The Subsidy Reckoning

Virginia's electricity consumption tax, enacted as part of the state's biennial budget (HB 30, signed by Governor Abigail Spanberger on June 22, 2026), applies at $0.011 per kWh on all electricity consumed by data centers with at least one megawatt of capacity. There is no threshold that exempts smaller facilities; the tax applies from the first kilowatt-hour. The annual cost for a 100 MW campus running at 85% utilization: approximately $8.2 million per year in new operating expense, flowing directly to Virginia's general fund. The levy is capped at $600 million annually in aggregate, with any excess refunded pro rata to operators, and sunsets on July 1, 2028, unless renewed.

The state made no effort to disguise what it was doing as the political case for the tax was explicit in the budget debate. Data centers had consumed an outsized share of grid capacity while contributing a smaller share of the infrastructure costs that capacity required. A Gallup poll conducted in early 2026 found that 71 percent of Americans oppose building a data center in their local area. Virginia's legislators had the polling, the utility commission cost allocation data, and the constituent pressure. That combination, organized local opposition producing a measurable political signal, is what moved the fiscal argument from a utility commission study into a budget bill. In this case, political influence was a direct function of community opposition. What operators failed to model was that relational dynamic, and how quickly it converts into law once it reaches a budget negotiation.

Virginia is notably the first state to enact a consumption-based electricity tax on data centers, and they will not be the last. Ohio and Illinois have suspended new tax exemption agreements; Arizona passed a three-year pause on new exemptions; and North Carolina's General Assembly has active proposals to repeal the state's $50 million annual sales tax exemption entirely. The form varies by state, but the underlying reasoning remains the same, and is steady, everywhere.

The mechanism is different from a zoning veto or a permitting delay, which are the risk types we examined in Issues 01 and 02. Those are gate events: binary, visible, dateable. A state legislature voting to kill a special-use permit triggers a specific response in a specific project. An electricity consumption levy is a continuous cost event. Although it does not necessasrily kill a deal, the levy will consequently restructure every deal already in the ground of that jurisdiction.

This is why the Virginia July 1 date matters so much. Operators who had executed leases in 2024 and early 2025 (when the tax was already in committee), had in most cases priced power at the pre-tax rate. The leases did not contain pass-through provisions specific to consumption-linked taxes (as opposed to property taxes, which are routinely passed through). The operators absorbed the first invoice. But while some tried to renegotiate, most could not. The tax went directly to NOI compression, which flows directly to cap rate sensitivity.

The pattern at Northern Virginia's Data Center Alley is instructive. Roughly 2,800 MW of operational capacity sits in the region. At $0.011 per kWh, the aggregate annual tax burden across operating facilities in Data Center Alley, before the $600 million cap kicks in, would approach the cap itself. Virginia's budget office projected the tax would generate $600 million annually. That figure is not necessarily the problem. It is that the figure was not in anyone's model on the day the leases were executed. The tax created a nine-figure annual cost event in a market where most sophisticated participants believed they had already modeled all known risks.

Although community opposition does not explicitly enact a tax, it does generate the political conditions that make one possible. The activation chain of the relational dynamic matters as it gives legislators cover. It creates electoral pressure and produces the equity argument that shows up in budget testimony: data centers account for a disproportionate share of grid load but contribute a smaller share of the infrastructure costs that load creates. That argument, once documented in a utility commission cost allocation study, becomes the legislative raw material for a consumption tax. In the case of the residents of Loudoun and Prince William counties, who did not write HB 30, but they created the environment in which it passed through organized opposition against new campuses .

The electricity tax story has a second chapter that most analysis has not yet reached. The incentive packages that justified many of the Virginia deployments in 2022 through 2024 (the job-creation tax credits, the sales tax exemptions on equipment, the utility rate concessions) were structured with sunset clauses. Some were three years, some five years. In other cases, two years, with a renewal option that was always discretionary and never guaranteed.

Those clocks are now running simultaneously with the tax clock. A facility that qualified for Virginia's Job Creation Tax Credit in 2022, on the strength of a construction workforce that has since normalized to the skeleton operational staff a hyperscale campus actually runs, is now in the credit recapture window. The employment threshold that unlocked the credit is no longer being met. The credit is expiring into an operating environment that now includes the consumption tax. So now, the incentive package that made the deal pencil is not just running out, but it is running out into a higher-cost environment than existed when it was originally granted.

This creates what we describe as the incentive inversion window: the period, typically spanning years three through six of a facility's operational life, during which incentive expiration and new fiscal obligations overlap. For facilities that modeled incentive value as a permanent feature of the operating environment, the inversion window represents a structural repricing event that was embedded in the deal from the first day of construction.

