How Will Data Centers Pay for Power?

The American electric power sector has not grown appreciably for twenty years. To be sure, consumers pay plenty to replace infrastructure, to “transition” the sector away from the most carbon-intensive sources of energy, and to find ways to allow utilities to cram a wide variety of underutilized capital spending (think “smart meters”) into their regulated “rate base.” But demand has been stable or declining.

To the degree that profits in the sector have grown, it is because the economic regulation of the utility industry provides for a “spend more, make more” ecosystem whose profits are a function of its capital investment. The sector is still one of the few that is actively regulated through government price-setting, even in places sometimes mistakenly termed “deregulated.” So, even when demand is not rising, the business must find ways to grow earnings by spending more to serve the same level of demand. This has meant that rates, which otherwise might be declining, have been at best steady, even without increasing grid capacity, and in fact, much utility spending has been undertaken to retire reliable capacity.

This was the uninspiring landscape of American electric utilities on the eve of the boom in data centers needed to fuel the technological revolution in AI. Electricity demand forecasts are now sharply up for the balance of this decade, and a majority of this growth is concentrated in data center power needs, with growth in manufacturing a distant runner-up.¹ For a power system that serves as a basis for Americans’ everyday lives and the economy writ large, it is unusual to see such a concentration amid one particular sector for its growth.

The rising power demand of the data center industry almost appears like an industry running within the integrated grid but outside the usual paradigm of the traditional electric utility sector. Indeed, it should be treated as such. Doing so calls for a variety of policy solutions that accurately price grid capacity in order to facilitate efficient usage of that scarce asset, impose regulatory requirements to furnish power generation to the system, and in the alternative, allow power demand that is more flexible to better use residual capacity. Such reforms can accomplish two important policy aims simultaneously. First, they would insulate legacy customers who have already paid their fair share and then some for the grid. Second, they would allow power industry growth in support of data centers to be unchained from traditional utility practices, which often do not reward speed or innovation.

Both of these aims, customer protection and growth, are embodied in the Ratepayer Protection Pledge, a March 2026 declaration at the White House undertaken jointly by the Trump administration and seven major hyperscaler AI companies. The 485-word document begins with an endorsement of data center infrastructure as “the foundation of the internet, cloud computing, and artificial intelligence (AI),” noting the national security implications of this. But it qualifies that “the American people should not be footing the bill for the benefit of private companies.”²

The central proposals of the Pledge are that AI companies “will pay for all new power delivery infrastructure upgrades required to service their data centers” and “will bring, build, or buy the new power generation resources and electricity needed to satisfy their new energy demands.” By directly incurring these costs, the “companies agree to protect American consumers from price hikes due to data center energy and infrastructure requirements, and lower electricity costs for consumers in the long term.” These ambitions are easier to proclaim than accomplish.

Power grids are characterized by joint costs: poles and wires, transformers, and substations that together form a network. Both the typical practice and the financial incentives of most local utility monopolies militate toward a broad socialization of costs to consumers. They do this by having rates set by utility commissions on the utility’s average, embedded costs, rather than pricing based on marginal costs or on a new customer’s willingness to pay. The Pledge wisely points the way toward value-based pricing for grid access that recovers at least the incremental cost of serving customers. This seemingly mundane change, if well implemented in an open season where data centers vie for grid access, is capable of not just protecting legacy ratepayers, but producing massive investment in the American power grid.

Meanwhile, for the power plants that generate electricity for data center consumption, the proposition that the AI industry furnish its own supply is both straightforward and, sadly, unlawful in a majority of states, which maintain local monopolies that prevent this. In these places, many proposals purporting to fulfill the Pledge fall well short of the mark. But this is not to say it is an easy story of letting the market go to work. Even in those regions where competition has been introduced to the sector, investment has been slow to materialize. The Pledge suggests clearing away barriers to new power generation, but with a corresponding regulatory mandate to match the Pledge’s ambition that AI companies bring their own generation to the grid.

The purpose of this essay is threefold: to examine the broader economic and institutional context that the Pledge must address; to put some meat on the bones of its spare but purposeful declarations; and also to take aim at some bad ideas masquerading as fulfillments of the Pledge’s ambitions. Since the two sides of the industry—price-regulated grid costs and more commoditized power generation—operate on so different a basis today, it is best to consider them in turn, but with reforms that ultimately come together, like the grid itself, in sound operation.

The Economics of Regulation

Amid a framework of economic regulation that exists for the electric power industry and very few others, utility commissioners at the state and federal levels fix prices based on a utility’s “cost of service.” This form of price regulation allows the utility to recover both its invested capital and a regulated return on that capital, while generally passing operating expenses through to customers without any markup.

The math that results from cost-of-service regulation is, at its core, one big division problem. The numerator is a sum of the utility’s costs; the denominator, the volume of services the utility sells; the quotient is the rate you pay. Pricing in competitive markets settles around the cost to serve marginal demand, at least according to the basic principles of microeconomic theory. Utility pricing, however, is principally concerned with recovering the sunk costs of infrastructure, which usually serve to flatten and socialize the volatile tendencies that would be expressed in a competitive, commoditized market. “Notice how, at once, the traditional practices of public utility price regulation diverge from economic principles,” the economist and utility regulator Alfred Kahn once dryly observed of the difference between marginal-cost and average-embedded-cost pricing.³

Kahn’s ironic observation has great import today. This divergence between competitive and utility pricing has substantial implications. Consider what happens when incremental demand manifests in price-regulated utility service. In the division problem, if the numerator (costs) rises more slowly than the denominator (demand), then all other customers’ rates would decline as a result of adding a new customer to the grid when utility commissioners next reset utility rates.

