PJM's Market-Reform Push Turns Data Center Power Into a Capacity-Finance Problem

One of the strongest power-market stories of May is not a utility earnings beat or another giant load forecast. It is PJM saying out loud that the market design beneath the grid is no longer aligned with the physical build timeline needed to serve data centers and other large loads. In its May 6 paper, Powering Reliability Through Market Design, PJM said the region is now dealing with a convergence of three structural forces: rapid growth in large-load demand led by data centers, accelerated retirement of dispatchable generation, and supply-chain and permitting frictions that have made new resources slower and more expensive to build.

That framing matters because it moves the AI power debate beyond vague scarcity language. PJM is not only warning that demand is rising. It is arguing that its existing capacity-market structure was built for a world where supply could show up on a timeline that matched the market's forward signal. PJM says that is no longer true. The report states that a new natural-gas turbine plant now requires at least four years from financial investment decision to commercial operation under optimistic assumptions, while the market's design was calibrated around a shorter development cycle.

PJM's real message is that data-center load is no longer just an interconnection problem. It is now a capacity-finance problem where market timing, construction timing, and affordability no longer line up cleanly.

The second important detail is cost. PJM says installed costs for new combined-cycle gas plants now regularly exceed $2,200 per kilowatt for projects targeting the late 2020s and early 2030s. That is a useful number because it turns the story from abstract reliability stress into capital formation math. If the market signal arrives on one schedule while equipment, permitting, and financing move on another, then large-load growth becomes a problem of who takes construction risk, who gets paid for speed, and who absorbs the consequences when supply arrives late.

PJM's March 2 filings help show what this means in practice. The grid operator filed to extend its auction price collar for two more delivery years and to create an expedited interconnection track for up to 10 generation projects per year that can deliver at least 250 megawatts of accredited capacity, commit to commercial operation within three years, and cover their own needed network upgrades. PJM also said a one-time reliability backstop procurement is being developed to secure future supply and be paid for by large loads, while some data-center growth may be subject to curtailment if it does not bring associated power supplies to the grid.

Those details are what make the story stronger than a generic market-reform headline. PJM is effectively sketching a new deployment logic for AI infrastructure in its footprint. Large loads may not just need an interconnection path. They may need a supply strategy, curtailment tolerance, and a willingness to fund reliability protections that keep the rest of the system from eating the cost of their speed. That is a more specific and more consequential operating environment than the old assumption that demand could simply show up and the market would eventually clear it.

For operators, the practical takeaway is that site selection in PJM is becoming inseparable from market structure. A viable campus is increasingly one that can pair load growth with dependable supply timing, tolerate procedural constraints, or contract around them. Power readiness in the region is no longer only a land-and-substation question. It is a question of whether the project fits the revised rules for scarcity, curtailment, and generation procurement.

For investors, PJM's paper is one of the more useful public documents for understanding where AI infrastructure economics are heading next. The Grid Report view is that the region is telling the market to stop thinking only in terms of megawatts requested and start thinking in terms of bankable delivery. The winners are less likely to be everyone exposed to data-center demand and more likely to be the developers, generators, utilities, and structured-power counterparties that can bridge the mismatch between auction timelines and actual energization timelines.

The broader implication is that AI load is now shaping wholesale market design itself. Once data-center growth starts forcing price collars, expedited supply tracks, backstop procurement, and curtailment frameworks, the story is no longer just about new demand. It is about whether the market can still coordinate the physical system fast enough to serve that demand without breaking affordability and political support. PJM's answer so far is that the old design is not enough.

PJM, “Powering Reliability Through Market Design,” May 6, 2026: https://www.pjm.com/-/media/DotCom/library/reports-notices/special-reports/2026/20260506-powering-reliability-through-market-design.pdf

PJM, “PJM To Lead Market Reform Effort To Support Generation Investment and Reliability,” May 6, 2026: https://www.pjm.com/

PJM Inside Lines, “PJM Files Price Collar, Expedited Interconnection as Part of Large Load Plan,” March 2, 2026: https://insidelines.pjm.com/pjm-files-price-collar-expedited-interconnection-as-part-of-large-load-plan/

PJM Inside Lines, “PJM Board Outlines Plans To Integrate Large Loads Reliably,” January 16, 2026: https://insidelines.pjm.com/pjm-board-outlines-plans-to-integrate-large-loads-reliably/