The Auction That Didn't Happen

September 30 was supposed to be the day the bids arrived.

This newsletter had tracked the backstop since July and circled the date itself since Issue #30. PJM Interconnection, the largest grid operator in America, would open a three-week window in Valley Forge, Pennsylvania to buy 6.8 gigawatts of new power plants on contracts of up to 15 years, with every dollar billed to data centers and large loads. The first auction in the history of American electricity markets designed to make the industry buy its own reliability. Issue #30 called it one half of a double verdict. Issue #32, after Newsom signed early, called it the single verdict left.

On September 29, the day before the window opened, the Federal Energy Regulatory Commission issued its order. It accepted PJM's plan. Then it suspended the plan for five months, to an effective date of February 28, 2027, subject to refund and further proceedings, after finding that the parts deciding who pays may be unjust and unreasonable.

On September 30, PJM confirmed what that meant: the auction would not open. No new timeline. All previously announced milestones under review.

The chair of FERC, Laura Swett, filed a concurrence that the trade press described as scathing, and the headline writers reduced to one word: "mess."

So the double verdict became a single verdict, and then no verdict at all. One governor acted nine days early, and one grid operator didn't act on time. This issue covers what FERC actually did, why it matters more than the delay itself, and what the people who build should take from a reliability tool that now takes effect in late winter for a shortfall scheduled for 2028.

Let's go.

The order is more precise than "rejected," and the precision is where the story lives.

Accepted: the procurement itself. A one-time Reliability Backstop Procurement to close the 6,831 MW shortfall from the 2028/2029 capacity auction, minus any capacity that utilities can document through bilateral contracts with data centers. Only new resources may offer: new plants, uprates, repowerings, and storage able to reach commercial operation by June 1, 2032. Pay-as-bid, with offers ranked first by earliest commercial operation date and then by lowest levelized cost.

Accepted: the price. The $555 per megawatt-day offer cap, on a megawatt-weighted average basis over the delivery period, stands. The number this newsletter has used since Issue #30 survived.

Suspended: the effective date, five months, to February 28, 2027, with a paper hearing in between and a list of potential modifications FERC wants addressed.

Struck: PJM's plan to let qualifying cooperatives and municipal utilities opt out of the procurement entirely. FERC called that discriminatory against other load-serving entities with data centers in their footprints. If the bill is going to data centers, it goes to all of them, not just the ones served by investor-owned utilities.

Redirected: the cost allocation. FERC offered an alternate approach that ties the backstop's costs to updated load forecasts rather than the forecasts PJM used, and told PJM to employ the most current load and resource data to set the procurement target itself. And if a transmission owner exits PJM before the costs are recouped, FERC said those costs land on the load-serving entities in that zone, not the departing owner.

And the sentence that explains the tone: Swett noted that PJM had filed its proposal on the last possible statutory day that would still allow a September 30 opening. The grid operator gave its regulator no room to fix anything without blowing up the schedule, and the regulator declined to be rushed. FERC encouraged PJM to come back with a revised filing within 30 days under its own tariff rights, and commissioners wrote separately urging speed.

PJM's response, through spokesman Jeffrey Shields: the order provides a "clear path," and the grid operator intends to address the remaining concerns quickly.

A five-month suspension sounds like a scheduling problem. Three consequences make it more than that.

First, the December auction goes blind. PJM had planned to take backstop offers through October 21, pick winners by December 2, and publish results before its regular capacity auction for the 2029/2030 delivery year in early December. That sequence is gone. Bidders into the December auction will now price it without knowing what the backstop bought, or whether it will buy anything. The two auctions were designed to inform each other. One of them will now run in the dark.

Second, the self-generation thesis just got five more months of confirmation. This newsletter has argued since the New York order (Issue #22) that every signal in this industry points toward data centers bringing their own power. PJM's own framework includes a registry of every load of 50 MW or more and an Interim Resource Adequacy Service under which new large loads arriving after June 1, 2027 without their own capacity are curtailed first. The backstop was the alternative to that: the grid buying capacity on the industry's behalf. The alternative is now suspended until a month before the curtailment rule's own trigger date. One trade analysis put it in a sentence: building on site looks better every month the backstop slips. It slipped five.

Third, the forecast fight is now the whole fight. Read FERC's redirection carefully. The commission's central objection wasn't the price or the mechanism. It was the load forecast underneath the cost allocation. PJM's independent market monitor has argued for a year that data center load in the forecast is uncertain and that much of it won't materialize (the same point SemiAnalysis made from satellite data in Issue #21). FERC just made the forecast the legal hinge of the entire procurement. Whoever controls the demand number controls who pays, and PJM has 30 days to produce one its regulator will accept.

Add the context from this newsletter's own archive and the pattern is plain. The regular capacity auction hit its price cap three times running and left a 6.8 GW hole (Issue #23). The backstop was the fix. The fix is now suspended. The curtailment registry is what's left. For a developer in the 13-state footprint, the operative question in October 2026 is no longer "what will the backstop cost me." It's "can I bring my own power before June 1, 2027."

If the forecast is the fight, here is why it's worth fighting over.

PJM's independent market monitor, Monitoring Analytics, has tallied what data center load did to the region's capacity costs across the last four auctions: an increase of $29.4 billion. That figure is before the backstop buys a single megawatt, and before the regular auction's cap rises any further. It is the bill that already landed, spread across 67 million people's utility statements, for demand the monitor says may not fully materialize.

Twenty-nine billion dollars is the reason a forecasting dispute between a grid operator and its regulator is the most consequential argument in American energy right now. Every gigawatt in the forecast that turns out to be vapor is a share of that bill paid by someone who never used the power. Every gigawatt that's real and unforecast is a reliability hole nobody procured for. FERC's instruction to use the most current load data isn't bureaucratic housekeeping. It's an attempt to find out which of those two errors the region is about to make, with a nine-figure price on getting it wrong in either direction.

