WEBVTT - Why Paul Kedrosky Says AI Is Like Every Bubble All Rolled Into One 0:00:02.720 --> 0:00:16.439 Bloomberg Audio Studios, Podcasts, Radio News. 0:00:18.520 --> 0:00:22.400 Hello and welcome to another episode of the Odd Lots podcast. 0:00:22.480 --> 0:00:24.800 I'm Joe Wisenthal and I'm Tracy Alloway. 0:00:25.120 --> 0:00:28.320 Tracy covering the AI boom is actually reminding me a 0:00:28.360 --> 0:00:31.240 little bit of the tariff boom in April, simply because 0:00:31.280 --> 0:00:33.880 every day they are new headlines, like they're just today 0:00:34.040 --> 0:00:38.519 we're recording this November twelfth, Anthropic commits fifty billion dollars 0:00:38.520 --> 0:00:41.160 to build AI data centers in the US. So the 0:00:41.240 --> 0:00:44.479 advanced model companies are vertically integrating more to build their 0:00:44.479 --> 0:00:46.879 own data centers. Every day some new development. 0:00:47.080 --> 0:00:49.760 Yeah, it's becoming pretty hard to keep up. So I 0:00:49.760 --> 0:00:52.400 think we're probably just going to talk in terms of 0:00:52.440 --> 0:00:54.520 billions and trillions. We're just going to say lots and 0:00:54.560 --> 0:00:57.280 lots of money is going into the space. But the 0:00:57.320 --> 0:00:59.840 way I've been thinking about it is, Okay, at this 0:01:00.240 --> 0:01:04.640 everyone agrees that the AI buildout is super expensive, and 0:01:04.760 --> 0:01:08.240 all these companies are spending massive amounts of capex to 0:01:08.360 --> 0:01:12.200 do this, and I'm starting to think that AI capex 0:01:12.319 --> 0:01:16.399 is kind of like the Schrodinger's Cat of markets in 0:01:16.440 --> 0:01:20.400 the sense that it could either be a massive strength 0:01:20.600 --> 0:01:23.920 for these companies because the capex is so expensive and 0:01:23.959 --> 0:01:26.200 it takes so much money to build out, and so 0:01:26.319 --> 0:01:28.320 anyone who manages to do it kind of builds a 0:01:28.360 --> 0:01:32.400 moat around their business. Or it could be a massive weakness, 0:01:32.480 --> 0:01:34.720 right if you're spending all this money and then that 0:01:34.760 --> 0:01:37.640 doesn't end up generating the revenues that you actually need 0:01:37.680 --> 0:01:41.640 to justify it. And going back to the Schrodinger's analogy, 0:01:42.560 --> 0:01:45.000 it seems like we just don't know what's going to 0:01:45.080 --> 0:01:47.720 come out of the box, right, Like it's simultaneously a 0:01:47.760 --> 0:01:51.240 strength and a weakness, and until we build out AGI 0:01:51.400 --> 0:01:53.000 or whatever, like, we're just not going to know. 0:01:53.720 --> 0:01:56.080 I told her, right, there's so much at stake here, 0:01:56.280 --> 0:01:59.520 and obviously we know the numbers are absolutely enormous. They're staggering, 0:01:59.600 --> 0:02:02.720 and we could talk about them too. The financing structures 0:02:02.760 --> 0:02:04.240 are also very interesting. 0:02:04.520 --> 0:02:04.680 You know. 0:02:04.760 --> 0:02:09.320 It's one thing if you just have Meta or Alphabet 0:02:09.440 --> 0:02:11.440 and they make a ton of money already and they're 0:02:11.480 --> 0:02:14.160 spending money on data centers whatever. That's one thing. It's 0:02:14.200 --> 0:02:18.320 another thing when you start seeing these SPVs where the 0:02:18.440 --> 0:02:20.560 hyperscaler puts in this amount of money and then the 0:02:20.600 --> 0:02:22.800 private credit puts in this equity and then they borrow 0:02:22.880 --> 0:02:25.840 a bunch and then there's all these questions about the payback. 0:02:25.880 --> 0:02:28.440 And we think of tech as from years and years 0:02:28.480 --> 0:02:31.280 as basically being this equity story, and when it becomes 0:02:31.320 --> 0:02:33.960 a credit story. Yeah, and when you know people are 0:02:33.960 --> 0:02:37.440 talking about quoting Oracle CDs, I always forget these companies 0:02:37.480 --> 0:02:40.520 even have CDs because I'm so unused to thinking of 0:02:40.520 --> 0:02:43.760 big tech companies as credits. So when I see people 0:02:43.800 --> 0:02:47.440 starting to tweet Oracle CDs charts or core Weave CDs charts, 0:02:47.440 --> 0:02:50.040 It's like, Okay, we are in a different level of 0:02:50.120 --> 0:02:51.560 capital intensity. 