The most dependable way to reduce electricity and cooling costs in an AI data center is to measure where energy goes, remove avoidable airflow and cooling-system waste, and then match cooling technology to rack density, climate, water constraints, and reliability needs. Start with operational changes that fit the installed facility; consider liquid cooling or major retrofits only after engineering review. No single cooling design suits every site, and published examples do not establish a universal AI-data-center savings rate.
Start with a baseline you can use to judge changes
Meter IT equipment separately from the rest of the facility, and record total facility energy and cooling-system performance over a representative operating period. Calculate power usage effectiveness (PUE) as total facility annual energy divided by IT equipment annual energy. Track water as well where it matters: water usage effectiveness (WUE) is annual site water use relative to IT equipment energy. These are the definitions used by U.S. Department of Energy Federal Energy Management Program (DOE/FEMP) guidance published in 2019.
As an Amazon Associate I earn from qualifying purchases.
Use the ratios to understand facility overhead and track trends, not as a substitute for the utility bill. PUE is not an electricity price or a cost forecast. Compare metered energy with actual bills and operating conditions, and record the workload, weather, and equipment configuration when evaluating a change. Otherwise a workload shift or cooler season can look like a cooling improvement.
Reduce cooling demand before replacing equipment
Separate server intake air from hot exhaust
Arrange racks and supply and return airflow so cool air reaches server intakes without mixing unnecessarily with hot exhaust. Hot-aisle/cold-aisle layouts and containment barriers or flexible curtains around rack areas can help isolate the air streams. DOE/FEMP identifies barriers above and along rack sides as one airflow-isolation approach. Before installing barriers, check rack dimensions and layout, access and egress, fire protection, and site requirements.
Airflow isolation can make it possible to reduce airflow and operate with higher chilled-water temperatures, depending on the system. DOE/FEMP’s 2019 cooling guidance reports that the practices it references can result in 20% less chiller energy. That is a conditional figure for the cited practices, not a guaranteed saving for every facility or an estimate of total data-center electricity reduction.
Review temperature and humidity settings against equipment limits
Facilities are sometimes run colder and within narrower humidity ranges than necessary, increasing chiller demand and potentially water use. Review control settings against the applicable current ASHRAE guidance, equipment specifications, facility classification, altitude, and measured IT inlet conditions. Change setpoints only with appropriate monitoring and operating controls; a room sensor reading by itself does not establish that every server inlet is within its allowable range.
DOE/FEMP’s 2019 resource summarizes ASHRAE ranges, but operators should consult current applicable guidance and equipment documentation before changing limits. The usable range depends on the installed equipment and site conditions.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problemsUse outside-air or water-side economizing where conditions allow
Economizers can reduce or avoid mechanical cooling during suitable conditions. Air-side economizing uses cool outside air; water-side economizing can use a heat exchanger and cooling tower to bypass or reduce chiller-compressor operation in a suitable configuration. The benefit depends on local climate, hours of availability, air quality, humidity, operating setpoints, and system design.
Assess the effect on total energy and water, not just chiller runtime. An economizer is not automatically beneficial year-round, and a design appropriate for one climate or facility may not suit another. DOE/FEMP’s cooling guidance describes these options as conditional on the facility and environment.
Optimize fans, pumps, UPS, and controls
Once airflow and environmental conditions are understood, tune fan and pump speeds to the actual cooling requirement and review controls for unnecessary operation. Assess UPS performance and other electrical overhead alongside cooling, while keeping reliability and total cost of ownership in view. DOE/FEMP’s efficiency guide places efficient IT systems and environmental conditions first, then free cooling where suitable, followed by optimized fan and pump speeds and UPS performance.
Rank #3
Make changes in a controlled sequence: establish the baseline, adjust one relevant control or operating practice at a time, verify thermal and reliability limits, and compare results under comparable workload and weather conditions. This makes it easier to identify which change produced an improvement and to detect unintended effects.
Evaluate direct liquid cooling for high-density racks
Direct liquid cooling transfers heat from IT equipment through a recirculating liquid loop to a coolant distribution unit (CDU), rather than relying only on room air to carry equipment heat. It may be relevant where rack heat density challenges an air-cooling design, but it is a facility and IT-system design decision—not a universal drop-in way to lower costs.
Before adopting it, verify compatibility with the servers and cooling plant, design the heat-transfer and facility-water interfaces, and plan for controls, operations, and maintenance. DOE says some implementations show promise for PUE and WUE, while emphasizing the need for additional controls and an operations and maintenance plan. A lower facility PUE or WUE should not be assumed without site-specific design and measurement.
Consider water, heat reuse, and carbon alongside electricity
Cooling choices can shift impacts between electricity, water, and heat rejection. DOE/FEMP’s design guidance prioritizes reducing energy use, then places waste-heat reuse after those reductions and favors dry heat rejection where possible to save water. Renewable energy supply and carbon performance are further considerations. Evaluate these together with thermal performance, airflow, water availability or stress, and reliability rather than optimizing a single metric in isolation.
For cooling towers, water-treatment practices can affect water consumption without being an electricity-saving measure. DOE/FEMP’s 2019 cooling resource, citing its Cooling Tower Best Management Practice, reports that increasing cycles of concentration from three to six reduces cooling-tower makeup water requirements by 20% and blowdown by 50%. Treat that as a water-management result under the stated change, not as a promise of lower power use.
Interpret published savings as examples, not guarantees
DOE’s 2021 case study reports cooling-energy savings of 53% at a Florida pilot and 74% at a Massachusetts pilot, as well as a $110,000 cooling retrofit at the Massachusetts site. The project proposal had predicted 30% cooling savings; DOE reports that optimizing cooling and airflow together proved essential in the pilots. These are results from two pilot cases, not a forecast for another facility or an AI-specific benchmark.
DOE/FEMP reports PUE 1.06 and WUE 0.7 for the National Laboratory of the Rockies data center. Those are facility-specific metrics, not a generally attainable target. Separately, the DOE Better Buildings & Better Plants Initiative reported in 2020 that 21 Data Center Accelerator partners had a goal of reducing infrastructure energy intensity by 25%; their average improvement was 36%, with annual cost savings of $3.9 million. These historical program results are not current market averages or a prediction for an individual site.
The published examples use different measures and contexts, so they should not be combined into a representative AI data-center savings figure. No comparable AI-only savings statistic is established here.
Choose changes by fit, risk, and measured outcome
DOE/FEMP emphasizes that no single design is the most efficient for every data center. Its guide, updated July 26, 2024, describes its recommendations as applicable across a variety of scenarios rather than as one universal design. In practice, prioritize options using these facility-specific checks:
Recommended Free Tools
- Electricity and cooling: Can you separately meter the affected system and verify the change in energy use?
- Thermal limits and reliability: Will IT inlet conditions remain within equipment requirements under expected loads and operating conditions?
- Climate and availability: How often will economizing conditions occur, and are air quality and humidity suitable?
- Water and site constraints: Does the option increase water dependence where water is scarce, or enable dry heat rejection?
- Compatibility and complexity: Does it fit installed IT and cooling equipment, and can the operations team support its controls and maintenance?
- Total cost of ownership: Do installation, operation, maintenance, and reliability considerations make sense for the site?
The practical order is to measure, improve airflow and environmental controls, use economizing when local conditions support it, tune cooling and electrical overhead, and then assess major design changes such as liquid cooling. Re-check energy, water, and thermal performance after each change rather than relying on a design label or a published case-study percentage.
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.
