The defining fight in the American electric-power sector is no longer whether the grid can be built fast enough to serve the surge in artificial-intelligence computing. It is who will pay for it. That question moved from abstraction to concrete regulatory consequence on September 29, 2026, when the Federal Energy Regulatory Commission largely accepted—but then suspended until February 28, 2027—PJM Interconnection’s one-time Reliability Backstop Procurement. The order halted an auction that was scheduled to begin the next day and left the nation’s largest grid operator to rewrite a mechanism it had spent months designing. In doing so, FERC crystallized a tension building across every regional market: the need to procure firm capacity quickly enough to keep the lights on for roughly 67 million people, set against the imperative that the staggering costs of that capacity fall on the customers driving demand rather than on households already straining under rising bills. PJM’s reliability backstop procurement has become the clearest test yet of whether regulators can reconcile speed with fairness. FERC’s answer this week was that PJM had not yet struck the right balance.
A Backstop Born of Scarcity
To understand why PJM reached for an extraordinary tool, consider the arithmetic behind it. The grid operator’s most recent base capacity auction, for the 2028/2029 delivery year, fell 6,831 megawatts short of its reliability requirement—the quantity of resources PJM needs to maintain a standard of no more than one loss-of-load event every ten years. That shortfall was not the result of unusual weather or a one-off modeling quirk. It reflected a structural collision between rapidly rising demand and declining supply. PJM now forecasts that large loads, overwhelmingly data centers, could add as much as 70 gigawatts of demand by 2038, even as roughly 15 gigawatts of generation has retired across its footprint since 2022. After two consecutive capacity auctions failed to secure enough supply to meet reserve requirements, the PJM Board of Managers concluded in late July that the situation required decisive intervention rather than waiting for the ordinary three-year-forward market to catch up.
The result was the Reliability Backstop Procurement, a one-time auction designed to fill the 6.8-gigawatt gap and meet longer-term forecast needs through commitments of up to fifteen years. PJM paired the central procurement with a bilateral matchmaking process, launched in June, intended to connect data-center developers directly with power suppliers so that privately contracted megawatts could reduce the amount the auction itself had to procure. The design was ambitious in both scale and price. PJM proposed capping accepted offers at $555 per megawatt-day, a sharp increase from the $325-per-megawatt-day cap that governed the last base auction and a reflection of how expensive new firm capacity has become in a market where gas turbines carry multiyear delivery backlogs. The Natural Resources Defense Council estimated that the program could commit PJM customers to as much as $20 billion for new plants built to serve data centers coming online through 2027. Those figures alone guaranteed intense scrutiny because every dollar committed through the auction would ultimately flow to someone’s electricity bill.
FERC Draws a Line
When PJM’s filing reached FERC in Docket No. ER26-3380-000, the Commission did not reject the concept of a backstop, but it refused to approve the version before it. FERC found eleven of the proposal’s fifteen key elements just and reasonable, including the $555-per-megawatt-day offer cap and the collateral requirements imposed on power suppliers. What it would not accept was the architecture that mattered most to consumers. The Commission concluded that PJM had failed to demonstrate that its cost-allocation plan was just and reasonable or that costs would be assigned to the parties causing them. It also rejected the transmission-owner exit provisions, the collateral requirements proposed for load-serving entities, and a carve-out that would have allowed certain cooperatives and municipal utilities to opt out of the procurement—a feature FERC found discriminatory toward other load-serving entities that also host data centers.
The order’s tone was unusually pointed. In a concurring statement, Chairman Laura Swett wrote that PJM’s filing had arrived at the last minute and left the Commission no time to, in her words, “rehabilitate the mess” it received. She added that FERC would not be forced to accept a deeply flawed, eleventh-hour procurement mechanism with major implications for consumers. Commissioner Lindsay See emphasized that regulators and grid operators alike need far better data on where and why load is growing—information that bears directly on who should bear the resulting costs. Rather than terminate the program, FERC suspended it until the end of February 2027 and offered PJM a path forward: use updated load and resource forecasts to set the initial procurement target; allocate costs according to those forecasts; and, if a transmission owner exits before costs are recovered, assign the stranded costs to load-serving entities within that owner’s transmission zone. The five-month pause, in other words, was not a dead end but a demand for a better answer to the problem’s hardest question.
The Cost-Allocation Crucible
Nothing in the proceeding carries higher stakes than the question of data-center cost allocation because it will determine whether the AI build-out is financed by the companies profiting from it or subsidized by ordinary ratepayers. PJM has maintained that costs should flow to the customers driving them, yet it has repeatedly said it is premature to quantify how the burden would ultimately be divided between existing customers and new large loads, arguing that states must determine which retail loads bear the costs. That ambiguity is precisely what alarmed consumer advocates and drew FERC’s scrutiny. The Maryland Office of People’s Counsel, which represents residential utility customers, estimated that the proposed procurement could impose up to $562 million in costs on Baltimore Gas and Electric and Potomac Edison customers alone. People’s Counsel David Lapp welcomed FERC’s finding that PJM’s allocation method was not just and reasonable. He cautioned, however, that the Commission’s eventual substitute could prove worse for customers than PJM’s original design and expressed disappointment that FERC approved a price cap for what is effectively a fifteen-year procurement based on only one year of cost data, without requiring market-power mitigation.
