Most of the battery queue never gets built

Every year Lawrence Berkeley National Laboratory publishes a report called Queued Up, and every year a large number gets pulled out of it and put in a headline. The 2026 edition, covering data through the end of 2025, landed in June. The headline number this time is that roughly 2,061 gigawatts of capacity is actively seeking interconnection: 1,312 GW of generation and about 749 GW of storage, across more than 8,200 proposed plants.

That storage figure is roughly fourteen times the entire operating utility-scale battery fleet in the United States. Written down that way it sounds like an avalanche is coming. I want to explain why that reading is wrong, using the report's own numbers, and then explain what the queue is actually good for, because it is genuinely useful once you stop treating it as a forecast.

An interconnection queue is an application pile

Before a generator or a battery can connect to the transmission grid, the transmission provider studies what upgrades the grid needs to accept it, and assigns who pays for them. The list of projects that have applied and entered that study process is the interconnection queue. Berkeley Lab compiles these lists from seven ISO/RTOs and dozens of non-ISO balancing areas, covering the large majority of U.S. generating capacity.

The critical property is that entering a queue is an application, not a commitment. It is cheap relative to building a power plant, and a developer can hold multiple positions on multiple sites while deciding which one is worth financing. So the queue measures interest, and interest is a real and interesting signal. It is not a construction schedule.

Here is how thoroughly it is not a construction schedule. Berkeley Lab states it plainly:

The majority (>70%) of interconnection requests are withdrawn. Just 13% of capacity submitted into queues from 2000-2020 had come online as of the end of 2025.

Counted by project rather than by capacity the figure is about 19 percent. That analysis covers 21,927 requests submitted between 2000 and 2020, from 7 ISO/RTOs and 33 non-ISO balancing areas with comprehensive status information. So when you see 749 GW of storage in a queue, the historical base rate says something on the order of one part in seven of that capacity becomes a real battery, and the rest evaporates.

The base rate varies a great deal by region, which matters if you are reading a state-level number. For requests from 2000 to 2020, ERCOT completed 24 percent of projects and ISO-NE 18 percent, while CAISO managed 8 percent and the West about 10 percent. ERCOT is the outlier for a structural reason the report names: under its "connect and manage" approach, generators do not pay for network upgrades, which removes the cost shock that kills many projects elsewhere.

The number that actually moved

Now the part that got much less attention than the 2,061 GW total, and that I think is the real story in this edition.

Storage in the queues did not grow in 2025. It fell. Active storage capacity is 749 GW, down 16 percent year over year. Solar fell 19 percent to 773 GW. Wind fell 19 percent to 220 GW. The one category that grew, and grew violently, is natural gas: up 86 percent, to 253 GW.

Across the whole queue, more capacity left than arrived. Over 750 GW withdrew during 2025 while about 600 GW submitted new requests, and Berkeley Lab notes this was the second year running that withdrawals exceeded new requests.

I would be careful about the causal story here, and so are the report's authors, who list interconnection process reforms, market conditions, transmission constraints and permitting barriers as factors that "may be contributing to this trend." Some of the decline is almost certainly housekeeping rather than retreat: several ISOs have run reforms that force developers to resubmit or drop out, which clears out speculative positions that were never going to be built. A queue getting shorter because the junk was purged is a healthier queue, not a shrinking industry. The report is explicit that withdrawn-year data was only available for 63 percent of withdrawn requests, so even the withdrawal chart understates the totals.

But it does mean one thing for certain: if you were using queue volume as your growth indicator for storage, your indicator just went down while actual storage installations went up. Those two numbers have decoupled, and the queue is the less reliable of the pair.

Five years, and getting longer

The other durable finding is time. The median project that reached commercial operation in 2025 took 61 months from interconnection request to operation. In 2015 that figure was 36 months. In 2008 it was 22 months.

Five years is longer than most of the policy cycles and procurement cycles that surround these projects. It is longer than the useful life of a battery cost forecast. A project that entered a queue in 2021 was priced against a cell market that no longer exists.

That duration also quietly corrupts the queue snapshot itself. Because recent entrants have not had time to fail yet, the current queue is always flattered relative to its eventual outcome. The report is careful about this, noting that "final outcome for projects entering the queues in recent years may not yet be determined; some take 5 or more years from request to COD."

The column that isn't there

There is one line in the report's footnotes that I want to flag, because it connects to something I have written about before. Discussing the storage capacity figures, Berkeley Lab notes: "Storage duration is not provided in interconnection queue data."

So the 749 GW is a power number with no energy number attached. Nobody filing these applications is required to say whether a proposed 100 MW battery is a two-hour unit or an eight-hour unit, and those are different machines doing different jobs. I wrote a whole report on why a gigawatt of storage is half a number, and this is that same gap appearing at the front of the process rather than the back. The queue cannot tell you how much energy is coming, only how much power applied.

Why I care about this particular number

I help design the AI battery-cycling systems for a veteran-owned (HUBZone) energy-storage integrator. I do not own that company and earn nothing from this link. Full policy here.

The reason the queue matters on that side of the work is unglamorous: interconnection timelines set what a control system has to be designed against. If a project's connection date can move by years, the cycling strategy, the warranty math and the revenue assumptions all have to survive a schedule that is not knowable in advance. People outside the industry tend to assume the hard part of grid storage is chemistry. Quite often the hard part is a study queue.

What I could not confirm

This is a data compilation, not an experiment, and it has stated gaps. Berkeley Lab notes that a valid in-service date was available for only 73 percent of operational projects, that withdrawn year was available for only 63 percent of withdrawn requests, and that hybrid storage capacity is estimated for some projects using storage-to-generator ratios from projects that report the two separately. The 749 GW therefore includes modeled components, not only reported ones.

The completion-rate analysis covers requests submitted from 2000 to 2020, which is a deliberately lagged window, and past completion rates are not a guarantee of future ones. The authors say directly that these rates "are variable over time, and trends may be shifting as queue volumes and reforms evolve." FERC Order 2023 and various ISO reforms post-date most of the data, and their effect is not yet visible here. I have applied a historical base rate to a current number, which is a reasonable way to read a queue and is not a prediction.

I did not verify Berkeley Lab's compilation against the underlying ISO queue postings, did not reproduce any figure, and did not inspect the accompanying dataset or the interactive tools at the project site. I read the published slide deck. Figures quoted here come from its text; I have not quoted values read off charts.

The report also excludes large loads. Data center interconnection requests sit in separate queues that are not included, which matters because the demand story and the supply story are being counted in different places.

None of this is my own research. My published work is in microwave spectroscopy. My grid background is engineering and operational rather than in transmission planning, and the analysis here belongs to the Berkeley Lab authors.

The signal

A queue number is a measure of intent, and intent is cheap. That is not a criticism of the queue, which does its job well: it tells you what developers wanted to build, where, and in what direction their appetite is moving. Read as a sentiment indicator it is excellent, and this year it says gas appetite roughly doubled while every clean category pulled back.

The failure is only in the translation. When 749 GW gets reported as a wave of batteries about to land, the reporting has silently converted an application pile into a delivery schedule, and dropped a 13 percent completion rate and a 61-month median on the way. The honest sentence is longer and less exciting: this much capacity applied, roughly one part in seven of applied capacity has historically been built, it takes about five years, and the storage share of the pile just shrank for the first time in a while.

Ask of any pipeline number: is this what someone intends, or what someone is building? Those are different quantities, and only one of them keeps the lights on.