How Data Centers Can Measure and Reduce Their Environmental Impact
Data centers are consuming increasing amounts of electricity, water, equipment, and construction materials. The International Energy Agency expects global data center electricity consumption to increase from about 485 TWh in 2025 to approximately 950 TWh in 2030, equivalent to around 3% of global electricity demand by the end of the decade.
The challenge is no longer to state a sustainability ambition. Operators need to be accountable with auditable measurements that show where resources are consumed, what emissions result, and whether an efficiency project produced a real improvement.
Uptime Institute’s 2025 sustainability survey found that 79% of data center owner and operator respondents compiled and reported electricity use and 72% reported power usage effectiveness (PUE). However, only 45% reported water use, 31% reported e-waste or equipment lifecycle impacts, and 26% reported Scope 1, Scope 2 and Scope 3 carbon emissions.
Closing that gap requires two connected disciplines:
- Sustainability measurement quantifies electricity, greenhouse-gas emissions, water, fuels, refrigerants and equipment lifecycle impacts.
- Operational environmental monitoring measures the physical conditions that drive risk and efficiency, including rack inlet temperature, exhaust temperature, humidity, airflow, differential pressure, cooling status and leaks.
A temperature alarm protects equipment; it does not calculate a carbon footprint. Conversely, a monthly utility bill can support carbon accounting but cannot show which rack is experiencing recirculated hot air. A useful monitoring program connects these two layers.
Why data center environmental measurement matters now
The scale and density of modern computing make small inefficiencies expensive. AI clusters can place high, rapidly changing loads on electrical and cooling systems. At the same time, customers, investors, and regulators expect operators to disclose environmental performance using consistent definitions.
In the United States, the Department of Energy and Lawrence Berkeley National Laboratory estimated that data centers consumed 176 TWh in 2023, or 4.4% of national electricity use. Depending on market and technology developments, demand could reach 325–580 TWh in 2028, representing 6.7%–12% of US electricity consumption.
In the European Union, reporting is no longer entirely voluntary. Under the Energy Efficiency Directive and Delegated Regulation (EU) 2024/1364, data centers with installed IT power demand of at least 500 kW must report energy and sustainability indicators annually to the European database. The required information includes energy use, power utilization, temperature set points, water use, renewable energy, and waste-heat reuse.
What should a data center measure?
A practical environmental-impact inventory should cover six areas.
1. Electricity and energy
Measure electricity at the utility entrance and at major downstream loads. At minimum, separate IT equipment from cooling and other facility infrastructure. More granular monitoring can isolate chillers, cooling towers, pumps, computer-room air handlers, UPS losses, lighting, individual distribution panels and tenant or rack loads.
Collect real power and energy, not just nameplate ratings. Time-synchronized interval data reveals peaks, partial-load behavior and energy that a monthly bill conceals. It also provides the numerator and denominator needed for PUE.
2. Greenhouse-gas emissions
The Greenhouse Gas Protocol divides corporate emissions into three scopes:
- Scope 1: Direct emissions from sources the operator owns or controls, such as backup-generator fuel and leaked refrigerants.
- Scope 2: Indirect emissions from purchased electricity, steam, heating or cooling. Electricity should be calculated using the applicable location-based and market-based methods where required.
- Scope 3: Other value-chain emissions, potentially including construction, purchased servers and network equipment, upstream fuel and energy activities, transport, waste and end-of-life treatment.
Operational sensors can provide activity data such as kWh, fuel consumption, run hours or refrigerant leakage alarms. Turning that activity into a defensible greenhouse-gas inventory also requires appropriate emission factors, ownership boundaries, procurement records and accounting controls.
3. Water
Water usage effectiveness (WUE) is important for facilities using evaporative cooling, but site water is only part of the picture. The electricity supplying a data center can also have an indirect water footprint through power generation.
The 2024 LBNL US Data Center Energy Usage Report estimated that US data centers directly consumed about 66 billion liters of water in 2023. The estimated indirect water associated with their electricity use was nearly 800 billion liters. Those national estimates should not be treated as a site benchmark; they demonstrate why operators should consider both direct and source water.
A lower site WUE is not automatically the best environmental outcome. Replacing evaporative cooling with a waterless system may reduce local water withdrawals while increasing electricity use. Evaluate water availability, local scarcity, seasonal conditions, energy consumption and carbon intensity together.
4. Cooling and thermal performance
Cooling measurements explain a large share of non-IT energy consumption. Useful data includes rack inlet and exhaust temperatures, rack temperature difference (Delta T), humidity, pressure differential, fan speed, chilled-water supply and return temperatures, cooling-unit status and external weather conditions.
Room averages are not enough. A data hall can have a reasonable average temperature while individual racks experience hot-air recirculation, cold-air bypass or overcooling. Top-, middle- and bottom-of-rack readings make these gradients visible.
5. IT utilization and useful work
PUE measures facility overhead, not IT load productivity. A data center can report an excellent PUE while operating large numbers of underused servers. Track server utilization, workload throughput, idle equipment, virtualization or consolidation ratios, and an application-appropriate measure of useful work per unit of energy.
