A (5:04)

All right, so I want to about the rationality of it. I have participated in one of these bubbles, as you know well and certainly a lot of the decision making then to me did seem rational and then it ended in this huge collapse. But before we do that, you did say that in these build outs or enthusiasms there are some winners and then some losers. So in within the A I build out who are, who are the winners and losers.

C (5:30)

So the losers are almost certainly most of the major frontier, a model frontier AI model companies. They're almost certainly going to be losers that most of them will probably disappear, be replaced by small language models, consolidate something else. I'm on the record as saying I don't think there'll be a chatgpt 8 I don't think OpenAI will be around long enough just to do that. But nevertheless. So those companies are going to be commoditized. It's already happening. They have no mode. It's highly capital intensive. The pace of developments is slowing and we can get into that. But these are the unlikely ones. And then everyone who's essentially subsidized and building on top of that stack, if you will. So this collapse collapsing, and I think this is an important element here, but the collapsing token prices that are the underlying sort of atomic units of this new economy, if you will. Anyone who's building on top of that and a recipient of that subsidy is an interesting place. So it's not clear to me yet who those people will be. It may turn out to be robotics companies. I don't think it's likely to be some of these thin wrapper companies we see like people who are building lawyers in a box. I think that's commoditized nonsense. And again, I don't think it'll be anyone. Almost no one in the current Mag 7 strikes me as a likely winner here. Not Nvidia, not OpenAI. Possibly anthropic only because they're oriented much more towards companies. But that's the thing that I think is very typical of this moment is that you're at the build out and it's not. Wasn't obvious at the build out of say railroads who were going to be the big shipping companies. No, wasn't at all obvious. And instead it just led to a series of economic crashes. And eventually out of that emerged some really interesting companies that were transporting goods over large distances. And the same thing was true with rural electrification or and the same thing was true with the telecom build out. It wasn't obvious that Netflix was going to be a big winner. It took time for that. And so the mistake people make is they want to anoint someone who's a current sort of mag7 company to say, oh, they're the winner, and I'll take the other side of that bet all day long.

A (7:21)

But I mean, you're connected with all of the folks who are putting just astronomical amounts of money still into OpenAI and anthropic and all the others. There's this big race. And what I hear is, look, one of these guys is going to be the Google. We got to own it now, and the rest are toast. You just don't think so.

C (7:39)

No, I don't. And I think that that's, it's a classic mistake. You and I have been around these cycles so many times. People always want to fight the last battle when it comes to geopolitics. And the same thing is true in tech. They want to anoint the next Google because that was the thing they missed in the last cycle. And I swore to myself I would never let that happen again. And it just doesn't play out like that. These mirror echoes partly don't play out in the same way because it's not the same businesses. But an infrastructure build out like this is not analogous to what happened in the dot com period. Not at all. As I say, it's much more like rural electrification in the 1920s where the main beneficiaries were sellers of refrigeration and air conditioning decades later. Now that's not to say that whoever benefits this time around won't show up for decades. It'll all happen much faster. But it won't look like General Electric or it won't look like someone who was building out telephone poles in North Dakota. That's the wrong way to think about what happens when you build out a new substrate on top of which people build companies based on a collapsing commodity price. And in this particular case, then it was power. In this particular time, it's this, these, these units we call tokens. Not the crypto kind.

A (8:47)

All right, so for those of us who are not totally embedded in the jargon, don't really know what tokens mean and that kind of thing, I, I, I think I know what you're describing, which is basically revenue units. Ultimately, I pay Chat GPT the same as when I first subscribed two years ago or, or whatever it was. What is it that's declining in that? Because all I hear is that they're losing more and more money, in fact more and more money on every search or every question and so forth. So what if, what's, what's actually declining?

