Britain urgently wants to be an AI maker, but data centres and model training demand more than ambition. Without immediate reform of how energy is priced and delivered, the UK risks becoming a mere consumer of intelligence produced elsewhere, rather than an economy that creates it.
The UK continues to frame its AI future around innovation, skills and regulatory leadership. Yet the decisive constraint sits elsewhere. AI is limited by physics long before it is by software, and electricity has become the defining input, shaping how intelligence is trained, deployed, and monetised.
The gap between ambition and infrastructure reality is widening at an alarming rate. Large-scale AI infrastructure is now dictated by power availability and power economics. On that basis alone, the UK is rapidly pricing itself out of contention for the most valuable layer of the AI stack.
According to Spencer Lamb, Managing Director and Chief Commercial Officer at Kao Data, the consequences are already visible in how hyperscalers decide where to invest. “The UK now has possibly the highest energy costs globally, and AI training is inherently energy intensive,” he says. “The more intensive the compute, the higher the electricity bill.
“Countries with high energy costs are therefore structurally unattractive for organisations deploying large-scale AI. Compute will still be placed in the UK to serve customers here, but large-scale AI training is happening elsewhere, predominantly in the United States, in the Nordics, where energy is cheap, in the Middle East, where gigawatt-scale data centres are being built, and in Spain, where solar has fundamentally changed electricity economics.”
That distinction is critical. Serving only domestic demand captures limited value and risks falling behind. Urgent action to train models is essential to secure the economic upside, from infrastructure investment through to intellectual property, skills development, and downstream innovation. Where energy prices diverge materially, no policy signalling can override the arithmetic.
Energy economics decide everything
AI training workloads operate at densities that have made electricity the primary factor influencing data centre locations, surpassing labour, land, or connectivity. For Lamb, this shift has fundamentally altered how technology strategy is formed. “When people talk about compute today, they immediately talk about the power required to sustain it,” he adds. “Utilities have become central to technology decisions in a way they never were before. In the United States, it has gone to extremes, with hyperscalers openly discussing recommissioning nuclear assets to support compute demand.”
AI infrastructure is now deployed primarily where power is affordable and accessible, and the UK lags due to historic underinvestment and persistent high prices. This is the result of years of infrastructure decisions. Power constraints now dictate data centre location, size, and competitiveness, so infrastructure shifts to regions with abundant, cheap, and supported power. In the UK, data centres cluster around London based on past demand, not current AI needs.
The government’s AI Action Plan aimed to make the UK an AI maker, but without reforms to energy pricing, the ambition risks remaining rhetorical. Lamb argues that the ambition to be an AI maker is right and should be applauded, but without action on energy costs, the UK will consume rather than create AI models, losing out on associated economic value.
This distinction is not academic; it will soon define whether AI bolsters domestic productivity and technological sovereignty or forces the UK to depend on external platforms. Energy pricing is critical, and unless immediate reform happens, current mechanisms will continue to undermine competitiveness for high-consumption industries, regardless of their strategic importance.
A broken pricing model
The UK’s electricity pricing framework links consumer costs to the most expensive marginal source of generation. In recent years, that source has been gas. At the same time, renewable generation has expanded rapidly, creating a structural contradiction at the heart of the system.
Lamb is blunt about how that contradiction plays out. “The way electricity is priced in the UK is based on an outdated model that links what consumers pay per kilowatt hour to the most expensive source of energy on the system,” he states. “Since the Ukraine war, there has been gas. We import a significant amount of gas because we have not invested sufficiently in domestic generation capacity.
“Before the war, that was manageable. After it, costs escalated dramatically. Yet over the past decade, the UK has built substantial renewable capacity. Around 45 per cent of electricity consumption now comes from renewables, which cost roughly a third of gas, but consumers still pay the gas rate for that electricity.”
The result is a system that fails to reward renewable generation with lower prices while penalising continuous, high-load users irrespective of sustainability performance. “Those green electrons are fed into the grid and blended with gas-generated power,” Lamb adds. “Consumers receive a mixed supply but pay the highest marginal price. The system does not reflect how electricity is produced, and it fundamentally distorts investment decisions.”
For data centres operating around the clock, this erodes global competitiveness and undermines the economic case for building near renewable energy sources.
Why growth zones fall short
AI Growth Zones were meant to urgently accelerate infrastructure investment through planning support and improved access to power. But these measures only address symptoms, not causes. Without rapid reform of energy pricing, these zones cannot fundamentally change the investment equation.
“The incentive must be energy pricing,” Lamb says. “If electricity costs were linked to the cost of renewable generation rather than gas, it would justify the level of financial commitment required. Without that, the business case simply does not work.”
Responsibility for energy pricing and industrial strategy remains fragmented across government departments, creating urgent and damaging friction precisely where alignment is critical. To address this, Lamb urges that the government must immediately align responsibility for energy pricing and industrial strategy under a single, coordinated authority. “The issue is not whether people understand the problem,” Lamb continues. “The issue is that energy pricing sits with one part of government, and industrial strategy sits with another. Until those are aligned, initiatives like AI Growth Zones will struggle to deliver meaningful change.”
Infrastructure myths and real constraints
Public debate around data centres often focuses on planning delays, water consumption and environmental impact. While scrutiny is legitimate, much of the narrative is poorly grounded in operational reality. One persistent misconception concerns water usage.
“There is a widespread belief that data centres consume enormous amounts of water, which in the UK is simply not true,” Lamb explains. “We operate closed-loop cooling systems, meaning water is not evaporated. From a utility’s perspective, data centres are not draining lakes or rivers in the way they are often portrayed.”
From a technical standpoint, the UK can support advanced AI infrastructure. Higher density compute requires different cooling approaches, including direct liquid cooling, but these are engineering challenges that have already been solved. Facilities capable of supporting AI workloads exist. The skills base exists. “The constraint is not technology,” Lamb says. “The constraint is whether energy economics make large-scale deployment viable.”
AI does not replace traditional cloud computing. It complements it. Enterprise workloads and CPU-based cloud services will continue to grow steadily. AI inference services sit alongside those platforms, augmenting productivity rather than displacing existing infrastructure. The strategic question for the UK is not whether it will host data centres, but what role they will play.
“If current energy costs persist, the UK will host consumption infrastructure rather than production infrastructure,” Lamb adds. “Applications will run locally, but models will be trained elsewhere.” That distinction shapes long-term competitiveness, economic value creation and technological sovereignty.
Without urgent reform of how electricity is priced and delivered, such as adjusting tariffs to support AI innovation or investing in grid infrastructure, decisions about the future of British AI will continue to be made outside the country, not because of a lack of ambition, but because the power economics no longer add up.