There are many examples of this happy phenomenon in the utility sector, beginning in its 1920s heyday, where investment and sales volumes soared, even while rates fell.⁴ More recently, unassuming North Dakota emerges as the winner of the demand growth Olympics in a magisterial study conducted by Lawrence Berkeley National Laboratory and Brattle Group that evaluated retail electricity rates from 2019 to 2024; the state simultaneously notched the highest percentage demand growth and the steepest percentage reduction in retail electricity prices.⁵ New Mexico and Nebraska are in much the same situation.

Some have taken these historical occurrences to stand for a general principle that a rising tide of demand lifts all boats. Would that were so. The early industry’s victories on economies of scale have long been priced in. Indeed, for decades now, the sector’s new classes of capital assets have been trending smaller, predicated on diversifying risk and modular, nimble deployment. When studied closely, these recent successes are idiosyncratic demonstrations of the ingredients one would need to make a return to those halcyon days a reality. North Dakota, for example, had residual grid capacity remaindered from previous oil booms, and on the commodity side, cheap fuels and a surplus of power generation stimulated by federal tax incentives pointed at renewables. In both situations, the marginal cost to serve was lower than the average, embedded cost rate, and this supported what was, in effect, a subsidy from newcomers to legacy customers: the type of subsidy everyone cheers.

These happy conditions no longer obtain. In the circumstances that have coalesced lately, a grid with little residual capacity during peak demand conditions means that a lot of new, uninterruptible demand for power placed upon it necessitates capital spending to expand the grid. The materials on which such an expansion is predicated have inflated in price rapidly. Wires and cables, transformers, switchgear, and wood poles have inflated 152 percent, 89 percent, 77 percent, and 50 percent, respectively, since the beginning of 2019, while the overall consumer price index recorded only 29 percent cumulative inflation in the same period.⁶

The cost of financing these capital assets has also become more expensive. Although utilities have earned generous returns on their equity investment, their debt costs have only captured a modest premium over Treasuries, with many utility customers paying rates that reflect historical debt costs in the 3–4 percent range. With Treasuries well above that today, electric utility issuances of ten- and thirty-year debt so far in 2026 have approached 5 percent and 6 percent, respectively. This double whammy of materials’ price inflation and higher capital costs means that nearly every megawatt of demand added to an American utility will incur costs that exceed the embedded, average cost to serve the same unit. Under such circumstances, if new customers are brought online paying the same rates as legacy customers pay, it will axiomatically result in a cost shift from legacy customers to new customers: the type of subsidy no one can stomach.

In the normal operations of utility regulation, that is usually what would happen. Utilities typically have a legal obligation, in exchange for their monopoly, to serve new customers under their prevailing rates. New data center customers would be classified into existing “rate classes” and begin paying the same rates as, say, a paper mill or chemicals refiner.⁷

For much of the past several years, the data center industry’s hill to die on at public utility commissions was an insistence that they should not be treated in fundamentally different ways than other customers. It is hard to think of a more arcane subject for outsiders to the craft of utility regulation than the procedures by which customers are separated into rate classes. But to long-time practitioners, this debate raised the question of whether regulators were going to labor under the premise that data centers were just another category of customer that fit within the extant practices of utility ratemaking.

A Utility for Data Centers

The end of this fiction has not been formally acknowledged, but it has essentially arrived in the form of the Ratepayer Protection Pledge. The projected growth in data center demand is beyond anything (short of wartime industries) ever asked of the American power sector. There is an obvious need to distinguish between legacy loads and new data center loads, if only because the projected demand not only exceeds existing system capacity but is beyond anything the existing power sector could reasonably achieve.

This is not mere pessimism. There is simply no conceivable path, for instance, by which Texas’s ercot, the state’s large intrastate grid and power market, could grow from today’s levels, a little less than 90 gigawatts of peak demand, to 278 gigawatts by the end of the decade. That growth, of which 149 gigawatts can be attributed to data centers who have requested interconnection, would require the sector to add the equivalent of three Californias to Texas in less than five years.⁸ Indeed, it is deeply unlikely that it could realistically achieve the market operator’s “more reasonable” forecast of a total market-wide demand of 145 gigawatts by 2031.⁹

Thus, unlike other existing and new customers who might reasonably be accommodated in the normal course of utility business, there is going to have to be some kind of rationing to serve data center growth. Even those who doubt that AI will produce such dramatic surges in electricity demand should support good policy. A sound approach can test such claims and contain the risks if demand surges prove exaggerated, just as readily as it can support growth if the demand materializes.

The conventions of cost-of-service regulation are poorly suited to the task. Data centers should be essentially treated as a parallel industry, subject to an entirely different regulatory practice. Such an approach will be necessary in order to identify the data centers that have the highest value and greatest ability to efficiently utilize grid service, to facilitate their interconnection to a sometimes slow-moving utility sector, and to shield legacy customers from the adverse economic and reliability consequences that arise when large new customers plunge abruptly into the grid. A policy that accomplishes these tasks would be superior to the current regulatory practice, which fails on all three counts.