0:02:51.160 --> 0:02:53.840 Right, and some of those swaps have been going up lately. 0:02:54.160 --> 0:02:56.960 I'm going to say one more thing, thinking back to 0:02:57.000 --> 0:02:59.880 the two thousand and eight financial crisis. I remember the 0:03:00.040 --> 0:03:03.320 economist at Raymond James I it is Jeff's out who 0:03:03.400 --> 0:03:06.440 went on to become a very big name. Yeah, we 0:03:06.480 --> 0:03:08.600 should have him on the podcast. But he made the 0:03:08.639 --> 0:03:13.200 point that historically when you had real estate crashes property crashes, 0:03:13.600 --> 0:03:16.960 it was usually because of a problem in the economy. 0:03:17.160 --> 0:03:19.080 But then what happened in the run up to two 0:03:19.120 --> 0:03:21.720 thousand and seven two thousand and eight is the housing 0:03:21.760 --> 0:03:25.960 market crash became the proximate cause of the troubles in 0:03:26.000 --> 0:03:28.520 the economy. And if you think about how much money 0:03:28.600 --> 0:03:33.000 is being spent on AI right now again billions, trillions 0:03:33.040 --> 0:03:37.800 possibly of dollars, it's very easy to see how AI 0:03:37.880 --> 0:03:40.760 could borph into a problem for the wider real economy. 0:03:40.800 --> 0:03:43.280 Totally just on this note, and then we'll get into 0:03:43.360 --> 0:03:46.160 our conversation. The Center for Public Enterprises out with a 0:03:46.200 --> 0:03:49.600 great report today called Bubble or Nothing by Ed vat Aarun, 0:03:50.000 --> 0:03:52.440 pointing out one of the things that makes data centers 0:03:52.480 --> 0:03:55.480 interesting is how they sit at this intersection of essentially 0:03:55.880 --> 0:04:00.119 industrial spending and real estate. It's an interesting ascid class 0:04:00.360 --> 0:04:02.280 for its own right. So much to talk about. We 0:04:02.320 --> 0:04:04.600 could never do a justice in one episode, but that 0:04:04.680 --> 0:04:06.640 means we got to do more. Anyway. I'm very excited 0:04:06.640 --> 0:04:08.680 for today's episode. We really do have the perfect guest. 0:04:08.920 --> 0:04:11.480 Someone who's been writing about this for a long time, 0:04:11.640 --> 0:04:14.000 someone who's just been writing about the Internet and all 0:04:14.040 --> 0:04:16.600 things for longer than any of us, someone who's been 0:04:16.640 --> 0:04:20.160 blogging and investing for far longer than either of us 0:04:20.240 --> 0:04:22.560 or anything like that. Way more knowledgeable about how these 0:04:22.600 --> 0:04:25.840 businesses worked, and most very focused on the data center 0:04:25.880 --> 0:04:28.200 buildout we're going to be speaking with Paul Kadrowski. He 0:04:28.279 --> 0:04:31.240 is a fellow at the MIT Institute for the Digital Economy, 0:04:31.480 --> 0:04:38.159 also a partner at sk Ventures, and longtime internet blogger, writer, newsletter, yapper, etc. 0:04:39.000 --> 0:04:42.360 Someone we've never never had on the podcast before. So Paul, 0:04:42.440 --> 0:04:44.320 thank you so much for joining us. 0:04:44.440 --> 0:04:46.880 Hey, guys, thanks good to be here. Other than the 0:04:46.920 --> 0:04:47.599 blogging part, but. 0:04:47.800 --> 0:04:50.839 No, it's all. It's all. You're a true pioneer in 0:04:50.920 --> 0:04:53.920 that and it's impressive that you still write with the 0:04:53.960 --> 0:04:57.320 output that you do. At some point in the last year, 0:04:58.000 --> 0:05:00.240 I feel like you really got laser focused, maybe in 0:05:00.240 --> 0:05:02.880 the last two years, really got laser focused on the 0:05:02.960 --> 0:05:05.560 data center's story is this is where the action is. 0:05:05.880 --> 0:05:07.960 Yeah, I did, and in part just because I caught 0:05:08.000 --> 0:05:09.719 myself by surprise with it. 0:05:09.720 --> 0:05:10.200 It was weird. 