MTS

Wiki/Primary documents

JLARC Virginia Data Centers Report (December 2024), Annotated

Document: Data Centers in Virginia — Report to the Governor and the General Assembly of Virginia, JLARC Report 598. Released December 9, 2024. ~140 pages of main text plus twelve appendices. Issuer: Joint Legislative Audit and Review Commission of the Virginia General Assembly. Project leader: Mark Gribbin. Consultants: Weldon Cooper Center for Public Service at UVA (economic impact + demand forecast); Energy + Environmental Economics, "E3" (grid modeling + rate impacts). Primary source: https://jlarc.virginia.gov/pdfs/reports/Rpt598-2.pdf Commission briefing slides: https://jlarc.virginia.gov/pdfs/presentations/Rpt598Pres-1.pdf

Why this report matters

Northern Virginia hosts the largest data center market in the world — "13 percent of all reported data center operational capacity globally and 25 percent of capacity in the Americas" (Summary, p. i). When the U.S. debates whether AI-driven hyperscale buildout is sustainable, it is debating, in practice, whether the Virginia model works. JLARC is the only government entity that has answered with primary data.

A caveat before the figures. The most-cited number in journalist shorthand — "$2.7 billion in foregone tax revenue" — does not appear in the report. The $2.7 billion figure JLARC reports is the annual GDP contribution from data center operations (Table 2-1, p. 15), not cumulative tax forgone. The actual exemption value is $928.6 million in FY23 alone, of which $683 million is the state share (Ch. 2, p. 20). Cumulative state-and-local foregone revenue since 2010 is plausibly in the $2.7B range, but the report does not provide that headline number. Treat citations of "$2.7B foregone" as an inference, not a JLARC finding.

Methodology JLARC used

Five evidence streams:

  1. Over 250 interviews with 150+ stakeholders (Ch. 1, p. 1).
  2. VEDP exemption-recipient filings. Beginning in 2021, recipients are required to report eligible expenditures and tax benefits. The disclosure produces "higher estimates of the tax revenue impact of the data center exemption than was reported in prior years" (Ch. 2, p. 20 sidebar) — this is what unlocked the $928.6M FY23 figure.
  3. A Weldon Cooper / IMPLAN economic-impact model. JLARC customized IMPLAN sector 436 rather than the generic 518210 NAICS code, which produces "aggregation bias": "only 15 percent of the total employment in the sector was data center employment... most employment in this sector involves other IT services, such as document scanning and software development, particularly in federal IT contracting in Northern Virginia" (App. D, p. 109).
  4. An unconstrained-demand forecast from Weldon Cooper, plus an E3 grid model running two scenarios (unconstrained, half-unconstrained) with and without Virginia Clean Economy Act compliance.
  5. An E3 cost-of-service review of Dominion, Mecklenburg Electric Cooperative, and Northern Virginia Electric Cooperative.

What's outside the model matters: the IMPLAN model "does not account for the cost of some potential externalities, such as health and environmental costs associated with increased carbon emissions" (Ch. 2, p. 14); the grid model is "based on current state and federal laws and regulations. Some costs, such as the social cost of carbon, were not explicitly included in the model" (Ch. 3, p. 28).

Headline findings (with page cites)

Economic and fiscal: the 74,000 jobs claim, qualified

The 74,000-jobs figure is JLARC's, but heavily qualified. Table 2-1 (p. 15) breaks it down: construction phase 59,000 jobs (35,000 direct); operations phase 15,000 jobs (4,400 direct); total 74,000 (39,400 direct). Labor income: $4.3B construction + $1.2B operations = $5.5B. Virginia GDP: $6.4B + $2.7B = $9.1B.

Chapter 2 makes clear what the 74,000 jobs actually are:

"Data centers were estimated to contribute 59,000 jobs annually during the construction phase, accounting for 80 percent of total annual jobs resulting from data centers. This estimate includes 35,000 direct jobs, most of which were construction workers (28,000), although some were IT-related workers manufacturing and installing equipment (7,000). Another 24,000 jobs were estimated to be in supporting sectors, such as materials suppliers, and 'induced jobs' in businesses that benefit from worker spending, such as restaurants and retail" (Ch. 2, p. 15).

