Feeding the Beast: Inside the High Stakes Race to Power AI

 What is the real cost of powering our future? Welcome to Watts the Point, the energy podcast exploring the trending issues in the electricity and energy industry globally. In this episode, we dive deep into the energy sector, from grid modernization and renewable energy to the massive power demands of AI data centers.  We are witnessing an unprecedented boom of AI today. The approval of the world's largest data center in Utah a few days ago certainly stands a testimony to this AI boom. While AI certainly has its benefits for the humanity, these can quickly be offset if the impact of AI on the four pillars - Reliability, Resilience, Affordability and the Environment is not managed well by the regulators and the utilities. Amongst these four, the minimizing the impact on reliability is the most important.

We at WATTS THE POINT (a newbie in the podcast world) are pleased to share that our first podcast "Feeding the Beast: Inside the High Stakes Race to Power AI" is now online.   Our host, Sunny Chhabra engages the two industry experts - Elliot Roseman, Managing Director, Strategic Energy Advisors and Viresh Shah, P.E., CISSP, CCSP, CEM, CMVP, PMP, LEED AP, Principal Advisor- Technology Consulting, Pioneer, LLC on the implications of AI boom for our energy security

Watts the Point cuts through the noise of the energy transition. Every month, we break down the complex intersection of energy tech, policy, geopolitics, and global commodity markets. We don't just report on utility bills and policy changes—we connect the dots on how the race for electricity impacts global business, corporate sustainability, and your daily life. 

Whether you are an industry professional, a clean energy investor, a data center, an utility or a regulator trying to understand the infrastructure bottlenecks of tomorrow, we always ask the ultimate question: What’s the point?

Hit follow to navigate the future of energy with us.

Transcript

SPEAKER_02 0:39

Okay. Hello and welcome everyone out there listening, and welcome to our first episode of What's the Point? My name is Sonny. I will be your host for this podcast, is brought to you by Pioneer LLC. I would first like to introduce this episode and topic we will be diving into today. And the episode is called Feeding the Beast Inside the High Stakes Race to Power AI. This is going to be a four-part episode series on AI, so more topics to follow coming up. As we dive into this topic, I would first like to introduce our guest today, Mr. Elliot Roseman, and one of our co-hosts from Pioneer, Mr. Vare Shah. Not featured here with us today is Dr. Shree, who will be joining us on the next episode. Gents, before we dive into the episode, please do a quick introduction of yourselves and give our audience a little bit more about your background, who you are, what you'd like them to know about you. Elliot, could we start with you?

SPEAKER_00 1:31

Sure. Thanks, Sonny. I appreciate the opportunity to participate in this podcast. I've had an extended career in the electric power industry and in energy more broadly. I've worked for a wide spectrum of the participants in the power sector, in particular, the utilities, the research the RTOs, the international organizations like the World Bank, USAID, the regulators, project developers. I've done that work uh internationally and domestically extensively. And it's really been about working on the cutting-edge issues, uh, of which uh, as your listeners may know, uh there have been many over the last four or so decades. Uh resource planning, queue management, project finance, a whole host of different new technologies, strategic planning. And so AI, artificial intelligence, and the data centers are the latest of those. Uh it's kind of the flavor of the month, if not the flavor of the decade. And we need to figure out how to deal with that. And uh I'm looking forward to the rest of the conversation.

SPEAKER_02 2:31

Thank you. Yeah, looking forward to it as well. Viresh, over to you.

SPEAKER_01 2:34

Thank you, Sunny. Hello, everyone. Viresh Shah here. I am in Northern Virginia, USA. Uh, have been in the industry for about 34 years, uh, doing a variety of energy focused activities. Um, first few years of my career, I spent uh about two and a half years, two and a half decades in the uh area of industrial control systems, power generation, um, and uh very, very high availability systems. Then I spent a few years in focusing on the Department of Defense activities for design of buildings, facilities, with a very, very strict focus also on energy efficiency and resilience. Um, I've had uh a pretty good career just like Elliott did, um, and I'm able to see that from a different vantage point of practicality, pragmatism, as well as affordability. Uh, and then the underlying scope of this conversation is uh pretty close to what we believe in. A good system, strong system, at the same time, uh one that has the ability to expand correctly. So over to you, Sunny.

SPEAKER_02 3:32

Oh, perfect. Thank you guys for the intros. Um, so shall we dive into the topic? Absolutely. Let's do it. All right, all right, let's do it. Elliot, let's start with you on the first question. Um, how are AI and data centers changing our energy future?

