22-minute listen/watch | 13-minute read
As electricity demand grows and becomes harder to predict, power grid planners face a difficult question: how do they build for a future shaped by extreme weather, renewable energy, and the rapid rise of data centers? A new research project is developing tools to help planners identify the most important risks without having to model every possible future. The work could help utilities make better long-term infrastructure decisions while reducing the chances of costly overbuilding or major power outages.
Transcript
Anirudh Subramanyam
A small data center consumes as much power as a small city does. And so this is kind of breaking, again, old assumptions. In the past, demand was never like this; we could very reliably predict how much electricity people were going to consume. And now, because data centers and AI are coming up, no one knows, in ten years, how many data centers we're going to have. No one has a clue.
Host
Welcome to Growing Impact, a podcast by the Institute of Energy and the Environment at Penn State. Each episode of Growing Impact explores the projects of Penn State researchers who are solving some of the world's most challenging energy and environmental issues. Each project has been funded by the Institute's seed grant program that grows new research ideas into impactful energy and environmental solutions. I'm your host, Kevin Sliman.
For decades, delivering electricity from power plants to homes and businesses followed a fairly predictable path. But today, that system is being tested by new demands and new uncertainties, from extreme weather and renewable energy to the explosive growth of data centers. Utilities and planners are now making billion-dollar infrastructure decisions years in advance because of the long lead times required to build energy infrastructure, all while trying to prepare for an unpredictable future. On today's episode of Growing Impact, we talk with two researchers developing a new tool designed to help energy planners make those decisions with greater accuracy and confidence.
Thank you both for coming on to Growing Impact and having this discussion about your research. My first question to you, before we even jump into the project, is: can you introduce yourselves?
Anirudh Subramanyam
I'm Anirudh Subramanyam. I'm an assistant professor of industrial engineering and operations research at Penn State. My research builds tools for people that help plan our energy infrastructure and helps them decide how to plan for uncertainty. So basically, helping them decide what they should consider in the future and how they should plan for it.
Mort Webster
Hi. Mort Webster. I'm a professor of energy engineering in the Department of Energy and Mineral Engineering, which is in the College of Earth and Mineral Sciences at Penn State. Very similar to Anirudh, my background is working, in general, on methods of making decisions under uncertainty, but applied mainly to energy systems, energy systems design, market design, and really helping with long-term planning under uncertainty — different methods for doing that.
Host
Let's jump into the first question. And this is just looking for an overview of the project. Can we discuss the overview and the items that it's looking to accomplish?
Anirudh Subramanyam
Transmission planning is basically about where, when, and what type of generation and/or transmission the grid should build over a long horizon. You can think of this as spanning 10 or 20 years. And this is fundamentally a question that has a lot of trade-offs. So a typical trade-off is you have a wind plant that is located very far away, and you have to build a long transmission line to connect this wind plant, located very far away, to bring it closer to where people need it, versus building a generation source that is much closer to where people live.
Right. So do you build a long transmission line that connects people far away, or do you build generation closer to where people are? And as you can imagine, this is a billion-dollar investment. It's a decades-long commitment. So we have to make sure we get it right. And what makes it even more difficult is that, because it spans such a long horizon, the future is uncertain. And we don't know what, 10 or 20 years from now, the future is going to look like .
What makes this project, I would say, timely and relevant is that federal regulators like FERC — FERC stands for the Federal Energy Regulatory Commission — have required utilities to explicitly plan against the future grid. They have explicitly required utilities and planners to plan against many possible futures of what can happen in the future.
But as we can imagine, the number of futures is essentially infinite, and they cannot possibly consider an infinite number of futures. So in the past, the tools that these planners have been using tend to oversimplify what happens in the future. Our method is trying to help them with that. And so the way our method does this, at its very core, is that it does two things.
It tries to help these planners find the scenarios that matter the most. It gives them a short, ranked list of: these are the 10 scenarios that you should consider for the future and that could potentially break the grid in the future. And the second thing it does is it does this by modeling the future grid more realistically.
So it essentially captures the physics of how electricity flows on the grid and how operators actually manage the system day to day, so that the scenarios that we evaluate in the future are evaluated as to how operators will actually face them in the future.
Host
What does grid resilience actually mean in everyday terms, and why should people care?
Mort Webster
First of all, people mean two different things when they talk about resilience, depending on who's talking. So it can get a little confusing. Grid resilience in one context means we don't want some shock to the system, or when the system is stressed — meaning very high demand — we don't want that to cause a failure that then cascades through and a huge part of the grid has to go down, and people go without energy.
So, avoiding those big events. The other thing that some people focus on is if something bad happens and it goes down, how do we get it back up quickly? The project we're talking about today, we're focused on the former, which is how to prevent those events. But people do mean two different things about resilience.
In terms of why people should care, this is all about avoiding large-scale power outages — people not being able to get energy, particularly in times of very high demand. You think about last summer. So June of 2025, we had a week of 100-degree-plus temperatures. The grid was truly stressed to its limit. Every available generator was on. If one or two components on the grid had gone down, we might have lost everything in the Northeast for one to two days in the middle of the heat, which means no air conditioning.
So we've managed, for many decades, to have an extremely resilient grid. We have very few widespread power outages. Most people who experience a power outage — 99% — it's because the local distribution lines, those last lines that go down the street to your house, which are the responsibility of the local utility. When those go down from a storm, that accounts for about 99% of when one of us has lost power. And then, you know, that takes a day to get back up.
We're talking about events that take down the eastern two-thirds of the U.S., and that's what we're trying to prevent.
Host
To your point, right, the grid has been resilient and has been able to hold. In my lifetime, I can't think of any times that we've lost power over huge swaths of the nation.
Mort Webster
In the 1960s, there were a number of these very large-scale events where the eastern two-thirds of the U.S. lost power. New York City was very famous — these big power outages in the mid-60s. And after several of those, which had very negative impacts on people, the electric utility industry got together. They didn't want government regulation. So they said, we will regulate ourselves to preempt Congress doing something.
They formed something called the North American Electricity Reliability Council. It's all of the utilities in the U.S. who get together, and they agree on reliability standards that everyone has to meet because everyone shares an interconnected grid. A failure in one place will take down all of your neighbors. So they agreed on these protocols, and that has been the way they have managed since the mid-1960s up to today to make sure that we have very few, if any, of those events.
The last thing I'll say is that NERC, just recently, a week or two ago, issued a warning that they see a real danger of events coming if there isn't more investment because of the growing demand from data centers. It needs to be more investment in generation and in transmission. So they're kind of putting up the flag saying we need to do something.
Host
So your project focuses on rare but high-impact events. Could you give an example of what one of these events might look like and how it could affect communities?
Anirudh Subramanyam
Two examples come to mind in the most recent past. The first is February 2021. During COVID, the Texas freeze. Millions of people in Texas lost power. It was an unusually cold day, which basically drove heating demand up, so all the natural gas was going toward heating.
What made it worse was that the natural gas wells where you actually pump gas from froze. Wind turbines froze. So a lot of power in Texas comes from wind, and they had frozen. And to make matters even worse, the power plants were not appropriately winterized, so to speak. So it was a combination of all of these events that essentially led to a complete blackout in Texas.
The other example, which is much more recent, is not in the U.S., but it's in Spain. So this was last year. And what makes this event very different is that this happened on a normal, sunny afternoon. So in Texas, you can attribute it to the very cold day. But in Spain, it happened on just a day like today, nothing unusual, where people in Spain and Portugal lost power for about 10 hours.
And there, what happened was it was essentially a combination of increasing voltage in the grid, which caused a lot of power plants to trip offline. And so essentially, you can think of this like a cascade. As voltage increases, power plants slowly start tripping offline one by one. It's like a series of dominoes falling. And all of this happened in a very short amount of time. This happened in, like, within a second. So you don't even notice. You blink, and it's down.
In both of these events, what is a common pattern is that it was a cascade of unlikely events that stack up, which ultimately caused the grid — the system — to tip over. These are the kinds of events that we are trying to capture in our project and to alert planners that there are these scenarios that can happen. It can happen because of an extreme weather event, or it can happen not because of an extreme weather event. Both of them can happen.
And essentially, we need to give planners a tool where they can actually help identify this for their own grid so that they can make better planning decisions.
Host
Right now, planners can't look at every possible scenario. How does your new tool help identify the most important risks without getting overwhelmed by data?
Mort Webster
What we're trying in this project to help the grid planners with is the way they have traditionally done, for example, transmission planning — and it has worked for the last 50 years — is they take from recent years and say, what was the highest demand hour of the past year? And then, if in 10 years we have some growth in electricity, that hour will also be higher. Let's just plan for when everything's on at full. And traditionally, that has been helpful to make sure there was enough capacity.
What's new because of all the new technologies — the wind, the solar, more reliance on gas, which is shared with home heating, and increasing frequency of extreme weather events — all of those are combining such that, as Anirudh said a minute ago, the problems may not be that highest-demand hour. The problems may be something else: when combinations of problems in several different places create a situation where you can't get power, even if there are enough power plants that could turn on somewhere else. You can't transmit the power over to where it's needed, and it is very difficult to figure out all the possible combinations.
It's not just, is demand higher or lower? It's all these different things in different parts of your region at the same time. So what we're developing in this tool is optimization to essentially look at all the different combinations. What could stress the system? We simulate the system and try to identify what are the more likely scenarios where you might have these big events, where you can't get the electricity from supply in one part to demand somewhere else because of some combination.
The other thing that we're trying to do in this project is, it turns out that for most of these systems, you can find lots of examples. If you just keep randomly trying different situations, you may find a lot of examples of a moment in time where you have a shortage of supply or you can't get it to demand.
But what you also see is that if you then organize all of those events that you've found, they tend to fall into a few small groups that have a very similar pattern. So one of the ways that we're trying to help industry planners, because they don't necessarily have the tools to do this today, is to not only look at all the combinations, but to figure out, really, when there is a problem, it's always when these three things go wrong, or it's when those four things go wrong. One is at one time of year, one's another time of year.
And then all they actually need is one example of that pattern. If they can add enough transmission, for example, so that you actually could meet demand in that situation, you're addressing all the other ones that have a broadly similar pattern of flows.
Anirudh Subramanyam
Imagine an insurance company, and instead of buying insurance for every possible accident that can go wrong on the highway, you're essentially going to price your product in such a way that you only plan for the most likely causes that are going to lead to that accident. That's a very rough analogy, but that's kind of roughly what we're going for.
The way we are doing it is, as Mort mentioned, using a tool called optimization in combination with a statistical model. So we have a lot of data on weather, renewables, demand, and we can use all of this data to essentially build a statistical model, which tells us how plausible or how likely something is. This data essentially informs us how likely something is to happen.
In combination with the physical model of the grid, or the way operators actually run their system, we can essentially put these two together to say: what is the most likely thing that can happen, given this is how operators run their grid?
Host
Extreme weather, renewable energy, and the addition of data centers — they're increasing, and they're increasing rapidly at this point. How do those trends make transmission planning more challenging than it used to be?
Anirudh Subramanyam
These are the three big challenges right now. As was mentioned, in the past, planners would just plan for the hottest summer day, which is, let's say, sometime in July, and the coldest winter day, which is sometime in January. And now we are noticing extreme weather happening all throughout the year. So you have cold days, unusually cold days, throughout the year. You have heat waves. You have wildfires. And we are not able to predict when these events happen.
We are not even able to predict wind properly, right? If you look at your iPhone app and you look at the wind speed, even tomorrow's wind speed forecast is actually quite off, right? So this brings us to the second aspect, which is renewable energy. So a lot of the U.S. now is transitioning to wind and solar, and we are not able to predict when the sun is going to shine, when the wind is going to blow.
And in a long-term perspective, because of policy, you don't know what kind of policies the government is going to implement and what will get enacted. We don't know what the grid in 20 years is going to look like, so we do not know if there's going to be a lot of wind, if there's going to be a lot of solar. And planners essentially are like, you know, it's like driving into the dark without headlights.
You don't know what the future grid composition is going to look like in terms of conventional coal and natural gas versus modern wind and solar.
And there's a third elephant in the room now, which is, as you said, data centers because of AI. And listeners may or may not know this, but they consume a lot of electricity. A small data center consumes as much power as a small city does. And so this is kind of breaking, again, old assumptions. In the past, demand was never like this; we could very reliably predict how much electricity people were going to consume. And now, because data centers and AI are coming up, no one knows, in 10 years, how many data centers we're going to have. No one has a clue.
Yes, we have certain confirmed contracts in place. We know certain areas where data centers are coming up. But this is just the beginning, and we don't know what the future is going to look like.
Mort Webster
The other thing that is new that is challenging is the uncertainty now, compared to the last 20 to 50 years, is where things will change. So if you build a transmission line, we're talking about the very large ones, like 765,000 volts. Those are those giant towers that you see running along I-80 that go hundreds of miles. Billions of dollars.
You put it up, and if you put it up, you're going to use it for 30 to 50 years. And traditionally, as we said, planners would say, in 10 years, how will the worst hour in August look then? And let's make sure we're still OK.
The difference now is that with all of these examples — so, for extreme weather, where will it hit? Where will the next storm, like the Texas example that Anirudh gave, hit? For renewable energy, we don't know. In 20 years, will New Jersey or Maryland build very large offshore wind farms, which they are considering, or not? It takes on average 10 years to put up one of these lines, these large, high-voltage, long-distance lines. You're talking a decade before it's actually up and operational. So you need to start planning now if you want the power from a wind farm coming into New Jersey to not be bottled up and not able to get anywhere. But we don't know if they're going to build it or not.
And then the problem with data centers: they are able to announce in pretty much every city in the U.S., oh, we're going to build one here. So every planner and every utility across the U.S. is panicked about how to meet those large demands that, as Anirudh was saying, are the size of doubling the amount of power in the next few years.
Should they be planning now for transmission to meet a lot of data centers being built in the Northern Virginia area, which is expected? Or will those companies end up saying, it's too expensive there; we're going to build a cluster in Cleveland, Ohio? And then you build different lines. But they need to build the lines first. Where do they build them?
This is where uncertainty is very new, because for the last 100 years, as electricity use grew, it grew proportionally to where population lived. And now everything's moving around, and they don't know which direction they need to ship power. If they panic and say, we must prepare for everything, that will be very, very expensive. So do we overbuild to be safe, but it's very expensive? Or do we underbuild but risk power outages?
Host
So what will people notice in the future in the absence of this research? In other words, what could we avoid by pursuing and implementing this research?
Anirudh Subramanyam
Our project is essentially helping people answer — or planners, I should say more precisely — navigate this trade-off between overbuilding and underbuilding. And the way that is going to effectively end up impacting the community, people like you and me, is in terms of lower bills on the grid side. This also is going to lead to fewer blackouts.
So this also means reliability. It means when you flip on the switch, the light does come on. If we do not do this, as Mort also mentioned at several points, we are going to be having more and more incidents like the Texas event in February 2021, or what happened in Spain last year, in 2025.
And the frequency is just increasing, which is a bit counterintuitive because we are also technologically becoming more advanced. So you would imagine that these events would go down. But the converse is happening. The research also can enable a clean energy transition in a more smooth way. It can help meet policy targets effectively.
Host
So thank you both. Thank you for coming on Growing Impact. Thanks for discussing your research.
Anirudh Subramanyam
Thank you. Thanks a lot, Kevin. Appreciate you having us.
Host
This was season six, episode five of Growing Impact. Thanks to Anirudh Subramanyam and Mort Webster for joining me. For a deeper look at their story, visit iee.psu.edu/podcast, where you'll find a video version of this episode, graphics, transcripts, as well as past episodes. Our communications director is Chris Komlenic, with graphic design and video production by Brenna Buck, marketing and social media by Tori Indivero, and web support by John Stabinger.
Join us next time as we continue our exploration of Penn State research and its growing impact. Thanks for listening.
