If you live anywhere from Illinois to the District of Columbia, this is not an abstract market story. The grid in question serves about 67 million people across 13 states, and it is also home to the densest cluster of data centers in the country. A 6.8 gigawatt gap in the supply stack for the 2028/29 delivery year is the number that started this scramble. Six point eight gigawatts is not a rounding error. It is the first time the entire region failed to clear its reliability requirement. The auction that was supposed to paper over that miss is now on ice.
Why The Biggest Regional Grid Hit Pause
The operator had filed a Reliability Backstop Procurement at the end of July. Think of it as a one-shot, 15-year shopping trip for new capacity, with the bill meant to land on the large loads driving the shortage rather than on every kitchen table in the footprint. Offers were lined up to run from September 30 through October 21, with selections expected by early December. Then the federal energy regulator accepted pieces of the filing and suspended the framework for five months. The chair said the commission would not be forced into blessing an eleventh-hour design.
I have watched capacity markets for years, and this sequence still feels abrupt. A procurement built to answer a historic shortfall was halted before a single offer could be logged. That does not mean the shortfall vanished. It means the argument over who pays, who can walk away, and how much collateral a local utility must post just got a longer runway. Perhaps the most interesting part is how little of the physical problem moved while the paperwork did.
The Shortfall That Forced The Issue
Back in July, the base capacity auction for 2028/29 cleared short. The price cap was the only thing holding the result near $325 per megawatt-day. Strip that ceiling off and the same auction would have landed around $554.72. A few hours of arithmetic made the point better than any press release: the region was already under its reliability threshold, and the cap was doing the work the supply stack could not.
Capacity prices did not creep up. They sprinted. Two auctions took the clearing price from $28.92 to the cap. That kind of move usually shows up when new plants are late, old plants are leaving, and a new class of customer is arriving faster than either side of the market can adjust. Data centers are that customer. They do not sip power. They drink it, steadily, and they want it on a timetable that transmission lines and combined-cycle plants simply do not share.
A price cap can hide a shortage from the headline. It cannot hide the shortage from the people who still have to keep the lights on.
The independent market monitor has been blunt that the backstop, even if it had launched cleanly, would cover only part of the gap. Rapid load growth means the 6.8 gigawatt miss is a floor, not a ceiling. Earlier this year the grid operator itself warned the shortfall could swell toward 60 gigawatts over the coming decade if nothing structural changes. I keep coming back to that range. Seven months later we are still arguing about the rules for the first slice.
Who Has Been Paying So Far
Here is the part that rarely makes the server-hall ribbon cuttings. According to the market monitor, data centers accounted for $6.3 billion, or 38 percent, of the $16.4 billion in charges from the latest auction. Across the last four auctions the same class of load is tied to $29.4 billion, or 46 percent, of a $63.6 billion tab. Nearly half of four years of capacity cost traces back to facilities most households will never tour.
That money does not stay in a spreadsheet. It lands on bills across 13 states and the District. People who have never heard the word hyperscaler are already underwriting a slice of the buildout. In my view that is the political fuse. You can debate interconnection queues all day. Once a winter bill shows a capacity line that did not exist three years ago, the debate moves to kitchen tables, and kitchen tables vote.
- Latest auction charges tied to data centers: about $6.3 billion, or 38 percent of $16.4 billion
- Four-auction total tied to data centers: about $29.4 billion, or 46 percent of $63.6 billion
- Customers in the footprint: roughly 67 million across 13 states and the District
- Reliability miss for 2028/29: 6.8 gigawatts, the first full-region shortfall
None of those figures say data centers are villains. They say the old cost-sharing habit was built for a slower world. A factory here, a subdivision there, a hospital expansion every few years. A campus that wants several hundred megawatts on a fixed date breaks that habit. The backstop was an attempt to retarget the bill. The regulator did not reject the goal. It rejected the rushed design.
What Regulators Actually Objected To
The suspension is not a blanket no. Parts of the plan were accepted. The objections cluster around three practical problems: how costs are allocated, the rules that let transmission owners exit the arrangement, and the collateral load-serving entities would have to post. That last item is the one that makes treasurers sweat. One northern Virginia cooperative alone would have needed to post roughly $2 billion in collateral. Two billion, for a cooperative, to stand behind a procurement that had not even opened.
