How $31 Trillion In Datacenters Get Built By 2050 — With Dallas Dolen

0:00How does the projected 31 trillion in data center construction coming our way actually get [music] built? And will it ever get built? Let's talk about it with PwC TMT leader Dallas Dolan in a conversation brought to you by PwC. And Dallas, it's great to see you again. Welcome back to the channel. It's great to be here, Alex. Great to

0:19see you. All right. So, recently PwC released a report on the projected data center spend. And you know, I'm recently out of looking at Anthropic's S-1 where they're planning to spend 500 and something billion dollars in the coming years on AI infrastructure. So, I thought like when you looked at where spending is going to come, maybe it will just top out there. Uh but your report tells a very different story. And the headline is pretty wild. So, the headline uh of this report says that global investment in AI infrastructure could hit 31 trillion through 2050 with uh data center spending in 2050 coming out at 1.8 trillion per year. So, clearly PwC

1:04doesn't think given the reporting that it's done that this data center construction boom is going to slow down slow down anytime soon. So, it'd be great to hear from your from you uh your perspective on um where we go from here to there and why uh you think that it ends up getting to those large numbers. Yeah, happy to go into it. And you know, it's amazing that the 31 trillion is the headline number, right? As you as you mentioned, but the possibility for growth on top of the 31 trillion, it should be the thing that's even more remarkable about it, which is the possibility it gets to, you know, 50

1:39trillion depending on the way certain cards fall um over the next uh over the next few years. I think at the end of the day, there's a couple things to keep in mind, Alex. I mean, first and foremost is there's this really big and you mentioned the the S-1, you know, leak, if you will, coming out with Anthropic. There's this big focus on exactly how much is being consumed from a compute point of view and from a token point of view or however you want to measure it by the frontier model firms. So, the Anthropic and the Open AIs obviously and

2:08and and others. That's only part of the story. They are in my you know perception you know one potential user of future compute but not the only user of future compute. And candidly there's sort of a replacement feature dynamic built into it as as it relates to frontier models. Of course we're talking about an hour and a half after you know Google just released Argon which is their latest model Gemini 4 and you know we're a week out from like Astra 6.1 and a couple weeks out from you know 5.5 from from you know from Anthropic. You know the reality is there's always going to be like

2:42something you know something bigger something better something faster that's coming out from both the frontier firms but also from the open you know the open models. And and at the end of the day I think the thing we have to keep in mind is it's less about the throughput related to the model itself and much more in my view about the throughput as it relates to total consumption that's occurring. What is the amount of data that's flowing through? What are the number of prompts that are flowing through? You know candidly how much of this is cached versus you know new prompting that's happening. So, there's so many different other variables that

3:12come into play. So, the build out of infrastructure I think maybe just separating that concept out from you know even just pure AI itself is a really critical piece of the equation which is we simply don't have enough infrastructure overall. So, when you start like bifurcating the market and the consumption level that's happening the 31. you know 6 trillion it seems like a gigantic number but it's still only part of the story and there's so

3:32much more that goes into it. Can I pressure test this a little bit? I mean so, does this report assume that the amount of compute that we have today to generate tokens is going to be the same in the future? So, for instance I would imagine let's say we hit 2050 that the AI companies will find a way to run these models much more efficiently, and therefore we don't need as much compute as we have already, because we've seen efficiency improvements, you know, across the board to date. Uh so, is it is the presumption then that the consumption is going to get so intense that even at more efficient rates,

4:15there's still going to be this boom in demand? I think that's absolutely the the assumption, or maybe it's even just the known the known fact pattern that's happening in the market today, right? Demand for compute is vastly outstripping the supply for it, just full stop. And, you know, whether you measure that by the way in just, you know, computational value as it relates to, um you know, units of compute that the hyperscalers or neo clouds would use, whether you use it in measured value in terms of token access, we've seen this, you know, deceleration or throttling that's occurred with some of the model companies over time, or you

4:48look at it on a separate metric, but a very relevant metric, which is just energy or power. Um the reality is that looking out 4 or 5 years, there simply isn't enough of any of it uh in order to to to, you know, to power the the demand that's there, and it it's already happening, right? You definitely have those experiences where things seem a little bit slower, and it doesn't matter what platform you're on, including, you know, we I should fail to mention like platforms like Groq and and some of the others, which are fantastic as well. You look, it just it happens because use is