SPEAKER_00 3:49

Well, I could answer it uh rhetorically and say, how are they not? You know, but uh the answer is in many different ways. The first and the most obvious, I would say, is looking at the demand impact. I mean, AI and data centers are going to accelerate our demand like nothing we've seen uh over the last 40 years or so. I remember back in the 1970s and 80s when we had tremendous load growth, but we haven't seen anything like that in a while. So we need to kind of flex our muscles again on how to absorb that much new capacity or that much new demand at the same time. So that's, I guess, number one. Secondly, we need to recognize, I would say, that uh all data centers are not created equal. Some of them are going to be uh significant amounts of capacity, a gigawatt or more, coming online perhaps all at the same time, but others are going to be a bit intermittent in terms of their need for power or come online incrementally. So you need to differentiate and not just uh sort of think of data centers in a monolithic way. Uh third, we need to make sure that as we uh look at integrating data centers, as we should, as we've integrated sort of all other needs in the power sector, that we uh think about um the reliability questions. And that is certainly a risk. I'm gonna share a slide here, uh here, and uh let's just share my screen real quick. Okay. Come back over here, share screen. There we go. And share. Can you guys uh see that? You can see that. Great. Okay. So what this shows is this is a projection by the North American Electric Reliability Corporation, or NURC, which is the entity tasked uh under FERC jurisdiction with uh maintaining and and signaling the risks to reliability as we look out into the future. And it's pretty obvious just from the red that you can see they are projecting as of uh late last year that there are high risks in significant swaths of the country, including the region of the what's called the PJM, Pennsylvania, New Jersey, Maryland area, which is the largest uh electricity market in the country, if not the world, large parts of the Midwest ISO, ERCOT, um, and out of the Northwest. And most of this risk to reliability is driven by the data center phenomenon. Uh so what they're saying is red light flashing, guys, let's figure this out. We gotta not uh take this lightly. And and people aren't taking it lightly. I'm gonna stop sharing my screen here, but just wanted to share that. So, in terms of the impacts, uh, that is certainly one of them. The need to anticipate what's coming down the bike and make sure that it doesn't uh unduly affect reliability or other facets of the power sector that we're going to get into. Uh, also, there's an element of U.S. leadership. We do want, I believe, to accommodate the data centers. If they don't locate here, they might locate elsewhere. So we do want to find ways in which we can uh certainly maintain the leadership that the U.S. has uh in that regard. Um and lastly, I guess I would say we need to understand that um that this is on top of a number of other challenges that the power sector already faces, right? This is not happening in a vacuum. We have issues of resilience, maintaining uh against uh natural disasters. We have an aging system that needs to be invested in. We have issues of integrating other types of capacity of new generation that wants to get onto the grid, solar and wind in particular, and a whole host of new technologies. So let's uh we can't forget about all of those things. AI is the uh the latest item and it's accelerating the pace of change, and we need to manage it carefully. I'm confident we can do it, but um we need to manage that transition.

SPEAKER_02 7:45

Oh, thank you for lots of.

SPEAKER_01 7:49

The the the data center industry, the industry is driven by demand for computing, demand for fast computing, and demand for proximity. All of that is what leads to this location selection, site selection. Um and it happens in a sometimes in a clustered way, where the capacity demand that is incremental to what was already planned by the utility companies in that region, in that in that zone, uh was not prepared for these surges. And uh just looking at the designs themselves, there's two types. There's the hyperscaler type, and then there's co-location companies that are the companies that have ability to host a variety of data center solutions in it, could be multiple tenants in it. Um all said, these these uh data center clusters technically provide a challenge because it's an I wanted now kind of demand for water and cooling. Uh then and at the same time, the biggest biggest challenge is that the extra power needed to fund the activity, to run the activity, to energize the activity. On top of it, uh, if you take an average uh data center design compared to what was five years ago, the compute need per rack is 10 to 12 fold. For example, each rack used to be about 100 kilowatt, now it's 110 kilowatt. Um 10 kilowatt before, now it's 110 kilowatt. 120 kilowatt is the demand per rack. So you can see because of the density of the computing and the ability to uh not tolerate um any kind of loss of power, all of that leads to a power density that was unheard of. That is what leads to the actual magnitude of the demand. And the other thing is I would call um so density is another thing I would I would emphasize is what leads to the the scale of the challenge itself. And then um when when these data centers come in a community, they compete for finite pre-planned resources, which could be power capacity and water capacity. Um there are solutions that are being looked at in a mid-mid-cycle stage right now that will get into low water use, low uh low other cooling needs use, but it will take time, it's not fully mature yet. So I think we are coming across a situation where the demand for now, here and now, is already outpacing what's available. And that that leads to competition for or contention for finite resources. It's basically that supply-demand. Uh demand is outpacing the supply of out of it, and the demand is unfortunately and fortunately one that's probably going to lead us to the fourth industrial revolution per se, where the need to upgrade and transition is not decent, it's not um discretionary per se. Uh, one last thing I'll say is um the rationale of why the data center market, which is generically called data center, because it it accommodates storage in the cloud, it accommodates cloud solutions, which is all the IT services, which used to be mainframe, on-premise. So storage, IT solutions, and then there's um all the new AI applications, which is one that is generative AI, uh, the machine learning type. They all, and then there's a training loops in the AI itself. Those generate additional synthetic but multiple, multiple loops of computing. These things were not heard of a long time ago. Now it's here and now, and uh that's that's what is leading to the actual severity of the magnitude of the demand is all the co-hosting, everything. And then it comes with uh requirements that we will go over in a couple of minutes here, uh the resilience and the availability and the timing to market. So all these things are adding additional pressures. So it's going to be an interesting competition with the limited resources. Uh, so we'll examine these things as we go forward.

SPEAKER_02 12:00

Sounds good. Elliot, any other um any thoughts on this topic before we shift gears?

SPEAKER_00 12:05

You know, the only other thought I have, Sonny, is that um uh I'll I'll show a short chart in a minute on this, but um, the increasing uh demand for and the need to both strategize uh about how to accommodate it is is is is really firing on so many different burners at the same time. It requires an engineering and technical response. Sunny was addressing some of those questions right now, and it requires a regulatory, a policy, and a strategy type of response too. It requires a supply chain response. You know, I mean, there's just so many different elements that it's hitting on that um, you know, that's what I think makes it uh as challenging or as fascinating to be in the energy industry right now as it is.

SPEAKER_02 12:48

No, these are all good thoughts, right? So as we go a little deeper into it, you know, supply and demand clearly seems to be a thing we're hitting on. So, Varesh, let's jump over into question two and start with you. You know, um, with the supply and demand discussion we were just having, what what do you think has led to the acceleration of data centers and AI?