This is especially important for AI and high-performance computing, where chip, server and cluster utilization can materially affect energy used per training job, inference request or scientific calculation.
6. Equipment lifecycle and e-waste
Hardware production and replacement create embodied emissions and electronic waste. The ITU’s Global E-waste Monitor 2024 reported that the world generated 62 million tonnes of e-waste in 2022, while only 22.3% was formally collected and recycled in an environmentally sound manner.
Data center operators should record equipment purchase dates, expected and actual service life, reuse or refurbishment, spare inventory, vendor take-back arrangements, disposal route and recycling certificates. Extending the safe useful life of equipment, redeploying it and selecting repairable hardware can reduce both capital expenditure and lifecycle impact.
The core data center sustainability metrics
No single ratio describes the full environmental performance of a data center. Use a small set of complementary indicators and retain the absolute totals behind them.
Power Usage Effectiveness (PUE) | Total data center energy ÷ IT equipment energy | Facility energy overhead relative to IT energy | Does not measure IT productivity, carbon or water; comparisons require consistent boundaries and conditions |
Carbon Usage Effectiveness (CUE) | Annual data center CO2e emissions ÷ annual IT equipment energy | Carbon emissions associated with each unit of IT energy | Depends on inventory boundary, emission factors and electricity-accounting method |
Water Usage Effectiveness (WUE) | Annual data center water use ÷ annual IT equipment energy | Site water consumed per unit of IT energy, commonly L/kWh | Does not by itself show water scarcity, water quality or indirect water from electricity |
Renewable Energy Factor (REF) | Renewable energy attributed to the data center relative to total data center energy | Share of energy supplied from renewable sources | Procurement quality, timing and additionality are not captured by one percentage |
Cooling Efficiency Ratio (CER) | Cooling-system heat removal relative to cooling energy, using the applicable ISO boundary | Efficiency of the cooling system | Requires consistent measurement points and operating-condition context |
IT utilization or useful work per energy | Workload-specific output or utilization ÷ energy | Whether IT energy produces useful computing work | There is no universal workload unit for every data center |
PUE, CUE and WUE are standardized in the ISO/IEC 30134 series. In 2026, ISO published the second edition of ISO/IEC 30134-2:2026 for PUE. ISO/IEC 30134-8:2022 defines CUE, ISO/IEC 30134-9:2022 defines WUE, and ISO/IEC 30134-7:2023 covers CER. Use the current edition applicable to the reporting period and document the measurement category and boundary.
Always report absolute figures alongside ratios. If IT load grows faster than efficiency improves, PUE may fall while total electricity, emissions and water still rise. Both views matter.
How to build a reliable measurement program
Define the boundary and baseline
State exactly what the assessment includes: the IT room, an entire facility, a campus or a portfolio. Record treatment of shared cooling, offices, renewable generation, exported heat and colocation, and tenants. Select a baseline period long enough to represent seasonal and workload variation.
Instrument the major resource flows
Start with revenue-grade or appropriately accurate meters at the utility service and IT distribution. Add submeters to major cooling and power-conversion loads. Place water meters at relevant incoming, cooling-tower, makeup and discharge points. Deploy thermal, humidity, differential-pressure, leak and equipment-state sensors where they can explain energy and water behavior.
Integrate time-series data
Bring utility, power, cooling, environmental and IT data onto a common timeline. Store units, time zone, sampling interval, device identity, calibration history and data-quality flags. Normalization by weather, IT load or operating hours can help teams compare periods without hiding absolute consumption.
Validate before reporting
Check that submeter totals reconcile reasonably with utility meters, identify gaps and duplicated readings, review anomalous values and preserve the original data. EU rules for covered data centers require retaining information about measurement points and devices, reinforcing the need for a traceable data trail.
Set both absolute and intensity targets
An intensity target such as PUE or kWh per compute unit supports operational optimization. An absolute target for electricity, water or CO2e shows whether total impact is falling. Use both so growth does not make an efficiency gain look like an overall reduction.
Implement, measure and verify
Record conditions before a change, make one controlled adjustment where practical, and compare the result over a representative period. Verify that a cooling improvement did not create rack hotspots, humidity problems, or reduced resilience. Then standardize successful changes across similar rooms or sites.
Practical ways to reduce environmental impact
Eliminate cooling waste
Use rack-level temperature and pressure data to identify bypass air, recirculation, missing blanking panels, containment leakage and excessive fan speed. Adjust cooling set points and airflow only within the supported environmental envelope of the installed equipment. A higher room set point is not a goal in itself; safe rack inlet conditions matter.
Improve IT utilization
Find and decommission unused servers, consolidate suitable workloads, use power-management features and schedule flexible workloads when lower-carbon electricity is available. Procurement should include performance per watt, expected utilization and lifecycle impact—not only peak computing performance.
Optimize water in local context
Measure cooling-tower cycles, leaks, blowdown and treatment performance. Compare cooling modes using electricity, direct water, source water, local scarcity and operating risk. The best option in a cool, water-abundant region may not be appropriate in a hot, water-stressed location.