5:18up. Uh use is up across the board. Now, whether it continues to grow at the rate that we've seen, right? Some of the AAR compounding is pretty remarkable. You know, that part of it is only, you know, part of the story again for a few of the companies. As you've seen as it relates to broader usage of AI, again with the open models in particular, the growth feature on uses of tokens and the growth feature from a projected point of view on on just overall compute, you know, continues to be an up and to the right line, um and that's something that, you know, candidly I think, you know, by the way,

5:50gives a lot of confidence to the builders of infrastructure for them to say, "Hey, it's worth it for me to invest this time and the money. I'm less concerned about exactly who's going to use it as in, you know, is it a model builder, is it a neo cloud, is it a specific enterprise or company or is it a consumer? The broad belief is that like there's always going to be the need for that. And importantly, like we're looking at this, you know, this forecast on demand as it relates to, you know, this being a global number, keeping in mind like you have really untapped markets. I was just in

6:21Southeast Asia, you know, hanging out with my friends in in the Philippines and also in in Japan. I mean, yes, they're they're at the level as far as understanding the tech and they certainly are are, you know, all over it from a use case scenario, but the proliferation isn't nearly there, right? So, as the maturization happens in those markets, demand will continue to go upwards in in those markets as well. And I think Europe will will, you know, stand to gain. And of course, there's a lot in America, too. You can't forget about any of these things. So, before you get even get into like East versus

6:49West and which models there are, you just look at the global framework and you say there are, you know, approximately 9 billion people and a lot of these folks at some point in time or another are going to be looking to access, you know, this technology and, you know, the conduit to to providing that access is going to be, you know, the the pipes and the data centers and the chips and the memory, not to forget about the memory component of the equation, too. The the future's in all

7:14of this. Yeah, and I'm glad you mentioned power because in order to power 31 trillion in data center construction, you're going to need a lot of power. So, have you thought about like how the power comes about in order to make this investment actually happen? Yeah, I mean, we've we've certainly thought about it. I think you have a bigger, you know, bigger story here as it relates to, um, you know, the political system, um, the the underlying, you know, who should pay for it and and the broader, you know, dynamic as it relates to uh impact on on pricing as well. That's been a big community discussion, um

7:50which I'm sure we'll get into, but I think I think it's good to be having the conversation about power right now as it relates to future demand because it's actually addressing a problem that's persisted for a very long time in a number of places. If we just want to be selfish for a second, speaking of the United States, we've seen what happens when demand outstrips supply in certain markets, right? This has happened multiple times in California, it's happened in Texas, like these have been widespread, you know, reported and very

8:19unfortunate tragic events. Um it is a, you know, a lack of of of uh reinvestment in infrastructure, but of course that's, you know, call it like part of the story and and that part is largely like the government and taxpayer and, you know, what what do the the municipalities and and bureaucracies, you know, state and federal governments

8:37do? I think that really great part about the conversation that's happening today as enterprises are getting involved and the free market gets involved, you have this opportunity for, you know, those folks to work with the power companies and to work with the governments and say, "Oh, I have an idea. Like, we want to build a

8:55data center. You guys probably want us to build a data center here, maybe maybe not, but if you assuming that you do, so that here's a situation where you have a, you know, a a natural a natural buyer or a natural seller, you know, coming together and looking into entering a contract for that, you know, coming together and making a decision about like what to buy around it. That includes the power piece. We've seen these deals done, you know, whether it's uh it's it's, you know, Google down in the Southwest, you've certainly seen it with Microsoft um up in New York and and obviously a ton of a ton of these uh

9:23deals being made, um you know, in the Midwest with some of things that XAI has done there, you know, for both their manufacturing and for their data centers and for, you know, the power dynamic as well. So, I think what's what the the the market is basically forcing these companies and these governments to do is actually to work collaboratively, even though it doesn't feel like it on the face, to actually solve for the power problem that, genuinely speaking, it's actually very difficult, I think, at the taxpayer level, at the individual level within these communities, um to really solve for. And so, chances are it actually gets better, you know, as