SPEAKER_01 13:09

I think it's uh tantamount to my first round of questions. Uh, it covers why things are converging into data centers. It's all these societal-driven demands for computing, for storage, for quick answers on the go run. Now it's going to also have quantum computing migrate into cloud. So that's gonna add additional on-site load in the cloud. Cloud is not a not a virtual activity, it's it's actual being performed in a data center, which is a congregation of all these and consolidation of all these activities. It has um it has two things hunger and haste. Hunger for power and water, cooling, typically for cooling, um, and then haste is this is what decides ability to meet market goals, growth goals for companies that are purely cloud providers, like Netflix, could be Amazon. For them, time to market is paramount. Um, and there are so many other business models that have come up that is based on AI, ability to do generative AI, agentic AI, as well as machine learning, right? Which which creates additional millions and millions of compute loops. Like if you search for something on Chat GPD, for example, it will lead to a few other loops churning things for you, researching things for you, finding answers for you for one simple question that you ask. All of that is hosted somewhere, and that is the exponential growth I'm referring to, which is what something we could not really estimate is coming, and the magnitude of it. Typically in uh power system planning, um, there is an outlook for growth. There's an outlook for where and how much, right? Um the the point at which the society is these days, uh, especially in the developed nations, is where the compute needs have grown exponentially. People are also getting hastier, a little bit more impatient. I want it now. I don't want slow latency, like high latency times, slow response times. Uh, it also means that are activities which leads to GDP issues directly because it's work gets done, something gets booked, something gets bought, something gets sold. You haven't even addressed the other aspect for what they call cryptocurrency or blockchain, which also will add additional, additional, additional incremental load to this whole uh circus, I would say. The need for immediate and fast uh compute capacity. So all of this is adding to a fact a set of factors that are undeniable. Number one, number two, it's almost like a one-way transition. Uh you will see this in even in developing nations that the mobile device usage, AI, uh from home computers uh usage is also gone up quite a bit. So it has to be hosted somewhere. So even they will start to see the spray as train on the resources. So I think that's uh some of those things that I would consider important to consider in um understanding why this happened and what the time pressures are, this hunger and haste part of it is is true. Um there's a lot of things that could be done about it, and we'll address that in future sessions. Uh so what do you see if Elliot has additional thoughts, Elliot?

SPEAKER_00 16:23

Uh yes, thanks. Um sure. Uh just to underscore some of the things that uh Varesh was just talking about, you know, I step back for a second and I think of uh how do we ever survive without data centers? Right? I mean, it's the last few years, all of a sudden, wow, you know, they just seem to be some that's what everybody is talking about when you go to the conferences. So, but clearly we're uh on the cusp, if not deeply into the need to accommodate them now. Uh as Raresh mentioned, uh, the growth of uh activities such as the large language models, the chat GPTs, the open AIs, the claws of the world, um, but also other interactions, other um, you know, just the vast increase in online meetings like this one of crypto, of social media, of transactions online. All of that is leading to more of an electrification of our society. I mean, we are becoming more electrified and using less of other energy sources. That has implications for what the electric system has to be prepared to respond to because reliability is paramount, and we can't afford to have a system, electric power system, that's not reliable. We'll talk more about that in a minute. But uh, as Varesh also mentioned, uh, these uh data centers are highly dense, voracious users of power land water, and they create a fair amount of noise if you're local. I'm gonna share one other chart here uh that uh indicates. Okay. That's the one I showed you a moment ago. I'm gonna share this other one here. Okay. I'm presuming you can see this. Uh if you look along the uh the very bottom here, you can see the number of megawatts we're talking about, number of gigawatts, actually, when you get up to like 10 gigawatts here, eight gigawatts, you can see down here on the left-hand uh side on the y-axis, you know, the uh the different participants, the Googles, the Meta, the Amazon, Microsoft, et cetera. This was as of uh the end of last year. This picture has certainly changed. And it's interesting, I think, um, actually informative to tell to see that a lot of those plans at those times were to have a lot of the capacity to power them coming from solar and wind. Uh Varesh talked about co-location also, and that shows up certainly in some of the uh the data centers shown here. Nuclear. Nuclear, I think, uh, is an option for the future. I don't think that the SMRs are certainly ready for prime time, but uh, you know, this gives you an idea that we are talking about a huge burst of activity and of demand that needs to be taken into account. I'm gonna stop sharing here for a second while I continue. Uh, this is um this is not so this is something we faced uh similar before. This is not completely unprecedented. We we have for a number of years had to deal with a question that I call Q management. And that refers to the integration of new power generation projects that have wanted to get access to the grid. I've been working a lot on this in the last five years or so. A lot of it was solar and wind. Uh, and the question was how do you prioritize and rank order and set the standards and criteria for which projects should actually get to connect so that we can bring the capacity online, but also not disrupt uh uh the reliability of the grid and impose undue costs on those that shouldn't be required to bear them. And so we have a number of lessons learned from the utilities, from the uh from the RTOs, uh, nationwide and even worldwide. I did a study on sort of worldwide uh best practices of queue management. And I think that uh that um you know, in advising uh the uh utilities and uh and the developers, frankly, of these centers, it behooves us to look at uh the lessons learned from uh the queue management work, that good work that's been going on for the last uh better part of a decade or so. The other thing I would point out is that uh we've experienced before what I call the land rush, you know, where developers of power projects of any type were looking for the sweet spot, looking for this the best locations to put their facilities, where there was access to water and gas pipelines or fiber optics or transmission capacity, et cetera, et cetera, where there was maybe hosting or headroom in the in the grid. Uh, and where they can get financing. You know, that's also um financing, though most of the AI companies don't have any problems with uh with capital availability. So we have dealt, my point is that we've dealt with these kinds of situations before. It needs to be applied freshly and and and tailored to artificial intelligence and to data centers, but we should we should use the lessons uh from those prior experiences to inform how we deal with data centers.

