The Hidden Footprint of AI: How Data Centres Affect Energy, Water and Emissions

Introduction

Artificial intelligence is rapidly becoming part of everyday business operations. Companies are using AI to analyse data, automate customer service, improve logistics, generate content and accelerate scientific research. However, every AI request depends on physical infrastructure, including processors, servers, cooling equipment, electricity networks and large data centres.

As AI adoption grows, so does the environmental footprint of the infrastructure supporting it. The challenge is not simply that AI uses electricity. Data centres can place concentrated pressure on electricity grids, water supplies and local communities, particularly when new facilities are developed faster than supporting infrastructure.

Understanding these impacts can help businesses make more informed decisions about AI providers, cloud services, procurement policies and sustainability reporting.

Why AI Requires So Much Computing Power

Traditional data centres already support websites, business software, video streaming, financial transactions and cloud storage. AI adds another layer of demand because developing and operating advanced models requires large numbers of specialised computing chips.

AI-related energy consumption generally occurs during two main stages:

• Training, when a model processes large datasets and adjusts its internal parameters.

• Inference, when the trained model responds to questions, generates images, analyses documents or performs other tasks.

Training a large model can require substantial computing resources over an extended period. However, inference can ultimately create a larger cumulative footprint when millions of people and organisations use the same service repeatedly.

Individual AI interactions may consume relatively little electricity, particularly as models and processors become more efficient. The wider concern is scale. A small amount of energy multiplied across billions of requests can create significant aggregate demand.

The International Energy Agency reported that global electricity demand from data centres increased by 17% in 2025. Demand from AI-focused facilities grew even faster. The agency expects overall data centre electricity consumption to double by 2030, while consumption from AI-focused centres could triple.

The Growing Electricity Demand of AI Data Centres

Data centres operate continuously and require reliable power for servers, networking equipment, storage systems, security infrastructure and cooling.

AI servers often contain high-performance graphics processing units and other accelerators that use more electricity and generate more heat than conventional computing equipment. As a result, AI facilities can have much higher power densities than older data centres.

This matters because a large data centre campus may require hundreds of megawatts of capacity. Some planned campuses are comparable in electricity demand to industrial complexes or medium-sized cities.

In the United States, data centres consumed approximately 176 terawatt-hours of electricity in 2023, equivalent to around 4.4% of national electricity use. The US Department of Energy estimated that the sector could consume between 325 and 580 terawatt-hours annually by 2028, representing approximately 6.7% to 12% of total US electricity demand.

Although these figures cover all data centres, AI is expected to be one of the principal drivers of future growth.

The environmental impact of this electricity depends heavily on how it is generated. A data centre supplied by low-carbon electricity will generally have lower operational emissions than an identical facility powered by a grid dominated by coal or natural gas.

However, renewable energy contracts do not always mean that a facility is using carbon-free electricity every hour. A company may purchase enough renewable electricity over the course of a year to match its annual consumption while still relying on fossil-fuel generation during periods when wind and solar output is low.

For this reason, some technology companies are moving from annual renewable energy matching towards round-the-clock carbon-free energy procurement.

How Data Centres Can Affect Electricity Grids

The electricity challenge is not limited to total consumption. Location and timing are equally important.

Data centres represent large, concentrated loads. When several projects are proposed in the same region, utilities may need to build new substations, transmission lines and generation capacity. Transformers and grid connections can take years to plan and construct.

Rapid data centre development can therefore create several risks:

• Delayed grid connections for other businesses and renewable energy projects

• Higher infrastructure costs for utilities

• Greater dependence on natural gas generation during peak periods

• Pressure on household and commercial electricity prices

• Increased difficulty balancing electricity supply and demand

The IEA notes that data centres can create particular affordability challenges because they scale quickly and may require major investment in generation and network infrastructure. However, these costs do not automatically have to be passed on to other electricity users. Regulators can introduce tariffs and connection rules requiring developers to cover the infrastructure costs directly associated with their facilities.

