Finding Flexibility in Data Center Use, with Johanna Mathieu

Notable Quotes:

• The story behind flat and rising electricity demand: “It’s amazing that electricity demand was relatively flat over the past time period, given population growth, because we would’ve otherwise expected it to grow. But because of efficiency gains, we’ve really been able to curtail that growth … It is the commercial and industrial load that are growing quite quickly. In terms of commercial load, it’s data center growth that is a big contributor to it.” (3:53)
• Locational flexibility leads to grid flexibility: “The cool thing about data centers is that it’s not just that you can shift things in time, computing in time, but that you can also shift it spatially. So, if you have a company that owns multiple data centers across the grid in different locations, you can actually move the computing to other locations if they have capacity.” (11:15)
• A billion-dollar infrastructure-savings idea: “If you can better shift workloads, align workloads, and better utilize the capacity that is there, you don’t need to build as many [data centers], because they’ll be working at higher capacity factors … But the real savings and infrastructure costs are not the data center infrastructure costs—it’s the grid infrastructure costs.” (16:35)

Top of the Stack

• The Center for Informed Voices for Infrastructure Choices (CIVIC) Forum
• “Data Centers” episode of the Behind the Meter podcast

The Full Transcript

Kristin Hayes: Hello, and welcome to Resources Radio, a weekly podcast from Resources for the Future (RFF). I’m your host, Kristin Hayes.

My guest today is Dr. Johanna Mathieu, an associate professor of electrical engineering and computer science at the University of Michigan. Dr. Mathieu’s research focuses on the electric power sector, and, in particular, on ways to reduce the environmental impact, cost, and inefficiency of electric power systems via new operational and control strategies. She thinks a lot about distributed energy, flexible loads, and other fun things like stochastic capacity scheduling. Don’t worry—she’ll explain the technical terms.

All of these thoughts and research are in service of making our power system as flexible and responsive as possible. So, today we’re going to be talking specifically about how data centers—which are notorious for their vast consumption of electric power—might actually be able to help the power grid, if deployed in certain ways. It’s an intriguing twist on the typical conversation about artificial intelligence (AI) and electricity demand, and I’m looking forward to learning a lot. So, stay with us.

Hi, Johanna. Thank you so much for coming on Resources Radio.

Johanna Mathieu: Thank you for having me.

Kristin Hayes: Yeah, of course. Well, first and foremost, we always love to start with a question that allows our listeners to know the guest on the show a little bit more. So, I’d love to ask you about how you got interested in working on the power grid in the first place.

Johanna Mathieu: Great! Thank you. When I was an undergraduate student, I studied ocean engineering, which is pretty different. But for my senior design project, we had to build a machine to harvest energy from waves in the Charles River in Boston. It was a fun project where I got to learn about energy and renewable energy, and that piqued my interest in working in energy.

So, when I got to grad school, I knew I wanted to work on energy, although it took me a little while to find my path there of what energy I wanted to do. But I got connected with researchers at the Demand Response Research Center at the Lawrence Berkeley National Laboratory, and that launched me into a career in that direction, where I was studying demand response, which is how we can shift or shed electric loads to help out the power grid, especially in periods when there’s insufficient supply, or other kinds of reliability issues in the grid where you need to change the demand.

It was soon after the California energy crisis, so it was a really exciting time to be working on demand response and understanding different sources of flexibility on the grid. So, I did my PhD on that topic, which was looking at how we can coordinate the power consumption of distributed air-conditioning systems to provide balancing support to the grid. And then now, I work on a variety of different techniques and tools to increase grid flexibility.

Kristin Hayes: Okay. I have to ask just one quick follow-up. What is a distributed air-conditioning system?

Johanna Mathieu: Distributed. I meant air-conditioning systems distributed across the grid, so that could be residential air conditioners. Yes. Or, a commercial building HVAC system. So, these resources are all over the grid. And why we say distributed is just because, when we think about normal flexible resources, they’re usually big power plants that are all in a single location.

