Britain is being promised an artificial intelligence revolution. But every new AI data centre needs electricity — enormous quantities of it, reliably, every hour of every day. At the same time, the Government wants to decarbonise electricity, electrify transport and heating, and bring down bills. The collision between those ambitions is no longer hypothetical. It can be seen in Norfolk, Devon, Yorkshire and in the investment decisions of major technology firms.
This is not an argument against computers, digital services or innovation. It is a demand for a credible national energy plan. If Britain wants the jobs, investment and technological sovereignty that data centres might offer, somebody must explain where the power will come from, when the grid will be ready, what it will cost — and who pays.
THE NEW POLICY: FAST-TRACK THE VERY DEMAND THE GRID STRUGGLES TO SERVE
On 28 October 2026, the Government published its Data Centre Covenant. Strategically important projects may be given help to secure grid connections more quickly, access reallocated capacity and, exceptionally, use temporary on-site electricity generation until a grid connection is ready.
The Covenant says operators should meet the additional costs they impose on the energy system and should protect households and other businesses from those costs. It also argues that more large users could help spread the grid’s fixed costs across a wider base. Both claims deserve testing: lower bills are not an automatic consequence of adding a vast new load if that load also requires new generation, reinforcements, substations or balancing services.
The Government’s own Clean Power 2030 plan seeks a predominantly clean electricity system, while explicitly retaining gas-fired backup when needed. The question is not whether a data centre can ever operate in a lower-carbon system. It is whether the investment, construction timetable and dispatchable capacity needed to serve that demand have actually been demonstrated.
LONGMARSH, NORFOLK: A GIGAWATT OF DATA CENTRE BESIDE 1.3GW OF SOLAR
In west Norfolk, EPL Norfolk Limited has announced Longmarsh Green Energy: a proposed 1GW data-centre campus alongside up to 1.3GW of solar generating capacity and battery storage. The development spans two sites near King’s Lynn, with a reported combined area of roughly 4,695 acres. The developer refers to a possible 2031 grid connection and investment exceeding £25 billion, including computing equipment. Consultation runs to 26 November 2026.
The essential engineering question is simple. 1.3GW of solar panels is not 1.3GW of guaranteed electricity at midnight, on a dark winter evening or during prolonged poor weather. That figure is the panels’ maximum rated output under specified conditions. A data centre needing a large continuous supply requires firm grid capacity, backup, or another dependable source when the sun is not producing. Batteries can shift electricity for a limited time, but their power rating alone tells us nothing about how many hours they can sustain the load.
For perspective, a facility actually drawing 1GW continuously for a full year would consume 8.76 terawatt-hours of electricity. That is an illustrative full-load calculation, not a forecast of Longmarsh’s eventual consumption. Any application should publish a realistic annual energy forecast, an hourly supply assessment, the battery’s energy capacity and duration, and a credible connection date — not just an impressive solar nameplate figure.
There is another public interest: agricultural land and landscape. The developer’s own FAQs say the solar installation is envisaged for a 60-year life, while parts of the western data-centre infrastructure would remain. Local communities deserve to know precisely what land will be permanently transformed, how the two sites connect and what the costs and benefits will be.
DEVON: XLINKS TURNS TO A 1.5GW AI CAMPUS
At Alverdiscott in North Devon, Xlinks is proposing the Devon Data Campus, centred on a data-centre scheme called Valeon. The promoter describes up to 1.5GW of AI-optimised compute capacity with a separately proposed 1.8GW battery energy storage facility. Xlinks estimates capital investment of around £12.2 billion to £13.8 billion and hundreds of permanent jobs. These are developer estimates for proposed developments, not realised results.
The developer says the data centre and battery proposals are separate and can operate independently. That distinction matters. A 1.8GW battery power rating does not mean 1.8GW is available around the clock. Without the accompanying gigawatt-hour capacity, operating strategy and recharging plan, the public cannot judge the duration of support. If the data centre’s computing load reaches its proposed scale, the additional electricity for cooling and other systems must also be included when testing total site demand.
The local debate also concerns land, water, visual impact and grid capacity. CPRE Devon has raised concerns about the size and countryside effects of the development. Xlinks maintains that its proposal could bring jobs, skills and investment. Both positions need to be examined against a transparent statement of the site’s full electricity demand and its physical connection requirements.
Xlinks previously pursued the Morocco–UK electricity interconnector. That history is relevant to the company’s change in project emphasis, but it does not establish that funds or contractual rights from the former scheme have been transferred to the Devon campus. Such financial claims would require separate documentary proof.
LEEDS: THE RETURN OF MAJOR POWER DEMAND TO A FORMER POWER-STATION SITE
At Skelton Grange, Leeds, Harworth Group and Microsoft are progressing plans for a hyperscale data centre on land formerly associated with a power station. Harworth reported on 27 April 2026 that Leeds councillors had resolved to grant planning consent for roughly 500,000 square feet across three data-centre halls and auxiliary buildings, together with a separate logistics component.
Reusing previously industrial land is materially different from converting thousands of acres of working countryside. That is a genuine advantage. It does not, however, eliminate the need to demonstrate available electrical capacity. The council application is 25/06139/FU. A committee resolution is not the same thing as an issued final decision notice, and the reliable public material examined for this article does not establish an agreed megawatt demand for the Leeds data centre. It would be misleading to invent one.
That missing figure should be central to scrutiny. What is Microsoft’s phased peak electricity demand? Which reinforcements are required? When can the site genuinely be energised? Who pays for any dedicated infrastructure? Those questions are especially important as the Government considers special treatment for strategically significant data centres.
