The environmental impact of data centers: From megawatts to water drops

Data centers convert electricity into computational work and heat, and operational impacts are hence dominated by electricity and water use.

AI-accelerated servers can draw four to five times more power than traditional x86 machines, placing greater demands on cooling systems (MIT News, 2025). Data centers and climate crisis narratives therefore focus on this operational growth. The European Union now treats them as “energy-intensive installations” subject to mandatory sustainability ratings (European Commission, 2024). According to the International Energy Agency, the sector consumed roughly 460 TWh of electricity in 2022—almost 2% of global demand—and could double by 2026 if current AI trends continue (IEA, 2024, 2025).

Beyond energy, UNEP guidelines identify four further aspects: water, refrigerants, solid waste and land use (UNEP, 2025). The World Bank considers social factors, such as community noise and strain on municipal grids, when assessing the ecological impact of data centers (World Bank, 2024). Collectively, these dimensions shape a facility’s data center sustainability profile and determine its compliance exposure.

Key operational variables: energy, water, refrigerants, e-waste

Operators routinely focus on tracking the four high-impact variables that most directly influence utility bills, compliance and brand reputation: energy, water, refrigerants and e-waste.

• Energy

The IMF warns that emerging generative-AI workloads could add up to 85 TWh of additional demand by 2030, emphasizing the environmental impact of AI data centers (IMF, 2025). Power-use effectiveness (PUE) remains the headline metric for data center energy efficiency. The U.S. Department of Energy (DOE) reports that facilities adopting hot-aisle containment, high-efficiency uninterruptible-power supplies and liquid cooling can lower PUE from a global average of 1.58 to below 1.2 (DOE, 2024). Yet PUE alone says nothing about the data center´s carbon footprint: a site running on coal-fired power may post a respectable PUE but still have a high carbon intensity, hence the added value of monitoring the Carbon-Usage Effectiveness (CUE) as well.

• Water

Hyperscale facilities can consume an average of 20–26 million liters of water per year for evaporative cooling (Financial Times, 2025). Regions facing droughts have already delayed cloud projects, making data center water usage a constraint equal in importance to electricity procurement.

• Cooling Refrigerants

Legacy chillers often rely on hydrofluorocarbons (HFCs) with a global warming potential thousands of times higher than that of CO₂. The DOE guide lists non-ozone-depleting refrigerants and recommends leak-detection programs to reduce fugitive emissions (DOE, 2024).

• Hazardous & Electronic Waste

Server refresh cycles of three to five years generate substantial flows of printed circuit boards and lithium-ion batteries. Harvard researchers estimate that e-waste now contributes up to 20 % of a facility’s lifecycle emissions when Scope 3 categories are included (Harvard SEAS, 2024).

*KPIs referenced in the EU rating scheme and DOE guide.

Real-time dashboards that combine these metrics enable operators to flag anomalies and report under the forthcoming EU disclosure regime (European Commission, 2024). The environmental cost of the cloud is increasingly benchmarked by investors using such data, creating a competitive incentive for transparency.

Design phase

Site selection determines grid carbon intensity, water availability and potential for renewable power purchase agreements (PPAs). Recent deals such as Google’s USD 3 billion hydroelectric PPA in Canada illustrate how siting strategy can offset operational emissions (Financial Times, 2025). Designing for modularity and equipment reuse can lower embodied carbon even before the first rack is installed, a trend highlighted in the Uptime Institute’s 2024 survey (Uptime Institute, 2024).

Construction phase

Concrete and steel account for the bulk of embodied emissions during construction. The World Bank estimates that construction can account for up to 45% of a data center’s lifecycle CO₂ emissions in regions where grids are already low-carbon (World Bank, 2024). Larger facilities intensify these impacts but also unlock economies of scale for on-site renewables and heat-recovery systems.

Newly built facilities often have lower PUE and WUE, but their absolute footprints can exceed those of smaller legacy sites due to a higher total IT load. Overall, lifecycle analysis shows a U-shaped curve: small edge sites have modest embodied impacts but poorer efficiency, hyperscale sites have significant construction impacts offset by high efficiency, and mid-sized corporate facilities often fare worst on both counts (IEA, 2024).

Underwater data centers

Pilot projects in Northern Europe and Asia place sealed server pods on the seabed, using natural convection to dissipate heat. Benefits include near-free cooling and proximity to coastal users, which reduces latency. Early trials reported PUE values below 1.1 and zero freshwater use (Uptime Institute, 2024). Risks, however, include corrosion, biofouling and uncertainties about marine ecosystems—issues flagged in UNEP’s precautionary guidelines (UNEP, 2025). Retrieval for hardware refreshes could also spike embodied emissions.

Edge and modular facilities

Prefabricated modules allow rapid deployment closer to end users, cutting backbone traffic and the environmental cost of the cloud. Yet hundreds of small sites can collectively consume more energy than a single hyperscale campus if they lack sophisticated air-flow management. The DOE notes that containerized edge sites rarely achieve PUEs below 1.4 unless liquid cooling is employed (DOE, 2024).

These emerging models offer pathways toward green data centers but also add complexity to corporate sustainability reporting.

Comparative Outlook

The environmental storyline of the digital economy is no longer confined to server closets; it spans global supply chains, local watersheds and the upper atmosphere. Energy, water, refrigerants and e-waste define the operational heartbeat of any facility, while siting and construction lock in impacts for decades. Non-conventional architectures—from underwater pods to edge boxes—promise efficiency gains but introduce fresh ecological uncertainties.

For enterprises aiming to quantify and manage these risks, Applus+ brings an integrated suite of environmental assessment, environmental-and-social impact studies, environmental monitoring, environmental inspection and ultimately minimization of environmental impact services tailored to data-center realities.

In the next article, we explore how a data center’s environmental impact can be effectively monitored and mitigated, translating insights into actionable steps.