In our view, Wales is already a quiet giant of British compute. On the UK Government’s latest estimate, for autumn 2024, no nation or region in Great Britain outside London had more colocation data centre capacity. Now, a much larger wave of development is arriving, targeted at former heavy industrial land in Bridgend and Newport, and at a hospital park-and-ride site on the eastern edge of Cardiff.
That Cardiff East proposal has drawn 1,602 objections on the council’s planning register, and Welsh ministers have frozen the council’s decision while they consider calling the application in. In our view, the row has become a proxy war over whether Wales wants this infrastructure at all.
Here, we set out the strongest case for why Wales should actively court data centres, separate the hard economic truths from the weak technical excuses, and test the criticisms of the Cardiff East scheme against the official planning documents.
What Wales has, and what is in the pipeline
According to the UK Government’s most recent official estimate (for autumn 2024), Wales holds 154 MW of colocation data centre capacity: more than the South East of England (128 MW) and second only to London’s 1,048 MW, out of a 1.6 GW British total (DSIT, May 2025, background). Because those figures only count colocation sites (where operators rent space to third parties) and exclude bespoke facilities built by tech giants for their own use, the true Welsh total is likely higher.
Electricity use tells a different story. On a broader official measure that includes sites big cloud firms build for themselves, the South East used more electricity for data centres in 2024 than any other region, 1.8 TWh, ahead of London at 1.7 TWh. In 2024, data centres in Slough alone consumed 1.3 TWh of electricity: 29% of all data centre power in Britain, and nearly two-thirds (65%) of Slough’s entire grid demand (DESNZ, June 2026). Across Wales and South West England combined, data centres used roughly 1% of electricity. On that measure, Wales is nowhere near saturation.
Wales is not short of data centres, nor of developers wanting to build them. It is short of the grid connections to power them.
Wales’s flagship site is Vantage’s CWL1 campus, offering 148 MW of critical IT load (Vantage). Although Vantage brands it “Campus Cardiff” and Senedd Research (July 2026) lists it among “two developments in Cardiff”, the site is actually at Celtic Way in Coedkernew, Newport.
What comes next is much larger, and it is hitting a single bottleneck: the grid.
- Bridgend (Vantage CWL4): Planned across 55 hectares of the former Ford engine plant at Waterton. Council officers report it will be roughly four times the size and capacity of the Newport campus, though grid constraints mean only around 20% can become operational before 2032 (Bridgend Council, 2025, background).
- Newport (Microsoft): Senedd Research lists a proposed 200 MW scheme in Newport (a figure we could not confirm from Microsoft).
- South Wales AI Growth Zone: Announced in November 2025. The UK Government says the zone could harness over 1 GW of power by the early 2030s (UK Government, background). That is an ambition rather than a firm commitment.
- Cardiff East (26/00109/HYB): A proposal by developer Curtis Hall for a data centre drawing up to 150 MW of power (120 MW IT load) (water usage datasheet, October 2026). It would replace the Cardiff East park-and-ride, which serves about 1,000 cars for the University Hospital of Wales (BBC, June 2026). Cardiff Council agreed to sell the land to Curtis Hall, subject to planning permission, but on 24 August 2026 Welsh ministers issued a holding direction pausing approval while they decide whether to call the application in (Written Question WQ100388).
Wales is not short of data centres, nor of developers wanting to build them. It is short of the grid connections to power them.
The affirmative case: why Wales should build
Critics often look at a data centre and see a windowless shed with remarkably few workers inside. In our view, that misses what a data centre actually is: heavy foundational infrastructure, closer to a port or a power station than a call centre. The case for welcoming them into Wales rests on four pillars.
1. Massive private capital injection
In our view, data centres mobilise private investment on a scale almost no other modern industry matches. Using Savills industry benchmarks per megawatt, the developer’s consultants estimate the Cardiff East build alone represents £1.11 billion to £1.39 billion in private capital (Environmental Statement, chapter 7, December 2025). Across the wider South Wales AI Growth Zone, the UK Government has pointed to a potential £10 billion investment pipeline supporting at least 5,000 jobs, including construction roles (UK Government, November 2025, background). Even treating announced figures with caution, this is billion-pound private capital.
2. High-productivity, high-GVA employment
In our view, the structural weakness of the Welsh economy has long been less a lack of jobs than a lack of high-productivity, high-wage roles. Once operational, Cardiff East is projected to support 100 to 125 permanent on-site jobs, following 404 to 505 direct construction roles. Crucially, the economic assessment values each operational job at £106,794 in gross value added (GVA) per year, generating £10.7 million to £13.3 million annually from a single site. At the larger CWL4 campus in Bridgend, officers project 600 permanent on-site FTE roles, 350 supply-chain jobs, and over 1,500 construction roles across a 15-year build. These are skilled engineering, maintenance, security, and network operations careers.