C (9:16)

So tokens are the representation of the underlying data that dates back to, I mean we've had tokens for probably 15 or 20 years now in machine learning, but they're the atomic unit of how you capture the data. So think about them in a sense as words, right? So loosely. They're words loosely. They're all of these units of information that's in my training data that in turn. And whenever you ask a question, and if you want to think of these things as giant auto complete engines, the tokens are the things that are being generated. So they consume them, they're trained on them to generate frontier models, whether it's OpenAI or whether it's Claude or whatever else, but they also produce them. So they consume and produce these things called tokens. So what's happening is initially the cost of training models was very, very high and it stayed high, but for kind of weird reasons. But the cost of inference of producing tokens is what's in full, full outright coll. And that's for a bunch of different reasons. Partly competition, partly improved efficiency in data centers, partly because of technological changes, partly because this particular technology went from lab bench to billion users faster than any technology in history. So as a result, our ability to produce these autocompletes, if you will, to your prompts will continue to get cheaper and cheaper and cheaper.

A (10:33)

Isn't that good for the companies though?

C (10:35)

Well, it's horrific. It's good for the company if you can grow commensurately. So literally just to stand still, still they have to grow at more than 3,000% because of this thing. Imagine being General Motors and the cost of cars is falling 60% year over year. I better find some new planets to sell cars on because I'm in big trouble selling on this planet. That's the problem.

A (10:55)

So it's me, the ChatGPT user, who's getting way more tokens for the same amount of money every year. That's right. Basically that's right. Okay, so it's not that their cost to produce them are plummeting and therefore their profit margins are rising or lack of profit margin.

C (11:08)

No. So there's this tremendous benefit to you as a consumer, which is great It's a transfer of wealth, right, being transferred to you and you're able to do more with them. But it's a lousy business from the standpoint of being the producer of these things because they're facing this exponential declining cost with respect to the units on top of which they generate their revenues and sell it. And so now let's go out a level and think about. Frontier models don't run in isolation. They run on these things we call data centers, these great big buildings full of liquid cooling and humming Nvidia chips and rack planes and so on. And most of those are funded from equity or funding from cash flow from large tech companies. But increasingly, and it's really took off last year, more and more of it's being funded by debt. Well, if that's the case, think about what happens with respect to my debt payments in future that are tied to this center, the revenues of which are driven by token economics. I don't get to shrink my interest payment in terms of what I'm paying back to Blue Owl or whoever extended the loan to me that I'm using to pay for the data center on the basis of collapsing token prices. You don't get to do that. So the economics of data centers were bad in the first place. In first place and getting worse because of this collapsing token economics problem. So all of this stuff going on under the surface makes the business of being in the data center side of things, the frontier model side of things, difficult, if not impossible because you're trying to build at this kind of pace with high capex on top of a commodity whose prices are collapsing.

A (12:38)

And so where are we in this whole process? If you look at some of the eras that you're talking about, and usually there's anywhere between five and seven years of this huge enthusiasm where the early folks make tons of money and grow really fast and then everybody piles in, feel like it's going to go on forever, then ultimately it crashes and then we rebuild from there. Like. Like where are we now we in 1997 for the telecom build out? Are we?

C (13:02)