The auction is suspended. One part of PJM's framework is not, and it's the part worth a phone call this week.

PJM's bilateral contract matchmaking is already underway. The design, approved by stakeholders in June and left intact by FERC's order, lets new supply resources contract directly with large-load customers, and every confirmed bilateral deal reduces the backstop's procurement target. The eligibility rules are the same as the auction's: new plants, uprates, repowerings, and storage that can reach commercial operation by June 1, 2032.

Translate that for the people who build. A generation developer with a shovel-ready project in the 13-state footprint no longer needs to wait for a February window that may move again. The counterparty is sitting across the table right now: a data center operator reading the same curtailment rule, with a June 1, 2027 date circled, looking for a capacity contract that keeps it off the registry's front line. The auction was going to intermediate that match at a $555 cap. The suspension just removed the intermediary and left the two parties, the deadline, and the incentive.

Three specific markets open up in the gap:

Bilateral capacity contracts. Developers of new gas, storage, and uprate projects can sell directly to hyperscalers and colocation operators, on terms the parties set, with PJM's own process documenting the deal. The lawyers and advisors who structure power purchase agreements have a new product with a built-in buyer urgency.

On-site generation, again. The curtailment rule did not move. The backstop did. Every month the gap widens, the behind-the-meter case gets stronger, and the equipment, switchgear, controls, and fuel supply that go with it (the Power Equipment Map's nine categories) get ordered sooner. Texas narrowed this lane in September by halting environmental permits; PJM just widened it by delaying the alternative.

Forecast and interconnection advisory. FERC made the load forecast the legal hinge of the whole procurement. Utilities, developers, and state regulators across thirteen states now need defensible demand numbers for their own filings, and the firms that can produce them have a client list that just doubled.

The buyers are motivated by a date. The sellers are motivated by a suspended auction. The matchmaking line is open. If your company sells into any of it, reply "MAP."

This newsletter promised, from the inaugural ratings forward, to be wrong in public when it's wrong. Two entries this week.

September 30 as the double verdict day. Called in Issue #30, repeated in Issue #32. Newsom signed September 21; PJM didn't open on September 30. The calendar was accurate as published by both parties at the time, and the sources are in the archive. But the framing treated two scheduled events as settled ones, and neither held. The correct framing was always "scheduled," and future "What to Watch" entries will say so when the date depends on a party that can move it. Lesson logged.

The curtailment registry and the self-generation read. Called in Issue #30, held up. FERC's order left the large-load registry and the Interim Resource Adequacy Service intact while suspending the backstop, which is the exact asymmetry the thesis predicted. That one stands.

Net: one framing miss, one call confirmed, both on the record.

Virginia remains on downgrade watch. Loudoun County's final vote on its 12-month application pause is October 20. The watch resolves there, as published in Issue #31.

Texas holds at Frozen with the widened scope from Issue #32. TCEQ's compliance report to the governor's office is due October 19. ERCOT's audit target is December 10.

California sits at Build with Conditions as of Newsom's September 21 signatures. The CPUC's tariff deadline under SB 886 is January 1, 2028.

New York: at our last check, the community-benefits guidance promised under the July 14 order still had not surfaced; the 60-day clock expired September 12. If a reader in Albany has seen it, reply "RATING" and we'll correct the record.

The Q4 State Risk Ratings edition will publish after October 20, so the Virginia watch resolves inside it rather than a week before it. It will carry the Texas scenario call, California's debut, the Texas poll results, the first reader-driven revisions, and a new line item: PJM-footprint exposure, now that the backstop has a February date and the curtailment rule has a June one.

Thirty-one days to the election. The Midterm Watchlist stands as published.

Every date below is as scheduled by the responsible party and subject to that party moving it. We'll say so from now on.

The most anticipated power auction in the history of the data center industry was supposed to open on Wednesday. It didn't, because the grid operator filed on the last possible day and its regulator refused to be rushed into deciding who pays for 6.8 gigawatts.

That's the surface. Underneath it is the thing this newsletter keeps finding at the bottom of every story: nobody has an agreed answer to the question of who pays for the power. California answered it with a ten-year tariff. New Jersey with an 85 percent floor. Virginia with a consumption tax. PJM tried to answer it with an auction, and the federal government sent the question back with a note that said: your demand number is wrong.

The industry will keep building through the five-month gap, as it has through every freeze, pause, audit, and veto this year. But the one mechanism that would have let it buy reliability collectively is now on a shelf until late February, and the one mechanism that penalizes not having it takes effect in June. Between those two dates, every developer in thirteen states is doing the same arithmetic, and it comes out the same way it has all year.

Bring your own power. The backstop will be there in February. Maybe.

On Wednesday, the same day the auction meant to fix PJM's reliability shortfall failed to open, the PJM Members Committee voted down a proposal from PJM's own management that would have given the thirteen states a formal "jump ball" to raise resource adequacy concerns for regulatory attention. In one morning, the region's grid operator both postponed its emergency fix and declined to let the governors ask about it. The states will have to keep raising their concerns the way they have all year: in bills, in orders, in audits, and in freezes.

The DC Pipeline tracks data center construction, policy, and market intelligence across North America. Home of the State Risk Ratings and the Midterm Watchlist.

If someone forwarded you this: we score our own calls in public, misses included. Subscribe at thedcpipeline.com.

Challenge a rating: reply "RATING." Join the supplier list: reply "MAP." Flag a local race: reply "BALLOT."