Castor’s subcommittee just made residential ratepayers underwrite the AI data center buildout. The Ratepayer Protection Act, which moved through a House subcommittee in mid-June on a bipartisan vote and is co-sponsored by Representative Kathy Castor, frames itself as a measure to prevent “wealthy corporations” — Castor’s word — from shifting their growing energy demands onto residential customers. The bill’s operative provisions, which are the only provisions that bind, are largely voluntary. The provisions that are not voluntary benefit the data-center developers. This is the structure of the bill. The structure of the bill is also the structure of the cost shift the bill is named to address.
The bill contains language about large-load tariffs — the mechanism by which a state utility commission can require a new data center to pay for the grid upgrades its connection triggers, instead of letting those costs be socialized across the existing rate base. The language is, in the precise sense, advisory. State public utility commissions, which are the bodies that actually set electric rates, can ignore it. Jim Walsh, the policy director at Food and Water Watch, who has been working the data-center-cost question longer than most of the current subcommittee has been in office, points out that the same state commissions are, by the critics’ own account, already broadly captured by the utilities and large-load customers they regulate. To write a bill whose protection consists of suggestions to a regulator that the bill’s own critics describe as already leaning the other way is to write a bill whose protection is, in the precise sense, advisory. The bill does what its title says. It is the ratepayers, in this case, who are being advised.
It is true, and the bill’s sponsors are at pains to say it, that nothing in the bill forbids the state commissions from doing the right thing. It is also true that the bill, in the same operative provisions, contains benefits for the data-center developers that are not advisory. Walsh identifies three. There is streamlined siting and grid-connection priority for new data centers, which is a direct subsidy to the developer’s timeline. There is a loosening of the review required under the National Environmental Policy Act — the 1970 federal statute that requires environmental impact statements for major federally funded projects — for the transmission lines that connect the new centers to the grid. And there is, in Walsh’s reading, a loophole allowing data-center operators to claim they are “self-supplying” their power, by building or contracting their own adjacent generation, in a way that exempts them from the cost-allocation rules the rest of the bill’s voluntary provisions are nominally nudging toward. The asymmetry is structural. The bill’s costs to developers are written into statute. The bill’s protections for ratepayers are written as recommendations to a regulator the bill’s own critics describe as already captured. The bill’s structure is the cost shift it is named to address. This is what is being voted on.
About two hundred new data centers have come online in the United States over the past three years, most of them to house infrastructure for artificial intelligence — the model-training compute, the inference servers, the chip stacks and cooling systems and power-distribution apparatus that the current generation of large language models requires. A single large data center can draw as much power as the largest U.S. cities. The Federal Reserve has documented the effect on wholesale electricity prices: an average increase of up to six percent in regions with high data-center concentration, with increases running as high as fifty percent in some local markets. Regions with higher numbers of data centers have seen electricity costs rise 267 percent over the past five years, on the figures cited in the markup. These are not contested numbers. They are the figures the bill’s supporters cite when arguing for the bill. The figures the bill’s operative provisions address, by Walsh’s reading, are largely not the figures driving the price increases.
The Federal Reserve’s six-percent figure describes the wholesale price effect — the cost the utility pays for the power, before it is allocated to retail customers under the state commission’s rate design. The bill’s voluntary large-load-tariff language lives at the retail-allocation step, which is downstream of the wholesale price. The bill is aimed at the second of two questions. The first question — who pays the wholesale cost of the power the data center draws — is the question the bill does not touch. The reason it does not touch it is that the answer to the first question, in the current arrangement, is partly the ratepayers, by a cost-allocation choice the state commissions have made and that the bill declines to disturb. The new data-center load gets priced at the utility’s average embedded cost, not at the marginal cost of serving the new load, which is the cost of the transmission upgrade, the fuel infrastructure, and the marginal generation the new load requires. The difference between the average embedded cost and the marginal cost is the difference the ratepayer pays. The bill, by leaving the wholesale-cost question alone, leaves the difference alone. The bill’s voluntary retail-allocation language is, in this light, a covering note on a wholesale-cost decision the bill declines to revisit.
A real ratepayer protection bill would, at minimum, do four things the Ratepayer Protection Act does not do. It would make large-load tariffs mandatory at the state level, rather than advisory, and would tie the tariff’s cost basis to the marginal cost of serving the new load — including the cost of the transmission upgrades the new load triggers, the cost of the fuel infrastructure the new load requires, and the cost of the water the new load consumes, because the most consequential costs of a new data center are not, in practice, the electricity bill the operator pays. It would foreclose the self-supply loophole, under which an operator can stand up an adjacent gas plant, contract to buy its own output back, and route the rest of its load through the utility’s grid on the cheaper embedded-cost basis, in a way that exempts the operator from the very allocation rules the rest of the bill is nominally nudging toward. It would either retain the National Environmental Policy Act review the bill loosens or replace it with a sector-specific environmental review regime, for the reason that most new data centers use PFAS compounds — a class of long-chain fluorinated chemicals, sometimes called “forever chemicals” because they do not break down in the environment or in the human body — for cooling, in a way that almost certainly contaminates the surrounding watershed. Georgia regulators, in one case Walsh cites, initially did not charge a data center for thirty million gallons of water use. The bill does not address this. The bill does not address the PFAS. The bill does not address the water. The bill does not, in the operative sense, address very much. And a real ratepayer protection bill would, fourth, ask the question the ratepayer-protection framing is designed to make it impossible to ask.
That question is whether the buildout is happening at a pace the country has decided to accept, or at a pace the federal government is now choosing to subsidize. The bill’s premise, in the words of Camden Weber at the Center for Biological Diversity, is that “datacenter buildout is inevitable.” The bill’s protections, such as they are, are designed to cushion the buildout, not to slow it. The choice to build at the current pace is a choice. It is a choice the federal government is making through the tax credits, through the accelerated grid interconnection, through the National Environmental Policy Act rollback the bill embeds, and through the unspoken guarantee that, whatever the data center’s nominal tariff, the cost of the marginal kilowatt will be shared across the rate base. The buildout is not a weather pattern. The buildout is a federal policy choice. The bill’s premise — that the policy is fixed and the only remaining question is who pays for it — is itself a policy choice. Food and Water Watch and the Center for Biological Diversity have called for a moratorium on new AI data center projects. They may or may not be right. The first thing to notice is that the question of whether the buildout should be slowed, rather than subsidized, is the question the bill’s structure makes it impossible for the subcommittee to ask.
The bill’s sponsors are not wrong that the data centers’ power costs should not be shifted onto ratepayers. The sponsors are wrong, in the precise way, about which bill accomplishes that. A bill that addresses the wholesale-cost question, forecloses the self-supply loophole, retains the environmental review, treats the buildout as a policy choice rather than a force of nature, and gives the state commissions a mandatory cost-allocation rule rather than a set of suggestions to a regulator the bill’s own critics describe as already leaning the developer’s way, would protect ratepayers. The bill that advanced out of subcommittee in mid-June does none of those things. It advances, instead, on a press release. The mandatory provisions are the developers’ benefits. The voluntary provisions are everyone else’s protection. Both are written down. The committee chose which is which. Deadlines on the full committee markup are tight. The deadline that matters more is the one for the next time the utility commission files a rate case and a residential customer opens the bill to find the number higher than it was last year, and wonders, briefly, what the Ratepayer Protection Act has been doing. The answer will be in the bill. The answer is in the bill.