Introduction: The hidden infrastructure behind AI

Artificial intelligence can help people write, compose music, create images, analyse information and develop new services. Yet the technology does not exist only on a screen. Behind many AI services are data centres filled with servers, specialist computing equipment, networking systems and cooling infrastructure.

These facilities require electricity. Depending on their design and cooling systems, they may also use water. Their construction and operation can create additional demand for land, power networks, equipment and other resources.

As governments and technology companies invest in AI, an important question emerges: who pays for the infrastructure that makes this revolution possible, and who benefits from it?

The issue is not whether AI should exist. It is whether its development can be managed responsibly, with environmental costs measured, public interests protected and benefits shared fairly.

1. What powers the AI revolution?

AI services rely on computing infrastructure that processes and stores data. Training some advanced AI models can require substantial computing resources, while operating services for millions of users creates continuing demand.

Data centres also support cloud computing, online banking, streaming, communications, public services and many other activities. Their electricity use cannot therefore be attributed entirely to AI. Understanding the additional demand created by AI requires careful measurement rather than treating every data-centre activity as an AI workload.

The International Energy Agency's Energy and AI report examines the relationship between computing demand, electricity generation, energy security, emissions and affordability. The IEA also considers the potential for AI to improve energy systems.

This creates a double question: how much energy will AI require, and can AI also help reduce waste or improve efficiency elsewhere?

2. What does the UK evidence tell us?

In October 2026, the Office for National Statistics reported that data centres in Great Britain consumed an estimated 4.5 terawatt-hours of electricity in 2024, equivalent to approximately 2% of the 249.2 terawatt-hours of electricity consumed from the grid that year. The estimate was 1.3 terawatt-hours higher than in 2020.

These figures cover data centres generally, not AI alone. They provide useful context for the scale of digital infrastructure, but they should not be presented as a direct measurement of AI's individual electricity consumption.

The Department for Energy Security and Net Zero also commissioned studies comparing the electricity used by digital services with physical alternatives. Its July 2026 follow-up study found that, under the scenarios examined, AI-assisted worker productivity typically reduced electricity consumption per task. Results for other digital services varied according to the activity being compared.

This is an important qualification: AI does not automatically increase energy use for every task, nor does efficiency in one task prove that total demand across the industry will fall. Greater efficiency can make services cheaper and encourage wider use, potentially offsetting some savings.

Sources: Office for National Statistics and the UK data-centre energy studies.

3. The water question: cooling a digital world

Servers generate heat. Some data centres use water-based cooling systems; others use air cooling, closed-loop systems or alternative technologies. Water use varies significantly with location, facility design, operating conditions and the cooling method selected.

Water can also be used indirectly when electricity is generated. This means an assessment of AI's water footprint may need to consider both the data centre itself and the energy supply supporting it.

It would be misleading to suggest that every data centre uses the same amount of water or that every AI request consumes a fixed quantity. Estimates depend on the facility, energy source, workload, cooling technology and method used to calculate consumption.

On 28 October 2026, the UK Government stated that data centres currently account for less than 0.2% of non-household water consumption in England, according to the evidence it had gathered. It also warned that future growth could increase pressure in water-stressed areas, particularly during hotter periods.

The Government has called for developers to prioritise reducing water demand, consider recycled water and rainwater where appropriate, and use additional freshwater only after reasonable alternatives have been assessed.

Source: Government statement on sustainable data-centre growth.

4. Who pays for the infrastructure?

Building and operating data centres can involve major private investment. However, the wider infrastructure needed to support them may also require electricity network upgrades, planning resources, water-system planning and other public or regulated services.

The central financial question is whether developers pay the costs attributable to their projects or whether some costs are transferred to households, other businesses or local communities.

On 28 October 2026, the UK Government published its Data Centre Covenant. It sets out expectations concerning energy costs, local community investment, environmental standards, jobs and apprenticeships, and national security.

The Covenant says data centres should pay their fair share of energy-system costs and that strategic facilities must meet specified tests to qualify for additional government action. It also encourages environmental reporting and opportunities to reuse waste heat.

These are important policy directions, but publication of a covenant does not by itself demonstrate that every developer has complied or that household bills have fallen. Those outcomes require evidence and monitoring.

5. Public funding: what should taxpayers expect in return?

Governments may support digital infrastructure through planning decisions, research programmes, skills initiatives, investment mechanisms and measures designed to improve access to energy or computing capacity.

Such support may have legitimate public benefits. Domestic computing capacity can support research, business innovation, resilience and the development of services. But when public policy helps accelerate private investment, it is reasonable to ask what the public receives in return.

A responsible assessment should examine:

• Financial transparency: What public support is provided, through which mechanism, and under what conditions?
• Environmental performance: Are energy use, water use and emissions measured consistently?
• Community benefits: Are promised jobs, training opportunities and local investments actually delivered?
• Infrastructure costs: Who pays for new connections and wider network improvements?
• Fair access: Do smaller businesses, independent researchers, educators and community organisations gain meaningful opportunities to use AI?
• Accountability: What happens if a developer fails to meet its commitments?

Public support should be assessed against clear outcomes, not just the amount of investment announced or the number of facilities proposed.

6. Can AI help solve environmental problems?

AI may support energy forecasting, building management, industrial efficiency, renewable-energy integration, transport planning and the detection of faults in infrastructure. These applications could contribute to environmental improvements when implemented effectively.

However, a proposed benefit is not the same as a verified result. An AI system should be evaluated against a credible baseline, taking account of its own energy use, the infrastructure it requires and any additional activity its adoption creates.

