Data centers can use less energy by reducing unnecessary IT load, improving airflow and cooling controls, and matching flexible computing work to efficient times or locations. The best measures depend on a facility’s climate, equipment limits, water availability, workload requirements, and operating plan.
Start with measurement and the IT load
Cooling energy follows the heat produced by IT equipment. That makes server and application efficiency a useful first place to look: reducing avoidable compute demand can lower both IT electricity use and the cooling burden it creates. The U.S. Department of Energy’s Federal Energy Management Program (FEMP) puts measures involving IT systems and their environmental conditions first because savings can carry through to mechanical and electrical systems.
As an Amazon Associate I earn from qualifying purchases.
Find idle capacity before buying new cooling equipment
- Build an inventory. Record servers, applications, utilization, power draw where available, and operational or compliance constraints. DOE FEMP recommends tracking hardware and applications and identifying unused or underused servers.
- Consolidate where the workload allows. Virtualization can run applications in separate environments on shared servers, potentially reducing the number of machines that need to be powered and cooled. Reassign or turn off equipment only after confirming that capacity, resilience, and service requirements remain covered.
- Review the equipment itself. Efficient processors, fans, power supplies, and networking, along with storage consolidation, can reduce IT load. Assess purchases against the performance the workload actually needs rather than treating peak capacity as the normal operating state.
DOE’s 2024 guide reports that average server utilization in enterprise settings is generally 20% to 40%; this is a broad range, not a benchmark for every organization. The same guide cites Rahkonen and Dietrich (2023) for a result that server efficiency rises by about 50% when processor utilization is doubled from low levels of 20% to 30%. That cited comparison is not a guaranteed saving for a particular server fleet: workload behavior, hardware, and the definition of efficiency all matter.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Improve airflow before assuming a major cooling replacement is needed
Cooling systems can spend energy moving air that does not effectively remove heat from equipment. In hot-aisle/cold-aisle layouts, supply air is directed toward equipment inlets and hot exhaust is kept separate. Poor separation can let hot air recirculate to inlets or let cold supply air bypass equipment, forcing fans and cooling systems to work harder.
Check what the equipment actually receives
- Measure temperature at equipment inlets across racks and locations, not only at a room sensor. Look for hot spots and uneven conditions that may be hidden by an acceptable room average.
- Inspect for bypass paths and recirculation between supply and exhaust air. Seal gaps where appropriate and check whether rack or aisle arrangements are undermining separation.
- Consider airflow accessories, such as blanking panels or containment, when they address a measured problem and suit the facility layout. They are interventions to evaluate, not a guarantee of a particular energy saving.
DOE’s data-center cooling toolkit explains that temperature differences and uncontrolled airflow can waste energy. Use measurements before and after an airflow change to establish whether conditions improved and whether fan or cooling controls can be adjusted safely.
Tune cooling controls and use economizing when conditions permit
Once airflow and inlet conditions are understood, review fan and pump operation, control sequences, and cooling setpoints. DOE FEMP’s 2024 best-practices guide recommends optimizing fan and pump speeds, maximizing compute inlet temperature within applicable IT thermal guidelines, and maximizing compute leaving temperature where that supports heat reuse or dry heat rejection. These are operating goals bounded by equipment specifications and site conditions, not permission to exceed thermal limits.
Use outdoor air only when it is suitable
Air-side economizing uses cool outdoor air in place of some or all compressor-based cooling when conditions allow. Its value depends on climate, operating hours, temperature and humidity settings, air quality, and the IT equipment’s tolerance for those conditions. Evaluate filtration, humidity control, and operating procedures as part of the design; outdoor air is not automatically a safe or efficient choice year-round.
Compare liquid, hybrid, and air-based options as systems
Direct liquid cooling and hybrid approaches may reduce PUE and WUE in some applications, but they add control loops and maintenance requirements. A choice that reduces facility overhead may have different water, reliability, serviceability, or heat-reuse implications. DOE FEMP cautions that there is no single most energy-efficient data-center design for every scenario. Compare options against site-specific operating conditions rather than selecting a technology on its name or a single metric.
Rank #3
Use workload flexibility for suitable jobs
Some computing work can move in time or between locations; some cannot. DOE/Lawrence Berkeley National Laboratory demand-response material identifies job shifting or queuing, power capping, server power management, and virtualization or migration to another facility as flexibility options. LBNL’s Center of Expertise also describes work on optimized controls, workload management, and energy storage to support flexibility while meeting operational requirements.
Match the tactic to the service requirement
- Queue or shift jobs when deadlines permit running them at a different time.
- Power-cap servers when a temporary reduction in peak demand is acceptable and the impact on throughput is understood.
- Use server power management to manage equipment that does not need to remain at full operating capacity.
- Migrate or virtualize workloads only where security, data location, network capacity, and the destination facility’s efficiency and capacity support the move.
Before changing scheduling or placement, account for latency, deadlines, security, availability, and service-level requirements. Shifting computation can change when or where electricity is drawn without reducing the total electricity needed to complete the work. Where algorithms can be improved, they may also affect energy demand: DOE/NREL’s 2024 guide notes that efficient algorithms can have a large impact, particularly in AI and machine-learning fields, while noting that its guide focuses on hardware rather than algorithm design.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Track energy, water, and useful work together
Facility energy, IT energy, water use, and computing output answer different questions. Track them together so that a lower facility-overhead ratio is not mistaken for lower total consumption or better service.
Recommended Free Tools
| Measure | What it tells you | How to interpret it |
|---|---|---|
| PUE (Power Usage Effectiveness) | Total facility energy divided by IT equipment energy. | A lower PUE indicates less non-IT overhead relative to IT energy. It does not show how much useful computing was completed, and it can improve even while total energy rises if IT load grows. |
| WUE (Water Usage Effectiveness) | Site water use relative to IT equipment energy, expressed in liters per kWh in the cited DOE guidance. | Pair it with energy measures when cooling choices affect water use; an energy result alone does not describe water performance. |
| Work per watt | Computing output relative to power; DOE’s guide discusses server efficiency as transactions per second per watt. | Choose an output measure that reflects the workload, then compare it consistently across the systems or operating states being evaluated. |
DOE FEMP reports that the National Laboratory of the Rockies data center achieved a PUE of 1.06 and WUE of 0.7 in a hybrid-cooling application. Those are results for that specific site and application, not predictions for another facility. Separately, a DOE cooling-controls case study attributes more than 2.3 million kWh of annual energy savings to a demonstration at California data centers; it is a case-study result, not a general savings estimate.
Quick Recap
Build a site-specific improvement sequence
- Establish a baseline. Gather facility and IT energy, inlet temperatures, airflow observations, cooling-system operation, water use where relevant, and workload utilization. Record service and reliability constraints alongside the readings.
- Address avoidable IT demand. Identify underused equipment and consolidation opportunities, then confirm capacity and resilience before changing production workloads.
- Correct airflow and control issues. Resolve measured bypass or recirculation, then tune fans, pumps, and setpoints within equipment limits.
- Evaluate cooling choices. Assess economizing, liquid or hybrid cooling, and heat-reuse potential against climate, air quality, water, maintainability, and operating capability.
- Test workload flexibility selectively. Start with jobs whose latency, deadline, security, and service requirements allow queuing, power management, capping, or migration. Measure energy and service outcomes after the change.
- Review the combined result. Compare total energy, IT energy, PUE, WUE where applicable, useful work, and reliability against the baseline. Keep measures that improve the outcome that matters without creating unacceptable operational risk.
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.