The investors most exposed are those who acquired existing facilities or portfolios at cap rates that reflected incentivized operating economics. The asset was priced on the assumption that the stack was permanent (it was not).

The exposure distribution across the capital structure is not uniform, and it does not follow the conventional risk hierarchy. In a standard underwriting, equity absorbs volatility and debt is protected by coverage ratios. In an incentive inversion scenario, the mechanics invert. Fixed-rate debt issued against incentivized operating economics is fully exposed to the repricing event because the coverage ratios that justified the debt terms are now unachievable without the incentive stack that was assumed to be permanent.

The highest-exposure category is sale-leaseback portfolios acquired in 2022 and 2023, when the incentive environment was at its most favorable and deal volumes were highest. These transactions were underwritten on a projected NOI that embedded two assumptions now simultaneously wrong: that no consumption tax would land on operating facilities, and that the incentive stack would hold through the hold period. The consumption tax reduces NOI directly, while the incentive expiration reduces it further. Additionally, the acquisition price does not change. The effective cap rate implied by current NOI at that acquisition price is materially worse than what was underwritten, and the full incentive expiration impact is still twelve to thirty-six months out for the 2022-23 vintage.

The second exposure category is single-tenant net-lease structures with electricity cost pass-through provisions that were not drafted to anticipate consumption-linked taxes. These leases typically pass through utility costs above a base rate, but Virginia's consumption tax is not structured as a utility cost; it is a state tax collected by the utility on behalf of the state. Whether the pass-through applies is now a live contract interpretation question for every Virginia lease signed before July 2026.

The underwriting correction is not as technically complex as it may seem. It requires three inputs that were previously treated as stable and should now be treated as time-varying: the effective electricity cost (inclusive of all enacted and pending levies in the jurisdiction), the incentive benefit schedule (modeled as expiring on the contractual sunset date, not extended), and the lease pass-through probability (which is now a legal opinion, not an assumption). With those three inputs recast, the NOI profile of any existing or prospective data center investment changes materially in years three through seven.

For existing portfolios, the correction is backward-looking but not academic. Investors who acquired Virginia assets before July 2026 should now be running a tax-adjusted NOI sensitivity for every asset in that market, not as a stress test, but as the base case. The incentivized economics turned out to be running on borrowed time; building business models on subsidies that were always meant to be temporary. Now, most current environment must operate without them.

For new development, the framework is forward-looking and more tractable. Virginia's tax rate, structure, and sunset date are known quantities. Incentive expiration dates are contractually fixed. The genuine uncertainty is legislative trajectory in states where proposals are active. But, the signal is there before the legislation: community opposition filings, utility commission proceedings, bill committee movement. This information accumulates for months before a vote takes place. The political environment of a site is not fixed at the moment of underwrite. It should rather be calculated as a time-varying input, and it can be modeled as one.

The states recommended to watch most closely: North Carolina, Georgia, and Texas. North Carolina has active bills to repeal its data center sales tax exemption and impose grid cost tariffs on large-load facilities. Georgia and Texas are earlier in the cycle but share the same utility cost allocation dynamics and the same density of hyperscale concentration that drove Virginia's legislature to act. The timeline from first legislative introduction to effective date in Virginia was approximately eighteen months. There is no structural reason the same process would take longer in Raleigh, Atlanta, or Austin.

In retrospect, the incentive packages that drove the 2021–2024 data center deployment wave contained the seeds of the current repricing event. They created concentration, which then led to grid pressure. Grid pressure initiated political leverage, and in turn, political leverage created levies. The incentive that justified the deal became a real pressure point; once the facility was built, it was too late to exit. .

This is a structural feature of any infrastructure investment, which is not a unique American dynamic. It relies on temporary public subsidy to achieve target returns. The subsidy signals to legislators that the investment is profitable, which invites the question of who is capturing that profit, which eventually produces a fiscal mechanism designed to share it. The reckoning is not a policy failure but a policy cycle completing itself.

The implication for investors is not to avoid incentivized markets. It is to model incentives correctly, as temporary, conditional, and reversible, and to understand that the same legislative dynamics that created the incentive can, with sufficient political activation, create its opposite. The veto economies we examined in Issue 01, the permitting arbitrage we traced in Issue 02, and the fiscal transfer we have analyzed here are not three separate risk categories. They are three phases of the same activation chain: community opposition generates political cover; political cover produces legislative action; legislative action reprices the asset. Political influence, in each phase, is not an independent variable. It is a function of community opposition, which is why the residents who organize against the next campus are not the counterparty risk. They are the leading indicator of it. Measuring that signal, before it completes the chain, is what LANDMARQ was built to do.

The operators who model that dynamic accurately will price it correctly. The operators who do not will find out on the first electricty invoice. By then, it's too late to do anything about it.