Seventy Gigawatts Is Not Load: Dominion’s Data-Centre Queue, Contracts and Meter

Four numbers, one label

The "70 GW" attached to Dominion Energy's Virginia data-centre demand is not load. It is the total of large-load delivery-point requests described in Dominion's 2 February 2026 filing to the Virginia State Corporation Commission: about 25,000 MW for facilities projected to connect by the end of 2031, plus about 45,000 MW with no connection date set³. A planner who reads it as demand, or even as committed capacity, is working from a number Dominion's own forecast does not use.

Dominion's 29 September 2026 submission to PJM's Load Analysis Subcommittee lets the label be taken apart¹. Under "data-centre demand" sit four quantities with different bases. They cannot be added, netted against one another or quoted under the same word.

How far apart they are

Billed load today is 4.8 GW¹. Against 53.8 GW of executed contracts, that is about 9% of contracted capacity showing up as a billing peak; against the 70 GW queue, under 7%. Both ratios are this brief's arithmetic on Dominion's own figures, and they compare different dates.

The gap is not a discrepancy. Dominion's model builds it in:

• Ramp and utilisation. Dominion uses an "industry average ramp period" of 5 years, "maturing at a utilization rate of 63%", defined as demand divided by contract capacity¹. A fully energised contract is expected to draw about two-thirds of its contracted megawatts, five years after connection.
• Non-firm contracts. Engineering Letters of Authorization are treated as non-firm load for 2032 and beyond. Of 32,473 MW of ELOAs, 19,044 MW (59%) are excluded from the forecast entirely¹.
• Energisation timing. The official billing-demand forecast is the average of a High case built on current energisation dates and a Low case built on historical energisation trends¹. The forecast does not assume customers connect when they say they will.

The output is far below the headline. Dominion's forecast table shows 11,590 MW of contract capacity and 5,075 MW of demand in 2027, rising to 34,299 MW and 19,466 MW in 2047¹. These are projections, plotted on a chart of coincident demand, not measurements. Even the 2047 endpoint is under 28% of the 70 GW queue. It is roughly the size of Dominion's entire current load-serving-entity peak of 19,775 MW on 30 June 2026, which is a measured all-load figure, not a data-centre figure¹.

Dominion's own forecast discounts its contract book before it reaches load: 59% of ELOAs to zero, the remainder to a 63% mature utilisation after a five-year ramp. The 70 GW queue does not enter the forecast at all.

Where the label slips

The conflation is not confined to social media. Reuters' report of Dominion's second-quarter results opened by describing "demand from server warehouses" that "grew beyond 50 gigawatts", and then gave the figure correctly in the body: Dominion's Virginia segment "had contracted nearly 53.8 gigawatts (GW) of data center capacity as of July, up 5.3 GW from December"⁴. MarketBeat's automated earnings-call summary gave "more than 53 gigawatts of data-center capacity in various stages of contracting, including about 12 GW under electric service agreements"⁵. That is consistent with the 11,997 MW of ESAs in Dominion's slide.

The 53, 53.8 and 54 GW figures are the same quantity, rounded differently by different publishers. They are not three measurements.

Every figure in this brief originates with Dominion. As a regulated utility whose investment plan and earnings narrative rest on data-centre growth, it is an interested party. Its own forecast, though, is the most conservative of the four numbers. The inflation happens downstream, when a queue or contract total is relabelled as demand.

What the record does not establish

• No forecast-versus-actual series for Dominion. The slide headed "Comparison of Forecasts" compares the current-year forecast with the prior-year forecast, not forecast with outturn¹. This brief therefore makes no claim that Dominion over- or under-forecasts.
• PJM's accuracy check is a share, not an error rate. PJM's May 2026 Load Forecast Accuracy Report shows the Dominion zone at 64.3% and 68.1% of PJM data-centre load at the July 2025 peak, under the headings "Actual Peak 2025" and "Forecast"². A share of a pool is not a megawatt comparison, and PJM states no percentage error. The actuals are "provided to PJM by EDCs/LSEs", not metered by PJM².
• Billing demand is undefined. The slides do not define it, and its relationship to coincident peak is given only as a modelled ratio. The 4.8 GW peak, the roughly 500 metered accounts and the 63% utilisation rate are operator figures with no independent confirmation¹.
• No cost data. Nothing in these sources prices the connections, the network build or who pays for capacity that is contracted but not drawn.

The planning judgement

Before using any Dominion data-centre figure, ask which of the four it is. For load, capacity and supply planning, the operative figures are the 4.8 GW billed today and Dominion's own forecast path, which starts at about 5.1 GW in 2027. Neither 54 GW nor 70 GW is a load figure.

Queue and contract totals measure intent to connect. Ramp, attrition and partial utilisation stand between them and the meter, and Dominion itself applies all three. A planner who treats 70 GW as demand has not found a bigger version of the forecast. They have found a different quantity.

This does not show that Dominion's forecast is right. No realised series exists in the public record to test it. It does show that the largest number in circulation is the one with the weakest claim to be called demand.