Kristin Hayes: Very fascinating. I thought, “Ooh, I’ve already learned a new term this morning.” I, in particular, love having a conversation about data centers and AI. It’s a particular interest of mine, but I do want to ground our conversation (no pun intended) in some basics about power consumption these days. I think it’s valuable to still remind all of us, myself included, about where we are with power demand in this country.

So, I think our regular listeners will have heard that for many years, power demand in the United States was largely flat. That’s really changed over the past decade. And rather than asking you a generic question of what’s causing that, I just want to jump to the chase: Is AI the culprit there?

Johanna Mathieu: It’s a piece of the puzzle, but it’s not the whole thing. First, I just want to say that it’s amazing that electricity demand was relatively flat over the past time period, given population growth, because we would’ve otherwise expected it to grow. But because of efficiency gains, we’ve really been able to curtail that growth, and residential demand is still growing really slowly, but yeah—it is the commercial and industrial load that are growing quite quickly. And in terms of commercial load, it’s data center growth that is a big contributor to it.

Although, I will say when we hear about it on the news, we hear about this massive growth. I think that’s predicted growth versus actual growth. There has been some actual growth of course, but I think that when you’re hearing these news stories about massive load growth, a lot of it hasn’t come yet—it’s in the plans. But there’s also other load growth that I just want to mention. So, industrial load growth: manufacturing load has also increased comparably to commercial load growth. And then, there’s also growth in terms of vehicle electrification and heating electrification like heat pumps.

Kristin Hayes: Well, that’s actually really helpful context to know—that some of this demand growth is actually projected, rather than here quite at this moment. But I would say even knowing that it’s coming, or having the strong sense that it’s coming, my feeling is that it’s led to some—I’ll refer to it as hand wringing—some real concern about the state of the power grid, and its ability to incorporate all of this new demand and the new load that’s going to be required to meet it—that it’s ready for that to do so reliably and affordably.

So, let me ask you to lay out some of the challenges that this either current or projected increased power demand for data centers, other causes, but largely for data centers for the purposes of this conversation: What challenges is that causing?

Johanna Mathieu: Yeah, that’s a great question. Some of the challenges, again, we hear about are speculative, but there are real challenges also that are on the horizon, or are already starting right now. And there’s a range of different impacts, from technical impacts to policy-markets impacts.

One of the things we hear most about is the need for new generation resources in the system to supply the data centers. And so, one way electricity markets ensure that there’s sufficient supply—so that we build enough power plants, essentially—is through capacity markets, where different power producers basically compete in a market to ensure their sufficient capacity to meet expected peak demand over the next year or three years, depending on the market design. And so, this is a little bit speculative in the sense that you’re looking at all of the different data centers that are in the queue, and guessing how much might come, and what the peak demand would be of the entire system—including those data centers—to ensure they’re sufficient generation.

And then, those capacity markets settle, and there’s prices for capacity that realize. And then, what happens is that, basically, utility companies have to pay for that capacity that they then are going to lock in. And so, we’re hearing in the mid-Atlantic region that the prices for capacity are really high right now because of the expected data center build-out. So, it’s somewhat of an economic and technical issue intertwined. And then there’s other impacts.

So, we often think about just making sure there’s enough generation in the system to supply, but that we also need to make sure that the grid is big enough and strong enough to basically move the power from the power plants to the data centers. And so, grid constraints—meaning like each power line has a specific maximum flow that it can handle—if the grid isn’t big enough, it can’t move that power.

So, we also need to think about building out the grid to make sure power can move through the system. There are these technical issues that can arise, too, like what we call “power quality,” where the quality of the power is defined by whether the frequency and voltage are correct.

Basically, all of our electric loads need a precise frequency of 60 hertz and voltage within a narrow range. Data centers, because they consume power in this … a little bit, well, it still looks like the grid a little bit, like a random way, increasing and decreasing quite quickly. It can affect power quality, and that can affect the performance of different generations and loads on the system that are connected that are expecting to see nice power quality. So, all of these things together are the range of effects that you might see.

Kristin Hayes: Interesting. Can you say just a little bit more about the randomness in the way that data center power consumption shows up?