Judging a billion-pound data centre by its headcount is like judging a wind farm or a power station by how many people sit in the control room.
3. Permanent municipal revenue
Data centres are high-value commercial properties, so they should pay substantial non-domestic rates. In Wales, properties with a rateable value over £100,000 pay a multiplier of 0.515 in 2026-27 (Business Wales). But neither the developer nor the council has published a business rates estimate for Cardiff East, so we cannot yet say how much it would raise.
4. Capacity to grow without straining the nation
Where Ireland saw data centres swallow 23% of all metered electricity in 2025 (CSO, July 2026) and Slough sits at 65%, Wales and South West England together sit near 1%. In our view, Wales has the headroom to capture the upside of the AI boom without overwhelming its national energy balance.
Our view: Data centres are the physical factories of the AI economy, and Wales should unapologetically want its share of them. Measured crudely by jobs per hectare, they will never match 20th-century manufacturing. The Bridgend CWL4 site once employed around 2,000 people making car engines, compared to 600 planned for the data campus. But those 20th-century assembly plants are not coming back. Judging a billion-pound data centre by its headcount is like judging a wind farm or a power station by how many people sit in the control room. The honest positive case rests on massive capital inflows, high-GVA engineering work, building the foundational compute of the modern economy and, in time, rates.
Why build them in Wales rather than England?
Even if Britain needs data centres, why should they sit in South Wales rather than Swindon or Slough? There are three strong reasons, and two popular myths that proponents should stop using.
The strong reasons
- Capturing the capital and the cluster: When a developer spends £1.1 billion to £1.4 billion building a facility (the Cardiff East developer’s estimate), that capital, those engineering contracts, and those business rates land somewhere. In our view, if Wales turns them away, the compute will still be built. Wales will simply export the investment, the high-wage jobs, and the tax base across the Severn Bridge while still using the cloud services they power.
- Anchoring western renewable energy: Wales already has around 7.9 GW of generating capacity (Senedd Research), with vast floating offshore wind planned for the Celtic Sea. Right now, the Welsh grid is constrained (Senedd Research), which notes that “locating data centres close to large generation sources can improve efficiency by reducing losses during transmission”. Placing large, steady 24/7 industrial demand in South Wales creates a natural local sink for Welsh clean energy in the 2030s. Admittedly, the grid is currently a bottleneck rather than an asset, forcing Bridgend to phase its build and Cardiff East to propose interim gas, and we found no official Welsh grid study that quantifies the transmission savings. Today, the grid is a constraint. In our view, for the 2030s, co-locating Welsh compute with Welsh renewables is a major strategic prize.
- National resilience: In our view, concentrating Britain’s digital backbone tightly around West London and Slough is a strategic vulnerability. Dispersing capacity along the M4 corridor into South Wales protects the UK against regional grid faults, extreme weather, or physical disruption.
The real prize is anchoring Welsh renewable energy to high-value industrial demand, and capturing the capital that comes with it.
The weak reasons: speed and law
- The latency myth: It is sometimes claimed that Wales needs local servers for faster digital services. The physics say otherwise. Microsoft measures round-trip network latency between its Cardiff region (UK West) and London (UK South) at just 7 to 8 milliseconds, compared to 14 ms to the Netherlands, 16 ms to Ireland, and 80 ms to the US East Coast (Microsoft, August 2026). Moreover, Senedd Research notes that training large AI models is not latency-sensitive. For NHS records, public services, and everyday AI, an 8-millisecond hop to London is invisible to a human user. Latency is an argument for hosting in Britain rather than Virginia, not Cardiff over London.
- The data sovereignty myth: We found nothing in law that requires Welsh data to sit on Welsh soil. UK GDPR permits transfers across the EEA and beyond with safeguards (ICO), and all countries in the EEA count as adequate (ICO). NHS England’s guidance sets the direction that health data at rest remains within the UK, with exceptions, and it does not apply in Wales (NHS England Digital). The UK Government’s Cloud First policy allows overseas cloud hosting with due diligence. In fact, Digital Health and Care Wales aims to “move all our live services to the cloud and close our datacentres” by 2030 (DHCW performance report, July 2026), although its internal auditors found the move “materially behind” plan (DHCW internal audit, October 2026). And we found no rule that requires Welsh-language services to be hosted in Wales.
Our view: Supporters of Welsh compute do their cause no favours by relying on flimsy claims about milliseconds or data laws. The real prize is anchoring Welsh renewable energy to high-value industrial demand, and capturing the capital that comes with it. Wales needs data centres to capture billion-pound capital, to anchor the clean energy grid of the 2030s, and to diversify Britain’s digital backbone. That economic case is strong enough to stand on its own.