To be clear, a new regulatory arrangement should not be disadvantageous to data centers, legacy customers, or to utilities. Data centers should reasonably expect a “get” for their “give.” In the policies thus far proposed for their regulation, data centers are practically expected to cough up tribute to politicians and regulated utilities on a basis expected of no other customer, but that has not meant faster service, higher reliability, or more flexible arrangements for them in turn. We may consider three reforms to make data centers pay a fair share of their grid costs, while also creating a kind of regulatory funnel to rationalize and expedite access to the system for very large new users of electricity.

An Open Season

The first—and arguably best—reform idea in this vein is to conduct recurring open seasons that align grid capacity with those willing to pay for it. A well-designed open season will maximize the usefulness of grid expansions that do occur, while conferring clear rights to grid access that better define the prospects of a highly speculative industry.

The current processes by which data centers, and for that matter large power generators, are brought online are governed by a kind of first-come, first-served rationale. You file an interconnection request, and then you take a place in line, waiting to see how it plays out in a utility engineering process to get an answer on when you might be able to interconnect and at what ramp toward full production. The utility modeling that governs this process is not entirely a black box. It is sufficiently vague, however, and more to the point, interdependent on the other requests waiting in the line, so much so that this information asymmetry propels developers seeking grid access to hedge their bets by staking multiple claims. This speculation, in turn, produces a kind of “gold rush” on grid access, resulting in today’s bureaucratic snarl and the kind of inflated demand growth projections we are now seeing compound upon the grid.¹⁰ Even when one tries to group these claims together in “clusters” or “batches,” as some utilities have done, problems having to do with the “first-in-time, first-in-right” assumption linger. This is because such claims produce a set of common costs, which then need to be reallocated (to the disadvantage of other projects), if one claimant drops out after getting a number it does not like.

By contrast, an open season squarely acknowledges the reality that the current grid is not some abundant Lockean commons where users may simply claim rights by being the first to apply some labor to the pasture, vein, or groundwater. Instead, open seasons are a tool to allocate scarce capacity in the best possible way. This is appropriate not only because the current grid capacity is inadequate but also because even incremental grid capacity that might be built out likely underserves plausible demand from the data center sector.

At their core, open seasons are characterized by an early give-and-take of those expressing interest from the demand side and those who would build the infrastructure to serve it. Yet the end result is a plan with the financially binding commitment of a set of infrastructure users that justify the investment. Gas pipeline projects use open seasons in this way to determine the diameter of their pipes, the locations of compressor stations, and the new regions they wish to serve. Because the open season iterates periodically, experience in one can produce learnings that inform successive processes in a manner that right-sizes the grid relative to demand. An open season in the electric power sector might work along the following lines:

(1) The transmission provider develops a plan that would offer to interconnect data centers, based on informal expressions of interest it has obtained, to any incremental capacity it can reasonably build within a defined period to accommodate that demand.¹¹ The transmission provider tenders that plan to the data centers bidding.

(2) Parties representing demand submit bids that represent a willingness to make a binding financial commitment to pay for grid access in a zone set out within the plan.¹² Successful bidders would obtain a defined right to make a “firm” withdrawal of a maximum number of megawatts from the system.¹³ Regulators may determine whether more than one round of bidding should occur to increase competitive tension in the process.

(3) The open season resolves in favor of the configuration of auctioned demand that returns the highest net present value on a unitized basis of megawatts of end-use demand served. This net present value is a function of the maximum demand able to be served by onsite or grid-delivered energy, the incremental grid cost of serving that demand, the speed at which the demand can be served, the tenor of the offtake commitment, and, of course, the amount of the bid. In other words, the open season is designed to reward both speed and the quantity of computational power unlocked, while minimizing grid costs and maximizing surplus revenues from data centers’ bids.

(4) An open season is declared a success if the revenues it returns exceed the incremental cost in at least one configuration of bids, with any surplus revenues applied against the utility’s revenue requirement to the benefit of other ratepayers. This affordability measure is incidental to the auction’s main purpose, which is to assign a scarce and precious resource, but it acts in a simple and complete way to leave all other ratepayers better off from data centers’ grid access.

(5) Transmission providers iterate open seasons, every one to three years, depending on the scale of data center development interest in their service territory.

A successful open season awards real property rights to winning bidders, which are tradeable or assignable, although these rights, like the rights of those who develop a Lockean commons, would have limited, usufructuary properties that guard against hoarding and require them to be developed for use. This trait acknowledges that the grid is a type of regulated commons devoted to public use. Indeed, privately owned “public service corporations” that provide transmission routinely exercise the power of eminent domain. Given this, it would not do to have rights of access to that infrastructure not also predicated on usufruct. The existence of a tradeable right means that, in the event of a certain user’s default, bankruptcy, or technological obsolescence, the right of use nevertheless remains property that can be reassigned. The importance of this attribute should not be underestimated if we assume, like most technological revolutions, that the bevy of AI firms now in the market will ultimately coalesce into only a few.