Johanna Mathieu: Yeah. So, different data centers are going to have different randomness, but we’ve looked at traces of power consumption from data centers that are training large language models (LLM), for instance. And you can see the power moving up and down very quickly on the timescales of seconds, or even sub-seconds, as the different jobs are completing and moving to different pieces of the process of training these models. And as the power is going up and down very quickly, the voltage is also going up and down quickly, because power and voltage are related to each other. And so, that’s affecting how the grid is behaving.

Our power grids work with alternating current (AC) power, where we have these nice, sinusoidal waves of current and voltage that are enabling us to transmit power. But as you add more of this type of fluctuation, it makes those waves not look nice anymore. So, instead of just up and down, up and down, there’s all these things on top of it called harmonics, and that can cause problems with different equipment, especially power-electronic equipment in the system. Some of that power-electronic equipment can actually solve the harmonics problem by mitigating the effect, but some of it can be impacted negatively.

Kristin Hayes: Interesting. There’s some new terms for sure. Fantastic.

Well, another thing that, anecdotally, I feel like I read a lot about in the news is just this frankly, nationwide, bipartisan, and pretty broad concern about the pace at which data centers are being built—community concerns about data centers. And so, I wanted to flag: I’m really interested in the work that you’re doing as part of a team of researchers at the University of Michigan, leading something called the CIVIC [Center for Informed Voices for Infrastructure Choices] Forum. I think it’s such a nice name (sorry to be cheesy for a second), but you guys really focus on the civic aspect of this, even though it’s quite technical work.

So, my summary of the CIVIC Forum is that it’s a body of work, and a body of engagement, that aims at improving evidence-based decisionmaking on the part of communities about AI infrastructure development.

I hope that was a decent summary. You’re welcome to correct that. But you and your team identified four primary data gaps that underpin all of the CIVIC Forum’s activities. And one of those reads, “Interconnection studies treat data centers as rigid loads, ignoring that AI workloads can shift training by hours, or defer inference by seconds. This flexibility could reduce infrastructure costs by billions, if coordinated.”

So, lots to unpack in there, but since that was one of these key pieces that was called out, and it seems right at the heart of what we’re talking about here, I wondered if you could just unpack that a little bit.

Johanna Mathieu: Yeah, sure. That’s great. This is why I’m really excited about the opportunity that data centers present now, because I think that they are these massive flexible loads. So, flexible loads, again, relates to this concept of demand response, where loads can change their behavior to support the power grid. So, it could be, again, reducing consumption, or just shifting consumption. And usually, we think about shifting consumption in time.

Data centers can do this easily, depending on what you’re using the data center for. Some computing tasks within the data center are harder to shift. But the cool thing about data centers is that it’s not just that you can shift things in time, computing in time, but that you can also shift it spatially. So, if you have a company that owns multiple data centers across the grid in different locations, you can actually move the computing to other locations if they have capacity.

And so, in that sense, they become this very flexible resource. The trick is, of course, understanding how the computing jobs relate to flexibility. Like I said, when you’re doing certain types of computing, you might not have a lot of flexibility. So, for instance, if you’re going to put a query into your favorite large language model to ask it a question, you probably don’t want, on the other end of them, to think about how much to delay the response so that you have to wait for your answer.

So, that sort of thing may not be super flexible, but if you’re going to send a larger job to it—for instance, if you needed to do some data processing for you, and you’re going to send that overnight, and tomorrow morning you’re going to wake up and look at the results—it might be okay. It may be that the whole job takes 6 hours and you’re giving it 12 to do it, and do it in the best possible way, at the right times and locations to support the grid.

So, that’s the type of flexibility that I’m talking about here. So, like I said: some processes and jobs are flexible, and some are not. And we’ve actually been using data centers for the better part of 20 years, I would say, to do this sort of thing. And the data centers … We talk about data centers like they’re a new thing. They’re not, of course. They’re just becoming much more hyped right now because of AI and large-language-model training, and all the different platforms that we’re using to leverage AI in our workflows.

But data centers have been there for a long time to store data, to do computations, to do scientific computing, but also commercial computing and different things like that. So, this type of thing has been done, and now we can do it at scale with these really large-scale resources. So, data centers are flexible in terms of their computing, but they also have other flexible loads.