So this is going to happen? Yeah, that's a great question. So that's going to happen. Even this is going to happen even faster because we just don't know how to do things slow anymore. And in part because the nature of this thing is that it connects things up and accelerates the pace at which people produce new products and companies. And all you have to do at things like Claude Code or Claude Cowork to see how people are this kind of accelerationist phase we're going through. So I think it'll all play out very quickly. And I think my guess is we're probably in early 99 in terms of. My guess is that this year we'll probably see. And we've seen this already in China to a degree. The first few truly speculative AI IPOs just came out over the last three months in China and both of them did moonshots. And so you will see this year the first non Frontier Co. IPOs. I think you'll probably see at least a dozen. It could be two dozen. And there the retail investor appetite for those is immense, in part because of what's happened recently with Mag 7 stocks, but also because that's the nature of the beast. People become really fatigued with quote, to use the Wall street parlance you and I know too well, with old names, they want new names. Very exciting. Just get me new names. And everyone wants new names. And so the stock market, the equity sides, the sell side is nothing if not terrific at providing new names when asked. So you want new names, I will get you new names. And so we'll see a host of those new names emerge this year, which will be very spec of, most of which will fail. But that will also begin to soak up some of the risk appetite in here, much like happened in 99 when we had dozens and dozens more than 200, 300 IPOs at one point. Forget the numbers now, huge numbers. And so that'll begin to soak up some of the risk appetite at the margin, which will pull some of the capital out of the Mag 7, which will make it more obvious that there's going to be real winners and losers among them. And so that natural evolution is already beginning. There was a little bit of a regime change, a sea change in sentiment about the Mag 7 stocks in October. And that's really continued up to now. Like if you look at the X Mag, like the S&P493 and compare it to the Mag 7, I love that this is S&P493. Anyways, compare the two of them. The X Mag is up about 2% to year to date, a little more. The Mag 7's off more than 5%. And you can trace that back to not just this year, but to the October change with respect to how people began thinking about how should I reward companies that are spending this much? Oracle, others Meta in terms of the capital spending that they're doing. Because we went through this phase, right? That was much like the, the Crypto treasury companies where you were rewarded for spending, for doing something silly and the crypto treasury companies, my valuation went up by $2 if I put a dollar's worth of crypto in Treasury. The same perverse phenomenon was happening early on with capex. The more I promise to spend and dilute you as a shareholder and soak up cash flow, the more you rewarded me. So I know what to do whenever that's the case, if I'm a board of directors, I spend more money. So they began spending much more money. But that moment changed in October when people stopped being rewarded for exc of capex. And so there's a these things are all coming together in answer to your.

A (15:57)

Question and do you think that's a permanent change or is it cyclical? And again, going back to the dot com bubble, there were pretty much every year from 1995 to 2000 you had lots of incredibly smart people like you coming out and saying, oh this is ridiculous, a bubble, it's going to end in disaster. This is also silly and often we would get a 30% correction in the big stocks and then they double and triple from there and then we get another one. So do you feel like we're done? Was that the turn?

C (16:24)

Yeah, I think that was the beginning of the turn. And this year will mark the change when we rotate out of the people who are building the infrastructure. And there'll be a second like an echo bubble if you will, in the app related services that sit on top of this AI fabric that we've built out. But I think the days of being rewarded for grotesque amounts of capex and how many gigawatts you promise to build in like New Mexico or whatever the case may be, the days of being rewarded for that are largely over. And there's some very good reasons for it, not least of which is that the other thing that's unusual about this bubble is this moment is that we're investing into a collapsing price commodity that wasn't true in the dot com period and wasn't true in telecom. So think about that. The token prices on average have been declining 70% plus year over year for at least three, four years now. The math's pretty simple in terms of what that implies. If I want to produce a make Wall street happy growth percentage, percentage of growth over the next few years, let's say I want to produce 30% growth on a compounding annual basis over the next five years to keep Wal happy. Leaving aside, what happens with earnings, just revenues, if I'm doing that as a frontier company or say a data center, a public data center provider, think about what I have to do. If my underlying commodity, the thing that I'm selling that's driving revenues in one form or another is declining 50 to 70% year over year, that's unprecedented. So the implication is, let's say I want to do 30% year over year. And the underlying token prices are falling 60% year over year. The five year growth required is around 10,000%. Hello. That's a faster growth profile than Facebook in its best five years, than cloud services ever, than mobile phones ever. So while it's possible, it's so wildly unlikely that you at least should be a little more skeptical about the likelihood of producing that kind of unit growth, that would justify the capex that requires that amount of growth.

A (18:23)

Well, the holidays have come and gone once again, but if you've forgotten to get that special someone in your life a gift, well, Mint Mobile is extending their holiday offer of half off unlimited wireless. So here's the idea. You get it now, you call it an early present for next year. What do you have to lose? Give it a try@mintmobile.com Switch limited time.

