An analysis of the contested economics of data centre tax incentives in the United States — the arguments for, the arguments against, and the wider implications for American fiscal policy.

A subsidy machine built for a different era

For two decades, US states treated data centres the way they once treated auto plants and film studios: as trophy investments worth almost any price. At least 37 states now offer targeted tax incentives to lure them, and the most common tool — a sales and use tax exemption on the servers, chips, cooling systems and electrical gear that fill these buildings — was designed in an age when a large facility might cost a few hundred million dollars and the fiscal exposure felt manageable.

That era is over. The generative-AI boom has turned the data centre into the single most capital-intensive structure in the modern economy, and the tax carve-outs written for the cloud-computing 2010s are now metastasising into some of the largest business subsidies in American history. In Texas, the comptroller projects the state will forgo roughly $3.2 billion in sales tax revenue over two years, rising toward $1.8 billion annually by fiscal 2030. In Virginia — home to 232 facilities and the largest data centre market on earth — the value of the equipment sales tax exemption reached nearly $2 billion in fiscal 2025. Georgia recently raised its own cost projection for fiscal 2026 by an eye-watering 664 percent, to $2.5 billion, with fiscal 2027 losses forecast near $3 billion. Virginia, Texas and Illinois have each seen the value of their exemptions spike by more than 1,000 percent in just a few years.

These are no longer rounding errors in state budgets. They are line items large enough to fund school systems, and they have transformed a once-sleepy corner of economic-development policy into one of the most genuinely contested questions in American public finance. The debate is not a simple morality play of greedy tech giants versus put-upon taxpayers. It is a real clash between two defensible views of how governments should tax capital, court investment, and share the costs of a technological transformation that no state can afford to ignore — or, increasingly, to bankroll.

The case for the incentives

Supporters of data centre tax breaks make an argument that is more serious than critics often allow, and it rests on three pillars: the nature of the tax being waived, the scale of capital that follows, and the transformation of local fiscal capacity.

Start with the tax itself. A well-designed sales tax falls on final consumption, not on business inputs. When a company buys servers to produce a service it will later sell, taxing that equipment layers a tax on a tax — precisely the kind of cascading, non-neutral levy that public-finance economists have warned against for generations. Seen this way, a sales tax exemption on data centre hardware is not a special favour at all but a partial correction of a badly structured tax. The Tax Foundation and other neutrality-minded analysts have long argued that if states simply refrained from taxing business inputs in the first place, they would not need a patchwork of industry-specific exemptions to stay competitive. The carve-out, in this framing, is a clumsy fix for an underlying design flaw rather than a giveaway.

The second pillar is the sheer magnitude of investment. A single hyperscale campus can represent two billion dollars or more of taxable capital dropped into a county that may never have attracted anything larger than a distribution warehouse. Even with generous abatements, the incremental property tax base, construction activity, and downstream spending can dwarf what a rural or exurban jurisdiction could otherwise hope for. In Virginia, industry-commissioned analysis credited data centres with generating $2.1 billion in total state and local tax revenue across fiscal 2024 and 2025 and supporting more than 31,000 direct and indirect jobs. Advocates point out that the relevant comparison is not "subsidy versus no subsidy" but "this investment versus the empty field it replaced."

The third pillar is fiscal transformation at the local level. Many of the counties competing for these facilities lean heavily on residential property taxes, which means homeowners shoulder the cost of schools, roads and public safety. A large commercial taxpayer that demands relatively little in the way of public services — no classrooms full of children, minimal traffic, few emergency calls — can broaden the tax base, lift a county's bonding capacity, and relieve pressure on households. Where communities have negotiated hard for infrastructure contributions, water-stewardship commitments, workforce funds and transparent reporting, the public case for hosting has been genuinely strong. Virginia's program, for instance, conditions its exemption on a memorandum of understanding requiring at least $150 million of capital investment and 50 jobs paying 150 percent of the local average wage. Georgia similarly gates its full exemption on a $100 million investment and at least ten new jobs. These are not no-strings handouts; they are conditional bargains, and in the right hands they can pay off.

