Grid interconnection queues hold 2,061 gigawatts — 75 percent of past requests were withdrawn

What is happening?

At the end of 2025, 2,061 gigawatts of generation and storage capacity were actively waiting for permission to connect to the electric grid in the United States. The queue shrank 10 percent during that year, but it shrank because projects gave up rather than because they were connected. Of the capacity that entered a queue between 2000 and 2020, 13 percent had reached commercial operation by the end of 2025 and 75 percent had been withdrawn.

The wait is measured in years. In 2025 the median time from submitting a request to signing an interconnection agreement was well over three years, and the median time to reach commercial operation exceeded five years. The composition of the line also flipped in a single year: natural gas capacity in the queue rose 86 percent to 253 gigawatts, while solar fell 19 percent to 773 gigawatts, storage fell 16 percent to 749 gigawatts and wind fell 19 percent to 220 gigawatts.

By the middle of 2026 the active queue was down to 1,919 gigawatts. In that half year 245 gigawatts across 1,319 projects were withdrawn and 224 gigawatts of new applications came in.

Whose problem is this?

The party that sets the deadline is not the party that runs the clock. FERC wrote a single national rule; more than 50 separate transmission providers each file their own compliance plan and each operate their own queue, and none of them can build the transmission a study calls for without a state siting decision.

Where does this problem end?

The boundary matters here because this is not a permitting problem and not a supply chain problem. It is the step where a project asks the grid operator what connecting would cost, and then waits for the answer.

What is the state now, and what should it be?

Now

Should be

FERC Order No. 2023 gives a transmission provider 150 calendar days to complete an initial cluster study, and a late study now carries a penalty of USD 1,000 per business day for a cluster study and USD 2,500 per business day for a facilities study. That is the only binding federal clock in the chain, and it covers one step of the process rather than the whole path. One large multi-state RTO tells applicants under its reformed process to expect one to two years for a generation interconnection agreement.

How big is it?

What is measured is the capacity. The line held 2,061 GW at the end of 2025 and 1,919 GW at the end of June 2026, and in the first half of 2026 more capacity left than arrived. One monetized estimate exists for one region: an advocacy analysis argued that connecting 15 percent of the capacity proposed in the queue of one large multi-state RTO would have supplied more than the roughly 10 GW of accredited capacity needed to return the 2025 and 2026 auction to historical cost norms, worth as much as USD 7 billion in that single auction. The method behind that figure was not verified.

Fills with researchthe affected population was not researched. The queue is counted in gigawatts and in projects, and no source we opened converts a gigawatt-year of interconnection delay into a number of people who paid more for electricity or went without service. The affected population is therefore left `not-derivable`. Details in `population/us-grid-interconnection-queue.json`.

Under what conditions does it arise?

Four conditions hold this state in place, and no single party controls any of them.

1. The queue is a shared line. What one project is told it must pay depends on which other projects sit ahead of it, so every departure forces a restudy and moves the numbers for everyone behind. The withdrawal rate and the study delay are the same fact seen twice. 2. Entry was cheap relative to the option it bought. A request costs far less than the project it reserves a place for, so filing many requests and keeping the best has been rational for each applicant and has clogged the queue in aggregate. Order No. 2023 raised the price of a place in line — a USD 5,000 application fee, a study deposit between USD 55,000 and USD 250,000, and control of 90 percent of the site acreage at submission. 3. One author, more than fifty implementers. FERC wrote the rule, but each transmission provider files its own compliance plan on its own schedule. In July 2025 FERC accepted the compliance plan of one large multi-state RTO only in part and gave it 60 days to propose further changes, two years after the rule issued. 4. Finishing the study is not the same as finishing. More than 40 percent of capacity is still withdrawn after an interconnection agreement has been signed, so speeding the study up does not by itself deliver a power plant.

What has been tried?

What was found?

Why is it still unsolved?

Coordination failure — the wait is produced by how four sets of parties interact, and none of them can shorten it alone.

The queue is a shared resource whose answers are interdependent. A cluster study prices the network upgrades a group of projects would need together, so the number any one applicant receives depends on who else is in the group. When an applicant leaves, the study is redone and every remaining number moves. That makes speculative entry cheap for the individual and expensive for everyone else, and it makes late withdrawal the most damaging thing a participant can do. Order No. 2023 attacked exactly this by raising the cost of entry and by penalizing the study side for lateness, and the entry discipline has visibly worked — the median project now abandons the queue after 21 months rather than 40.

