Data centers and sustainability: the big picture

Data centers are among the fastest-growing consumers of electricity in the world. The rise of cloud computing, streaming and, above all, artificial intelligence has turned them from a modest share of demand into a major factor in utility planning, climate targets and local water use. At the same time, they are one of the most efficient ways to deliver computing, and the industry has led many advances in energy efficiency and clean energy procurement. This course examines how data centers use energy and water, how their impacts are measured, and how designers and operators reduce them: through efficient power and cooling, efficient IT, water stewardship, carbon accounting and clean energy matching, heat reuse, embodied carbon and circular practices, and the reporting and regulation that increasingly govern the industry. This first lesson sets out the scale of the issue and the main levers.

How much energy do data centers use?

The International Energy Agency (IEA) estimated in its 2025 Energy and AI report that data centers consumed around 415 TWh of electricity in 2024, about 1.5 % of global electricity use, and projected that this would more than double to around 945 TWh by 2030, driven largely by AI. The United States accounted for about 45 % of global data center consumption in 2024. Growth is concentrated geographically: in some regions, such as Northern Virginia, Ireland and parts of the Nordic countries, data centers use a much larger share of local electricity, and their growth shapes transmission planning, generation investment and electricity prices.

In the United States, a 2024 report by Lawrence Berkeley National Laboratory estimated that data centers used about 4.4 % of national electricity in 2023 and could use between about 7 % and 12 % by 2028. Growth is concentrated in Virginia, Texas, Oregon, Arizona, Georgia and Ohio, where utilities and regulators are revising tariffs and interconnection rules for very large loads.

Impacts beyond electricity

Energy use is the largest environmental impact of most data centers, but it is not the only one:

• Greenhouse gas emissions: mostly from the electricity used (Scope 2), plus emissions from backup diesel generators and refrigerant leaks (Scope 1), and from manufacturing the building, equipment and IT hardware (Scope 3, including embodied carbon).
• Water: evaporative cooling consumes large volumes of water, and electricity generation itself consumes water. In water-stressed regions, this is a major local concern.
• Land and community: large campuses occupy land, change local landscapes, and bring noise from cooling equipment and generators, as well as jobs and tax revenue.
• Materials and waste: servers are replaced every few years, producing electronic waste, and buildings use carbon-intensive concrete and steel.
• Grid effects: large loads can strain grids, but flexible loads can also help integrate renewable energy.

A sustainable data center strategy addresses all of these, and recognizes that improving one can worsen another: evaporative cooling saves energy but uses water; dry cooling saves water but uses more energy.

Comparing designs

The interplay between efficiency, grid carbon and water is easiest to see with numbers. The example compares a 100 MW campus under three designs: evaporative cooling on a carbon-intensive grid, air cooling on an average grid, and free cooling on a hydroelectric grid.

# Annual energy, emissions and water of a 100 MW (IT) campus under three designs
it_MW, hours, load_factor = 100, 8760, 0.75
designs = {
    # name: (annual PUE, WUE in L/kWh of IT energy, grid gCO2/kWh)
    "Evaporative cooling, coal-heavy grid": (1.25, 1.8, 700),
    "Air-cooled, average grid":             (1.40, 0.1, 380),
    "Free cooling, hydro grid":             (1.15, 0.2, 30),
}
it_MWh = it_MW * hours * load_factor
print(f"IT energy: {it_MWh:,.0f} MWh/yr")
for name, (pue, wue, g) in designs.items():
    total = it_MWh * pue
    co2_t = total * g / 1000
    water_m3 = it_MWh * 1000 * wue / 1000
    print(f"{name:<38} total {total:>9,.0f} MWh  CO2 {co2_t:>9,.0f} t  water {water_m3:>9,.0f} m3")
homes = it_MWh * 1.4 / 10.5      # about 10.5 MWh per North American home per year
print(f"Roughly the annual electricity of {homes:,.0f} homes at PUE 1.4")

IT energy: 657,000 MWh/yr
Evaporative cooling, coal-heavy grid   total   821,250 MWh  CO2   574,875 t  water 1,182,600 m3
Air-cooled, average grid               total   919,800 MWh  CO2   349,524 t  water    65,700 m3
Free cooling, hydro grid               total   755,550 MWh  CO2    22,666 t  water   131,400 m3
Roughly the annual electricity of 87,600 homes at PUE 1.4

The campus uses about 660,000 MWh a year at the IT equipment, comparable to the electricity of tens of thousands of homes. The evaporative design uses the least energy of the first two but the most water, over a million cubic meters a year, and on a coal-heavy grid it emits over 570,000 t of CO₂. The air-cooled design uses about 12 % more energy but almost no water on site. The free-cooled campus on a hydro grid emits about 4 % as much as the first design. The lesson is that location and grid often matter more than any single efficiency measure, and that water and carbon must be weighed together.

