American utilities are sitting on requests for more than a trillion watts of data center power. Most of it is not real, and everyone involved already knows it.
An interconnection queue is just the waiting line a utility keeps for anything that wants to plug into the grid, a solar farm, a factory, a data center. Getting in line costs almost nothing. Nobody has to prove they have financing, permits, or a customer first.
That low bar is exactly why the topline numbers are so distorted. A developer racing to lock down a ten billion dollar campus before a competitor grabs the same substation capacity has every incentive to file everywhere at once and sort out the details later.
Interconnection queue. Entry fee: paperwork and a modest deposit.
To hedge against multi-year wait times, developers routinely file the same 500 to 1,000 megawatt project with three or four different utilities at once, planning to build wherever the transformer comes through first and simply abandon the rest.
Some of the identities behind these filings are deliberately hard to trace, with land purchases often made through anonymous LLCs. Even the largest, best-capitalized tech companies contribute to the noise. Latitude Media has reported that Microsoft and Meta have each walked away from gigawatts of previously requested capacity as permitting dragged on or strategies shifted, meaning even a real company's request isn't a guarantee.
Same project, four filings. Only one substation ever gets built.
Texas shows the scale best. The state's grid operator went from tracking 63 gigawatts of large load requests at the end of 2024 to well over 400 gigawatts by the middle of 2026, more than five times the grid's own record peak demand.
In Houston specifically, CenterPoint Energy watched its data center interconnection requests jump from 1 gigawatt to 25 gigawatts in a single year. The mismatch got large enough that Governor Greg Abbott ordered a formal audit of data center projects in August, citing roughly 13 billion dollars in industry revenue riding on getting the real number right. Exelon, meanwhile, cut its own pipeline by nearly 40 percent, down to an 11 gigawatt queue it actually considers credible, with one industry voice summarizing the problem bluntly: grid operators don't know which ones are real.
Exelon has already cut its own pipeline by nearly 40%, to an 11 GW queue it calls credible.
Some analysts push back on the word phantom itself. Unlike past interconnection gluts driven by thinly capitalized solar developers chasing tax credits, the serious data center requests come from Microsoft, Amazon, Google, Meta, and Oracle, companies with the balance sheets to actually build and a strategic reason to keep building regardless of subsidies.
That distinction matters for how the story ends. A tax credit lottery ticket and a hyperscaler's genuine, if hedged, land grab are not the same kind of speculation, even if both flood the same queue. VaaSBlock's analysis argues the disagreement isn't really about whether the topline numbers are inflated, almost everyone agrees they are, it's about how much of the gap represents wasted paperwork versus real demand still finding its final address.
The "phantom load" framing: most requests are inflated, filed for hedging, and will never be built.
The counterargument: Microsoft, Amazon, Google, Meta, and Oracle have real balance sheets and strategic reasons to keep building, subsidies or not.
This isn't a victimless numbers problem. Every phantom request a utility has to study and plan around is staff time, grid modeling, and construction decisions spent on a project that may never exist, while genuinely ready projects sit queued up behind imaginary ones.
One utility told the interconnection software company GridUnity that nearly 30 percent of all its 2024 applications ended up canceled. A public database tracking queue activity since 1996 shows 162 of 521 recorded requests eventually withdrawn. Multiply that pattern across every utility in the country simultaneously fielding dozens of large load proposals a year, up from the one or two utilities used to expect annually, and the planning problem compounds fast.
The fix utilities are converging on is simple to describe and slow to implement: stop treating every filing as equal, and start requiring proof before a request counts as real.
Belfer Center research details the screening process Dominion Energy now applies to separate serious applicants from speculative ones. In Texas, ERCOT is moving toward rules that require financial commitment before a large load request gets treated as firm rather than aspirational. None of this resolves the underlying uncertainty quickly. Grid planning happens years in advance, and until the screening catches up with the filing habits it's trying to correct, the gap between a trillion watts requested and a few hundred actually built will keep shaping decisions it probably shouldn't.
A tighter gate than the one this story opened with.