80% of the 74,000 figure is construction. Of the ~15,000 operations-phase jobs, only 4,400 are direct data-center employees. The briefing slides are plainer: "During operations, typically employ small number of workers relative to facility size (~50)" (Presentation, slide 16). The Table 2-1 note concedes the figures "likely underestimate the impacts in more recent years given the growth of the industry" — the 74,000 is anchored to FY21–FY23.

The sales-tax exemption: largest economic-development incentive in Virginia

"Data centers and tenants reported saving $928.6 million in sales taxes in FY23 because of the exemption, including state, local, and regional portions of the tax. The state portion of the exempted amount was an estimated $683 million, making it by far the state's largest economic development incentive, with the next closest incentive valued at $74 million" (Ch. 2, p. 20).

Distribution is concentrated: "the median savings for a data center company using the exemption was $5.4 million in FY23, and all but six companies saved $1 million or more" (Ch. 2, p. 20).

JLARC's comparative scoring (Table 2-2, p. 23) is unusual for a tax-incentive critique because it concludes the exemption performs typically, not poorly, against other Virginia incentives: 84 jobs added per $1M (vs. Virginia incentive average of 58); income added $6M (vs. $5M); GDP increase $10M (vs. $9M); return in revenue per $1 spent: 48¢ (vs. 41¢). "Like most economic development incentives, the data center exemption does not pay for itself when considering just the state portion of the exemption cost and the state return in revenue" (Ch. 2, p. 22–23). That last clause is load-bearing: the program is rated "moderate" only because most state economic-development tax incentives also do not pay for themselves.

Capital investment, not jobs, is the story

"Capital investment in Virginia data centers is substantial, exceeding $24 billion in FY23" (Ch. 2, p. 13).

Electricity demand: double by 2034, more by 2040

"JLARC's independent forecast shows that unconstrained demand for power in Virginia is expected to double within the next 10 years, driven primarily by the data center industry's growth. Almost all of the demand growth is expected to occur in the Dominion transmission zone... JLARC's forecast largely matched the most recent PJM forecast" (Ch. 3, p. 27).

"By 2024, PJM forecast an unprecedented 5.5 percent year-over-year growth in the Dominion transmission zone, mainly because of increasing data center demand... The state's energy demand was essentially flat from 2006 to 2020 because, even though the population increased, improvements in energy efficiency offset that increase" (Ch. 3, p. 27).

The infrastructure-build implication is unusually candid for a state report:

"Building enough infrastructure to meet unconstrained energy demand will be very difficult to achieve, with or without meeting the Virginia Clean Economy Act (VCEA) requirements... new solar facilities would have to be added at twice the annual rate they were added in 2024, and the amount of new wind generation needed would exceed the potential capabilities of all offshore wind sites that have so far been secured for future development. Large natural gas plants would also need to be added at an equal or faster rate than the busiest build period for these facilities (2012 to 2018)" (Summary, p. iii).

Even half-of-unconstrained is hard: "New gas would need to be added at the rate of about one large 1,500 MW plant every two years for 15 consecutive years, equal to the busiest period of the last decade." "Both Scenarios 1 and 2 would rely on energy from as yet unproven nuclear technologies" (Summary, p. iv).

Rate impacts on residential customers

Table 4-2 (p. 48) shows residential rate impacts in constant 2024 dollars for a Dominion residential customer using 1,000 kWh/month. Typical 2023 generation-and-transmission charges: $90/month. Unconstrained demand (very difficult to achieve): +$23 by 2030, +$37 by 2040 with VCEA; +$22 / +$33 without. Half-unconstrained demand (difficult to achieve): +$7 / +$14 with VCEA; +$6 / +$14 without.

JLARC flags that this is only the generation-and-transmission slice:

"Dominion's total residential bill projections, from its integrated resource plan, show much larger overall increases than the numbers reported here. Dominion's projections apply to the whole residential customer bill and include several costs that are not captured in JLARC's analysis, such as distribution costs and the cost of some additional transmission and generation projects that may not be solely attributable to data centers" (Ch. 4, p. 47).

