Oncor's transmission system carried a record 33.3 gigawatts on August 24. The Dallas utility's connection queue, as of September 30, held 795 large-load requests. Data centers account for 293 gigawatts of them. Another 19 gigawatts comes from industry, including 4 gigawatts of crypto mining.
ERCOT set its own record on July 22, at 91.1 gigawatts. Oncor's peak rose from 31.4 gigawatts in 2025, a gain of roughly 6% in a single summer. The requests in the queue are not load. They are letters asking for the right to become load.
That distinction is where the argument starts. Texas opened a process this year that classifies large customers before it studies their requests. Oncor submitted 45 gigawatts into it, about 135% of the utility's entire system peak. ERCOT conditionally classified 24 gigawatts as base load, 10 of it already energized, and 21 gigawatts as studied load. The classifications survive only if an audit in December confirms the developers were telling the truth.
The Line Is Longer Than the Grid
The national backlog is bigger than any single state's imagination. Lawrence Berkeley National Laboratory counted about 2,061 gigawatts of generation and storage seeking connection at the end of 2025, spread across roughly 8,200 projects. That is 1,312 gigawatts of generation and 749 gigawatts of storage, and it is 10% below the prior year's total.
Withdrawals explain the drop. More than 750 gigawatts left the queues in 2025. Natural gas added 253 gigawatts of active requests, up 86%. Solar fell 19% to 773 gigawatts, storage slipped 16% to 749, and wind dropped 19% to 220.
Another 549 gigawatts already holds a draft or executed interconnection agreement and still is not operating. The pile includes 256 gigawatts of solar, 161 of storage, 76 of wind, and 45 of gas. An agreement is a milestone. It is not a switch.
Time is the part that compounds. Berkeley Lab puts the median duration from request to commercial operation above five years for projects that finished in 2025. Brattle Group and Grid Strategies, writing for Advanced Energy United in June, put the median queue stay at 60 months, against 20 months in 2005.
Only 13% of the capacity that asked for a connection between 2000 and 2020 had reached commercial operation by the end of 2025. Three quarters of it had withdrawn. A queue that converts one request in eight is not a pipeline. It is a filter with a five-year delay.
The equipment behind the filter has its own line. As we wrote in November, GE Vernova paid $9 billion for Trench, a business built around the 5% of grid hardware that decides where the rest can connect.
ERCOT's Batch Zero is the first US test of classifying large load before the study, and Oncor alone put 45 GW into it.
The December 10 audit report is the filter. Every later date depends on what survives it.
Pole One: Wires as Public Infrastructure
The first camp treats connection capacity as a public good. If the wires are shared infrastructure, the argument goes, then planning them is a public function, and the cost belongs on the system rather than on whichever developer happens to arrive first.
FERC Order 2023 moved the country partway there. It replaced first come, first served with first ready, first served, and it clusters projects so one study answers many requests. Advanced Energy United's June scorecard credits the order with real progress while noting that withdrawals remain high and that fast-track workarounds have mostly helped expensive thermal projects.
The cost of delay is not abstract. One analysis cited by the group found that PJM's 2024 capacity auction could have cost consumers $7 billion less with a working connection process. Connecting just 15% of the projects then waiting would have added more than 10 gigawatts.
Part of the delay is clerical. Berkeley Lab reports that a pilot test of automated studies using Pearl Street Technologies' SUGAR software cut a study timeline by 98% with very high accuracy. SPP launched a consolidated planning process in March 2026 to join transmission planning with interconnection review, and MISO now caps total queue capacity. These are process fixes. They do not add copper.
The physical version of the argument is being built in the Aegean. Nexans signed a subcontract on November 6 to supply a 420-kilometre cable system for the Corinth-Kos link, a 1 GW, 320 kV HVDC connection between the Dodecanese islands and mainland Greece. The full project runs about 1,260 kilometres under water and 30 kilometres on land. Nexans will use the cable lines from the end of 2026 into mid-2028.
