Moody's Talks - Inside Economics - The AI Series: The Data Center Buildout

Episode Date: September 1, 2026

Continuing the AI series, the Inside Economics team talks with ConstructConnect Chief Economist Michael Guckes about the data center boom. Michael’s granular view of the buildout sheds light on both the opportunities and risks, including whether financing has outpaced construction and whether cheaper AI models could leave some premium data centers stranded. And as a bonus, he has some thoughts on construction in Area 51. Guest: Michael Buckes, Chief Economist, ConstructConnect View our latest articles and research on AI- https://www.economy.com/ai-insight-hub Hosts: Mark Zandi – Chief Economist, Moody’s Analytics, Cris deRitis – Deputy Chief Economist, Moody’s Analytics, and Marisa DiNatale – Senior Director - Head of Global Forecasting, Moody’s Analytics Follow Mark Zandi on 'X' and BlueSky @MarkZandi, Cris deRitis on LinkedIn, and Marisa DiNatale on LinkedIn Questions or Comments, please email us at [e-mail]. We would love to hear from you. To stay informed and follow the insights of Moody's Analytics economists, visit Economic View. Hosted by Simplecast, an AdsWizz company. See pcm.adswizz.com for information about our collection and use of personal data for advertising.

Welcome to Inside Economics. I'm Mark Sandy, the chief economist of Moody's Analytics, and I'm joined by my two trusty co-host, Marissa Dina Talley, Chris DREDIES. Hi, guys. Hey, Mark. Hi, again. Hey, again. Yeah, this is daily.

We're meeting up here at this Inside Economics podcast. We recorded a podcast yesterday. Today's Friday, August 21st, but we had a podcast yesterday that will be released today. We've had a couple of podcasts around our AI series, and we've got a, another guest to talk about a different aspect of artificial intelligence in the construction markets more broadly, and that's Michael Gukas. Good, welcome, Michael. Good to see you. Thank you for having me.

Did I get your last name right? Did I pronounce that incorrectly? Gukis is exactly it. And is that a Greek name, Greek heritage, or where's that from? Yeah, it's Eastern European, actually. Eastern European. It's a mishmash. It's broad eastern European.

So there could be some Greek in it. Who knows? It sounds Greek to me. Isn't that if people say that, it's Greek to me? Yes. When they don't understand. They don't understand. It's Greek to me.

Yeah, that's good. And you're the chief economist of Construct Connect. Correct. Right. So Construct Connect is a software company, but we specialize in pre-construction software. And our goal is to help every project start on a strong foundation. So it's neat.

It's a company where our real bread and butter, our real core value ad is that we capture hundreds of thousands of active projects in some phase of pre-construction. And then we're able to structure that data and then package it up and offer it to firms in the construction space and in the building products manufacturing space so that they don't have to do that work, right? They can keep their pipelines full. They can keep their revenues topped up through the tools that we support them with. So you're able to collect all this data on what's going on in the construction markets.

And you do that because folks in the industry are using your software to do their work. Is that kind of roughly right? Right. So it's all kinds of different ways that you can. There are different ways to use it. Right. So if you're a general contractor or a trade contractor, it's really easy to find projects that would suit your company's best offerings, right? So if you're an electrical contractor or plumber or whatever, you can use it to find the electrical and the plumbing projects that either you could serve directly or you could try to become a subcontractor to a larger project, you know, by filtering through all the projects. again, hundreds of thousands of projects that we have actively curated and we, you know,

actively monitor at any given time. So that's the goal. And if you're a BPM, a building products manufacturer, right, it helps you to level said no, how much demand might there be for your product in the near term or even in the more distant future. And then also, of course, we do, I like to do this a lot. I cross my hands like a grid because it's all about your subcategories, things like, you know, fire and police stations, you know, office buildings, hospitals, et cetera, et cetera. And then you have to marry that up with your geographies, right? And you want to pick the right grid squares that are going to be the most profitable for your

business. And I think that's really where we can come into play and be a huge value ad for the construction industry. So how did you find your way to construct connect? How did you, what's your path? Started off with me loving Legos as a kid, right? I've always loved building and manufacturing. manufacturing things. So, yeah, it starts with Legos, like with so many of us. But I actually

started in. Really? So many of us? Marissa, did it start with Legos with you? Oh, sure. My path to economics didn't start with Legos, but I, but I love Legos. I still love Legos. Yeah, I do it with my nephew all the time. Chris, what about you? Did, were Legos in your path to becoming an economist? Absolutely. Absolutely. Yes, yes. Before the kits, right? Yeah. Standard Legos that, you know, a little bit more creativity. But, yeah, I don't know.

I haven't looked at modern day Legos recently. Have you? They don't look like, they look like models or, you know. Oh, Michael, are you, do you still play with Legos, Michael? My kids, my kids do. So I have a teenage daughter and I have a tween-aged son. And, of course, they look at Legos very differently. The daughter, she loves the new plant ones.

Have you seen these botanical Legos? And they're beautiful. They're these intricate, you know, roses and other flower types. I couldn't name any of them. But they're gorgeous. And she loves them and she uses them to decorate her room because they're that nice looking. And then, of course, my son, he likes anything that has wheels and an engine and, you know, the classic stuff that all of us probably remember from way back. Right.

So it all started with Legos. And then what happened? Yeah. Yeah. So for me, it was applying to. many, many, many jobs right out of college and getting nowhere with that. I was able to finally get a job with the state of Ohio working as one of the very first hired economists in the Department of Transportation. And I was there for quite a long time. I got my master's degree

while I was there. And then moved through banking and insurance. And that was when I first got to interact with Moody's was way back in the early 2010s as part of C-Car. So very memorable times. C-Car being, no one knows what C-Car, but you and I know. Chris obviously knows, Brisson knows, but C-Car is the stress testing process, the capital planning that banks have to do that came out, born out of the financial crisis. And that was a pretty hard, banks complained endlessly, I'm sure. And I'm sure you did as well about this process, but we helped banks with that, and that's how we got to know each other.