But the rule has one author and more than fifty implementers, and the pieces do not move together. Each transmission provider files its own compliance plan on its own schedule; two years after the rule issued, FERC accepted the compliance plan of one large multi-state RTO only in part on 2025-07-24 and directed a further filing within 60 days. Meanwhile the transmission a study calls for is sited by states, not by FERC and not by the grid operator, so the operator can complete a study on time and still be unable to deliver the connection.

The measured improvement is therefore ambiguous, and that ambiguity is itself part of why the problem persists. The queue got shorter mainly because projects left it. Withdrawals exceeded new applications in the first half of 2026. A shorter line and a faster median are exactly what a healthy process would produce, and they are also exactly what a process that discourages everyone from entering would produce.

What observation would mean it is solved?

Candidates — (a) the active queue falls toward the volume actually being connected each year (b) the median time from request to commercial operation falls below three years (c) the share of entering capacity that eventually operates rises well above the historical 13 percent.

(a) alone is satisfied by mass abandonment. The queue fell 10 percent in 2025 and another 3.5 percent in the first half of 2026, in a period when withdrawals of 245 GW outran new applications of 224 GW.

(b) alone is satisfied by survivorship. If readiness screens push out everything that would have been slow, the median falls without anyone being connected faster. The drop from 5.5 to 4.8 years is not separated from that effect in anything we read.

(c) alone can be met by admitting less. A queue that only lets in pre-approved projects would post a high completion rate and say nothing about how much generation the country actually added.

Read all three against the capacity energized per year. One large multi-state RTO connected 2.1 GW in 2025 while holding roughly 46 GW of signed agreements.

What is it connected to?

Fills with researchtransmission expansion and state siting · data center load growth · capacity market prices · equipment supply chains for transformers and turbines. Relation type and evidence grade were not confirmed in this round.

What these sources do not say

• How much of the 75 percent withdrawal the queue itself caused. Financing, offtake contracts, siting and equipment supply also kill projects, and none of the sources read here separates those causes from study delay. Without that split the queue cannot be assigned a share of the loss.
• What network upgrade costs the studies actually assigned. The 2024 cluster study recorded above shows 28.5 GW leaving immediately after the study results were issued, which implies the cost number is decisive, but the distribution of those numbers is not published in anything we opened.
• Whether the late-study penalties have ever been assessed. Order No. 2023 sets a per-day amount but we found no count of penalties actually billed or paid, and no transmission provider named as having paid one.
• The distribution behind the medians. A median of 4.8 years says nothing about the tail, and regional queues differ by a factor of more than two in volume, so a national median may describe no region.
• Whether the swing to gas is new demand or reshuffling. Gas rose 86 percent in a year when every other resource fell, and no source distinguishes genuinely new gas projects from capacity that displaced other requests in the same interconnection position.
• What became of the capacity that left. No source tracks whether a withdrawn project re-entered a different queue, so the withdrawal totals may count the same megawatts more than once.

See the evidence

The Berkeley Lab report itself was not read directly — the PDF exceeds the fetch size limit and the emp.lbl.gov landing pages return 403 — so the end-of-2025 figures come from two independent summaries of it, and the OSTI record was opened to confirm the publication date, the authorship and the coverage claim. Two numbers cross-check cleanly: the 2,061 GW reported for end-2025 as a 10 percent decline implies about 2,290 GW for end-2024, which is exactly the figure the summary of the prior edition gives.

Two discrepancies remain open. The 2024 withdrawal total is reported as more than 700 GW in the summary of the Berkeley Lab edition and as 296 GW by the independent tracker, a gap we could not resolve because the counting bases are not stated side by side. And the tracker puts the queue at 1,919 GW in mid-2026 after a 3.5 percent decline from 1 January, which implies about 1,989 GW at the start of 2026 rather than the 2,061 GW Berkeley Lab reports for end-2025; the two datasets clean and de-duplicate the same operator filings differently. The same gap shows in the gas total, which Berkeley Lab puts at 253 GW for end-2025 and the tracker at 247 GW, so the tracker rise to 305 GW by mid-2026 is measured against the smaller base. Figures attributed to a nonprofit research institute, to a commercial data provider and to an industry advocacy group are graded low because each publishes its own analysis and none of the underlying methods was verified; the third of those is additionally writing about a market it participates in. This is a Path A output, so observation_refs is empty and provenance_mode: press-derived.

People affected

Estimated range Not derivable

The reason and what is missing are listed under “What is missing” below