The cost-benefit conclusion, actually

JLARC does not issue a single net verdict. It splits the question across actors:

  • State economy: positive, but "mostly because of the industry's substantial capital investment. The primary benefit comes from the initial construction of data centers" (Summary, p. i). Most of the GDP and income gains are construction-phase and require continued growth to sustain.
  • Host localities (Loudoun, Prince William, Fairfax, Henrico, Mecklenburg): clearly positive. "Data center revenue ranged from less than 1 percent to 31 percent of total local revenue" (Summary, p. ii). Loudoun: $733M (31%); Prince William: $110M (7%) (Presentation, slide 19).
  • Non-data-center utility customers: prospective negative through 2040.
  • State taxpayers in aggregate: the exemption "does not pay for itself" on a state-revenue basis (Ch. 2, p. 22).

This split — state-level moderate-positive, locality-level large-positive, ratepayer-level negative — is the actual policy substance, and is what makes the report easy for any side to quote selectively.

Underrated findings (with page cites)

Water use is, presently, fine

A finding that surprised commentators expecting an alarm:

"Most data centers use about the same amount of water (or less) as an average large office building (6.7 million gallons per year), although a few require substantially more, and some require less than a typical household. In 2023, 11 data center buildings each used over 50 million gallons, including one building that used 243 million gallons (10 percent of the industry's total use)... In 2023, the data center industry used an estimated 2.1 billion gallons of water, with just over a third coming from reclaimed water instead of new withdrawals. Data center water use accounted for less than 0.5 percent of total state withdrawals" (Ch. 5, p. 62).

For comparison, "the state's largest industrial water user in 2023 used about 36.5 billion gallons of water annually" (Ch. 5, p. 61) — roughly 17 times the entire data center industry combined.

Co-op stranded-cost risk and the bankruptcy scenario

The report names a risk that has not made it into press coverage:

"Data center customer demand could account for 80 percent or more of energy sales for some co-ops by 2030... If a data center customer delayed, disputed, or failed to pay an energy generation bill and the co-op was unable to recoup these costs from the customer, they would ultimately have to be paid by all other co-op members. A large enough bill could potentially result in a co-op defaulting and going bankrupt" (Summary, p. vi; Presentation, slide 46).

Overbuild risk if AI doesn't pan out

"Generation could be overbuilt if a substantial portion of the expected data center demand does not materialize, or if there is a decrease in that demand overtime. As a result, non-data center customers would pay a larger share of the fixed costs for this new generation... much of the data center industry is concentrated in a small number of companies... if one of the major hyperscaler companies decided not to pursue development of new artificial intelligence (AI) products or has a line of AI products that fail to be commercially viable, then energy demand from that company could decrease substantially" (Ch. 4, p. 50).

This is the most explicit acknowledgment in any state regulatory document to date that the AI capex bull case may collapse and ratepayers would absorb the loss.

"Industrial uses [...] are incompatible with residential uses"

Quoting Loudoun County's own ordinance, JLARC writes: "industrial uses [...] are incompatible with residential uses due to the prevalence of outdoor storage and emissions of noise, odor, and vibrations" (Ch. 6, p. 74).

"One-third of data centers are currently located near residential areas, and industry trends make future residential impacts more likely... Inadequate local planning and zoning have allowed some data centers to be located near residential areas. In some cases, this occurred because local zoning ordinances did not consider data centers to be an industrial use... Local elected officials have also granted data centers exceptions that led to adverse residential impacts, such as approving rezonings that would allow data centers next to sensitive locations" (Summary, p. vii).

Several proposed sites abut elementary schools: "at least two proposed developments where the property also abuts an elementary school" (Fig. 6-3 note, p. 78). Resident descriptions, quoted in sidebars: "a giant monolith in the wrong place"; "a prison" (Ch. 6, p. 74). The decibel data shows the noise is rarely loud — "an estimated 40 to 59 decibels" — but constant. JLARC characterizes it as a "drone or hum" residents say causes migraines, disrupted sleep, and inability to concentrate (Ch. 6, pp. 75–76).

The diesel-generator finding cuts the other way

JLARC does not find a regional air-quality emergency: "Data centers' diesel generators are a relatively small contributor to regional air pollution — in Northern Virginia, they make up less than 4 percent of regional emissions of nitrogen oxides and 0.1 percent or less of carbon monoxide and particulate matter emissions" (Summary, p. vi). Emissions are "only 7 percent of what permits allowed (2023)" (Presentation, slide 53).

