OpenAI Pauses UK Data Center Deal Citing Energy Costs and Regulation

On April 9, 2026, OpenAI confirmed it was pausing Stargate UK, the flagship AI data centre project it announced just seven months earlier in partnership with Nvidia and British cloud provider Nscale [1]. The company's stated reason was blunt: "We continue to explore Stargate UK and will move forward when the right conditions such as regulation and the cost of energy enable long-term infrastructure investment" [2].

The pause landed as a public embarrassment for the Starmer government, which had trumpeted the project during President Trump's September 2025 state visit as centrepiece evidence that Britain could become an "AI superpower" [3]. It also raises harder questions — about the structural competitiveness of the UK as a location for energy-intensive AI infrastructure, about whether political ambition has outrun regulatory reality, and about what happens to a country's bargaining position when the companies it courted start building elsewhere.

The Deal That Was

Stargate UK was announced in September 2025 as a sovereign AI infrastructure initiative. OpenAI committed to deploying up to 8,000 Nvidia GPUs at sites in North East England — primarily Cobalt Park in North Tyneside and facilities near Blyth — with the potential to scale to 31,000 GPUs over time [4][5]. The project was designed to anchor the government's newly designated AI Growth Zone in the North East, which officials projected would create 5,000 jobs and attract £30 billion in private investment across the region [6].

OpenAI also committed to making the infrastructure available for public services and national security applications [7]. Former Chancellor George Osborne was recruited to lead Stargate's international expansion, while former Deputy Prime Minister Nick Clegg joined the board of Nscale, which had recently raised a record $2 billion Series C at a $14.6 billion valuation [5].

The project sat within a broader £31 billion package of UK tech investment announced during the state visit [3]. Now, with the anchor tenant stepping back, questions surround the viability of both the AI Growth Zone and Nscale's UK ambitions. Nscale declined to comment on the pause [5].

The Price of Power

The most concrete obstacle OpenAI identified is the cost of electricity. UK wholesale electricity prices averaged approximately $115 per megawatt-hour in the first half of 2025, according to the International Energy Agency — nearly 2.4 times the US average of $48/MWh and significantly above France ($73/MWh) and the Nordic countries ($40/MWh) [8].

Source: IEA Electricity Mid-Year Update 2025

Data as of Jul 1, 2025CSV

For a data centre consuming 100 megawatts or more of continuous power, this differential compounds from a line-item cost into a structural disadvantage. OpenAI's own framing described UK industrial electricity prices as "more than four times" those in the United States, Finland, Norway, and Sweden [4]. At scale, the gap translates to hundreds of millions of dollars annually in additional operating costs — money that buys more GPUs, more training runs, and more competitive advantage in markets where power is cheaper.

UK futures prices for 2026 indicate roughly 10% lower prices compared to 2025 [8], but that still leaves Britain well above most competitor markets. The underlying cost drivers — the UK's gas-dependent generation mix, network charges, and environmental levies — are structural rather than cyclical.

The Grid Bottleneck

Even if electricity were cheaper, getting connected to the grid presents its own multi-year obstacle. UK grid connection requests surged from 41 gigawatts in November 2024 to 125 gigawatts by June 2025, with approximately 75 GW of that queue attributed to data centre projects alone [4][9].

Source: National Grid ESO

Data as of Jun 1, 2025CSV

The mismatch is stark: a data centre can be built in 18 to 24 months, but securing a grid connection takes three to eight years under current timelines [4]. The Department for Science, Innovation and Technology (DSIT) has acknowledged that "connection to the electricity grid for AI data centres is the single biggest blocker for establishing AI Growth Zones" [10].

The government has taken legislative action. The Planning and Infrastructure Act 2025, which received Royal Assent in December 2025, enables large-scale data centres to be classified as Nationally Significant Infrastructure Projects (NSIPs), and introduced a "first ready, first connected" model to replace the previous "first come, first served" queue [11]. Officials estimate these reforms could cut major project "time-to-power" delays by up to five years [11]. A "Connections Accelerator Service" was piloted in late 2025 for projects within AI Growth Zones [10].