0:05:10.240 --> 0:05:12.160 I was looking at first half GDP day it actually 0:05:12.200 --> 0:05:15.719 first quarter GDP data earlier in the year, and you know, 0:05:15.760 --> 0:05:17.760 this has become a commonplace that people know this, but 0:05:17.800 --> 0:05:20.480 I hadn't realized what a large fraction of GDP growth 0:05:20.640 --> 0:05:22.720 in the first quarter data centers were was on the 0:05:22.800 --> 0:05:25.680 order of fifty percent, much larger if you included all 0:05:25.680 --> 0:05:27.760 sort of externalities all the other things that data center 0:05:27.800 --> 0:05:31.240 spending in turn kind of accelerates. And then obviously the 0:05:31.240 --> 0:05:33.080 same thing was true in the second quarter, and it 0:05:33.120 --> 0:05:35.360 was I got back to thinking about my dog, and 0:05:35.400 --> 0:05:36.960 I was my analogy is that. 0:05:36.800 --> 0:05:38.680 As one does, as one does. 0:05:38.960 --> 0:05:41.359 I got to get like my dog barks when the 0:05:41.400 --> 0:05:44.200 mailman comes to the house and keeps barking, and then 0:05:44.240 --> 0:05:47.200 the mailman goes away. And I'm convinced he thinks he 0:05:47.279 --> 0:05:49.600 makes the mailman go away, right, he has this really 0:05:49.640 --> 0:05:52.800 screw causality, and it's like, dude, if you don't bark, 0:05:52.839 --> 0:05:54.720 it goes away. Anyway, this is part of the job. 0:05:54.760 --> 0:05:58.280 They just go away. And I think about macro policy 0:05:58.320 --> 0:06:00.160 in the same way that if you don't understand and 0:06:00.279 --> 0:06:02.960 the drivers of GDP growth, you're likely to think to 0:06:03.000 --> 0:06:05.200 whatever it is you would most like to be causing 0:06:05.240 --> 0:06:07.200 GDP growth is doing that. So in the case of 0:06:07.200 --> 0:06:09.520 the US in the first half of the year, you know, 0:06:09.680 --> 0:06:11.920 was this puzzle was, well, maybe it's terroifts, maybe tariffs 0:06:11.960 --> 0:06:14.120 are actually contributing to it, maybe consumers are much. 0:06:14.040 --> 0:06:15.400 More resilient than we expected. 0:06:15.880 --> 0:06:18.080 And as it turns out, a huge factor, probably the 0:06:18.160 --> 0:06:21.920 largest factor, was this sort of unintentional private sector stimulus 0:06:21.960 --> 0:06:26.000 program otherwise known as data centers, and for me that 0:06:26.080 --> 0:06:28.840 I'll start it. So that started this puzzle of understanding 0:06:29.400 --> 0:06:33.080 this sort of disconmisserate size, the consequences of that size, 0:06:33.440 --> 0:06:36.880 and the acceleration's consequences in terms of where where the 0:06:36.920 --> 0:06:37.960 money is coming from, and all. 0:06:37.920 --> 0:06:38.719 Sorts of other things. 0:06:38.760 --> 0:06:41.760 But just to reframe in terms of something you guys 0:06:41.800 --> 0:06:44.320 were already talking about, and this I think is super important, 0:06:44.320 --> 0:06:47.960 and understanding why this particular episode is likely to turn 0:06:47.960 --> 0:06:49.640 out to be historically really important. 0:06:50.200 --> 0:06:53.679 Wait, when you say you're referred to this podcast episode, 0:06:53.720 --> 0:06:57.040 you're not referring to the broader episode of AI data Center. 0:06:57.160 --> 0:06:58.960 Entirely, just the podcast. 0:07:01.080 --> 0:07:03.480 Who Cares about data centers at the ten year anniversary 0:07:03.480 --> 0:07:06.680 of bad Law. So the reason why sort of it's 0:07:06.680 --> 0:07:08.800 going to be historically important is because, for the first time, 0:07:08.839 --> 0:07:12.160 we combine all the major ingredients of every historical bubbles 0:07:12.160 --> 0:07:14.280 in a single bubble. We have a metabubble no pun 0:07:14.320 --> 0:07:17.400 intended for meta. We have real estate. You guys just 0:07:17.400 --> 0:07:19.640 talked about this, right, Some of the largest bubbles in 0:07:19.720 --> 0:07:22.640 US history had some relationship to real estate. We have 0:07:22.680 --> 0:07:25.880 a great technology story. Almost all the large modern bubbles 0:07:25.880 --> 0:07:27.200 has something to do with technology. 