This is the first paragraph of one of my posts from earlier in the year, “AI is Giving Me Gas: The Collision of Tech Hype and the Carbon Budget.” It may be bad form to start a Substack post citing another Substack post, but clearly these two posts are related. The sub-title of the above referenced post” “We are scraping the bottom of the 1.5°C carbon budget. Big Oil’s response? Build 252 gigawatts of new gas power to feed the AI boom.”

I also wrote about this back in September of last year, in a post titled “New Gas Generator Plants and the Plan to Flood the (Electricity Demand Growth) Zone.” In this post there’s a link to an AI analysis I did in a report called “The Future of U.S. Natural Gas Power Generation: Projections, Accuracy, and the Confluence of Limiting Factors to 2030.”

The electricity demand growth tied to AI and data centers has Big Oil salivating, with plans—dreams?—of 100-plus new gas-fired generation plants in place by 2030. Not that the supply chain and turbine manufacturing capacity can deliver, but the explosion in small diesel or natural gas generators nonetheless seems a happy enough ending, boding well for Big Oil sales.

This doesn’t bode well for the rest of us, unfortunately. In the news of late is Elon Musk’s xAI company, which has used a large fleet of mobile, trailer-mounted gas turbines (rather than diesel generators) to power its “Colossus” AI data center in Memphis, Tennessee. These turbines are deployed as a temporary, “quick and dirty” solution to bypass power grid constraints while constructing the facility, which houses Nvidia H100 GPUs for training the Grok AI model.

To get the data center operational in just 122 days, xAI used mobile turbines (approximately 35 to 62, depending on the report and timeline). Each turbine is capable of providing 2.5 MW of power, with reports indicating a total capacity exceeding 35 MW to over 100 MW. There’s been pushback from local residents and environmental group. It turns out that neither noise pollution or emissions of nitrogen oxides and formaldehyde are being welcomed, and in January 2026, the EPA ruled that xAI violated the law by operating dozens of these, at times, unpermitted, gas generators. I’m guessing any fines actually levied against xAI will be just part of the cost of doing business. The broader AI data center industry is facing a shortage of power, with many companies increasingly using on-site, fossil-fuel-based generators to bridge the gap.

By the way, if you want to dive deeper into the issue of power strategies and developments for AI data centers, check out Distilled, on Substack. Distilled is written by Michael Thomas and he’s undertaken as series of articles on the issue of AI data centers and approaches being pursued for power, including data centers building their own power plants.

Wet Dreams and Sloppy Seconds

The accuracy of future demand predictions itself is highly questionable, considering the wide-ranging numbers and, among other issues, the double, triple, or greater duplicate counting of generation sources among data center hyperscalers. These eager corporations reach out to more than one potential generation source to cover their bets. Accurate forecasting seems hindered by some combination of wishful thinking and double counting. Here’s an AI summary of this issue:

The “duplication issue” (often called “phantom load” or “speculative queuing”) refers to the practice where data center developers submit multiple applications for electrical grid interconnection for the same single project. 

Because securing power is now the primary bottleneck for AI and hyperscale facilities, developers “spam” the queue to hedge their bets. They might file requests for the same 500 MW project in three different states (or three different sites within the same utility territory) to see which one gets approved first. Once one is approved, the others are withdrawn, but in the meantime, they clog the study queue and artificially inflate demand forecasts. 

Grid operators warn that these “phantom” requests make it impossible to accurately plan for new power plants, as the requested demand on paper is often 5x to 10x higher than what will actually be built. 

Estimate Ranges of Duplication

Industry data suggests that the vast majority of current interconnection requests are duplicate or speculative. 

• • Overall “Phantom” Rate: Experts estimate that 80% to 90% of the data center capacity currently in US interconnection queues will never be built.
  • Realization Rate: Utilities often project that only 10% to 20% of the requested data center load in their pipelines will actually materialize.
  • Speculative Ratios (Firm vs. Requested):
    • AEP (American Electric Power):Reported 24 GW of firm commitments but has requests for 190 GW of additional load—a ratio of nearly 8:1 (speculative to firm).
    • Oncor (Texas Utility):Reported a queue of 186 GW of data center requests. For context, the utility’s entire current peak demand for all customers is only ~50 GW, suggesting the queue is inflated by nearly 400% of the grid’s total existing capacity.
    • PSE&G (New Jersey):Reported a 9.4 GW large load pipeline but expects only 10–20% of those inquiries to result in actual projects.
    • ERCOT (Texas Grid):Has received requests for over 220 GW of new load by 2030 (mostly data centers), which is more than double the state’s all-time peak demand record. 

Why This Is a Problem

1. 1. Planning Paralysis: Utilities cannot distinguish real projects from “zombie” projects. If they build transmission lines for all 190 GW (in AEP’s case), they would bankrupt ratepayers. If they wait to see which are real, they risk being too slow for the 24 GW that is real.
   2. Queue Backlogs: The “phantom” requests force grid engineers to perform complex impact studies for projects that don’t exist, delaying the connection of viable power plants and real factories by years.
   3. Artificial Scarcity: The illusion of zero capacity drives up power prices and panic-buying of land, further fueling the cycle of speculative multiple-filing. 

Why is this a Problem?

You might also ask, from the perspective of Big Oil, “Why is this an opportunity?”

Big Oil loves the high estimates of power demand and the expanded market for their products, and not just more volume, but more over the next several decades, just when we need to reduce carbon emissions, not raise them. Big Oil is in a frenzy to keep their business going for decades more, despite the counter need for this industry to decline.

There’s temptation, too, for the utilities who contract or build new generation capacity. While solar/wind/batteries can meet new energy needs (and be quicker and cheaper), power utilities remain drawn to action that follows business-as-usual thinking, and for many utilities, especially without governmental and regulatory guidance, that means more power plants.

The opportunity is now for Big Oil to future-proof the industry.

Does the Trump Administration strike you as leaning on governance and regulation to push for clean energy?

I don’t think so, but then maybe I’m as wrong as all those wild estimates about power requests to feed the AI industry.

The post Fossil Fuel Demand Growth Uber Alles first appeared on David Guenette.