B (18:43)

50% off regular price for new customers. Upfront payment required $45 for three months, $90 for six months or $180 for 12 month plan taxes and fees. Extra speeds may slow after 50 gigabytes per month when network is busy See Terms.

C (18:57)

Right now in the world of AI, two things are happening simultaneously. One, the technology is getting better fast, people are finding new uses for it, it's more popular than ever. And two, every company that makes AI is absolutely hemorrhaging cash. On the Vergecast this week we're talking about what OpenAI and other other companies.

A (19:15)

Are doing to try to finally figure.

C (19:17)

Out how to make some money off of this technology. Spoiler alert, it's mostly ads and we're.

B (19:23)

Talking about whether any of it's actually going to work.

C (19:25)

All that, plus some stories about the Chinese company that appears to be beating Tesla on the Vergecast.

B (19:31)

Wherever you get podcasts. This week on Net worth and Chill, I'm breaking down the essential money tips for every stage of your relationship from the first date to forever. Who actually pays on the first date? How do you split rent when you move together if one of you makes way more than the other person? And yes, we need to talk about prenups. Plus, I'm sharing why I believe in equity over equality when it comes to splitting costs. Whether you're single and swiping moving in with your partner or planning your wedding. This episode will help you navigate the uncomfortable money conversations that can make or break your relationship. Listen wherever you get your podcasts or watch on YouTube.com YourRichBFF so you've written.

A (20:15)

And talked a lot about just how deep this extends into the economy in terms of construction and, and the percentage of GDP growth and everything else. So give us a snapshot of that and, and what happens if that starts to reverse or if we start to max out here, which is, I assume, what ultimately ends the cycle.

C (20:34)

Yeah, well, one would think, but that's being rational. I've learned not to be rational. So yes, I'll go with that. Let's take that as. As the. As a given. So, yeah, so first half of this year, data center related capex writ large, was about a little more than half of US GDP growth, which is a really big deal because in the first half of the year, and certainly in the second half as well, there's been so much speculation in the analyst community, especially with respect to strategists, about why the US Economy is behaving so strangely. Job growth is tepid, and yet GDP is holding up. And so what's the disconnect? A huge chunk of the disconnect was this misunderstanding of the nature of what was driving GDP growth. More than half of it was coming from this anomalous spending on the single thing that we call data centers. And so the reason why I bring this up is not just because it's like, good party talk. It's because if you don't understand the things that are driving GDP growth, you're likely to come up with any random idea about what's driving GDP growth. Maybe you'll decide it was tariffs. Maybe you'll decide it was my own personal charm. Maybe you'll decide it was something else when the reality was it was this one anomalous thing. This disconnect in our understanding of what's driving the current macro cycle is a huge source of risk because you're likely to double down on things that are actually risk increases, like, say, for example, tariffs or whatever else, not realizing that what's driving this is actually this anomalous spending on GDP growth. So, and again, now diving into that, you can say this is like 60% of the cost of data centers is chips, but there's lots of other things going on, right? There's architectural design firms, there's construction. Roughly 30%, give or take, is cement now, which is ridiculous. I mean, so there's people making super cycle arguments that we're entering the era of perpetually growing cement demand. I mean it's really, it becomes madness, right? So you get, you people are finding their own favorite bull cases across all the underlying commodities that create these things we call data centers and their derivative products.

A (22:29)

And so what ultimately stops it? So in other words, you're painting a picture where if the demand drops and if the, whatever the real spending driver here drops and it doesn't, it stops filtering out into all other areas of the economy, it's going to hit a lot of the economy. What stops it? What have you learned from other enthusiasm cycles like this?

