Greg Abbott's August 3 letter did not stop a single data center from being built in Texas. What it did was reprice them. It sorted the market into two piles — megawatts that already hold an approved interconnection, and megawatts waiting on a verification and audit that ERCOT now tells the Commission "will take several months" — and then it let everyone work out which pile was worth more.

That took about a week. On Monday, Riot Platforms leased 191 megawatts at Rockdale for $9.1 billion, and the load-bearing phrase in the announcement was five words long. On Wednesday, IREN handed Microsoft a finished 50 MW at Childress. In between, ERCOT asked the Public Utility Commission for more time on three separate deadlines and, in the same filing, disclosed 8,766 megawatts of Texas data center and mining load that was never in Batch Zero to begin with. Nobody was counting it. It is still coming.

"Already Fully Approved and Energized"

Riot's August 10 filing describes a 20-year, 191 MW critical IT build-to-suit at Rockdale, Tier 3, running through June 2048. Total initial contract revenue is approximately $9.1 billion, or $16.1 billion if the tenant exercises both five-year extensions. Ninety-six IT megawatts land in December 2027; the full 191 by June 2028. Morgan Stanley is fronting a $573 million interim facility until an investment-grade credit backstop is finalized.

The tenant is identified in the filing only as "one of the world's leading frontier AI labs." Bloomberg reported it as Anthropic. Riot has not confirmed that, and I am not going to pretend it has.

What Riot did say, in CEO Jason Les's own quote, is the whole thesis of the week: the company's platform stands apart through "multi-gigawatt-scale power capacity that is already fully approved and energized, in-house data center development expertise, and the ability to engineer custom infrastructure." Two of those three are things a competitor can hire. The first one is now, functionally, unobtainable in Texas until the audit clears.

Run the arithmetic yourself, because it is worth seeing: $9.1 billion over 20 years is about $455 million a year, against 191 megawatts of critical IT. That is roughly $2.38 per critical IT watt per year, on a site where the interconnection was already done. Morgan Stanley took its price target to $43 from $36 the same week. It did not need a new thesis to do it — it needed a calendar.

The rest of the quarter was ugly and largely beside the point. Revenue was $174.2 million against $153.0 million a year ago, but Riot booked a net loss of $237.2 million and adjusted EBITDA of negative $69.7 million. Bitcoin mining revenue fell to $113.7 million from $140.9 million. Data center revenue went from nothing to $23.2 million. That is the transition, showing up in the line items exactly as advertised. Also disclosed: a non-binding LOI covering the full ~1 GW Corsicana campus, single tenant, unnamed. And the AMD lease from January now runs to 50 MW when Phase 4 lands in May 2027.

Two days later, IREN delivered Horizon 1 at Childress — 50 MW of critical IT, direct-to-chip liquid cooled, accepted by Microsoft, the first of four such deployments due in 2026 under a five-year, $9.7 billion contract signed last November. Same pattern. Existing Texas campus, existing power, finished product, paid for. The audit does not touch it.

ERCOT Is Aware of 258 Medium-Sized Loads

I wrote two weeks ago that ERCOT had missed its own August 7 classification deadline. Here is what actually happened next, because it is more interesting than the miss.

On August 10, ERCOT filed in PUCT Project No. 59142 requesting good cause exceptions to three Planning Guide obligations at once: the missed August 7 classification, the August 1 quarterly stability assessment, and the November 1 quarterly stability assessment. It received more than 200 dynamic data submissions and could not review them in time. It has temporarily paused Batch Zero, and it will refrain from granting energization approvals to large load data centers and crypto facilities until verification completes. Note the verbs: ERCOT asked. The Commission has not yet granted anything. That happens August 20.

The new number is on page seven. ERCOT is aware of 258 medium-sized loads — peak demand of at least 25 MW but less than 75 MW — seeking interconnection by 2032, representing 13,473 MW. Of those, approximately 157 are data centers or virtual currency mining facilities, representing 8,766 MW. Every one of them sits below the 75 MW Batch Zero threshold. None of them were in Batch Zero. ERCOT surfaced them from an earlier RFI tied to the 2026 Regional Transmission Plan, and it now intends to collect community impact information from all of them.