Yeah, it was important work, but not by any stretch of the imagination, the most fun part of my career for sure. But then I got to move on to manufacturing economics. I worked for a boutique consulting group out of Cincinnati, and I was chief economist there for five years. And then back in 2022, I shifted from manufacturing and came back to construction. And so Construct Connect, in particular, focuses on non-residential construction. So that is your non-residential building plus your civil construction, right? So all of your big towers and then all of your your you know, big infrastructure, your roads, your bridges, water, you know, water, sewer, dams, all that kind of stuff, airports, et cetera. So everything but basically single family homes.

Everything but single family homes. Right. Got it. Got it. So we want to talk about data centers. I mean, obviously all the other construction that's going on related to the buildout of the AI infrastructure. That's what I think we want to spend a bulk of our time on. But before we get there, let's.

Let's talk about the construction markets generally. My 30,000 foot take on it is it's pretty ugly out there, right? I mean, construction put in place as the value of construction across all types of construction. It's pretty big. It's a little over $2 trillion per annum, which is, what, 5, 6% of GDP. So it's a consequential industry. But it's actually declining. In aggregate, I would characterize it as being.

in recession. Is that, is that fair to say? I think it's, it's definitely in a difficult spot. You know, from where I sit, one of the things that we do is we look at starts. And there's an important difference. And I can explain it in 30 seconds for the audience, right? The dollar amounts are roughly the same, right? A start is taking the total value of a project and assigning its value to the first day of construction. So when construct connect reports are data, we're talking in starts. So if a billion-dollar project is going to break ground on August 21st, right, we associate all $1 billion with August of 2026. A put-in-place report would say, hang on, let's look at the dollar flows essentially from the owner-operators to the general contractor, right? So if you

have a billion-dollar project and it's slated to take four or five years, right? And there's time curves where they look at project type and dollar value and they say, okay, well, the average project takes so many dozens of months to complete if it fills that particular grid on the chart. And they'll stretch those dollars out. And it's not a linear flow. It's not like it's, you know, $100 million every month. They show a distribution curve. And so you add up all those projects and their distribution curves. And it's a neat offering because put in place really allows firms to, I think, get the most accurate sense of what their own revenues will look like. Whereas starts is more interesting because it gives you, I think, a sense of where the industry will go, right? So we see all

these projects starting, right? And we're just better able to visualize and capture with starts. So it's, it maybe gives a little bit of a canary in the coal mine look, right? You can see things happening faster and starts than you can with put in place. Yeah, I mean, the way I kind of think about it is there's permits. So if you want to build a project, build something, you got generally got to go get a permit and you get the value of the permits. That's kind of the first step, and that's the best leading indicator. Then that permit ultimately turns into a start, which is what you've been talking about. So you actually put a shovel in the ground and you begin to erect something. And then the put in place is what's actually happening over time as you go from

the start to completion. And then finally, there's the completion, you know, the, you know, what's actually been completed. Is that, is that right? Yeah, I think that's a great, high level overview of it all. Right. I think, yeah, actually, we, the best leading indicator that I found is actually around housing is single family. You can look, look at multifamily as well, single, single, multi-family construction, the completions themselves are, you know, a very good barometer of GDP, you know, kind of the valuable things that we produce. Yeah. So very, very good indicator to watch.

So what, and if I look at, you're looking at starts, are you sensing, that's a head, that's, as you said, that's a step before put in place. Are you sensing any, if I look at put in place, it's declining. So that's my way of saying recession, the actual level of put in place is declining. Are you sensing any turnaround here in terms of starts? Is that starting to pick up at all? Yeah. So the interesting part with the starts is that they're bifurcated.

It's the construction market, the way that we track it, we track over 30 different subcategories of construction. Two of those, of course, are single family and multifamily. So if we just push those aside for just a moment, right? We still have about 30 subcategories we look at. Many of those subcategories, if you look at our monthly reports right now, a handful are growing tremendously fast, and we have a handful that are contracting significantly. And so what we're seeing, it's almost, I don't want to call it casehaped. Everyone keeps using the word K-shaped. It's the new fad word among us.

I think you used it. Didn't you in one of your papers? Did you say, did you say K-shaped construction? I think that was your, it all goes back to you, Michael. No, no. I'm very rarely try to use it. I'll call it bifurcated. That's my favorite.

That's my favorite claim of bifurcated. But what we're seeing is that, you know, data centers, power, water, to a lesser degree of water, but data centers power, some of the other infrastructure power, gen, is really pushing the market ahead. At the same time, we've seen in the last four years manufacturing Wipsoe back and forth when you measure how it starts, right? And part of that is a mega project start. And again, mega projects is just another fancy word for saying projects over a billion dollars in total value.

And so what we're seeing on the manufacturing side this year is a real pullback. Big double-digit pullback in manufacturing representing many, many billions of dollars. At the same time, that pullback is being offset by more than offset by the growth in data centers. So when you look at non-residential building, right, which is, you know, all construction, less all the housing, unless all the big civil projects and bridges, roads, et cetera, were up double-digit percentage points. But the only reason that we're up so much, when you look at NRB on the whole,

it's because of what's happening with data centers and that ecosystem there. If you pull out the data centers, or if you were to pull out the megaprojects, which, of course, data centers and megaprojects, there's a very high overlap. But when you pull those out, you see a construction economy

that is basically flat to slightly contracting. depending on how you measure it out. So that is a concern. You know, again, the industry, I think, has rarely been more uneven or more volatile. And going back to what I said before about my grid, it is important to understand, hey, as a business owner in the construction space right now, you have to be very discerning in which markets you're going to pursue and which geographies. Not every company has the luxury.

of getting to pick both their subcategories and their geographies. Some firms are stuck with a single geography. And so they may have to struggle a little bit there and say, hey, gosh, I need to enter into a different subcategory, you know, a different part of the market than maybe I'm used to, just to help keep revenues where they want, you know, keep them on target. Hey, Chris, so how would you interpret what Michael just said? I've got my own interpretation, but I'll let you interpret it.

What did you just say to you? What was the frame that he just provided? It's not a quiz. I'm just for the listener, just so the listener can, you know, get their minds around it. Yeah, it sounds to me as though there's quite a bit of heterogeneity, right? The bifurcation, broadly speaking, but then even within geographies or within category, you know, broad sectors of the construction industry, there's a lot of differentiation.

