Texas has stopped issuing new permits for data centres while it works out which of them are real. At the end of 2024, the state’s grid operator, ERCOT, was weighing 63 GW of requests from big new power users. By June the figure was 474 GW, over five times the state’s all-time peak draw, and about 90 percent of it data centres.
Those figures come from an essay by Ryan McEntush of Andreessen Horowitz (a16z), published 7 October. A16z is a venture firm that backs energy hardware companies, and McEntush says as much: the firm invested in Heron Power, which makes solid-state transformers. He argues against a blanket pause. Read the policy recommendations below as his, not as findings.
The more useful part is his account of why the queue is so large. Developers routinely file the same project at several sites, and many filings are speculative, with no tenant signed. Some applicants, he writes, have never plugged in a GPU. Planners cannot tell which projects are genuine, and they do not want ratepayers funding upgrades for buildings that never arrive. So 474 GW is a count of applications. Any “data centres will need X gigawatts” forecast built on queue data inherits the same inflation.
Filing is cheap, which feeds the problem. Under new Public Utility Commission of Texas rules effective 8 October, an applicant pays a $100,000 study fee and a deposit of $50,000 for each MW requested. A project that loses its place after falling two years behind forfeits only 20 percent of that deposit, $10 million at minimum for a 1 GW campus. McEntush notes the deposit typically backs just that project’s own upgrades, not the costlier regional lines whose bills are shared by every customer.
What Texas has actually done, per the essay: it escalated from instructing data centres to fund their grid upgrades themselves to freezing new permits. ERCOT has also paused approvals to energise data centres of 75 MW and up, among them 17 that had cleared every other step. Its “Batch Zero” process, which studies large loads together, cannot proceed until an audit reports in December. Chevron, which plans its own gas plant next to a planned Microsoft campus, now says the freeze may delay its final investment decision until 2027, though it still expects first power in 2028.
Bringing your own power is the obvious workaround, and the essay is blunt that it rarely avoids the grid. SemiAnalysis tracks 75 GW of orders for on-site equipment, yet McEntush says every site he knows of plans to connect when it can, because generating your own electricity usually costs more than buying it from the grid. Texas law also limits how far a private network can go: a campus may supply itself, but serving multiple customers across its own wires is not allowed. A proposal from the Cato Institute would loosen that. Nothing in the essay says Texas is moving to adopt it. Utah’s SB 132 already permits fully off-grid systems for loads from 100 MW upward.
The politics are local as much as technical. As of June, just 28 of 377 companies had replied to a state questionnaire about resource consumption, according to the essay, and commissioners in Hood County were handed a tax waiver request for a project known only as “Project Patriot.” McEntush cites an August poll in which 56 percent of Texas voters expected local energy bills to suffer from more data centres.
His own prescriptions: carve out generators permitted for emergencies only, and any project whose opening phase is already financed, hold utilities accountable for delays as developers are, and let customers who agree to cut demand connect sooner. Batch Zero already offers a version of that last idea, and 11 projects have chosen to count their own power plant toward their size.
For anyone planning compute capacity in Texas, queue position is not a date. The next real marker is the December audit report, and a campus that can fund its first phase and shed load on request is better placed than one that cannot.
Based on “Why Texas Is Making Data Centers Wait” by Ryan McEntush, a16z (Andreessen Horowitz), published 7 October 2026.
