Energy, water and sustainability

Data centers use a growing share of the world's electricity, and in some regions a noticeable share of water, while also enabling services that can save energy elsewhere. Their environmental footprint has become a major issue for communities, regulators, customers and investors, and it increasingly decides whether a project is approved. Lesson 1 introduced power usage effectiveness (PUE) and the scale of data center electricity demand. This lesson looks at the full environmental picture: the metrics used to measure performance, how to reduce energy use at each level, how carbon emissions are counted and reduced, how water is used and saved, how waste heat can be reused, and the embodied impacts of buildings and equipment.

The metrics family

PUE measures facility overhead, but it says nothing about where the energy comes from, how much water is used, or how useful the computing is. The Green Grid and ISO/IEC 30134 define a family of metrics:

• WUE (water usage effectiveness): liters of site water used per kWh of IT energy. Values range from near zero for dry-cooled or closed-loop facilities to about 1.8 L/kWh or more for heavily evaporative designs in hot climates.
• CUE (carbon usage effectiveness): kilograms of CO₂ equivalent emitted per kWh of IT energy, which equals PUE times the carbon intensity of the electricity used.
• ERF (energy reuse factor): the fraction of facility energy reused outside the data center, for example as district heating.
• REF (renewable energy factor): the share of energy from renewable sources.

Each metric captures one aspect; improving one can worsen another. Evaporative cooling lowers PUE but raises WUE; a dry-cooled facility saves water but may use more energy in hot weather.

Reducing energy at every level

Energy savings multiply through the chain, so savings in IT equipment are the most valuable: every watt saved at the server also saves the cooling and electrical losses attached to it. In order of leverage:

1. Useful work: run software efficiently, schedule workloads well, and avoid idle capacity. For AI, choose model sizes and inference methods appropriate to the task.

2. IT equipment: use efficient servers and accelerators, consolidate workloads through virtualization, retire unused "zombie" servers, and enable power management.

3. Power distribution: reduce conversion stages (Lesson 3), use high-efficiency UPS modes where risk allows, distribute at higher voltages and size transformers to operate near their best efficiency.

4. Cooling: manage airflow well (Lesson 6), raise supply temperatures within ASHRAE guidelines, maximize economizer hours, use liquid cooling with warm water, and use variable-speed fans and pumps.

5. Building: efficient lighting and controls, and good envelope design.

Carbon emissions

Greenhouse gas emissions are counted under the GHG Protocol in three scopes:

• Scope 1: direct emissions from sources the operator owns, mainly diesel burned in generators during testing and outages, and leaks of refrigerants, which can have very high global warming potentials.
• Scope 2: emissions from purchased electricity, usually by far the largest operational source for a data center.
• Scope 3: all other indirect emissions, including the manufacturing of servers and other equipment, construction materials and transport. For efficient facilities on clean grids, scope 3 embodied emissions can exceed operational emissions.

Scope 2 is reported in two ways. The location-based method uses the average carbon intensity of the local grid. The market-based method uses the emissions of electricity the company has contracted for, such as renewable power purchase agreements (PPAs) or renewable energy certificates. Companies increasingly aim for 24/7 carbon-free energy, matching consumption with clean generation every hour in the same grid region, which is more demanding but more meaningful than annual matching (DCE-420).

Grid carbon intensity varies widely across the US, from regions dominated by hydro, nuclear and wind (such as the Pacific Northwest) to regions still relying heavily on coal and gas. Siting and procurement therefore strongly affect a US data center's emissions, and large operators sign PPAs for new wind, solar, geothermal and nuclear capacity.

Water

Data centers use water directly, mainly for evaporative cooling (cooling towers and adiabatic coolers), and indirectly, through the water used to generate their electricity. Two quantities should be distinguished: withdrawal (water taken from a source) and consumption (water evaporated or otherwise not returned). Evaporative systems consume roughly 1.5 to 2 liters of water per kWh of heat rejected by evaporation in practice, including blowdown (water drained to control mineral concentration).

Water strategies include: dry coolers or air-cooled chillers where climate allows; warm-water liquid cooling that rarely needs evaporation (Lesson 6); hybrid systems that evaporate only on hot days; reclaimed municipal wastewater or captured rainwater; and careful water treatment to reduce blowdown.

Heat reuse

Almost all the energy a data center uses ends up as low-grade heat. Reusing it can displace fossil heating elsewhere. Air-cooled data centers produce heat at 25 °C to 40 °C, usually too cool for direct use without heat pumps; liquid-cooled systems return water at 40 °C to 60 °C, which suits low-temperature district heating, greenhouses, aquaculture or industrial preheating. Heat reuse is common in Nordic countries with district heating networks, and is beginning to appear in North America.

Embodied impacts and circularity

The buildings and equipment themselves carry environmental costs. Concrete and steel account for most of a building's embodied carbon; servers, especially their chips and memory, carry significant manufacturing emissions and use scarce materials. Strategies include:

• Lower-carbon concrete (supplementary cementitious materials such as slag and fly ash), recycled steel and efficient structural design.
• Longer service lives for buildings and infrastructure, designed for adaptation rather than replacement.
• Extending server life where efficiency gains from replacement are small, refurbishing and reusing components, and recycling e-waste responsibly.
• Choosing refrigerants with low global warming potential and preventing leaks.

Measuring and verifying

Metrics are only as good as the measurements behind them. A credible PUE requires metering of total facility energy at the utility connection and of IT energy as close to the IT equipment as practical: at UPS outputs (the simplest, but it counts PDU losses as IT), at PDU outputs, or at the rack or server (the most accurate). The Green Grid describes these as PUE categories of increasing rigor, and ISO/IEC 30134-2 requires the measurement point to be stated. Water metering should separate cooling make-up water, blowdown and domestic use. Annual values are reported, since seasonal variation is large, and trends are tracked monthly to spot problems such as failing economizer controls or leaks.

Reporting and regulation

Disclosure requirements are expanding. The European Union's Energy Efficiency Directive (2023) requires data centers of 500 kW and more to report energy, PUE, water, waste heat and renewable energy to a public database. Some jurisdictions restrict new data centers where grids or water are constrained, or require heat reuse.

In the US, there is no federal data center reporting requirement as of 2026, but states and utilities increasingly set conditions on large new loads, some states require disclosure of energy and water use for incentives, and local governments scrutinize water and noise during approvals.

Large technology companies publish sustainability reports with energy, water and emissions data, and many have committed to carbon-free or carbon-negative goals; independent analysts scrutinize these claims, especially as AI-driven growth has raised some companies' emissions.