Can Britain’s grid keep up with the datacentre boom?

Anatomy of a £1bn grid upgrade

The North West London Upgrade shows just how much physical infrastructure is needed to accommodate new datacentre facilities.

The project will see the installation of around 60km of new high-voltage cable, 35km of re-cabling between Sundon and Elstree, and upgrades to more than 200km of overhead lines. At Elstree, the agency is also extending an existing 400kV substation, and building a new 132kV substation at Letchmore Heath to connect the new datacentres.

The upgrade is only one part of a much bigger programme of investment in Britain’s electricity network. Ofgem’s RIIO-3 price control, for example, includes up to £10.7bn for electricity transmission between 2026 and 2031, with flexible mechanisms potentially taking investment to as much as £90bn as projects mature.

Elisabeth Cremona, energy infrastructure lead at energy think tank Ember, tells Computer Weekly that the scale of investment is a response to the unprecedented speed at which these new facilities come online.

[Datacentre] developers are behaving differently to any previous class of large energy user. Rather than wait patiently in established queues, they’re moving to find loopholes, to find a place where they can connect immediately

“Datacentres represent a huge load,” she says. “You’re essentially looking at putting 500,000 houses on the grid in a single location really, really fast, and that’s just never happened before. Households don’t come online that fast, factories don’t come online that fast.”

Typically, says Cremona, building out or upgrading a grid takes between five and 15 years, while a datacentre can be ready to come online in 18 months. Bringing datacentres online quickly is also fast becoming a competitive advantage for businesses.

“Developers are behaving differently to any previous class of large energy user,” she says. “Rather than wait patiently in established queues, they’re moving to find loopholes, to find a place where they can connect immediately.”

For years, the UK’s datacentre industry has been heavily concentrated around London and the M4 corridor, driven by proximity to customers, existing fibre networks, established datacentre ecosystems and access to major business markets.

Now, with increased onus on speedy connection times and a backlog of connection requests, projects are shifting north.

Analysis of data from construction project information provider Barbour ABI finds that of the UK’s top 10 datacentre projects, only one – at East Havering in Essex – is in the London area. The remainder, totalling just under 4GW, are spread north from Oxfordshire to Lincolnshire, North Wales, the north-east and Scotland.

A changing grid landscape

According to Ofgem, connection offers rose from 41GW to 125GW between November 2024 and June 2025, with datacentres accounting for at least 80GW of that demand. Connection wait times have stretched as a result, some by as much as 15 years.

Part of the backlog is self-inflicted. Developers, aware of how slow the process is, submit multiple speculative applications for the same site to see which clears first, and that creates “zombie applications”.

Ofgem has proposed reforms intended to remove speculative and non-viable projects from the queue. Large datacentre developers would have to pay an upfront commitment fee when accepting a connection offer, which would range from £237,500 to £712,500 per megawatt. The fee would be refunded when a project reaches energisation but forfeited if it leaves the queue early.

Developers would also have to demonstrate that projects are commercially and technically ready to progress.

Matt Salter, global head of datacentres at Onnec, says the proposals reflect how tight the market has become.

“Ofgem’s ‘Pay to Stay’ proposals are a sign of just how tight the UK’s datacentre power market has become,” he tells Computer Weekly. “Asking developers to prove projects are real and ready to move makes sense, but it adds another pressure point for operators already being pushed to build faster.”

Research by Onnec shows that 92% of operators are compressing build timelines to keep pace with artificial intelligence (AI) demand, while 75% say speed-to-market pressure is forcing design decisions before infrastructure requirements are fully understood.

Salter says this creates the risk of adding to grid capacity without properly coordinating it with the rest of a facility’s infrastructure.

“The £1bn National Grid upgrade shows just how much investment is now needed, but more grid capacity on its own will not fix this,” he says. “Power, cooling, cabling and compute need to be planned together from day one, or we risk spending billions connecting new datacentres only to spend more fixing them after go-live.”

The issue is not unique to Britain. In March, Denmark’s grid operator suspended new connection requests after receiving roughly 60GW of applications in a country whose peak demand is around 7GW.