The case against

The critics' case has grown sharper as the numbers have grown larger, and it too rests on three pillars: the collapse of cost control, the thinness of the jobs promise, and the hidden transfer of costs onto other ratepayers and taxpayers.

The first and most damaging critique is that these incentives have become budgetary black holes with no ceiling. Unlike a discretionary grant capped at a fixed dollar amount, a statutory sales tax exemption is open-ended: every server a company buys is exempt, so the state's exposure scales directly with an investment boom that no legislature anticipated. Watchdog group Good Jobs First has argued that this structure represents a loss of spending control — states have effectively written a blank cheque whose value is set by private capital-expenditure decisions, not by public appropriators. The 664 percent overnight revision in Georgia and the four-figure percentage spikes in Texas, Virginia and Illinois are the predictable result. Money that flows out through an uncapped exemption never appears in a budget debate, never competes openly against schools or Medicaid, and is extraordinarily hard to claw back once an industry has organised around it.

The second critique targets the jobs math. A $2 billion hyperscale campus typically employs only 300 to 500 people in permanent operations once construction crews leave. Divide even a modest annual exemption by that headcount and the implied subsidy per permanent job can run into the hundreds of thousands or millions of dollars — figures that would be politically radioactive if offered as a direct cash grant to any other industry. Defenders respond that jobs are the wrong metric for what is essentially infrastructure, but that reframing concedes the point that the traditional economic-development justification — good local jobs — largely does not apply.

The third and increasingly central critique concerns who bears the costs the balance sheet does not show. Data centres are voracious consumers of electricity and water, and their demand is now large enough to move wholesale power prices and force utilities to spend tens of billions on new transmission and generation. Under conventional utility ratemaking, those costs are typically socialised across the entire customer base — meaning ordinary households and small businesses can end up subsidising the grid build-out for facilities that are simultaneously receiving tax exemptions. A community can thus find itself paying twice: once in forgone tax revenue and again in higher electricity bills. That double burden, more than any abstract argument about tax neutrality, is what has moved data centre policy from the economic-development committee to the front page.

The federal layer that supercharges everything

The state-level fight does not happen in isolation. Federal tax policy has poured accelerant on the fire, most dramatically through the One Big Beautiful Bill Act, which restored 100 percent bonus depreciation for qualifying property placed in service after 19 January 2025. For data centres this is enormously valuable, because it lets owners immediately deduct the full cost of their most expensive components in year one. Servers and chips can account for roughly 60 percent of the total cost of ownership of a large facility, so the ability to expense them instantly rather than depreciate them over years is a substantial federal subsidy layered on top of whatever the state provides.

The federal package went further. It permanently removed depreciation, amortisation and depletion from the calculation of adjusted taxable income for interest-limitation purposes, which lets heavily leveraged data centre developers deduct far more of their financing costs — a meaningful benefit for an industry funding hundreds of billions in construction with debt. Provisions boosting rural investment vehicles added still more, with certain rural opportunity fund investments now eligible for a 30 percent basis step-up after five years. The upshot is a stacked incentive structure in which a single facility can enjoy immediate federal expensing, enhanced federal interest deductions, a state sales tax exemption, and a local property tax abatement all at once. Notably, several states — California, Massachusetts and New York among them — have decoupled from federal bonus depreciation, forcing operators to keep separate state and federal books and underscoring how contested even the mechanics have become.

A tale of two approaches

The clearest way to see the trade-off is to compare the states at either end of it. Virginia represents the maximalist bet. Over roughly two decades it built the densest concentration of data centres on the planet, and its exemption has become correspondingly enormous — on the order of $2 billion a year in forgone sales tax and, by some accounting, $1.6 billion in annual property tax abatement. Defenders can point to tens of thousands of direct and indirect jobs, billions in aggregate tax revenue, and a genuine cluster effect that keeps drawing investment. Critics can point to the same figures inverted: a subsidy so large it now shapes the state budget, concentrated grid strain in Northern Virginia, and local communities increasingly resistant to the next campus. Virginia is not a failure or a success so much as a demonstration of what full commitment actually costs and yields.