The offer cap inside the procurement sat near $555 per megawatt-day. That is almost exactly the uncapped price the July auction would have printed. Coincidence or signal, it told developers the backstop was willing to pay scarcity prices for a decade and a half. Developers like clarity. They do not like posting rules that can strand a project if a transmission owner walks, or collateral math that freezes a local utility before a shovel hits dirt.
I’ve found that eleventh-hour market designs fail in the same place every time. The economics can be directionally right and the legal plumbing still leaks. Cost allocation is plumbing. Exit rights are plumbing. Collateral is plumbing. Federal regulators, warned in July that they would impose reforms if the region did not move by September, decided the filing on the table was not ready to become the rule. A five-month suspension, with an effective date pushed out to late February 2027 under the order as analysts read it, is their way of saying rewrite it or sit through a paper hearing.
The Auction That Never Opened
There is a dry comedy in the calendar. Offers were supposed to start September 30. The suspension landed the day before. By the next morning the operator had pulled the launch. No bids, no selections in December, no 15-year contracts signed under this version of the rules. For anyone keeping a private scorecard of delayed data-center power, add one more line: the auction meant to pay for the plants is delayed too.
That scorecard was already long. Campus schedules have slipped. Financing windows have slipped. At least one large project has invoked force majeure. An offering meant to fund a flagship site has been pushed. None of that is gossip from a single campus tour. It is a pattern. When the grid operator then delays the mechanism built to buy the missing megawatts, the pattern includes the market itself.
Does a delay of the auction equal a delay of every data hall? No. Some projects already hold interconnection rights, some have bilateral deals, some will simply wait. But a region that missed its reliability target does not get to treat time as free. Every month the rules stay unsettled is a month when a turbine order, a transformer slot, or a gas interconnect can slip to the next window. Those windows are not weekly.
| Item | What was planned | Where it stands |
| Backstop offers | September 30 to October 21 | Launch date not set |
| Selections | Early December | Not applicable until relaunch |
| Term | 15-year procurement | Design under rewrite or hearing |
| Offer cap | About $555 per megawatt-day | Likely revisited with the filing |
| Framework effective date | Near-term start | February 28, 2027 under the order |
| Collateral example | Cooperative posting near $2 billion | Central objection in the suspension |
Two Filings, And Why One May Matter More
Analysts who cover independent power producers were quick to call the suspension net bearish but mixed. Lack of a finished framework can stretch the regulatory overhang and cool the appetite for long-term power purchase agreements. At the same time, the shortage that makes existing plants valuable did not get fixed by a press statement. Tight markets and higher pricing remain the base case until new capacity is actually in sight.
Names with heavy exposure to this region sit closest to the rewrite. Others with a broader fleet are a step removed. As a group, the independent producers have been trading at a clear discount to what a settled rulebook would normally support, something like the mid-single digits on forward enterprise value to earnings before interest, tax, depreciation, and amortization, with free-cash-flow yields that look generous precisely because nobody knows the rules. A significant discount is what uncertainty looks like when it has a ticker.
The more important filing, in the view of several utility researchers, may not be the backstop at all. It is the Interim Resource Adequacy Service, a framework for large loads that want to connect without bringing their own capacity, in exchange for being curtailable. Filed in August, it would let data centers waive curtailment compensation, keep new large load off the capacity demand curve so residential customers are shielded, and give bilateral contracts a path forward. The operator asked for a ruling by October 12 even though the service would not take effect until June 2027.
Many developers prefer bilateral deals to a central procurement. A constructive ruling on that interim service could shrink the need for the backstop. In other words, the next couple of weeks of docket-watching may matter more for contract flow than the suspended auction did. I would not bet the house on any single order date. I would bet that lawyers on both sides are reading the same paragraphs twice.
- The operator can sit through a paper hearing under the suspended framework.
- Or it can file a fresh proposal within 30 days and try to accelerate approval.
- Either way, the effective date currently read by analysts is February 28, 2027.
- The interim large-load service, if approved cleanly, could pull some demand out of the central auction.
The Market Did Not Wait For The Docket
While collateral arguments filled the comments, power traders repriced the curve. Over the past month the region has seen a substantial move in both cash and forwards. September peak cash was set to clear near $128 per megawatt-hour, up about 47 percent from where it traded entering the month, helped by an odd heat pattern and scheduled transmission outages. October pinned above $100 into options expiry. Calendar strips for 2027 through 2029 have all traded above $90, with load-serving entities hedging intermediate commitments more aggressively.