0:07:27.200 --> 0:07:28.120 We have loose credit. 0:07:28.480 --> 0:07:30.880 Most of the major bubbles in some sense have a 0:07:30.920 --> 0:07:34.480 loose credit aspect. And one of the other exacerbating pieces 0:07:34.520 --> 0:07:36.800 that some of the largest bubbles, thinking about even the 0:07:36.800 --> 0:07:40.240 financial crisis, is some kind of notional government backstop. You know, 0:07:40.280 --> 0:07:43.520 think about the role in terms of broadening home ownership 0:07:43.520 --> 0:07:45.320 in the context of the real estate bubble, and the 0:07:45.400 --> 0:07:48.160 role that Fanny and Freddie played and loosening credit standards 0:07:48.160 --> 0:07:50.560 and all of those things. This is the first bubble 0:07:50.600 --> 0:07:52.760 that has all of that. It's like, we said, you 0:07:52.800 --> 0:07:56.040 know what would be great, Let's create a bubble that 0:07:56.160 --> 0:07:58.040 takes everything that ever worked and put it. 0:07:58.000 --> 0:07:59.720 All in one. And this is what we've done. 0:08:00.120 --> 0:08:02.480 Got a speculative real estate component is probably one of 0:08:02.520 --> 0:08:07.440 the strongest technology stories we ever had back to rural electrification. 0:08:07.560 --> 0:08:09.840 In terms of a technology story, we have loose credit. 0:08:09.880 --> 0:08:12.800 You guys talked about what's happening with respect to not 0:08:12.880 --> 0:08:14.840 just the role of private credit, but how private credit 0:08:14.880 --> 0:08:17.840 is largely supplanted commercial banks with respect to being lenders here. 0:08:18.000 --> 0:08:19.560 So we have all of these pieces that have all 0:08:19.560 --> 0:08:21.200 come together at once, and I think in terms of 0:08:21.240 --> 0:08:24.360 framing what's going on right now. It's really important to 0:08:24.480 --> 0:08:28.040 understand that it brings together all of these components and 0:08:28.080 --> 0:08:29.720 ways we've never seen before, which is one of the 0:08:29.720 --> 0:08:31.720 reasons why the notion that we can land this thing 0:08:31.840 --> 0:08:33.560 on the runway gently is nonsense. 0:08:34.360 --> 0:08:37.160 I love that framing the metal babble is perfect. Also, 0:08:37.520 --> 0:08:40.120 I had an epiphany earlier. I already told Joe, so 0:08:40.160 --> 0:08:43.040 you can attest to this, but I realized private credit 0:08:43.640 --> 0:08:47.000 kind of supplanted shadow banking as the term. Right like 0:08:47.240 --> 0:08:49.800 after two thousand and eight, we called it shadow banking, 0:08:50.240 --> 0:08:52.480 and then at some point it flipped to I guess 0:08:52.559 --> 0:08:54.480 the couplier private credit. 0:08:54.559 --> 0:08:57.120 Shadow bank always owned it sinister right away that private 0:08:57.120 --> 0:08:57.560 credit is. 0:08:57.559 --> 0:08:59.560 Well, someone figured that out and they're like, well, now 0:08:59.559 --> 0:09:00.439 it's private credit. 0:09:00.520 --> 0:09:01.640 I like to think of it as a kind of 0:09:01.679 --> 0:09:05.880 financial witness protection program. It was like, oh, you're those guys. 0:09:06.200 --> 0:09:07.480 That's great, now who you are? 0:09:07.920 --> 0:09:10.120 Yeah, it's kind of like that, And it's now like 0:09:10.200 --> 0:09:13.000 one point whatever. It is one point seven trillion dollars 0:09:13.080 --> 0:09:16.560 is the size of which is larger than many components 0:09:16.559 --> 0:09:19.040 of the orthodox lending market combined. In terms of the 0:09:19.080 --> 0:09:21.839 private credit industry itself, so that's a huge new piece 0:09:21.840 --> 0:09:24.520 of this that sometimes escapes notice how big it is 0:09:24.520 --> 0:09:26.520 and why it emerged, So all of those pieces. 