it into a liquid loop. A CDU commonly separates or controls the IT loop and the facility loop, regulating flow, pressure, temperature and chemistry. The facility still must reject that heat outdoors or to a useful heat sink. A hybrid design may send processor heat to liquid while fans remove memory, storage, power-supply and other residual room heat.
Air cooling takes all server heat into room air. Computer-room air handlers, chillers, dry coolers, cooling towers, economizers and containment then move and reject that heat. The room system, not just the server fan, determines energy use and operating limits.
Build a like-for-like comparison
Use the following axes in a requirements document before requesting proposals. Record measured values separately from modeled values and state the year, climate, load profile and boundary for every result.
| Axis | Questions to answer | Evidence to request |
|---|---|---|
| IT and rack compatibility | What servers and accelerators are supported? What rack density is planned? How much heat is captured by liquid, and how much remains in room air? Which supply-temperature class and flow, pressure and chemistry limits apply? | Server and rack specifications, CDU data sheets, commissioning limits and the facility-loop design. |
| Energy | What are annual facility and IT energy use, PUE, fan, pump and chiller loads, and economizer hours? | An hourly or sub-hourly model using the same IT load and weather year, followed by operating measurements. |
| Water | How much site water is used, from which source, for evaporation, adiabatic assist, treatment and discharge? | WUE calculation with an explicit site-water boundary, makeup and blowdown accounting, and local water constraints. |
| Climate and heat rejection | How often can dry, evaporative or hybrid equipment operate, and at what ambient temperature and humidity? | Climate-specific modeled performance and, where available, measured annual results from a comparable installation. |
| Resilience and maintenance | What fails closed or open? How are leaks detected? Which pumps, controls, water-treatment systems and heat-rejection units are redundant? | Failure-mode analysis, maintenance procedures, spare-parts plan, alarm matrix and integrated-system test results. |
| Deployment and expansion | Is this a greenfield build or retrofit? Can the facility loop, electrical service and heat-rejection plant support later racks? | Capital and operating estimates tied to a defined scope, phasing plan and connection points. |
| Useful work and environmental impact | How will you report CUE, WUI, DCRE, IT Work Capacity or heat reuse in addition to PUE and WUE? | Definitions, boundaries and raw data for each selected metric, following the terminology described by ASHRAE’s AI framework introduction. |
Air cooling: when it still fits AI
Air cooling removes heat from servers into the room and relies on airflow management and room-level equipment. It remains a practical choice when rack heat, server inlet limits and expansion plans stay within the capability of the air system. Do not use a single watts-per-socket or rack number as a universal cutoff; equipment generations and facility designs differ.
What to examine
- Containment and airflow: Separate supply and exhaust paths, seal bypass openings and size fans for the actual pressure drop. Poor containment can make a capable cooling plant perform badly.
- Setpoints: Excessively low room temperature and unnecessarily narrow humidity control increase cooling demand. The U.S. Department of Energy’s cooling-water guidance identifies these controls as opportunities to reduce energy and water use.
- Density distribution: A moderate average rack load can hide a few high-density racks that create local hot spots. Check row-level airflow and worst-case rack conditions rather than relying on room averages.
- Service model: Air systems are familiar to many operators, but filters, fans, coils and chillers still require inspection, cleaning and redundancy testing.
Air cooling avoids an IT liquid loop and CDU, which can simplify a retrofit. Its limitation is heat-transfer capacity: as component power rises, moving enough air can consume fan energy, require larger distribution paths and make inlet-temperature control more difficult.
Direct liquid cooling: what changes
Direct liquid methods place a cold plate or another liquid interface close to processors and memory. DOE describes cold-plate and immersion technologies as direct liquid cooling and lists treated water, glycol-based solutions and dielectric fluids among possible IT-loop fluids. In a cold-plate design, the CDU interfaces the controlled IT loop with the facility loop. In immersion, equipment is surrounded by a dielectric fluid and the heat is transferred through a tank-side loop.
DOE’s explanation is explicit: “Direct liquid cooling systems transfer the heat generated from the IT equipment directly to a recirculating chilled water loop rather than transferring the heat to the room air and then moving the heat from the air to the chilled water loop.” Read the full Maui High Performance Computing Center liquid-versus-air evaluation for the system context.
Benefits to test, not assume
- Heat is captured at the source, which can support higher compute density and reduce the amount of room air that must be moved.
- Higher coolant temperatures may create more hours for dry heat rejection or heat reuse, depending on the equipment class and outdoor conditions.
- Liquid can remove heat from components whose power would make an all-air design impractical.
Trade-offs to quantify
- Cold plates, manifolds, hoses, quick disconnects, CDUs, pumps and controls add equipment and leak-management tasks.
- Residual heat from fans, memory, drives, power conversion and other components may still enter the room, requiring an air system.