What the critics say about Cardiff East
While, in our view, the strategic case for Welsh data centres is strong, the specific row over Cardiff East has been fierce. By 3 November, the council’s planning register listed 1,602 objections against just two letters of support (Cardiff Council planning register). Here are the six main charges levelled by critics:
- Gas power and climate: Friends of the Earth Cymru’s interim director, Jenny Lloyd, called the plan a “climate-wrecking piece of infrastructure” (BBC, 16 October 2026). Caerdydd Ffynnon Taf MS Nick Carter said “underneath the headlines is a small gas-fired power station”, while Llanrumney ward councillor and cabinet member Lee Bridgeman wrote that the emissions were “hard to reconcile” with Cardiff’s climate emergency and Planning Policy Wales (BBC).
- 24/7 generators and household bills: A campaign page by the community union Acorn, used to email councillors, claims the facility will “run gas and diesel generators 24/7” and “potentially increase local resident utility bills” (Acorn).
- City-dwarfing electricity scale: The Guardian (24 October 2026) reported that the site “will use up to 1,314,000 megawatt hours (MWh) of electricity a year, more than triple the demand of a city the size of Cardiff”, adding that five proposed Welsh data centres would use more electricity than all Welsh homes combined.
- Water strain: Acorn’s campaign page warns of “severe strain” on local energy, grid and water infrastructure, while three of the area’s Members of the Senedd, in a joint formal objection (objection, 26 May 2026), noted that data centres are “among the most energy- and water-intensive land uses in existence”.
- Hospital parking, noise, and low job density: The MSs’ objection cites the “premature removal of essential transport infrastructure serving Wales’ largest hospital, without a confirmed replacement” and “unacceptable noise impact from 24/7 industrial operations”. One midwife told the Guardian she probably wouldn’t continue working at the hospital without the park-and-ride. Acorn adds that data centres “have no place in residential areas” and offer a “lack of employment opportunities”.
- Process and transparency: Critics note Cardiff Council is both the landowner selling the site and the planning authority judging it. Nick Carter told the Guardian that developer Curtis Hall has no track record in data centres and “could be a front for someone else: Google, Microsoft, whoever” (Curtis Hall told the Guardian it never claimed it would operate the facility itself, and that it had been working on data centre projects for five years). Meanwhile, Questions Before Concrete, a campaign website whose author has sent the council detailed representations criticising the application, says, as its own inference, that it has “found no published assessment of whether the very large investment figures announced for data centres include the cost of the grid reinforcement” (Questions Before Concrete).
Fact-check: testing the Cardiff East criticisms
We audited each claim against the technical planning filings and official energy statistics.
Where the critics are wrong or overstate the case
- “More than triple” Cardiff’s electricity demand: NOT SUPPORTED. The Guardian’s 1,314,000 MWh figure is simply 150 MW multiplied by every hour in a year (and real data centres rarely draw 100% of their peak supply 24/7). Even at that theoretical maximum, official DESNZ figures (December 2025, background) show Cardiff consumed 1,384.8 GWh of grid electricity in 2024, of which households used 489.9 GWh. Flat out, the data centre would equal 95% of Cardiff’s total electricity use, or 2.7 times its domestic use, not triple the city’s demand.
- Higher local utility bills: UNSUPPORTED. We found nothing in how retail energy prices are set to suggest a new data centre in Llanrumney raises local household bills, which are set nationally by suppliers and regulated by Ofgem (developer newsletter). The valid policy question is national: how Ofgem apportions the cost of wider grid upgrades between developers and consumers.
- “Run gas and diesel generators 24/7”: PARTLY FAIR. The diesel units are standby backup generators, tested weekly and semi-annually, designed to run for up to 72 hours only if the main grid fails (utility services report). Only the interim gas turbines would run continuously, and only until the grid connection arrives.
Unproven so far
- Air quality and noise breaches: UNPROVEN. Under normal operation, the developer’s modelling puts annual nitrogen dioxide at the most affected home at 19.62 µg/m³, less than half (49%) of the legal objective. If the grid failed completely for 72 hours and all 40 standby generators ran, the developer’s consultants predict that hourly nitrogen dioxide would rise above 200 µg/m³ at every modelled location, and that seven homes nearby could breach the legal hourly objective. They put the chance of that breach at about 0.1% in any year (generator air quality note, 8 September 2026). Routine six-monthly tests could take some homes and a primary school above 200 µg/m³ for up to four hours a year, within the 18 hours the objective allows. We have not yet seen the council’s own air quality view. On noise, the developer’s latest noise note predicts plant noise at night at or below background at the nearest homes, by 0 to 9 decibels, and proposes a limit of 35 decibels outside any home, with a stricter test for tonal and low-frequency noise (noise technical note, September 2026). The developer’s newsletter says daytime levels would be 12 to 25 dB below background. The council has yet to decide whether to impose those limits.
