Methodology: a modelled estimate, not a meter

What this site does

The Cloud Kettle Index estimates GB data-centre electricity demand and expresses it as kettle-boils per second. GB electricity demand falls by roughly 15–20% on bank holidays and quiet days. Data-centre demand is generally much flatter - so when national demand dips, data centres represent a larger share of the grid.

The figures are modelled, not metered. The central estimate comes from NESO’s 2025 national figure; the scenario range from DSIT’s published capacity data. Both sources are cited throughout.

The kettle-boil unit

One kettle-boil is defined as 0.1 kWh - the energy a 3 kW electric kettle uses to boil approximately one litre of water from 15°C to 100°C.

One kettle-boil per second is a rate: it means 0.1 kWh of energy is consumed every second. Continuous power = 0.1 kWh × 3,600 s/hr = 360 kW = 0.36 MW. Therefore:

kettle-boils per second = facility load (MW) ÷ 0.36

Note: this is an energy-rate metric, not a count of simultaneously boiling kettles. "2,000 kettle-boils per second" means the data centres are consuming enough energy every second to boil 2,000 one-litre kettles - not that 2,000 kettles are boiling simultaneously. That would be a different, and much larger, number.

The dual-baseline model

We cannot directly measure live data-centre electricity demand. The model combines two independent public sources: a national demand estimate from the grid operator (NESO), and a capacity-based scenario range from DSIT.

Central estimate - NESO FES 2025

NESO estimate: 7.6 TWh/year from 2.4 GW connected GB data-centre facilities
Average continuous power: 7.6 TWh ÷ 8,760 h/year = ~868 MW
Kettle-boils/sec: 868 MW ÷ 0.36 = ~2,410 kbs

Estimated share of GB demand
  = ~868 MW ÷ live Elexon BMRS demand (MW)

Source: NESO Future Energy Scenarios 2025 (July 2025), confirmed in UK Parliament POSTnote 762 (March 2026). The 7.6 TWh figure represents total facility electricity consumption (grid draw), not IT load alone. It covers all connected data-centre facilities in GB, including enterprise and colocation.

Scenario range - DSIT capacity model

Facility load (MW) = DSIT IT capacity × capacity-utilisation × PUE

  DSIT GB colocation IT capacity: 1,600 MW (maximum rated, autumn 2024)
  Capacity-utilisation: 40% (low) · 55% (central) · 70% (high)   - assumptions
  PUE:                  1.2 (low) · 1.4 (central) · 1.6 (high)   - assumptions

  Low:  1,600 × 0.40 × 1.2 =   768 MW  →  2,130 kbs
  High: 1,600 × 0.70 × 1.6 = 1,792 MW  →  4,980 kbs

The NESO central estimate (~868 MW) sits within this range.

Why two sources? NESO’s figure starts from measured electricity demand and covers all GB data centres. DSIT’s capacity data covers colocation facilities only and requires utilisation and PUE assumptions. Together they bracket a plausible range. The NESO figure is the more directly relevant demand estimate; the DSIT scenarios provide uncertainty bounds.

Intraday shaping - UKPN measured profiles

The NESO 7.6 TWh figure is an annual average. To reflect how data-centre load varies through the day, the NESO central estimate is scaled by a relative utilisation multiplier derived from UKPN’s Data Centre Demand Profiles dataset - half-hourly apparent-power readings from 96 anonymised data-centre sites (78 colocation, 18 enterprise) in the UKPN licence area (London, South East, East of England), January 2023–April 2026. The dataset covers 28 bank holiday days in that period. Readings are utilisation ratios, not absolute megawatts.

The multiplier is looked up by day type (weekday, Saturday, Sunday, or bank holiday) and half-hour slot. Timestamps are parsed as UTC and converted to Europe/London before slot assignment, ensuring correct handling of British Summer Time. Each site’s shape index is normalised to its own overall mean across all data, preserving genuine day-type differences: weekdays run ~1.6% above the overall baseline, while bank holidays run ~3.6% below. Intraday peak-to-trough variation is ~5.8% on weekdays and ~4.0% on bank holidays. The DSIT scenario band is intentionally left flat - it represents structural uncertainty in capacity and utilisation, not intraday variation.

UKPN profiles cover UKPN licence areas only and may not be nationally representative. The shaping affects the NESO central estimate only; the DSIT scenario bounds remain unaffected. The raw UKPN parquet data is not published here; only the derived aggregate profiles are committed to the repository.

Important: The NESO 7.6 TWh is an annual average, not a live reading. The actual load at any given moment will vary. DSIT’s 1.6 GW is maximum rated IT load for colocation data centres as of autumn 2024 - not total facility draw, not 2026 capacity, and not all data-centre types. Neither figure is a live meter.

Data sources

NESO Future Energy Scenarios 2025

Published July 2025 by the National Energy System Operator. Estimates current GB data-centre electricity consumption at 7.6 TWh/year from 2.4 GW of connected facilities. This figure covers all connected data-centre facility types (colocation, enterprise, hyperscaler). It is an annual estimate, not a half-hourly measurement. Repeated verbatim in UK Parliament POSTnote 762 (March 2026).