They use a lot of cooling because they need to cool the chips, and everything inside the data center that’s producing heat-cooling systems are inherently flexible, just like the air conditioners that I was talking about before. Cooling systems in data centers, traditionally, were just basically HVAC or air-conditioning systems as well. And you can also change a little bit how they work, consuming power, and make them flexible resources.

Now, data centers are using liquid cooling, and we need to understand a bit better how flexible those processes are, and data centers are companies that are often investing in behind-the-meter resources like batteries and behind-the-meter generation, and that also makes them a flexible resource.

Kristin Hayes: Great. Well, two quick follow-up questions for you. One is about the last sentence of what I read from the website, which was about reducing infrastructure costs by billions. So, I wondered if you could just say a little bit more about that piece—is the idea that by harnessing this flexibility, we could build fewer data centers and potentially address some of these cost concerns, land use concerns, but also real community concerns about having one in their backyard?

And then, I guess my other follow-up question, if you’ll bear with me, is that you mentioned this idea of, you might be able to give a system 12 hours instead of 6 hours for a job. Who’s actually making that decision? Is that a user who’s actually able to decide that, or is it really guided by the data center companies, or the companies that own those data centers, or the grid operators? Who’s able to operationalize that flexibility?

Johanna Mathieu: That’s a great question. So, ultimately whoever’s doing the computing needs to be able to have some say, of course, in how flexible they’re willing to be. So, it’s the user, but the grid can incentivize it.

For instance, with traditional demand-response programs, the utility is financially incentivizing customers to behave in a certain way to shift their load, to reduce their load, and so forth, by either exposing them to time-varying electricity rates, so that they encourage them to shift their load to times when electricity is cheap, or by actually paying incentives for reductions or shifts.

And so, you can do the same thing in this case where the utility could do that with the data center company, and the data center company could then offer that as a service to the folks inside their facility that are using the computing, but it could be something that a user sees.

So, in that example I gave you, where you need data processing overnight, maybe you have a bunch of options that it’s going to cost this much, this much, or this much based on how much flexibility you enable. So, you make the choice. But if you need it done right away, you say that, and you pay for it. And it could be tied to payments, but it could also be tied to other incentives. Maybe instead of a payment, you learn that if you allow more flexibility, it reduces the carbon emissions by this much, and you make that choice.

So, it has to be based on some kind of explicit consent and agreement. And so, the way they do this with the companies is through contracts. But with users, it would be hopefully to give you a menu of options that you would select.

Kristin Hayes: Oh my gosh. I’m just thinking about my economist colleagues who would have a field day with all that data about willingness to pay for various—

Johanna Mathieu: Exactly.

Kristin Hayes: Yeah, interesting.

Johanna Mathieu: Lots of folks in my field work directly with economists to figure out all of these mechanisms and designs for how you think of the economics together with these technical issues, because you can’t pull them apart. They’re so intertwined.

Kristin Hayes: Yeah. Well, can I ask you to tackle that maybe larger, second question?

Johanna Mathieu: Infrastructure?

Kristin Hayes: Yeah.

Johanna Mathieu: Yeah. So, how do we reduce the infrastructure cost by billions? So, there’s two components of infrastructure costs.

One is the data centers themselves. So, if you can better shift workloads, align workloads, and better basically utilize the capacity that is there, you don’t need to build as many, because they’ll be working at higher capacity factors. So, basically, they’ll be utilized at a higher amount than if everyone’s doing their own thing and not doing this kind of coordination. But the real savings and infrastructure costs are not the data center infrastructure costs—it’s the grid infrastructure costs.

So, it’s reducing how much new generation you need to build, because generation needs to supply the peak demand. So, reducing the peak always helps immensely, and you can reduce the peak by this shifting—but also by reducing transmission, and subtransmission, and distribution costs. So, building out the lines and the wires and the transformers that enable us to transmit power, and so forth.

Doing all of that means that you have this flexible resource, you can control power flow more, and you don’t need to build out a massive system that is just there for the worst case.