- Coolant chemistry, filtration, pressure, flow and material compatibility become part of commissioning and maintenance.
- Changing server generations can require new cold plates, connectors or CDU capacity; verify the upgrade path before selecting a rack standard.
ASHRAE identifies liquid cooling as an enabler for high-compute-density AI installations and links warmer loops with possible heat reuse and water-free heat rejection. Those are design opportunities, not automatic outcomes. Confirm the actual server limits and annual plant performance with the ASHRAE tools, standards and resources and your equipment documentation.
Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Evaporative, dry, adiabatic and hybrid heat rejection
Evaporative rejection
Cooling towers and other evaporative systems reject heat by evaporating water. They can provide efficient heat rejection in suitable weather but consume makeup water and discharge concentrated water through blowdown. Include treatment, drift control and local discharge requirements in the water account.
Dry rejection
Dry coolers pass outdoor air across a heat exchanger and can avoid routine evaporative consumption. Their electrical demand and available capacity vary with ambient temperature, approach temperature, coil size, fan speed and redundancy. A dry system that performs well in a cool climate may need more fan energy or supplemental equipment during hot hours.
Adiabatic and hybrid operation
Adiabatic systems add water to incoming air under selected conditions; hybrid systems switch between dry and evaporative modes. This can preserve efficiency during heat events while using little or no water for much of the year. The controls, water quality and transition sequences must be commissioned as part of the complete plant.
Uptime Institute’s April 2026 analysis reports that, in its examined sample, well-designed dry and adiabatic systems could match evaporative PUE across ASHRAE climate zones 2–6 with zero or near-zero water consumption. The report also emphasizes variation within cooling types. Treat that finding as a result for the studied systems and conditions, not a guarantee for a proposed facility; review the Uptime Institute analysis.
Check coolant-temperature compatibility
ASHRAE liquid-cooling classes W17, W27, W32, W40, W45 and W+ express increasing upper supply-temperature limits, with W+ extending beyond 45°C. The class name alone does not prove that a server, CDU or facility loop can operate at that temperature. Verify the equipment’s allowable inlet range, transient limits, flow, pressure, fluid chemistry, filtration and dew-point controls.
| Compatibility item | Why it matters |
|---|---|
| Supply and return temperature | Determines whether dry, evaporative or heat-reuse operation is possible and whether component temperatures remain within specification. |
| Coolant chemistry | Incompatible inhibitors, metals or dielectric fluids can cause corrosion, deposits, seal damage or warranty problems. |
| Flow and pressure | Insufficient flow raises component temperature; excessive pressure can stress hoses, manifolds and quick disconnects. |
| Dew point and condensation | A cold surface below room dew point can condense moisture. Controls must coordinate coolant temperature and room conditions. |
| Connection and service design | Dripless disconnects, isolation valves, drain-and-fill points and leak detection affect maintenance risk and downtime. |
Use PUE and WUE together
PUE is total annual facility energy divided by annual IT-equipment energy. A lower PUE means less overhead energy for the measured boundary, but it says nothing about water use or how much useful computing the site delivered.
WUE is annual site water use in liters divided by IT-equipment energy in kilowatt-hours. Publish the site-water boundary: include or exclude potable, reclaimed, process, blowdown and water-treatment streams consistently. A closed IT liquid loop does not establish zero facility water use; the heat-rejection plant may still evaporate or consume water.
ASHRAE also identifies WUI, CUE, DCRE and IT Work Capacity as useful measures. Select only metrics for which you can state the definition, time period, boundary and data quality. For a fair annual comparison, report at least IT energy, total facility energy, site water, peak demand, IT load and weather conditions.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Interpret facility examples carefully
DOE reports a PUE of 1.06 and WUE of 0.7 for the National Laboratory of the Rockies data center using a thermosyphon hybrid system. DOE describes the arrangement as direct liquid cooling with air-cooled heat rejection and an open cooling tower, and notes that additional control loops require operational attention. These figures describe that facility and its measurement conditions; they are not typical values or guarantees for every thermosyphon, liquid or hybrid design.
Account for climate and annual operation
Compare hourly operation across a representative weather year rather than a single design-day efficiency. Record ambient temperature and humidity, economizer or free-cooling hours, dry-mode hours, adiabatic hours, water-treatment operation and the percentage of time at each IT load.
Uptime Institute’s 2026 cooling survey collected responses from 1,034 participants about cooling-system use and continued direct-liquid adoption. The respondent count is a survey sample size, not the market share of any technology. Use survey evidence to understand adoption questions, then use site-specific modeling and measurements for a design decision.
