Context

Power and the grid

Data centres take large, steady electricity loads. Where the local grid is already tight, new connections for housing and other uses can wait years.

UK. In 2022 the Greater London Authority was alerted that parts of West London (Ealing, Hillingdon and Hounslow) faced electricity capacity constraints. Investigations linked the queue partly to rapid data-centre growth along the M4 corridor; some applicants were quoted connection dates into the 2030s (including figures up to about 2035–2037 in GLA materials). There was no formal ban on housing. Short-term fixes and connection reforms have since unlocked many homes, while transmission upgrades remain necessary for the long term. (GLA West London constraints; GLA Planning & Regeneration Committee, Dec 2025.)

US. Northern Virginia’s “Data Center Alley” is the world’s densest cluster. Loudoun County supervisor Mike Turner, citing Dominion Energy figures, puts the county’s data-centre power consumption at about 5.33 GW in 2025, up from 4.14 GW in 2024 (his earlier years are read off a Dominion chart). The presentation is his own and states it does not represent the county. Dominion Energy has reported tens of gigawatts of data-centre capacity under contract across its Virginia service area. Residents and regulators have argued over how much of rising bills and grid spend is driven by that load; Dominion disputes a simple causal story, and a new large-load rate class (GS-5) is due from 2027 to shift more cost onto big users. (Turner, “Data Center Growth”, July 2026; DCD on Dominion contract pipeline, 2025; Loudoun Now on GS-5, 2025.)

London on this map. The west corridor (Slough–Heathrow) still holds most operating megawatts; large proposed schemes push north and east of the M25. See pipeline.

Mitigations on record here: grid reinforcement and flexible connection products (the GLA/SSEN work above), and Manor Farm’s consented ~100 MW battery energy storage alongside the data centre.

Water

Cooling can use large volumes of potable water. UK operators are not required to disclose water use, and data centres are excluded from the Environment Agency’s National Framework for Water Resources, so most dossiers here show water as not disclosed, not as zero. Full method note: methodology § water.

UK. An April 2026 Global Action Plan analysis reported that about 84% of proposed UK data centres sit in areas already water-stressed or projected to be by 2040 (Water Magazine). Water UK has told Parliament that some proposed centres in the Affinity Water region have requested up to about three million litres a day, and that the sector’s demand is not built into national water planning (Guardian / Water UK, Jul 2026).

US. In The Dalles, Oregon, a public-records fight over Google’s data-centre water use ended in a 2022 settlement that forced disclosure; later reporting showed Google taking a large share of city water as campuses expanded (Oregonian, Dec 2022; Oregon Capital Chronicle, Jan 2026). Drought-constrained states such as Arizona show the same conflict pathway when cooling competes with other users. Treat these as precedents for politics and disclosure, not as identical hydrology to the Thames basin.

Mitigations on record here: operator claims of air or closed-loop cooling with low on-site water at Ada Docklands, Google Waltham Cross, and Vantage, all flagged as claims until metered. Ada and Google also state heat-reuse or district-heat provisions, which we show as claims. “Waterless” on site still shifts water to the power station through higher electricity use (methodology).

Jobs and the local economy

Promoters emphasise capital investment, business rates, and construction employment. Permanent on-site staffing is usually small relative to the land and power take.

Example on this map. Manor Farm’s economic statement and inquiry materials put construction at about 430–490 full-time-equivalent jobs over the build, and operations at about 65 FTE once running (Economic Statement; BBC). Those are applicant figures in the consent trail, not independent audits.

Benefit side. The same facilities underpin cloud and AI services used across the wider economy. That benefit is real. It is also spread thin: the rates and construction jobs stay local, but most of the digital gain is used nationally. We do not score “net public value” on this site.

Noise and health

Continuous cooling plant and periodic diesel (or other) generator testing are among the impacts residents cite most often in objections. Emissions from testing add an air-quality concern beside noise.

US. Suburban complaints in Northern Virginia and places such as Chandler, Arizona, have turned on overnight noise and generator exercises. The pattern is useful: once campuses sit beside housing, acoustic conditions become a planning lever.

Mitigations: planning conditions on plant enclosure, hours for generator testing, acoustic barriers, and (where proposed) HVO or electric backup. Check each dossier’s planning documents rather than assuming a London-wide standard.

Land

London’s build-out mixes brownfield industrial reuse (Docklands, Trading Estate halls) with green-belt and farmland schemes on the fringe: Manor Farm, Equinix South Mimms, and Digital Reef East Havering. Brownfield-first is the usual mitigation in policy language; green-belt loss is decided case by case, often at appeal or by the Secretary of State.

The land question. Land is finite, and each consent decides what a parcel serves once it changes hands. A scheme can convert green belt or working farmland into a fenced, single-use facility that takes large power and water while employing few people locally: Manor Farm’s own figures put operations at about 65 FTE (jobs). On any planning horizon that change is hard to reverse; a hall of racks and cooling plant does not go back to being a field.

Rival claims on scarce capacity. The pressure is not only about the ground. Land, grid headroom, and water are shared resources, and a data centre’s claim on each one competes with other uses. In West London, data-centre grid demand sat in the same constrained queue that delayed new housing connections (power). Green belt does work while it stays open, checking sprawl and holding land in reserve, and a consent spends that.

Who carries the cost. The local costs concentrate: the land itself, the noise, the traffic, and sometimes water use fall on the immediate area. Most of the benefit is national: the cloud and AI capacity these sites host is used across the economy, not mainly nearby (jobs). Opponents call this a land-justice problem.

Benefits. Clustering compute near existing hubs and fibre can lower running costs, and building on hard-to-redevelop industrial land puts derelict sites back to work. On this map the brownfield conversions (Docklands, the Trading Estate halls) are where that case is strongest: industrial reuse that raises no green-belt question at all.

Whether a given scheme is a good use of its land isn’t this site’s call: each dossier just logs the prior use of the plot and who signed off.

How London compares with the US

Single US AI and cloud campuses are often quoted in the multi-gigawatt range, and Loudoun County’s data-centre consumption alone was about 5.33 GW in 2025 (Turner, citing Dominion). London’s entire mapped build-out, running plus proposed, is about 3.0 GW stated; its largest single scheme is about 600 MW (Digital Reef East Havering). London is smaller per site and far more constrained on land, water, and grid headroom, so friction such as connection delays, water stress, and green-belt fights shows up at lower absolute megawatts. Some of it is already here.

Metric caution: both sides are cluster-scale, but they are not the same measurement. London’s figure is the stated total of every mapped site, running and proposed, and mixes IT load with a few whole-site figures. Loudoun’s is power consumed across the county cluster in 2025, as reported by Supervisor Turner from Dominion data. Read it as scale, not a like-for-like league table.