9:54opposed to, quote, getting, you know, getting worse necessarily as these companies come together and try to frame out, you know, future demand and and and also forecast what that means for the communities themselves. Right. And there will be international competition, as you mentioned, right? There are going to be countries that are going to be like, um you know, for instance, I'll just throw it out there, as the United States deals with moratoriums and kill switches and political protests, uh we want you to come, build their your data centers here. And in fact, your report uh shows that the expectation is only 48% of data center construction is

10:26going to happen in the United States. I say only. Um I don't know, from the headlines, you would think it would be 70, 80, uh but there is a lot of international opportunity. So, it'd be great to hear your perspective just on where the power is going to flow, both the watts and the political power.

10:43Yeah. Um so, it's funny. I I I I love the commentary on the split, and I think it's really important data point to keep in mind, because when you look at um when you look at what's happening as it relates to, you know, growth of markets, and I mentioned that, you know, for Latin America, Southeast Asia, and other places, um and you also look at the current

11:03footprint of data center and compute. What's really notable is that the amount of actual compute that exists in the United States today, not necessarily the number of data centers, but actual compute, is significantly higher than exists in most other parts of the world. Arguably, that is why people would say, "Hey, at some level, you know, the US is in the lead, um as it relates to AI and, you know, the future of, you know, things that are happening, you know,

11:31specifically in in this space. What I think that sort of ignores a little bit is our pace and ability in the US to actually build things quickly. And importantly, um, to get people to work together to go to go off and do that. And so, you have one bit of it being like, if you can't get something done fast in the US, you can also serve other markets. And you have a lot of demand, a lot of demand. I was meeting with with with folks who are in positions of of power, so to say, in Southeast Asia, who were telling me, "Hey, we would love to have more of the

12:07construction, right? Whether it's to to build ships or to build memory. We would love to have more data centers, and we would love to bolster that, you know, from an economic point of view, in in incentive point of view, to have it happen in our country." And there's there's more than a handful of of of uh of places where that's happening, and I think Southeast Asia is probably the number one example of that. Um, you know, the Middle East is a little bit of a of a tough one to to lock down right now just because of the war, you know, that's going on there and the

12:33uncertainty um, associated with maybe getting a construction project started at this point. But, I see that likely opening up assuming, you know, you have, you know, further further containment or curtailment, um, you know, of of the activities that are happening in the Middle East. So, even if you just look at those two territories, or those two jurisdictions, if you will, parts of the world, gigantic populations, uh, growing demand for for services for services like this. And also a really high degree of willingness, candidly, to attract that foreign direct investment, and a willingness also to invest on the

13:06parts of the governments themselves. That that's a natural equation, Alex, for them to start saying, "Aha, I'm going to go build another data center in United Arab Emirates, or in the Philippines, or in, you know, Vietnam and a lot of these places." We're going to see a lot more of that activity happening, um, over the next couple of years. And look, it's going to put some pressure, rightfully so, um, you know, on on the the you know, the the global dynamic, right? And and hopefully that also unlocks, you know, some of the some of the hold-ups that are there, you know, as it relates to the speed and the

13:35execution that's happening. Again, the the good news is if you look at it again from a a US lead point of view, is that we already have, you know, way more compute and way more data centers, but you know, there's always that need for more. And as I said, I mean, the the curve moves up into the right as it relates to demand, and I think there's no sense

13:51that there's going to be a slow down. The only challenge is, do they have, frankly, the inputs, right? Do you have the chips and do you have the memory, and do you have the transformers and everything else that you need to actually do all these things? And then, of course, you know, can you get folks to approve building of the uh the power

14:05facilities to do it, too? Right. I was in Indonesia over the summer, speaking of which, and everybody there is using ChatGPT, except for this one cab driver that I had a very long ride with. He didn't speak any English, I didn't speak any Indonesian, we had 3 hours in the car together, and I was

14:22like, I need to get to know this guy. And so, I just pulled out ChatGPT voice, and I said, "Your job is to act as a translator between the two of us." And the the guy was like, "What is going on here? How How is this working?" And we really got a chance to get to know each other, and then afterwards, he said,

