The capacity being built, what it delivers, and the sourced footprint of constructing and running it — CO₂e, water and hardware lifecycle.
There is no single official figure. The government does not routinely collect capacity data; its 1.6 GW estimate covers co-location sites only, which is why IT-power methods come out higher (Oxford Economics: 2.9 GW; NESO counts 2.4 GW of connected data-centre demand using 7.6 TWh a year). The government's own preliminary range for 2030 is 3.3–6.3 GW. This page uses the Computer Weekly / Barbour ABI project data for the pipeline so the charts are internally consistent. Summary of all of these in the House of Commons Library, CBP-10315 (27 May 2026).
Cumulative installed capacity (area) versus the full announced pipeline (line). The government's 6 GW AI-capable target for 2030 sits above the planning-approved trajectory of ~4.9 GW — the gap is the build that is announced but not yet consented. Figures are approximate, drawn from the sources below.
Capacity & factors (editable)
Operational capacity is heavily concentrated. London and the M4 corridor hold over half; the next planned giants shift weight toward the Humber, North East and Scotland.
Largest single planned builds announced
A full geographic heatmap (production vs consumption by postcode/region) is a sensible next step, but reliable per-site consumption data is scarce — operators self-report inconsistently. This cluster view uses the regional figures that are actually published.
NESO's data-centre demand estimate against the theoretical ceiling of the operational fleet (MW × 8,760 h). The demand line sits at an implied fleet-average load factor of roughly 0.4–0.55 — shown as the shaded band. Per-site utilisation is not public anywhere and is not synthesised here. The system-level tension: about 73 GW of data-centre projects (around 315 of them) sit in the connection queue, against 4.9–6.2 GW likely to be built by 2030 and a national peak demand of 45 GW. Ofgem is consulting on fees to clear speculative projects out of the queue.
Fleet emissions can be counted two ways: location-based (grid average × consumption) or market-based (netting renewable certificates & PPAs). Vendors lead with the second; the difference is what the certificates and contracts net off. The GHG Protocol allows it, but a certificate match is not the same as zero-carbon power at the time it was used. This site stays location-based throughout — handbook §4 explains why.
Scope 2 only, Mt CO₂e, latest reporting year: Microsoft FY2025 (to June 2025) and Google calendar 2025. Both figures are the companies' own, from their 2026 reports. Microsoft's market-based figure rose tenfold in the year after it stopped using unbundled renewable certificates. Context: scope 3 was 86% of Microsoft's FY2025 footprint, and Google's electricity use grew 37% in 2025.
| Figure | Value | Source |
|---|---|---|
| UK capacity now | 1.6 GW co-location, Great Britain, 2024 (government estimate) | House of Commons Library, CBP-10315 (27 May 2026), citing DSIT |
| UK pipeline | 4.9 GW consented for 2030 · 6.2 GW incl. unconsented · 8.1 GW announced to 2037 · govt preliminary range 3.3–6.3 GW by 2030 | Computer Weekly / Barbour ABI (Apr 2026) · House of Commons Library, CBP-10315 (27 May 2026) |
| 6 GW AI-capable target | "at least 6GW of AI-capable data centre capacity by 2030" | DSIT / UKRI, UK Compute Roadmap (Jul 2025) |
| Number of data centres | 266 sites (2024) | Oxford Economics (2026) · IT-power capacity estimate 2.9 GW |
| Construction embodied carbon | ~200 tCO₂e/MW (10 MW ≈ 1–2 kt concrete CO₂) | Texas DC LCA, arXiv (2025) · ATLAS (200–800 kg/m²) |
| Hardware embodied (per MW IT) | 1,100 tCO₂e/MW · range 750–1,500 · AI-dense halls likely >1,500 | ADW / CIBSE TM65 (2025) |
| Server embodied carbon (alt. preset) | ~1.8–3.3 tCO₂e/server × servers/MW | Google, Life-Cycle Emissions of AI Hardware (2025) — note AI-accelerator servers draw >1 kW, so AI halls run ~100–200 servers/MW, not 1,000 |
| Hardware refresh cycle | 3–5 years | Sustainable LLM Serving (2025) |
| Water (WUE) | UK ≈0.5 · EU ≈0.6 · global evap ≈1.9 L/kWh · 1 MW evap ≈ 26 ML/yr | WRc for MOSL (2026) · EESI / Green Grid · Univ. of Manchester (2025) · UK Gov water report (2025) |
| What the water figures exclude | off-site water · seasonality | On-site WUE only: water consumed generating the electricity (thermal-plant cooling) typically exceeds on-site cooling water severalfold. And the UK concern is peak coincidence — hybrid-cooled sites draw water during heatwaves exactly when domestic demand peaks (WRc/MOSL) — which annual averages hide. |
| Grid connection queue | ~73 GW of data centres (~315 projects, 1–1,500 MW each) within 125 GW of contracted demand · GB peak demand 45 GW (2025) · earlier NESO count: ~140 projects, 50 GW, only ~20 GW at final investment decision | Ofgem, Curate consultation (29 Jul 2026) · earlier figures via the House of Commons Library, CBP-10315 |
| GB data-centre demand (NESO) | Now: 2.4 GW connected, 7.6 TWh/yr · 2030: 20.1 TWh (ten-year forecast; pathways 13.2–24.7) and ~5.2 GW connected · 2035: 33 TWh (20–41) · 2050: 30–71 TWh, 6.1–14.6 GW | NESO, Future Energy Scenarios 2025 (report and data workbook, fig. 49) · 2030 capacity from NESO evidence to the Environmental Audit Committee (2026) |
| Scope-2 gap (location vs market) | Microsoft FY2025: 12.03 vs 2.71 Mt · Google 2025: 15.15 vs 2.82 Mt | Microsoft 2026 Environmental Data Fact Sheet · Google 2026 Environmental Report (both vendors' own figures) |
| UK site map | ~23 curated sites (of ~190 EPC-identified) | GOV.UK AI Growth Zones · Nscale/Microsoft/OpenAI (2025) · operator pages · press-cited planning applications — per-marker source & confidence in each tooltip |
| World map — capacity shares | ≈55 GW global IT (mid-2025) · US ≈50% · Europe ≈18% · China ≈10% · US 180 TWh / China 102 TWh (2024) | IEA, Energy and AI (2025) |
| World map — grid-CI choropleth | display only; UK/US/France anchors match gridFactor | Ember Yearly Electricity Data (2024) — non-anchor countries VERIFY |
All factors are editable and represent academic/industry approximations — the literature spans wide ranges (server embodied carbon alone varies ~40× across published LCAs), so treat outputs as order-of-magnitude. The CO₂e/water bases here are read from the same shared factors as the AI Impact Calculator (one gridFactor, one WUE set) so totals reconcile across the site. Runs in your browser.