Which is the tell. The 474 gigawatts Abbott cited in his directive letter, roughly 90 percent of it data centers, was always the visible queue. Underneath it there is a second population sized deliberately to duck the threshold, and the state has just decided to look at that too. Eight-point-eight gigawatts is not a rounding error. It is more than half of what CenterPoint put into Batch Zero.

The collateral damage is stacked up behind it. ERCOT cannot begin the system-wide Batch Zero study in September as planned. It recommends delaying the Long-Term Load Forecast until final classification, which delays the 2026 Triennial Reliability Assessment results expected in November, which could delay the December Capacity, Demand and Reserves report. ERCOT is not asking the Commission to rule on any of that — it raised it "for the Commission's awareness," which is regulator for you should know this is on fire. And it has pointedly not asked to move the April 9, 2027 deadline for study results. The start slips; the finish does not. Somebody eats that compression, and it is not going to be ERCOT.

"If It Can't Get Met, It's Not Coming Online"

Into that, a genuinely useful counter-forecast. Ascend Analytics gave Utility Dive a market report on August 12 putting ERCOT's 2030 peak at about 120 GW. For context, ERCOT said in April that transmission service providers had reported 208 GW of 2030 load based on contracts and officer letters, and ERCOT's own adjusted forecast came in at 138 GW.

Ascend gets to 120 by assuming a 55.4% success rate for proposed loads, and its core finding is that more than 80% of new large loads seeking interconnection will not have matching generation online by 2030. "Even though that appetite is enormous, if it can't get met, it's not coming online," said Brent Nelson, Ascend's senior managing director of market intelligence. His colleague Robert LaFaso, director of market intelligence, put the constraint plainly: "The ability of the grid to add new generation is much, much smaller than the demand of queued large-load facilities."

This is the argument that has been missing from the queue fight. Everyone has been litigating whether the demand is real. Ascend's point is that the demand's reality is not the binding constraint — tier-one gas turbine slots, EPC crews, high-voltage equipment and permitting are. You can cull the queue or not cull the queue and you still cannot buy a turbine.

Vistra CEO Jim Burke, on the August 7 earnings call, would like it culled regardless: "I'd like to see the queues culled, at the end of the day." Vistra posted $1.767 billion of adjusted EBITDA, up about 30% year over year, and trimmed its ERCOT annual load growth outlook to 4%–6% from 5%–6%. Its 2.4 GW Comanche Peak starts delivering 1.2 GW to Amazon in 2027 under a 20-year PPA, which is to say Vistra is calling for a thinner queue from a position inside the part of it that already cleared. Everyone in Texas is now arguing their own book, and at least Burke does it out loud.

One number to keep straight while all this gets argued: ERCOT's all-time peak record stands at 91,089 MW, set July 22, 2026. That is the integrated hourly figure, not an instantaneous one, and it is still unofficial pending final settlement. The record it beat, 85,508 MW, took three years to fall.

"I Think It's a Mistake"

The most notable political event of the week was a Republican president criticizing a Republican governor's grid policy. In a Punchbowl News interview released August 7, Donald Trump said: "I saw Texas the other day sort of is against data centers." Then: "I think it's a mistake. And I'm not taking positions, I just think it's a mistake, because there are other communities that want it." And: "For Texas to say no to data centers is a mistake in the sense that it could be bigger than oil."

Abbott's office did not blink. Spokesperson Andrew Mahaleris cited ERCOT tracking a more than 500% increase in peak demand and closed with "Simply put, Texans must come first." He also produced the number that makes the audit hard to argue with: less than 10% of data centers responded to the state's requests to report their power and water usage. A separate survey tells the same story — the Texas Water Development Board has asked for monthly water data since 2020 from facilities using at least 10 million gallons a year, and fewer than one-fifth of 341 data centers reported in 2025, despite reporting being required by law.