An open season achieves a range of goals for both legacy and new consumers, for regulated utilities and other market participants, and for the national interest. For existing ratepayers, an open season mechanically ensures that new demand pays at least its incremental cost, which fulfills the goals of the Ratepayer Protection Pledge. For data centers, an open season provides a clear plan and ultimately a defined right to interconnect to the grid for the purpose of being able to withdraw a defined amount of power at or around a particular location. For utilities, an open season allows an engineering exercise that unfolds according to an orderly and iterative plan to serve demand, which also clarifies the ultimate quantity of demand that may be accommodated, in distinction from the “gold rush” scenario of suboptimal expansion. For power generators, like utilities, an open season clarifies the amount of total demand interconnecting to the grid, which then provides a bogey for new investments in supply, a parallel policy issue discussed below.

Finally, for the national interest, an open season can use regulatory policy to propel the AI revolution by treating the grid not as a constraint but as an engine that unlocks higher volumes of compute in an organized and sustainable manner. In addition to making new capacity available, open seasons reward technological innovations for flexibility in data center operations, since the logic of making awards on a net-present-value basis gives due advantage to those data centers who do not impose demand at peak times when the grid would require additional, time consuming, and costly infrastructure to accommodate it.

Fronting Capital to Cover Costs

Of course, not every utility will have the scale, capacity, or ambition to meet the demand posed by data centers. The United States has hundreds of transmission-owning electric utilities, and some of them will see little data center development in their service territory. In that case, a simplified approach is called for, which can accommodate ad hoc data center development in an efficient manner. Under this approach, the second of the three reform packages proposed, data centers would ask the utility to interconnect to its system; the utility would then deliver an estimate of incremental cost to serve the data center, and the data center would prepay these costs, holding the utility and its other customers harmless.

This sounds deceptively simple, and of course, this straightforward approach has occurred to data center developers, who know enough to flash a wallet. But even when large customers have offered to furnish their own capital, or even to build their own utility service assets to industry specifications before transferring control to the local monopoly, many utilities have vigorously opposed their doing so. These utilities even go so far as to litigate the matter at regulatory commissions throughout the country. The reason is because, under cost-of-service regulation, utilities’ profit is a function of the amount of capital they invest in their system. Utilities, of course, want a monopoly on customers in their service territories, but this is actually an indirect way of stating what they really want, pursuant to the incentives laid down under their economic regulatory model: a monopoly on the opportunity to invest the capital necessary to serve those customers.

This is as clear a perverse incentive of the American style of utility regulation as one could imagine, especially in the present moment. Most other businesses would be tickled pink at their customers’ offer to front capital to their business. Many member-owned electric cooperatives (a not-for-profit model devoid of this perverse incentive) actively seek not to add capital assets to their balance sheet if they are in service to data centers, which these often smaller utilities wisely regard as a risky, large, and distinct group of customers who should pay their own way.

Customer-contributed capital, as a general rule, should serve to expedite service to the customer. Although utilities want to be the ones to spend the capital, they will do so at their pace because of both financial and regulatory hang-ups that attend their business model. On the financial side, utility managers are constrained by the performance metrics they and their peers have presented to Wall Street and the ratings agencies. Those metrics emphasize a balance between long-term capital spending, which generates future profits, and short-term cash flow, which can be weakened by higher capital expenditures. The balance sheet of a single firm may thus gatekeep an entire local economy that depends upon electricity service.

Utilities’ toxic codependency with their regulators is also a problem. Many utilities have stopped being businesses in the sense of taking managerial accountability for their own decisions and look instead to regulatory preapproval for even minor business decisions. This relationship assures the utility of riskless profits but at the expense of being paralyzed by regulatory indecision in big moments. In the current political milieu, whole swathes of the country might be redlined to data center development if the lame managerial sentiment of modern utilities coincides with the loud but meritless opposition to data centers that camps out in utility regulatory proceedings.

When a customer wishes to contribute their own capital as full prepayment for the utility’s own estimate of the cost to serve, this should obviate any real concerns about cost-shifting that otherwise would exist. Beyond this simple approach, regulators and policymakers should encourage modifications to the regulatory model in recognition of the fact that this infrastructure can have joint benefits to the data center as well as to other customers.

One way to do this is to allow the utility (after its costs are prepaid by a data center) to submit a proposal to its regulator that estimates benefits extended by the project to other customers. This might be the case for a capital project, like an electric transmission line, that is oversized relative to the data center’s own needs and has spillover benefits that other customers should pay for. Under this approach, the utility could, over time, partially reimburse the data center as a rebate to the ongoing rates paid; the expense of that rebate would then be recorded in the utility’s own rate base, which other customers, in turn, would begin paying for, including a return to the utility.

Finally, regulators should be mindful that the infrastructure needed to accommodate data centers may not necessarily belong in the utility’s typical corporate structure; they should instead encourage joint ventures and other creative approaches to infrastructure ownership, so long as the infrastructure can be safely and efficiently operated as part of a wider utility system. The regulatory model should be flexible enough to encourage utilities who want to participate actively in serving growth, but it also must not allow utilities to be a gatekeeper to growth in the American economy.