UK Parliament POSTnote 762 - Data Centres (March 2026)

Parliamentary Office of Science and Technology briefing confirming the NESO figure: “All grid-connected data centre facilities in the UK consumed an estimated 7.6 TWh annually” and “data centres use around 2% of the UK’s electricity.”

DSIT Estimate of Data Centre Capacity Great Britain 2024

Department for Science, Innovation and Technology publication (GOV.UK, published 2025). Provides maximum rated IT load of colocation data centres at national and NUTS1/ITL1 regional level. Data collected from public sources, industry engagement, and modelling. Covers colocation facilities; does not include enterprise on-premises data centres and has limited coverage of hyperscaler (cloud provider) in-house infrastructure. Regional figures are published in aggregated form only; underlying site-level data is commercially sensitive. The national headline figure (1.6 GW) and regional table figures are committed as static JSON in this repository and verified against the live GOV.UK publication.

Elexon BMRS Initial National Demand Outturn

Public API from Elexon, the electricity balancing and settlement code manager for GB. Provides half-hourly GB electricity demand in MW, updated within approximately 15 minutes of the end of each settlement period. No authentication required. Used for the live "share of GB demand" figure and the demand chart. We fetch this directly in the browser; the displayed timestamp reflects the settlement period start time.

Carbon Intensity API

Public API jointly operated by National Energy System Operator (NESO) and Oxford University. Provides national and regional carbon intensity (gCO₂/kWh) at half-hourly resolution. Regional carbon intensity is shown as context alongside postcode lookups; it does not affect the modelled data-centre load figure.

postcodes.io

Public API mapping UK postcodes to administrative geographies including ITL1/NUTS1 region codes. Used to map a user-entered postcode to its DSIT ITL1 region for the regional capacity lookup.

Geographic scope: Great Britain, not the United Kingdom

Elexon and the Carbon Intensity API cover Great Britain (England, Scotland, Wales). Northern Ireland operates a separate electricity system under the Single Electricity Market, managed by SONI (System Operator for Northern Ireland), and is not included in any figure on this site. DSIT's dataset similarly covers Great Britain.

Postcode lookup

The postcode lookup uses postcodes.io to derive an ITL1/NUTS1 region code, then applies the same scenario model using DSIT's regional IT capacity figures. Regional carbon intensity is fetched separately from the Carbon Intensity API.

The Carbon Intensity API uses Distribution Network Operator (DNO) regions; DSIT uses ITL1 statistical regions. These geographies do not perfectly align. The regional capacity and carbon intensity figures are drawn from different geographic boundaries and should not be treated as precisely co-located measurements.

Postcodes in Northern Ireland, the Channel Islands, and the Isle of Man return an explicit "not supported" message. The site does not silently degrade.

Limitations and known assumptions

• The model assumes DSIT's autumn 2024 colocation capacity figure is broadly representative of current capacity. Actual capacity may have changed.
• Utilisation and PUE are scenario assumptions applied uniformly. In reality, different facilities have different utilisation rates and PUE values, which vary by operator, age, and season.
• The model does not capture intra-day variation in data-centre load. Evidence suggests data centres are predominantly baseload - relatively flat - but this is assumed, not measured.
• DSIT's dataset covers colocation data centres and does not include enterprise on-premises facilities or provide comprehensive coverage of hyperscaler (cloud provider) owned infrastructure. Total data-centre electricity demand in GB is therefore likely higher than the figures shown.
• The regional model uses ITL1-level DSIT figures, not local-authority-level data. Site-level or local-authority-level data is not publicly available without registration.

What this site does not claim

• It does not claim to show live metered data-centre electricity demand.
• It does not claim to measure AI-specific electricity use. Data centres power AI, cloud storage, streaming, banking, SaaS, and all internet infrastructure.
• It does not identify nearest data centres from a postcode.
• It does not use UKPN profiles to estimate absolute GB data-centre MW or live data-centre demand. UKPN measured profiles are used only for relative load shape (intraday and day-type variation).
• It does not claim regional carbon intensity and DSIT capacity regions are the same geography.
• It does not claim DSIT's 1.6 GW is a current 2026 live figure.
• It does not include Northern Ireland.

DC + Mid-Atlantic page methodology

The DC + Mid-Atlantic page uses a different model from the GB page. The core data-centre load estimate is not constructed from IT capacity assumptions; it comes directly from Dominion Energy Virginia’s own published billing-demand forecasts.

Sources

Dominion Energy Virginia 2025 data-centre forecasting presentation (Indiana Utility Regulatory Commission filing, 2025) and PJM Dominion data-centre large load request (PJM Load Analysis Subcommittee, September 2025). Both provide year-by-year data-centre billing demand for Dominion’s Virginia service territory: 2024 actual (~3,584 MW), 2025 forecast (~4,149 MW), and annual projections through 2030 (~6,992 MW). The 2026 projection is used as the primary DC + Mid-Atlantic page figure. Dominion also states an industry load factor of approximately 90%.