Climate can reverse an apparent ranking. In a mild, dry location, a dry cooler may reject heat with modest fan energy for much of the year. In a hot or humid location, adiabatic assistance or evaporative equipment may reduce electrical demand while increasing water use. Compare the same reliability configuration and control strategy in both cases.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsEvaluate resilience and operations
Cooling architecture is an operating system as much as a collection of hardware. Define the failure response for utility loss, pump failure, CDU failure, water-treatment upset, leak alarm, control-network loss and extreme weather.
- Redundancy: Document N, N+1 or 2N capacity for pumps, CDUs, chillers, dry coolers, towers, controls and electrical feeds. Confirm that a maintenance event does not remove the required reserve.
- Detection: Use leak sensors, pressure and flow alarms, isolation zones and automatic or operator-directed shutdown sequences appropriate to the equipment.
- Water quality: Monitor conductivity, inhibitor concentration, biological control and filtration where a water loop is used. Define sampling frequency and corrective action.
- Controls: Test transitions between dry, adiabatic and evaporative modes, staged equipment, setpoint resets and loss-of-sensor conditions. DOE notes that control sequences and operating plans matter in the described hybrid approach.
- People: Train operators on liquid service, confined equipment access, chemical handling, alarm response and recovery after a leak or loss of heat rejection.
- Commissioning: Perform integrated-system tests at low, normal and peak IT load, including failure injection and restart. A component test cannot prove that the rack, CDU, facility loop and heat-rejection plant work together.
A practical comparison procedure
- Inventory the IT load. List server models, accelerator count, memory and storage heat, rack power, inlet limits, expected utilization, growth stages and the fraction of heat that each proposed liquid system captures.
- Set environmental requirements. Record allowable air temperature, humidity, dew point, liquid supply and return temperatures, flow, pressure and chemistry. Map each requirement to the proposed equipment documentation.
- Draw both heat paths. Show component-to-rack, rack-to-CDU, CDU-to-facility-loop and facility-loop-to-outdoor rejection paths. For air systems, show server-to-room and room-to-plant paths. Mark residual room heat and all pumps, fans, chillers and towers.
- Model a full year. Use the same weather file, IT-load profile, redundancy condition and control philosophy for air, direct-liquid, dry, evaporative and hybrid alternatives. Produce hourly energy, water and operating-mode results.
- Calculate comparable metrics. Compute PUE and WUE with stated boundaries, then add peak power, water by source, free-cooling hours, IT Work Capacity or other metrics that have reliable data.
- Test abnormal conditions. Simulate the hottest and most humid hours, partial-load operation, a failed pump or heat-rejection unit, a water restriction, a leak alarm and a control-network failure. Record remaining capacity and recovery time.
- Price the defined scope. Separate server changes, CDUs, facility-loop modifications, heat-rejection plant, electrical work, controls, commissioning, water treatment, maintenance and expansion provisions. Do not apply a generic payback period to a project whose scope and tariffs are unknown.
- Validate with operations. Have the people who will run the site review service access, spares, training, alarm workload, chemical handling and planned maintenance windows before approving the architecture.
- Measure after turnover. Install meters for IT energy, facility energy, water makeup and blowdown, temperatures, flow and operating mode. Compare measured annual results with the model and correct control sequences that miss the design intent.
Common comparison mistakes
- Calling liquid cooling water-free: Liquid at the server does not describe the outdoor heat-rejection plant or the site-water boundary.
- Comparing only PUE: A low PUE can coexist with high water use, poor compute utilization or inadequate resilience.
- Using one facility’s result as a benchmark: The DOE 1.06 PUE and 0.7 WUE example is installation-specific.
- Ignoring residual room heat: Hybrid liquid systems still need air capacity for uncaptured loads and people, lighting and power equipment.
- Assuming a rack threshold is universal: ASHRAE’s 50–120 kW guidance is contextual; server design, climate and expansion plans determine the actual decision.
- Modeling average weather only: Heat waves, humidity and water restrictions often determine capacity and operating cost.
- Leaving controls until the end: Poor sequencing can erase the modeled advantage of dry, adiabatic or hybrid equipment.
- Requesting unsupported savings claims: Require a named facility, year, climate, load, boundary and measured-versus-modeled status for every percentage.
How to make the final decision
Choose air cooling when the equipment and rack plan fit its airflow and inlet-temperature limits, the retrofit value of existing room infrastructure is high, and the annual model shows acceptable fan, chiller and water performance. Choose direct liquid when component heat density, rack expansion or room-air constraints make source-level heat capture valuable and your organization can operate the additional loops and controls. Choose dry, evaporative or hybrid heat rejection according to the climate, water policy, electrical capacity and reliability design—not according to the cooling label alone.
The defensible selection is the one supported by a complete heat-path diagram, equipment compatibility records, hourly energy and water analysis, failure testing, a defined operating plan and metered performance targets. Revisit the comparison when accelerator generations, rack density, utility tariffs, water constraints or the facility’s expansion schedule changes.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