SPEAKER_01 21:14

Yeah, I'll add one more data point there. About 10 years ago, uh about 10 megawatt was considered, or 30 megawatt was considered like the norm of a data center. Today the norm is about 200 megawatts, seven times full, seven-fold increase. This is the average. And we are gonna we're going in a direction where it's one gigawatt is gonna be the norm sometime soon with the way things are clustering in their heading. So just the scale of each data center itself, the hunger for the power itself, is um it's itself pretty astonishing. So I think we need to just acknowledge that as the reality now.

SPEAKER_00 21:51

So go ahead, Elliot. Oh, one last thing that I'll just add here. I was gonna save this for later, but it seems appropriate now. Um the utilities and uh the data. Center developers are not unaware of a number of these issues, right? The uh you know, the epicenter of a lot of data center development has been Northern Virginia, an area that several of us are actually in the uh living in the area of. And, you know, Virginia uh electric power, Dominion, uh, has developed and has had numerous conversations with the developers, as they should. This has got to be an interactive, uh, it's got to be a conversation. You know, the utilities for years have had relationships with their largest customers. They assign representatives for their largest industrial customers. They need, and as Dominion does, uh, have representatives to work with the developers of data centers so that they can accommodate that. They need to think about, you know, what is going to be and work collaboratively to figure out where the best locations are, what the right rates are, who's gonna have to pay for what. And they have Dominion has something like that in place. Um, the PJM has got uh now uh a new policy, I think it was just uh announced earlier this week. It's called bring your own generation. Like we used to think about BYOB, right, to uh to a party, bring your own bottle. But this is uh very different. So if you want to hook up in the near term to the uh to the PJM grid, they're saying you've got to be responsible for most, if not all, of your own generation. And that you know, imposes a rigor on the developers as to where they can actually put their facilities. So we'll get into that uh supply chain a little bit more, but I thought I would raise those questions now.

SPEAKER_02 23:32

No, those are all good points. And thank you guys for thoughts on that. You know, part of what I heard you say and underscoring a number of things was again, we we kicked off the question around acceleration, right? Voresha, just heard you talk about you know, 7x the demand over the past decade, and it's only certainly going to increase from here. Um, and you know, we had talked about this a little offline before we started the recording, but one of the things that I'm thinking through, you know, is the defense sector side of it, right? Like the the potential of again, us trying to stay ahead and be the world leader, not only in this space. I think it was mentioned that you know, Virginia certainly is like one of the largest data center clusters in the world. Um, and you know, the AI arms race with the defense sector, you know, us trying to outpace our peers in other countries, I can only see that the demand's gonna increase even further and further and faster and faster, especially as defense spending. If you look at the 27 projections for what's coming, you know, from a financial standpoint and where defense is pointing their budget, certainly going to be pouring a lot of money into these types of topics. So uh it'll be interesting to see as this unfolds over the next 18 to 24 months for sure. But you know, back to you guys if you have any thoughts on that before we jump into the next.

unknown 24:45

Okay.

SPEAKER_01 24:45

Uh so I think it's it's important to kind of see what what other aspects of data center market we're going to talk about. So let's go over the next next kind of evaluation.

SPEAKER_02 24:55

Yeah, for sure. So, uh Elliot, let's kick off the next topic with you. You know, one of the things we had talked about was uh impact on the four pillars, right? Reliability, resilience, affordability, and emissions. Um, let's let's talk about that a little bit from your perspective.