Data centers are currently paying their full cost of service

The cleanest pro-industry finding:

"JLARC staff commissioned an independent study of electric utility cost recoveries under current rate structures to see if the data center industry is paying its share of current costs. The study found that current rates appropriately allocate costs to the customers responsible for incurring them, including data center customers" (Summary, p. v).

The problem is forward-looking: "However, data centers' increased energy demand will likely increase system costs for all customers" (Summary, p. v).

Agency responses

Only the SCC and VEDP submitted written responses (Appendix C, pp. 105–108). The SCC's letter is one paragraph and does not engage substance. VEDP submitted a longer letter, key passages:

"VEDP strongly agrees with the report's finding that the sales and use tax exemption has been an important part of the industry's growth and continues to drive site selection and expansion decisions... Allowing the existing exemption to sunset would result in development shifting to competing markets, and those effects are likely already beginning to be felt given the long timeframes the industry uses to analyze their investments... we strongly agree with the report's warning that saddling an incentive program with competing policy priorities is not sound economic development practice. Furthermore, VEDP would caution against any action that could constitute a legal or moral failure to deliver on commitments to companies that have chosen to invest in Virginia" (Jason El Koubi, President & CEO, November 21, 2024).

JLARC's analysis is empirical and skeptical; VEDP's response is protective of the exemption and warns against conditioning it. The Department of Taxation did not submit a response. Dominion Energy, NOVEC, and Rappahannock Electric Cooperative were given the chance to comment but their letters are not reproduced.

Recommendations to the General Assembly (with page cites)

JLARC issues eight formal recommendations and ten policy options (pp. xi–xiii). Recommendations are staff-endorsed; policy options are framed as judgment calls reserved to elected officials.

The eight recommendations, in compressed form:

  1. VEDP should clarify that data center sites are eligible for Virginia Business Ready Sites Program grants (Ch. 2).
  2. Clarify that electric utilities have authority "to delay, but not deny, service to customers when the addition of customer load cannot be supported" (Ch. 3).
  3. Expand the Accelerated Renewable Buyers program to include battery storage (Ch. 3).
  4. Require utilities to establish a demand response program for large data center customers and require participation (Ch. 3).
  5. Direct Dominion "to develop a plan for addressing the risk of generation and transmission infrastructure costs being stranded with existing customers" filed with the SCC (Ch. 4).
  6. Authorize localities "to (i) require proposed data center developments to submit water use estimates and (ii) consider water use when making rezoning and special use permit decisions" (Ch. 5).
  7. Authorize localities to require sound modeling studies before approval (Ch. 6).
  8. Authorize localities "to establish and enforce maximum allowable sound levels for data center facilities, including (i) using alternative low frequency noise metrics and (ii) setting noise rules and enforcement mechanisms in their zoning ordinances" (Ch. 6).

The three core policy options on the exemption (Ch. 7): Option 8 extends 2035 → 2050; Option 9 allows expiration in 2035; Option 10 extends a partial exemption 2035 → 2050. JLARC's framing of Option 10 is the most analytically pointed passage in the report:

"A partial exemption would also better align the economic benefits the state receives with the exemption's value. Most economic benefits occur during construction, and switching to a partial exemption in 2035 would reduce the value of the exemption in later years when the economic impacts of current and planned data centers could be expected to slow. A partial exemption would also generate more revenue for the state. For example, a 1 percent partial sales tax would have generated approximately $160 million in state tax revenue in FY23" (Ch. 7, p. 90).

Additional options bundle environmental, historic, and residential conditions onto exemption eligibility (ISO-50001 energy management; ISO-14001 environmental management; Tier 4 generators in NoVA ozone nonattainment areas; Phase I historic and viewshed studies; sound modeling).

2025 legislative response

The 2025 General Assembly session produced roughly 20 data-center bills explicitly responsive to JLARC. Four survived passage. Governor Glenn Youngkin vetoed the most substantive ones in May 2025.

