Hyperscalers' Private Gas Plants Sidestep Ratepayer Protection, Leaving Grid Costs Behind

Research firm Cleanview has documented what amounts to a regulatory escape hatch: 59 hyperscale data centers with combined capacity of roughly 90 GW are planning to build their own natural gas plants and disconnect from the public grid rather than wait for interconnection.^[5] The originating reporting^[1] framed this as a speed play (grid connections take years; private generation takes months). Correct. But the cost allocation story is darker and already embedded in law: when these projects go "behind the meter," they avoid regulation, avoid ratepayer tariffs, and most critically, avoid cost-sharing for the grid infrastructure they would have needed. The bill for that infrastructure does not vanish. It gets socialized.

Start with the carbon math. Even conservative emissions assumptions put these 59 plants at more than 200 million tons of CO2 annually, equivalent to more than 46 million gasoline-powered cars.^[1] That's not incidental: it's the climate cost of regulatory arbitrage. But the financial arbitrage is the beat. A data center that would once have demanded a special contract negotiation with a utility, triggering tariff review and (at best) some form of cost isolation, now simply builds onsite. No negotiation. No tariff. No ratchet protecting residential ratepayers if the load underperforms. The developer avoids the interconnection queue entirely.

What does the grid lose? Consider the Louisiana example: a single data center consuming 2.2 GW, roughly twice New Orleans' peak demand.^[3] If that load had come through the grid, the utility would have had to justify new transmission, possibly new generation. The utility would file for a rate increase. The state commission would open a docket. Intervenors (residential customers, competitors, public-interest groups) could inspect the contract, demand ratchets, collateral, cost isolation for that customer class. Not perfect; the Harvard Electricity Law Initiative's work on data-center tariffs shows utilities still extract rents even under scrutiny.^{[background library]} But there is a gate. When the load goes behind-the-meter, the gate is gone.

The grid infrastructure does not disappear either. Transmission upgrades that would have served that 2.2 GW load may still be needed to keep the rest of the system stable, to handle imports from distant wind farms, to route power around congestion. The utility still builds them. Ratepayers still pay. The difference: the entity consuming the power that justified the investment is not contractually responsible for its cost. This is cost socialization by default. The Cleanview portfolio of 59 plants represents roughly 90 GW; if even half actually build, and if grid upgrades that would have supported 40 GW of interconnected load are now spread across a smaller base of regulated customers, the per-unit cost to homes and small businesses rises. No rate case. No docket. No visibility.

There is also the question of which projects are real. Cleanview notes that 92% of the 59 projects, representing roughly 82 GW, were announced since the start of 2025.^[5] In the data-center space, an announcement is a press release. Historical realization rates for announced (not contracted) capacity routinely fall 40, 60% short of what gets built, because developers shop the same project to multiple utilities and jurisdictions to pressure interconnection timelines, then abandon dupes. The interconnection queue is now so crowded with phantom load that grid planners cannot distinguish signal from noise. Behind-the-meter generation makes the problem invisible: if half these 59 plants never materialize, no one files an amended tariff, no ratepayer complaint emerges, and the grid has absorbed risk for capacity that does not exist.

The immediate policy frame sits at state level. Virginia, Ohio, Oregon, and roughly 20 other states have approved large-load tariff schedules that isolate cost responsibility for loads above 20, 25 MW and demand minimum takes (ratchets of 60, 85%) so underused capacity isn't socialized.^{[background library]} These tariffs are not perfect, but they make hidden costs visible. A hyperscaler choosing to build behind-the-meter is choosing to opt out of that scrutiny. If a state's regulatory model is strong enough, the utility should have the standing to insist that large new loads take service under the tariff or, if they prefer self-generation, post collateral and accept responsibility for grid upgrades triggered by their exit. If the state allows utilities to simply lose customers and socialize the stranded infrastructure cost, the incentive to build private plants will keep rising. The fix is not a ban on self-generation; it is cost discipline: name the grid cost of the exit, and assign it to the entity causing it.