PJM turns to accelerated backstop procurement after 6.8 GW capacity shortfall
A grid operator's struggle to plan for data-centre load shows why capacity markets, interconnection queues and lead times decide where AI infrastructure gets built.
PJM, the regional transmission organisation that runs the wholesale power grid across a group of US states, is preparing an accelerated procurement after its July capacity auction fell 6.8 GW short of the reliability requirement, Data Center Knowledge reported on 26 August. The Base Residual Auction procured 138,318 MW for the 2028/29 delivery year, leaving a gap of 6,831.3 MW of unforced capacity.
The shortfall follows a stakeholder process, launched in 2025, that was meant to settle how PJM should handle large new loads, principally data centres. That process produced 12 competing proposals, and stakeholders voted every one of them down, so no recommendation reached the PJM board. Jacob Finkel, Pennsylvania's deputy secretary of policy, called the outcome a failure, according to the report.
In its place PJM filed a Reliability Backstop Procurement with the Federal Energy Regulatory Commission on 31 July. The procurement opens on 30 September, runs through September and October, and is expected to return results in December. It sets a maximum weighted-average willingness to pay of $555 per MW-day and allows contracts of up to 15 years, with the aim of closing the 6.8 GW gap. PJM expects roughly 70 GW of new large load could connect by 2038.
The report quoted a range of participants, including Joseph Bowring of Monitoring Analytics, PJM's independent market monitor, Brian George of Google and Mason Emnett of Constellation. Neil Osnato of Persistence Analytics Group argued that adding more alternatives had not created consensus, and that speed and verification should not be treated as opposites.
Why it matters
Data centre demand is now large enough to move the reliability planning of an entire grid region, which changes who bears the risk of a project being built before power exists. A backstop procurement with 15-year terms commits customers to long contracts rather than waiting for market signals. How this process survives regulatory scrutiny will shape how other regions respond to the same pressure.
Data centre demand is now large enough to move the reliability planning of an entire grid region, which changes who bears the risk of a project being built before power exists.
July auction falls short
Base Residual Auction procures 138,318 MW for 2028/29, 6,831.3 MW UCAP below the reliability requirement.
Large-load process stalls
Stakeholder process launched in 2025 produces 12 proposals; all are voted down and none reaches the PJM board.
31 July: FERC filing
PJM files a Reliability Backstop Procurement with the Federal Energy Regulatory Commission.
30 September: procurement opens
Runs through September and October; cap of $555 per MW-day; contract terms of up to 15 years.
December: results expected
PJM aims to close the 6.8 GW gap while forecasting about 70 GW of new large load by 2038.
What you can learn from this
- Capacity markets buy availability, not energy. An energy market pays generators for the megawatt-hours they actually produce, while a capacity market pays them to be available on the peak days years ahead. Prices are quoted per MW-day because the product is a promise to be there for each day of the delivery year. Unforced capacity, or UCAP, discounts each unit's nameplate rating by its historical outage rate, so a 100 MW plant with a 10 percent forced outage rate counts as 90 MW. The two markets settle separately, which is why a region can have cheap energy and expensive capacity at the same time.
- The reliability requirement is a probabilistic target. Grid planners size the capacity they need by modelling the chance that demand exceeds supply, commonly targeting one loss-of-load event in ten years. When forecast load rises, the requirement rises with it. A shortfall means the auction did not attract enough committed capacity to meet that target. It does not mean the lights will go out, but the margin of safety has thinned. Reserve margin, the percentage of supply above expected peak, is the everyday expression of that target.
- Interconnection queues create the timing mismatch. New generation must pass studies to prove it will not destabilise the grid before it can connect, and those studies take years. Data centres can be built in a fraction of that time, so load can arrive well before the supply planned to serve it. This gap between load lead time and generation lead time is why operators reach for procurements outside the normal auction cycle. Some operators now let large loads connect sooner if they agree to curtail during emergencies, trading firm service for speed.
- Long contracts shift risk onto customers. A 15-year term gives a developer the revenue certainty needed to finance new plants, which is why such terms attract supply the annual auction did not. The trade-off is that all customers in the region underwrite that commitment through their bills for the length of the term. Forecast error cuts both ways: if the load does not arrive, the capacity is paid for anyway. Regulators therefore scrutinise the load forecasts behind such procurements at least as closely as the prices.
- Consensus governance struggles with structural change. Stakeholder processes work by weighted voting among generators, utilities, consumer advocates and states. When a question redistributes cost and risk between those groups, every proposal creates a losing coalition large enough to block it. That is the mechanism behind twelve proposals failing, and it explains why regulators end up receiving filings rather than agreed recommendations.
We teach this
How to use this in practice
- Price a hypothetical campus at the backstop cap. Multiply $555 per MW-day by 365 days and by the size of a site, for example 100 MW, and record the result. Done looks like: an annual capacity cost figure, roughly $20 million for that example, sitting beside the energy cost you get from multiplying 100 MW by 8,760 hours by a local wholesale price. Change the site size and the price to see how sensitive the total is to each, and record the assumptions in a comment cell.
- Find your own region's numbers. Identify the grid operator or utility for where you live, locate its most recent capacity or resource adequacy report, and pull out three figures: forecast peak demand, the reserve margin target, and the volume of proposed generation in the queue. Done looks like: a three-line note with sources, so you can check next year whether the margin is growing or shrinking. Most operators publish these figures in an annual load forecast or a state-of-the-market report, usually as a PDF you can search for the words reserve margin.
- Draw the delivery chain and label who pays. Sketch generation, transmission, distribution and a data centre campus, then annotate each link with which cost it carries: capacity, energy, transmission tariff, or connection charge. Done looks like: a single-page diagram where you can point to where a large-load tariff or a backstop procurement changes the bill. Add the typical build time for each link, since the mismatch between them is what the PJM story is about.
- Read one large-load proposal. Grid operators publish stakeholder materials; pick any one proposal on large-load rules and summarise in five bullet points who it asks to bear the risk of load not showing up. Done looks like: a short summary you could explain to a colleague, which is the same skill you need to read any infrastructure contract. Note which stakeholder group would have voted against it, and why, using the categories from your diagram.
Sources
- PJM's Power Shortfall Puts Data Center Growth in Focus — Data Center Knowledge
Our reporting is an original summary; full coverage is at the links above.
Don't just read about it — build it.
Square 1 teaches the skills behind the headlines, with every line of your work graded by AI. Find your starting point in 3 minutes.
Get your free skill report