14:37Yes. "How did you do that?" I was like, "Sir, this is ChatGPT." Yeah. I'll tell you, the technology is so incredible. And and you know, um like not not to be undersold on everything that's going on. Like, isn't it amazing how technology can actually bring us together? Um I I I was in Beijing about a month ago. I met up with

14:59uh my father's best friend's nephew. Wow. We met over like, "Hey, WeChat, we should get together. Great, let's go to dinner." He's like, "I'll bring my wife." Okay, phenomenal. He drove to my hotel, he picked me up. He meets me in the lobby, he says, "Dallas, I'm Ming. I don't speak any English besides what I'm going to tell you right now, which is, please put this AirPod in your ear, and I'm going to speak to you in Mandarin, and you're going to speak to me in English, and WeChat is going to translate our entire conversation. We did that for 3 hours, just like you, 3 hours. We drove, we went to dinner, we

15:31drove back, we had a wonderful conversation between the three of us, talked about our kids, talked about our outlook on life. And you know, honestly, the ability to communicate at that level like demonstrates to me that a lot of people are basically like we're the same. Like, you know, like we we like each other, we like interacting. It doesn't matter where you're from. It it it actually was like probably the most empowered I felt in the time that I was in in China. Um And yet here I was with two people who

15:56I'd never met before. Uh and I'm in the backseat of their car, by the way, [laughter] like just driving around. Um and you know, you'd think that might be a high-risk situation, but it wasn't at all. It was just an amazing amazing opportunity to to interact, and it was all because of technology. And otherwise, it would have been an awkward

16:113 hours. No, that is such a cool story, and it's a it's amazing. Yeah, and I think that uh as AI gets better, speaking of like where this technology is going, uh translation, which is kind of where it all began if you think about the attention is all you need paper and the efforts within the Google to get better at translation, um that is just going to

16:29become way more seamless and and faster. All right, so before we move on, give me your gut feeling on this one. Uh 31 trillion is the projection. Where are you? Do you think it's going to be more or less in data center construction when all said and done by 2050, which which seems far away, but

16:46it's only 24 years from now? Yeah. I mean, 2050, I think it's like it's only like 4 years after like I'm supposed to retire, too, by the way, which is kind of a like crazy thought to put to put out there. Um Look, I think we'll have some hiccups along the way. And what's really difficult to predict is, you know, the political landscape, including we have midterms coming up in in just a few weeks, and like what do the midterms mean for the broader equation? Um and and and you know, just you know, like is someone going to come in and say, "No, we must stop. Um you

17:18know, that could be done in a lot of a lot of ways. It doesn't have to be legislation. I think that's also true at the state level, right? Keep in mind, you know, we always get hyper obsessed with with federal level elections here in the US, but of course at the state level that is far more influential of a lot of the things that most of these companies want to do. A lot of the the quote fights that are happening are at that level. Um I think maybe the good news is we're starting to see some really good pathways on on production and and development and building on these

17:46things. I I was reading um you know, like 30% of of a lot of the the builds are taking place today already have a localization requirement as it relates to power supply to data centers. You start thinking about that, like if every state copies the ones that are doing a good job to say, oh, let's do something that's helpful to like to the community in building in this way. If we have like something becomes like the normal or the the expected water level um on these types of deals, then we get to a place where we should be a lot more comfortable with the outlook. So, I'm

18:17actually largely a buyer of the 31 trillion. I actually think we're going to get pretty darn close to that number barring massive economic displacement. And I the number one thing I worry about on the economic side is interest rates. How much of that 31 trillion do you think will actually end up being built,

18:34more or less? Yeah, look, I I I'm I'm largely a buyer on the 31 trillion. If this was a an investment that I was making, I'd say this is something that's going to happen. I think we're going to have two potential called disruptions in in the broader flow, which could maybe displace the timeline a little bit, but not the

18:53total number. Um the first displacement is elections. I mean, you know, we we talk about federal elections. Obviously, we have midterms coming up here. Um generally speaking, we know what happens in midterms as it relates to a sitting president. Uh is it a possibility that you have uh you know, a switch in control as it relates to one or two uh chambers of uh of of Congress. Uh seems you know, pretty high right now, but we'll have to see what ends up happening. So, that's the first