That's what I heard. So that the overall number might be. bit misleading or doesn't give the full picture. We really need to dig into the detail to understand what's going on with construction. Right, right. Mercy. You get that, is that your interpretation? Yeah, I also get the impression that data centers are driving most of the top line number, right?

Yeah. Yeah, so, okay, so at a 30,000 foot level, in aggregate, not looking at all the things going on underneath from zero to 30,000 feet. you know, it looks weak. And I know to some degree, you're saying, I'm not looking at housing or single family housing. But if I put in single family housing, because that's kind of the weakest part of construction,

it's weak. But you're saying, look, that really belies what's going on underneath the hood. I've got this bifurcation, this heterogeneity, meaning anything AI-related, data centers, power, you mentioned water, at least on the margin, that's, I think the word is booming, isn't it? Isn't it? It's like booming.

And that's, but on the other side of it, excluding the AI-related stuff, particularly data centers, but, you know, broadly, that's where you see the, you know, obviously the weakness. That's the bifurcation that you're observing. Is that, is that, do we have that roughly right? I think that's roughly right. I think what we're seeing with a lot of the categories, again, of those 30 categories in the center.

Yeah. Yeah. Certainly the housing part, we're in our third. fourth year, really, of contracting on the residential side. So they would be on the bottom of that, you know, that matrix. In the middle of the matrix, we do see a lot of non-residential building that is growing, you know, low single digits. And those are in nominal terms, right? So with construction inflation, which another topic we could talk about, but it certainly has risen in this last year for all sorts of various reasons, whether it's tariffs, you know, energy shocks, etc. But we have

seen material costs move up 9% year over year, labor is up about four and a half percent, right? So you combine those two extremely important components of construction costs, and you can see where if we're seeing a four, three, four, five percent nominal growth in a subcategory, like a courthouses or fire stations, whatever it should be, but you see inflation growing at a higher rate, you know, we start to see that real impact is almost nil. It may even be contractionary. Got it. You're saying all these numbers we've been talking about are nominal.

To put in place, we're looking at the thing about nominally, that's flat to down, but on a real basis after inflation because construction industry is experiencing very significant inflationary pressures related to the tariffs, related to immigration and labor costs, related to the Iran War and energy prices, diesel, everything else. that on a real basis is actually even weaker, you know, on a, right, maybe the clock. It's one of those things where there has been so much attention given to the data centers part of this, right, Mark, that a lot of other things that are really important have been left to the side. They've just been left off the newspapers. And so that's why I'm so glad to be on here.

We can talk about some of those things that really do need to be given attention and talked about because they do have a significant impact. to everybody's top and bottom lines. Yeah, so before we turn to AI and data centers and all the other related stuff, interest rates, you know, obviously, this was the subject matter of yesterday's podcast that it's going to be released today on Friday on the 21st is the run-up in long-term interest rates. We've got the 10-year now at, what, four and three-quarters almost, at 30-year at 5 and a quarter. These are rates we haven't seen since the global financial crisis. You must be sweating about this, no? I mean, this must be causing some fair number of consternation in the construction trades.

It makes me nervous. You know, we just went through a big financial challenge, right? Within the commercial real estate market, not very long ago, right? We had all of these institutions, all these. call them institutions, we had a lot of CRE debt, right? That was tied to offices and other property types where they were no longer penciling out. They couldn't refinance, right? And so we had this overhanging debt. And I think some of that issue has been resolved. But, you know, as we think about maybe a next wave of higher interest rates, right, moving higher than expectations, which is exactly

what happened in 2023, 2024, you know, everyone was expecting rates. I think to, come down and they didn't, right? Inflation was much more persistent. We can, I'm sure you've covered the whole transitory inflation thing at nauseam and other podcast, right? But to your point, if we once again see higher rates than are expected, it really does become a concern for the industry, right? Because so much of the AI buildout has been through debt financing, whether it's on the books or, and a lot of it's been off book financing, right? But it's, It's huge. And one of the things I think makes this economic boom around AI and data centers unique is just how fast the capital has been accrued and accumulated and gathered, right? We have financial tools that are getting way out ahead of the ability to construct things.

And if you look at the put in place data like we were talking about before and look at the starts data, there's this divergence occurring in real time. And it's a massive divergence in terms of how much money is being borrowed and earmarked for future construction that may not happen for many years to come. And so I do wonder, you know, what will happen in the future when we have these economic booms where the capital can be accrued so much faster than the actual construction, the actual creation of the asset? You know, I think it'll be fascinating to see and to think about the ramifications for that. And to just bring it back, sorry, just to bring it all back, right? Interest rates are going to play a really important part in all of this, right, when we have the financing, getting out ahead of the physical construction. Now, the higher rates, I think they do significant damage very quickly to a lot of the construction market. I mean, you mentioned commercial real estate, CRE.

obviously the housing market's incredibly sensitive to the 30-year fixed mortgage rate. Do you think, though, that the higher rates are going to matter at all to the financing necessary for the data centers and power that are related, anything related to AI? Does it matter that, you know, long-term rates are up 50, 75 basis points, you know, in the grand scheme of things, given the amount of capital that's being thrown at the industry? Does it matter at all? Well, sure, right? Because we have to, I think, I think so, right, because we have to, at the end of the day, the hyperscalers, for example, I'll use as a generic term for all the firms that are building these data centers, right?

But at the end of the day, you're borrowing all this capital. You have to be able to repay it at some point, right? And it's a huge amount of money that they're going to be investing, right? That they've already invested, that they are planning to invest. if you look at their quarterly reports, right, and their financial statements and all those other things, right? I mean, we're talking on the order of trillions of dollars over the next few years, in addition to what they've already invested, right?

Eventually, they have to make that money back. And that's the concern because if you look at what's happening with those AI models right now, we have premium AI models, but then you have a whole bunch of models underneath them that are much, much cheaper to operate. I was talking to some of our own IT executives and others, IT experts. And you know, you can talk about cost per compute, right? And it may be several dollars for so many tokens, a million tokens, for a really advanced model. But if you look at a model that is a little less advanced, right, the cost of it might be one-tenth of the premium model's cost.

