This is the fifth installment of our Science at the Edge series, where we explore the benefits, tradeoffs, and risks associated with innovative solutions while unpacking questions about ethics, policy, or public perceptions.

The AI boom is driving surging electricity demand from new and expanded data centers, putting pressure on Colorado’s clean energy goals. At the same time, AI may enable us to optimize the grid in new ways. In this panel discussion from August 11, 2025 with Chris Hansen, CEO of La Plata Electric Association, and Kyri Baker, Associate Professor of Civil, Environmental, and Architectural Engineering at the University of Colorado, Boulder, moderated by Max Neumeyer, Institute Deputy Director of Policy and Engagement, we discuss the emerging challenges and opportunities AI presents for our energy grid and the decisions facing Colorado utilities and decision-makers.

Watch a video of this discussion on our YouTube channel.

 ​AI and Energy

MAX NEUMEYER: Hello everyone, and welcome to today's Science at the Edge webinar. My name is Max Neumeyer, Deputy Director for Policy Engagement at the Institute for Science & Policy, which is a project of the Denver Museum of Nature & Science. As many of you know, our mission at the Institute is to be a catalyst for thoughtful dialogue, working towards solutions on society's greatest challenges with scientific thinking, empathy, and inclusivity.

Our Science at the Edge Webinar series is one way we bring that mission to life, and we do so by exploring new ideas and innovations in science and technology that impact society, looking specifically at trade-offs, risks associated with innovative solutions, and unpacking questions around ethics, policy, and public perception. As always, we're not here to prescribe answers, but to create space for informed and nuanced discussion. Today's installment is on AI and energy, a conversation that sits at the intersection of two massively important and rapidly changing fields.

As we all know and are experiencing, the AI boom is changing our lives. It's also fueling unprecedented growth in electricity demand, driven by new and expanding AI data centers. As we'll hear today, that growth poses some real challenges for Colorado's clean energy goals. At the same time though, it also holds promise for better ways of managing our electric electrical grid.

To unpack this topic today, we're thrilled to be joined by Kyri Baker and Chris Hansen. Dr. Kyri Baker received her BS, MS, and PhD in Electrical and Computer Engineering at Carnegie Mellon University. Since 2017, she has served as an Assistant Professor at the University of Colorado, Boulder in the Department of Civil, Environmental and Architectural Engineering with a courtesy appointment in the Department of Electrical, Computer and Energy Engineering. Her research interests include power system optimization and planning, building-to-grid integration, smart grid technologies, and renewable energy.

Chris Hansen is the CEO of La Plata Electric Association, based in Durango, Colorado, and has spent over 25 years in the global energy sector. Previously, Chris served in the Colorado Senate representing District 31, which includes east, central and southeast neighborhoods of Denver. During his time in the Colorado Senate, Chris served on the joint budget committee, on capital development, house transportation and energy, appropriations, agriculture, livestock, and natural resources committees. He also holds degrees in nuclear engineering, civil engineering, technology policy, and economic geography. Kyri and Chris, thanks so much for joining us.

CHRIS HANSEN: Thanks, Max.

KRYI BAKER: Thank you.

Current Grid Infrastructure & Energy Demand

MAX: Kyri, to set the stage and give us some context, can you explain what the difference is between traditional data centers and AI data centers, and in what ways are the challenges of AI data centers new and different?

KYRI: Yes, I think in this whole discussion today, we should be clear about what we mean by “data center.” The discussion today will focus on AI data centers, which are data centers used to train or use AI models. You also have things that are conventional data centers, like internet data centers, that process or store information related to the internet. And a bunch of others, like there are technically Bitcoin data centers.

There are a lot of different types of data centers, but what makes AI data centers particularly interesting is that they consume very, very high levels of energy, and they also have quite high utilization rates. So you're running GPUs, or graphics processing units, which are high-power processors that help train or teach these AI models to learn. These are popping up on the grid en masse and consuming huge amounts of energy, which is why we're so interested in them. We want to know what their effect is on energy consumption, cost and emissions.

MAX: Absolutely. Chris, can you walk us through how data centers driven by AI are impacting the demand for energy in Colorado? And I understand you have a few slides for us to help us understand the landscape.

CHRIS: Yes. Thanks Max. I thought it might be good to pick up where Kyri left off on what it's doing to the grid. Where is this demand, how does it fit into our current infrastructure and where might the challenges be, is a good way to set up our conversation today. So, let me start with the spaghetti map of Colorado [Figure 1]. 