C (22:49)

It's the roi. What happens is the market event says you're making a huge investment that's no longer just being funded out of cash flow. And I didn't like that in the first place because that was dilutive to me and EPS was declining. And so fine, you wandered off and began doing all this spending using increasingly amounts of debt and that's being provided by private credit providers and others. But that in turn also increases the risk here because now you've got some risk with respect to the stability of your cash flow because that in turn is being driven by this collapsing price. Your obligations are tied to this collapsing price commodity. So the bottom line is the ROI is now getting huge focus in terms of here' the capex and here's the likelihood of me being able to deliver a competitive return on that. Which gets back to the point I made earlier that say 30% year over year compound 5 year growth would require this sort of like around 10,000% unit growth greater than the growth than we've seen in any, under any comparable tech product in the history of such tech product. Doesn't mean it can't happen. It just means that suddenly people are actually being much more, I'll say careful about it. And I see this increasingly in data center spend. I see this increasingly as I talk to hedge funds and others who are active in this area and are actually looking at it. There's a lot more attention being paid to how does this generate a competitive ROI that justifies the valuations that we're seeing? And the answer is I don't see it.

A (24:08)

And for a very simple example I've been watching for years, the OpenAI financing numbers go up in terms of the amount they're raising and the valuations. Amazon IPO was very controversial. Everybody said, oh, it's a money losing Internet bookstore, who cares, it's worth zero. But when it went public it was only 400 million. This is like this tiny little number. And so if they got a lot of things right, you had a huge upside from that. I look at OpenAI now, last headline I saw was they're raising money at $800 billion valuation.

C (24:38)

Yeah. As a dropping. Similarly. Yeah, yeah.

A (24:41)

As a professional investor, which you are, what kind of return do you need to see to justify that? What are you expecting?

C (24:48)

So there's two different answers to that. One is that I don't care as long as retail investors give me 4x on the IPO. So. Right. I mean, there's a very cynical answer to this, which is if I can unload at the IPO at a significant premium to what I paid for it in the last post, the last pre IPO financing round, then I don't really care what their underlying unit economics look like. That's a problem for retail investors in future. And I think it's really important to keep that in mind because that's one of the reasons why people got this all wrong during the dot com episode was that, oh, you know, Amazon will never generate a competitive return. That mattered later. That mattered later for holders post ipo. But early on, what mattered more was, how does this perform and can I flip my shares? So we're going to go through that phase first. And so the fact that calculating what a reasonable rate of return that they'll have to generate in terms of a cash flow that would justify the valuation I pay in a series double z round on OpenAI is kind of irrelevant in the sense that most of the people buying here are buying to try and get rid of it, to unload it post ipo.

A (25:52)

So you said a few minutes ago, which I think is the key to all of this, that most of the folks making the decision think that they are rational decisions.

C (26:00)

Is that true? Oh, absolutely. That's the thing that's really funny to me is that I get asked all the time by some of the hedge funds I consult. It's like, they must know inside. This is ridiculous. It makes no sense. He's like, can we find someone like a whistleblower who says it's fraud? I'm like, dude, you don't understand how people drink the Kool Aid inside these companies. Not only do they not think it's fraud, they actually think they're changing the world. So you'll never find that whistleblower who says, last week in a meeting at OpenAI, I heard Sam say, boy, oh boy, are we ever pulling one over on those rubes at Blue Owl. That's not happening. No one's doing that. These are true believers. That's the nature of tech. Sure, you'll find people who are screwing around with really speculative data centers. There's a few out there that are like politically driven, that make absolutely no sense and are probably very cynical. Inside of the large frontier companies, these are uniformly true believers in terms of what they're doing. So you will not find somebody who's going to tell you that, oh, this is all just dodgy nonsense that we're doing to screw people up over.

A (26:59)

And so if the decisions are rational, because I think you can go to lenders too, you can certainly go to the investors. You explain the logic there, which is, oh my goodness, I missed Google last time. I got to get the Google this time. I'm not going to. It's not going to be me who misses it. They all make sense. And I think you can look at OpenAI and say, look, you know, guys, if we have any chance of doing this, we've got to be the one that crashes through here and makes it so spend absolutely everything we can. We've got the ability to raise money so it looks very rational. So given that, can it happen any other way? I mean, you've been, you've seen these cycles, you're an amazing writer, so you make fun of everybody as you go. You're so sardonic about it. I mean, it's like, look at all this stupidity. But is there another way that we could be doing this, these innovation cycles?