Manage fuels and refrigerants
Track generator testing and runtime, fuel consumption and refrigerant inventory. Maintain cooling equipment to prevent leaks and evaluate lower-global-warming-potential refrigerants when systems are replaced, subject to safety, performance and regulatory requirements.
Extend hardware life and improve circularity
Match refresh cycles to operational need instead of a fixed calendar. Redeploy serviceable equipment, retain repair and component histories, sanitize data securely and use qualified recycling or manufacturer take-back programs. Request lifecycle and embodied-carbon information from suppliers where it is available.
Treat renewable energy as part of a larger strategy
Renewable procurement can reduce reported electricity emissions, but it does not remove the need to reduce waste. Evaluate the location, timing and quality of energy claims and report them consistently under the selected accounting standard.
Turning sensor facts into efficiency decisions with AKCP
Environmental performance improves when the team can connect a high-level KPI to a physical cause. AKCP monitoring provides the operational layer for that process.
AKCP Cabinet Thermal Map Sensors measure intake and exhaust temperatures at the top, middle, and bottom of a rack, together with humidity and rack Delta T. This helps reveal hotspots, recirculation, poor heat removal, and areas that may be overcooled. For airflow investigations, the Cabinet Analysis Sensor adds differential-pressure measurement.
AKCP power monitoring sensors and powerProbeX+ can provide circuit-level electrical data for energy audits, rack or tenant monitoring and load analysis. Select the right metering architecture and accuracy class for the intended operational, billing, or regulatory use.
Quicklime DCIM brings sensor and infrastructure data into a common monitoring environment. Its sensorCFD capability uses physical sensor readings to constrain thermal and airflow analysis, helping operators identify issues such as hot-air recirculation, cooling imbalance and wasted airflow. The analytical output remains grounded in measured conditions: sensor facts first, AI-supported interpretation second.
This data can support PUE analysis, cooling optimization and verification of efficiency projects. A complete CUE, WUE, or corporate greenhouse-gas inventory will also require data beyond the monitoring system—for example, utility and water meters, contractual electricity information, approved emission factors, fuel and refrigerant records, and equipment-lifecycle documentation.
For a broader deployment overview, see AKCP data center monitoring and optimization.
Frequently Asked Questions (FAQ)
What is the environmental impact of a data center?
It includes electricity use, greenhouse-gas emissions, direct and indirect water consumption, refrigerant and backup-fuel emissions, embodied impacts from buildings and equipment, and electronic waste. Material impacts vary by facility design, energy supply, climate, workload, and accounting boundary.
Is PUE enough to measure data center sustainability?
No. PUE measures facility energy overhead relative to IT energy. It does not show whether servers are well utilized, how carbon-intensive the electricity is, how much water is consumed, or the lifecycle impact of equipment. Use it with absolute energy, CUE, WUE and workload or utilization metrics.
What is a good PUE?
There is no universally fair target for every facility. Climate, redundancy, age, load level, and measurement boundary affect the result. A consistent, standards-aligned trend for the same facility is often more actionable than an unqualified comparison with a very different data center. The best result is the lowest practical overhead that preserves equipment requirements and resilience.
How is data center water use measured?
Install meters at the defined site boundary and important process points, then calculate WUE using water consumption and IT energy over the same period. Also assess water source, local scarcity, seasonal use, and indirect water associated with electricity generation.
What environmental sensors does a data center need?
Most facilities need rack inlet and exhaust temperature, humidity, water-leak detection, power and equipment-state monitoring. Differential pressure, airflow, chilled-water temperatures, and other sensors may be needed depending on the cooling design. Placement should reflect risk and thermal gradients rather than relying only on room size or a single ambient sensor.
Does liquid cooling automatically reduce environmental impact?
No. Liquid cooling can improve heat transfer and enable high-density computing. Still, the result depends on pumps, heat-rejection design, supply temperatures, water use, refrigerants, workload utilization, and whether recovered heat is used. Compare total system energy and water under actual operating conditions.
From sustainability claims to measurable performance
Data center environmental impact cannot be managed with an annual utility total or a room-average temperature alone. Operators need consistent KPI boundaries, calibrated resource meters, granular physical sensors and accounting records that can withstand review.
Start with the largest flows: facility and IT electricity, cooling, water and equipment utilization. Add carbon, refrigerant and lifecycle records. Then use measured operational data to locate inefficiencies, implement changes, and verify results.
AKCP’s sensor-backed monitoring and Quicklime analytics help connect sustainability targets to the physical conditions inside the data center. Contact AKCP to discuss an environmental and power monitoring design for your facility.
- About the Author
- Latest Posts
For over two decades at AKCP, I have been focused on a single mission: bringing complete visibility, security, and efficiency to the world’s critical infrastructure.
I believe that in the modern data center, AI is only as good as the data it receives. My goal is to ensure facilities have the precise sensor facts needed to control AI opinions, ultimately reducing PUE, releasing stranded capacity, and ensuring maximum uptime.