Figure 1: Colorado's 10-Year Transmission Plan (courtesy La Plata Electric Association)

Here is the 10-year transmission plan for the state of Colorado. This is a couple of years old now, but all of the black lines are the existing lines in the state. And then the colored lines are the places where the engineers have identified that we need more capacity. So, this is what it looks like in our neighborhood here in Colorado. If we zoom out, you can start to see what the larger picture looks like [Figure 2]. 

Figure 2: Uncoordinated Grids in the West (courtesy La Plata Electric Association)

We've got the basic problem in the West that we have a lot of uncoordinated grids. For most of the country, they sit inside what’s called an RTO, Regional Transmission Organization, or ISO, Independent System Operator. And that is true for most of the rest of the country. And it's really just the Western interconnect, from the Colorado/Kansas border to the Pacific where we have this problem of uncoordinated grids. And it's something that I've spent a lot of time working on, and I know Kyri studied the hurdles that we face because we live in the region without that grid coordination.

Figure 3: Renewable Energy Generation through the US (courtesy La Plata Electric Association)

Here [Figure 3], you can see the other fundamental problem that we have, which is that we've got lots of new resources that we can bring to bear. This happens to be a map that NREL put together. We've got the wind maps, the solar maps, and the purple areas where we have great wind and solar.

But you can quickly see the problem, which is we don't have a lot of grid resources to bring those existing wind and solar, or new wind and solar and battery storage options into the market. And so, we’ve seen big growth in those areas over the last several decades as the prices of wind, solar and batteries have fallen very rapidly. And then you match that up with the existing coal, gas, and hydro facilities that were already on the grid.

But lots of disconnect, literally, between where the resources are and how we get those resources to things like data centers. And if we go to our last map [Figure 4], we try to overlay all of this in one place. 

Figure 4: Data Center Infrastructure in the U.S. (courtesy NREL)

And this is a map that was recently published by some great folks at NREL, looking at the major transmission highways, if you think of it that way. And the darker and thicker the line, the more capacity that line has. And then we overlay that with where the big data centers are happening--in the Phoenix area, the Reno area, and Northern Virginia are the places where this is happening the most. But you can see big growth in Texas too.

In Colorado, it's relatively small right now, but there's tons of interest in bringing data centers closer to the demand for those data centers. And I'll let Kyri explain the difference between high latency and low latency data centers in just a second. But I think these maps give us a good understanding of the status quo of where we start from when it comes to the grid. Pretty weak in the center of the country, and stronger on the edges, because historically that's where the supply and demand has been biggest.

But now with this massive new demand coming from data centers, we really need a better connected, more coordinated grid so we can move power back and forth. And the phrase that I like to use is, we really need a grid bigger than the weather and bigger than multiple time zones. And that's one of the reasons that I've been working so hard to try to coordinate those grids and create an RTO or ISO in the West, because that would help us to solve this problem.

And of course, the extra pressure that's coming from the new demand of AI and data centers, is just adding to the scale of the problem. With that, I'll pause, but hopefully those maps give people a bit of an idea of what we're up against, and I look forward to this conversation.

AI Data Center Demand in Colorado

MAX: Great. Thank you. Kyri, can you put the demand growth that we're seeing in perspective? How much more electricity will we need for AI, and can the current grid handle this kind of growth, or will we need major new infrastructure investments?

KYRI: I don't have the numbers off the top of my head, but if you look at the entire country and the projected amount of energy consumption from AI data centers, it will be multiple Colorados, so quite large. And the demand for that energy is location-dependent. We have areas of the country where there's honestly overbuilt renewables that are getting curtailed every day. The prices are going negative.

I saw a question in the chat, "What does it mean for a grid to be coordinated or uncoordinated?" That's like, is your portion of the grid in a competitive market or not? If you're in a competitive market, these ISOs that Chris mentioned, everyone submits supply and demand curves, you clear the market, and you match supply to demand. Everything is coordinated across the region.

And in uncoordinated areas, there's a lot of power plants that schedule themselves, or utilities that aren't communicating with each other in the same way. So, in areas where you have these markets, sometimes there's just so much excess power or there's constraints in the grid where you can't send power where it needs to go, where the price is going negative.

There aren’t many other things in economics where you actually get paid to consume more of a good. Electricity is one of those where it happens every day in Texas and California. So in terms of do we need more power? Yes, but we can also utilize some existing resources a little bit better, depending on where those data centers are located.

