Texas Data Center Audit: Who Pays for the Phantom Load?

Texas Gov. Greg Abbott's August 3 moratorium on data center approvals ^[1][2][3] is a rare moment of regulatory friction in the AI rush. But the freeze itself, while politically smart, masks the deeper mechanism at work: the phantom-load problem baked into how utilities count demand.

Here's what Abbott ordered: PUCT and ERCOT must audit the 1,800 projects in the interconnection queue before any more connections are approved, demanding proof of tax breaks, power use, water cooling, and local impact plans. ^[6][7][8] The headline figure is staggering: 474 gigawatts of requests, roughly five times ERCOT's record peak demand. ^[6][8] Data centers make up about 90 percent of it. ^[1][6][8] But the moratorium's framing, grid overload, reliability risk, tells only half the story. The other half is the one nobody in the approval queue wants audited: how many of those 474 gigawatts are actually contracted with binding offtake agreements, and how many are speculative press releases?

This is the inflation mechanism. Developers file interconnection requests across multiple utilities simultaneously, shopping the same project or pipeline to ERCOT and sometimes to other regions, knowing that visibility, a high number in the queue, is political and financial currency. A handful of genuine, signed commitments get diluted in a crowd of announced, exploratory, or redundantly filed inquiries. When a utility invokes the pile of requests to justify new capacity investments (transmission, distribution, generation), those costs get handed to all ratepayers, not just the data centers. If half the queue never materializes, the other half of Texans picks up the tab for stranded infrastructure. That's the socialized cost of privatized optionality.

Abbott's audit is a needed check, but it stops short of forcing the underlying tariff change. The audit will surface information, tax breaks, water use, ownership, that are properly public. What it probably won't do is break open the interconnection queue to force developers to segregate contracted versus announced load, or to require that new capacity investments be assigned to a dedicated, cost-isolated customer class rather than spread across residential and small commercial. ^[3] And it doesn't require ERCOT or PUCT to mandate bring-your-own-generation or flexible-load options that could absorb genuine AI-driven load without building new firm capacity at all.

The window is open now. While ERCOT and PUCT conduct the audit, Texas lawmakers and regulators should condition any reconnection on three things: (1) segregation of contracted, committed, and speculative load in the public record; (2) assignment of all new transmission and generation costs to the data-center customer class, not to residential ratepayers; and (3) a requirement that any project approved must either bring its own generation or accept a flexible interconnection service with defined curtailment availability (similar to Duke Energy's model ^{[background library]}). Without those tariff changes, the audit becomes a theater of accountability while the cost-shifting mechanism stays intact.