Take-or-Pay Contracts

A third proposal for how data centers should pay for power usage is through the use of a financial contract known as Transmission Security Agreements (TSAs). This “take-or-pay” contract requires would-be customers of the grid to guarantee that they will pay utilities at least a certain amount of revenue over a fixed term, usually ten to fifteen years. The data center incurs this obligation before the utility undertakes significant capital spending and would be in effect regardless of the data center’s usage of the grid at peak times, which typically drives capital spending needs. Utilities in favor of TSAs have generally proposed this revenue guarantee to be calculated on the basis of the average, embedded cost rates then in existence, multiplied by the contracted demand of the data center. This contract is secured by collateral or a parental guarantee.

TSAs accomplish a few things. They help winnow the field to data centers that have a very high probability of being developed, limiting the speculation engendered by the first-in-time queue. To some extent, they derisk utility spending which, if a data center did not materialize, would cause rate increases for other customers or regulatory disallowances for utilities.

The problem, however, is that TSAs do not bear any relationship to incremental cost and so, by definition, fail to vindicate the Ratepayer Protection Pledge. Because the TSA uses the socialized average, embedded cost rate to price the revenue guarantee, a five-hundred-megawatt data center built in a greenfield, requiring $1 billion in capital upgrades, would make the same revenue guarantee as another five-hundred-megawatt data center that redevelops a brownfield, requiring only $100 million in new capital spending. Assuming industry average transmission rates, the former guarantee would not be adequate to cover the incremental cost, and the latter would far exceed it. While these numbers are illustrative, the problem is not hypothetical but a documented reality in one of the first places where this approach is being pioneered.¹⁴

A TSA is the utility industry’s preferred approach to provisioning grid access to data centers, and one can be forgiven for thinking that it does more to protect and further secure the role of incumbent regulated utilities than it does to protect their other customers. The TSA allows utilities to monopolize the capital spending opportunities that exist in the sector and obtain a revenue guarantee from large customers relative to utility service. It also allows utilities to price service at the same rate even to customers that cost more in capital spending to serve. This feature ultimately increases utility profits even while limiting the risk they are exposed to. In those situations where the TSA does not cover the entirety of incremental costs for some data centers, and they then default, the TSA becomes a regulatory shield for utilities. In that eventuality, utilities will surely point to TSA as the structure the regulator itself blessed to supposedly mitigate that risk, which presumably will entitle the utility to recirculate the balance of stranded costs onto the rest of their legacy ratepayers.

In a moment that seems to call out for a new paradigm for utility service, the TSA is an attempt to extend average-cost pricing to a place that is probably beyond its breaking point. This will have a variety of undesirable consequences beyond its failure to fulfill the Ratepayer Protection Pledge, which, once again, calls upon AI companies to shoulder the full cost of the new power delivery infrastructure needed to run their data centers. By failing to send an efficient price signal, TSAs will not do any work to guide where data centers should be sited, even though it would make sense to locate data centers in places with more robust, preexisting electrical infrastructure. They also create an effectual disincentive on data center flexibility at peak times, since a data center will have to pay regardless of whether they use grid power at those times.

State utility commissions, as well as their federal counterpart, nevertheless have started to adopt the TSA approach. Despite being less than ideal policy, it is something that can be advertised as protective of legacy ratepayers in the data center era, and something is better than nothing.

Markets for Power

Turning now to the other side of the electricity industry, the supply of energy that transits through the grid, things are directionally similar: an increasingly tight market that is seeing a boom in demand and which, unlike the elements of the regulated grid, has also seen active retirement of capacity due to environmental regulation, renewables subsidies, and the closure of coal-fired plants that have been unable to weather the former two factors in combination with intense competition. This competition is mainly with natural gas, which was the source of 43 percent of U.S. power production in 2023 and became the dominant source of American electric power in 2016.¹⁵

Even as coal plants closed, a gradually dawning awareness of the interrelated risks of a system dependent on Mother Nature for renewables and gas pipelines for reliability caused grid operators to derate the effective capacity thought to be available from the configuration of these power plants. This has effectively meant less capacity, at least in the judgment of the technocracy that governs the grid, than one might have imagined. And this all might have been fine because—except for Texas, seemingly always a boomtown—the largest electricity markets in the United States last recorded peak demands in the mid-2000s. Now, however, we are returning to such demand levels with the advent of data centers for AI computational uses.

Everywhere, the power plants that we depend upon for reliable service require fuel to operate. Whatever the regulatory arrangement of the power plants downstream of it, the fuel market is a lightly regulated commodity, and everyone faces the same price exposure. In the power markets known as Regional Transmission Organizations (RTO), near-to-real-time and day-ahead auctions for power are centrally administered, forming a spot price which then forms a basis in futures trading handled through third-party arrangements. In the non-RTO markets, it works the same way, except spot trading occurs bilaterally, at a handful of trading hubs at particular locations on the grid and without the transparency of the RTO’s administration. As new demand materializes, power plants with higher production costs are called into service or more frequently relied upon, and it is their marginal cost that sets the clearing price for the electricity markets writ large.