And one more thing. Towards the type of AI data centers, there's actually two more subcategories that are important to distinguish. One of them is when you're training a model like ChatGPT or something, you're teaching it to learn English language. You're teaching it grammar, you're teaching it historical facts, and then after you've taught it that, after you've trained the model, you use it. So, people send a query, it gets sent to the model, the model processes it, and it produces an answer.

So there's difference between a training load and a demand load. For training, people specify, I want my model to be trained on this specific type of hardware. I want it to be done by this specific time. I'm willing to pay this much. That's hardware specific. You can't really shift that around to different data centers, whereas things like internet requests can be shipped around to different data centers, whoever can process the requests the fastest. So that makes these interesting in terms of load curves and shapes and locations.

MAX: Fascinating. Chris, can you give us the view from Colorado specifically? Are we looking at the need for new major infrastructure investments in our state? And if so, who's gonna pay for these updates?

CHRIS: The short answer is yes. There are lots of folks who would like to build data centers in Colorado. There are a few hurdles that we can talk about today as to why Colorado is behind some other states when it comes to these data centers.

First and foremost, our electricity prices are kind of in the Goldilocks zone. They're not the highest in the country. They're not the lowest in the country. And of course, data centers want to go where they can get the best price possible for electricity because they'll use megawatt hours of it during the operation. So they want that as cheap as possible. The other thing data centers have to look at is, can they get enough fiber capacity, the backbone of the internet. That is largely a problem that can be solved with more modest investment.

The biggest hurdle really is, can we get enough electricity? Colorado does have a little bit of excess capacity, not a lot. And our prices are in the middle of the pack. And so we haven't been a first port of call for some of these big data centers, but some of the data centers need to be close to the customer, where they want low latency. Some of the companies are like, we need to be right next to Metro Denver because there are applications that we want to serve with milliseconds of lag. There are other places where you can do the training of the models where you don't need to worry about latency. And so Colorado is on the map in a sense, but not at the top of the list for new data center developers.

Now that said, the good places have kind of been used up: the places where there was immediate power available, where there was great confluence of fiber and power availability. Those have largely been staked out. And so now data center developers are going everywhere in the country looking for where can we get the things we need. Cheap land, cheap electricity, and fiber access are at the top of their list. And can they hire the employees they need to run the data center.

So lots of attributes. The data center developers are looking at Colorado's middle of the pack as far as how good we are and how attractive we are for those data center developers. But at this point in the demand, it's kind of anywhere they can get the power. And so I guess the other thing I would mention is, the other big hurdle are the physical transformers. You might be able to get electricity on the grid, maybe you've got that problem solved, but then you've got the next problem of being able to get the right voltage at the substation and being able to deliver it to the data center. And that in and of itself is a giant hurdle, taking several years to get some of those pieces of equipment.

So lots of factors here. It's of course a demand story, and supply trying to catch up. But it's also just a physical infrastructure story that you can see on the maps that we displayed.

KRYI: And the physical infrastructure is often overlooked. It's not just that you need enough installed power plants to supply these. You need the lines, you need the transformers. So even the cylinders on the power poles, the green boxes in your neighborhood--there are back orders for those in some areas of the country of years just because we have demand increasing, such as people getting electric vehicles, and other things that are increasing demand.

And they weren't sized for that in Colorado. A lot of the transformers weren't even sized for air conditioning load. A lot of them are consistently reaching their limit. So you can imagine for a large data center, the supply chain issues and backlog of trying to order a big transformer, it can be many, many years.

That's why you'll see a lot of these companies--Google, Meta--putting data centers that are already in existing locations with large transformers. They're restarting Three Mile Island. They're using infrastructure that was already built out for power plants because they can reuse those poles, wires and transformers.

Funding Physical Infrastructure Changes

MAX: When it comes to this new physical infrastructure, who’s going to pay for it? Is it going to land on rate payers, or are data center operators going to have to foot the bill? What does the conversation look like from your perspective, Chris?

CHRIS: This is such an important question, and I think we've got examples that are positive in this sense, and we have examples that are not so positive as far as how costs get allocated.

And what do I mean by that? I have the honor to lead the La Plata Electric Association. We serve about a hundred thousand people, about 50,000 meters in the southwest part of the state. And I'm trying to do that and maintain great affordability for my members. So, the lowest cost electricity possible, while meeting our reliability and environmental goals.