0:09:26.640 --> 0:09:28.880 Yeah, it's stunning the growth that we've seen. Let me 0:09:28.920 --> 0:09:31.840 ask a very basic question before we go further. But 0:09:32.160 --> 0:09:35.719 one thing I've been wondering is Joe mentioned that anthropic 0:09:35.760 --> 0:09:39.160 headline that we heard before. We've seen Meta raising financing 0:09:39.200 --> 0:09:42.720 for data center builds, all that stuff. Why do these 0:09:42.880 --> 0:09:47.679 massively profitable and cash rich companies have to raise financing 0:09:48.120 --> 0:09:49.040 at all? 0:09:49.240 --> 0:09:51.880 Well, they don't, but there's these irritating shareholders out there 0:09:52.480 --> 0:09:55.560 get all pissy whenever you start diluting earnings pre shared 0:09:55.559 --> 0:09:57.840 too much and diverting it towards a single source. Now 0:09:57.840 --> 0:10:00.280 that's not the case with private companies obviously, but by 0:10:00.320 --> 0:10:03.000 the same token, open ai doesn't have the luxury of 0:10:03.000 --> 0:10:05.600 having cash flows via which they can do any of 0:10:05.640 --> 0:10:09.200 the things we're describing, so anthropic open Ai and everyone 0:10:09.200 --> 0:10:11.280 else they have no option other than to do exactly 0:10:11.280 --> 0:10:13.560 what we're describing. It's a different story with respect to 0:10:14.360 --> 0:10:16.840 how what percentage of Google's free cash flow or Amazon 0:10:16.880 --> 0:10:19.240 free cash flow that they want to continue to divert 0:10:19.440 --> 0:10:22.040 towards data centers. So in terms of the privates, this 0:10:22.080 --> 0:10:25.000 is the only option that they have. The public's obviously 0:10:25.000 --> 0:10:27.959 increasing the hyperscalers increasingly. We've got up to the point 0:10:27.960 --> 0:10:30.520 where around five hundred billion dollars or fifty percent of 0:10:30.520 --> 0:10:33.600 their free cash flow is going directly towards spending on 0:10:33.760 --> 0:10:36.400 data centers, and that's obviously a point at which you know, 0:10:36.480 --> 0:10:38.120 we have other things we have to do with free 0:10:38.200 --> 0:10:41.440 cash flow, and including having some of it be earnings 0:10:41.440 --> 0:10:44.160 per share, and so we increasingly it's become the option. 0:10:44.320 --> 0:10:46.240 You see what METT is doing recently with respect it 0:10:46.360 --> 0:10:47.119 is SPVs. 0:10:47.400 --> 0:10:50.760 We bring in other participants, create new financing vehicles, and 0:10:50.800 --> 0:10:52.920 then we play this entertaining game of it's not really 0:10:53.000 --> 0:10:53.440 our debt. 0:10:53.480 --> 0:10:55.320 It's in an SPV. I don't have to roll it 0:10:55.360 --> 0:10:56.280 back onto my own. 0:10:56.160 --> 0:10:58.839 Balance sheet and then bring in new lenders, new private 0:10:58.840 --> 0:10:59.760 credit firms and others. 0:11:00.000 --> 0:11:02.200 So that's the reason. Obviously it's partly because of the scale. 0:11:02.240 --> 0:11:04.400 It's probably because the privates who have no other option, 0:11:04.760 --> 0:11:07.400 and it's probably we've kind of tapped out the public 0:11:07.440 --> 0:11:09.520 companies in terms of the fraction of free cash flow 0:11:10.080 --> 0:11:10.920 that they. 0:11:10.760 --> 0:11:14.920 Feel as if they can spend with impunity on these projects. 0:11:14.640 --> 0:11:16.959 Explain to us for those who don't know. You know, again, 0:11:17.160 --> 0:11:19.880 SPV one of these terms that we really haven't heard 0:11:19.960 --> 0:11:22.400 in a while. And there's nothing inherently bad about an 0:11:22.480 --> 0:11:25.720 SPV except that you only hear about them typically after 0:11:25.840 --> 0:11:27.800 there's something, you know, some sort of crazy. 0:11:27.679 --> 0:11:29.640 Ride, which is weird obviously, But yes, tell. 0:11:29.559 --> 0:11:31.679 How would you U say in the broad strokes, how 0:11:31.720 --> 0:11:34.800 would you characterize what these financing vehicles are? 0:11:35.280 --> 0:11:37.959 So Mechanically, it's just a way of making sure that 0:11:38.000 --> 0:11:39.720 I don't have to roll data onto my balance sheet. 