Who Gets Paid When AI Comes to Town

In rural Georgia, a homeowner may soon stop receiving a property tax bill because an AI company built a warehouse next door. OpenAI's Project Camellia — a $20 to $30 billion data center complex announced this week — comes with a 15-year property tax abatement, but local officials expect the remaining tax revenue to fund "Project Zero," an initiative to eliminate county homestead property taxes entirely ^[1]. The trade is explicit: AI infrastructure in exchange for homeowner tax relief. This is not a one-off. It is the leading edge of a shift that has moved faster than the policy debate has acknowledged. Six months ago, the question animating every statehouse and public utility commission was whether AI firms should be required to cover the grid costs their data centers impose. That question has been answered — the cost-recovery consensus is now federal policy, backed by a FERC order, a White House pact signed by more than 220 companies and 22 governors, and bipartisan legislation in both chambers ^[2][3]. What has emerged since is something different: a subset of governments has moved beyond making AI firms pay their own way and begun using them as a revenue base. The White House has given this logic its bluntest expression. In a policy announcement this week, Trump called data centers "money machines for the state in which they are built," citing Google's Georgia investments eliminating property taxes and Meta's $27 billion Louisiana project funding teacher bonuses ^[2][4]. The framing is not rhetorical — it is operational. The same federal government that designated data centers as critical national security infrastructure in a July 2025 executive order separately required tech firms to fund their own power plants through a February 2026 pact ^[5][6]. The posture is coherent: build as much as you want, but the public must be paid. The states have taken the invitation. Oregon's POWER Act raised data center power rates by 29 percent, and the state's public utility commission layered a 1-cent-per-kilowatt-hour surcharge on loads exceeding 100 megawatts — the proceeds fund low-income energy programs and a 1.3 percent residential rate decrease, saving non-data-center customers an estimated $900 million over 30 years ^[7][8]. Texas lawmakers are moving to repeal a data center sales tax exemption that has grown from under $30 million annually to $1.3 billion this year, with projected losses of $3.2 billion over two years — a clawback driven by the realization that AI infrastructure is consuming more public resources than it contributes ^[9]. New York paired its moratorium with a plan to repeal sales tax exemptions for large data centers, extracting revenue even while restricting construction ^[10]. Each case goes further than cost recovery: the money is being redirected to purposes that have nothing to do with the grid. The shift is not universal. Tennessee's law is pure cost recovery — data centers pay for substations and power upgrades, utilities cannot raise rates on other customers, and no revenue flows to the general fund ^[11]. Maine's statewide moratorium genuinely pauses construction to study impacts without extracting revenue, and even restricts tax incentives ^[12]. Alabama and Arizona have seen utility oversight become a political battleground over rate freezes and energy mix, but neither is using data center tariffs as a revenue stream ^[13]. The extraction model is emerging in specific states — Oregon, Texas, Georgia, New York — not sweeping the country. But the direction of travel is clear, and it is being led from Washington. The mechanism that makes extraction politically viable is the same one that makes it geographically unstable. The fiscal benefits — tax revenue, rate reductions, teacher bonuses — accrue at the state and federal level. The physical costs — water consumption, noise, land use, the transformation of rural landscapes — remain local. That split explains why moratoriums keep spreading at the city and county level even as state capitols and Washington accelerate expansion. Two cases make the point precisely. Applied Digital spent $75 million to self-fund a substation in North Dakota, covering every infrastructure cost a state regulator could ask for. Neighboring counties — Oliver and Mercer — enacted moratoriums anyway ^[14]. Meta is directly funding a 366-megawatt natural gas plant in El Paso to power its $1.5 billion data center; the utility confirmed the customer requested and is funding the resource. The city of El Paso opposes the project as a violation of its climate plan ^[15]. In both cases, the fiscal bargain satisfied the state-level question — who pays for the power — and did nothing to resolve the local one: who wants this in their backyard. The industry's largest players have accepted the pay-to-play frame, which is precisely what has made extraction possible. Microsoft's Brad Smith said in January that the industry needed to cover the cost of electricity itself ^[16]. Anthropic went further in February, committing to directly compensate ratepayers through credits, payments, or stabilization funds — a private subsidy program that Senator Van Hollen now wants to make mandatory by federal law ^[17]. Once the leading AI labs accepted the cost-recovery consensus, it became politically easier for governments to ask for more. The industry did not have to embrace revenue extraction for the logic to advance; it only had to stop fighting the premise that it should pay. The fight over whether AI firms should pay is over. What is replacing it is a fight over who collects — and the answer so far is state capitols and Washington, not the county commissioner fielding noise complaints. That is why the moratoriums and the acceleration are happening at the same time. They are two sides of the same bargain, and only one side of it is being offered to the people who live next door.