Not one operator sided with Trump. The compliance announcements instead came in a queue of their own: QTS on August 6, Skybox and Digital Realty and MARA on August 7, Meta and OpenAI on August 10, Oracle and EdgeConneX and Stream on August 11, Google and CleanSpark and Core Scientific and Vantage on August 12, Microsoft and Hut 8 and Equinix on August 13, Amazon and Lancium and Cipher Digital today. Compass Datacenters founder and CEO Chris Crosby, who got there first with a Fortune op-ed, wanted exactly one thing from the state: "Just measure every developer by the same yardstick."

CenterPoint took the more interesting angle. On August 11, nine days before the Commission takes up the audit, the utility announced a Customer Savings Initiative projecting more than $5 billion in statewide customer savings over the next decade, driven by up to 14 new gigawatts of ERCOT-eligible base load and studied load. Chairman and CEO Jason Wells called it "a once-in-a-generation opportunity."

Read that as what it is: an affordability argument aimed at the exact political nerve Abbott is pressing. Large loads absorb fixed grid costs that residential customers currently carry. The 14 GW comes from CenterPoint's July 28 quarterly release, where it disclosed submitting over 17 GW into Batch Zero, of which roughly 14 GW is expected eligible — "more than a 65% increase from our current Houston Electric peak system demand of 21 gigawatts." The pitch is that the thing your neighbors are protesting is the thing that lowers your bill. Whether that lands in an election year is a separate question.

One Hundred Watts in Caldwell County

Last item, and the one nobody put in a queue. At 9:19 p.m. Eastern on August 5, in a 35-foot-deep concrete cavity southwest of Lockhart, Oklo's Groves isotope test reactor achieved first criticality — the fifth DOE-authorized advanced reactor to do so this summer, and the first on private land.

Maximum design power is 100 watts. It will never produce an isotope; Oklo's own communications lead confirmed as much. The DOE determination calls it a zero-power critical assembly running five Framatome GAIA low-enriched uranium dioxide assemblies, commercially procured, never to be refueled, on less than an acre inside a 47-acre parcel.

So why does it matter. Because of the clock. Oklo built the facility in 229 days and went from groundbreaking to a self-sustaining chain reaction in under a year, on private Texas dirt, with commercial fuel and no government material. "Reaching criticality in less than a year is an incredible milestone for our team," co-founder and CEO Jacob DeWitte said. In a week where the operative timeline for a Texas megawatt is "several months, and we'll get back to you on the study," somebody just demonstrated that the physics is not what takes ten years. The paperwork is.

What to Watch Next Week

Wednesday, August 19, 8:00 a.m. — Texas House State Affairs, JHR 140. Chairman Ken King has both charges on one agenda: data center development in Texas, explicitly including SB 6 implementation and the Large Load Batch Study Process, and the regulatory and planning processes for the proposed 765 kV transmission lines. One day before the Commission meets. That sequencing is not an accident.

Thursday, August 20, 9:30 a.m. — PUCT open meeting. ERCOT's good cause exception request is Item 22, Project 59142, with the large load interconnection standards rulemaking, Project 58481, sitting at Item 21. ERCOT has said it will detail the audit's scope and timing here. This is the meeting that decides whether the pause has an end date or a shrug.

August 31 — the dynamic-model cure deadline. Entities notified of deficiencies must resolve them by month-end or drop out of Batch Zero. ERCOT has asked the Commission to let anyone notified late cure within 24 days of notification instead. If that relief doesn't come through on the 20th, some queue positions die on a technicality nine days before September.

September — the study that cannot start, and the month that gets tight. ERCOT has said it cannot open the system-wide study in September. Separately, Ascend flags September as an emerging risk window: earlier sunsets cut solar output while evening wind runs below August levels. Watch evening net-load peaks and scarcity pricing.

Riot's Corsicana LOI and the credit backstop. A non-binding LOI on a full ~1 GW campus is not a lease, and a $573 million interim facility from Morgan Stanley is not permanent financing. Both convert or they don't, and the second one tells you what the first is really worth.

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.