C (27:48)

Oh, I don't think so. That's one of the, and that's Carlotta Perez and lots of other people's arguments that these kinds of cycles require require grotesque overspending at this point in the process. And all you can do is try to either stay out of the way or caution people that, you know, here's how this part of it is likely to end. But you don't know if you've built enough until you've built too much. And I think that's really key to understanding. And people who come into this thinking that there's some kind of rational, clean way of doing this sort of thing are naive. And not because you know the animal spirits or whatever else, but it's because you just don't know. You don't know when you've built built enough till you've built too much. You don't know when you've got enough supply of tokens until you've got too much. I mean, I'll give you an example. People often say oh, I'm hugely bullish about AI because ChatGPT is super useful to me. And I'm like, that's terrific news. I'm glad it's really working out for you. But I said, how important do you think consumer chat is to the AI story? And they're like, well, it's huge. Everyone I know is now using Claude or ChatGPT or whatever else. And I said, so that's inference traffic. So, so that piece of inference traffic you think is really important to the story and is driving your bullishness? And they're like, yes. I said, okay, fine. I said, how much of current global inference capacity is required to satisfy consumer chat? And they are usually off by a factor of 10. They say, oh, you know, 60, 70% of all of the data centers out there. It's five. It's less. Well, it's actually a little less than 5%. You could satisfy all of global inference, consumer chat related global inference from a single data center in Northern, in Virginia. All of global inference. We are not the story here, you and I. As much as we want to be part of this as good narcissists, I want to be part of the story. I'm not part of the story. The story is this build out for training and for something else that they're hoping will come along. And so we're already at a point where the level of overbuild is titanic in service of things that have yet to be identified. So we're already at that moment, if you think about it in terms of, of the relevance of the thing that makes most people bullish, which is the utility of consumer, of consumer chat with respect to these apps.

A (29:58)

Okay, so I want to actually restate that because I think that people, it's so easy to get lost in the jargon of the training and inference and everything else. So what you're saying is the actual product that I'm consuming when I use ChatGPT or Claude, when I ask a question or I ask it to do something, that's the inference piece and that actually uses a single data center in Virginia and that's it.

C (30:19)

It could is my point. It uses. It's not. My point is that we've got like 5,000 plus data centers out there. We could satisfy it all from the racks of a single day in Northern Virginia. It doesn't happen that way. But the point is to drive home how much excessive supply we have already if the world is moving towards infrared. So then the question is, what the hell is going on inside of all those other data centers, is it like laser tag or hide and seek or whatever else? And about 60% of the traffic in those data centers in terms of the workload, training. And you're like, oh, that's terrific news. So you're spending all that money to make, you know, better things. Better and shinier things. For me, that makes me very happy. And of course, then you get into the thing. Well, no, that's actually not the problem. Not what's happening. It's training loosely, but most of training is just experiments, which is to say the large frontier companies are trying want to create better models, but they're at a bit of a log jam where we've kind of hit an architectural cul de sac, a dead end with respect to the progress of these models. And you can see that in how people react to new models. Like, oh, this gpu, I can't tell the difference. It's not that much better than five. We've all become like Yelp reviewers of these things, right? We were like, oh, this is only four and a half stars. This isn't as good as I expected. And that's. But if you look under the hood, what's really going on is we're reaching a natural point with respect to the richness of the underlying training data that's used to produce these models, that it's taking longer, producing more marginal improvements and costing more to produce the things that are being created. And so that's the moment that we're at now. And those models, those are running at these data centers, but that are scaled up in anticipation as if we will be able to do this forever. That's a huge error. That's not going to happen. And it's already. It's already. We're already seeing that with respect to the. To the models themselves.