Industry voices framed the same issue from the demand side. Aaron Tinjum of the Data Center Coalition argued that data centers should cover the costs of their own growth rather than shift those costs to households, while insisting that fair cost allocation must be paired with concrete action to build infrastructure. PJM’s independent market monitor, Joseph Bowring, offered a sharper critique, warning that FERC’s cost-allocation directions still do not go far enough to keep data-center costs off existing residential bills. The structural tools PJM has proposed to address this concern are notable. Beginning with the 2029/2030 delivery year, incremental new large loads above the 2028/2029 forecast would be excluded from the demand used in future capacity auctions. Existing customers therefore would not shoulder those capacity costs unless the large loads bring or contract for their own new supply. Paired with the new Interim Resource Adequacy Service—formerly branded “connect and manage”—which would curtail new large loads lacking dedicated generation during grid emergencies, PJM is attempting to create a regime in which the largest new consumers either finance their own firm power or accept interruption. Whether FERC’s revised allocation formula ultimately enforces or dilutes that principle will define the February order.
Reliability on the Clock
Suspending an auction intended to shore up a capacity shortfall might appear to invite reliability risk, yet the most credible voices in the proceeding argued otherwise. Bowring, the market monitor, stated plainly that the delay poses no reliability or price risk. His assessment reflects the fact that the backstop targets the 2028/2029 delivery year and that PJM retains other tools, including its bilateral matchmaking process and ongoing base auctions. PJM has indicated that it still intends to run its regularly scheduled capacity auction for the 2029/2030 delivery year in December, even as it has declined to set revised dates for the suspended backstop. Lapp, the Maryland consumer advocate, likewise characterized a five-month delay as no major setback for what is fundamentally a fifteen-year procurement. PJM’s resource-adequacy challenge is real and severe, but it is a multiyear problem. FERC’s wager is that a few months spent getting the cost framework right will not meaningfully erode the margin of safety.
That calculation, however, rests on assumptions that could shift quickly. The original backstop sought roughly 6.8 gigawatts of new capacity, with developer-funded transmission upgrades incorporated into offers and all resources required to come online by June 1, 2032. If data-center interconnection requests continue to accelerate and the bilateral track underdelivers, the gap PJM must fill could widen before the redesigned auction runs. The operator’s commitment to restoring the traditional three-year-forward capacity timeframe, along with its broader market-design review under the “Powering Reliability Through Market Design” framework, represents a longer-term effort to make emergency backstops unnecessary. For now, however, the grid is being managed on a compressed timeline, and the February 28 deadline has become the pivot on which both affordability and adequacy turn.
What It Means for the Industry
For utilities, developers, and large loads across PJM’s thirteen-state footprint and the District of Columbia, this week’s events signal that the era of broadly socializing data-center costs across the ratepayer base is meeting firm regulatory resistance. The Commission’s insistence on cost causation, its rejection of the municipal and cooperative opt-out as discriminatory, and its demand for updated, defensible load forecasts point toward a market in which large new consumers will increasingly be expected to internalize the cost of the firm capacity they require. That shift has direct implications for how hyperscalers structure power procurement, pushing them toward bring-your-own-generation arrangements, long-term bilateral contracts, and co-location strategies that were once optional but are becoming, in effect, the price of admission to the queue. Developers of new generation, meanwhile, gain a clearer—if delayed—demand signal and a $555-per-megawatt-day ceiling that, whatever its flaws, shows PJM is willing to pay substantially more than before to bring new capacity online.
The proceeding also reaffirms the pivotal and contested role of the states. FERC can police cost allocation among load-serving entities, but translating those allocations into retail rate classes remains a matter of state jurisdiction. That is why PJM has committed to providing registry and billing data to support state-level cost-allocation efforts and why consumer advocates such as Maryland’s are fighting on two fronts at once. PJM’s capacity shortfall and proposed remedy have therefore become a template that other regional transmission organizations are watching closely, because every market serving concentrated data-center growth will eventually confront the same trilemma: speed, reliability, and affordability. The electricity-affordability pressures driving consumer-advocate interventions are unlikely to ease, and FERC’s cost-allocation ruling, expected in early 2027, will be scrutinized as a precedent far beyond PJM’s borders.
Conclusion
This week’s order is consequential not because FERC paused a single auction, but because it refused to let urgency override the principle of cost causation at the moment when pressure to do so was greatest. PJM built a backstop because its markets could not keep pace with demand; FERC suspended it because the plan for allocating its enormous costs could not withstand scrutiny. Both institutions are responding, in different ways, to the same underlying reality: the AI-driven load boom is testing the foundational bargain of regulated power markets, under which those who cause costs are expected to bear them. The redesigned procurement that emerges before February 28, 2027, will show whether that bargain can withstand the strain of 70 gigawatts of projected new demand—or whether the path of least resistance, spreading the bill across households in thirteen states, proves too tempting to avoid. For an industry accustomed to measuring progress in megawatts and miles of wire, the most important number in the coming months may be the dollar figure on a residential bill. And the most important question may be the oldest one in utility regulation: Who pays?