Texas spent this week doing two things that look contradictory and aren't. On June 18, the Public Utility Commission approved ERCOT's "Batch Zero" process — the first real rulebook in the country for connecting the wall of data center load now stacked against the grid. Four days later, Microsoft and Chevron announced they would skip that rulebook entirely, building a 2.67-gigawatt gas plant in West Texas that never touches ERCOT at all.

That is the whole story of the AI power era in one week. The state is writing the rules for who gets to plug in, and the largest buyers have already decided not to wait in the line the rules govern. The grid is not the bottleneck, because for the biggest players the grid is no longer the power source.

"Built to Run Independently of the Texas Grid"

Chevron's subsidiary Energy Forge One, alongside Engine No. 1, will build a co-located natural gas plant in West Texas to feed a Microsoft AI campus directly, under a 20-year power purchase agreement announced June 22. The project, internally called Kilby, runs to roughly 2.67 GW of phased capacity and about $7 billion, with first power targeted for 2028 and a final investment decision expected by the end of this year. The turbines come from GE Vernova, with additional capacity from Caterpillar's Solar Turbines, all of it fed by Permian Basin gas sitting more or less underneath the site.

The same day, Microsoft confirmed the load this plant exists to serve: a 2 GW data center campus near Pecos, which it called one of the largest single capacity additions in the company's history. Microsoft is funding the energy infrastructure itself. So this is not two announcements — it is one deal with the wiring shown. Kilby is the power; Pecos is the load; ERCOT is not in the diagram.

For three years the standard move was to argue your way to the front of the interconnection queue. Microsoft and Chevron made the same structural bet everyone else is now making, only bigger and with a balance sheet most developers can't match: build your own generation next to your own demand, and let the grid be a backup you may never call. When a major drills its own gas plant to run a data center, "energy company" and "cloud company" stop being separate businesses.

"An Energy Transformation Unlike Anything We Have Seen Before"

That was ERCOT CEO Pablo Vegas on the day the PUCT approved Batch Zero, making ERCOT the first grid operator in the nation to study large loads in batches instead of one at a time. The process groups every project 75 MW and larger into a single study so ERCOT can size the transmission build against the whole picture at once. The picture is large: the queue now holds more than 438,000 MW of large-load requests, roughly 89 percent of it data centers. Applicants get their classification in August, with the final transmission plan due in fall 2027.

Read the fine print and Batch Zero quietly blesses the Kilby model. The framework carves out a path for customers who want to build their own onsite generation and self-supply, and another for loads willing to be curtailed when the local grid is tight. Texas isn't fighting the build-your-own-grid trend. It is writing it into the rules.

It got to, because nobody could stop it. The same June 18, FERC issued show-cause orders to six RTOs — CAISO, ISO-NE, MISO, NYISO, PJM, and SPP — finding their large-load rules inadequate and giving them 60 days to fix cost allocation and co-location. Notably absent from that list: ERCOT, which sits outside FERC jurisdiction. So while the rest of the country got a federal ultimatum to go design something, Texas had already voted its own version into effect. I flagged the federal cost-allocation fight a few weeks back as the thing Washington couldn't keep ignoring. The shoe dropped — everywhere except the one grid that answers to no one in Washington.

Washington Bets $17.5 Billion on Reactors That Don't Exist Yet

On June 23 the Department of Energy conditioned up to $17.5 billion in loans to accelerate ten large reactors — five loans, each funding a site with two 1.1-GW Westinghouse AP1000s. The money targets the long-lead parts, the reactor vessels and steam generators that take years to forge, with the stated goal of shaving up to three years off deployment. Westinghouse and its partners have to put up $1 billion in equity per project first, which is the part that separates the press release from the groundbreaking.

DOE didn't name the sites, but Westinghouse says it has signed letters of intent with seven potential partners, and the math of AI demand keeps pointing at Texas. The state has its own $350 million nuclear development fund, two operating plants with owners who have openly studied expansion, and the load growth to justify it. None of these reactors will produce an electron before the early 2030s. The deals announced this week — Kilby, the gas plants, the batteries — are what carries the load until then. Nuclear is the bet on the decade after the one we're living in.

The Bill for Going Off-Grid: Ask xAI

One cautionary note for everyone now racing to pour their own gas turbines. On June 15 the Department of Justice moved to dismiss the NAACP's suit against xAI over 27 allegedly unpermitted gas turbines at its Colossus 2 site outside Memphis, arguing in court that shutting them down "threatens American national, economic, and energy security." The plaintiffs want the turbines idled until permits and pollution controls are in place, plus penalties of up to $124,000 per day per violation.

The site is in Tennessee and Mississippi, not Texas, so it doesn't move the ERCOT map directly. But the principle travels. Every behind-the-meter gas deal — Kilby very much included — inherits the same air-permitting exposure the moment it fires a turbine. Chevron and Microsoft have the lawyers and the lead time to permit Kilby properly. The cautionary tale is for the dozens of smaller off-grid projects that will try to copy the model without the same patience, and discover that "skip ERCOT" does not mean "skip the regulators."

What to Watch Next Week

August Batch Zero notifications. ERCOT tells the first cohort of 75-MW-plus projects whether they made the initial study. This is the first official read on which of the 438 GW is real and which is a placeholder.

The 60-day FERC clock. Responses from the six RTOs are due in mid-August. Watch whether PJM and MISO ask for extensions, and read the cost-allocation language closely — that is where the fight over who pays for data center transmission actually gets decided.

Chevron's final investment decision on Kilby. FID is expected by year-end and is not yet locked. If the conditions slip, the flagship behind-the-meter deal wobbles, and a lot of copycat term sheets wobble with it.

The five nuclear sites. DOE still has to pick which of Westinghouse's seven LOIs convert. A Texas site in the final five would put real reactors behind the load-growth story.

The heat. ERCOT is forecasting a summer peak above 92 GW against an all-time record of 85.5. If late-June heat sets a new record, the forecast stops being a forecast.

Disclaimer: The Grid Report is Barrio Energy's market intelligence product. Nothing here is investment advice. Links go to primary sources wherever possible; form your own view.