And if you think about a new data center coming into a part of the grid, that's a big new lumpy load. It's 100 – 200 megawatts. And the hyperscalers are looking at gigawatts--a thousand megawatts--at a time of new load. So, the infrastructure to supply those new loads has to be built, and somebody's got to pay for it.

Now, in most cases and most systems, whoever's coming in and needs the new thing has to pay for it. So the hyperscale developers, the big tech companies, they're going to write a big check. And they'll go off and order the transformers and the new lines and all the things that are needed, and they pay for that new development.

But there are cases like in Northern Virginia where the incumbent customers who were there on the system before the data centers, in some cases, are getting some of that cost loaded onto them because now their price of power and price of capacity has gone up. Or there are shortages of the transformers that they need for the existing load or the native load, which gets squeezed out by the new data center development.

So lots of ways where costs can get shifted onto those existing customers if you are not careful. And I think that's one of the things that Colorado's trying to be super careful about with the conversation at the public utility commission that applies to Black Hills and Xcel. And then in the individual co-ops and municipal systems, we're talking with our boards of directors about, if we bring in a new load, we have to make sure we're not hurting our existing customers. That's a super important conversation, and we're working really hard to make sure that the right person is responsible, and the right company is responsible for those new costs.

Speed of Renewable Energy vs Speed of Load Growth

MAX: One more question about the challenges we face. Colorado set some really ambitious clean energy and decarbonization goals. To what extent will the growth in energy demand from tied to AI jeopardize these goals? I'd love to hear thoughts from both of you.

KRYI: I think Chris is better to answer this, but my take on that is just, it's so hard to predict because, like Chris said, we don't know right now how many data centers will be installed in Colorado. So it's the speed of renewable energy installations versus the speed of load growth. And load growth also means electrification: it's getting hotter, AC loads are going up, people are consuming more electricity. So all those things together strain the infrastructure as for where the physical electrons are flowing. That will depend on where the data center is placed and what other resources are being installed at the same time.

CHRIS: Yeah, I think that's a really great way to think about it. The pacing element is the one that I think a lot about. We have projects in La Plata Electric Association where we're looking at new generation hydro, wind, solar batteries, new gas, fire generation, geothermal. So we're looking at all these different options. Some have emissions, some don't. Could we use CCS, carbon capture and sequestration? So we're really trying to look at every option under the sun and look at the entire portfolio of choices to meet our goals.

And state law right now requires 80% decarbonization by the end of this decade. That applies to all the electricity operators in Colorado. And that task is now more difficult than it was before the federal changes. There's no doubt about it. It will be more expensive. There are tax benefits for carbon capture and sequestration, but there are not tax credits for wind and solar projects after this year. That is a more difficult thing to do from the federal policy standpoint. The good news is Colorado still has a lot of great resources and a lot of great places to do zero-emission projects. And so really what I'm thinking of, in my responsibility at La Plata Electric, is how do we create a lot of optionality, a lot of flexibility.

And it's not just generation, it's got to include the upgrades on the distribution system, the upgrades to the transmission system so that you have lots of choice and can take advantage of different things during different parts of the day, different parts of the year, all of it coming together in concert to be able to meet these goals.

Now, no doubt about it, data centers add pressure to that. But, there's also the opportunity that the data center developers will pay for more generation projects and more transmission projects that will make it overall easier for the system. And I think that's really the goal for Colorado, is how do we channel this new investment, this new demand in a way that really helps the whole state to meet its goals?

It's tough. That is a tough engineering challenge. That is a tough policy challenge. But I think that's really the goal for Colorado over these next 10 years.

AI Data Centers & Clean Energy

MAX: Turning now to some of the potential solutions and opportunities that AI brings with it, Kyri, I want to start with you. This has been asked in the chat: to what extent can these large AI data centers be run on clean energy, either by equipping them with solar panels, siting them near other renewable energy resources, building a small modular reactor (SMR), or equipping them with energy storage? Can you talk to us a little bit about some of those solutions?

KYRI: It's extremely location specific. For example, with solar, you just need a lot of land to power something as large as a gigawatt-hour data center. That's a lot of solar. The ideal solution's probably going to be a combination of resources. It would be ideal to power everything with wind and solar, but there are land use issues.