Kristin Hayes: Well, and I can see that, in particular, helping address one of the other real, public concerns around all of this build out is rising electricity costs, which are driven—at least to some degree, I think—by the need for more of this grid infrastructure. So, I can definitely see how that would also matter a lot to a wide constituency.

Well, let me ask you a little bit more about your role in the CIVIC Forum, which focuses on grid-impact evaluation, which we’ve been talking about here. What are these different strategies for AI training and inference workloads? I think inference just means, basically, anytime that you ask an LLM to answer a question for you—that’s considered an inference incidence. Is that right?

Johanna Mathieu: Yep.

Kristin Hayes: Okay. So, what sort of strategies … You’ve referenced a couple of things already, but I’m curious if you can say a little bit more about what sort of strategies—on the part of the AI companies themselves or the data centers—what are you able to test?

And I’m really curious to know … I think I saw the word “testbed” on the website, and I’m really intrigued by what a testbed looks like in practice. So, I’d love to know more about what you’re testing.

Johanna Mathieu: Yeah, that’s a great question. So, I mean, ideally what we’re testing is how to do this in real life. So, our test bed ideally would be a large-scale data center that we could play with, and we could manipulate the resources. What we actually have in the lab … Actually my colleagues, Vladimir Dvorkin and Mosharaf Chowdhury, they’ve built a small test bed in the lab with a few ground power units (GPUs), which basically emulates a data center, and they have a grid emulator that’s modeling the grid at high fidelity, and other simulators to test the flexibility of the computing system. So, both to measure how these data centers are consuming power as you give them different types of jobs, and then understanding the flexibility in that. So, I think the first challenge is really to characterize the flexibility as a function of the job type, and then to understand across different platforms how that flexibility changes.

But to get a little bit more nuance there, you can imagine that when you’re doing these types of things like training, for instance, they run these training algorithms for very long periods of time, weeks, months, and so forth to get to a point where they can output a new model. And through that process, you can imagine that if you’re running something for months, there must be some flexibility there, but you might just think, “Well, they’re running it at full power all the time, and they’re just using everything they can.”

But you can imagine that there could be cases where you need to compute a bunch of things simultaneously, and all these things are being computed in parallel by different units, and then they’re going to come together to produce an answer that will then enable you to go to the next step of the computation.

In practice, all these things are going to end at slightly different times, and you’re going to be waiting on that one last job to finish so that you can move on to that next step. And that’s where this inherent flexibility is in saying, “Well, if things are waiting, then I could have shifted when that was happening, so that I could have better aligned the endpoint such that they could have moved to the next step together at the same time.” And so, that’s the stuff that we’re trying to understand without getting into too many of the further nitty-gritties. I know that was already a little bit nitty-gritty.

Kristin Hayes: No, that was fascinating.

Johanna Mathieu: Yeah. And so, I think that’s what we’re trying to understand now: What is the inherent flexibility? We don’t necessarily want to compromise on time to finish tasks, or the quality or the performance of the output, but we want to understand, in the computing procedures, what part is flexible.

I’ll also say that this goes along with research that’s looking at just reducing energy usage of each of these steps as well, because I think that has to happen. That’s necessary. We want to reduce the consumption as much as possible, and then leverage the remaining flexibility to do this kind of shifting.

Kristin Hayes: Okay. And what does reducing the consumption look like in this case? I feel like I’ve got a pretty good handle at this point in our conversation about the flexibility side. You’ve actually done a fantastic job of explaining these things that are very complicated, but what is reducing the energy demand for any individual step, or a combination of steps, look like? Is it about chip efficiency? Is it about what else could go into play there?

Johanna Mathieu: Yeah, it’s everything. It’s the hardware side, for sure. I have colleagues in my department here at the University of Michigan that are working on better hardware that basically has less losses, less … These things generate tons of heat. All of that heat is losses. It’s efficiency loss. How can you reduce that? So, it’s the hardware design, and it’s a software design. It’s how we think about doing computing in ways that mitigate usage of power as opposed to just producing code that “works.”