19:15variable. Variable one part two is what happens as it relates to state elections and how does that play out? I think that's actually probably more impactful candidly as it relates to you build out of data centers and power and infrastructure and the like cuz the reality is it's the deals that are being done you know, in that space which are a lot more relevant than necessarily at a federal level. I mean yeah, we've seen some activity in the last couple of days you know, with the president and others around data center and AI build out but it doesn't mean that that's the primary thing that's happening. So, that's sort

19:45of one one one sort of potential roadblock and issue that arises. The second one is actually broader macroeconomics. I mean we have not seen COVID aside a significant you know, economic downturn in a while. Um, you have rising interest rates which are being used you know, to combat you know, a market you know, challenge if you will as it relates to both the fiscal situation here in the US from you know, just broad you know, broad budget and the second one of course is is CPI and rising inflation and the fact that the Fed is you know, going to use monetary policy to defeat that. The challenge of course then being

20:19if money becomes more expensive and there's a heavy debt financing as you mentioned as it relates to a lot of these companies and the requisite build out you know, of infrastructure being debt oriented, you know, will that slow things? Absolutely. It'll change the timeline. It'll change the mix. It may even change who the investors are and where money is coming from. I still think there's a lot of dry powder as it relates to where they get into the you know, the broader flow of things. Um, and I am you know, generally speaking relatively optimistic that people see like the big you know, the like the long

20:49upward right hand you know, tail so to say of growth and say hey, this is still worthwhile. But look, I mean those two those two macro features um, could absolutely be a challenge but I'm I'm mostly a buyer on 31. Right. And I think the political side of things uh, you know, with the broader what the economy does and doesn't do that's a little bit tougher to control or the tech industry, you know, for certain doesn't have, you know, as much of a bearing on that because, you know, if you look at the S&P 500, uh you might have those, you know, magnificent seven at the top, but there's 493 that need to

21:18pull their weight as well, and you know, they've been doing, you know, that with various uh varying degrees of success, shall we say, in in recent years. It's a narrow base. It's a super narrow base on which we're relying upon, 100%. Yep. Yep. But then you have the political side, which they do have an impact on, right? They can influence uh the way that people feel about it, A because they're the ones messaging about it, and B because they're building the data centers in communities. So, I'm curious to ask, do you have, cuz you've, you know, obviously PwC has done the research, you've had a lot of these

21:49conversations, you know, both in the US and abroad, why do you think data centers have become so unpopular among populations? I mean, where they're where they're, you know, have a prox- where they have a proximity, you know, that makes sense. Like if a data center's built in your backyard, fine. Um Yep. But they seem to be unpopular everywhere. So, what's your take on

22:09this? Yeah, it's Look, I think it's an interesting thing. Sometimes um you know, political issues take on a life of their own. I I lived and worked in DC for a number of years and had an opportunity to work in the space. Uh it was actually around regulatory policy and tax policy for uh for cross-industry issues, but a special, you know, focus of mine was on tech, and it was fascinating the things that people locked into. I mean, sometimes it's just like something that looks like a good headline, if you want to use it for political motivation, you can go off and grab it. We certainly saw that, you

22:39know, over the last uh you know, 25, 30 years um in in the policy front, and certainly in the area that I was working in, like in tax. I think on this one, on data center, I think you have a couple things going on. And first and foremost, you know, credit credit to where credit is due, which is, you know, if you can find a divisive issue, like roll with it, because there's an opportunity to raise money off it, there's also an opportunity to gain votes from it. So, like that's just sort of bare, you know, bare-bones like just political tactics

23:05that you're going to find everywhere. The fact that it's data centers is notable maybe to you and I and and certainly the hyperscalers and others care about it. Uh but it's, you know, maybe not that interesting of an issue for most people except it's getting its day in the sun because it has a lot of different things that I think actually speak to folks which are,

23:24you know, who's involved in it, right? Who are the companies? What are they doing? Right? In particular, what are they doing as it relates to resource usage, right? You mentioned and we've talked a lot about power, but you talked about, you know, water usage. You certainly have the dynamics of land usage and broader investment thesis including incentives that are coming, you know, from governments. And people say, "Well, wait

