In April, President Donald Trump issued two Defense Production Act determinations covering grid infrastructure and large-scale energy infrastructure. One found that the country’s constrained electric grid posed an increasing threat to national defense and made grid infrastructure and its supply chains essential to national defense. The other found that financing risks, regulatory delays, and market barriers were preventing large-scale energy infrastructure essential to national defense from being provided in a timely manner. Less than three months later, New York Gov. Kathy Hochul imposed a one-year moratorium on new hyperscale data centers. Her order said the pause would keep New Yorkers from paying for transmission and other infrastructure built for large loads that might never fully materialize.

The defense stake is clearest in PJM Interconnection, the regional market at the center of the cost dispute examined below. A March 2026 Center for Strategic and International Studies analysis found that the market contains 55 percent of U.S. titanium capacity, 50 percent of aluminum capacity, 31 percent of steel capacity, and 12 percent of semiconductor fabrication capacity. The same analysis estimates that, under its peer war mobilization scenario, increased production of munitions and drone systems would account for roughly 60 percent of additional defense manufacturing energy demand. The region is simultaneously facing 48.5 gigawatts of planned data center capacity, roughly 30 percent of its 2025 summer peak load. The defense capability at stake is industrial surge: whether the regional grid can support that mobilization while absorbing tens of gigawatts of new commercial load.

Washington has treated the power needed for artificial intelligence mainly as a physical buildout problem. Those constraints are real. But another constraint is forming around who carries the cost and risk of demand that may not arrive. In March, major technology companies signed the White House’s Ratepayer Protection Pledge, promising to build, bring, or buy the power they need and pay the full cost of the infrastructure required to serve them. The promise crosses a jurisdictional line. Federal regulators and regional grid operators govern the wholesale side of the system, while states largely oversee retail rates and local distribution. No institution has authority spanning both the wholesale treatment of uncertain demand and the retail allocation of its costs. Each can address the part it controls without settling who carries the remaining risk.

How Forecasts Become Costs

The regional grid operator runs a capacity market for a region serving more than 67 million people across 13 states and the District of Columbia. The market secures power resources up to three years in advance. A data center does not need to be operating today to affect what the region prepares to buy for a future delivery year. One reason those forecasts can overshoot is that developers may shop the same prospective project among different utilities or connection points before deciding where to build. The Federal Energy Regulatory Commission has warned that these duplicative requests can inflate expected load growth and lead to double counting. The grid operator now screens large load requests for their likelihood of proceeding and the risk that the same demand appears more than once.

In its analysis of the 2027/2028 auction, the regional market’s independent market monitor calculated actual capacity market revenue of about $16.4 billion. Its counterfactual found that removing existing and forecast data-center load from the peak forecast would reduce modeled revenue by about $6.5 billion, or 39.6 percent. About $6.2 billion of that modeled reduction came from forecast new load rather than data centers already operating. The estimates are the monitor’s, not figures adopted by the grid operator.

The auction also cleared at the $333.44-per-megawatt-day ceiling of a commission-approved price collar. The collar constrained both ends of the exercise: Its ceiling limited the actual clearing price, while its minimum price limited how far modeled prices could fall when data center load was removed. The market monitor says that the floor significantly reduced the calculated impact of data center load. The market monitor separately estimated that, without the restricted demand curve and holding everything else constant, total auction revenue would have been about $26.3 billion rather than $16.4 billion. The data center counterfactual therefore needs to be read narrowly. It does not mean data centers alone caused the price, and it largely holds supply in place. If generation had expanded more quickly, the same demand would have faced more available supply. The grid operator itself says demand growth is outpacing new supply. The $6.5 billion estimate shows the scale of exposure created when planners act on large loads that have not yet materialized. It is not a measure of what an unconstrained market would necessarily have charged.

Who Owns the Risk

The price collar itself shows how that jurisdictional split works in practice. In December 2024, Pennsylvania Gov. Josh Shapiro filed a complaint with the Federal Energy Regulatory Commission challenging the regional market’s capacity-market rules, arguing that the coming auctions could impose billions of dollars in unnecessary costs on consumers. The dispute produced a January 2025 settlement that lowered the auction’s price ceiling, which federal regulators approved in April. The collar was later extended, and Shapiro’s administration now attributes roughly $45 billion in cumulative consumer savings to its interventions in the regional capacity market. But the tradeoff runs in the other direction as well. The grid operator describes its capacity market as providing long-term price signals needed to attract investment in power supply and has argued that the temporary price collar should balance consumer costs against the need to preserve incentives for new generation. Pennsylvania therefore reached into the federal wholesale layer to protect its retail customers, while leaving unresolved how much of that price signal the region needs to attract a new generation.

Monitoring Analytics President Joseph Bowring wants much of that risk separated from the ordinary capacity auction. The market monitor has proposed a dedicated mechanism for new large data center load that would directly link that demand with new generation rather than leave the cost and risk with other customers. The grid operator has now moved partway in the same direction. On Aug. 13, 2026, it filed a new large load framework for customers that bring their own power and an interim service for those that do not. Beginning with the 2029/2030 capacity auction, the proposal would exclude new large loads without new supply behind them when calculating the future power needs to be procured in the auction.

The regional grid operator can decide how unsupported load enters a regional capacity calculation. It cannot write the retail service contract between a utility and a data center or decide how a state allocates local distribution costs. The same project can therefore be treated one way in a regional market and another way by the state that has to explain the bill.