Usually, when we produce code, we’re not thinking about making it as efficient as possible unless we need to think about the speed and costs of electricity. And traditionally, when you’re prototyping something, you don’t worry about that. So, the first versions of all of this didn’t worry about that. And then, they become much more efficient as they realize that the huge cost of doing business here is an electricity cost. So, all of that comes together, but yeah, it’s like an “everything and” approach of trying to reduce energy.

Kristin Hayes: Interesting. Well, I wanted to just go back to something you said earlier about referencing that companies often have data centers in a number of locations—maybe a wide set of geographies, I’m not sure—but it strikes me that geography, generation mix, and all of those factors are going to contribute to how any given company or facility is going to tailor its strategies to actually take advantage of some of this flexibility.

So, I’m wondering if you can just give an example or two of how those strategies might be tailored to reflect some of those distinct, again, geographies, generation mixes, et cetera.

Johanna Mathieu: Yeah, that’s a great question. I think when I think about geographies, I think about places that are dry and don’t have access to a significant amount of water for cooling, for instance, versus other places that have more access. And so, that affects how they design their cooling systems, and whether they use an open-loop or closed-loop water-based cooling system, for instance, or if they can even use an air-based cooling system or not.

Modern data centers are moving away from air-based cooling because it just doesn’t scale at the sizes that we’re building out these data centers. And when I think about generation mixes varying across the country, I think about how some of these companies are really interested in understanding how to mitigate their carbon emissions, and they’re looking at trying to align power consumption with times of low carbon on the grid.

And so, that’s when wind and solar are producing a lot such that they’re on the margin of the production curve. And so, in that sense, you can imagine them scheduling things to be at different times at night if you’re in a wind-dominated area, where the wind tends to produce more at night, or during the day for solar. But I think that what’s driving differences the most across the country is policy and regulation in different states.

Kristin Hayes: Of course.

Johanna Mathieu: Yeah, of course. Exactly. I’m an engineer, so I don’t want to talk too much about policy, because I’m wading into this space where I’m not an expert. But the interesting thing from an engineering standpoint is that it affects how you design the system, too, if you have to worry about policy about how you can interconnect with the grid, and how long that might take based on the utilities interconnection queues, but also the utilities responding to state policy because they’re regulated by public service commissions.

And so, that affects how fast they go, and what they require, and how they negotiate, and everything. And so, that’s been really, really interesting to see. In a specific example, in Michigan, there’s a large data center going in not too far from where I am in Ann Arbor, and they’ve negotiated … The utility here has negotiated with the company, with support from the Michigan Public Service Commission, to basically have the company also buy grid-scale batteries.

And the batteries are not going to be located at the data center. They’re going to be located across the state of Michigan, and owned and operated by the utility to improve reliability and resilience of our entire grid, which will also support the data center, which needs high reliability and resilience. But it’s a different model: instead of the data center taking care of itself behind its own meter, it’s having benefits across the entire system.

So, they have all these, what we call “front-of-the-meter resources.” Again, they’re not the data centers’ resources—they’re going to be the utilities’ resources—but as part of the interconnection agreement, this is happening. And so, this leads to a different kind of development than in certain states, where they’re building very large-scale power plants, or assigning contracts with nuclear power plants to get gas power plants behind the meter with data centers—it’s becoming popular in part because of this issue with interconnection.

It’s taking too long, and there aren’t enough resources, so they’re just buying their own gas plants and putting that behind the meter. But that can’t happen in every state, because it depends on regulation. I find this really fascinating, because all of these rules are affecting how we as engineers do our job, and design these data centers, and figure out how best to have them interact with the grid.

Kristin Hayes: Yeah. Boy, that’s such a great illustration of how we talked, already in this conversation, about how the economics and the engineering really intersect in lots of ways, but these layers upon layers of policy, engineering, economics, and consumer behavior—which I guess is economics—but all of these pieces put together, it’s an incredibly complex system. And without looking at each of those pieces in turn, and then combining them to look at them holistically, that combination of disciplines seems really, really important.

