|
Part I
The Mechanism
|
Friday the 13th seems like the right day for this one. Texas power.
The story you hear on every earnings call: AI needs electricity, Texas has land and gas, the data centers are coming. The proof offered is always the same number. ERCOT's large-load interconnection queue.
ERCOT's all-time peak demand is 91,089 MW, set on 22 July 2026. The queue is more than five times the entire grid at its hottest hour. That should be your first clue.
A queue entry isn't a load. It's a request. Some are real projects. Many are the same project filed in three counties, or a land developer holding a spot in line the way people used to hold concert tickets.
On 3 August, Governor Abbott ordered a full audit of every data center in the queue. ERCOT plans to file its verification report with the PUCT by 10 December. That's four weeks from now, and almost nobody is pricing what it might show.
Here's the mechanical point. The grid doesn't care how many requests are on file. It cares how many megawatts can be safely connected, studied and served at peak. That number is much smaller, and it moves much slower.
|
Part II
The Diagram
|
Here's what happens to the queue when you ask people to put money down.
Under Batch Zero, projects of 75 MW and up had to post $50,000 per megawatt in financial security and prove site control. Then ERCOT sorted the pile. In early September it conditionally included 204 projects (66.4 GW) as base load and 158 projects (127.9 GW) as studied load.
So the first real filter cut the headline by more than half. That's the speculative froth, measured. And the survivors are only conditionally in.
The audit is grinding through them now. By 5 November ERCOT had reviewed first-round answers from 257 developers, about 5,140 questions. A second round is going to roughly 80% of developers, with five business days to explain the gaps.
The problems it found are the boring kind that kill projects: affiliate relationships, site control, long-lead equipment, and not enough financial security posted. Paperwork, in other words.
Now the part that actually plugs in. As of August, 17 mostly data center projects had passed stability assessment, about 6.6 GW of peak demand, ramping over roughly five years. Energization approvals for big data centers have been paused since 3 August.
BNEF warns nearly 50 GW faces delay risk and that 2026–27 additions could fall to just 1.2 GW. Read the flow bottom to top and the boom looks a lot narrower than the headline.
|
Part III
The Weak Link
|
The weak link is the timeline. And ERCOT has already told us it's broken.
Batch Zero study results were due on 9 April 2027. At a PUCT meeting in August, ERCOT's Chad Seely said it plainly: "We will not have the study done by April 9, 2027." In September the board materials still listed the schedule impact as to be determined.
Meanwhile the financial layer is priced off the big number. Chips get ordered, buildings get financed and power deals get announced as if the connection is a formality. The connection is the constraint.
And there's a clause in SB 6 that most bulls haven't read. New large loads energized after 31 December 2025 can be curtailed during firm load-shed events, and new projects have to disclose on-site backup covering at least half their demand. In a Texas emergency, the newest data center is the first thing switched off.
I've seen this movie in other forms: pipeline capacity, port slots, refinery permits. The demand is real. The paperwork turns out to be the scarce asset, and the market finds out late.
The bulls aren't wrong that the demand exists. They're wrong about the speed. A plan to plug in a gigawatt doesn't become a gigawatt of load until the wires, the substation and the study all say yes.
|
Part IV
The Chain Reaction
|
10 December is the trigger, with the PUCT's 17 December open meeting right behind it. Here's the likely sequence when the audit lands.
First wave: the duplicates and placeholders get cleared out. The 474.7 GW figure shrinks, maybe a lot. Second wave: the lazy headline arrives, something like "Texas AI power demand collapses." That read will be wrong.
The real demand doesn't disappear when the fake requests do. It just gets stuck behind the same interconnection wall, competing for a few gigawatts a year. Third wave: everything that already has a connection, or can bring its own power, gets more valuable.
You can already see developers adjusting. This week one Texas AI compute campus announced it is designed to run fully off the ERCOT grid. When the wire is the bottleneck, people start building around it.
The machines reading headlines will sell the AI power story on that print. The people reading the interconnection documents will be buying what's already connected.
Where does capital go? Not into developers whose value is a spot in line. The edge sits with sites that are already energized or past stability review, and with the on-site generation SB 6 effectively requires, because a data center that can run on its own power doesn't wait for the grid.
Honest caveat: ERCOT hasn't set a new study date, and the audit could slip like everything else. But the direction is clear. The market is pricing the size of the queue. The machine runs on the rating of the wire.
Sources: ERCOT, POWER Magazine, PUCT, Stoel Rives, AIX Energy, Utility Dive, BNEF via Morgan Downey's Commodity News, GlobeNewswire