There's the fact that you can't control the wind and the sun, so that opens up the door for other things like energy storage. Energy storage is a great solution to backup power but also balancing fluctuations from renewables. And then co-locating with things like existing nuclear sites or SMRs, that's always very interesting. I think the SMRs in particular. I don't know my opinion on them yet because not enough have been built, so the cost estimates for them are still noisy. And then the level of energy that they produce is still not quite enough for some of these large data centers. So, you have to couple it with something else, like the gas turbine or something.

The realistic answer is probably a combination of renewables, potentially nuclear, probably some kind of backup, like energy storage or maybe a backup gas turbine. And then something that can handle the peaks. So, a connection to the grid or something like a gas turbine. It is quite hard. You have to think about things practically. If you get 98% of the way there with clean energy, that might be half the price of getting 100% of the way there with clean energy. I'm not always adverse to non-renewable energy saying, “we can never have gas turbines.” Sometimes it's just going to happen and it's not necessarily the worst thing in the world.

CHRIS: I would pick up on that. I mean, Kyri’s absolutely right. The last 5%, the last 2%, the closer you get to 100%, the more expensive it gets. There's no doubt about that. That's an important part of this conversation. And so small amounts of backup fossil will be necessary in the near term for sure.

But the good news is wind, solar, and batteries keep falling in price. Geothermal, I think, could be a really great addition to the overall portfolio. And of course, that's zero emission. I started off in nuclear engineering. I'm a big fan of the technology. However, as Kyri said, there's a lot of noise in the cost estimates. We don't know what these reactors will cost. I think there's a lot of excitement around them. It would certainly be a great addition to the portfolio in a lot of places in the country, in the world. But right now, the cost estimates are extremely high. They're three or four times the cost of the alternatives in the market right now.

So, there are significant challenges there, and timescale issues with these nuclear reactors. Even if everything went perfectly from a technology standpoint and the cost came down, it is still 7, 8, 12, 15 years away from being able to get scale out of a new technology. And so we've got to be realistic about that as we meet this moment of what do we do in the next five years? What do we do in the next 10 years? And the constraints that we have to deal with, which is why I think I've been such a big proponent of coordinated grids, because we have curtailment right now. We have wind farms and solar farms that get shut off because they can't get the power exported. We have the need for more battery storage. We have the need for more transmission interconnection to bring different types of generators to get the power to the right place. And those inefficiencies will be easier to stamp out than counting on a new generation technology 10 or 15 years from now.

So let's go after the low-hanging fruit, so to speak. And that's why I think an RTO or ISO in the West is so critical for us to be able to meet this challenge.

AI to Improve Grid Efficiency Optimization

MAX: Kyri, we've talked a little bit about some of the potential solutions of siting AI data centers near clean energy. What about the potential for AI to improve grid efficiency optimization? To what extent could it help offset some of its own emissions?

KRYI: What we've been talking about so far has been really focused on LLMs, large language models, and that's a little pet peeve of mine. When people say AI, now they refer to LLMs. If you've ever used Excel and you fit a line to a scatter plot, that is technically AI. Anything where you take data and you learn a pattern from it is AI. AI is a very broad term, not to be confused with just LLMs. From that perspective, a lot of my research group's work has been focused on how we can use AI to lower emissions.

One concrete example is, inside these ISOs, these markets that clear supply and demand, they solve an optimization problem, and they solve one that is based on an approximation of the grid. It is quite challenging to model all of the physics of the grid. It takes too long, it's unreliable to solve, it's a very challenging math problem. So what we're doing is, we're telling power plants to act based on these approximations of the physics. If you use AI, you can learn a better approximation of the physics and operate assets in the grid more efficiently.

I'm just talking like a software upgrade inside the ISO, an approximation of the physics that is slightly better. That alone, in some of our estimates, could improve grid emissions by 5%, which doesn't seem like a lot, but if you literally just upgrade the software and it shaves off 5% of nationwide emissions, it's quite significant.

I think we generally need to be careful when we say AI when we actually mean LLMs, because AI can do a lot of other things: operate a combined cycle plant more efficiently when you have more data, estimate parameters of the power grid, learn where the locations of transformers are, learn when their lines will fail. AI can do a lot of different things that actually help grid assets too, and that's just collecting data and learning patterns.

CHRIS: I'll extend that because it's not only grid optimization, but also right now we have very large reserve margins in the system.

MAX: Can you explain to us what a reserve margin is?

CHRIS: Absolutely. So, you have to have a reserve margin in the grid you're operating in case a power plant fails or a line goes down, right? You've got extra capacity to immediately be able to meet demand.