And I think that’s one of the things that the CIVIC Forum also does, is that it is this interdisciplinary look at all of these interconnected issues. Again, I think that’s really, really intriguing. And it does sound like you have a home base in Michigan, where some of these issues are playing out in practice, so you can really see how communities are responding, how companies are negotiating, and how the utilities are responding. So, to me, it’s a really fascinating set of real-time and very important issues that you guys are tackling.

Johanna Mathieu: Yeah, exactly. Looping back to your question about testbeds, that is our testbed, I guess, is the whole system—the real system—and trying to react to that. But what we realized with the CIVIC Forum—and I should credit my collaborator, Mosharaf Chowdhury, with framing it as a CIVIC Forum and not sticking data centers in the title of that—but he’s a computer scientist who studies how to make these data center systems as efficient as possible, specifically the chips and the GPUs. But we also have on our team lawyers, and we have people who are energy planners and think about energy siting. We have engineers, of course, and economists.

And I think you just need to bring together that group of people to fully understand the issue because otherwise, I think that what commonly happens in engineering, at least, is we develop solutions and look at trade-offs—technology trade-offs—and say, “Well, if the policy was like this, this is what we could do. And if it was like this, this is what we could do.” And then, it often ends there, where we put it out in the world and we say, “Policy people, please look and see.” Or, “Economists, do an analysis with this.”

Kristin Hayes: Just please do it this way.

Johanna Mathieu: Yeah, exactly. But if you’re not all working together, it doesn’t necessarily go anywhere. And so, I think that’s where we really need to pull together this very multidisciplinary team to make progress on this issue.

Kristin Hayes: Yeah. Well, great. I definitely encourage people to take a look at the CIVIC Forum website. It’s very easily understandable, too, and really gives a pretty comprehensive look at what you guys are doing, which is just fantastic. And you obviously have much more work beyond this, too, but yeah, I just really appreciate your time today talking us through these broad issues, as well as some of the applications within the CIVIC Forum.

So, with that, I will close with our regular feature, Top of the Stack. I would love to ask you to recommend some content that you think our listeners might enjoy—of any media. So, Johanna, what’s on the top of your stack?

Johanna Mathieu: So, I thought long and hard about this, but I’m going to totally nerd out with a podcast by the Michigan Public Service Commission, which is called Behind the Meter. You can access it on the Michigan Public Service Commission website, but it’s a super neat look behind the scenes of how state regulators think about regulating electricity and the utilities. They’ve recently had an episode on data centers, and how they’re thinking about data center regulation, and how Michigan differs from the rest of the country.

For your listeners who are, obviously, probably all across the world and not just based in Michigan, maybe this will seem too Michigan specific, but I think it’s fascinating how much they’ve opened up an understanding of their decision processes, and how they interact and relate to stakeholders. And I’ve learned a lot about state regulation of the electricity industry this way.

Kristin Hayes: That’s a great recommendation. And given how much of this is place based, I acknowledge that every place is going to be different, but that level of transparency, no matter what place it’s in—I’m sure our listeners are going to love it.

Johanna Mathieu: Great.

Kristin Hayes: Yeah, I really appreciate it. It’s a great recommendation.

Well, thank you again, Johanna. This has been really fascinating. I really appreciate your taking the time.

Johanna Mathieu: Thank you very much. It’s been fun.

Kristin Hayes: You’ve been listening to Resources Radio, a podcast from Resources for the Future, or RFF. If you have a minute, we’d really appreciate you leaving us a rating or a comment on your podcast platform of choice. Also, feel free to send us your suggestions for future episodes.

This podcast is made possible with the generous financial support of our listeners. You can help us continue producing these kinds of discussions on the topics that you care about by making a donation to Resources for the Future online at rff.org/donate.

RFF is an independent, nonprofit research institution in Washington, DC. Our mission is to improve environmental, energy, and natural resource decisions through impartial economic research and policy engagement.

The views expressed on this podcast are solely those of the podcast guests and may differ from those of RFF experts, its officers, or its directors. RFF does not take positions on specific legislative proposals.

Resources Radio is produced by Elizabeth Wason, with music by RFF’s own Daniel Raimi. Join us next week for another episode.