SPEAKER_00 25:11

Yeah, sure. Um, you know, you just laid them out there, Sonny. You know, those of uh uh are kind of the uh the high priests of uh the power system, you know, we need to certainly, and I I would say that of the those uh royalty that the uh that reliability is king. Uh we have to keep the lights on. We have come in this country to rely on when I turn the light, when I flip the switch, that that light above me is going to go on almost all the time. Uh we can't have a hundred percent reliability as we know. Sometimes there are conditions that are so um unanticipatable or would be so costly to accommodate that we don't try to plan for every possible contingency. But 99.99, you know, add your favorite number of nines amount of time, we expect the lights to stay on. And that means the reliability of the power system to be there for all users, not just for homes, but for businesses and for data centers and for hospitals and for, et cetera, et cetera. So we don't want, I'll put this in the negative, we don't want the data center phenomena to have a spillover effect on the reliability for other customers, including what has become common throughout the world in power systems, a reserve margin. We have to have some amount of uh in reserve in case something untoward happens. So, what that implies is that to maintain that level of reliability that we've come to expect, and that our economy needs, because we we said, as I mentioned, that we're getting more electrification over time, that we either need to build out the generation and the grid. The grid is a fundamental element of all of this, or we need to figure out how to pace the level of data centers that come online. And to do that, we need to uh know which ones are real, which projects actually are going to come to fruition as opposed to those that are just proposed. There are a whole set of techniques from Q management that I mentioned a few minutes ago, where there are indicators, there are payments, there are standards and criteria such as a deposit, such as having the land, such as having the permits, such as having the equipment that you can demonstrate that your project is actually going to happen as opposed to ones that are just um hoped for or that are not quite ready for prime time. So, you know, the reliability question depends on what is actually going to come to fruition. You know, the um in Delaware, they have uh recently uh uh put in place a requirement that the data centers, if they are going to locate or propose to locate in a certain area, they need to pay for the upgrades to the grid that would be required for the next decade. And whether or not the data center uh actually comes online, whether or not it it goes, um it varies, or it might who knows 10 years from now, whether that same data center is going to be there, or well, they have moved on. But that kind of that's a commitment that they are looking for for the data centers to make to demonstrate that they are actually going to make it. And that causes, in turn, the developers to think about hmm, is this one that we want to place our chips on, or is this one that we um want to maybe locate somewhere else? And that's good. We have to have an orderly way of figuring that out. Uh, in addition, you know, the reliability is affected by by how much uh these data centers actually need. We're already looking at second generation data centers that are a lot more efficient than the first ones. They can use 20 or maybe 30% less power per process, you know, and that's going to actually mitigate, soften the effect of reliability. The um the new generation options, you know, uh are actually um are actually limited in terms of how much can actually be available immediately or in the next couple of years. You know, there's uh these data centers have a lot of rooftop, right? You can probably put solar on them. They may be able to put some um diesel generators nearby, which is dirty and noisy, but you can have them. There's uh combustion turbines if you can get them, but there's a long backlog, long backlog for uh transformers also for connecting to the grid. So there's a reality check, I guess is what I'm saying. But the um the underlying uh fundamentals is that um that is that we need in uh in collaboration between the developers and the utilities and the regulators, we need to be guardians of reliability, right? We need to make sure that what we're doing doesn't affect uh the overall reliability. And I'll just roll into a brief mention of resilience, because um resilience is kind of the cousin of reliability. Uh it's the ability to protect against disasters, against attacks, against things that are you know uh quick, usually one-time type of events. And the need for that has been increasing substantially if you've followed anything that's going on with regard to wildfires in California or other parts of the country or uh disasters, hurricanes, floods, you know, storms, et cetera, that take place. So utilities are already thinking about how to accommodate or harden the system, uh, which has costs associated with it, so that you can minimize the outage. Outages are very, very costly, not just to the system, but to our society. There's a huge value of electric power. It's often called, uh sometimes the term you hear is the value of lost load, vol. And it's it's it's in the billions, even with minutes or hours of outages, now that we are so becoming so electrified here. So we need to try to mitigate and minimize that impact. And that includes, as we plan for data centers, seeing that the resilience um is not unduly affected. That the data centers that come online don't make the system more brittle uh and less able to accommodate those. Again, I think we can do it. It will just take some uh careful planning.

SPEAKER_01 31:24

And I guess I'll touch on the two other two aspects of the question you asked, Sonny. Uh, one is about affordability, and the other one is about emissions, right? So on the affordability, like there is no free lunch, as we all know. Somebody picks up the tab for the expansion of the transmission distribution grid. I believe there's enough capacity in the system. I'm talking about actual units of energy available as the power generation side. It's the ability to transport it and deliver it. That's where the choke points are, that's where the bottlenecks are, that's where the the laggardness is, the where it's lagging behind. At the same time, it's that's where not it's not one of those glamorous investments. It's something that is hard to do, it's not very attractive. Um, so sometimes we don't have the ability to kind of plan for the ability to maximize transmission distribution for these upcoming loads, is probably because these scenarios for AI growth and the concurrent demand surges like this uh was not anticipated. And then on top of it is also a bit more stochastic in terms of even able to predict where the computing load is going to be highest. If you just use weather, you can estimate whether economic activity, production activities, manufacturing activities, there are ways to estimate the growth demand in different pockets of the nation. But as far as AI demand and things that are driving the exponential growth, those are those things are very, very hard to model. So you can't prepare for it if you can't model. You can approximate, but you can't model. On top of that, um, I think the other aspect is the traditional growth of T and D, transmission and distribution, is also one where it's highly supply chain driven at the same time. Somebody has to pay for it. It usually ends up being the right payer, the surrounding communities that might see feel the pinch. Just like competition for finite resources, for water, same thing, gas, same thing. Electricity is nothing different. Um, so the affordability outcomes, affordability second-order effects could be uh uncomfortable to the surrounding communities. That's why sometimes you hear in news articles that they're talking about not here, not in my backyard kind of approach, right? Um so I would I would just kind of cover that from an affordability, those aspects. And then on the uh the emissions aspect, I I can think of two different ways. One is just a pure um, if it's a on-site generation, either through combustion turbines or through reciprocating engine-driven generation, which is the most common ones for independent power production on site. The third type is the fuel cell base, which has low PTU, low, low methane. Um they can use hydrogen, natural gas to burn, release uh low carbon emission, low NOx emissions. But the other two that I mentioned, combustion turbine as well as the reciprocating engine-driven ones, are the most conventional on-site generation sources. All of them are not carbon neutral, right? So there's always NOx, CO2, uh, and carbon monoxide that also comes out. Those are all going to be affecting the air quality in the surrounding areas. So, as you we have seen in refineries and petrochemical plants, in those surrounding areas, people have um health effects that are considered as a result of. So, I would consider that to be another little factor. If the data center industry is getting impatient to choose to have to do on-site generation, they have to choose these sources of on-site generation. If they cannot wait for the grid connection, this is what they will start with, and that's what's going to likely happen. So we'll see some emission-related concerns come out because of the insistence of doing it on-site, number one. The other slightly mitigating factor is the noise emission, like Elliot mentioned. Uh, some of these systems do make noise, and there's a humming sound, just like some folks have issues with large wind turbines, there's always this ongoing humming sound, noise, it's uh generators kicking on and off and things like that. Um, these two aspects which would consider, I would consider they are more ergonomic to the surrounding communities, and that may be directly connected to either noise uh or air emissions or exhaust emissions that are a result of the activities. So I think we have a multifold uh areas of coverage to address. It's not one formula that fits all. So it's all these are very important topics that we'll we'll be facing.