Data centres can also support electricity systems when designed to operate flexibly. Some computing workloads can be shifted to periods when renewable electricity is abundant or when overall electricity demand is lower.

Battery storage, onsite renewable generation and intelligent workload scheduling can also help facilities reduce peak demand. The US Department of Energy has identified flexible consumption, onsite power and energy storage as important measures for integrating data centres without making them a burden on the grid.

The Water Footprint of AI

Electricity consumption is only one part of the environmental impact. Data centres also use water, primarily for cooling.

Computer processors generate heat while operating. If that heat is not removed, equipment performance and reliability can decline. Data centre operators therefore use cooling systems to maintain controlled temperatures.

Some facilities rely largely on air cooling, while others use cooling towers or evaporative systems that consume water. Water may also be used indirectly by the power stations supplying electricity to the facility.

This creates two different categories of water impact:

• Direct water consumption, associated with cooling the data centre itself.

• Indirect water consumption, associated with generating the electricity it uses.

Evaporative cooling can be highly energy efficient, but it may consume substantial quantities of water. This creates a trade-off between reducing electricity demand and protecting local water resources.

The location of a facility is therefore important. A cooling system that may be appropriate in a water-abundant region could create environmental and social risks in an area facing drought or long-term water stress.

Water impacts are also highly localised. National data centre water use may appear relatively small when compared with agriculture or total industrial demand, but a single large facility can still place considerable pressure on a local water system.

Companies evaluating cloud and AI providers should therefore consider where data centres are located, what cooling technology they use and whether potable, recycled or reclaimed water supplies the facility.

Carbon Emissions Beyond Daily Operations

The operational electricity used by servers is often the most visible component of AI’s carbon footprint. However, a complete assessment should also include embodied emissions.

Embodied emissions arise from producing and constructing:

• Semiconductors and AI accelerators

• Servers and networking equipment

• Cooling systems

• Backup generators and batteries

• Data centre buildings

• Transmission and electricity infrastructure

Manufacturing advanced chips is an energy-intensive process that also requires water, chemicals and specialised materials. AI hardware may also be replaced frequently as companies compete to deploy newer and more efficient processors.

This can increase electronic waste and create additional supply-chain emissions.

For companies using AI services, these impacts may fall within Scope 3 emissions rather than direct operational emissions. They can be difficult to calculate because cloud providers do not always disclose emissions at the level of an individual service, workload or customer.

Businesses should avoid assuming that moving computing activities to the cloud eliminates their environmental impact. In many cases, the impact has simply moved to another organisation’s infrastructure and supply chain.

Efficiency Improvements Are Not Always Enough

AI processors, software and cooling systems are becoming more efficient. Newer models may require less energy to complete a comparable task, while specialised chips can deliver more calculations per unit of electricity.

The IEA has found that energy use per AI task is falling rapidly. However, total data centre electricity demand continues to rise because AI adoption and the number of applications are growing even faster.

This is an example of the rebound effect. When a service becomes cheaper and more efficient, its use may expand enough to outweigh the savings achieved per transaction.

A company might reduce the energy required to analyse one document, for example, but then apply AI analysis to millions of documents that were not previously processed.

Efficiency remains essential, but it should be combined with controls that ensure AI is used where it provides meaningful value.

How Businesses Can Reduce the Environmental Impact of AI

Most organisations do not own data centres. Nevertheless, they can influence the footprint of their AI use through procurement, governance and operational decisions.

Select Providers With Transparent Environmental Data

Businesses should ask cloud and AI providers to disclose information such as:

• Energy consumption and power usage effectiveness

• Renewable and carbon-free electricity procurement

• Scope 1, Scope 2 and relevant Scope 3 emissions

• Water consumption and water usage effectiveness

• Data centre locations

• Cooling technologies

• Hardware lifecycle and recycling practices

Environmental claims should be supported by measurable data rather than general commitments.