For example, a system might reduce the energy required for one industrial process but encourage enough additional production to increase total energy use. A data centre might improve its cooling efficiency while increasing its overall water demand as it expands.

The relevant question is therefore not simply whether AI can make something more efficient. It is whether the complete system produces a measurable net benefit.

7. Why environmental reporting matters

People cannot properly assess environmental claims if the information needed to test them is missing, inconsistent or difficult to compare.

The Government's Data Centre Covenant encourages annual reporting on environmental performance, including energy and water consumption, efficiency, emissions, heat reuse and participation in electricity-system flexibility services. It says strategic data centres should report annually and encourages existing operators to do the same.

The Government has also said it will consider a statutory reporting framework if voluntary reporting does not provide sufficient information. This distinction matters: the current approach described in the statement is an expectation and encouragement for reporting, not proof that a universal mandatory reporting regime is already in force.

Useful reporting should include:

• Total and peak electricity demand.
• Energy-efficiency measures and the methodology used to calculate them.
• Total water consumption and water-use effectiveness.
• Carbon emissions and the boundaries used for reporting them.
• Cooling methods, recycled-water use and relevant local water constraints.
• Waste-heat recovery and whether heat is actually reused.
• Progress against published commitments, with independent verification where appropriate.

Publishing a target is only the beginning. Transparent reporting makes it possible to compare the target with the result.

8. What about local communities?

Large infrastructure developments can create employment and investment, but their effects depend on where they are built and how they are managed. Communities may have questions about land use, construction, noise, electricity connections, water availability, local services and the long-term distribution of benefits.

Meaningful consultation should take place early enough for concerns to influence decisions. Environmental assessments should reflect local conditions rather than relying solely on national averages.

In England, the Government published a draft National Policy Statement for Data Centres on 28 October 2026. It is intended to guide development-consent decisions for data-centre projects handled through the Nationally Significant Infrastructure Projects regime. The consultation is open until 8 January 2027.

Because this is a draft policy document, it should not be described as a final policy statement. Its eventual effect will depend on the consultation and subsequent decisions.

Source: Draft National Policy Statement for Data Centres.

9. What should responsible AI companies do?

Responsible AI development requires more than a statement about sustainability. Companies should connect environmental commitments to operational decisions and public evidence.

1. Measure the energy and water demands of their infrastructure.
2. Choose locations and cooling systems with local resource constraints in mind.
3. Improve computing efficiency without assuming efficiency alone will solve rising total demand.
4. Use renewable electricity and other lower-carbon options where credible and practical, while being transparent about how claims are calculated.
5. Assess whether waste heat can be reused by nearby buildings or networks.
6. Disclose environmental performance and progress against commitments.
7. Engage with local communities, regulators and utility providers before major development decisions.
8. Explain the public benefits of any public support and report whether those benefits have been achieved.

10. What can users and smaller creative businesses do?

Individual users are not responsible for solving the environmental impact of the AI industry on their own. The largest infrastructure and investment decisions sit with technology companies, energy providers, regulators and governments.

Nevertheless, users and smaller creative organisations can make considered choices:

• Choose AI tools that genuinely help achieve a useful goal.
• Avoid unnecessary repeated generation when a result is already suitable.
• Compare tools' published environmental information where reliable information is available.
• Ask suppliers about data protection, transparency and sustainability commitments.
• Support public policies that encourage clear reporting and fair allocation of infrastructure costs.

For creative communities, including disabled and neurodivergent artists, the aim should not be to discourage useful technology. AI can support experimentation, accessibility and new forms of expression. The challenge is to make that innovation responsible, affordable and environmentally sustainable.

11. The questions governments and investors should answer

As AI infrastructure expands, decision-makers should be able to answer a straightforward set of questions:

• How much public money or other public support is being committed?
• What environmental and financial conditions are attached to that support?
• Who bears the costs of additional electricity, water and network capacity?
• What measurable benefits are being delivered to local communities?
• Are environmental reports complete, comparable and independently checked?
• Are promised efficiency improvements reducing total resource use, or simply enabling more consumption?
• What remedies are available when commitments are not met?

These questions make it possible to examine AI investment without assuming either that every project is harmful or that every promise of progress will be fulfilled.

Conclusion: Innovation needs accountability

Artificial intelligence may contribute to scientific discovery, creative expression, business productivity and better public services. It also depends on physical infrastructure whose energy, water and environmental effects deserve serious attention.

The evidence does not support a simplistic conclusion that AI is always environmentally damaging, or that efficiency improvements automatically make its expansion sustainable. The outcomes depend on how systems are designed, powered, cooled, used and governed.

The UK Government's October 2026 announcements create an opportunity to test whether growth in data-centre infrastructure can be matched by transparent reporting, fair infrastructure costs, protection of water resources and tangible community benefits.

The central question remains: if AI is being promoted as an investment in our future, how will we know that its financial, social and environmental benefits justify the resources it consumes?

Progress should be measured not only by how much AI can do, but by whether it is developed responsibly and whether the wider public shares in its benefits.

Sources and further reading

• UK Government — The Data Centre Covenant (28 October 2026)
• UK Parliament — Enabling sustainable data-centre growth through responsible water and energy use (28 October 2026)
• UK Government — Draft National Policy Statement for Data Centres (28 October 2026)
• Office for National Statistics — Data centres and the UK National Accounts
• Department for Energy Security and Net Zero — Impact of growth of data centres on energy consumption
• International Energy Agency — Energy and AI
• UK Government — Compute Evidence Annex