Interconnection-queue figures do not measure data-center waits

Berkeley Lab’s Queued Up research tracks proposed generation and storage seeking transmission connections. It does not measure the time a data center waits to receive electric service. A campus announcement needs a different set of records before a resident can judge its power timetable.

This is Part 6 of The Power Plant Next Door. This comparison uses published research and utility rules, rather than a site visit or a developer interview.

The Berkeley figures have specific populations

The 2026 Queued Up edition reports 8,244 active requests as of December 31, 2025, representing 1,312 gigawatts of generation and 749 gigawatts of storage, 2,061 gigawatts in all (Berkeley’s full report: the 8,244 appears on slides 7 and 13, and the 1,312 and 749 on slide 4). Those are proposed supply and storage resources, rather than a count of data-center customers. Berkeley says so itself on slide 2: its data “only include resources that supply electricity to the transmission grid (i.e., generation and storage); there are separate queues for large loads and those are not included in this report.” A gigawatt measures power capacity; the total is not an annual electricity-consumption figure.

For requests submitted in 2000-2020, 13% of capacity had reached commercial operation by the end of 2025. The figure divides capacity that reached operation by the capacity originally requested; it does not count projects equally. Calling it 13% of projects changes the denominator and can change the conclusion; Berkeley’s slide 4 puts it as “about 19% of projects (13% of capacity).” It also does not give the probability that a newly announced campus will open.

Berkeley reports that the median project built in 2025 took 61 months (5.08 years; 61 divided by 12) between a grid-connection request and commercial operation, when a generation or storage project begins operating. The median is the middle of the recorded times, and Berkeley’s footnote on the same slide says “valid in-service date was only available for 73% of all operational projects,” which limits what the median can describe. It is neither a waiting-time estimate for every request nor a measured delay for data-center service. Withdrawn projects and unfinished requests do not become completed observations simply because their applications were filed earlier.

Identify the type of application

A generator adds electricity supply; a data center ordinarily seeks service as a large electricity customer. The word interconnection can appear in both settings. The application type and the governing process matter more than the label alone. A campus with associated generation may also need records for that generating project; the generator’s queue position does not itself establish the campus’s service date.

In Texas, the Electric Reliability Council of Texas (ERCOT) publishes separate Planning Guide chapters for generator interconnection or modification and for large-load interconnection or modification. As of October 29, 2026, its index lists the generator chapter (Section 5) dated August 1, 2026 and the large-load chapter (Section 9) dated July 11, 2026. That separation illustrates why a generator dataset cannot substitute for a customer’s service record. The applicable utility and regional requirements must be checked for the actual location.

A records test for a campus timetable

The following is a reporting checklist, not a finding about an unnamed developer. It asks which documents support each milestone. A requested date, an estimated date and a committed date should be labeled separately. If the record is unavailable, that limits what can be concluded; it does not prove misconduct or establish that the project will fail.

A useful public answer identifies the document, its date and the milestone it supports. An application number alone cannot answer all of these questions. Neither can a national queue median. Where the utility serves a campus in phases, the public description should identify the phase rather than treating first energization as proof that the full announced capacity is operating.

Virginia: what the billing order says about cancellation risk

The Virginia State Corporation Commission’s November 25, 2025 order approved Dominion’s GS-5 rate class, effective January 1, 2027. These are Dominion Virginia provisions for qualifying large-load customers. The table lists the terms that decide who carries the cost if a campus is delayed, shrinks or is cancelled.

The threshold is inclusive. The order says “25 megawatts (‘MW’) or greater,” so a customer at exactly 25 megawatts falls in the class if its load factor is at least 75%.

To see what the collateral term means in dollars, take a hypothetical 100-megawatt customer; this is our arithmetic, not a real contract. At $1.5 million per megawatt the requirement is $150 million. With the maximum 70% reduction for credit it is $45 million. One-fourteenth of the requirement is released each year, which is $10.71 million a year on the full amount and $3.21 million on the reduced amount. The actual figure for a campus is in its executed agreement. On the same hypothetical 100 megawatts, the capacity-reduction terms would let the customer cut contracted capacity by 20 megawatts at no cost, and by a further 30 megawatts if another customer agrees to take it over, each cut on its own 36-month notice. On our reading of the order, the 30% is a further share of the same contracted capacity, so together that is up to 50 megawatts of the 100 (20 plus 30); the order does not work out that sum.

These financial obligations answer who bears cost if a customer leaves. They do not establish that a campus can be energized on its requested date, and the terms above do not cover who funds network upgrades for a particular campus, which remains a question for the agreement and the study. The queue process is the record that does decide order and timing. DatacenterDynamics reported that Dominion has filed a proposed process counting 70 gigawatts of large-load delivery points in its queue (DatacenterDynamics’ wording is “approximately 70GW”), with 25 gigawatts assigned projected connection dates through the end of 2031 and a further 45 gigawatts under study; we did not open the filing itself. Dominion’s own May 2026 report on the GS-5 class and large-load customers gives the same two figures in its words, 25 gigawatts with assigned energization dates through 2031 and 45 gigawatts for which it has “not yet offered future connection dates,” and states no total; the two add to 70 (our sum). A hearing should ask for the terms, the queue position and the service evidence, and should not treat approval of the billing class as approval of a connection. The SCC also published a February 2026 summary of its data-center work.

Hardware needs its own evidence

The national generator dataset contains no delivery schedule for a campus’s servers. A general supplier-capacity announcement does not identify what a particular customer ordered or when that order will arrive. For an opening-date claim, the useful question is which procurement record supports the proposed phase. If delivery information is not public, the uncertainty should stay visible instead of becoming a prediction that a building will be partly empty in 2028.

Update, October 29, 2026: The earlier story presented generator-queue statistics as a measured data-center wait and described a capacity share as a project share; this version separates the populations and takes the 61-month median, the 19% project share, Berkeley’s footnote that a valid in-service date existed for 73% of operational projects, and the 8,244 requests from Berkeley’s 2026 report (slides 7, 13, 4 and 2 respectively cover the count, the active-request tables, the split and the large-load exclusion). The Virginia section was rebuilt from the commission’s order (printed page numbers throughout): the threshold is “25 megawatts or greater,” and the collateral, exit fee, load ramp, post-term charges, exemptions, capacity-reduction terms (page 27), the page-26 exclusion of PJM capacity expense from the minimum generation demand charge, and a 100-megawatt dollar example are listed. Chip sell-out and campus-delay predictions were removed, the 5.08-year conversion and DatacenterDynamics’ queue figures are stated in their own terms, and the ERCOT index dates were rechecked. Later on October 29: the 50-megawatt example is labeled as a worked reading, the 70 gigawatts is set beside Dominion’s own report of 25 and 45 gigawatts, and the table notes that no later order changing the terms turned up.

Editor written. AI used for spelling and grammar only.