Why UTILISATION × PUE is not applied

The GB model applies capacity-utilisation and Power Usage Effectiveness (PUE) assumptions to DSIT’s IT capacity figure to estimate total facility electricity demand. That step is necessary because DSIT reports IT load, not facility draw. Dominion’s billing-demand figures are already power demand - the grid draw in MW. Applying utilisation and PUE again would double-count. The conversion used is simply:

kbs = billing_demand_mw / 0.36

Billing demand vs average continuous load

Dominion’s billing demand figure represents peak or contracted demand used for utility tariff calculations, not average energy use or coincident grid peak. Dominion separately derives coincident demand - the actual grid draw at the moment of system peak - from its billing-demand forecasts; coincident demand is generally somewhat lower. Dominion reports an industry load factor of approximately 90%, implying that data centres draw close to their billing demand on average, but the figures are not identical. The headline kettle-boils/sec figure on the DC page translates billing demand directly as a power-rate comparison, not an annual-average energy estimate. Average continuous draw can be estimated as billing demand × 0.90, yielding approximately 4,280 MW (~11,900 kbs) for the 2026 projection.

Why Northern Virginia / Dominion zone for a DC-area estimate

I could not identify a public data-centre load figure for the District of Columbia itself. Northern Virginia - primarily Loudoun, Fairfax, and Prince William counties - is reported by JLARC (2024) as representing roughly 13% of global data-centre operational capacity and 25% of Americas capacity. It sits within Dominion Energy Virginia’s service territory and powers much of the cloud infrastructure used in and around Washington, DC. The Dominion service-territory billing demand is the most concrete and well-sourced proxy used here for a DC-area cloud-load estimate. Note: the scientific scope of this page is the Dominion Virginia service territory, not Northern Virginia counties alone; “DC + Mid-Atlantic” is reader framing reflecting where this load is most relevant.

What the DC + Mid-Atlantic estimate is not

• A measurement of data centres physically inside Washington, DC.
• Total PJM Interconnection data-centre load (PJM covers 13 states and the District of Columbia).
• A full Northeast model - PJM, NYISO, and ISO-NE would each require separate treatment.
• A live grid demand share - this page shows modelled billing-demand estimates only.

Intraday load shaping

The DC + Mid-Atlantic page includes a shape-only intraday comparison. I could not identify a public data-centre-specific half-hourly load profile analogous to UKPN’s dataset for the Northern Virginia region. The chart therefore keeps two things separate: a conservative synthetic data-centre daily shape anchored to Dominion’s approximately 90% industry load-factor context, and generic Dominion Energy Virginia customer-class hourly profiles for selected large-customer classes (GS3, GS4, LGEMLP, and MS). Those Dominion profiles are real utility settlement/load-research profiles, but they are not data-centre-specific and are used only as context. COMM6VA is omitted because Dominion publishes it as a weather-sensitive formula table rather than a simple hourly profile.

Latest-reported PJM grid demand comparison

The GB page shows a live “share of GB demand” metric using Elexon’s public half-hourly demand API. The DC + Mid-Atlantic page includes the closest static-site equivalent: a latest-reported PJM demand comparison using EIA Balancing Authority Areas hourly operating data. The numerator is still modelled Dominion data-centre billing demand; the denominator is PJM balancing-authority demand. This is not live data-centre load, not Dominion-zone load, and not county-level demand. EIA data can lag, so the page labels this as latest-reported grid context rather than live demand.

National US context

For broader national context on US data-centre electricity use, the following sources are relevant but are not used as model inputs for this site:

• LBNL 2024 US Data Center Energy Usage Report (Lawrence Berkeley National Laboratory / DOE, 2024). The closest US analogue to a national baseline, estimating total US data-centre electricity consumption and projecting significant growth. DOE has separately noted that data-centre load could double or triple by 2028.
• EPRI 2026 data-centre electricity analysis (Electric Power Research Institute, 2026). Estimates data centres could consume 9–17% of US electricity by 2030, using state-level capacity data. This is a scenario/forward projection, not a measured load figure, and should be read as a range of possible futures rather than a near-term demand estimate.

Neither the LBNL report nor the EPRI analysis is used as a model input in this site. They are cited as context only.

What would make this estimate better

• Site-level disclosure from operators. If major data-centre operators published their grid connection utilisation at half-hourly resolution, we could replace the scenario model with measured demand. Several European jurisdictions are moving towards mandatory reporting.
• National-scale measured data-centre load profiles. UKPN's half-hourly anonymised load shapes exist but require registration. An openly licensed national equivalent would allow the modelled band to be replaced by a real time series.
• Updated DSIT capacity estimates. The current dataset is autumn 2024. Annual updates at regional and sub-regional level would reduce the uncertainty in the capacity figure.
• US county-level or zone-level demand data. No source used here provides public data-centre-specific load figures at county or PJM-zone level for Northern Virginia. Regulatory filings, Virginia SCC dockets, and county planning records may hold better evidence.

Public authorities may already hold relevant evidence - demand forecasts, planning documents, grid-impact assessments, and connection assumptions.