0:11:39.760 --> 0:11:42.960 But legally it's a structure into which I and my 0:11:43.040 --> 0:11:46.439 partners contribute capital that in exchange for which they retain 0:11:46.559 --> 0:11:49.600 legal title to the project that we've created, which allows 0:11:49.679 --> 0:11:52.280 us to all contribute capitalists but not have to put 0:11:52.280 --> 0:11:54.160 it back on my balance sheet and therefore not to 0:11:54.200 --> 0:11:55.359 have that debt rated. 0:11:55.640 --> 0:11:56.920 Which is really the key. 0:11:57.040 --> 0:11:59.520 Now, if you look at the actual intrinstics, say, for example, 0:11:59.559 --> 0:12:01.640 the reason that a project that they did in conjunction 0:12:01.720 --> 0:12:04.720 with blue Out, it's wild and byzantine. It looks like 0:12:04.760 --> 0:12:06.400 something you might have seen and what was that in 0:12:06.440 --> 0:12:08.240 Harry Potter or the forest with all the spider webs. 0:12:08.240 --> 0:12:09.720 It looks a little like that, right where. 0:12:09.559 --> 0:12:11.640 Everything's connected to everything and all I know is something 0:12:11.640 --> 0:12:14.160 and here's going to get me. So there's incredible complexity, 0:12:14.200 --> 0:12:16.680 but at the core, it's a mechanism via which I 0:12:16.679 --> 0:12:18.720 can raise more capital and keep it off my balance 0:12:18.800 --> 0:12:21.280 sheet by creating a legal entity that controls the actual 0:12:21.360 --> 0:12:23.559 data center and I don't. Therefore I have to put 0:12:23.559 --> 0:12:26.400 it back, roll it all back onto my balance sheet, navierated. 0:12:26.920 --> 0:12:31.320 Now there's weird intricacies obviously, So for example, what happens 0:12:31.440 --> 0:12:34.400 if at some period in the future this thing isn't 0:12:34.440 --> 0:12:37.120 performing the way we expect who owns it at that point? 0:12:37.320 --> 0:12:38.960 Is there a payment exchange, does. 0:12:38.840 --> 0:12:41.200 It become metas, does it become blue ouls, does it 0:12:41.200 --> 0:12:45.000 become someone else? And these things will turn out to matter. 0:12:45.080 --> 0:12:46.840 Right now, no one cares. If you go through some 0:12:46.880 --> 0:12:49.040 of the documents on these things, it's not entirely clear 0:12:49.480 --> 0:12:51.600 what the recourse payment will be when it ever, if 0:12:51.600 --> 0:12:54.120 and when it ever has to revert back to another owner, 0:12:54.160 --> 0:12:55.760 and it's not going to be held on to by 0:12:55.760 --> 0:12:57.440 the SPV. And I think this will turn out to 0:12:57.440 --> 0:12:59.480 be really important four or five years down the road, 0:13:00.080 --> 0:13:14.200 but right now nobody cares. 0:13:16.960 --> 0:13:21.199 So Number one, the lifespan of data centers is actually 0:13:21.360 --> 0:13:24.440 not that long. I can't remember the exact estimate, but 0:13:24.480 --> 0:13:27.000 maybe like three or four years something like that. And 0:13:27.040 --> 0:13:30.160 then also you have this risk that tenants are sort 0:13:30.200 --> 0:13:33.240 of rolling through and no one knows what that actually 0:13:33.280 --> 0:13:35.200 means for the structure of the debt, and you kind 0:13:35.200 --> 0:13:37.080 of get this asset liability mismatch. 0:13:37.760 --> 0:13:40.280 Yeah, so I'll start with the first one first. So 0:13:40.520 --> 0:13:42.760 this gets into something Michael Berry was tweeting about the 0:13:42.800 --> 0:13:46.319 other day, which was sort of entertaining that back about 0:13:46.360 --> 0:13:51.120 four years ago, tech companies changed the appreciation schedule or 0:13:51.160 --> 0:13:53.040 the assets inside of data centers. 0:13:53.240 --> 0:13:57.200 They extended them somewhat. Now, that wasn't an error. 