The levers

The main ways to reduce a data center's environmental impact, each developed in later lessons, are:

• Measure: establish metrics such as PUE, WUE and CUE, and measure them properly (Lesson 2).
• Reduce infrastructure losses: efficient electrical distribution and cooling (Lessons 3 and 4).
• Use IT efficiently: high utilization, efficient hardware and software, retiring idle servers (Lesson 5).
• Manage water: choose cooling technologies and water sources suited to local conditions (Lesson 6).
• Account for carbon correctly and procure clean energy, ideally matched hour by hour (Lessons 7 and 8).
• Reuse waste heat in district heating, buildings and industry (Lesson 9).
• Reduce embodied carbon and practice circularity in buildings and hardware (Lesson 10).
• Report transparently and comply with regulation and certification schemes (Lesson 11).

Lesson 12 brings these together into a sustainability strategy for a real or hypothetical site.

Efficiency gains and rebound

Between 2010 and 2018, global data center workloads grew many times over while energy use grew only modestly, thanks to large efficiency gains: the shift from small, inefficient server rooms to hyperscale facilities with low PUE, server virtualization, and more efficient processors. That period is often cited as evidence that efficiency can decouple computing from energy growth. The AI era has changed the picture. Accelerated computing is very efficient per operation, but demand for AI training and inference is growing so fast that total energy use is rising sharply. This is an example of the rebound effect (sometimes called Jevons' paradox): making a service cheaper and more efficient can increase total demand. Sustainability strategies therefore cannot rely on efficiency alone; they also need clean energy, careful siting and attention to whether the computing itself creates value.

Who cares, and why

Pressure for sustainable data centers comes from several directions:

• Customers: large technology companies and enterprises have climate targets that extend to their suppliers, and they ask colocation providers for renewable energy, emissions data and efficiency commitments.
• Investors and lenders: they assess climate risk, regulatory risk and sustainability performance, and offer green bonds and sustainability-linked loans with conditions.
• Governments and regulators: they require reporting, set efficiency standards, condition grid connections and planning approvals on sustainability measures, and in some places restrict new data centers.
• Communities: residents care about water, noise, land use and electricity prices, and local opposition can block projects.
• Employees: many engineers want to work for organizations that take sustainability seriously.

For engineers, this means that sustainability is no longer a separate topic but a design requirement, alongside reliability and cost.

How this course is organized

The section at the end of the lesson shows how the twelve lessons fit together.

• Lesson 1 (this lesson): data centers and sustainability: the big picture.
• Lesson 2: efficiency metrics: PUE, WUE, CUE and beyond.
• Lesson 3: power distribution efficiency.
• Lesson 4: cooling efficiency and free cooling.
• Lesson 5: IT efficiency: utilization, hardware and software.
• Lesson 6: water use and stewardship.
• Lesson 7: carbon accounting for data centers.
• Lesson 8: clean energy procurement and 24/7 matching.
• Lesson 9: heat reuse and community energy.
• Lesson 10: embodied carbon and circularity.
• Lesson 11: reporting, regulation and certification.
• Lesson 12: a sustainability strategy: case study and project.

The course builds on mechanical systems (DCE-330) and connects with power procurement (DCE-420) and operations (DCE-450).

The role of the engineer

Engineers make most of the decisions that fix a data center's environmental footprint for decades: where it is built, how it is cooled, how power is distributed, how efficiently it runs at part load, whether heat can be exported, and what materials go into it. Many of these choices are made early, during site selection and concept design, when changes are cheap; later, they are locked in by concrete and contracts. Engineers also shape operations, through commissioning, control settings and maintenance practices that determine whether a well-designed facility actually performs as intended. Developing the habit of asking "what does this choice do to energy, water and carbon over the facility's life?" alongside "is it reliable and affordable?" is the central aim of this course.

That question rarely has a single right answer. Good sustainability engineering means quantifying the options, being explicit about assumptions and uncertainties, and explaining trade-offs clearly to the people who must decide.