And the question becomes one of, you know, at what point do firms, especially I think the enterprise firms start to realize, look, I don't need to be paying, you know, tens of dollars per millions of tokens, I can be paying 30, 40 cents for a million tokens maybe. And what happens to the financing, right? What happens to the repayment schedules when firms aren't able to sell a premium model to all of their enterprise customers because the enterprise customers have figured out, hey, I can, I can get 98% of the value at one-tenth of cost, right? And so that's where I start to wonder about how this market, how the financing of all of this really plays out. Yeah, so you sound, you sound bubble.

Yeah, that's like this is word, right? I'll tell you what, here's my part. Are you saying that? This is a, this isn't going to work out. Economically, it's not going to work out? No, I think it's just going to look different, right? One of the things that gives me great hope for AI is how it can help small firms, right? Say you're a one to five person firm, right?

Well, oftentimes you need a marketer, you need a lawyer. You need, you know, a bookkeeper, et cetera, et cetera. You need a little bit of all these roles. And so I think the, what AI can do for the economy is going to be tremendous, especially at the smaller firm sizes. I think at these really big enterprise levels, it might actually be harder, right? They have deep pockets. So it's great to get a big win early if you're a hyperscaler.

Say, hey, I've got, you know, a big contract with a big firm's bringing in a lot of money. But I'm not convinced, right? And we've seen this in the news, right? I think the big enterprise firms are realizing, hey, you know, buying AI at all cost, you know, token maxing, et cetera, et cetera, as they call it, right? Just spending as much money as you possibly can on AI. It's not having the return that maybe some of them are hoping for. But I think where the real value will be made and it's going to be unlocked in a country like ours

or in an economy like ours where you've got strong entrepreneurial spirit, And imagine how much more powerful one or just a small number of people could be if they have access to just even a decent AI model that can be, you know, a sufficiently good, you know, document writer for them, you know, legal consult, not an expert, right? All of those kinds of things. So I really, my greatest interest is to see how AI unlocks creative ingenuity and entrepreneurship within this country in a way that hopefully we've never seen before. I'd love to know what you guys think about that. Well, I mean, I agree with you. I mean, I think you can see it in the business formation data. I mean, the applications to form a business, this is data we get from the IRS, and it's real time or close to real time and a lot of granularity.

And there has been a significant pickup. And I think that goes to significant degree to empowerment related to AI adoption. So I think that's the case. But I want to come back to the point about financing. Chris, what do you think? I mean, how big a deal or worry is this? I mean, a lot of credit, both debt and equity, a lot of capital is being thrown at the industry

to build out the infrastructure, including the poster child, is the data centers. And, you know, the question is, will the future revenues generated from AI be sufficient to cover the, the payments needed on that capital, on that debt. And if not, you know, obviously that has all kinds of ramifications, you know, for the

AI, but also for the financial institutions that are providing that credit. So how big a deal is it? The higher rates themselves? Well, just generally, I mean, it sounds like I listen to Michael. He sounds pretty concerned about it, not to put words in your mouth, Michael, but it sounds like that you're pretty nervous about that. Yeah. I mean,

it sounds to me like an internet story, right? Internet story. It's a very similar type of trajectory here. A lot of capital goes in initially. There's, you know, it's trying to figure out the optimal allocation. Things blow up. But that doesn't mean the Internet goes away. It's the second round of companies that actually benefit and figure out how to use this new technology.

in a cost-efficient way. So that's the story I kind of heard here that, you know, there's a lot of capital going in. My take is that to your earlier question, the long-term rate actually, I think, does not matter in the short-term here. There's so much kind of faith-based economics going on that the returns are going to be so astronomical that, who cares, 50 basis points, 75 basis points compared to the returns that are being projected. I don't see that as having any short-term impact here. But as the reality sets in, I think Michael's absolutely right, cheaper models, Chinese models, what have you coming in. I think that that starts to unravel some of those lofty expectations. And I do expect to see at least some type of a consolidation phase here.

I just don't know how dramatic it will be, right? So maybe we're overbuilding, but I don't know if we're overbuilding by 10x or, you know, a slight amount. and we will use that capacity within a few years. That's the mystery to me. That's sound right, Michael. Did you get the sense of what you're trying to say, right? Yeah, I think so. I think, you know, like Chris is, I think, pointing out without, again, putting words in anyone's mouth.

No, you can do that. I do that all the time. Feel free. That's what we do here. That's what we do here. But what is the market structure? Is this an oligopoly with a very competitive fringe? Is this a market that's going to be?

be very high fixed upfront cost with extremely low marginal cost. You know, I mean, it is the cost of me running an AI query essentially going to come down to no more than the cost of the electricity, the couple of pennies at cost to run that query, right? So I think those are the two big questions that we need to, as economists, wrap our head around is what's the market structure? And then within that, what is the microeconomic structure for these firms? if it's high fixed cost, very competitive or very low marginal cost,

you know, what can that tell us about the future of this market, right? Do we look at this like an airline model thing? I'm not saying that it is, right? But, you know, will we see these large firms constantly, you know, 10 years from now doing amazingly? Or is it going to be like every 18 months somebody is declaring bankruptcy or something? And I know that's very extreme. I'm not saying that that's going to happen, but you just get the idea, right?

There's a whole spectrum of outcomes here depending on how we understand the market structure. All right, well, let's stick deeper into the data centers because that is the poster child for all the AI build out. And there's obviously a lot of controversy building, or no pun intended, around the data center construction. But before we get there, how big a deal is it? Let me give you the numbers that I look at and then maybe tell me if I have them, if this is, I'm looking at the right numbers.

You know, if I go back to look at the census data, this is where the put in place numbers come from. And I look at put in place. And the Bureau of Census says recently, I don't know how recently, but in the last year or two or three, started releasing data on put in place for data centers. You go back and look before chat GPT was put on the planet, you know, before November of 2022. it was about $10 billion per annum. It's now $70 billion per annum. So, yes, that's a very large percentage increase, 10 to 70, but that's over four years. So you do the arithmetic, you know, you're talking, what, 15 billion, maybe 20 billion per annum.

In the grand macroeconomic scheme of things, that registers, but it's pretty darn small. I mean, it's, you know, a tenth of a percent of GDP, you know, something like that. So it's a deal. I'll take it, pretty going to the context of the rest of construction really getting nailed. So we need the growth. But is that a, how big a macro deal is that? You know, not the, not the AI that's generated from it, but the actual physical construction of these data centers. Is it really that big a macroeconomic deal? Or am I looking at the wrong data, Michael? Well, I think when you're looking at put in place, you're looking at, you're looking at just the construction of the physical infrastructure, right?

