The Auction That Did Not Run

The facilities director at a Northern Virginia colocation operator learned on 30 September 2026 that the PJM backstop procurement scheduled to open that morning would not run. FERC had suspended it five months the night before. The capacity she was counting on to energize two halls in Q2 2028 now has no price, no timeline, and no vendor obligated to deliver it. The number that had been carrying her board update since July is 6.8 GW, the shortfall PJM wanted to buy to cover load growth it forecast between the 2026/27 and 2028/29 delivery years. By 21 September, PJM's independent market monitor had counted 4,017 MW of datacenter load canceled or delayed, 42% of the growth PJM had forecast. The gap she was relying on may have already closed by the time anyone was allowed to bid.

1. What the auction was supposed to do

PJM designed the Reliability Backstop Procurement as a one-time buy of new capacity to address a resource adequacy shortfall driven largely by rapidly growing demand from large loads such as datacenters. The auction was set to accept offers from 30 September through 21 October, with selection occurring 22 October through 2 December. PJM expected to announce results before its standard capacity auction for the 2029/30 delivery year in early December. The mechanism let PJM lock in supply before the regular auction, giving developers who had filed interconnection requests a bridge between feasibility and finance. The colocation operator had three customers with signed letters of intent, all of them conditional on power delivery by June 2028. The backstop gave her a date she could underwrite.

On 29 September 2026, FERC accepted the proposal but suspended it for five months with an effective date of 28 February 2027, subject to refund and the outcome of a paper hearing. The bid submission window, proposed to open 30 September, is accordingly on hold until the issues FERC raised are resolved. FERC zeroed in on cost allocation concerns, refusing to approve a mechanism that might shift datacenter procurement costs onto residential ratepayers if the datacenter load forecast proved wrong. The chair called the filing "a mess".

2. The load that is no longer there

Monitoring Analytics reported that over 4,000 MW of datacenter load had been canceled or delayed by 21 September. That figure is a mix. Some projects could not secure interconnection agreements within the construction finance window and withdrew. Others hit local opposition and were refused permits. A third category signed conditional PPAs with nuclear restart projects whose timelines slipped, making the datacenter's own go-live date a floating variable. Take the 4,017 MW out and the backup capacity PJM wants to buy shrinks by more than half.

The operator knows her project is not in that count. Her interconnection application is active, her site has zoning approval, and her customer pipeline is live. But FERC does not know that, and the suspension order treats forecast demand as inherently suspect until datacenter developers prove otherwise. Six governors and the mayor of Washington, D.C., pressed PJM in an April letter to protect consumers so that they do not pay backstop costs if datacenters default or PJM overprocures capacity. The regulator is doing what the governors asked. The cost is that projects with actual contracts and actual timelines now wait in the same indefinite queue as speculative filings.

3. The queue she is already in

In PJM and MISO, the two grid operators covering the densest concentration of new datacenter demand in the country, the wait between filing an interconnection request and flipping the switch now runs seven years or more on average. PJM interconnection averages 8+ years from request to operation as of 2026, up from under 2 years in 2008. Her project filed in Q3 2023, giving her a provisional energization window of mid-2030 to early 2031. The backstop was meant to pull capacity forward, letting her de-risk 2028 delivery without waiting for the 2031 grid connection.

According to PJM, projects spent an average of more than three years reaching an interconnection service agreement, and another four years waiting to come online after approval. The bottleneck has moved. Large power transformers used in substations and transmission infrastructure are facing sharply longer procurement timelines, with average lead times increasing from roughly 50 weeks in 2021 to about 120 weeks in 2024, with some large substation and generator step-up transformers ranging from 80 to 210 weeks. A signed agreement no longer guarantees energization on schedule. Equipment that was ordered in 2024 for a 2027 install is now delivering in 2028 or later, and developers who did not pre-order against an unsigned ISA are stuck.

4. The power demand that may not arrive

TrendForce estimates that global datacenter power demand capacity will reach 161 GW in 2026, up approximately 31% year-on-year. Gartner forecasts global datacenter electricity consumption to reach 565 TWh in 2026, up from 447 TWh in 2025. Every one of those forecasts carries the same assumption: that demand converts to load at the rate developers are currently projecting. Nearly 40% of U.S. datacenter projects scheduled for completion this year could miss their targets by more than three months, according to geospatial analytics company SynMax. Proposed U.S. datacenter power requests have topped 700 GW, more than 10 times the estimated electricity currently consumed by U.S. datacenters.

ERCOT is tracking 474 GW of large-load requests, approximately 90% of them datacenters. Governor Abbott paused all new Texas datacenter grid connections in August 2026 pending a comprehensive audit. The mismatch between filed requests and realized load is forcing regulators to treat the entire category as speculative until construction starts. The operator's project has steel going up and cooling towers on order, but the regulatory filter does not distinguish between a financed site and a letter of intent.

5. What she can do

FERC encouraged PJM to submit a revised filing as soon as possible, noting that if PJM submits a new filing within 30 days (by 29 October 2026), FERC would likely pause the paper-hearing schedule while it considers the new filing. PJM has said it intends to submit a revised filing by 29 October and has scheduled a special Members Committee meeting for 22 October to consult stakeholders. The soonest the auction could reopen is late November, assuming PJM's revision satisfies FERC and no intervenor files for rehearing. The latest is February 2027, and only if the paper hearing resolves cleanly.

Her customer contracts have a force majeure carveout for regulatory delay, but only for 180 days. After that the off-take price resets to market, and market has moved. She can wait for PJM to fix the auction, renegotiate her interconnection request to accept a different substation tie with a shorter upgrade scope, or sign a behind-the-meter gas peaker PPA at a strike price 2.3 times what the grid would have cost. The peaker gets her to commercial operation on the customer's original timeline, but it turns the project from 24% to 11% IRR at the term sheet hurdle rate, and it does nothing for the Scope 2 emissions target three of her customers put in their RFPs.

6. The number after the number

Delayed is not the same as canceled, and the monitor's count puts the two in one bucket. A project that slips a year or two may still need power well within a 15-year contract. The 4,017 MW that dropped out of PJM's 2028/29 forecast has not disappeared. Some of it moved to 2029/30, some renegotiated for ERCOT or MISO, and some is sitting in the same suspended state she is in, waiting to see if the auction reopens or if the queue clears. What FERC decided is that it will not buy capacity to cover demand until the demand proves it can pay. The operator who already has contracts, a construction loan, and an energization dependency now carries the same regulatory discount as a speculative filing with no site and no tenant. The auction that was supposed to separate the two did not run.

Tarry Singh is the founder and CEO of Real AI (realai.eu), an enterprise AI advisory and deployment firm working with global enterprises on production agent systems, model risk, and AI sovereignty strategy. He also leads Earthscan (earthscan.io) for Energy AI, and is a founding contributor to the EU-funded HCAIM and PANORAIMA programmes for responsible AI education across European universities. He writes at tarrysingh.com.