SPEAKER_02 36:00

So yeah, for sure.

SPEAKER_00 36:03

Uh jump in if it's okay, Sunny, uh to again underscore some of the things that uh that Varesh was just talking about, particularly on the affordability side. I mean, we're in a world right now where the cost of electricity is going up. I don't know how to say it any more plainly. We just have a lot of uh costs that are required to maintain the system, to improve the system, to accommodate all this new generation data centers. Uh, we're becoming, uh, as we've mentioned a few times, um, we're going to greater electrification. All of that costs money. And we uh and the regulators for decades have been deciding how to take the costs, the reasonable costs, which they also uh judge and evaluate, of the utilities and determine how to pass those on through rates to customers, residential, commercial, industrial, data centers, um, military, et cetera. Uh there are most of us pay on the basis of how much we use per kilowatt hour in homes, uh, because we're not using all of that much power on the scale of the grid. But the large users are paying a capacity charge as well, an amount per kilowatt or megawatt, as well as a charge per use for megawatt hour, kilowatt hour. So that whole rate design is under pressure as a result of this. Um, I mentioned uh Dominion and um Delaware, Delmarva, you know, with some of their own rates now, data center rates and data center uh requirements. This is uh a process that is uh sometimes messy, uh contentious, because no one wants to pay more than their fair share. But the data centers, to their credit, at least all those I've talked to, uh indicate they're willing to pay their fair share. They're not trying to uh hide the fact that there are additional costs uh imposed on the system as a result, and they're willing to shoulder them. So that's good. Um, but uh we should not be surprised. Um you might not be comfortable all the time, but we're but you know, you can draw some comfort if this is comforting to you, that by paying more for electricity and the reliability thereof uh and the resilience thereof, that we're probably paying less for some other forms of energy. Right? We're paying less maybe for for natural gas if we have uh if we have um uh electric heating or if we have electric water heating. Um if you have an electric vehicle, uh, then obviously you're paying less for gasoline or nothing for gasoline for that vehicle. We're going much more to batteries. So, you know, those um all I think are gonna help to mitigate maybe the costs of other fuels, um, but in general, uh the cost is is going up. Uh, and that's because we're getting more benefits from the electric power, which uh which we do. So uh it's um it is something that we're gonna have to accommodate. The regulators, I think, are really uh on the hot seat right now. They're the ones who sort of stand between the customers and the uh consumers and the and the and the demand side. And they're the ones that um are properly going to be feeling you know some heat. That's their job, you know, to try to figure out how to divide up this pie uh from a uh cost perspective.

SPEAKER_02 39:22

No, it's it's good to hear that, you know, from an affordability standpoint, that the data centers seem like they're willing to shoulder their fair share of it, which I think is important. But you know, back to a larger point you made around electrification, right? Um, you know, and and more and more things. We didn't really underscore things like EVs and other, but you know, we had also talked about this prior to the recording of other additional load from the electrical perspective. And we had mentioned, you know, uh tankless water heaters, things like that, right? More things that are just gonna factor into, again, a larger demand on the electric grid, and something you had touched on previously, Elliot, of having the load and then also having that margin for what's the additional beyond that, right? That's got to get factored in. And it seems like a really tricky exercise at this point because we can't really scope and build good planning factors for even the past, you know, if if we look back two to three years to look at the exponential growth in AI and data centers, it would be hard to model back to what Varish is saying. It'll be really interesting to see as more and more push goes in this direction over the next two to three years. What does that look like? You know, is it uh how do you model out something like that? But it's it's uh again, it's it's a challenging exercise, certainly. Uh I would hate to be in the regulator's shoes at this stage of it, to be completely honest. You know, I'm certainly thinking about that, because I think we would all agree that the the need and the demand happens first. You know, the the fact of how are we going to meet the need, the intent, the capacity, the regulator stuff's multiple steps behind now trying to catch up, typically. I would I would say. Would you guys agree with that? So, you know, it's it's one of those where it's okay, you're trying to solve what was probably a year ago problem, and now being asked to run even faster. You know, trying to keep up with that seems to be a very difficult exercise for those folks as well, I would imagine.

SPEAKER_01 41:22

Yeah, so I would say there's a clear timeline expectation mismatch or timeline horizon mismatch. Typical data center wouldn't go active in 18 to 24 months from the time the first paid goes in the ground to the time that the compute power is delivered. That's 18 to 24 months to revenue. That's their expectation. The average power infrastructure growth or the ability to provide a proportional amount of power, that the planning cycles are three to five years for growth, for being able to deliver that. So there's a clear mismatch right there, right? That's one part. Um I would also compare the same thing with the demand for water in cool in cooling data centers, you know, even for the building, the white space areas right, even if you go all liquid for the the co the compute aisles, you still have a lot of need for cooling, right? And um average uh data centers each each simple data center can take about it can take the water use of 30,000 to 50,000 homes. That is the substitution it takes away. That's the quantum it removes from the rest of the supply. So 30 to 50,000 homes is that capacity of water is diverted. Obviously, somebody has to plan for it, especially in drought hot drought-head areas or water-strained zones, right? Um, and with some of those activity, some of the climate phenomena happening and a little bit more severity as well, that is even further contested naturally speaking, right? So uh all in all, it's a very intense competition for resources, essentially. Um, and also the timeline mismatch, like like I mentioned, the horizons are completely vastly different, and that's why some sort some data center companies are choosing to grow their own, do their own IPP, independent power production on site. Like, and then similar to what um Elliot mentioned, uh that is driven by ability for the equipment to come to them on time. Um, there's only so many X number of gas turbine vendors in the world. Everybody's order books are full. Think of so it's kind of comparable to what Boeing goes through. Somebody orders a jet now and gets it after two and a half, three years. It's a very healthy backlog for them, but it's a lot of wait time for a waiting customer, right? So it's a problem that even from the ability to stand it up that fast, that is becoming a hard to solve issue. So it's kind of an interesting combination of factors at play here. Elliot, you have any thoughts?