An increasingly common response to lengthening connection times is for developers to simply move where there is capacity.

Follow the power: One operator’s move north

For Spencer Lamb, CEO of datacentre operator Kao Data, the shift described above is already shaping investment decisions.

“Grid availability is one of the most important factors in determining when and where new datacentres are built,” he tells Computer Weekly. “If you cannot secure power within a commercially viable timeframe, it becomes extremely difficult to justify investing in a location, regardless of its other advantages.”

As a result, he says, operators are increasingly choosing sites where capacity already exists rather than waiting years for network upgrades. It is one reason Kao Data is investing £450m in a new facility in Greater Manchester. Lamb says the city region offered tech and AI research heritage, government support, a brownfield site that could be redeveloped for the benefit of the community, and a realistic timeline for power delivery.

We need to deliver new network upgrades before the UK’s AI and economic opportunity is lost

He cautioned against reading the National Grid’s North West London Upgrade purely as a response to AI. Some substations, he says, have been in planning for many years, driven by demand for cloud, enterprise computing and digital services.

“What AI has done is accelerate the conversation and intensify pressures,” he says. “The reality is that the UK’s electricity infrastructure has been under increasing strain for some time, and significant long-term investment will be required if the country wants to maintain its position as a leading destination for datacentre and AI investment.”

Who pays for the new infrastructure?

While developers do bear some of the cost of these new infrastructure buildouts, the costs of expanding the grid are ultimately recovered through regulated network charges paid by households and businesses. So, the question of who pays for new infrastructure is increasingly significant for the public as power demand ramps up.

An Ofgem spokesperson tells Computer Weekly it is alert to the risk that “consumers should not unfairly shoulder the costs of connecting” datacentres, and that its bid to clean up the connection queue aims to mitigate this risk.

The regulator’s Demand Connections Reform programme is also designed to reduce the likelihood of consumers funding infrastructure that is ultimately not needed because projects fail to progress.

“Our recent consultation will help shape reforms that will help to ensure viable projects, including datacentres, connect faster and network costs are shared fairly,” says the spokesperson.

Concentration of datacentres also matters, with location materially affecting network costs, reinforcement requirements and connection timescales.

“If you suddenly get a massive cluster of datacentres,” says Ember’s Cremona, “the system operator has to build out a huge amount of transmission lines, substations, and all these very, very expensive things to supply them with power.”

With developers pushed to areas of less concentrated connection, she warns they run the risk of prioritising speed over everything else – including cost for businesses and consumers. Additionally, datacentres rarely draw as much power as they request, meaning bills could rise by more than the actual load justifies.

“They operate on average at around 50%,” says Cremona. “So that means you build a huge amount of transmission infrastructure, and they only use half of it. They’re not even using it fully, and you’re billed in a way that isn’t the most cost-efficient.”

Managing datacentre buildout is therefore a necessity in terms of not overwhelming the grid and in keeping costs down for the public. One solution is to make datacentre demand more flexible, though this remains a difficult pill to swallow for operators facing consumer demand for always-on cloud and enterprise workloads. Cremona says it’s still something worth aiming for, though the road to get there may be long.

“If you could enforce flexibility, developers would get online faster and you would bring down network requirements,” she says. “It’s really a win-win for everyone.”

Can the grid keep up with AI?

Britain’s datacentre ambitions are increasingly becoming an electricity infrastructure problem. The North West London Upgrade demonstrates the physical scale of investment required to connect just five datacentre facilities, while the wider connection queue shows how quickly demand is outstripping available capacity.

Asked whether the UK is doing enough to keep power availability in step with planned capacity, Kao Data’s Lamb is blunt: “In short, no.”

Too often, he says, discussions focus on securing capacity in a handful of established markets, rather than taking a more strategic view of where power is available today and how future demand can be distributed across the country.

“The remaining bottleneck is therefore not simply generation capacity itself, but the speed, transparency and coordination needed to connect demand with available power,” he says. “We need to deliver new network upgrades before the UK’s AI and economic opportunity is lost.”