Ohio offers a direct test of whether financial commitments separate firm demand from expressions of interest. Before American Electric Power Ohio’s Data Center Tariff took effect, developers had submitted requests totaling more than 30,000 megawatts. Once American Electric Power Ohio required customers to pay for formal load studies, about 13,000 megawatts remained. After the utility presented service plans, collateral requirements, and binding agreements, data centers signed contracts totaling 5,642 megawatts under the new tariff. Another 12,219 megawatts of load was already under contract prior to the tariff. The process turned a broad expression of interest into a much smaller set of new projects willing to put money behind their forecasts.

The tariff did not settle the wholesale forecast question. The Ohio Manufacturers’ Association argues that the utility’s tighter customer commitments did not produce a corresponding reduction in the load forecast sent to the regional grid operator. The utility says it is updating its forecasts to report only load backed by signed contracts. The dispute captures the jurisdictional gap: A state tariff can discipline retail commitments without itself determining how the regional market counts future demand.

There are reasonable objections to treating this friction as a policy failure. The Federal Power Act deliberately preserves state authority over retail sales and local distribution, so some of the conflict is federalism working as designed. Nor does Ohio’s filtering mean the remaining demand is speculative: Many projects have made binding commitments. And the manufacturers’ challenge argues that singling out data centers is discriminatory and sends the message that some businesses are unwelcome. The problem is not that every forecast is wrong, but that different institutions can apply different tests of how firm it is and assign the remaining risk differently.

When Costs Become Delays

By August, the dispute had become an electoral problem. A private campaign memo from the Senate Republican campaign arm warned that opposition to data centers was hurting Sen. Jon Husted in Ohio and could spread to other races. It urged the industry to explain who benefits, who pays, and why communities should want these projects. Energy planners count megawatts, but campaign staff count votes. Communication cannot allocate the cost of a transmission project, decide what a data center should commit before it enters a forecast, or determine who is left with the bill if it disappears.

That pressure also shapes what governors and state regulators can accept. A cost arrangement that works inside a regional market may still be difficult to defend if voters believe households are carrying the risk.

Pennsylvania’s August order was not its first intervention. A November 2025 law requires the state Public Utility Commission to review and validate the load forecasts Pennsylvania utilities submit to the regional grid operator, giving a state regulator oversight of inputs used in the regional capacity market. Nine months later, Shapiro moved from forecasts to permits: His Aug. 18 executive order required data center developers to make binding commitments to cover the infrastructure their projects require and removed data centers from the state’s fast-track permitting program.

The White House’s April large-scale energy determination cites regulatory delays and market barriers as reasons needed infrastructure is not arriving quickly enough. Its March AI policy framework calls for faster permitting while also saying residential ratepayers should not bear higher electricity costs from new data center construction.

Those goals can collide long before a transformer is installed. Generation, transmission, equipment, and supply chains remain real constraints, but unresolved cost disputes can add delays of their own. The federal government can accelerate one part of the process while states slow another because the financial risk has not been settled. That is the political constraint: An unresolved argument over who pays when Washington wants speed and the demand behind the buildout is still uncertain.

What Washington Should Do

The Federal Energy Regulatory Commission’s June proceedings are a necessary start. The commission ordered all six regional grid operators under its jurisdiction to justify or reform rules governing large loads, including protections against cost shifting and greater transparency. But the problem identified here crosses the federal-state boundary. The Commission should use Section 209 of the Federal Power Act to bring regulators from affected states into a formal process on large load forecasting and the interaction between wholesale rules and state cost allocation.

That process should not set state retail rates or impose a national market design. Its purpose would be narrower: make the wholesale and retail sides answer the same questions. When is a proposed load firm enough to enter a regional forecast? What financial commitment stands behind it? Who carries the cost when it does not materialize? The commission has used that machinery recently. In 2021, it established a Joint Federal-State Task Force on Electric Transmission under Section 209(b), bringing federal energy commissioners and 10 state regulators together to address transmission planning, shared regulatory authority, and how infrastructure costs should be allocated. If some states decline to participate in a large load process, the commission should continue its existing proceedings within its own jurisdiction while leaving the joint process open to states willing to coordinate on the retail side.

The process also needs a common factual record. In March, Sens. Elizabeth Warren and Josh Hawley asked the U.S. Energy Information Administration for mandatory reporting on data center electricity use, rates, deposits, infrastructure costs, and how those costs are distributed. In April, the agency said it planned a mandatory national survey after completing its pilot studies. That survey should distinguish binding commitments from prospective demand and track what happens to infrastructure costs when projects are delayed, reduced, or canceled. That record would also give regulators and Congress a way to test whether companies are bearing the infrastructure costs they pledged to cover under the Ratepayer Protection Pledge. Better data will not erase the jurisdictional line. It can make it harder for forecast risk to disappear across it.

The United States still needs more power. Building it faster also requires federal and state institutions to reconcile how uncertain demand enters regional planning and who bears its costs. Without that coordination, states will keep answering the question through tariffs, lawsuits, permitting conditions, and pauses even as Washington treats faster energy infrastructure as a national security priority.

Javaid Iqbal Sofi is an AI governance researcher and policy consultant who has advised international organizations on regulatory frameworks. He can be reached at [email protected].

Image: Rsparks3 via Wikimedia Commons