The Cloud Kettle Index

Data-centre electricity demand in Northern Virginia, shown in kettle-boils per second

The US Cloud Kettle Index translates Dominion Energy Virginia's published data-centre billing-demand figures into kettle-boils per second. It is framed for DC + Mid-Atlantic readers because Northern Virginia is the region's dominant cloud-infrastructure cluster.

Northern Virginia cloud-load proxy - 2026 projection

~13,200 kettle-boils per second

Dominion 4,753 MW billing demand equivalent ÷ 0.36 MW/kbs · projection

The published demand data is public. Measured load is not. This page uses the best public proxy sources. - Full methodology

2026 projection: Dominion Energy Virginia publishes 4,750 MW of data-centre billing demand in its Virginia service territory. At 0.36 MW per kettle-boil/sec, that is ~13,200 kbs.

Scope and average-load context: Dominion Energy Virginia service-territory data-centre billing demand. Framed as DC + Mid-Atlantic because Northern Virginia, which dominates Dominion’s data-centre load, is the cloud-infrastructure cluster most relevant to DC readers. Dominion reports an industry load factor of approximately 90%, implying roughly 4,280 MW average continuous draw (~11,900 kbs). The headline uses billing demand because that is the public series Dominion provides.

Full methodology →

Northern Virginia and Washington DC county locator map Census county-boundary locator map highlighting Loudoun, Fairfax, and Prince William counties, with Arlington and the District of Columbia shown for orientation. Fairfax CountyLoudoun CountyPrince William CountyArlington CountyDistrict of ColumbiaLoudounFairfaxPrince WilliamArlingtonDCMarylandVirginia
Dark counties are the Northern Virginia focus: Loudoun, Fairfax, and Prince William. Arlington and DC are shown for orientation. Dominion data covers the wider Virginia service territory. Based on US Census Bureau 2024 Cartographic Boundary Files, county layer (1:20,000,000). This is not a Dominion service-territory map.

Source trail

The US page mixes several public sources. These cards show what each source does, and what it does not prove.

Billing demand, not metered load

4,750 MW is Dominion's 2026 projection data-centre billing demand - used for tariff calculations, not measured grid draw. Dominion separately derives coincident (metered) demand, which is generally lower. At ~90% load factor, average continuous draw is roughly 4,280 MW (~11,900 kbs).

What this is not

  • -Data centres physically inside DC.
  • -Total PJM data-centre load (13 states + DC).
  • -A full Northeast model (PJM + NYISO + ISO-NE).
  • -Live grid demand - modelled estimates only.

Published Dominion data-centre billing demand

Dominion's public figures show data-centre billing demand rising from 2024 actuals through the 2030 planning forecast. Select a year to inspect the conversion.

Published Dominion data-centre billing demand - 2024-2030

2026 projection: ~13,200 kettle-boils/sec

Billing-demand equivalent90% load-factor context
8k12k16k20kkettle-boils/sec2024202520262027202820292030~13,200

Selected year

2026 projection

Billing demand

4,753 MW

Average-load context

4,278 MW

View source table
YearKindBilling demand (MW)Kettle-boils/sec
2024actual3,584~10,000
2025forecast4,149~11,500
2026used aboveprojection4,753~13,200
2027projection5,322~14,800
2028projection5,863~16,300
2029projection6,419~17,800
2030projection6,992~19,400

Select a year to inspect the conversion. The GB page can show a live 24-hour share because GB grid demand is public through Elexon. Here, the public Dominion series is a forecast trajectory, not a live data-centre meter.

Sources: Dominion Energy Virginia and PJM Interconnection. Billing-demand figures converted using 0.36 MW/kbs; load-factor context uses Dominion's approximately 90% industry load factor.

Primary source: Dominion Energy Virginia 2025 data-centre forecasting presentation (Indiana IURC filing).

Also: PJM Dominion data-centre large load request (Load Analysis Subcommittee, September 2025).