WHEN THE ELECTRICITY DOES NOT ARRIVE — OR COSTS TOO MUCH
This is not merely a theoretical warning. On 9 April 2026, OpenAI confirmed that it had paused the planned Stargate UK infrastructure initiative, pointing to energy costs and the regulatory environment as conditions affecting long-term investment. The Government described this as part of a wider recalibration of the Stargate programme and emphasised OpenAI’s continuing UK presence. The pause was not a complete withdrawal from Britain, but the stated concern about energy costs was real. Reuters reported the announcement.
A different problem arose in Loughton, Essex. On 19 October 2026, Reuters reported that Microsoft had ended its agreement to anchor an Nscale data centre after expected grid power had slipped into the 2030s. Microsoft said the decision concerned that single site and its wider Nscale partnership continued. This was not the Leeds Skelton Grange scheme, and should not be confused with it.
Taken together, the cases illustrate two distinct investment risks: power that businesses may regard as too expensive, and power that the network cannot deliver on the required date. Announcing a data centre is not equivalent to demonstrating that it can be supplied competitively and reliably.
THE GRID NUMBERS THAT THE PUBLIC SHOULD SEE
In evidence to Parliament, the National Energy System Operator (NESO) said its analysis points to around 5.2GW of connected data-centre capacity and just over 20TWh of electricity demand by 2030. It also reported connection requests exceeding 70GW, while stressing that many schemes may fall away, build slowly or use less electricity than their headline requests. The queue is not a forecast of 70GW of actual consumption.
That distinction is critical. It is wrong to add every announced project together as if all will run tomorrow. It is equally wrong to assume an electricity system designed around older demand assumptions can accommodate any number of new hyperscale facilities without changing its costs, reinforcement needs and security margins.
Britain already has a competitiveness problem. According to the Government’s international non-domestic electricity price tables, the average UK price for the large-business consumption band in the second half of 2025 was around 23.37p/kWh, against roughly 11.39p/kWh for the median of the EU countries and UK in the published comparison. Those are national band averages, not actual tariffs paid by Microsoft, OpenAI or a particular hyperscale campus.
Nor can high prices honestly be attributed to just one policy. Gas often sets the wholesale electricity price; network costs, taxes, levies and financing also matter. Nevertheless, a policy programme that calls for ever more electricity consumption has to explain how it will make that electricity internationally competitive — rather than merely promising that lower costs will appear later.
THE NET ZERO PARADOX: WHO GETS PRIORITY, AND WHO PAYS?
Here is the emerging contradiction. Families and businesses are being encouraged or required, through different policies, to rely increasingly on electricity. The Government also wants rapid expansion of AI infrastructure that needs exceptional levels of dependable electricity. Meanwhile, some major investors say UK electricity is too expensive, and others have faced connection dates far beyond their intended opening.
The October Covenant proposes preferential help for the most strategically important data centres, including temporary on-site generation in exceptional circumstances. It promises that operators should cover their fair share of system costs. That makes the rules for defining a fair share a major public-interest issue: are new substations and network upgrades fully charged to their beneficiaries, or will some costs ultimately be recovered from everyone using the grid? The answer depends on the connection and charging arrangements, which must be published and tested.
It is possible to support AI investment and a lower-carbon electricity system together. But doing so is a matter of engineering, prices, land, time and public accounting — not slogans. Renewable electricity can supply valuable energy; gas, nuclear and other dependable capacity can support resilience; batteries provide time-limited flexibility. None of those facts excuses an incomplete explanation of how a 24-hour industrial load will be served during a windless winter evening.
SIX ANSWERS THE GOVERNMENT AND DEVELOPERS SHOULD PUBLISH
First: the credible peak megawatt demand, annual terawatt-hours and phased energisation timetable for each major campus. Second: the actual point of grid connection, all essential enabling works and the earliest physically achievable connection date.
Third: an hour-by-hour supply and security assessment, including winter shortages, backup generation, curtailment and the duration — in megawatt-hours or gigawatt-hours — of any battery. Fourth: the whole-system cost of dedicated and shared network reinforcement, with a clear account of how charges are allocated.
Fifth: a published land and water assessment that compares brownfield options against losses of agricultural land and environmental impacts. Sixth: a credible electricity price pathway that works for new technology investors and existing manufacturing, businesses and households.
BRITAIN DOES NOT NEED ANOTHER SET OF PROMISES. IT NEEDS THE POWER PLAN.
The AI economy is an opportunity. It also presents a hard test of whether Britain’s energy ambitions have been matched to its physical infrastructure. Longmarsh, Devon and Leeds should be assessed as real electricity and land-use proposals, not just as attractive investment headlines. OpenAI’s pause and Microsoft’s separate Essex exit show why the risk cannot be ignored.
The Government cannot claim to be delivering lower bills, faster electrification and an AI-driven industrial future without demonstrating — in public — the generation, grid connections, reliable capacity, land requirements and costs that make those promises compatible. That is the evidence taxpayers, communities and investors are entitled to see.
MAKE THEM SHOW THE EVIDENCE. Our parliamentary petition calls for an independent national assessment of the full costs and impacts of current climate targets, including the rising demand from data centres. Read and sign petition 780321 here.
Shane Oxer — Hope to Save Our Countryside. 29 October 2026.
Sources and further reading: HM Treasury Data Centre Covenant (28 October 2026); NESO evidence to Parliament; Longmarsh project announcement; Devon Data Campus; Harworth / Microsoft Leeds announcement; DESNZ electricity price statistics; Reuters on Stargate UK; Reuters on Essex / Nscale.
Leave a Reply