Who Pays for Virginia's Data Centres

Virginia's affordability problem does not show up in its price. At 14.41 cents per kilowatt-hour the residential rate is the twenty-third lowest in the country, well under the 16.48-cent national average. A regulator reading only that number would find nothing to act on. The strain is in the bill set against an income: 373,645 households at or below 30% of area median income carry a median energy burden near 16%, against a 6% standard, and Virginia recorded 461,586 residential disconnections in 2024.

Into that sits the largest load buildout in the state's history, most of it not yet built. The commission asked how affordability should be defined and built into planning. Our answer is that it is a cost-allocation problem before it is a load-growth problem — and the section below is the part I built to prove it.

Demand is a distribution, not a point. The fitted layer carries the measured over-forecast bias; beyond a ten-year horizon the curve is held flat rather than extrapolated, which understates the premium because the fitted line rises. The assumed layer gives the above-embedded slice a realisation rate centred on 0.70, bracketed by five published anchors including the California Energy Commission's milestone tiers and Berkeley Lab's queue-completion rates. It is assumed, and the report says so in the body rather than a footnote.

Surplus megawatts — those a perfect-foresight planner would not have built — are priced at PJM's Net CONE for capacity and its network rate for transmission. The two arms commit against different horizons: capacity at the three-year auction lead, because the auction re-clears against the then-current forecast, and transmission at the full planning horizon, because RTEP commits long-lived assets. The GS-5 minimum demand charges then split that cost, and the residue reaches households through the residential class share. Every figure is a median over 50,000 draws.

The estimate fell twice during the work — from $8.08 to $4.95 when a discount already applied upstream was corrected, and to $4.46 when the two commitment horizons were separated. Neither correction changed the ranking that follows.

A distribution is not a planning rule. Each row is a trigger tied to something observable, with the action named in advance, priced by re-running the same model with that rule in force. The gates overlap by construction, so the rows do not add. The top row is a loss avoided rather than a saving, since the order is already in force.

Source: Eggnomics (2026), Table 2. Priced at 2040 against the $4.46 baseline.

1. 01

   Allocation beats forecasting about fifteen to one

   Sweeping the transmission recovery share between zero and the ordered 85% moves the residential bill $9.89 a month. Removing PJM's forecast bias altogether — perfect forecasting, the ceiling on any forecasting reform — moves it $0.67. The comparison holds across the whole tested range of the least certain assumption in the model.

2. 02

   The order already in force is the most valuable action available

   Holding GS-5 against waiver, true-up or retroactive relief is worth $13.96 a month, more than the other four gates combined. It is defensive, it needs no new authority, and it is the commission's own docket. FERC supplied the precedent when it denied AEP's request to shift surplus capacity cost onto ratepayers after forecast load failed to appear.

3. 03

   Most of the answer rests on a number nobody publishes

   About 91% of the reported spread at 2040 comes from the realisation-rate assumption rather than from fitted data. PJM's Board created a large-load reporting obligation in January 2026 with first submissions due three weeks after this report was filed. That is a dated fact about the record, not a gap in the analysis.

4. 04

   Three of the five gates are not Virginia's to pull

   What enters the zonal forecast is set by PJM's Manual 19 process, transmission need cases are staged through RTEP, and the capacity-market exclusion is a PJM Board decision. On those, the commission's role is advocacy and intervention at FERC. Only two are retail tariff terms it sets directly — and those are the ones it can move without anyone's agreement.

The recommendation that follows is not to repel the load. It is to price its risk to the customers who cause it, and to stop treating a forecast as a fact. A commission that adopts one rule from this work should adopt the first: refuse waiver and retroactive relief against terms it has already ordered, because the instrument only works ex ante and giving it away costs households more than every other lever combined.

The method travels. The model runs on public data any state commission holds — PJM or its own RTO's forecast tables, EIA-861, LEAD, and its own docket record — so any jurisdiction facing concentrated load growth can re-run it on its own numbers rather than borrowing Virginia's.