SPEAKER_00 43:56

Yeah, yeah, no, absolutely, Varashi, right on target. I would uh add a couple of points. Um, the heating demand, the the heating heat that's produced, uh, which leads to the requirement for cooling, which is a lot of the water use in these facilities. The data center developers are also beginning to think creatively, more creatively about that. I mean, that heat can be used to um uh to create to to produce more electricity, right? You can siphon off that heat and use it to meet some of your on-site demand. So, you know, that's also uh, I think a way that uh they're looking at adapting. The whole system really has to adapt to this the speed. Uh, traditionally, the regulators, I'll return to them here for just a second, the regulators have um have been able to receive from the utilities the uh the costs that they have expended, you know, for putting in place new facilities and approve or not in some cases, or or maybe reduce the amount that the utilities could collect, generally approved if those expenses were were approved in advance. But um the luxury of being able to wait until the money is spent and then have the rates adjust uh is slipping away, if not gone. The regulators and the utilities are going to need to be now looking proactively at what the expenditure is going to be in the next year or two, because the regulatory pro you don't have time as much time for the regulatory process to work its way through. Long hearings, these rate proceedings can sometimes last a year or two. And by that time, you know, the costs that you might have been required to absorb to build out the system could be uh enormous. So it's again sort of putting more pressure on the on the regulatory processes. Um uh and that's um something that uh needs to to um to flex, I guess, and and when we're bringing on so much new capacity as quickly as we're going to be.

SPEAKER_02 45:53

Gosh, just so much to unpack on some of this as we go, right? Um well let's let's do this. I think we're at a natural breakpoint on this topic for now. Um let's uh how do we feel about heading over to question four? Yeah. Okay, all right. Very let's start off that last question of the day with you. Um, you know, we've touched on planning already a bit, so let's talk about planning aspects of managing the tsunami of what's coming next.

SPEAKER_01 46:20

Okay, yeah, just like it's a multi multi-factor, multi-stressor situation for what's at risk and what's being demanded. Everything needs its own addressal process, right? So power availability, power supply, power delivery, that is one aspect that needs serious luck, right? Um, same thing goes with the water, is the other natural, naturally occurring resource that's needed. Uh, whether it's even if you go all liquid for for rack cooling, you still have area cooling and uh other processes that required lots of water. So the water is the other piece that requires serious planning, serious allocation, serious estimation. Right? And then there is um there's three, four other things. One is I would consider the ability to project the model the demand, growth. What is it now? Why is it now this way? What is it like to look for, look like in a couple of years, five years or seven years from now, that will decide ability to pre-plan some of this capacity uh constraints that we have today. Then there is this other human aspect, workforce for that area, uh, mechanics, electricians, I people who can do data center work, design, maintenance. I'm not sure if we are all fully ready to have the right amount of workforce ready and certified, ready to go to work. It might need a lot of retraining, a lot of incentive making through maybe Department of Labor. There's a there's a workforce piece as well, just like cybersecurity. I had a big big demand for people that are trained, cyber aware. It's gonna need uh people that do this kind of work. And then um uh I would say uh I I consider this to be also a good part of the equation, is this could be a good way to get rid of the old old approaches. So you optimize designs, you optimize your density of computing, the spaces. You can even go further into making like the heat reuse, like Elliot mentioned. Find ways to optimize if you just go a normal way and not think about optimizing your design approaches, make it smaller, make it more efficient, make it more packed. I would consider the data center design firms will be also uh at stake to complete that task, to fill you know, hit that challenge so that the demand itself um can be managed, not grow, not grow like a machine, right? Um so I think there's a lot of these areas of planning and estimating and projecting and modeling. Uh that's where a lot of the effort needs to go. So oh, Elliot?