One of the things that has happened as we have added more intermittent sources to the grid is that we've actually needed to increase our reserve margins in the near term because we have these small grids that don't communicate very well together. So, reserve margins in some cases can be 30% or more of extra generation you have to have on at all times to meet the current requirements of our operations. And I'm really excited to extend Kyri's point. AI can help us run that at a much tighter margin without losing any reliability. And that, in and of itself, could be another 5% or 10% of efficiency beyond the physical operation of the lines or the dispatch.

So yes, some really great opportunities to increase efficiency that then immediately have a huge impact on cost and emissions from the overall electric system. So, I love that point. AI can be harnessed here. All that capability can be harnessed to improve the system operation.

KRYI: One analogy for reserves I've thought of is insurance. When you buy a car, you get it insured just in case something goes wrong. You might buy a really, really expensive insurance plan where you're wasting money every year because you're not sure when the car will fail.

If you have better forecasts of when the parts are going to fail, you can buy a tighter insurance plan that makes you not spend as much money on it. AI can provide better forecasts, weather forecasts, failure forecasts, a bunch of things that give us more insight into the grid, so we're spending less money on these backup services.

MAX: Kyri, can you tell us--I know we can’t see the future of course--but to what extent do those improvements equal the increased demand from AI use? Are they close in scale? Are they within order of magnitude? Do you have any sense of what that looks like?

KYRI: No, not quite. The demand being added from AI data centers is quite large. I won't lie about that. But every little bit counts. A 0.1% cost savings nationwide is, I don’t know, billions or multimillions of dollars. So it means a lot, but it won’t offset the data center growth.

Planning to Meet Increased Demand

MAX: Chris, what kind of planning and coordination are companies like La Plata Energy considering to handle increased demand? We also had a question in the chat around a Western regional energy market or joining another RTO in our region or another region. Can you talk about some of those potential solutions?

CHRIS: For sure, I really love to get that question. La Plata Electric will be joining an RTO on April 1st of next year. We are excited to be in an RTO. We'll be joining SPP West, which is one of the main options we have here in the Western US. And we'll be joining alongside almost every other operator in the state of Colorado. The one exception is Xcel. They have decided to do a day-ahead market instead, the benefits of which are much, much smaller than joining the RTO.

The thing I worry about is if we have different approaches in the state of Colorado, we’ll get a lot of inefficiency, a lot higher reserve margins, and a lot higher emissions as a result. That was one of the reasons when I was in the legislature, I ran a piece of legislation, Senate Bill 72, that requires all of the operators in the state to be in an organized wholesale market or RTO at the end of the decade. So, that is a really important part of our strategy here at La Plata Electric. But it's also a huge part of the public policy strategy of the state of Colorado so that we can meet our goals, as required in state law.

To zero in on La Plata Electric, we will join the RTO next April. That will immediately give us access to lots of different, additional generation options and give us more flexibility as we put together our portfolio to decarbonize. That also helps us reduce costs because we're getting competitive bids from a larger region, more options, meaning more competition, and we see that translating into lower prices over time. So all of those are really great reasons that the RTO is part of the strategy here. And I think there's just no way the West broadly can cope with new data center demand if we don't have better coordination among our power grids.

Also, watching closely some of the legislation in California, they are looking at changing their governance model for CA-ISO, which was on the map that we looked at earlier. That's the other main RTO option in the West, is having California offer a multi-state version of that market. So we'll be watching that really closely. I'm feeling optimistic that they'll open up their governance model for a multi-state approach. And if they do that, now we've got multiple choices when it comes to organized markets. And that, in and of itself, helps to keep costs lower and provide more choice for electric customers and electric companies. So those are some of the big trends that I'm seeing and how I think they fit into this topic today of how we meet this challenge of significant new growth from AI.

Data Center Energy Consumption

MAX: Turning to some of the questions in the chat, we had one question around water use. Large data centers come with heavy water use due to cooling. I've also read that cooling is also a major energy cost as well. Certainly, in the West, we have major issues around water quantity and water quality. Kyri, can you comment a little bit on what the demand from data centers looks like for water, for cooling, and maybe what viable alternatives are out there? How do we address that issue?

KRYI: In terms of energy consumption, around 30% of the data center's energy consumption is just for cooling. These are hot processors that are running pretty consistently, so they're generating a lot of heat. In terms of water, I'm not as familiar with water resources and the use from data centers. I think there's a lot of new technologies coming out that are low water-cooling technologies. But I don't think I can speak intelligently to the water use aspect.