That is a market saying the shortage is not theoretical. A commodities desk summary put it plainly enough: meaningful repricing across cash and forward tenors. Translation for anyone who pays a bill, higher prices are already in the strips that suppliers use to lock supply. Regulatory delay does not freeze the forward curve. If anything, delay removes one path by which new supply might have shown up, and curves notice.
A rough read of the recent move: September peak cash near $128/MWh, up about 47% intra-month October pinned above $100 into expiry Cal-27 through Cal-29 strips above $90 Capacity cap still the lid on the 2028/29 auction print
Weather and outages explain part of the cash spike. They do not explain a whole strip of outer years lifting together. That is load, retirements, and a thin bench of projects that can actually reach commercial operation before the delivery year in question. Forward power is a blunt instrument. Right now it is blunt in one direction.
Bills, Midterms, And The Affordability Argument
Research desks that track customer affordability have flagged this footprint as the sharpest pressure point in the country, precisely because capacity prices have reflected the tightening. One forecast puts average utility-bill inflation near 3.7 percent a year through 2029. I will say the quiet part: that average can hide a much steeper local path if capacity stays pinned at the cap and energy forwards stay elevated. Averages are polite. Winter bills are not.
After sitting with a large utility’s federal regulatory lead, researchers reported a plain view from the company side. The current capacity construct is not, in that telling, adequate to incentivize new supply. Scrutiny of data centers and bill inflation is expected to outlast the November midterms. Voters in Maryland, Pennsylvania, and New Jersey have already noticed. Once a state commission hearing fills with residents holding printed bills, the design question stops being a specialist sport.
There is a fair counterpoint. Data centers also bring tax base, construction wages, and a slice of the computing infrastructure the rest of the economy now treats as ordinary. The tension is not whether they should exist. It is whether the cost of the marginal megawatt should be socialized across every meter or assigned, as far as the physics and the law allow, to the load that caused it. The suspended procurement tried to do the second. It did not clear the first draft.
The only schedule that still looks punctual is the one on the electric bill.
A grid watcher, after one too many slipped in-service dates
Bring Your Own Power, Or Keep Sharing The Queue
A year ago, as bills started their climb, the practical argument from the Texas model got louder: require large computing loads to show up with on-site generation, behind the meter, rather than assuming the shared system will stretch. A month later the same idea was put more bluntly. Make behind-the-meter supply a condition, not a courtesy. Wall Street has been catching up. One updated outlook lifted the 2030 global figure for behind-the-meter generation by 68 percent, from 40 gigawatts to 67. Gas turbines and fuel cells are each expected to take a quarter or more of the incremental build. The same work now sees on-site power serving about a quarter of all data centers by 2030.
That shift is not a slogan. It is a reaction to queue times, to capacity prices at the cap, and to customers who would rather own the megawatt than argue about it. On-site gas is faster than a new combined-cycle plant tied into a congested node. Fuel cells are cleaner at the stack and still depend on fuel logistics. Neither is free, and neither removes the need for the shared grid when the campus wants redundancy. They do change who carries the first dollar of new capacity.
Longer term, small modular reactors keep showing up in the serious decks. The path is bumpy, and a recent procedural loss shows why. Federal regulators sided with the grid operator and removed a 750-megawatt hybrid nuclear project in Virginia from the interconnection queue. The developer warned the decision would delay the effort by more than a year. A year, on a technology that already has a long licensing road, is not a footnote. It is another reason near-term supply will look like turbines, engines, and demand response rather than a row of new reactors.
What The Delay Does To Builders And Buyers
Put yourself in the chair of a developer with a site, a deposit on equipment, and a customer who wants a commercial operation date that already slipped once. The backstop was one way to underwrite the plant. A bilateral contract under the interim large-load service is another. A pure behind-the-meter build is a third. Each path has a different regulator, a different collateral schedule, and a different chance of being curtailable on the worst summer afternoon. The suspension does not kill the project. It forces a fresh choice among paths that all have friction.
Buyers, meaning the computing firms and their landlords, face a mirror image. Sign a long power purchase agreement into a rulebook that may change, and you own basis risk you cannot fully model. Wait for the rewrite, and your campus date moves. Accept curtailment in exchange for a faster interconnect, and you need a workload that can actually drop. Training runs can sometimes move. Inference tied to a consumer product often cannot. That split, more than any slogan about artificial intelligence, will decide which halls get built on the shared grid and which arrive with their own iron in the yard.
Local utilities sit in the middle, which is why the collateral number mattered so much. A cooperative posting $2 billion to back a regional procurement is not a theoretical credit metric. It is a balance-sheet event. If the rewrite lowers that burden and still assigns costs to the new load, you may see reluctant support return. If the rewrite socializes the cost again, expect state commissions to slow-walk approvals and expect more campuses to look at private wires and on-site machines.
A Plain Reading Of The Reliability Math
Reliability planning is mostly unglamorous arithmetic. You forecast peak load, subtract expected retirements, add projects that have a real chance of showing up, and apply a reserve margin. When the sum falls short, you are not in a philosophical debate. You are short. The July auction said the region was short for 2028/29 even after the price ran to the cap. The monitor says the proposed backstop would not close the whole gap. The operator’s own earlier warning said the gap could widen sharply over a decade.
People sometimes treat a gigawatt as a vibe. It is not. A modern combined-cycle plant might contribute on the order of several hundred megawatts to a gigawatt, depending on configuration. Filling 6.8 gigawatts of accredited capacity means multiple large plants, or a crowd of smaller ones, plus the transmission to move the power, plus the fuel or the fuel substitute. Do that while load is still growing and while older units exit, and the 6.8 starts to look like the easy year. That is why calling it a floor is not rhetoric. It is how the stack works when the additions lag the additions to demand.
Curtailment can shave the peak. It cannot run a region. Demand response, interruptible rates, and the interim service’s curtailable option are useful tools at the margin. They are a weak substitute for steel in the ground if the forecast keeps rising. I tend to trust the boring stack chart over the keynote slide. The stack chart, right now, is the one with the hole in it.
How Investors Might Read The Mixed Signal
For holders of generation, the suspension cuts two ways. Near-term contract momentum can slow if customers hesitate to sign while the framework is in flux. That is the bearish half. The bullish half is stubborn: scarcity rents do not require a perfect tariff. They require a tight system. Until accredited megawatts arrive, existing fleets in this region keep a supportive backdrop. Analysts have pointed to higher pricing and tight conditions specifically in the absence of a clear line of sight to new capacity.
The discount in the group, an average around 7.4 times enterprise value to EBITDA and roughly a 12 percent free-cash-flow yield on 2027 estimates excluding one large nuclear-heavy name, is the market’s way of charging for rule risk. If a cleaner filing lands within 30 days and a bilateral path opens under the interim service, some of that discount can compress without a single new plant synchronizing. If the hearing drags and state politics harden against large loads, the discount can stick even while cash flows stay strong. Both outcomes fit inside the same set of facts. That is what mixed means.
I would separate the trade from the civic question. A tight market can be good for a plant owner and rough for a household in the same month. Pretending those interests match is how these debates go sour. The rewrite has to serve reliability first. Who captures the rent is a second question, and it is the one legislators will actually hear about.
What A Workable Rewrite Would Need
Nobody outside the docket can draft the tariff, but the objections already sketch the minimum. Cost allocation has to be legible enough that a state commissioner can explain it without a whiteboard. Exit rights for transmission owners cannot leave a generator holding a 15-year promise that evaporates. Collateral has to be large enough to mean something and small enough that a cooperative is not asked to post a sum that rivals its entire rate base. The offer cap should reflect scarcity without becoming a blank check indexed to the last uncapped counterfactual.
A second requirement is speed that is real, not theatrical. A procurement that cannot award until 2027 does little for a delivery year that is already short, unless parallel paths, bilateral contracts, and on-site builds, are moving at the same time. The interim service is that parallel path. If it is written so residential load is insulated and large customers accept curtailment risk in exchange for speed, it may do more practical work than a perfect backstop that arrives late.
There is also a geographic honesty the rules should not dodge. Northern Virginia is not rural Illinois. Congestion, land, water, and community tolerance differ. A single regional product that pretends every node is interchangeable will mis-price the plants that can actually relieve the constrained pockets. Locational signals are messy. Ignoring them is messier.
The Slower Collision Behind The Headlines
Step back from the auction calendar and the pattern is older than this filing. Computing campuses have been announced on schedules that generation and transmission cannot match. Some of those campuses have already moved their in-service dates out by years. Estimates from earlier this year suggested that half of the US data centers scheduled to start in 2026 would be canceled or delayed, and that more than two-thirds of the power sought for these facilities might never materialize in the form first requested. Those are not comforting ratios if you underwrote the original press release. They are useful if you are trying to forecast actual load.
