U.S. electricity demand is growing again, driven by large loads from things like cloud data centers and cryptocurrency miners. With the largest facilities over 1 gigawatt of load, states and utilities court these large loads to attract tax revenues and well-paying jobs.
Every month, the Electric Reliability Council of Texas publishes its so-called Operational Overview, a report on power system supply, demand, and system operating conditions. Last week, we highlighted it (first on the list) as one of the interesting findings in US energy. It’s so fascinating, in fact, that I wanted to dig into it further, and specifically into the state’s queue of large electricity loads seeking interconnection in ERCOT.
The first, and most glaringly obvious, thing about the queue of large electrical loads in Texas is that it’s enormous. Currently, Texas has more than 172 gigawatts of large loads, more than double the state’s most recent reported peak demand.
However, a large number like this encompasses a multitude of possibilities while also obscuring important details. One of the key aspects to consider is which section of that queue can be considered high-likelihood. ERCOT categorizes projects in five statuses, ranked by their level of certainty. These range from “observed energized” (meaning that they are operationally part of the electricity network) to “no studies submitted” (meaning the project has only offered a statement of intent to the grid operator). Or to put it more succinctly, ¯\_(ツ)_/¯.
ERCOT’s first three categories combined amount is only 20 gigawatts by 2030. Although, the word “only” may be misleading here, as even just this small sliver represents nearly a quarter of the state’s peak demand from last month.
Obviously, this leaves another 150 gigawatts of large loads planned in Texas between now and the end of the decade. There’s a near-perfect 40/60 split in those planned projects in 2030: 61 gigawatts are under ERCOT review, and a further 90 are in the “no studies submitted” category.
To put it another way: more than half of the 2030 large load interconnection requests in Texas right now are far from certain, and are so far from certain that the state cannot study them thoroughly yet.
This particular distribution — a small volume of near-dated certain assets and a much larger cohort of longer-dated uncertain assets — paints a picture of a highly uncertain future.
We cannot completely predict the future, but we can measure its inputs with as much fidelity and as high a frequency as possible. In this sense, we should be grateful to ERCOT for publishing its queue every month.
But if we cannot know the future, we can at least consider its possibilities. To me, the combination of these charts and their underlying data urges us to think about possible distributions of large load interconnections in the 2030 timeframe. We can think in tails: fat tails, and long tails. Here’s a chart showing both of these distributions along with a Gaussian, or normal, distribution.
A normal distribution means that most outcomes cluster near the middle. To apply this concept to the Texas large load interconnection, we can consider a normal distribution providing a relatively clear understanding of the likelihood of any given event occurring. Anything outside of the middle, or what we broadly expect, is a fluke. A fat tail distribution is still fairly predictable, though with a greater chance of a big jump outside of the normal distribution. And a long tail distribution, finally, is one in which wild outliers are more likely.
What could a fat tail distribution of large electricity loads look like? Imagine a sudden surge in demand driven by a global need for an additional 200 gigawatts of computing power for training AI models. If a quarter of that demand were to converge on Texas, it’s possible that some projects, which had not previously submitted studies, might expedite their processes, resulting in more developments occurring than we anticipated.
And a long tail distribution? That would be something like another 200 gigawatts of data centers just in Texas, completely blowing up this chart in just a few years. But it could also mean the total opposite: training becoming massively more efficient, rendering many of the speculative, early-stage interconnection requests moot.
The point of these fat and long tail distributions is not that we know what will happen, but that we can think of what might happen. And while we’re thinking about the long future, we can measure the present and near future more accurately, with higher frequency and higher fidelity.
Subscribe for more content like this; reach out with questions: sayhi@halcyon.io; follow us on LinkedIn and Twitter