In certain markets—such as PJM Interconnection, which sprawls from Washington, D.C. to Chicago—investments in power generation capacity, the steel in the ground and not just the fuel throughput, are also formed through competitive auctions. The Independent Market Monitor of PJM, an office long led by the cantankerous economist Joe Bowring, estimated that the marginal-price effects of new data center load made up 45 percent of $47.2 billion in total capacity auction costs over the last three years.¹⁶ To be clear, the IMM has no visibility into what end-use customers’ ultimate financial exposure is. Bowring is totaling up all revenues that flow as a result of the centralized spot market’s sell-all, buy-all design, which clears all demand even if it has been economically hedged by a forward contract that will, in effect, rebate revenues to the demand side if it has contracted earlier for a lower price.¹⁷

In markets where electricity service has been demonopolized, customers have the right to enter into fixed-rate, long-term contracts with a supplier of their choice. As a residential customer, I signed a fixed-rate, five-year-long contract in early 2024 at a price well below the market’s current pricing.¹⁸ In my experience, however, most customers tend to buy power like they buy gasoline: at the pump whenever they need it, or in this case, at the meter whenever they use it. While not necessarily exposed to the spot price per se, the product residential and small-business customers default to, if not having made an active selection, gives about as much protection as filling up the gas tank does until you need to pay up again at the prevailing market rate.

In this market, it is large customers with strong balance sheets, such as data centers, who tend to be the most willing to sign up for long-term contracts for power-supply offtake as energy is such an influential input to their business model. As the AI boom accelerated, large nuclear power generators maneuvered to sell contracts for their output to these large customers. Just a few years into this cycle, in view of everyday consumers’ affordability concerns, new large users’ tie-ups of legacy power generation can seem a bit like Daniel Plainview declaring that the milkshake has been drunk, right out from under the nose of people who did not quite know the game they were playing.¹⁹ Whatever legal niceties obtain—no one has legal claim to uncontracted power supply that is available to any comer for purchase on the open market—this is clearly a politically implausible outcome. Nearly everyone in both the hyperscaler and power generation industries now sees this, albeit after a few sips of the milkshake; once again, the Ratepayer Protection Pledge is a recognition of this policy consensus, if not its details.

Some have claimed that the messy dynamic of the competitive power markets vindicates the enduring phenomenon one sees in many American states, which is a utility monopoly whose perimeter encompasses the supply of power and not just the grid’s poles and wires. This remedy is an illusion because “regulation” as a consumer protection measure is an illusion in the current circumstances, for at least two reasons. First, a “regulated” utility’s power generation assets buy natural gas out of a commoditized, lightly regulated upstream market just like any merchant power generator does. You can only vertically integrate so much into a monopoly, and although it once did, the United States abandoned the economic regulation of natural gas and oil two generations ago, and in the process, it liberated the gas industry’s production from the regulated utility segment of the business.

Together with the advent of hydraulic fracturing, this unbundling is why industry has been able to rapidly produce gas fields. Never say never, but price regulation of gas and oil is probably not coming back. In practice, one often sees utilities treating their captive customer base as their insurance policy on fuel purchases out of this market. Memorably, one utility in Oklahoma, facing the price blowout that attended Winter Storm Uri in 2021, spent more on natural gas purchases in a single week than the entire amount of capital it invested in its utility system. The result was not the utility’s bankruptcy, as would happen in the competitive market. Instead, the utility was allowed to pass this cost through to its customers, who, owing to the munificence of Oklahoma regulators, are in the process of paying it back through a roughly $7-per-month surcharge over the next two-odd decades. Public utility regulation cuts both ways, and in bad times, often serves to protect regulated firms, not their consumers.

Next, it is worth recalling the history of why policymakers began to “deregulate” power generation in many places when they did, since it bears at least passing resemblance to the present moment. In the 1970s and ‘80s, materials price inflation coincided with a rising cost of capital. At first, it seemed little to worry about. Demand had seen solid growth ever since the Second World War and a new era of manufacturing and suburbanization ensued. Chasing that growth, utilities funneled these escalating costs through utilities’ average, embedded-cost ratemaking logic, ballooning the numerator of costs. Then, consumer demand flagged and utilities lacked a sufficient denominator to spread those costs over. The resulting vicious cycle of rate hikes and halted power plants sent the entire sector into a tailspin for a generation. A few notable utility bankruptcies took place, but the norm was a series of bailouts undertaken to save utilities perceived as too big to fail from collapse. Key policy takeaways from that miserable era included limiting the surface area of the utility monopoly, and placing the business of investing in power generation within the hands of private developers who would own the associated risks. Whatever criticisms may be lodged against the merchant power sector that emerged in the aftermath, its bankruptcy-to-bailout ratio is a large number, with the risk of many bad or untimely investments shouldered by management.

Fifty years ago, regulated utilities, although ripe for reform, at least were chasing after largely organic demand growth spread across numerous customer classes, which declined as the wider economy did. In the current era, where demand growth is highly concentrated in one sector, the risks are further magnified and the proposition that utilities are some kind of indispensable partner for power generation consumed by data centers is, at best, a tenuous proposition.

Worse, there are some indications on the part of a subset of utilities that, contrary to their usual attitudes as monopolies, they understand that they now must, in some sense, compete for the largest and best hyperscaler loads. Usually, one should cheer the reversal of the lackadaisical stance that arises from monopoly, but since utilities also have customer bases that they don’t have to compete for, there is a risk that legacy customers in all sorts of subtle ways will be used as a sop for the risks inherent in chasing service to data centers. When this does happen, even certain gifted utility regulators may not realize it. Indeed, risky proposals may even be dressed up as a consumer protection measure.

