There is a fundamental question at the heart of the artificial intelligence revolution that governments have so far struggled to answer: how much of the world are we prepared to sacrifice to make machines more intelligent?
The construction of data centres is accelerating as technology companies compete to build ever more powerful AI systems. These facilities are becoming essential to modern life, supporting everything from scientific research and medical advances to online services and industrial automation. But they also consume enormous amounts of electricity, require sophisticated cooling systems, and place pressure on water supplies, electricity networks and local communities.
The International Energy Agency estimates that data centres consumed around 415 terawatt-hours of electricity worldwide in 2024. Its subsequent analysis found that consumption rose by 17% in 2025, with AI-focused facilities growing even faster. The agency expects global data centre electricity demand to roughly double between 2025 and 2030 if current trends continue.
These are not trivial numbers, particularly when countries are trying to reduce carbon emissions, protect nature and keep energy affordable.
The obvious response might be to oppose new data centres altogether. But that would be short-sighted. Artificial intelligence could help develop medicines, improve energy efficiency, accelerate scientific discoveries and solve problems that have resisted human ingenuity for generations. Denying ourselves these possibilities would be a mistake.
The answer is not to stop technological progress. It is to make sure that progress serves humanity rather than forcing humanity to serve it.
Who benefits, and who pays?
Imagine a large technology company proposing a new data centre on the outskirts of a British town. The developers promise investment, jobs and a contribution to the local economy. The government welcomes the project as evidence that Britain is competing in the global AI race.
But the local community has questions. Will the facility increase pressure on the electricity grid? Could it push up energy costs? How much water will it consume? What happens to the surrounding countryside? And how many permanent jobs will actually remain once construction is complete?
These are reasonable questions, not signs of hostility towards technology.
The problem is that the financial benefits of AI infrastructure can be concentrated in the hands of a relatively small number of companies and investors, while some of its environmental and infrastructure costs are shared by everyone else.
An AI politician would regard this as a question of fairness as much as environmental policy.
If a private company wants to build a facility that consumes as much electricity as a small town, it should not automatically receive priority over households, hospitals, schools or essential public services. Nor should ordinary bill payers be expected to subsidise the necessary grid upgrades simply because a technology company has decided that rapid expansion is commercially desirable.
The principle would be straightforward: the companies that create the demand must bear a fair share of the costs.
That includes the cost of connecting to the grid, providing additional electricity generation where necessary, managing environmental damage and helping to ensure that local communities benefit from the investment.
Public money may sometimes be justified, particularly where research or infrastructure provides a demonstrable public benefit. But subsidies should be transparent, conditional and proportionate. Governments should not socialise the risks while allowing private investors to retain almost all the rewards.
We should build more intelligently, not simply build more
One of the great weaknesses of modern politics is its tendency to measure success by the scale of what is announced rather than the value of what is achieved.
A government can boast about billions of pounds of investment, thousands of proposed jobs and vast new computing facilities. Yet none of those figures tells us whether the infrastructure is necessary, environmentally responsible or worth its total cost.
An AI politician would begin with a different question: what computing capacity do we actually need, and how can we provide it with the least possible environmental damage?
That would mean examining every major proposal on its merits. Is the facility needed to support genuinely useful applications? Could existing data centres accommodate the demand through better utilisation? Could the same work be completed using smaller, more efficient AI models? Would another location place less pressure on the electricity grid or local water supplies?
The answers would not always be the same. A data centre supporting medical research or the management of renewable electricity might deliver substantial public benefits. A facility devoted primarily to generating endless quantities of disposable digital content would require a different assessment.
This is not about a government deciding which ideas people are allowed to pursue. It is about ensuring that scarce resources are allocated intelligently, rather than assuming that every proposed investment must be welcomed simply because it involves AI.
The alternatives are already emerging
We do not have to rely on a single technological solution. Several approaches could reduce the environmental cost of AI, and an effective government would encourage them to work together.
First, we should make existing computing infrastructure more efficient. Better chips, improved software, more efficient cooling and smarter workload management can reduce the electricity required for many tasks. The International Energy Agency has highlighted the substantial potential for efficiency improvements to restrain future demand.
Not every AI application needs the largest available model. Smaller, specialised systems can often perform particular tasks with far less computing power. If a modest model can answer a routine question, analyse a document or manage a simple business process, deploying a vast system to do the same work may be wasteful.
Second, we should make data centres more flexible. Some computing tasks, especially the training of AI models, can be scheduled rather than performed immediately. Where practical, these workloads could be shifted to periods when renewable electricity is plentiful or the grid is under less pressure.
This would not eliminate the need for reliable power. Critical services must remain available, and not every workload can be moved at will. But greater flexibility could help data centres use electricity more intelligently instead of competing with households for capacity at the most difficult times.
Third, we should change where and how facilities are built. Locations with reliable low-carbon electricity, adequate grid capacity and suitable cooling conditions may be preferable to sites where new infrastructure would put additional pressure on an already constrained system.