23:45a second. I use water. I use power. I want incentives, right? Oh, and by the way, you know, like my tax dollars should be going to fund schools in my neighborhood or to build a library or to, you know, like pave the roads or whatever it might be." If you can, you know, more than casually connect these concepts and say, "Ah, this is something that's disadvantageous and disadvantageous to me." It's an easily divisive issue and I think that's something that's, you know, come into the fold. And I mean, if we're being like really, really candid about this, Alex, I think the thing that, you know, broadly speaking, whether it's the

24:17builders of data centers or importantly the folks who are owning the infrastructure, right? In the hyperscaler uh chip manufacturing, uh uh memory manufacturing, and of course, the the model companies themselves, they've not really gone out to own the narrative around this. I mean, we've seen some good stories and some smart acting on the part of like Meta with CBRE, some of the localized training and efforts that they've done there. Um but there isn't like this really compelling like we've all gotten together and we figured out like what the talk track is

24:44and getting that narrative out there. I think if these companies actually got organized around this and said, "Hey, this is our A, talk track. B, it's truthful. Right? Like this is actually what happens when we build a data center in the county, not in your backyard, but in the county or in your state or whatever it might be. You know, if we

25:01build power, like here's a benefit, too. I think if they got together and did that a little bit better, I think you'd see a different narrative unfolding. But that hasn't happened. Again, it's been in in spurts, like I said with Meta, but largely you turn on the television, you don't see anything other than, "Hey, I'm going to prevent a data center from being built in your backyard." And people say, "Oh, that's great because I heard data centers use water or I heard

25:21data centers make my power bill go up." Um it's not to say that there may not be facts that that support that, but I think the the biggest problem that we have is we have not gone through and actually really like lined up all the issues and either knocked them down in the right way if if you're the companies advocating for it or if you're anti-data center, like you haven't gone out there and like articulated a really clear case

25:41except to say, "I'm anti-data center." And there's nobody pushing back to say, "I'm for data center." except for maybe the companies, you know, they're out there saying, "Well, we really need to build this stuff because if I don't, I can't go public or if I don't, I can't serve my customers." Which most Americans, you know, don't really care

25:57about that. Do you think the leaders of these companies need to be out in these communities? Like do you think Jensen Huang, for instance, needs to get himself out to Texas and speak to the like the newly reticent, um you know, Texans, which are usually support free enterprise, about the story? Look, I think um I don't know if you if if the tech community needs to anoint a leader around this, but at some level, the construct of anointing a leader, I mean Jensen's done an amazing job, for example, on on AI safety and uh maybe the anti-doomerism kind of construct as it relates to um you know, some of the narrative that's

26:32really taken a hold and gripped uh both the market as well as um as as well as the front page news um over the last, you know, 6 weeks or so. He's done a great job of being kind of the the the the voice for why that it should not be the case. I do think something similar uh or someone similar to that as it relates to like why do this and why it makes sense is probably a wise thing that you know that broad tech and infra and other

26:58companies should be doing. And look, I mean there's a lot of things they could hang their hats on whether it is actually coming up and and displaying accurate data as you know in in terms of water usage and power and everything else. Like the localization, you know, agreements that could be entered into and why you know these companies would

27:15pay for it. That's one thing they can hang their hat on. Another thing they absolutely should be doing is talking about the broad economic impact. I mean there's been a lot of studies like Georgia Tech did one that found the counties actually saw like higher employment and higher wages and higher household income after data centers were built. So you start looking at these things you say hey there's

27:32actually really good talking points. It's not that you need somebody with like you know a slick tongue and you know way to sell this. You need somebody who can articulate what it is that's actually happening. And it probably would behoove them to like get together and say aha like who's it going to be right? Find at least divisive person find somebody new but either way like figure that part of it out. Is that person going to get heat? 100% like whoever she or he is will absolutely catch heat from the other side. But look if you have the facts to support it and if you know

28:03we want to make this like a grandiose thing if if like the future of the western world depends on this happening now might be the time to find your Luke Skywalker right? Like this is the moment. If you have a Luke I'd pull that Luke Skywalker out like right now because it's important to get this thing