At the other end sit the pause states. Ohio froze new tax-break requests while a commission studies the issue; Illinois suspended new exemption agreements after a repeal effort stalled; Oklahoma and Georgia have floated trimming credits that were, until recently, marketed as competitive advantages. These states are betting that a moment of restraint will let them redesign the bargain — capping exposure, adding ratepayer protections, and demanding better data — before the fiscal commitment becomes as locked-in as Virginia's. The risk they run is equally real: investment is mobile, the AI build-out is happening now, and a facility that chooses Texas or Indiana during a pause may never come back. Neither approach is obviously correct, which is precisely why the debate is contested rather than settled.

The wider implications

Step back from the individual numbers and the data centre debate becomes a lens on several deeper tensions in American fiscal policy.

The first is the classic problem of interstate tax competition. No single state can unilaterally stop offering incentives without watching investment cross the border, so each is trapped in a prisoner's dilemma that collectively drains public revenue while merely reshuffling where facilities land. This is the "race to the bottom" that economists have long warned accompanies subsidy competition — and the AI boom has raised the stakes by an order of magnitude. Absent federal coordination or an interstate compact, the logic pushes every state toward giving away more than it gets.

The second implication concerns democratic accountability and budget transparency. Open-ended tax expenditures are, by design, less visible than direct spending. They do not require annual reauthorisation, rarely face the scrutiny of the appropriations process, and often lack the kind of clawback and reporting provisions that would let voters judge whether the bargain is working. The current backlash is, in part, a demand that these subsidies be brought back into the light — capped, sunset, audited, and made to compete openly against other public priorities.

The third implication is about industrial policy in disguise. The United States has, without ever quite deciding to, assembled a sprawling, decentralised subsidy regime for AI infrastructure — assembled not through a deliberate national strategy but through the accretion of state exemptions, local abatements and federal depreciation rules. Whether one views this as a fortunate accident that is financing critical strategic infrastructure, or as an unaccountable transfer from the public to the most valuable companies on earth, depends heavily on how much one trusts that the promised long-term benefits will materialise where the costs are being paid.

Where the policy is heading

The politics have shifted with striking speed. In the first six weeks of 2026 alone, more than 300 data centre bills were filed across at least 30 states, and the centre of gravity has moved decisively from luring facilities to regulating them. Ohio's governor paused new tax-break requests pending a state study; Illinois suspended new exemption agreements after lawmakers failed to repeal the underlying break; Virginia, Georgia and Oklahoma have all floated trimming their credits. On the energy side, at least 18 states have introduced bills creating special rate classes for very large power users, and Congress is weighing a Ratepayer Protection Act that would push utilities toward "large load" standards requiring data centres to pay for the grid upgrades they necessitate rather than passing the bill to households.

What is notable is that almost no state has moved to abolish its incentives outright. The emerging consensus is not repeal but recalibration: keeping the door open to investment while capping the fiscal exposure, attaching enforceable community and ratepayer protections, and demanding transparency about what the public is actually buying. That is a reasonable place for the debate to land, but it papers over the unresolved core question. Is a tax exemption on business inputs a sensible correction to a flawed tax, or an uncapped giveaway to the richest firms in the world? The honest answer is that it can be either, depending entirely on how it is designed — and for most of the past decade, it was designed for a world that no longer exists.

The states now rewriting these rules are, in effect, deciding how the costs and benefits of the AI era will be distributed between the companies building it and the communities hosting it. That is a genuinely difficult trade-off, and the sums involved guarantee it will remain one of the defining fiscal fights of the decade.

Analysis prepared for us-taxdesk.com. This piece weighs competing policy arguments and does not constitute tax or investment advice. Figures are drawn from public reporting and government projections as of mid-2026 and should be verified against primary sources before republication.

Sources