Inside Texas’s AI Data Center Queue
How real is the 226 gigawatts of demand sitting on Texas' interconnection queue?
In November 2025, the Electric Reliability Council of Texas (ERCOT), which is the organization that manages most of the electrical network in Texas1, published a number that should make everyone in the energy industry sit up straight: 226 gigawatts of large load interconnection requests, up from 63 gigawatts just twelve months earlier.
To put that in perspective: Texas’s all-time peak electricity demand, set during a scorching August 2023 afternoon, was 85.5 gigawatts. The state’s total available generation capacity is around 103 gigawatts. Companies are now asking to connect loads totaling 2.6 times the state’s record peak demand, and 73% of those requests are from data centers.
These aren’t abstractions. Each gigawatt represents roughly the output of a nuclear power plant, or enough electricity to power 750,000 homes. Many individual requests exceed one gigawatt per site. This is on the scale of entire power plants, demanded by single facilities.
What’s driving this explosion? Two letters: A.I. This post makes extensive use of data soruced from ERCOT’s publicly available files. Paid subscribers can access a spreadsheet I put together which organizes all of this data.
The Queue Explosion
The growth trajectory is unlike anything grid operators have seen. In December 2022, ERCOT was tracking approximately 15 gigawatts of large load requests. These were mostly crypto mining operations and conventional industrial facilities. By December 2023, that figure had grown to 41 gigawatts. Then 63 gigawatts by late 2024.
Then the dam broke.
In 2025 alone, ERCOT received 225 new large load interconnection requests through mid-November, compared to just 152 requests across the entire 2022-2024 period combined. The second quarter of 2025 was particularly frenzied: 78 applications totaling more than 70,000 megawatts arrived in a single three-month span.
“We have outgrown the process that was established for reviewing these large loads,” Kristi Hobbs, ERCOT’s Vice President of System Planning and Weatherization, told the board of directors in December. “It was originally set up for, we were thinking about 40 to 50 loads back in the 2022 time frame.”
The average project size has ballooned as well. In the 2022-2024 period, the typical large load request was around 415 megawatts. In 2025, that average jumped to roughly 725 megawatts, and ERCOT notes that “many” requests now exceed a full gigawatt per site.
The Conversion Funnel
ERCOT tracks large loads through a multi-stage process, from initial request to actually drawing power from the grid. The October 2025 Monthly Operational Overview reveals a dramatic gap between aspiration and reality:
Read that again: of the 226 gigawatts sitting in the queue, only 1.8% is actually operational and drawing power from the grid. More than half of the requests haven’t even submitted enough information for ERCOT to begin reviewing them.
This conversion rate, from queue to reality, is the central tension in the AI data center boom. Are we witnessing the early stages of an infrastructure revolution, or the world’s most expensive game of musical chairs?
The Phantom Load Problem
Industry insiders have a name for speculative interconnection requests that may never materialize: phantom loads.
The phenomenon isn’t unique to Texas, but ERCOT’s independent market structure and relatively streamlined permitting make it a natural magnet for data center developers exploring multiple options simultaneously. A single hyperscaler might file requests in Texas, Virginia, Ohio, and Georgia, knowing full well it will only build in one location.
“There’s not enough stuff to serve that much load on the equipment side or the consumption side,” Beth Garza, former director of ERCOT’s independent market monitor, told CNBC in December. Joshua Rhodes, a research scientist at UT Austin, was more blunt: “There’s just no way we can physically put this much steel in the ground to match those numbers. I don’t even know if China could do it that fast.”
Texas lawmakers attempted to address the phantom load problem with Senate Bill 6, passed in 2025. Among other provisions, the law requires large load customers to pay a $100,000 screening fee before advancing in the queue and disclose whether they’ve filed duplicate applications elsewhere in the state. The theory: if you have to put real money down and reveal your other options, speculative requests will fade away.
Early results are inconclusive. The fee is pocket change for companies planning billion-dollar facilities, and disclosure requirements only apply within Texas. A developer filing in Virginia and Texas faces no obligation to reveal the duplicate.
What “Real” Looks Like: Stargate
While most of the queue exists only on paper, some projects are very real indeed.
The Stargate project, which is a joint venture between OpenAI, Oracle, and SoftBank, represents the most ambitious AI infrastructure buildout in American history. The first site, in Abilene, Texas, is designed to consume 1.2 gigawatts at full capacity. When operational, it will house up to 400,000 Nvidia GPUs across eight massive data center buildings, each the size of several football fields.
The Abilene campus is already partially operational. Crusoe, the AI infrastructure company building the physical facility, completed construction of all eight buildings in late 2025. Workers are now racing to install the computing equipment inside.
But Stargate’s ambitions extend far beyond Abilene. A second Texas site in Shackelford County, about 40 minutes away, will add another 1.4 gigawatts of capacity, powered entirely by an off-grid natural gas microgrid featuring 210 industrial generators. A third site in Milam County is in planning stages.
In total, OpenAI and its partners have announced plans for roughly 10 gigawatts of capacity across multiple U.S. sites, an investment they value at $500 billion. If completed, that single project would consume more than the current large load queue’s actually operational capacity by a factor of 2.5x.