28:17going. They need it. Definitely. You know and it might not be Mark Zuckerberg but the meta CBRE example does seem to be a program that may be heading in the right direction. I totally agree. I mean and by the way they started that like a year ago. You see it come up a little bit you know watching Monday Night Football whatever sometimes you see that ad and it's like hey here's what meta's doing in some of

28:39the local communities. They're good ads. They're really good ads. And the question is you know like how could you scale that? And if you were the other companies who are involved in this, right? Like why wouldn't you scale that, too? It seems like a pretty practical thing to go do. Uh we've seen it in the last week, right? Between Muse and Dot yesterday and a few of the things. There's been some really cool products that come out that are very consumer-oriented. It's like, here's how AI is going to affect

29:02me and make my life better. Great. That's like half the story. The other half of the story is, here's why it's not going to ruin your community and the world, you know, those types of things. Like those two to me are like two sides of the same coin. Yeah, I think you have to have both of them within the tech community in order for this thing to like completely float and and and really

29:21work for everyone. Yeah. And I I you know, Dallas, you know, now that we're talking about it, I think a big part of this is also that we've seen so much of the doom and the AI will take your job and will make your energy prices go up and it might kill you by the end of the decade. Um And then you saw all the tech companies go to enterprise. And consumer AI was kind of left behind. So, maybe in a weird way meta Muse, which is a consumer product and a movement among some to like see the momentum there and try to release products that appeal to

29:54consumers. Um Yeah. That might also change the reality on the ground, potentially. 100%. I mean, I like it's interesting. You know, if you think about the narrative that could be put out there, and this happened to me yesterday. It was actually very funny. I was at um I was at this big uh telco event. I was in the audience. I was like 25 ft from this panel that had some like

30:14tremendous speakers on it, you know. Bret Taylor, a few other people were in the room and it was really phenomenal. And um my phone starts blowing up, like my texts. And I'm like, okay, what's going on? And it's my 15-year-old son. And he had uh he had a test uh an algebra test that was starting like in, you know, a few minutes. He had like 10 minutes to go. And he said, "Dad, I don't know anything. I'm going to fail and I've been studying and studying and studying and I don't get it." And I said, "Look, there's nothing you can do about this test. But

30:41for the next one, can you do me a favor? Like, if you're not talking about it in school, you should be. I want you to come to me. We're going to work on a prompt for GPT or Gemini or whatever, like whatever you choose, son. We're going to work on this prompt, and you're going to use that to build the study course, right? The the 5 hours of study that you need to do to go do well in the next test. If you start thinking about like that use case where the consumer, right? 15-year-old kid in high school who wants to pass a test, like that use

31:13case to build a customized learning uh a program for him would be worth any amount of money I could possibly spend, but also would probably make me a lot more positive, generally speaking, on AI. And by the time he got over it, by the way, he did perfectly fine on the test. That's the funny part, Alex. The best part of all is that, right? It's like speaking of doomerism, like the doomerism from the 15-year-old on the text message was pretty extreme, but the other part of it is this is totally solvable. And like you look at that example, like extrapolate that into like every household, and AI becomes very very

31:45useful. There's no longer an excuse, "Oh, I didn't learn the math subject. My teacher wasn't good, or I missed a day, or what have you." These things work really exceptionally well. And by the way, most of it's pretty free right now, too. That's the coolest part. So, Dallas, you know, thinking about what we've talked about, 31 trillion in data center spend scheduled, but some political forces thinking about maybe pauses or kill switches, moratoriums, um uh pacing the frontier, which is coming

32:13from the companies themselves. Uh let's just do a thought experiment here. If everything stopped, the presses stopped, and the uh global economy, businesses, people were left with the version of AI that exists today and didn't progress at all from where it is, right? Because I think these bigger numbers are anticipating that AI is going to keep getting better. Let's say it doesn't. Um what do you think happens

32:37then? Yeah, it's a great question. Um, I think we have an exceptionally useful tool today, like right now. And broadly speaking, I think the the lack of ROI, right? We There's been a number of reports that Gartner came out with one a few weeks ago talking about, you know, 80% of enterprises had like widespread AI usage and only, you know, single digit 7 8% were seeing some benefit from it. Some reports have also said maybe 12%. But the numbers are