And so, yeah, sure, the physical infrastructure creation is a very small part, right? Right now, just for example's sake, when we look at data center construction starts as a percentage of all non-residential building, it's about one out of every $4 right now. So now you can understand when it's one out of every $4, we now understand why the entire construction industry is essentially following in the shadow of the trajectory. of data centers. On the other hand, like you said, when you look at put in place, put in place has been much slower to respond relative to the starts. And in our own forecast, when we look at put in place, we're seeing about a $20 billion lift year after year for the next several years out to 2029.

Our own forecast for starts peaks sometime around 2029, maybe into 2030. So we, we, see essentially a peak, we're already well on our way towards the apex, but we still have two or three years in terms of starts. What it means for those who are actually working in this industry doing the actual physical work, right, the put in place is going to take many more years after that to catch up. So if we're saying forecasting a starts peak in 2029, 2030, right, the put in place peak doesn't hit sometime until, you know, almost mid 2030. So for those of my customers and prospects and all those other people who follow us, right? If they say, look, I haven't done anything as a construction firm or a consultant in the

data center space yet, my response to them is, well, you still have a decade left to go, not quite a decade left to go before we actually hit peak, put in place construction possibly. So there's a long story still to be written here and one where I think people can still get into this from a contraction perspective. The question, to answer your point, well, how big is it really? Well, that depends on how much of a lift we all get from productivity gains in AI. And I think that's where the real excitement for a macroeconomist, right, such as yourself, comes from. It's like, well, how much of a force multiplier does this new IT technology become for every

industry, right? Yeah, so no argument there. I mean, yes, the productivity gains, if they They haven't, I haven't seen them yet, but, you know, assuming that they do come, but that's down the road. The herein now, you know, over the last four years, the next end of 26 into 2027, before these productivity gains take effect, we're hopefully come to take effect. We're now focused on what's the juice that AI is providing to the economy. And it's largely through the buildout, which the poster child, again, is data centers. And if I look at the data center construction put in place, which is actually what's happening on the ground, not a start.

A start is going to happen. It hasn't happened. You're saying it's a good leading indicator of something that will happen. But, you know, the actual reality of what's happening today is it's, it's meaningful, you know, but it's not hundreds of billions of dollars. It's tens of billions of dollars. Correct. Is that fair? Correct.

Yeah. So our put in place, 2030, our put in place forecast is about $130, $130,000. billion up from like you said, some around that 60, 70 billion dollar market. So it's in the tens of billions for the next several years, yeah. Got it. And of course, there's water. You mentioned, you mentioned power. There's communication.

There's other stuff going on that are so 70 billion doesn't cover it all, but that kind of gives you, you know, orders of magnitude. The other thing I mentioned just real quick on that is that we're talking the construction, the power shell, right? The building. We're not talking the rat. the servers. And the interesting thing there, right, look at how much more a chip costs, like semiconductor, the semiconductor PPI, right? I think it's up 25% year on year, right? And so what

we need to remember is that we're coming into a situation where it used to be that the cost of construction, the cost of filling that building with the chips and the racks and the servers and all the electronics equipment, it used to be almost one-to-one. So almost 50% was the construction, a little more than 50% was the IT equipment, but now we're getting more to one-third is construction, two-thirds, I think, is the chips and the servers and all that stuff. And so, you know, you take that $70 billion lift between now and 2030 on the construction side, we're not even yet starting to include all of the additional spending that goes into the chips, the servers, et cetera. Yeah, the problem with that argument, though, again, from a macroeconomic

perspective, you know, what does it mean for growth? is all that stuff is imported. I mean, you know, if you look at the servers and the chips, you know, most of it is imported. So the actual benefit to production here of GDP, again, it's positive another 10th or two, but it's not a percentage point or two, you know, so it's meaningful. But anyway, but here's the other thing. Oh, did you want to say something on that regard, Michael? Well, I just wanted to say that's changing. That's a story that I think we have to realize it's changing. the number of fabrication plants that are being stood up in this country, whether you're in Columbus, Ohio or somewhere out in Arizona, right? I mean, we're seeing tens of billions of dollars

going into now chip fabrication. So I think that's a, that's a, you have a really fair point in today's economy, but I think that story will start to evolve over the next several years, too. So that's just something I wanted to throw out there. Yeah, no, no good point. Here's the other thing that I don't think people realize, and I think I have this right, the actual juice to the economy from the Chips Act and the impact that had on, you mentioned manufacturing earlier, manufacturing construction. That's the factories, the facilities that are going up. That was even bigger. That's now we're on the other side of that and that's now a drag on growth. But for between 2022 and I think through almost the end of last year, we got more juice from that

than we've been getting from the data center construction. I mean, that was a lot of juice. For certain, it was. It was. I mean, we saw an incredible. amount of volatility in construction as a result of what came out of COVID. And then, like you're pointing out, you know, in the years subsequently slightly after that, for sure. Okay. So let's now, here we are today. The other thing about data centers that I think has come to the fore and really come on

very quickly, probably because it's political and we're coming up to an election, is all of the pushback against these data centers. I mean, political pushback on the data centers, you're now seeing moratoriums on data center development in many states and communities. We're in the state of Pennsylvania. We live in the state of Pennsylvania. Our governor, Josh Shapiro, just recently had an executive order. It wasn't a moratorium. I thought it was actually pretty interesting. You know, he came up with a set of rules and guidelines that need to be followed before a data center project can go forward in the state of Pennsylvania. But even the state of Texas, I think, Governor Abbott has put stuff in place.