The Supply Side Gap
While load requests have quadrupled, the supply side tells a more complicated story.
ERCOT’s generation interconnection queue, which the pipeline of power plants waiting to connect, totals 432 gigawatts as of October 2025. But the composition reveals the central challenge facing the Texas grid:
Battery storage and solar dominate the queue, comprising more than 77% of planned additions. This matters because AI data centers run 24/7/365, which requires firm, dispatchable power that solar panels can’t provide when the sun sets. Batteries help bridge the gap, but current technology limits most installations to 2-4 hours of storage. This is not nearly enough to power a data center through a windless night.
Natural gas, the only carbon-emitting but dispatchable option in the queue, accounts for just 48 gigawatts. This is up from 26 gigawatts in October 2024, suggesting developers are pivoting toward firm power. But even that expanded pipeline is dwarfed by the 165 gigawatts of data center requests alone.
Meanwhile, actual generation additions have been more modest. Between 2024 and November 2025, approximately 23 gigawatts of new generation capacity was synchronized to the ERCOT grid. This is impressive growth, but it is still less than 11% of the current large load queue.
The Kill Switch
Texas isn’t betting solely on market forces to manage the data center surge. Senate Bill 6 included a provision that data center critics call the “kill switch” and that grid operators view as essential insurance.
Under the law, large loads connecting to ERCOT after December 31, 2025 must install equipment allowing the grid operator to remotely curtail their power consumption during emergencies. Data centers must also maintain on-site backup generation capable of serving at least 50% of their demand, and must disclose these backup capabilities to ERCOT.
During extreme grid stress, like the conditions during Winter Storm Uri in 2021, when millions of Texans lost power, ERCOT can now, with 24 hours’ notice, direct data centers to either switch to their backup generators or reduce their load entirely.
The provision survived intense lobbying from the data center industry, which argued that their facilities often host critical infrastructure including 911 call centers and military operations. Lawmakers carved out exceptions for facilities designated as “critical loads,” but left the determination of what qualifies as critical to regulators.
For data center operators, the kill switch creates a new calculus: the cost of building sufficient backup generation versus the risk of being disconnected during the handful of hours per year when the Texas grid approaches its limits.
The Probabilistic Future
ERCOT publishes monthly outlooks projecting the probability of grid emergencies. The December 2025 report, covering January 2026, calculated a 1.13% chance of declaring an Energy Emergency Alert during the highest-risk morning hours, which is relatively low by historical standards.
But those projections assume today’s load levels. Add even a fraction of the data center queue, and the math changes dramatically.
Consider: if just 10% of the current queue (22.6 gigawatts) actually comes online by 2030, it would represent a 26% increase in ERCOT’s all-time peak demand. If 20% materializes (45 gigawatts), the grid would need to serve loads never before contemplated in Texas energy planning.
ERCOT’s own long-term forecasts, updated to include prospective large loads, project peak demand could reach 150 gigawatts by 2030. That would be a 75% increase from the current record. Transmission planning documents describe the need for “substantial new transmission infrastructure” to serve even a portion of anticipated growth.
What the Data Tells Us
After analyzing ERCOT’s public filings, the picture that emerges is one of extreme uncertainty bounded by a few hard constraints:
The floor is real: Approximately 7.5 gigawatts of large load capacity has completed all planning studies and received approval to energize. These projects have real developers, real financing, and real sites. Barring catastrophic changes, most of this capacity will come online.
The ceiling is imaginary: The 226 gigawatt queue is a collection of options, many held by developers who will build at most one or two of their proposed projects. Treating the queue as a demand forecast would be like treating homeowners’ Zillow searches as actual purchase commitments.
The middle is contested: Somewhere between 7.5 and 226 gigawatts lies the actual buildout trajectory. Industry analysts at Aurora Energy Research, commissioned by ERCOT to study resource adequacy, model scenarios with 35 gigawatts of data center load by 2030. This is a figure that would represent a 10x increase from today’s operational level.
The conversion rate will likely improve: As speculative requests wash out and serious developers advance through studies, the gap between queue and reality should narrow. The $100,000 screening fee, disclosure requirements, and simple passage of time will separate viable projects from placeholder applications.
But not by much: Even optimistic scenarios suggest conversion rates well under 50%. The physical constraints, including transmission capacity, generation availability, construction timelines, and equipment supply chains, impose hard limits that no amount of capital can fully overcome.
The Bottom Line
The 226 gigawatt queue is simultaneously one of the most important energy statistics published in 2025 and one of the most misleading. It captures genuine interest from the world’s largest technology companies in building unprecedented computing infrastructure. It also includes an unknown, but likely substantial, volume of placeholder requests that exist primarily to preserve optionality.
For investors, policymakers, and Texas residents watching their electricity bills, the relevant question isn’t whether 226 gigawatts will materialize. It won’t. The question is whether the 20-50 gigawatts that plausibly might arrive can be served reliably without destabilizing the grid or driving consumer prices to unacceptable levels.
That question remains open. The data centers are coming. The only uncertainty is how many, how fast, and who pays the price when supply and demand collide.
If you enjoy this newsletter, consider sharing it with a colleague.
I’m always happy to receive comments, questions, and pushback. If you want to connect with me directly, you can:
