Phantom data centers: the record test
Why grid queues fill with data centers that never get built, and the five public records that separate a real project from a speculative request.
On this page8 sections
A phantom data center is a project that exists in a request or an announcement and nowhere else. It holds a place in a utility’s forecast or a grid operator’s queue, and it never reaches a permit, a contract or a meter.
The public numbers show how much of a queue that can be. They also show which records thin it out, and those records are the test this article sets out.
Why phantom data centers exist
Requests are cheap compared with what follows them. A developer can ask several utilities for power for the same project, and each request appears in a forecast until it is withdrawn.
Texas law now names the duplicate problem directly. Senate Bill 6 (2025) requires each large load customer to disclose “whether the customer is pursuing a substantially similar request for electric service in this state the approval of which would result in the customer materially changing, delaying, or withdrawing the interconnection request.” The provision is aimed at requests that duplicate one another across utilities.
How large the gap is in Texas
The gap between requests and running load in ERCOT is about two orders of magnitude. ERCOT told its board in June 2026 that it was tracking about 438 GW of large-load interconnection requests. Its September 2026 overview reports 9,456 MW approved to energize and an observed peak of 4,132 MW from those loads.
The three figures are separate snapshots. Approved loads include sites that applied years ago, and the request total includes sites that will never apply for anything else. For scale, ERCOT’s historical peak for the whole grid is about 86 GW, by its own count in a 2025 filing.
What ERCOT does with the forecasts
ERCOT already discounts data center requests when it plans transmission. In a May 2025 filing at the Public Utility Commission of Texas (PUCT), it said utility forecasts “suggest an increase in demand to 218 gigawatts (GW) by 2031”, several times its historical peak.
The filing says “the vast majority of load” in those forecasts was attested in a letter from a utility officer “rather than being supported by an interconnection agreement.” ERCOT cut data center loads to 49.8% of requested MW and officer-letter loads to 55.4%, and pushed officer-letter in-service dates back 180 days.
ERCOT · PUCT Project 55999 · 2025-05-01
A 100 MW officer-letter data center load is planned at 27.6 MW.
The rates come from ERCOT’s own history. It says data center loads with in-service dates from 2022 through 2024 “came into service with 49.8% of the requested amount”, and that of officer-letter loads due in 2024, only 55.4% were in service by February 2025. Large loads in that period also came in about 220 days late on average.
The discounts are historical realization rates applied to the whole forecast. They tell you what a planner assumes about the group. The record test tells you where a given site sits.
What happens when requests cost money
When a utility attaches a deposit and a deadline, the queue shrinks fast. AEP Ohio’s data center tariff, approved by the Ohio regulator, ran a two-step process with fixed windows.
Before the tariff, data centers had requested 30,000 MW. Given 45 days to request and pay for an engineering study, 13,022.7 MW did. Given 60 days to sign binding agreements with financial commitments, 5,642 MW signed as of February 12, 2026.
A signed contract is still short of a building. It is the first point in this funnel where the developer has put money behind the megawatts, and the drop from request to contract is the phantom share for this one utility.
What SB 6 asks for in Texas
SB 6 builds similar filters into Texas interconnection standards for sites above the 75 MW threshold. Each one creates a record that a speculative request is unlikely to have.
- Disclosure of any substantially similar request for service elsewhere in the state (§37.0561(d)).
- A flat study fee "of at least $100,000" for initial transmission screening studies (§37.0561(f)).
- Proof of site control "through an ownership interest, lease, or another legal interest acceptable to the commission" (§37.0561(g)).
- Uniform financial commitment, which may include "security provided on a dollar per megawatt basis" or a contribution in aid of construction (§37.0561(h)).
The statute directs the PUCT to write the standards. How each requirement is applied, and when, depends on that rulemaking, so check the current rule before you rely on any one filter.
The record test
A project passes a stage when a document from someone other than the developer says it has. Five questions cover most of the pipeline.
| Question | Record that answers it | Where it usually lives |
|---|---|---|
| Is the site fixed? | Deed, lease or plat naming the parcel | County records; city plat approvals |
| Has it asked a public body? | Zoning, plat or permit application | City and county agendas; state permit portals |
| Has money been committed? | Study fee, security or signed agreement | Utility tariff filings; PUCT dockets |
| Is anything being built? | Issued permit plus dated inspection or imagery | Building and air permit records |
| Is it drawing power? | Approval to energize or in-service confirmation | Grid operator and utility filings |
A project with only an announcement and a load request fails every row. It may still be real, but the record gives you nothing to model yet. Announced isn’t built gives the full stage definitions, and data center megawatts covers which MW figure each record reports.
How to apply it to a pipeline
Count megawatts by the last row each project passes, and never sum across rows. Adding a request total to an energized total double counts every project that has moved forward.
- Give each project the stage of its latest independent record, with the document and date.
- Flag projects whose only records are an announcement and a load request.
- Check for the same developer, MW and county in more than one utility's queue.
- Re-date a stage when a newer record replaces an older one.
- Keep the MW basis with each figure. A request, an approval and an observed peak measure different things.
The flagged group is where phantom capacity sits. Report its MW separately from the rest. Its timing is covered in how long it takes to build a data center.
Sources
Accessed 2026-11-07.
- Texas Legislature, SB 6 (89th Legislature, enrolled), PURA §37.0561. https://capitol.texas.gov/tlodocs/89R/billtext/html/SB00006F.htm
- ERCOT, Interconnection and Grid Analysis Update, Board of Directors, June 1-2, 2026. https://www.ercot.com/files/docs/2026/05/24/8-Interconnection-and-Grid-Analysis-Update.pdf
- ERCOT, Monthly Operational Overview, September 2026. https://www.ercot.com/files/docs/2026/10/16/ERCOT-Monthly-Operational-Overview-September.pdf
- ERCOT, filing in PUCT Project 55999, Update on ERCOT’s Adjusted Load Forecast (May 1, 2025). https://interchange.puc.texas.gov/Documents/55999_121_1495046.PDF
- AEP Ohio, news release on data center tariff contracts. https://www.aepohio.com/company/news/view?releaseID=10753
Spotted an error? Corrections are logged on the article.
Send a correction