How big a deal is that? I mean, when you said you gave us a forecast, you know, how do you, when you do that forecast of future data center construction, how do you think about how this is all going to be resolved with the kind of the political backlash? Yeah. So I think we need to be really thoughtful about how we think about those moratorium. So, for example, in New York State, right, there was a moratorium. I think it's a one-year moratorium if I recall. call correctly the last time I read a headline about it. And that, I think New York was one of those

early sort of hard and fast deciders, hey, we're going to, we're going to really put the brakes on this. We have to remember that if you look at our data and you look at how many dollars are flowing into New York State compared to all the rest of the country when it comes to data centers, New York's getting like less than 1%. It's a really tiny piece of the pie. So if New York says, hey, we're not going to build data centers here forever and ever again, you know, just as a hypothetical, it doesn't scare me, right? It's, you know, less than one percent. Texas, that's 20 percent. That's huge. You know, I'm vastly more interested in monitoring what happens in Texas than I am in what would happen in a Vermont, a New Hampshire, New England state, where it's just it's not the place

where data centers are being built. And it's not the place where they're going to be built in the future. The places where they're going to be built in the future are going to be your classic rest belt states, the north-central east part of the country. If you follow the Census Bureau of regions, right, it's your Wisconsin all the way down through Ohio. That's going to be big next generation place. The other place that we need to be watching for geographically is your southeast states where we have a really strong natural gas pipeline and a really strong infrastructure already in place, which is huge for those companies that want to stand up a data center and then build power generation behind the meter, where they're doing their own on-site power, right? So everything from

Texas, all the way across to Georgia and then up to the Carolinas, those are the other states. We really want to be watching for legislative changes, right, and changes in those political wins, as you were alluding to, because those are the states that are going to drive the next many years of real data center growth. There's a handful of other states. You know, the West Coast doesn't do a ton of it right now. They're not building a multi-gigawatt facility. You know, there's a little bit happening in Arizona. North Dakota, of all places, has over a billion dollars of data center starts this year. You know, so I think those owners and developers are definitely thinking about not only power and water and space constraints, but like you say,

Now they have to think much more so about those political constraints. And I think I've now given us a good sense of which states are really going to matter the most. So if you can only follow a couple of governors and a couple of these political headlines, right, follow the ones that are in these particular geographies because those are the ones that are going to be the real guides to what happens with the data center rollout in the years to come. Okay. So it sounds like what you're saying is, look, Okay, so we've got these moratoriums, but just as long as the states in, you mentioned, the east-north central region, Wisconsin, down through Ohio, parts of the southeast, you know, over into Texas,

if those states continue to remain, uh, uh, willing to, yeah, favorable, I'm searching for the word, then we're okay. We're going to be able to, we're going to get the data center construction that we need, and it'll be consistent with your forecast of the, of the, of the, uh, well, I think, well, I think that our forecast or anyone else is for that matter, because, again, that's just where the growth is slated to occur. That's where the, like you, we said at the very beginning of this, you know, that's where the permits are occurring. That's where, you know, the efforts have been put in recent years, right, for that next generation of data centers. Yeah, I guess the other, the biggest state for data centers is Virginia. I don't know. Mercer, do you know, is there any

moratoriums or anything in Virginia? Not that I'm aware of. Not that I'm aware of. I don't know, Michael, if you've heard anything, but I don't think so. But, yeah, that has historically been the nexus of them. Yeah. Yeah. I think Texas is going to overcome Virginia,

but right now Virginia is still the largest. Hey, so let's talk about the substance of the pushback on data center construction. And you listen to the opponents of data center construction. And there's just a general kind of nimbism for any kind of construction. So it's not just data centers. It's just that's what's being built.

But there's worries about electricity, you know, the demand for electricity, driving up the cost of electricity. There's the concerns around water, you know, the need for water to cool the servers that are in the data centers that are powering a lot of AI. There's concerns about critical minerals, you know, that are needed to go into the, you know, constructing these facilities and filling them with all the electronics. There's labor issues, all kinds of labor issues, concerns about that. I'm sure there's other, and just a general, I think, loss of agency. I think people feel like, you know, these things are being decided. The data center is going to be built down the street here.

I had no, no one asked me about whether I thought that was a good idea, you know, and it was all done, it felt like in secret and therefore I have, I lost agency. What else? The noise. Noise. Yeah, noise. The other thing, people are complaining about noise. How real, and I'm confused. I'm hearing very conflicting perspectives on all of this from, you know, credible people. Do you have a sense of it? What's real here? What's real and not real? In terms of the complaints, the concerns. The concerns. Yeah. Well, I think the complaints are legitimate. I mean, there have been those horror stories that we've heard about where people say, hey, a data center was built. It's making, you know, 24-7 noise. It's ruining people's sleep. It's ruining their mental health. You know, they're saying, hey, they have so many diesel generators. They're creating all this pollution. And then it's covering my town. And so I can't breathe. I can't sleep. You know, that's enough to drive anybody crazy. And I think that's, those are concerns that can be addressed. It was really interesting. So I, I get to go around the country and talk to some of the greatest, you know, people who are, who are involved in data centers.

And one of them was an engineer with many decades of experience doing this stuff. And he said, you know, the one thing we have to understand is that you can engineer a lot of solutions. You know, yes, it costs more money, but you can engineer solutions for noise. You can engineer some solutions for power, right? We can come up with solutions for water, right? Close-loop systems. I think that's, from what I understand from the experts out there, I get to talk to across the country again, you know, closed-loop is going to become the thing, right? Especially as we increase the density of the server racks, they're going to create so much more heat, so much, you know, they're going to consume so much more energy, right? So you're almost going to be forced into liquid

cooling on a closed network system. It's more... You mean evaporation? The weather. Evaporation? I capture the evapuration and I bring the water back into the system. Is that right? However it works. It's recycling the water. Yeah. Right. Right.

There's problems with that. Every time it evaporate, but I'm not way outside my strike zone. But okay, go ahead. Yeah. But, you know, so that's the thing, right? We can engineer,

I think, a lot of solutions to the problems that the public are bringing up, right? It's going to cost more money. Right now, you know, the way that we look at construction costs per megawatt of compute, the current gold standard would be to get down below $10 million per megawatt of compute, right? So $10 billion will buy you about one gigawatt of compute is what we're seeing in the latest very, very large data centers, right? But that's the economy's the scale that I think these hypers are aiming for is how do we drive it

under $10 million per megawatt of compute? If you do that, right, with these super, or sorry, with these hyperscaler centers, right? I think you can do, you can come up with solutions at scale that will solve a lot of these problems, right? You can come up with noise solutions. You can come up with, you know, a behind-the-grid system, right? Whether you're using solar turbines, batteries, right? Nuclear, micro-nuclear power facilities, right, that are essentially the size.

of a trailer on a truck, right? I mean, there's all sorts of, and that's what I love about this country is the amount of creativity being put into these solutions knows no bounds, which is a phenomenal thing about capitalism, right? So the solutions are available. They're being designed today.

They're being tested, prototyped even put into early production. So, you know, I guess what we need to do is maybe help the legislators understand the concerns of the people. but then also be working with those people on the cutting edge of these solutions and say, hey, look, we get it, we hear you. There are solutions.

We will make the hyperscalers pay for the cost of these solutions, right, so that those costs aren't, there isn't a moral hazard, right? The cost of the data centers, we don't want the burdens being put on the public. There are technical solutions and just have patience.

I don't know if we can do that. I don't know if the American public, you know, how patient will we be? You know, and how quickly can these solutions come to be scaled up? You know, that's... It sounds like, though, what you're saying is, look, yeah, these are legitimate concerns and issues to some degree or another, but... And we, we work, the industry recognizes them and ultimately will address them.

Technology will improve, we'll find solutions to these issues and constraints, and that that will ultimately appease the folks that are opposed to at least sufficiently that we'll get the data center construction that we need to power AI in the future. That's kind of your general sense of things. It's not going to be perfectly graceful. There might be two steps forward, one step back. You know, it might not work out exactly to script, but, that's kind of the general direction of travel. Yeah. I think that's fair.

You know, unless the government, right, unless government officials become real, you know, perfect actors, right? But I'm with you. I think it's going to be two steps forward. Oh, good luck with that. Perfect. Yeah. I've never heard those two words put together, politicians and perfect actors.

Okay. Yeah. Right. But I think we'll come to a balance. The other thing, too, that we would also, I think, be remiss not to call out is the need from a national defense perspective for this. It definitely seems that there's a necessary incentive to build out a certain amount of data center infrastructure for reasons to go beyond the private sector. So then the question becomes one of, well, look,

if we need these things for national defense, do we just start putting them on missile ranges, right? Oh, I'm sure, you know, there's a little place called Area 51 that no one's ever allowed to go to. It's in the middle of nowhere by design. I don't know. can we set up a behind the grid power system in the middle of nowhere and the military, you know, and government needs can be met through systems like that, right? Isn't Area 51 where they keep the aliens? Body of the aliens? Is that what, am I wrong? I mean, yeah.

Oh, I'm right. Okay. Well, I love Area 51. I would think there would be a permitting problem to put the data site in with the alien facility. I think that would be a problem, no? Yeah. I don't know about the aliens. I love the aviation component to Area 51

where some of the greatest technological advancements in aviation were tested there. That's what I love about that area. I didn't know that. That's interesting. Okay, so what am I missing, Marissa? What did I fail to ask on this topic that we should ask, Michael?

Well, you know, the more we talk about this and the more I hear about the concerns and the resources that go, into it, it almost sounds like it's a public utility, you know, like it's infrastructure like we would have for electricity, water, gas. And so then I think about the financing of it, the taxation of it, which is another concern that people have, you know, my electricity bills going up, why do I have to pay higher electricity rates or taxes on it to build out these data centers. How do you have a sense for the, I guess we can just talk narrowly about the electricity

part of it? I mean, we now have some of the hyperscalers going off grid, right? Because there isn't enough capacity on the grid to tap into. How much of that might be a, I guess two parts of it. How much of that construction are you seeing so far? And then how much of that might be a, I guess, two parts of it. And then how much of that construction are you seeing so far? And then how much of that might be a hindrance or to building out the data centers or slow this down at all? Do you have any sense of that part of it? Yeah, you know, that's a great question. If I can just rephrase it slightly, I think the question is, what are the biggest barriers to data centers growing as fast as the hyperscalers want them to grow? And I think Mark's touched on, I think you've touched on it, right? It's things like access to power. That's huge. When we think about regulatory control over power,

utility control, right? The hyperscalers don't want to deal with that, right? And they don't want to deal with the public pushback that comes with people saying, hey, if you're going to demand this much more power out of the grid, not only does it make the grid more sensitive to damage and to failure, but it also raises their cost, right? So there's a lot of, I think, logical reasons for them to come up with that behind the grid solution, or behind the meter solution. Sorry. I think that almost goes without saying that that's the way that all this has to go. And there's also a need for technical solutions there again, right? I mean, the number of utility turbines being built out over the next several years across the planet, they've all been claimed already. You know,

there's a serious shortage of turbines to the point that I was talking to somebody who's looking at taking old military aircraft, taking engines out of helicopters, and converting them to utility use. I mean, that's how desperate we are. We have people considering those kinds of ideas. And so that's part of it, right? The behind the grid, the behind the meter solutions, I think we'll answer a lot of those questions. What else was there to your question? Oh, and then labor.

That's the other big component, right? It's going to slow us down. I can't tell you where I was, but I was here in the U.S. talking to a very large G.C. Area 51? Yeah, I was over in Area 51. Area 51 G.C. No.

But, you know, it was interesting because I said, you know, these really massive, the gigawatt centers, right? How do you find thousands of construction employees and bring them to a place like northern Louisiana to build a facility, right, or to one of these other sort of more rural areas? And they said, you know, the first thing that they need to start doing now is building essentially a work camp, right? First thing you need to do is find a quarter acre by a quarter acre, level it. fill up with gravel and then put all these RVs on it so that these people have a place to live.

It's fascinating how the scale up of these, to get the right of time as a scale is completely changing how they have to think about construction, right? Because there's nowhere in, you know, in these places where there's a drury inn or a holiday inn or one of those many motels, right, they can hold thousands of workers on a dime. Are there still holiday ends, by the way? I just as a sidebar. Yes. There are? Okay. Yeah. I love Holiday Inn.

Not so much that you know that it exists, though, apparently. Yeah, apparently. For my childhood, I have these fond memories of Holiday Inn, going across the country, staying at a holiday inn. But, yeah, that reminds me of the sort of the oil fracking boom that we had, you know, decades ago when they would have to bring people to North Dakota and just set up a town, a manufactured town, to bring labor in. And it's temporary, right?