Consider, for example, Dominion Energy Virginia, the utility that has served the largest concentration of data centers in the United States to date, which recently received regulatory approval for a generation-focused version of the Transmission Security Agreement described above. Under this arrangement, new large customers, over twenty-five megawatts in size, must pay average, embedded-cost rates at a level that is at least 60 percent of their maximum size, regardless of their actual usage during peak times, for a term of fourteen years of this take-or-pay contract. This is a bad idea in several different directions and is sadly representative of the way of thinking emerging across the industry, packaging noxious ideas as consumer protection measures, when under close evaluation they are not.

Dominion’s new rate structure shifts costs and risks to existing customers in a subtle way. While a take-or-pay contract covering 60 percent of demand over fourteen years fosters a feeling that money is on the line, those terms are not coextensive with the full cost and lifespan of a new power plant. Dominion is signing itself, or really its other customers, to significant tail risk. More importantly, by premising this rate on what Dominion charges all its customers for power generation capacity, the utility is commingling the legacy investments that Dominion made on behalf of legacy customers with the higher, incremental costs of the new power generation that is being built to serve rising demand, which is to say, data centers. Consequently, in addition to being an inadequate hedge, it is also redistributing the benefits of legacy investments to those who did not pay for them.

In the other direction, ironically, the terms of the take-or-pay are onerous enough to have the pernicious effect of snuffing out competition with Dominion in its service territory. Under Virginia law, large customers may take power supply from a third party, but under Dominion’s new rates, such a customer would essentially pay twice: full freight to the third party for generation capacity, as well as the minimum demand charge to Dominion, even though the utility is not the commodity supplier.

When regulation is your business, you usually emerge with real skills for the craft. And so it is with Dominion, which used the institutions of utility regulation to remonopolize a jurisdiction that had a limited measure of competition, and in doing so, concentrated significant tail risks onto a captive base of customers. The utility sold this package as a combination of consumer protection and economic development: a classic regulatory tactic that has, over time, turned into a veritable business model for a handful of utilities known for this showmanship. Crowing about Dominion, one analyst described it as “a straightforward and attractive Southeastern rate base and sales-growth story,” noting a compound annual growth rate of 9 percent in the capital assets that form the basis of its return, over a five-year trajectory, and the presence of regulatory “adjustment clauses” that allow the passthrough of 75 percent of its costs without normal regulatory procedures.²⁰

“Bring Your Own” as a Requirement, Not a Suggestion

Consider, by contrast, an alternative regulatory policy where data centers are required to fulfill their energy needs only through non-utility suppliers; by the same token, in this scenario, legacy utilities are forbidden from supplying data centers with energy and are instead confined to the already complex business of grid infrastructure, as described in the first part of this essay. Suppliers who are not regulated utilities have no captive customer base. There is nowhere to allocate the risk between them and data centers other than those two parties. And that allocation of risk occurs through terms in an arm’s-length contract between those two sophisticated parties, not a regulatory proceeding in front of a utility commission. An economic regulatory process can always become a conduit for risk to transit to other customers, as in Dominion’s example, regardless of good intentions. Properly understood, this policy of non-utility supply for AI companies, sometimes defined as “Bring Your Own Generation” (BYOG), is the remedy that the Ratepayer Protection Pledge is actually looking for.

The question remains how to ensure that the sector’s investments will actually rise to the moment that demand really requires. Say what you will about regulated utilities, they will build you a pyramid if their regulator allows it; it may simply take a while and come at significant costs to those other than its beneficiaries. But for producers in competition, as Alex Bronzini-Vender has noted in these pages, “capital fears overproduction more than it fears scarcity.”²¹ That observation is especially true in commodity markets that settle on marginal-cost pricing for a set of products that sellers and buyers accept as fungible.

Such dynamics are the rule in markets for power, even if government regulators set bounds for the market’s pricing. The delta between the price ceiling and price floor in PJM’s roughly 160,000-megawatt generation-capacity market is only a few thousand megawatts. No incumbent generator sitting atop a portfolio of legacy power plants wants to commit the double error of outlaying speculative capital to build the new plant that collapses the market price, consequently failing to earn margin on the new investment while also reducing profits on the legacy investments.

Indeed, power executives have seen that movie before. Each of the legacy generators as a corporate entity, or important management personnel within them, lived through prior bankruptcies propelled by these very dynamics. So, if they are to build at all, they have to be assured that new demand will pay for new investments while not impairing legacy investments. Herein lies the allure of BYOG, which, if instituted as a regulatory mandate, clarifies the investment opportunity in generation, especially when paired with one of the aforementioned approaches that makes a clear allocation of rights to grid access. This allows the private capital that BYOG is contingent upon to get over the hump of a data center market whose demand projections are so bullish that they are bearish.