Some data centres could also make greater use of recycled water, closed-loop cooling and systems that reduce dependence on drinking water. Waste heat could supply nearby buildings, swimming pools or industrial processes where the temperatures, distances and economics make this practical.
There is no universal answer. A cooling system that works well in one climate may be unsuitable in another. Every proposal should be assessed according to its actual environmental performance, not simply its marketing claims.
Finally, we should invest in alternatives to the assumption that every increase in demand must be met by constructing another enormous facility. Better use of existing capacity, more efficient algorithms, shared computing resources and carefully designed smaller systems could all reduce the pressure for continual expansion.
The objective should be to deliver the computing power society needs while minimising the resources required to provide it.
Clean energy is essential, but it is not a blank cheque
Technology companies increasingly point to renewable electricity as the answer to concerns about data centre emissions. That is encouraging, but it does not settle the argument.
A company purchasing renewable electricity certificates or signing a contract for wind power does not necessarily mean that its data centre is being supplied with additional clean electricity at every hour of the day. The wider electricity system still has to balance supply and demand, including when the wind does not blow and the sun does not shine.
An AI politician would therefore favour investment in renewable generation alongside grid improvements, energy storage, interconnection and other dependable low-carbon sources. Nuclear power may form part of that mix where it is safe, economically justified and appropriate to local circumstances.
The test would be whether the overall system genuinely reduces emissions and provides reliable, affordable electricity, rather than whether a company can advertise itself as green.
Nor should we accept the destruction of valuable habitats or the unnecessary loss of agricultural land on the assumption that a project is environmentally beneficial simply because it supports AI. Climate policy and nature protection must be considered together.
A green data centre that damages an irreplaceable ecosystem is not automatically a good outcome.
Give communities a genuine say
Too often, communities are presented with a development as though the important decisions have already been made. Consultation follows, objections are recorded, and residents are left feeling that their participation was little more than a formality.
That approach breeds resentment, even when a project might otherwise have been acceptable.
An AI politician would require clear environmental assessments, honest disclosure of expected energy and water consumption, and meaningful consultation before major planning decisions were finalised. Independent experts would scrutinise the figures, and the public would be able to see how competing interests had been weighed.
Communities hosting large facilities should also receive tangible benefits. These might include contributions to local infrastructure, investment in skills and training, and support for public services where the development creates additional demands.
But compensation cannot turn every environmental loss into an acceptable one. Some places should remain protected, and some proposals should be rejected.
That is what planning is for.
The same rules should apply whether the developer is a small domestic company or one of the world’s most powerful technology businesses. Wealth, political influence and promises of future economic dominance should not exempt anyone from reasonable standards.
Progress should be judged by what it gives humanity
There is a deeper issue here. AI companies are engaged in a race to develop increasingly capable systems, often encouraged by investors who expect rapid growth and enormous returns. That competition may produce remarkable advances, but it can also encourage a dangerous assumption: that more computing power, more facilities and more investment are inherently good.
They are not.
Technology is a means, not an end. Its value lies in what it enables us to achieve.
If AI helps discover a treatment for a serious disease, reduces energy waste or makes essential services more accessible, the environmental cost of that work may be justified. If vast amounts of electricity and water are consumed to produce marginal commercial advantages while ordinary people face rising bills and worsening environmental conditions, the case becomes much weaker.
We should also ask who gets access to the benefits. If publicly supported infrastructure helps create extraordinary private wealth, there is a legitimate question about how society should share in the resulting gains. Better public services, scientific research, education and lower-cost access to useful AI are all possibilities worth considering.
An egalitarian society does not require everyone to benefit identically from every technological advance. It does require us to take seriously the question of whether the gains are being shared fairly and whether the burdens are falling disproportionately on those with the least power to resist them.
That principle should guide the development of AI just as firmly as it guides housing, healthcare or energy policy.
The decision an AI politician would make
An AI politician would neither declare war on data centres nor give the technology industry an unrestricted licence to expand.
It would establish clear environmental standards, require developers to pay their fair share of infrastructure costs, encourage more efficient computing, favour genuinely low-carbon power and protect communities from unreasonable harm. It would support research into better cooling, energy storage and alternative computing methods, while rejecting projects whose costs clearly outweighed their public benefits.
It would also insist on regular reviews. AI technology is changing rapidly, and planning decisions made today could have consequences for decades. Governments should not commit enormous resources to facilities without considering how future improvements in efficiency, computing architecture and energy supply might change what is needed.
Above all, the policy would be based on evidence rather than fear, corporate lobbying or political competition.
The world does need computing infrastructure. But it does not need every proposed data centre, any more than it needs every proposed road, airport or shopping centre. Each should justify its place.
The real measure of technological progress is not how many machines we can build, how much electricity we can consume or how quickly a company can increase its valuation.
It is whether people’s lives become better.
If artificial intelligence is to help us build a more prosperous, peaceful and equitable world, then the infrastructure that powers it must be developed in a way that respects the planet and the people who live on it.
We should build the AI infrastructure we genuinely need, not sacrifice the world to satisfy an appetite for growth that has no natural limit.
Leave a Reply