0:13:57.240 --> 0:14:00.520 The reality is that data centers used for the purposes 0:14:00.559 --> 0:14:02.400 like at aws, where You've got a big S three 0:14:02.440 --> 0:14:05.160 bucket and I'm storing data inside of it. Those things 0:14:05.280 --> 0:14:08.280 generally speaking, the assets are long lived. I'm not running 0:14:08.280 --> 0:14:10.880 them flat out, it's not. These are not streetcar racers 0:14:10.880 --> 0:14:13.200 that I'm running around inside of a data center. These 0:14:13.200 --> 0:14:16.640 are relatively inexpensive chips that I'm using for really mundane 0:14:16.679 --> 0:14:20.680 purposes like storing large amounts terabytes exhibites of data inside 0:14:20.680 --> 0:14:23.080 of s three buckets, so it's not unreasonable to say 0:14:23.080 --> 0:14:26.040 their lifespans fairly long. They're not being taxed that heavily, 0:14:26.160 --> 0:14:29.160 so pushing out the depreciation schedule makes a lot of sense. 0:14:29.240 --> 0:14:32.640 But that was coincident with the emergence of GPU driven 0:14:32.720 --> 0:14:36.000 data centers using products like the chips from Nvidia, and 0:14:36.040 --> 0:14:39.200 those have much shorter lifespans, so depending on the usage. 0:14:39.200 --> 0:14:39.720 So there's two. 0:14:39.640 --> 0:14:44.400 Different reasons why the lifespan and therefore the depreciation schedule 0:14:44.440 --> 0:14:46.760 of a GPU inside of a data center is very different. 0:14:46.880 --> 0:14:49.560 So the reason most people think about is, oh, well, 0:14:49.640 --> 0:14:51.720 technology changes really quickly and I want to have the 0:14:51.800 --> 0:14:53.400 latest and greatest, and therefore I'm going to have to 0:14:53.480 --> 0:14:58.320 upgrade all the time. That's important, but it's probably about equal, 0:14:58.360 --> 0:15:01.600 if not maybe slightly less important the nature of how 0:15:01.640 --> 0:15:04.200 the chip is used inside the data center. So when 0:15:04.280 --> 0:15:07.560 you run using like the latest, say a Nvidia chip 0:15:07.600 --> 0:15:10.160 for training a model, those things are being run flat 0:15:10.160 --> 0:15:12.480 out twenty four hours a day, seven days a week, 0:15:12.480 --> 0:15:14.720 which is why they're liquid cool. They're inside of these 0:15:15.080 --> 0:15:17.520 giant centers where one of your primary problems is keeping 0:15:17.560 --> 0:15:18.120 them all cool. 0:15:18.600 --> 0:15:20.880 It's like saying I bought a used car and. 0:15:20.840 --> 0:15:22.720 I don't care what it was used for. Well, if 0:15:22.760 --> 0:15:24.440 it turns out it was used by someone who was 0:15:24.480 --> 0:15:27.240 doing like Laman's twenty four hours of endurance with it, 0:15:27.480 --> 0:15:30.160 that's very different. Even if the mileage is the same 0:15:30.200 --> 0:15:31.280 as someone who only drove to. 0:15:31.320 --> 0:15:32.200 Church on Sundays. 0:15:32.360 --> 0:15:36.720 Right, these are very different consequences with respect to what's 0:15:36.760 --> 0:15:38.920 called the thermal degradation of the chip. The chip's been 0:15:39.000 --> 0:15:42.440 run hot and flat out, so it probably it's useful. 0:15:42.480 --> 0:15:45.400 Lifespan might be on the order of two years, maybe 0:15:45.400 --> 0:15:48.680 even eighteen months. So there's a huge difference in terms 0:15:48.720 --> 0:15:51.800 of how the chip was used, leaving aside whether or 0:15:51.840 --> 0:15:53.600 not there's a new generation of what's come along. So 0:15:53.760 --> 0:15:56.920 that takes us back to these depreciation schedules. So these 0:15:56.960 --> 0:16:00.920 depreciation schedules change, just as the nature of how the 0:16:00.920 --> 0:16:04.360 lifespan of the chips changed dramatically, because I can use 0:16:04.360 --> 0:16:06.560 something for you know, storing things in s three buckets 0:16:06.600 --> 0:16:08.720 for a long time, six to eight years isn't unreasonable. 