Texas's grid operator is about to find out which AI data centers are real. On June 18, 2026, the Public Utility Commission of Texas approved a one-time, systemwide review, called "Batch Zero," that will force every project seeking 75 megawatts or more of grid capacity to show concrete proof of progress before its demand counts as committed (ERCOT release). It is the largest credibility test the state's grid has ever run, and the first designed specifically to discipline the AI data center buildout.
ERCOT, the Electric Reliability Council of Texas, manages the flow of electricity for about 27 million Texas customers and roughly 90% of the state's electric load (ERCOT large-load integration). For decades, the way a new industrial project got onto the grid was through individual negotiations with the local transmission utility: file a study, sign an agreement, take a queue position. That worked when large new loads arrived in a slow, predictable trickle. It does not work when 438 gigawatts of requests are sitting in the queue (Utility Dive).
To put 438 GW in scale: ERCOT's all-time record peak demand is about 85 GW. The queue, on paper, is more than five times larger than the system has ever been asked to serve. Most of those requests are not connected to anything yet. They are option-value plays: bets that a hyperscaler, a real-estate developer, or a special-purpose vehicle might someday sign a power purchase agreement worth billions. The grid cannot plan transmission, generation, or reserves against hypothetical demand. Something had to give.
Batch Zero is what gave. Under the new framework, ERCOT will first finalize the roughly 35 GW it has already classified as committed Base Load, meaning projects with executed interconnection agreements, purchased long-lead equipment, and active construction. Everything else, an additional ~65 GW of large-load requests, will go through a single, centralized evaluation. Projects have to show real progress to keep their place: completed interconnection studies, signed transmission agreements, ordered long-lead equipment such as transformers and breakers, site preparation, and physical construction (Foley & Lardner explainer). Those that cannot produce that evidence are downgraded from the active queue and lose priority.
"The load interconnection changes … represent a fundamental shift from prior processes," said Beth Garza, a senior fellow at the R Street Institute and former director of the ERCOT Independent Market Monitor (Data Center Knowledge). "There are simply too many requests to try and evaluate sequentially." Garza called Batch Zero a transition rather than a permanent rule, a one-time screen to clear the backlog, after which ERCOT will return to a more conventional process tuned to a hopefully more honest queue.
The framework's timing is notable because ERCOT's own staff have publicly hedged on whether the AI power boom is real at all. A separate analysis from the grid operator warned that the buildout "may not materialize" the way developers are advertising, citing uncertainty about hyperscaler siting decisions, water availability, and the eventual economics of large-scale AI training and inference workloads (Data Center Knowledge). Batch Zero can be read as ERCOT's institutional answer to that uncertainty. Stop trying to guess which announced projects will actually break ground, and force the market to prove it.
The mechanism matters because the same credibility problem is showing up inside ERCOT's own queue today: a roughly 438 GW request backlog against a system that has never been asked for more than 85 GW at peak, with most of the gap attributable to projects that have not yet demonstrated they are real. Batch Zero is, in effect, ERCOT admitting that the old process could not tell the difference, and building a new one that can.
For now, three things are worth watching. First, the deadline ERCOT sets for evidence submission, and how steep the documentation bar turns out to be. Second, whether any of the projects currently classified as committed Base Load lose that status as the screen runs. Third, whether downgraded projects re-file under the legacy utility-by-utility process or quietly exit the Texas market altogether. The 35 GW that survives Batch Zero as confirmed demand will be the most credible number in American AI infrastructure planning when this review closes.