SPEAKER_00 48:59

Yeah, yeah. Uh planning is uh it's uh the coin of the realm, I think, at this point. Um it really involves, when I think of planning, involves trying to look over the horizon. And we're gonna have to look over the near-term horizon. We're gonna have to look over the horizon that we is just uh you know out into the next decade as well as we do this. The electric power industry is a long-term planning industry. We're putting in place facilities. They're gonna be there for decades. So we need to do it smartly today so that it'll be resilient and reliable for decades to come. So that planning uh is gonna need to take place at the utilities, certainly, the grid planners, the generation planners, um, more sophisticated modeling. They, the utilities are very sophisticated at doing a lot of this scenario analysis, and they're gonna need to be on top of their game as the data centers come online and raise loads at what are non-traditional peak times, right? Raise them significantly. So there's gonna be sort of less margin unless they plan for it at those times as well. You know, you can't only worry about any more just the peak time, you have to look at a lot of the off-peak times as well. So planning at the utilities, uh, very, very critical. The distribution companies in particular, uh, where a lot of these facilities are located, are gonna need new switching stations and transformers if they can get them and circuits and system hardening and the margins that I just talked about. So uh, you know, the distribution versus the transmission side of things. The hyperscalers themselves, they need to be uh in communication with and honest with themselves and with communicating with the with utilities and the uh and the regulators also about where they're gonna go next and the where they're actually gonna come online, as opposed to where you know they might just be reserving sites for for future use. Um the customers, uh no active planning, just kind of a preparation. You know, you you mentioned in introducing this question a tsunami, um Sonny. Um I don't think that customers are gonna drown, um, but they might get their feet wet because there um there could be some modest impacts, hopefully not significant, but on reliability, on rates, and on the environment, as Sonny was uh as Faresh was talking about. Um the RTOs, the uh NERC, you know, all this takes a fair amount of planning. And then to return to our favorite um topic, uh it seems, or one of our favorite uh uh participants, the regulators, they're gonna be in the position of having to review and approve rates, uh, review resource plans, um, approve the costs of the build-out that's taking place, uh, and to figure out how to divide up this uh significant increase in the uh amount of costs and avoid undue stress. We need to undertake this so that we are, you know, looking back, you know, five years from now, we're able to say that we maintained America's confidence in the reliability of the power supply, that we've limited the amount of power costs that uh increased, though costs are likely to rise, that we were able to facilitate uh and meet AI's power needs because it's a key industry, and we enabled the U.S. to stay ahead in the computing arms race. Um it's uh planning is a fundamental part of all of that. It's the biggest challenge we pay face in the power sector right now. Uh and uh I am confident that we'll meet uh our goals, but it's going to take a lot of uh paying attention to make sure and and active diligence to make sure that we do.

SPEAKER_02 52:39

Yeah, lots, again, to to talk about in all these areas. You know, I think with that, it's um it's a good thing we've decided to make this a multi-part series, right? Because I feel like we're just scratching the surface on some of this for sure. Okay. Um, gents, you know, we've been going for close to an hour at this point, right? Uh we certainly appreciate your time and for joining us today. As we bring this discussion for today to a close, you know, what's one final thought from each of you you'd like to share with the audience? Whether that's something you've already said, you wanted to recap, or just one final thing you want to share for now. Elliot, could we start with you on that?

SPEAKER_00 53:15

Sure, sure. So what I would say, Sonny, uh first of all, I would say thank you for the opportunity to participate uh in this podcast. Uh it's an honor and a pleasure to be part of the debate. You know, the uh I've seen lots of issues come and go uh in the over the decades, uh, and this is certainly a vital and critical one that is uh is on our plate, right? And we uh we need to digest it. So, as with many things in life, um, especially with the power sector, planning. Planning is powerful. And uh all the different uh participants uh that are part of this are gonna need to step up to that plate that I just referred to and to be able to uh figure out how to up their game and accommodate this wave, whatever you want to call it, tsunami, of new demand and and new uh supply chain issues. So planning is powerful, and I would uh close with just saying that reliability. Uh one of the most three most important things in the electric power sector: reliability, reliability, and reliability. After that, yes, we do have the other pillars that we talked about, but keeping the lights on, if you don't do that, uh that's you need to do that, that's job one.

SPEAKER_01 54:26

Sure. And I think uh planning is certainly important, and I would consider this to be a case of classic change management. So if you can draw a direct line of benefit from what's needed to what it'll mean for everybody. So if let's say compute capacity growth leads to better outcome economically, more GDP, more transactions, more stuff getting done faster, i.e., productivity. This the if you can map cross cross-map those benefits, uh it also means there's also uh an incentive for creating in um innovation, better solutions, more innovative approaches, or utilization of spare capacity that might be sitting around. How do you get them from here to here? Um, and then third part will be creating models to uh show benefit of planning uh the uh the model of concurrent needs and how they can be met. Pretty much like how AI figures out something, right? If you give multiple factors, say, give me the best optimal solution, AI finds a solution for you, typically, right? Give us given uh best uh inputs available, can give you a recommended course of action. Probably we can put AI to work to also find better ways to do this and connect the benefits uh to what it means. Um so I would say that's those are the areas that inspire me. It's not in the technical area, but I think they are considered essential uh in parallel.

SPEAKER_00 55:48

So one last thing, if I might jump in. You know, I said reliability is king uh and um made a strong point of that just a moment ago. You may have customers, I think they're well, there are customers who are willing to accept a lower level of reliability for accommodation on rates. And figuring out who those customers are, whether they're data centers, whether they're industrial customers, whether they're some homeowners, we now have rates that are in place where if someone is willing to have their air conditioning cycle, you know, and have it not be on for 15 minutes an hour and during even during the summer, that they get a break on the rates, and that gives the utility a chance to take a little bit off the demand, you know, as well. So that kind of flexibility on the customer side and the offering uh uh you know rate flexibility, I'm sorry, reliability flexibility, I think might be something we're looking at in the future as well.

SPEAKER_01 56:38

And suddenly this is also a competitive advantage for a country. If you if you really shape it right, it can be a game changer too. So could if you tie this to a national strategy, it's it's doable. Yeah, it's not mandatory, but it's obviously very competitive.

SPEAKER_02 56:52

Yeah, no, those are those are all good thoughts, right? And um, as we again bring this to a close, we're certainly going to tease the next episode, but as we've discussed briefly about pep episode three and four, sticking with this same topic and just going in with it, you know, evaluating mitigations or broader enabling factors and things that could be done about it, these are certainly things we should touch on more at those points. Um so good thoughts. Thanks, gents. Um, okay. Well, thank you to everyone out there that's listening and joining and following along. You know, please drop us a like, um, share with others. If you'd leave us a comment or a question, we'll certainly follow up with you. For our next episode, we're gonna touch on Power Hungry, the hidden toll of AI. And with that, um, I'm gonna share our quick video as the segue here, and we'll bring this to a close.