A developer with five candidate sites in Texas files five interconnection requests, because filing costs relatively little and a place in the line does not [5]. Four of those sites will never be poured. All five sit in the total that now gets quoted as demand [1].
The queue was built to sequence connections for grid operators who have to commit generation and transmission years ahead [18]. It is being read as a forecast of compute. What filers are actually doing is buying options: the cost of holding one is paperwork, and the cost of skipping one is watching a better-funded project connect behind somebody who filed two years earlier [6]. Texas's queue shows that shape clearly: a free place in line, held against sites that may never be built.
The readings worth anything come from utilities that started charging. Exelon added stricter collateral requirements and its pipeline of high-probability data centre demand fell 40%, to 11GW [8]. Run that backwards: 11 divided by 0.6 is about 18.3GW before the change, so roughly 7GW of "high probability" demand declined to post money [2]. Ohio brought in connection study fees of up to $100,000 and AEP Ohio's pipeline then fell by more than half [9]. In Pennsylvania, more than 100 proposed data centres have produced 20 permit applications, under a fifth [10][4].
The awkward part is that real load is arriving at the same time. Thomas Gleeson, chairman of the Texas Public Utility Commission, says that when you do not know what is real, you do not know how to build the infrastructure for it [7]. Jeff Shields of PJM says the load is showing up and generation is not at the pace needed [14]. Both men are describing the same list. Across the Midwest, Mid-Atlantic and South, requests exceed 700GW, more than ten times current estimates of US data centre electricity consumption [4], which measures filing behaviour rather than servers. Daniel Farris, an attorney at Foley & Lardner, puts part of it down to entities that rushed into the space learning how hard construction is [15].
The arithmetic on the other side is less speculative. PJM's capacity costs have risen $29.4bn across roughly four auctions [13], an average near $7.35bn per auction working its way into bills [3].
So the usable version of the queue has two columns: megawatts, and non-refundable dollars per megawatt the applicant has already spent. That is how a pipeline gets sorted, megawatts against dollars spent, whether it is a company's own pipeline or a counterparty's before a power contract gets signed against their queue position. A request with posted collateral and a named end user is a project. A request with a filing date and neither is an option someone else is paying to keep open. Pennsylvania's August 18 executive order now demands the end-user answer above 25MW [11], and Texas ordered an audit in early August covering ownership, tax incentives, water use and generation plans [12]. Both actions raise the cost of filing vaguely rather than banning speculative requests outright [17]. A queue discounted too aggressively is itself a reason not to build generation [19].