0:16:09.200 --> 0:16:12.760 But if I'm doing the the Laman's endurance equivalent with 0:16:12.880 --> 0:16:16.200 a GPU, it might be eighteen months. That's a huge 0:16:16.240 --> 0:16:19.000 difference in terms of the likely lifespan of a product 0:16:19.040 --> 0:16:21.840 that I'm depreciating over a very different period. And so 0:16:21.920 --> 0:16:24.280 that's a huge part of the problem here with respect 0:16:24.320 --> 0:16:28.440 to understanding the intrinsics in terms of how data centers 0:16:29.040 --> 0:16:31.360 can and can't make money. How you have to think 0:16:31.400 --> 0:16:34.560 about the likely capex requirements because of this much shorter 0:16:34.640 --> 0:16:37.000 life span of the underlying technology, and then. 0:16:36.840 --> 0:16:40.400 Talk about the tenancy rollover risk. I guess we might 0:16:40.400 --> 0:16:40.760 call it. 0:16:41.200 --> 0:16:44.760 Yeah, it's really interesting. So one way to think about 0:16:45.160 --> 0:16:48.160 data centers is as giant apartment buildings. Right, They're essentially 0:16:48.200 --> 0:16:50.560 gigantic commercial pieces of commercial real estate with a bunch 0:16:50.600 --> 0:16:53.560 of tenants. Sometimes there's a lot of tenants, sometimes there's 0:16:53.560 --> 0:16:55.920 only one. Sometimes Google bought the whole apartment building and 0:16:55.960 --> 0:16:57.760 just moved in, Or it's a giant office building they 0:16:57.800 --> 0:16:59.840 just moved in. It's all theirs, right, So think about 0:16:59.880 --> 0:17:02.320 it in those sorts of terms. And the reason why 0:17:02.400 --> 0:17:04.960 as a sponsor of a data center I might take 0:17:05.000 --> 0:17:07.159 a different view on how many tenants I want is 0:17:07.200 --> 0:17:09.720 again you think about it in terms of what can 0:17:09.760 --> 0:17:11.919 I get Google to pay? But whereasus what can I 0:17:12.000 --> 0:17:15.000 get someone who's a much flightier tenant to pay? Well, 0:17:15.040 --> 0:17:17.399 I can get the flightier tenants, more of them and 0:17:17.480 --> 0:17:21.440 diversified as all leasing inside the data center, paying higher 0:17:21.560 --> 0:17:24.800 lease rates for GPUs over the period of tendency than 0:17:24.800 --> 0:17:26.840 I can get a Google to pay. Why because Google's 0:17:26.840 --> 0:17:28.880 got great credit, they don't have to pay very much 0:17:28.880 --> 0:17:29.600 and they know they don't. 0:17:29.600 --> 0:17:31.280 So if you look at the commercial real estate. 0:17:31.200 --> 0:17:34.520 Data, the cap rate, the blended cap rate for these 0:17:34.560 --> 0:17:37.320 for the largest data centers that are tenanted by hyperscalers 0:17:37.920 --> 0:17:41.000 is horrible. It's like four point eight five point three percent. 0:17:41.119 --> 0:17:43.920 It's like, why don't you just buy a treasure you're doing. 0:17:44.160 --> 0:17:47.440 So what happens then is people start blending in more 0:17:47.440 --> 0:17:50.359 different kinds of tenants to Tracy's point, as an effort 0:17:50.400 --> 0:17:53.240 to try and improve the yield the cap rate on 0:17:53.600 --> 0:17:56.240 the underlying instrument, which is the data center. So you 0:17:56.280 --> 0:17:58.440 could do all of this should start to sound familiar 0:17:58.520 --> 0:18:01.040 because it's this idea of a blend together all of 0:18:01.040 --> 0:18:03.359 these different tendencies. I can increase the yield of the 0:18:03.400 --> 0:18:07.479 securitized instrument, but that also changes the risk profile of 0:18:07.520 --> 0:18:09.400 what comes out at the other end, which just takes 0:18:09.480 --> 0:18:12.920 us to things like the increasing usage of these things 0:18:12.920 --> 0:18:15.639 in asset backed securities, which are these trench securities that 0:18:15.720 --> 0:18:18.359 have all the different pieces, We have different layers associated 0:18:18.400 --> 0:18:21.399 with it, and that's a reflection of well, there's different 0:18:21.440 --> 0:18:24.280 tenants inside these data centers, and people want different exposures 0:18:24.280 --> 0:18:26.240 to risks. So I may only want to buy the 0:18:26.280 --> 0:18:29.440 senior tranch. You may want to buy the mezzanine and trace. 0:18:29.520 --> 0:18:31.560 He may want to buy the equity charge. 0:18:31.800 --> 0:18:33.959 Can I just say, I know we already said this, 0:18:34.359 --> 0:18:39.719 but Paul is truly, truly the perfect guest. I remember 0:18:39.720 --> 0:18:43.359 reading his coverage of subprime and securitization in like two 0:18:43.400 --> 0:18:45.560