  • HB 1601 (Del. Josh Thomas, D-Prince William) / SB 1449 (Sen. Adam Ebbin, D-Alexandria, with Sen. Jennifer Carroll Foy, D-Prince William) — Required site assessments for facilities using 100+ MW: sound impacts within 500 feet of residences and schools, effects on water, agriculture, parks, forests, and utility-infrastructure needs. The most direct legislative implementation of JLARC Recommendations 6, 7, and 8. Vetoed by Youngkin, May 2, 2025. Rationale: "one-size-fits-all approach" undermines local autonomy; bills "could hinder data center development and chill future investment." The General Assembly had rejected Youngkin's substitute amendments.
  • HB 2084 (Del. Irene Shin, D-Fairfax) — Originally required the SCC to determine if non-data-center customers subsidize data center costs (JLARC Rec. 5). Weakened in conference; final version only directs SCC to "consider creating" new customer classifications — language already within existing SCC authority.
  • SB 1047 (Sen. Danica Roem, D-Prince William) — Originally mandated demand-response programs for 25+ MW customers (JLARC Rec. 4). Converted in committee to a study requirement; Department of Energy and SCC report due November 1, 2025.

About 16 of 20 introduced data-center bills failed in committee. Sen. Russet Perry, who carried the SCC customer-class bill before withdrawing it after it was gutted: "We didn't do nearly enough."

The 2035 sunset remained unchanged in the 2025 session. The exemption issue moved to the 2026 budget process: the Senate proposed accelerating the sunset to January 1, 2027; the House proposed retaining the exemption with new clean-energy conditions (PUE ≤ 1.2 or 90% carbon-free electricity procurement by 2028). FY2024–25 reported exemption claims totaled $3.2 billion against $80.6 billion in claimed capital investment — the program's annual cost has tripled since JLARC's FY23 baseline.

The net political result: of JLARC's eight recommendations, zero have been enacted as Virginia statute through May 2025; the strongest implementing legislation was vetoed; the exemption question was deferred to budget negotiations the following year.

Open questions and criticisms

Methodology: The IMPLAN model's exclusion of carbon externalities and the social cost of carbon (Ch. 2, p. 14; Ch. 3, p. 28) means the 74,000 jobs / $9.1B GDP / "moderate" cost-benefit rating are gross-of-externality numbers. A carbon-priced model would likely flip the state-revenue picture sharply negative. The Weldon Cooper analysis is anchored to FY21–FY23 averages and "likely underestimate[s] the impacts in more recent years" — that cuts both ways: economic benefits and foregone tax are higher than headline numbers. The aggregation-bias correction to IMPLAN sector 436 (App. D) is a methodological contribution: most state-level data-center economic-impact studies use the uncorrected 518210 NAICS code and substantially overstate jobs.

Scope: The report does not separately analyze AI training versus general cloud workloads despite naming AI as the principal driver of forecast demand growth. The forecast implicitly bundles both, with no sensitivity to AI commercial viability — except the brief acknowledgment (Ch. 4, p. 50) that hyperscaler AI failure is a stranded-cost risk. The "half of unconstrained demand" scenario is treated as a planning benchmark rather than an upper bound on what the grid can physically achieve. Rate-impact analysis focuses on Dominion residential customers using 1,000 kWh/month; APCO, NOVEC, and co-op customers may face materially different exposures.

Political reading: The asymmetry between Recommendations (technical, legislatively cheap) and Policy Options 8/9/10 (substantive, multi-billion-dollar) is itself a finding. JLARC declined to recommend a specific course on the exemption, but the structure makes Option 10 the staff-implied preference — it is the only option paired with an affirmative quantitative case ($160M in additional FY23 revenue at a 1% partial rate). The VEDP response letter is the closest the public record gets to inter-agency disagreement: where JLARC writes that the exemption "does not pay for itself," VEDP warns that expiration would cause "development shifting to competing markets" and cautions against conditions. The Department of Taxation did not respond on the record. Dominion's full response is not reproduced. The 2025 vetoes reframe the report's downstream effect: JLARC's recommendations were not analytically controversial — most simply authorized local governments to act — yet the executive branch judged even those modest authorizations as a chill on investment. Virginia is now committed to its data-center growth model not on the merits of the JLARC cost-benefit but on executive preference and existing performance agreements with hyperscalers.

Next: The 2026 budget cycle resolves the 2035 sunset. The SCC's December 2024 technical conference on large-load customers (Ch. 4, p. 49) is the likeliest near-term venue for rate-allocation. The Department of Energy / SCC demand-response study from the modified SB 1047 (due November 1, 2025) is the first formal follow-up document.

Sources: