This summer, Texas accumulated requests to connect roughly five electrical grids' worth of data centers, most of which will never be built.
ERCOT, which runs the Texas grid, reported in early August that developers had filed requests to connect about 474 gigawatts of large new electric load — more than five times the state's all-time record peak demand of about 91 gigawatts, with roughly 90 percent of the requests tied to data centers. Everyone involved understands most of it is fiction. ERCOT's own chief executive has said the forecast is higher than expected load growth. On August 3, Governor Greg Abbott ordered every data center in the interconnection process audited before another one advances, writing that any project failing to comply "must be denied connection to the Texas grid. Simply put, Texans must come first." In September he extended the freeze to environmental permits.
Texas is the vivid case, but the condition is national — and it matters more than almost anyone outside the industry realizes. Somewhere inside this administrative machinery is the institution that now rations the binding constraint on the AI buildout: not chips, not capital, but time-to-power. Because the queue is no longer a waiting room. It is the market.
What's being traded is priority access to the grid — really, time. The buyers are developers and hyperscalers, bidding against each other with deposits, collateral, and a willingness to lose both. There's no seller in the usual sense; the grid operator rations a fixed supply of interconnection, and the price of admission is whatever it costs to survive the process. It clears at decision points, where the undercapitalized get filtered out and the committed pay up. If that framing is right, it points to the questions worth asking: who does a market like this select for, and who ends up paying for the grid?
If you want to connect anything big to the grid — a power plant, a data center, a factory — you have to ask permission. The grid operator studies whether the wires near you can handle it, what upgrades would be needed, and who pays for them. This is called interconnection, and for most of the industry's history it was plumbing: administrative, boring, first-come, first-served.
The plumbing broke in the most predictable way imaginable. Under first-come, first-served, a queue position cost almost nothing and was worth a great deal: it was a free call option on future grid capacity. If power demand materialized near your project, you held a golden ticket. If it didn't, you'd spent little more than the application fee. So everyone filed — the serious, the speculative, and the fictional. Filing early and often wasn't gaming the system. It was the system working exactly as its incentives dictated.
Berkeley Lab keeps the authoritative scorecard, and the latest edition is staggering for a simple reason: the queue now holds more power plants, on paper, than the country has ever built. At the end of 2025, some 2,061 gigawatts of generation and storage capacity sat in interconnection queues across roughly 8,200 projects — more than the entire installed generating fleet of the United States, which is about 1,374 gigawatts. The country is being asked to connect more than it has ever built, all at once. The mix tells you where capital wants to go: 773 gigawatts of solar, 749 of storage, 253 of natural gas (up 86 percent in a single year), 220 of wind.
And most of it will never be built. Of every project that requested interconnection between 2000 and 2020, only about 19 percent — 13 percent of the capacity — had reached commercial operation by the end of 2025. More than 1.8 terawatts have been withdrawn since 2020; in 2025 alone, withdrawals topped 750 gigawatts while only about 600 gigawatts of new requests entered. Only around 311 gigawatts, about 12 percent of the active queue, had signed interconnection agreements — the closest thing the queue has to a commitment device. The typical project that did manage to get built in recent years waited something like four to five years from request to switch-on.
This is not a pipeline. A pipeline delivers things. This is purgatory with paperwork — and purgatory, it turns out, had a business model. The speculative mass slowed the studies for everyone, which lengthened timelines, which caused more withdrawals, which forced the restudies of everything queued behind them. A doom loop, administered one serial study at a time. And it had consequences beyond long waits. Viable projects died in the queue, squeezed between upgrade costs that ran 44 percent higher in 2019–2023 than in the five years before and timelines no business plan could survive. In effect, the country was rationing new power through waiting times. That couldn't hold — not with the scale of construction now being attempted — so regulators rebuilt the process around money: deposits to enter, collateral to stay, forfeiture for leaving late. And a filter with money in it started behaving like a market, whether or not anyone quite intended that.
In July 2023, the Federal Energy Regulatory Commission unanimously issued Order No. 2023, the biggest overhaul of generator interconnection rules in two decades. The diagnosis was blunt: the queue was clogged with speculative, commercially non-viable requests. But the cure was plumbing. Study projects in clusters instead of one by one. Replace first-come, first-served with first-ready, first-served. Require real site control, real study deposits, and commercial readiness deposits to enter and to stay. Penalize withdrawals. Penalize transmission providers for late studies. This July, the D.C. Circuit upheld the entire rule, writing that "FERC, in short, identified a real problem in the energy generation system, selected a reasonable alternative to the status quo, and provided rational policy reasons for its decision."
PJM, the grid operator for thirteen eastern states, is the live demonstration. Years ago it stopped taking new interconnection applications entirely while it worked through a backlog north of 300 gigawatts. When its reformed queue reopened this year, 811 projects applied for roughly 220 gigawatts; 715 were accepted into the cycle for 201.5 gigawatts; 617 entered the first study phase in late September for about 169 gigawatts. The screen filtered out a quarter of applicants before a single study was performed — the mechanism working as designed. The first real money decision, when developers must post additional security or walk away, lands at the end of November. Around the screen, PJM has built a whole apparatus of speed lanes: surplus interconnection service for unused headroom at existing plants, provisional service, transfers of capacity rights from retiring generators to replacements at the same site, a one-time initiative to fast-track eight or nine gigawatts of reliability-critical projects. The operator is, in effect, running a bazaar of quasi-markets around the main event.
PJM's price list is where the story turns.
To enter the queue: a study deposit of $75,000 to $400,000 depending on project size, ten percent of it non-refundable from the start. A readiness deposit of $4,000 per megawatt. Documented control of 100 percent of your site for at least a year. At the final decision point: security equal to 100 percent of your allocated network-upgrade costs, plus three years of site control, before interconnection-agreement negotiations even begin. Miss a decision point and escalating portions of your deposits are forfeited. Forfeited deposits are pooled across the cycle to backfill the network upgrades stranded by late-stage withdrawals. The quitters literally fund the grid.
There is even a circuit breaker, the way a real exchange has circuit breakers: if your allocated upgrade costs jump more than 25 percent — and more than $10,000 per megawatt — at the second decision point, you get your deposits back. The designers knew they were building something with prices in it. They just didn't call it a market.
But that is what it is. It has entry prices, collateral requirements, forfeiture, and something very close to a property right: your queue position determines whether you get studied, what share of shared upgrade costs you absorb, and whether you survive to the end. Nobody held a hearing on whether America should have a market for interconnection rights. FERC built a filter to kill speculation, and a filter with money at stake becomes a market. That is the mechanism — designed as a filter, functioning as a market.
Morgan Stanley's power analysts have reportedly estimated that a one-year head start on electricity is worth about $4.50 per watt to a data center developer. Time-to-power is the commodity now, and the queue is the exchange where it trades. A hyperscaler doesn't primarily compete on chips or capital anymore. It competes on interconnection.
And like every market, this one has an allocation criterion — a rule for who gets the goods. Under first-come, first-served, the queue allocated on punctuality: who asked first. Under first-ready, first-served with escalating collateral, it allocates on capital and readiness: who can post $4,000 a megawatt without blinking, hold site control for years, and swallow the full cost of network upgrades at the final decision point.
The winners select themselves: deep-pocketed developers and hyperscalers, obviously — the parties for whom a seven-figure deposit is a rounding error. Incumbents who can transfer interconnection rights from a retiring plant to a new one at the same site, converting a dead asset into a live queue position. And, notably, natural gas: nearly 100 of the 201.5 gigawatts accepted into PJM's first reformed cycle is gas-fired, the largest single fuel share. That is not a coincidence and not quite a conspiracy either. Under a market that allocates on speed, the fuel that can be permitted and built fastest is the fuel that wins. Gas doesn't have to be anyone's favorite. It just has to be ready.
The other side of the ledger is less comfortable. Smaller developers, for whom escalating at-risk deposits are existential rather than annoying. Technologies with long development lead times, which are structurally worse at demonstrating "readiness" on a regulator's timetable. And the speculative option-plays — the intended casualty, the one category everyone agrees should die. The trouble is that the filter cannot tell a speculator from a small developer with a good site and thin capitalization. Collateral requirements never can. They are a test of balance sheets wearing the costume of a test of seriousness.
Texas is the control group that proves the mechanism, because Texas spent years running the old system on the load side. Where queue position stayed cheap, fiction piled up five grids deep. So Texas is now building the same filter for loads that FERC built for generators. Senate Bill 6 requires large-load customers to demonstrate site control, self-fund their grid upgrades, disclose duplicate interconnection requests across the state, and accept curtailment during grid stress. ERCOT's new Batch Zero process studies big loads in clusters — first-ready, first-served, the Texas translation of Order No. 2023. And the governor's audit functions as a solvency test: demonstrate site control, funded upgrades, and full disclosure of load and water use, or be denied connection regardless of your place in line. Different state, different politics, same discovery. When the line is free, everyone gets in it, and the line stops working.
The same apparatus is now being turned on the buyers, which is where the story stops being about power plants and starts being about the entire economy. On June 18, FERC issued six show-cause orders — one for each regional grid operator — aimed at interconnection rules for large loads: data centers, advanced manufacturing, anything drawing more than 50 megawatts. The orders demand faster and more transparent study processes and, crucially, clear rules for who pays for the upgrades new loads trigger. As Commissioner LaCerte put it in a concurrence, "We are at an inflection point in the history of American energy infrastructure. The extraordinary and rapid growth of large loads in RTOs and ISOs is faster than the current system can handle." The proceedings were paused in August at the operators' request, to give them time to write real tariff proposals, and are expected to resume around mid-November — days before PJM's developers face their first real-money decision at the end of the month. The market is about to hold its first trading session.
Underneath every queue fight, though, is the same older fight: who pays. The interconnecting customer is supposed to cover the network upgrades its project triggers — the "but-for" test — but when a cluster of projects jointly triggers a billion-dollar transmission build, the allocation is politics done by spreadsheet. And the sums are getting large enough to notice.
In July, PJM proposed a one-time "backstop" auction to procure 6.8 gigawatts of capacity for reliability, at an estimated cost of up to $20 billion, with contracts running as long as fifteen years — an extraordinary departure for a market built on annual capacity auctions. On September 29, FERC partially approved the plan but suspended it for five months, ruling that the cost allocation might be unjust and unreasonable, and pointedly suggesting the costs be assigned using updated load forecasts so that new large loads bear them. "Existing customers should not be left paying costs attributable to new demand," Commissioner See wrote in concurrence. The chairman was blunter: "This commission will not be forced into accepting a deeply flawed, eleventh-hour procurement mechanism with billion-dollar implications for consumers." For context on what's at stake: data centers were $6.5 billion of the $16.4 billion total in PJM's last capacity auction — 40 percent, paid by everyone.
The parable is Susquehanna. In March 2024, Amazon paid $650 million for a data-center campus next to Talen Energy's nuclear plant in Pennsylvania and tried to take power straight from the reactors, behind the meter. Utilities objected that the arrangement let one giant customer rely on the grid without paying for it; FERC rejected the expanded deal 2–1 in November 2024, and the appeal was dismissed that January. Then the whole thing was rebuilt the boring way: a seventeen-year, grid-delivered power purchase agreement for up to 1,920 megawatts through 2042, worth something like $18 billion notionally, which took effect this April. Strip away the electrons and the story is pure queue economics. "Behind the meter" was never about physics. It was about who pays for the shared grid — and the answer, in the end, runs through the market.
Every era rations its scarcest input through whatever institution happens to be lying around. Ours happens to be interconnection paperwork: procedures designed for a slower grid, in a slower century, to answer a duller question — in what order do we study these power plants?
It now answers a different question: where does the AI buildout physically happen, which utilities get to grow their rate base, who earns the return on the largest capital deployment in a generation. And it answers not by willingness to pay for electricity, but by willingness to post collateral for a place in line. The old system — free options, five-year waits, fiction piled five grids deep — was worse. The point is to notice what we have built.
Nobody designed a market for the right to plug into America. We got one anyway, assembled out of deposits, deadlines, and forfeitures. It now rations the binding constraint on the AI buildout, and almost nobody outside the industry has ever heard of it. The queue was supposed to be plumbing. It turned out to be the market. The work ahead is deciding whether it's the market we want: who it should select for, what it should price, and who should pay.
Sources consulted for this essay: Lawrence Berkeley National Laboratory, "Queued Up: 2026 Edition" (slide deck); PJM Inside Lines reporting on Cycle 1 of the reformed interconnection process; FERC Order No. 2023 and the D.C. Circuit's July 2026 decision in Advanced Energy United v. FERC; FERC's June 2026 show-cause orders on large-load interconnection (e.g., EL26-72-000) and related concurrences; the August 2026 directive letter from Governor Abbott to the PUCT and ERCOT; PJM's July 2026 backstop procurement filing and FERC's September 2026 order; reporting from Reuters, Utility Dive, E&E News, PowerMag, and Latitude Media.