33:05exceptionally low. I don't think it has anything to do with the power of the models. I also don't think it has anything anything to do or much to do with the proliferation of access to models, especially in the enterprise space. I think the number one thing missing right now is actually top-down leadership vision for application of the new toolset within their organization. And it's really, uh, you know, candidly, it's it's a it's a failure to execute at the leadership level to actually decide the what and the where and the how on the tools and

33:35less about the the actual tool itself. The tool itself, while super interesting and it's super cool that like, you know, I got Aragon or I got, you know, Claude 55 or Mythos or, you know, or Astra, it's great to have the debate about those things and of course the Chemist and the Quants and the rest of them as

33:51well. But at the end of the day, the the commoditization, if that's the right word, has happened very quickly as it relates to, you know, these models. The place that they're at, they are already so useful to just about everybody. If you look at the enterprise space and you go into the application points around most of things that you can use them for, so whether it's sales, marketing, product development, you know, finance, uh, and and back office, other functionality, source to pay, etc., all those things, like the tools that are there,

34:20they're ready. They're ready to be used. It's again, it's a it's a top-down issue. And so, I'd actually say that progress as it relates to the model development itself, if it stopped right now, the adoption cycle, we're still in the early innings of it, and the ROI actually still needs to begin to tick up. I actually think that part of the issue of ROI is the fact that no one's really wed to one specific way to solve a problem because we're scared to death if we pick a path that tomorrow the next thing that comes out is going to be better than the thing before, right? The

34:52benchmark that came out on on Aragon about an hour ago said that it was like tops in 15 or 20 categories. Okay. So, if you implemented Claude 5.5 a week ago, are you missing out? Have you like completely missed out on the entire, you know, opportunity associated with using AI in your enterprise? The

35:10answer is no. Of course it's no, right? But that that that kind of FOMO construct I think is is part of the issue. And again, that that goes back to leadership. I think you really need really clearly articulated top-down. It's not CEO, just to be clear, but it's top-down within an organization, top-down within a business unit, top-down within a school. I'm just thinking of all the application points where you could really get maximization of of ROI. So, I I I'm actually, you know, very optimistic still with the use cases there. Now, what happens to maybe the the models themselves and what becomes popular and what becomes unpopular? Sure, like I

35:45think people are going to start making choices as it relates to cost trade-offs and and those types of things, but you know, look, if everyone is still using more AI because it proliferates and you're getting ROI out of it, the they go go back to our initial thesis, like you will have more throughput in the pipes. Even if it's more efficient AI, you will have more throughput into the pipes. And I think that's, you know, further argument for, you know, the the the fact that, you

36:12know, the build-out will continue. Um and and that's just a, you know, sort of maybe maybe basic example, but like one real example of I think what's happening in enterprise today. Maybe in a weird way pacing the frontier would actually be good for these AI businesses. I mean, it's crazy, but like they're like you said, their technology is getting so much better, so much more quickly that it's tough for companies to kind of land on what they should be implementing. And if they took, you know, a a short break, you know, maybe people see what's out there and they they start standardizing around it. And that is what inevitably what eventually

36:47ends up making this stuff, you know, profitable, but 100% they really want to do that. I know they don't. I mean, look at like every you know, every builder of something wants to sell more of it. A quick analogy, like like golf, right? Like Thank you, Dallas. Thank you. newer and newer and newer golf clubs, right? And and the reality is like that thing may allow you to hit the ball a little bit further, a little bit straighter, or what have you. Or you can make your swing a little bit better. And then therefore, you may actually hit the ball a little bit straighter or a little

37:17bit farther. Um I think the model like the model construct and what's like coming out of the models is not that different from like the type of equipment that goes into sports and, you know, goes in um to other recreational, even even hardwired into a lot of, you know, a lot of industry, too. Better tooling, sure,

37:32better tooling makes you more efficient. It doesn't mean that everybody needs the latest hammer or, you know, the latest, you know, ditch digger or whatever might be. Love that. Great way to end. Dallas, thank you so much for coming on. Always great to speak with you. Thank you, Alex. All right. Thanks everybody for listening and watching, and we'll see

37:47you next time here on Big Technology.