Actual load still rises. A canceled campus removes a line from a queue. It does not remove the campuses that already have steel up, nor the ones that will accept a higher power price and a later date. The collision is between announced capex and accredited megawatts. Capex can be revised in a board meeting. A turbine slot and a substation cannot. That asymmetry is why traders have been willing to lift outer-year power even as some project headlines slip. Slippage in announcements is not the same thing as surplus in the stack.
Perhaps that is the line I would underline. Delay is not relief. A delayed auction, a delayed campus, and a delayed reactor interconnect can coexist with a rising bill, because the demand that is already connected does not wait for the demand that is not. Households experience the first. Spreadsheets argue about the second.
What To Watch In The Next Few Months
Three clocks are running, and they do not tick together. The first is the 30-day window for a fresh filing that could bypass a long hearing. The second is the requested ruling date on the interim large-load service, which developers hope will unlock bilateral contracts even if the service itself waits until mid-2027. The third is political: state hearings, bill inserts, and a midterm season in which electricity has stopped being a background issue in this footprint.
- Whether a revised procurement lands inside 30 days or the paper hearing runs its course
- How collateral and exit rights are rewritten, especially for smaller load-serving entities
- The interim service order, and whether curtailment terms are strict enough to protect other customers
- Forward power and capacity prints, which have already moved ahead of the docket
- On-site generation announcements, the practical bypass when the shared queue stays slow
- State commission tone in the states where bills have moved the most
I do not expect a single order to settle this. The region needs plants, wires, and a cost rule that can survive a court and a campaign. It currently has a suspended rule, a tight forward curve, and a reliability miss on the books. That combination can persist longer than a news cycle. It cannot persist forever without either new supply or a sharper argument about which loads get served first on the peak day.
A Few Distinctions Worth Keeping Straight
Energy price and capacity price are not the same bill line, even though both hurt. The cash and forward moves discussed above are mostly energy. The $325 cap print is capacity, the payment for being available. A household sees a blended result. An investor should not. A plant can earn scarcity energy in a hot month and still depend on the capacity construct for the revenue that justifies a new build. The backstop was aimed at that second stream. Suspending it does not cap the first.
Accredited capacity is not nameplate capacity. A data center’s requested interconnect and the megawatts a plant can actually offer into a reliability auction are different numbers, shaved by outage rates, fuel risk, and deliverability. This is why a queue full of hopeful projects can coexist with a short auction. Hope is not accreditation. The July result was accreditation, and it came up short.
Behind-the-meter supply helps the customer who builds it. It helps the shared system only to the extent that the load would otherwise have drawn on the pool. If the campus still wants grid backup for the hours its on-site units are down, the region has not fully escaped the obligation. That residual obligation is where a lot of quiet arguments are going to live over the next two years.
Why This Story Is Larger Than One Auction
The paused procurement is a symptom. The underlying mismatch is a decade of digital load arriving on a planning cycle built for slower growth, in a region that also has to manage retirements and a transmission build that takes longer than a server order. Federal regulators were right, in my opinion, to refuse a rushed design with a multi-billion-dollar collateral quirk and muddy exit rules. They did not, and could not, manufacture 6.8 gigawatts by rejecting a filing.
So the practical question for the winter ahead is narrower than the headlines. Will a cleaner rule arrive fast enough to pull private capital into accredited plants before the outer years of that forward curve become next year’s cash? Will large loads accept curtailment, on-site generation, or both, in exchange for a place in the queue? And will the cost of the residual shortfall stay visibly tied to the load that created it, or drift back onto everyone else while the lawyers finish?
I keep the spokesman’s line on a sticky note. Launch date yet to be determined. It is accurate, and it is incomplete. The launch of the auction is undetermined. The bill for the tightness is not. Between those two facts sits the entire argument this region is about to have, in hearing rooms and on rate inserts, about who gets to plug in and who pays for the privilege.
Reliability gap, simplified: forecast peak + reserve margin − accredited supply = shortfall. July printed a 6.8 GW shortfall. A suspended auction does not change the subtraction.If you take nothing else from the paused sale, take the sequence. Prices ran to the cap. The region missed its target. A special procurement was drafted to assign the fix to the new load. Regulators called the draft deeply flawed and stopped the clock. Traders marked the curve higher anyway. Households remain the account of last resort until a rule, a plant, or a private wire says otherwise. That is not a tidy ending. It is the middle of the story, which is exactly where the grid sits this week.
]]>