But the reforms have not yet produced results at the speed the market demands. Final Gate 2 transmission offers are expected by March 2026, and distribution offers by May or June 2026 [11] — timelines that remain too slow for a company like OpenAI, which is simultaneously breaking ground on sites in Texas and Abu Dhabi.

The Copyright Wrinkle

Energy costs drew the headlines, but regulatory uncertainty over copyright is the second explicit barrier OpenAI cited. The UK government had proposed a text-and-data-mining exception for AI training that would allow companies to use copyrighted works with an opt-out mechanism for creators [12]. The proposal was rejected by a majority of consultation respondents, including creative industries, publishers, and news organizations who opposed compensation-free AI training on their works [4].

In its formal consultation response, OpenAI argued that the opt-out framework "would suffer from technical complexity and require a lengthy multi-stakeholder process," and that "a straightforward copyright regime is needed for AI businesses to have the legal certainty necessary before investing billions of pounds in long-term infrastructure" [12]. Legislation remains delayed, and no resolution is expected before late 2026 at the earliest.

This matters because copyright uncertainty creates a jurisdiction-level risk: if the UK eventually adopts restrictive training rules, infrastructure built there may be constrained in how it can be used, undermining the economic rationale for the investment.

Where the Money Is Going Instead

OpenAI is not short of alternatives. The company's global Stargate programme — originally announced at the White House in January 2025 as a joint venture with SoftBank and Oracle — is proceeding on multiple fronts [13]:

• United States: Five new data centre sites were announced in September 2025 across Texas (Shackelford and Milam counties), New Mexico, Ohio, and a midwestern location. Oracle appointed a new CFO to manage $50 billion in US data centre construction, with $40 billion in SoftBank bridge financing [4][13].
• United Arab Emirates: Stargate UAE, announced in May 2025, is a 1 GW data centre venture in Abu Dhabi spanning 10 square miles — the largest such deployment outside the US. Partners include G42, Oracle, Nvidia, Cisco, and SoftBank, with the first 200 MW cluster going live in 2026 [14][15].
• Norway: Stargate Norway, announced in July 2025, is sited in Narvik and uses abundant hydropower. Initial capacity is 230 MW with plans for 100,000 Nvidia GPUs by end of 2026 [13].

The contrast is instructive. Norway offers cheap, clean hydroelectric power. The UAE offers subsidised energy and fast-tracked permits backed by sovereign wealth fund co-investment through G42. Texas offers low electricity costs, minimal permitting friction, and an established grid infrastructure accustomed to large industrial loads. Each of these jurisdictions has structured its offering around the specific needs of hyperscale AI infrastructure — cheap energy, fast connections, regulatory certainty — in ways the UK has not yet matched.

A Pattern, Not an Outlier

OpenAI's pause is not an isolated case. Data Center Watch reported that as of June 2025, more than 36 data centre projects globally representing $162 billion in investment were either blocked or significantly delayed [16]. In the UK specifically, growth is constrained by power availability, equipment lead times, and local opposition [16].

Hyperscale data centre protests have intensified, with organised opposition documented by The Ecologist in early 2026 [17]. The government itself acknowledged a "serious error" in approving at least one hyperscale data centre without adequate environmental protections [16]. The tension between the government's AI Growth Zone ambitions and local planning concerns remains unresolved.

Meanwhile, the broader pattern of AI infrastructure investment is shifting toward jurisdictions that can offer both scale and speed. Microsoft committed $6.5 billion to cloud and AI infrastructure across Singapore and Thailand [18]. ByteDance is building a massive AI supercluster in Malaysia [19]. The countries winning this investment tend to share common features: affordable and abundant electricity, streamlined permitting, and government co-investment.

The Case for Caution

There is a credible argument that the UK's higher energy costs and regulatory deliberation reflect sound policy rather than failure.

Concentrating hundreds of megawatts of power demand for a single commercial tenant creates real grid stability risks. National Grid's CEO has warned that power used by data centres will increase six-fold by 2035 [20]. Over 140 proposed data centres in the UK pipeline require more aggregate power than Britain's current peak demand [20]. If those loads materialise faster than generation capacity expands, the consequences fall on ratepayers and other grid users.