- Five Welsh schemes using more power than all Welsh homes: PLAUSIBLE, BUT YEARS AWAY. All Welsh homes used 4,553.9 GWh in 2024 (DESNZ), an average draw of roughly 520 MW. If Bridgend CWL4 (roughly four times the 148 MW Newport site), Newport’s existing and proposed sites, and Cardiff East are all built out and run near full load, they would exceed 520 MW. However, grid limits mean only a fifth of Bridgend can run before 2032.
The critics are right about the local trade-offs and wrong about the scale.
Where the critics are on solid ground
- Loss of the hospital park-and-ride: FAIR. While the developer promises the car park “will stay open until an alternative car park is operational” (developer newsletter, October 2026) and the health board expects it to remain open until the end of 2026 (BBC), neither the council nor the developer has named a replacement site.
- Years of on-site gas power: FAIR, WITH LIMITS. Because the National Grid connection is not due until between 2030 and 2034 (BBC; developer newsletter), the plan includes roughly 49.9 MW of gas turbines (utility services report, November 2025). At 49.9 MW, the turbines can only power about a third of the 150 MW site, and the developer calls five years of sole reliance a “worst-case scenario” (noting fuel cells could be used instead). Even so, building an interim 49.9 MW gas plant because the grid isn’t ready is a real trade-off.
- Water and operator uncertainty: PARTLY FAIR. The developer has changed the design from evaporative cooling (chapter 3) to a closed-loop system that rejects heat to the air and typically needs only “tens of litres a week at most” in top-ups (water usage datasheet). Yet the application still requests a peak allowance of up to 300 cubic metres a day, and admits the final cooling system is “still to be confirmed” once an end operator is found. In our view, until a planning condition legally binds that operator to closed-loop cooling, the loophole remains open.
- Low headcount and woodland trade-offs: PARTLY FAIR. Against £1 billion-plus of capital, 100 to 125 on-site jobs is a small local headcount, and the developer’s own report grades construction employment as only a “minor beneficial” impact (chapter 7). Claims of “thousands of jobs”, which the Guardian reports were made by the previous Welsh Government, are unproven. Some local woodland will also be felled, though 95% of the land already has planning permission for industrial, retail, or logistics use, and the developer pledges about 8,500 new trees under a 25-year management plan.
- Council’s dual role and policy vacuum: FAIR. Cardiff Council chose Curtis Hall as preferred bidder for the site in 2020 (Environmental Statement summary) and is also the planning authority. That dual role is lawful and not unusual, and we have seen no evidence of wrongdoing, but it is a reason for close scrutiny. Wales also currently has no national planning policy on data centres (the planning minister says one will arrive before the end of 2026) (BBC). In our view, the ministers’ decision to pause the application is a sensible check on both issues.
In short: the critics are right about the local trade-offs and wrong about the scale. On the hospital car park, the interim gas plant, and the planning process, the documents back their concerns; on household bills and city-scale power comparisons, the numbers do not.
Our view: the right infrastructure, on the right terms
Wales should actively welcome data centres. They are the core infrastructure of a productive modern economy, no less than ports, railways, and power stations. As the Welsh planning minister put it: “These data centres are needed to sustain a modern economy”, provided they are built “on our terms” (BBC).
The question is not whether Wales should build the physical foundations of the AI economy, but where, and on what terms.
Where the case is effortlessly strong is on large, former heavy-industrial sites with planned grid connections, like the former Ford plant at Waterton in Bridgend and the industrial corridors of Newport, even though grid limits mean only about a fifth of the Bridgend campus can run before 2032. Where the case is hardest to defend is at a suburban hospital park-and-ride that removes a vital commuter link before a replacement exists, and burns gas for up to five years because the grid isn’t ready.
Yet “on our terms” is not an empty slogan; it is a checklist that Welsh ministers can write directly into planning conditions for Cardiff East and the national policy that follows:
- Lock in closed-loop cooling: Legally bind the eventual operator to the closed-loop cooling design and impose a strict daily water cap.
- Cap the gas hours: Set a hard ceiling and sunset date on the interim 49.9 MW gas turbines so temporary bridge power cannot become a permanent fossil-fuel plant.
- No car park closure without a live replacement: Stipulate that the existing park-and-ride cannot close until the replacement facility for University Hospital of Wales is built and open.
- Turn waste heat into a real plan: The developer has offered the site’s waste heat “free of charge to the Council for district heating”, yet we found no scheme to pipe or use it. Require a funded heat-offtake feasibility plan before construction begins.
- Mandate annual transparency: Require every large Welsh data centre to publish its actual annual electricity and water consumption.
Welsh ministers now have the leverage to set those ground rules. The question is not whether Wales should build the physical foundations of the AI economy, but where, and on what terms.