Use the Right Model for the Task

Not every business problem requires the largest available AI model. Smaller or specialised models may complete routine tasks using fewer computing resources.

Companies can establish policies that match model size and complexity to the task. A lightweight model may be sufficient for document classification, basic summarisation or internal search, while more advanced models can be reserved for activities requiring greater reasoning capacity.

Avoid Unnecessary AI Processing

AI can be valuable, but its use should not become automatic.

Organisations can reduce unnecessary processing by:

• Limiting repeated or duplicate requests

• Reusing previously generated results

• Processing documents in batches

• Setting limits on high-compute features

• Reviewing automated workflows that run continuously

• Removing applications that deliver little business value

These measures may also reduce cloud expenditure.

Schedule Flexible Workloads

Certain AI activities, such as model training, data processing and non-urgent analysis, may not need to happen immediately.

Where providers offer the capability, these workloads can be scheduled for locations or periods with lower grid carbon intensity and greater renewable energy availability.

Carbon-aware computing can help shift flexible demand without reducing the quality of the final service.

Include AI in Sustainability Reporting

Companies should consider whether cloud computing and AI services represent a material component of their environmental footprint.

Relevant information may need to be incorporated into:

• Greenhouse gas inventories

• Scope 3 calculations

• Environmental management systems

• Supplier assessments

• Digital sustainability strategies

• Climate transition plans

Businesses should also document estimation methods and data limitations, particularly when suppliers do not provide customer-specific figures.

The Potential Environmental Benefits of AI

AI’s environmental impact should be considered alongside its potential to support decarbonisation.

AI can help organisations forecast renewable electricity generation, improve building efficiency, detect methane leaks, optimise logistics, reduce material waste and manage electricity grids.

According to the IEA, proven AI applications could help energy-intensive companies reduce their energy costs by approximately three to ten percentage points.

However, these benefits are not guaranteed. An AI application creates a net environmental benefit only when the emissions, energy, or resources it helps avoid are greater than the footprint of developing and operating it.

Companies should therefore evaluate AI projects using measurable outcomes. Claims that AI is supporting sustainability should be linked to evidence such as lower energy consumption, reduced fuel use, fewer wasted materials or verified emissions reductions.

What Sustainable AI Infrastructure Could Look Like

Reducing the environmental impact of AI will require action across the technology, energy and policy sectors.

More sustainable data centres are likely to combine:

• Energy-efficient processors and servers.

• High equipment utilisation rates.

• Advanced cooling systems.

• Recycled or non-potable water.

• Low-carbon electricity.

• Battery and thermal energy storage.

• Flexible computing workloads.

• Heat recovery.

• Longer hardware lifecycles.

• Transparent environmental reporting.

Waste heat from servers may also be captured and supplied to district heating networks, greenhouses or nearby industrial facilities. The feasibility of heat recovery depends on the temperature of the waste heat, the distance to potential users and local infrastructure.

Data centre developers must also improve site selection. Access to electricity is not the only consideration. Water availability, grid carbon intensity, climate risks, community impacts and opportunities to reuse existing infrastructure should all influence development decisions.

Turning AI Growth Into a Managed Sustainability Issue

AI is not an invisible or purely digital technology. It depends on physical facilities, electricity systems, water resources, raw materials and global manufacturing supply chains.

The environmental impact of one AI request may be modest, but the rapid expansion of AI across businesses and consumer services is creating a significant infrastructure challenge.

This does not mean that companies should avoid AI entirely. It means they should use it deliberately.

Businesses that measure AI-related impacts, select transparent suppliers and apply efficient models can reduce both environmental risks and computing costs. They may also be better prepared for future disclosure requirements, energy constraints and stakeholder scrutiny.

The most sustainable approach is not simply to make every AI task more efficient. It is to ensure that AI is used for valuable purposes, supported by responsible infrastructure and measured against credible environmental outcomes.