The carbon accounting is equally fraught. Analysis by Carbon Brief found that CO2 emissions from planned UK data centres could be "hundreds of times higher" than government estimates. DSIT projected maximum emissions of 0.142 MtCO2 from 11.2 GW of AI computing by 2035 — a figure that implies less than 2 grams of CO2 per kilowatt-hour, which Carbon Brief called unrealistic. If even 5% of data centre power comes from gas, actual emissions would reach approximately 2 MtCO2. With heavier gas reliance, the figure could exceed 30 MtCO2 — equivalent to Denmark's total annual emissions [21].

Existing UK data centres already consume 1.8 GW, or roughly 2% of national electricity [21]. The 71 "mature" projects with funding require approximately 20 GW [21]. Approving this capacity without corresponding clean generation investment would put the UK's carbon budgets and 2035 decarbonisation targets at risk.

On copyright, the creative industries' resistance to blanket training exceptions is not obstructionism. Publishers, musicians, and visual artists face real economic displacement from AI models trained on their work. A regulatory framework that takes time to balance these interests may produce more durable outcomes than a hasty exemption designed to attract a single company's investment.

The Sovereignty Question

If the Stargate UK deal collapses entirely, the second-order consequences extend beyond lost jobs and investment.

The UK's ability to host sovereign AI infrastructure — systems that process sensitive government and citizen data on domestic soil — depends on attracting the companies building frontier AI. Without domestic compute capacity at scale, the UK becomes a customer rather than a host, reliant on American hyperscalers operating under the US CLOUD Act, which allows American law enforcement to access data held by US companies regardless of where it is stored [22].

AECOM has called for a sovereign data centre framework to "protect critical digital infrastructure and ensure the UK benefits from the rapid growth of AI" as AI becomes embedded across public services and regulated sectors [22]. The UK government's own sovereign AI fund — backed by £500 million from DSIT — aims to provide domestic alternatives, but its Edinburgh supercomputer is not scheduled to come online until early 2027 [23].

Forrester projects that 2026 will be the year governments take a "tech nationalism" stance in selecting AI suppliers [22]. If the UK cannot host frontier compute, it loses leverage in negotiations over data access, algorithmic transparency, and standards-setting with the very companies it seeks to regulate.

The Gap Between Ambition and Infrastructure

The UK government's AI Opportunities Action Plan, published in January 2025, committed over £1 billion to expand the AI Research Resource twenty-fold — from 21 AI ExaFLOPS to 420 ExaFLOPS by 2030 [23]. Five AI Growth Zones have been designated in Culham, the North East, North Wales, South Wales, and Lanarkshire [23]. Data centre projects worth £14 billion were announced alongside the plan, with companies including Vantage, Nscale, and Kyndryl committing to UK digital infrastructure and 13,250 jobs [24].

These are substantial commitments. But as one CNBC assessment noted in December 2025, "heavily restricted access to energy via the national grid and slow-moving buildouts" suggest the country "is at risk of lagging behind global rivals" [25]. The OpenAI pause is the most visible confirmation of that risk.

The government described the decision as "disappointing" and said it "remained in dialogue with OpenAI" [7]. OpenAI, for its part, maintains its London research hub — described as its largest international facility — and has committed to NHS AI adoption initiatives [5]. The relationship is not severed. But the infrastructure that would have made the UK a physical home for OpenAI's compute is, for now, on hold.

What Comes Next

The Stargate UK pause is a stress test for the UK's industrial strategy. The government has passed legislation, designated growth zones, and pledged billions. But the underlying economics — electricity prices, grid queues, copyright uncertainty — have not yet changed enough to close a deal with the most capital-rich, geographically mobile companies on earth.

Whether the UK can bridge this gap depends on whether the reforms of 2025 produce measurable results in 2026: faster grid connections, lower effective energy costs for strategic projects, and a copyright framework that provides legal certainty without abandoning creators' rights. The competing offers from Abu Dhabi, Texas, and Narvik are not theoretical. They are under construction.