To date, the clearest ringing of the bell on BYOG has occurred in the PJM market. Just as futures prices for generating capacity began to rise, Pennsylvania Governor Josh Shapiro filed a complaint against PJM at the Federal Energy Regulatory Commission, whereupon this sleepy bureaucracy unaccustomed to major political blowback capitulated to a much lower price ceiling for its auction.²² The rationale was that, without a cap, existing power plants—which, like all resources in the PJM auction, are paid the market-clearing marginal price—would receive windfall profits. But the new price cap was also below the cost of entry for a new power plant. This made it a foregone conclusion that new generation would have to be procured outside of the normal market.

The institution of a price cap further chilled the enthusiasm of power generation developers. And in attempting to restore that enthusiasm, the National Energy Dominance Council herded regional governors of both parties, including many likely Democratic presidential aspirants, into a joint “Statement of Principles” that called for an entirely separate auction to occur alongside PJM’s ordinary ones for the purpose of awarding fifteen-year contracts to new sources of generating capacity. These new procurements would be paid for by new data centers, not other customers.²³

This approach stands in contrast with the one that another large market for data centers, Texas’s ercot, has taken. There, policymakers have voiced similar sentiments about BYOG, but those pronouncements have so far lacked concomitant regulatory procedures to impose an affirmative BYOG mandate.

Meanwhile, in more than half of the other states, BYOG is not something a data center of its own volition may even do because of the abiding role of utility monopolies in furnishing this service. In those southeastern, midwestern, and western states, BYOG depends on the local utility’s willingness to accept into its fold a third-party arrangement to accommodate a data center. Usually, this means paying excessive rates to a utility or granting it an ownership interest in the new generation project, with the possibility that the costs and risks are subtly commingled on the balance sheet that underlies the utility’s service to other customers. Properly understood, BYOG-like policies that require investment from a regulated utility are not BYOG at all.

There are a handful of ingredients that need to be in place for BYOG to be a policy that works to fulfill the Ratepayer Protection Pledge. First, in the majority of jurisdictions that have maintained a monopoly on the supply of electricity to end-use customers, state legislatures have a rare opportunity to both promote investment and to protect customers. They could establish a right for data centers in their jurisdictions to contract for their own power supplies. States that have historically maintained monopolies on utility service, including Utah and West Virginia, have recently moved in this direction.

Second, demonopolizing power generation is not a guarantee of producing investment in it. Indeed, competitive exposure can keep people holding their wallets close, as described above. So an essential ingredient to BYOG is to make it not just a right, but a duty. Markets like PJM can establish and actively enforce requirements for a “showing” that those contracted to provide retail service to data centers have long-term, physical assets backing those contracts.

Third, any BYOG policy needs to address its major disadvantage, which is that it can risk failing to make use of the residual capacity that does exist on the system and which is available in most hours of the grid’s life, just not those at peak. Allowing data centers to be exempted from new-build requirements to the extent that they source flexible operations from their own site or from other customers in the vicinity is, therefore, a necessary provision of any efficient regulatory policy on BYOG. So too is some consideration for power-generating resources that exist but which would actually be threatened with retirement if the paradigm of serving new data center demand is one accomplished with exclusively “new” power generation. This legacy investment can act as a bridge to new development. These are messy qualifications that, in a world which treated power generation as the fungible commodity that it used to be, would not be necessary. But politically, the train has already left the station, and one can only hope it is headed somewhere worthwhile.

In short, a sound BYOG policy is one that imposes affirmative obligations on data centers to make available new generation or to make better use of unused parts of the existing grid as a precondition for data centers’ interconnection. By giving energy suppliers some limited optionality in how they serve new demand—by adding capacity or paying to free it up—well-designed BYOG policy can also encourage these intermediating, non-utility suppliers to get creative in an industry badly in need of that trait.

The competitive retailing of power has not taken off as in other industries. But there are other industries, such as telecom, where retailers market joint infrastructure under different models to try to maximize its economic use. Mint Mobile resells T-Mobile capacity and Priceline discounts hotel rooms; a whole industry has been built up around finding space amid the residual capacity of significant fixed cost investments and making sure it is utilized.

Differentiated service in the power sector may ultimately take several forms: flexible demand; demand whose reliability is supplied by on-site backup generation rather than the public grid; or firm demand at a fully subscribed substation where capacity has been freed up by batteries installed in nearby homes and businesses. The way to get all of these things is by having a BYOG model predicated on creative retail suppliers who are on the hunt for these solutions.

Toward a Fairer System

There is no world in which data center demand is hermetically sealed from that of everyday electricity consumers. Even where we assign grid access through an open season, or require data centers to bring their own capacity or to make their own space on the grid, the materials and fuels that supply these needs will be subject to demand originating from both data centers and ordinary consumers. It would require extraordinary price regulation to not make it so, and it would be bizarre to insist on this cordon, given the importance of AI and its own inherent characteristics as a network.

Nevertheless, the above approaches offer a clear way to vindicate the Ratepayer Protection Pledge. They also offer an approach for an industry in data center computational power, a national priority for the United States, to be expedited without disadvantaging ordinary ratepayers. While debates in the United States on the electric power sector often feature a posturing between monopoly and competitive business models, a keen observer may well be tempted to summon a pox on both their houses because neither of them moves with particular alacrity in an exigent time.

The industry, whatever its organizing principles, used to move more in tandem with the national interest. It is time for it to do so again.

This article originally appeared in American Affairs Volume X, Number 2 (Summer 2026): 3–23.

---

Notes