CHRIS: I'll just add, we've taken a look at this in southwest Colorado, where we are very water-constrained, as is almost the entire state. One of the things I find interesting that fits into our conversation today is that you can actually trade electricity for water and go to a dry cooling system. But, of course, that takes a lot more electricity to run the cooling equipment. The more water access you have, the less electricity you need. It’s a really fascinating nexus there. And there are open-loop and closed-loop systems. Where you have lots of water available, you use the open loop and evaporate; where you can't, you use closed loop, but that of course takes more electricity to run the closed loop. And then if that isn't available and you really are short of water, then you just run air cooling and use a lot of electricity to do the air cooling. So, really interesting trade-offs there that the developers are always trying to look at.

It's been interesting to me to see what's happened around the Phoenix area, where they have high ambient temperature and low water availability. Essentially, the data center developers are upping their electricity use to compensate. Whereas, if you do it in a more water-rich area with lower ambient temperatures, you get that trade-off. Lots of things go into that decision-making. But that's a really interesting one, I think, for this conversation we're having today.

KYRI: The nexus gets further complicated because a lot of the power plants themselves use a lot of water. Nuclear uses a ton of water. Coal plants use water, not just to cool the plant, but also these are steam turbines so they literally need water to make the plant generate electricity. So the electricity use increases, which then increases water use somewhere else in the system. So yes, water's a big consideration. It's something that will be important.

MAX: Kyri, staying with you for a moment, someone asks if 30% of the energy is for cooling, are there ways that heat energy can be captured and reused?

KYRI: Yes, I think there are a couple projects hypothesizing heat recovery from the processors and using it to heat domestic water, or co-locating it with buildings and using it for heating. I don't know the names of them off the top of my head, but people are definitely looking into that.

CHRIS: I am just back from an energy policy trip that I led to Denmark, and this is operational right now in Europe. The big data center developers--Meta, Google, etc.--are actually teaming up with district heating systems and heat pump systems to be able to use the waste heat for the domestic needs in the communities around them. That's a pretty interesting option. There are not a lot of district heating systems in the US. I wish we had more. We're exploring that in southwest Colorado as a way to be more efficient.

Of course, the heat pump technology has coefficients of performance that can be four, five, or six times what we're seeing from traditional systems. And if you add waste heat from a data center, then it gets even better. So, that's a really exciting option, but the Europeans are way ahead of us on that one.

MAX: We have another question here of concern that data center developers will promise to build a nuclear plant, but then end up switching back to coal and natural gas. Is there any talk about how to prevent what this person describes as a “bait and switch.” I think the plus side of that is a lot of the major developers have carbon-free energy goals for their data center portfolio. Maybe Chris can answer whether or not that's even possible.

CHRIS: I don't see anybody that wants to build a new coal plant. I don't see any of that on the horizon in the United States, so I'm not sure what that's referring to. If you're building a nuclear plant, I'm not really sure how that danger would come about. You're going to get permits for a nuclear plant, and then you build a nuclear plant. And once you do, you want to use that asset for 80-100 years. The Vogtle facility in Georgia is the newest nuclear construction in the US. Massive cost overruns, extremely expensive power. But now that it's built, we should run that thing for a hundred years, if it can be done safely, and of course, with testing and verification, but you want to use those assets for as long as you can once you have them built. It's a sunk cost problem when it comes to the financial model. And if you've got the nuclear facility built, you should run it as much as you can for as long as you can safely.

AI Use for Grid Planning

MAX: Kyri, we talked a little bit about using AI--not large language models--for grid optimization and integrating renewables and such. What about applying that technology to grid planning beyond just operations?

KYRI: One big challenge with planning is ideally you'd want to run millions and millions of hypothetical scenarios. What if this policy is implemented? What if this wind realization occurs? AI allows you to rapidly simulate things that you wouldn't be able to if you're running a model from scratch, like a high-fidelity model, because it can learn these patterns that are reduced approximations of this space.

Sorry, I'm trying not to be too jargony, but for planning you can simulate a wide range of future scenarios that you wouldn't be able to do otherwise. So it's really, really good for planning. I focused on optimization just because that's my main area of research, but there are a lot of people doing this for planning.

How Colorado Citizens Can Be Involved and Informed

MAX: How can Colorado citizens get involved if they care about these issues?