Generic Dominion large-customer load shapes

Dominion publishes generic hourly profiles for customer classes. These are not data-centre profiles, but they are useful local context. Each line is scaled to its own average, so 1.0 means a normal hour for that class. Dominion has not published an equivalent hourly profile for data centres.

Median large-customer class: Median of GS3, GS4, LGEMLP, and MS normalized profiles.. This profile varies by 38.4% peak-to-trough on a weekday.

0.80.91.01.11.200:0003:0006:0009:0012:0015:0018:0021:00Relative load (mean = 1.0)
Selected Dominion classAverage baseline (1.0)

These profiles are settlement profiles, not measured data-centre profiles. COMM6VA is omitted because Dominion publishes it as a weather-sensitive formula rather than a simple hourly shape. Source.

What are these Dominion classes?
GS3
Large general-service customers connected at secondary voltage; a useful commercial/industrial comparator, but not data centres.
GS4
Large general-service customers connected at primary voltage; flatter than GS3 in this dataset and closer to very large continuous loads.
LGEMLP
Large miscellaneous light-and-power customers; broad, but still a large-load class in Dominion's published profiles.
MS
Military service profile; included as another large institutional load shape in Dominion's published profiles.

The median option takes the hour-by-hour median of GS3, GS4, LGEMLP, and MS after each class has been scaled to its own annual mean.

Latest-reported PJM grid context

Here the Dominion data-centre estimate stays fixed while total PJM demand moves through the day. It is regional context, not a Dominion-zone meter.

Grid context: Dominion proxy as share of PJM demand

5.9% of PJM hourly demand (4,753 MW ÷ ~80000 MW PJM)

Dominion 2026 projection data-centre billing demand is fixed here. PJM demand comes from EIA and changes hour by hour.

Billing demand90% average context
3.5%4.0%4.5%5.0%5.5%6.0%6.5%7.0%7.5%May 30noon ETMay 3111am ETJun 111am ET% of PJM demand

The line moves because PJM demand moves: usually higher on weekday afternoons, lower overnight and on weekends. The Dominion data-centre estimate is fixed, not a live meter. Cache generated Jun 1, 8:11 AM EDT.

Source: EIA Balancing Authority Areas hourly operating data. EIA values can lag; this is not a Dominion-zone or county-level meter.

Data request scoreboard

The estimate is only as good as the public record underneath it. These are the datasets that would make the DC + Mid-Atlantic index much stronger.

Help make the US data open
Missing or partial data Likely holder Status Next request
Measured Dominion-zone data-centre load Dominion Energy Virginia / Virginia SCC filings Missing Ask for hourly or monthly data-centre load by tariff/customer class, aggregated enough to protect customers.
Dominion service-territory boundary as GIS Dominion Energy Virginia / Virginia SCC Missing Ask for a public polygon or shapefile suitable for non-operational mapping.
Northern Virginia county-level data-centre load Counties, PJM interconnection queues, utilities Missing Ask Loudoun, Fairfax, Prince William, and Arlington for planning assumptions and utility-impact evidence.
Measured regional data-centre load shapes Dominion, PJM, operators, research partners Missing Ask for anonymised hourly profiles analogous to the UKPN data-centre profile release.
Latest PJM demand comparison EIA Balancing Authority Areas hourly data Wired Let the scheduled EIA refresh populate the static cache; keep it labelled as reported data, not live data.

Future expansion

A fuller Northeast version would need separate treatment for PJM, NYISO, and ISO-NE. A future PJM comparison could use Dominion-zone demand, if a reliable public series is available, instead of PJM total.

Great Britain page

The original Cloud Kettle Index: GB data-centre load based on NESO and DSIT figures, with live Elexon grid demand and UKPN load shapes.

Go to GB page →

Request US records

Ask for better data using Virginia FOIA, federal FOIA, DC FOIA, or utility commission filings. Ready-to-send templates for DC-area data-centre electricity records.

US public records templates →

Methodology

The formula, data sources, assumptions, and what this site does not claim.

Read the methodology →

vetch

Open-source LLM inference energy measurement and circuit breakers for runaway cost and carbon.

GitHub →