The rapid expansion of artificial intelligence has pushed data centres from a relatively specialist infrastructure issue into the centre of the political debate.
That debate is often framed around electricity demand, grid capacity and water consumption, but those issues should not be treated as interchangeable. Power availability and grid capacity are already tangible constraints on data centre growth in the UK. Water, by contrast, is a more contested part of the domestic debate: there are no official figures for total UK data centre water consumption, and operating models, cooling technologies and local conditions vary considerably. Importing assumptions from other markets risks obscuring the challenges the UK sector actually needs to solve.
Calls to restrict new data centre development, however, risk framing the issue too narrowly. The more important question is not simply how many data centres the UK builds, but how efficiently we use the computing infrastructure and energy capacity already available.
If the UK wants to grow its AI economy while meeting its environmental and energy objectives, efficiency needs to become a fundamental part of its digital infrastructure strategy.
DATA CENTRES ARE CRITICAL NATIONAL INFRASTRUCTURE
It is easy to talk about data centres purely in terms of their energy footprint, but that overlooks the role they now play across the wider economy.
Banking, healthcare, telecommunications, public services, manufacturing, retail, cloud computing and an expanding range of AI applications all depend on digital infrastructure. As organisations digitise more processes and deploy increasingly sophisticated AI models, that dependence will only grow.
The challenge is that the physical infrastructure underpinning this transformation cannot expand without limits.
Power availability is already a significant consideration in where and how data centres can grow. Grid connections can take years, while adding generation and transmission infrastructure is a much slower process than deploying a new digital service or AI application.
AI therefore creates something of a paradox. The technology promises to make businesses and economies more productive, but delivering that opportunity requires significant amounts of physical infrastructure and energy.
Restricting infrastructure may suppress one side of that equation, but it does little to address the fundamental efficiency challenge.
MORE COMPUTE CANNOT ALWAYS MEAN MORE POWER
Historically, growing demand for computing capacity could largely be answered by adding more infrastructure.
AI is making that model increasingly difficult to sustain. Training and running AI models can require significant computing resources, while the move towards higher-density infrastructure places additional pressure on power and cooling systems.
If every increase in compute demand requires a broadly corresponding increase in energy consumption, the industry will eventually collide with physical constraints.
That means the relationship between compute and power has to change. The objective should increasingly be to deliver more useful computing work from each unit of energy consumed. That requires the industry to look beyond headline capacity and consider how effectively infrastructure performs against the workloads it is actually running.
There is significant scope to do this because server estates are dynamic environments. Workloads rise and fall, applications behave differently throughout the day and computing requirements rarely remain constant.
Yet infrastructure can still be configured around relatively static assumptions about the resources that might be required rather than what is required at any particular moment.
At the scale of a modern data centre, the resulting inefficiencies can become significant.
EFFICIENCY SHOULD COME BEFORE EXPANSION
This is where the debate around data centre development needs greater nuance.
New capacity will undoubtedly be required – the continued expansion of cloud services, AI and digital applications means it would be unrealistic to assume existing infrastructure can accommodate every future workload.
But before seeking additional power capacity or expanding their server estate, operators should first understand how efficiently their existing compute resources are being utilised.
That principle is commonplace elsewhere – a manufacturer facing higher production demand would not automatically build another factory without first understanding whether its existing production lines could operate more effectively.
The same thinking should increasingly apply to digital infrastructure. Improving utilisation, reducing unnecessary server energy consumption and matching resources more closely with real workloads can potentially create additional headroom within existing power envelopes.
That headroom matters because power is already one of the most valuable resources within the data centre sector, with access to sufficient grid capacity increasingly shaping where facilities can be built and how they can grow. In the US, major hyperscalers are even acquiring power generation assets to secure future supply, while in the UK, grid connection availability and capacity can be as significant a consideration as geographical location itself.
A megawatt that does not need to be consumed by an existing workload may effectively become capacity that can support additional computing demand. Efficiency therefore becomes not simply an environmental objective, but a capacity strategy.
AI CAN HELP ADDRESS THE PROBLEM IT IS CREATING
There is also an important role for AI itself. Modern computing environments are too complex and fast-moving for every optimisation decision to be handled manually. Thousands of servers can experience continually changing workloads, making static configuration or periodic adjustment increasingly inadequate.
AI and automation can enable infrastructure to respond more dynamically. By continuously interpreting workload behaviour, utilisation and other operating conditions, automated systems can make frequent adjustments to how computing resources consume energy.
The significance of this approach lies in its ability to operate continuously rather than periodically.
Human expertise remains essential for setting strategy, managing risk and determining the guardrails within which infrastructure operates. Automation can then handle large numbers of routine adjustments that would be impractical for teams to make manually and in real time.
This is an important distinction – the objective is not autonomous infrastructure for its own sake, nor removing people from data centre operations.
It is about applying automation where the scale and speed of decision-making exceeds what manual processes can reasonably achieve.
POWER EFFICIENCY AN AI COMPETITIVENESS ISSUE
The discussion also extends beyond sustainability. Countries around the world are competing to attract AI investment, develop domestic capability and build the infrastructure required to support increasingly compute-intensive economies.
Access to power is part of that competition. If electricity and grid capacity become limiting factors, the ability to extract more computing capacity from available energy could directly influence where AI workloads can be deployed and expanded.
Efficiency should therefore be considered alongside new generation, grid investment and data centre construction as part of a broader infrastructure strategy.
None of these measures replaces the others. Renewable generation cannot compensate for inefficient computing infrastructure, just as optimisation cannot remove the need for additional energy and data centre capacity.
The strongest approach combines them. New infrastructure should be designed to operate as efficiently as possible from the outset, while existing estates should be continually assessed for opportunities to reduce unnecessary consumption and release capacity.
WE NEED BETTER MEASURES OF DIGITAL GROWTH
There is also a broader question about how the success of the data centre industry is measured.
More buildings, more servers and greater power consumption are relatively straightforward indicators of growth, but they tell us little about the productivity of the infrastructure.
As energy constraints become more significant, metrics around compute efficiency should receive greater attention.
How much useful computing work is being delivered from the energy consumed? How much capacity is sitting unused or underutilised? How effectively can infrastructure adapt as workloads change?
These questions will become increasingly important as AI reshapes demand. The industry may ultimately need to think about digital infrastructure in the same way other sectors think about resource productivity: not simply how much input is available, but how much economic value can be generated from it.
MOVING BEYOND THE FALSE CHOICE
The debate around data centres should not become a choice between unrestricted expansion and restricting development.
The UK needs digital infrastructure. It also needs an energy system capable of supporting economic growth without undermining wider sustainability objectives.
Those goals are not inherently incompatible, but achieving both will require a greater focus on efficiency.
Building more renewable generation, expanding grid infrastructure and developing new data centres will all be necessary. So will reducing the amount of energy unnecessarily consumed within the infrastructure we already operate.
As AI adoption accelerates, power may prove to be one of the defining constraints on its growth.
The response should therefore be neither to build without limits nor simply to stop building. It should be to make every available watt of power work harder.
That is how the UK can support a growing AI economy while creating a more efficient and sustainable digital infrastructure system.