Greece is not Texas. The point is the sequence. A transmission operator decided the link was worth building, then bought the cable. In most US queues, the developer files first and the wire is discovered later.
Pole Two: Make the Requester Pay
The second camp starts from the bill. Oncor holds about $6.0 billion of customer collateral against active generation and large-load requests, including $2.0 billion tied to its Batch Zero projects. The money exists so that ratepayers are not left holding transmission costs after a project disappears.
That is the strongest case for charging requesters: a queue full of speculative letters wastes study capacity and steel. Collateral turns a request into a liability. Oncor's own filing says a request is not a guarantee that load will materialise, which is precisely why the deposit exists.
ERCOT's verification shows what happens when a regulator asks for evidence. The state-ordered review sent questions to 465 loads and got 433 responses, 93% of the total, covering about 177 gigawatts, or 97% of the requested capacity. All 245 large loads of 75 megawatts or more answered, representing about 166 gigawatts. Thirty-three projects of more than 1 gigawatt each make up 36% of the capacity. West Texas alone holds 41%, or 73 gigawatts.
The audit found the usual gaps. ERCOT reviewed responses from 257 developers and about 5,140 questions, then sent second-round requests to 80% of them. The recurring problems were missing proof of site control, unclear affiliate relationships, unproven long-lead equipment orders, and uncertainty over how much financial security had actually been posted.
Smaller loads are next. ERCOT issued 132 requests to utilities on October 26 covering loads of 10 to 75 megawatts, with answers due November 6. The concern written into the filing is that medium loads could absorb capacity already allocated to base load in Batch Zero, leaving little room for ordinary growth.
A standard large-load contract is also under discussion at the PUC. That turns the queue from a line into a lease with terms.
The Middle Is a Contract
Neither pole survives contact with the schedule. Socialised planning cannot compress a five-year study cycle into two years, and requester-pays does not build a 765-kilovolt line for a customer who may never energise.
What is actually being negotiated is the contract: who posts security, who funds the upgrade, and who absorbs the loss when a project dies. ERCOT has published the sequence. Initial validation closed on July 24. Conditional classifications came on September 3. The audit report is due December 10, with a capital plan update expected in February 2027. Nominal Batch Zero study results are targeted for April 9, 2027, developer commitments for summer 2027, and a final transmission plan for the second half of 2027.
Generation and storage waiting for a connection
About 8,200 projects at the end of 2025, split between 1,312 GW of generation and 749 GW of storage. · Berkeley Lab, 2026
The money is moving ahead of the verdict. Oncor spent about $6.3 billion through September against a roughly $9.0 billion plan for 2026, including $1.3 billion on transmission in the third quarter alone. The utility filed 16 new certificates of convenience and necessity this year. Eight have been approved.
Its 2027–2031 base capital plan is expected to grow roughly 20%, a figure that excludes Batch Zero work entirely.
Median wait from request to commercial operation
The median queue stay was 20 months in 2005. · Brattle Group and Grid Strategies, 2026
The arithmetic explains the caution. Forty-five gigawatts of conditional load against a 33.3-gigawatt system peak is a 135% increase in demand sitting on a process that has not yet passed its own audit. If a third of it fails verification, the wires still get built to a plan that assumed all of it.
What December Decides
December 10 is the first real filter. ERCOT must file its verification report with the PUC that day, and the commission meets on December 17 to consider good-cause exceptions for developers who missed deadlines. Loads that fail verification do not vanish from the map. They leave the base case, which is where the capital plan and the transmission studies live.
After that, the calendar belongs to commitments. The April 9, 2027 study results set what Batch Zero actually costs. The summer 2027 deadline decides who signs. The final plan in late 2027 turns signatures into steel and cable.
None of that answers the question the two camps are really arguing about. If the load shows up late or not at all, the upgrade is already in the rate base, and the recovery runs through the collateral, the tariff, or the next rate case.
The queue is not a line of projects waiting for permission. It is a five-year bet on which letters become buildings, priced by whoever holds the risk when the bet loses.