You said they build the data center, but then once the data center, But then once the data center is up and running, it's not like, I don't know how many people actually are at the data center when it's up and running. I can't imagine there's many. So, yeah, there's real labor concerns to it. Again, it just sounds almost like a public good to me or a public utility. The more we talk about the scale and the resources that have to go into it. Let's the other aspect of this. There aren't a whole lot of jobs, right?

I mean. Once they're up and running, I mean, there's nobody living in the middle of North Dakota. So you got to do what you just said. But once you do it, you look and see how many people are there. It's not certainly not hundreds of thousands of people. It's not even tens of thousands of people. It might be at best thousands of people. So it's not like this is a job creator.

You know, again, the infrastructure build out, not what AI is going to do in the future. That's a whole different ballgame and can cut in a lot of different ways. But in the near term, it's not a whole lot of jobs. So, okay, Chris, to you, what did what didn't we have? ask that we should ask, Michael, now that we have him. Anything? Well, let me propose alternative scenario based, going kind of back to the original arguments about alternative models and maybe, you know, these, these frontier models aren't

where the actual demand is going to be. And also, I'm also thinking back to the mainframe era in computers, right? As you described, the demands for land and energy, right, there were also some concerns back then because of these huge, you know, vacuum tubes. factory size computers, and then the microchip came along and revolutionized it, right? So that might be, so I'm thinking, well, maybe there's a technological breakthrough here. That doesn't look anything like just making the systems more efficient. It's actually some new model, more efficient model.

So my question to you is, what if indeed we have our overbuilding, we have all these empty data centers at some point, what do you do with an empty data center in the middle of North Dakota? A lot of aliens. You could put a lot of aliens in those data center. If they're sitting in area 51, another reason why you put them in area 51. It sounds like a risk of a lot of stranded capital potential, right? It's not like even an office building you can with some effort convert to an apartment building. But data center.

Well, yeah, I mean, you have any thoughts? You're building in the middle of nowhere that does nothing. And we already have, I mean, there have been storage about happening in California, right? you build a data center and then all of a sudden there's no power, the regular, you know, something happens. The regular says no power for you, and you literally have a very expensive brick. So, I mean, we've already started to see that in some very particular cases. I think the question, really, it's, I care less about the physical infrastructure.

If the building sits there, that's bad. The question I start to wonder about is, well, what happens to all the financing, right? How does this impact? When you've got a trillion plus, when you've got trillions, maybe a trillion, probably trillions of dollars that become at risk of default, right? What does that do to the overall financial market? I think that's where I would really focus my concerns, right, Chris? Yeah, you're saying, do you know, Michael, state-of-the-art data center,

how much does it actually cost to put one up? Not the stuff that goes inside, just the actual data center. Right. So, I mean, right now the average data center, and this is, you have to remember, But this is not normally distributed, but the average data center is running over $3 billion. $3 billion. That's the average, right? Because you have to remember, we have some that are running $10 billion, $30 billion, $50 billion, right?

Meta just said, hey, we built a $10 billion data center. It's not even finished and we're already going to expand it to another $40 billion, right? Does that include the power plant, the gas? I don't read the blueprints. I know the dollars, you know. Do I get to pick the color of the paint that? goes on the data center for the 30 billion? Is that part of the deal?

I would hope so. I would hope so. No. It's extra. It's extra. Yeah. The fascinating thing is that the scale is getting even more wild in the sense that you look at a project Kestrel, a project Jupiter. These things, we're now cracking the ideas for a $100 billion data centers and greater.

I mean, that's, I mean, I can't really wrap my head around. What does a $100 billion data center look like? You know, how many hundreds of acres of construction, you know, are we talking about here? So there's some. But you don't think that includes the electronics that go into the data center, the servers and all the electronics? You think that's the actual cost. Is it the physical structure? This is.

So at Construct Connect, we do our absolute best to make sure that we're only tracking the construction dollars. Okay. This stuff may be headlines right. now. I mean, some of these projects, again, they may exist theoretically, conceptually. You know, it's going to take time for us to figure out, you know, when they say $100 billion, to your point, are we talking just phase one? Are we talking 10 phases with the last phase occurring sometime in 2040? You know, I think for these really, really big centers, it's not only just, is it, what part of it is the power show and what part of it is the chips, but we also have to be careful

I'd ask ourselves, well, how many phases of construction are we talking about? You know, the last $50 billion of $100 billion project. If it's not slated to even break ground until 2045, you know, I think we need to start discounting these things, you know, so that we can keep, you know, our minds from exploding. You know, it's so amazing to me because I can remember not long ago, if you said a billion dollar semiconductor plant, chip plant, that like, are you kidding me? a billion dollar intel plant in New Mexico. I mean, it sounds so quaint, and that wasn't that long ago. I don't think. 10, 15 years ago, but anyway.

Okay, so we've taken a fair amount of your time, Michael. Really appreciate it. Maybe I'll just end by asking you, what didn't we ask that we should have asked you? You know, what are we missing here? Anything? Do we cover all the ground? You know, the only things that we need to maybe think about are, what happens, you know, if these data centers can do all that they say they're going to do, right?

Workers become vastly more productive, but we may need vastly fewer white-collar workers, right? And we're already starting to see glimmers of that, right? Entry-level white-collar job opportunities have already started to be impacted, I think, by large firms bringing in AI. So I think one of the most fascinating parts of this entire story will be, you know, what is the impact of AI on the job market, right? What do we do? We have a flat to shrinking workforce that's going to come head to head with tools that could vastly improve productivity, but who will capture those gains, right?

Hey, I guess the passing part. I think that's a great question. I got a recommendation for you. There's a podcast called Inside Economics. We just had David Otter, MIT professor on. It was actually one of the best podcasts we've ever, correct me if I'm wrong, guys, but it was a great podcast on the value of labor.

So it just came out. So you might want to take a listen to that because I thought that was pretty cool the way that was laid out. But anyway, but Michael, I want to really thank you for spending time with us. This is very, very engaging and informative. Really appreciate it. Anything else, guys, before we call this a podcast,

Marissa, Chris, anything? We good? Oh, I think we're good. Got the thumbs up. Okay, Michael, you're good, thumbs up. Oh, this is great. Thank you so much. Okay.

Yeah, well, with that, dear listener, I hope you enjoyed the podcast. We'll talk to you soon. Take care now.