Prime Minister Keir Starmer was early to the AI party when he announced in January the UK’s AI Opportunities Action Plan: a series of policies designed to jump-start the economy with artificial intelligence. The plan is ambitious; it articulates the government’s intention to boost “sovereign” compute power in AI and attract billions of pounds in data centre investment. But key questions remain over how to make the UK an attractive destination for data centre projects – and how to run those projects sustainably.
Organisations’ race for AI dominance has produced a boom in data centre infrastructure. Training and operating the large language models that underpin GenAI is resource-intensive. With more data centres needed to keep the wheels turning, Silicon Valley’s hyperscalers have increased their spending on physical infrastructure by billions of dollars.
Power really is the biggest problem
When it comes to attracting data centre investment, the UK is in a uniquely challenging position thanks to its energy supply and planning systems. According to the Social Market Foundation, a public policy think-tank, powering a 100MW data centre in the UK costs four times as much as an equivalent project in the US. What’s more, near-peer countries in Europe can provide greener electricity at a far lower price.
“Power really is the biggest problem, particularly in the south east of the UK,” says Cathal Griffin, chief revenue officer at Asanti, a data centre provider. In the UK, it’s not unusual for a data centre to pay between 24p and 28p per kilowatt hour. That’s a great deal more than in countries such as Norway, where power is available for 4p or 5p per kilowatt hour. Griffin says that last year the UK lost multiple data centre projects to Norway, largely thanks to the cost of electricity.
Power surge
Before the GenAI boom, a data centre would require “three, four, five, maybe six kilowatts (kW) per rack,” Griffin explains. But the latest AI servers from Nvidia require roughly 130kW to 250kW, and next-generation servers may need as much as 900kW each.
“You do the arithmetic and extrapolate the price difference,” says Griffin. “There’s no way we’re going to be attracting AI development to the UK when power costs that much.”
And there’s no reason to believe efficiency innovations will solve this problem. Analysts expect models that are allegedly more power-efficient, such as Deepseek’s R1 model, to have little impact on the total energy consumption of AI-factory data centres. Providers would satisfy their current demands more efficiently but would likely scale up the amount of servers in use, consuming the same amount of energy only with far more machines.
The seemingly inevitable growth in power demands has also brought data centre sustainability under the spotlight. Data centres come with considerable resource demands – water and electricity. The International Energy Agency expects data centre electricity consumption to double from 2022 to 2026. The National Grid, meanwhile, has warned that AI use could increase the UK’s total electricity demand by 500% in the next decade.
There’s no way we’re going to attract AI development to the UK when power costs that much
Westminster has acknowledged these problems. As part of its action plan, the government announced the creation of an energy task force called the AI Energy Council.
Chaired by the UK’s science and energy secretaries, the council will collaborate with utilities companies to understand how AI impacts energy consumption and the challenges related to increased future use. It will also investigate the possibility of using nuclear power as a sustainable energy source. It will even encourage research on nuclear fusion, a possible solution to energy scarcity for all humanity – of course, major scientific breakthroughs are needed to make the technology feasible.
Until such breakthroughs occur, there are “questions about how we’re going to coordinate our energy policy with this vision of becoming a globally competitive centre for AI,” says Oliver Johnston-Watt, an investment analyst at Par Equity, a venture-capital firm.
Without reductions in the price of energy, he says, private enterprises will struggle to keep pace with the development of the technology. “The reality is the UK has some of the highest energy costs in the advanced industrialised world and the G7,” says Johnston-Watt. “For us to be globally competitive, we are going to have to bring that cost per kilowatt hour down quite considerably.”
One possible solution is to implement dynamic pricing to reflect regional differences in energy costs, he suggests. Another option is subsidies, which may be worth considering if the UK is to keep up with “far better capitalised and larger economies, such as the US and, to some extent, the EU.”
Land grab
The vast majority of data centre capacity in the UK is in or around London. But capacity problems are delaying further development projects, with 400GW of data centres awaiting approval for grid connection.
The government hopes to alleviate the pressure by introducing AI Growth Zones (AIGZs) with fewer planning restrictions for data centre developments. The first of these zones will be located in Culham, Oxfordshire, near the UK Atomic Energy Authority.
But these development zones could massively impact communities in neighbouring areas, warns Alvin Nguyen, a senior analyst at Forrester, a research and advisory firm. A little over a decade ago, one small data centre might have run at only 200 kilowatt hours. But new AI-factory data centres, packed with the latest and greatest Nvidia AI servers, could easily demand as much as 1 gigawatt hours – that’s 5,000 times more electricity.
“If you put something requiring that much power next to other industrial, commercial or residential areas, it’s competing for space, water and energy – balance is needed, otherwise these data centres will harm, rather than benefit, the surrounding areas.”
AIGZ: a green opportunity
The government has also resuscitated projects that were denied by local governments owing to their encroachment on green-belt land. This is a mistake according to Spencer Lamb, chief commercial officer at Kao Data, a data centre provider that redeveloped a former industrial facility in Greater Manchester. “Unequivocally, we see no immediate need to build on greenfield sites, as ample brownfield locations across the UK are ripe for sustainable development,” he says, adding that “building on greenfield sites risks further backlash from the general public.”
A key challenge with data centres is ensuring access to reliable and sustainable energy. Data centre investments could create an opportunity for the UK to develop its clean-energy industry, so long as AIGZs are effectively aligned with the nation’s net-zero and sustainability goals.
“The success of the AI Opportunities Action Plan depends on getting both the data centre foundations and the energy supply right,” Lamb says. “That means embracing a clear planning framework, prioritising brownfield redevelopment and working closely with renewable-energy partners.”
Rather than stamping the seal of approval on any and all data centre projects to rack up headline-friendly investment figures, the government should carefully work to identify brownfield sites and regenerate them, ensuring good access to sustainable power and connectivity. The co-development of data centres and green solutions could help build towards Starmer’s vision for the UK and provide the competitive edge the nation so sorely needs.
Prime Minister Keir Starmer was early to the AI party when he announced in January the UK’s AI Opportunities Action Plan: a series of policies designed to jump-start the economy with artificial intelligence. The plan is ambitious; it articulates the government's intention to boost “sovereign” compute power in AI and attract billions of pounds in data centre investment. But key questions remain over how to make the UK an attractive destination for data centre projects – and how to run those projects sustainably.
Organisations' race for AI dominance has produced a boom in data centre infrastructure. Training and operating the large language models that underpin GenAI is resource-intensive. With more data centres needed to keep the wheels turning, Silicon Valley's hyperscalers have increased their spending on physical infrastructure by billions of dollars.
Power really is the biggest problem