CHRIS: I love that question. Well, I'd say you can run for office. There's that option, which I made the plunge myself 10 years ago to do that. But there are so many ways to get directly involved with the legislative process, with local decision-making. These decisions are going to be made at city council. They're going to be made at the Board of County Commission. They're going to be made at state legislatures in a very important way over the next 10 years. The federal government has decided to go one direction with H.R.1 recently. But now state and local governments can really step up and lead on these issues and make sure that new development and new growth is done in a very environmentally responsible way.

So folks need to chime in. They need to show up at city council meetings. They need to show up at county commission meetings, come and testify at the legislature, work with citizen groups to advocate for legislation. And the other thing I would say is, get to know your state rep and your state senator. I often would joke with people when I was in the Senate that you pass bills in September and October. It's because that's when you have time to get to know your representatives, take them out for coffee, join them for lunch, let them know what's important to you. And take advantage of the off-session as it were.

It's much harder to do when we're in the middle of the legislative session sometimes to have as many meetings as we might want because of the legislative schedule. So do it ahead of time and advocate and make sure your voice is heard. In my experience, the Colorado legislature is super accessible. Just reach out, introduce yourself. I'd love to grab a coffee. And 99 out of 100 legislators will say yes to that request. So, that would be my big exhortation is to get to know your representatives and show up and make sure your voice is heard.

MAX: Kyri, anything to add to that from the perspective of a citizen or as a consumer?

KYRI: I think the only thing is educate yourself as much as possible because these issues are really, really complicated, and respect the nuance. When I entered into energy, I was like, “Oh yeah, we should have 100% wind and solar. This is stupid that we don't.”

And then as I learned about the infrastructure challenges, the cost challenges, the policy challenges, so many challenges, I'm able to have a more informed opinion that I feel more confident going forward representing, which is that there are always sacrifices to be made. So, like I was saying about if you get 95% of the way there, but it's half the cost, cost isn't always to be sacrificed at other expenses. That’s people's livelihood we're talking about. There are a lot of different objectives. So education.

How Colorado can Prepare for these Challenges

MAX: I had one final question that I was hoping both of you could take, and that is, what else should Colorado be doing right now to prepare for these challenges and also get ready to take advantage of these opportunities? Chris, you want to take that first?

CHRIS: This is fresh in my mind, just having met with a bunch of different policy makers over the last few weeks about what is needed for my main responsibility now, which is delivering electricity to the great folks in southwest Colorado.

I think top of that list is we've got to get serious about having the ability to get stuff built. And that means permitting new transmission lines, permitting new projects, working through the local land use issues that come up at city councils and at board boards of county commissions, that is really slowing us down.

And to give an example, La Plata County just put in a moratorium on new battery projects, and I, as the operator of this system really need more storage capability in Southwest Colorado. So we're working closely to get through any issues that have been brought up around safety or fire hazard. The new battery designs are a very, very low fire risk. Some of the old battery designs are not as great. 10 years ago, they weren't as safe. The new ones, these issues have really been dealt with by the companies and the engineers. And so I'm hoping we can get the train back on track and get these types of projects done.

But I use that as an example of how we really need to come together and work through these permitting issues. We've got the technology, we've got the willpower, we've got great state goals in place. Now it's time to get things built and get projects done. And, the regional grid is an example, a battery project in La Plata County is an example, and we need all of it to be successful here.

MAX: Great. Thank you, Kyri, what can the state of Colorado be doing to prepare itself for these challenges and also meet the opportunities that AI brings with it?

KYRI: One thing we haven't talked about yet is that Colorado is actually a pretty big entrepreneurial hub. Colorado University is one of the top, if not the top, public school for startups that have come out of our universities. There's a lot of low hanging fruit in this area because there are a lot of challenges that we haven't had to think about before the past decade. So, if you have the energy to do something yourself, you can impact policy or you can also impact industry through creating new solutions.

AI is unlocking a lot of that. The availability of data and computing power is bringing new solutions to the table for startup companies so there is something you can do for reducing wait times on transformers or improving line loading capacity estimation. There are so many things that you can do to help the grid in terms of that.

MAX: Fantastic. Well, Kyri Baker and Chris Hansen, thank you so much for joining us today. I certainly learned a lot. You can visit the Institute's website to learn more about our work, see other Science at the Edge webinars, and find out more about future webinars, both in this series and some of our other public programming as well. Kyri and Chris, thanks again for joining us. It was a pleasure speaking with you.