A (32:06)

And this is something I was very struck a few months ago when Gemini, Google's thing, came out and suddenly got better ratings than ChatGPT. And all I can say is that never happened in the 1990s. Barnes and Noble never caught Amazon. You could say what you want about Amazon, but they were way ahead and accelerating. It's just simply not happened here. And you've gotten really smart. People like you and others in the tech industry are saying, look, the LLMs are commodities now and you can use them interchangeably, which is very striking.

C (32:39)

It's really striking. And you even heard yesterday. Was it yesterday? Davos Satya Nadella, the Microsoft CEO, was making the point that we live in a multimodal world and we need to build fabrics that can allow our users to mix and match and so on. And that's all really great, but it's like, dude, you're like a huge investor at OpenAI. You just said one of your largest investments is kind of meh to you. I mean, what the hell? This is a really unusual moment because you're essentially saying that something in which you've got a huge stake is commoditized, so you're placing bets across the landscape because you don't know whether or not any of them will actually matter. And that went by like people said, said, huh, Nice to see Satya Davos. And no one was saying, wait a minute, this is a remarkable statement that this is so thoroughly commoditized, that one of the largest investors is diversifying across models. Yes, but that's your point.

A (33:30)

Yeah. And you've got to say for Microsoft, what an incredible investment. They figured out a way to take tens of billions of dollars of cash sitting on their balance sheet doing nothing and turn it into revenue because they get a huge royalty off of all the revenue that ChatGPT generates. So it's tremendous.

C (33:46)

Although that expires in the early2030s, but.

A (33:48)

Nevertheless, yes, we're worried about the short term. So Satya Nadella, he sort of, he was asked on the last call, you know, hey, dude, like, everybody's talking about a bubble, like, what are you doing spending all this money? And his response was, we cannot keep up with demand. And it's not just about chips. We've got racks of chips, we can't even get power to them. There's an incredible demand from what you've just said, which I think is incredibly important for everybody to understand, is most of that demand. Demand is for training new models that we may not need, or where they.

C (34:23)

Are not just won't need, that are actually incremental in commodities. Like, I defy you in a Coke, Pepsi sense to tell the difference between most of the models other than Grok, which is regularly fairly offensive. So of the non offensive models, I defy you to tell the difference between them in a blind taste test. That's a really big deal. So, yeah, okay, fine, they have no choice. So they're continuing to do it. Because Oscar Wilde, the writer, had this tremendous line years ago where he said, a man's second marriage is the triumph of hope over experience. This is a little bit of that, right? This is the triumph of hope over experience. I know my last training round didn't produce a model with all kinds of wild new functionality, but maybe this time it will. Well, that's back to the Oscar Wilde line, right? So it doesn't really work that way. And the second point I'd make is that one of the illusions underlying the current wild expansion that Satya is citing in terms of demand is that, okay, fine, 5% of traffic is consumer chat and 60% of overall traffic is training. That's still a lot of unaccounted for traffic. What's the biggest piece of that? It's coding. So about 85% of inference traffic is coding. And that's exploding because developers increasingly are like, ah, screw it, I'll just throw all my code in here and say, you know, do this and I'll check back periodically on the tab and see how it's doing. With respect to producing code, coding is two things. One, models are very, very good at doing code, but they're also incredibly wasteful at doing code because you have to ingest the entire code base every time to make incremental improvements. So it's like having to ingest an entire book every time you want to read the next paragraph. So this is really important. Important, but that's as a signal with respect to demand in the marketplace. It's not exogenous, it's not external to the market, it's internal to the market. Because the largest users of coding tools are inside the frontier AI companies themselves. So the notion that they're radically expanding their use of AI IDEs development environments, and that in turn is a very wasteful use of tokens, and that's creating the impression that token demand is growing. That's not an external signal that doesn't validate anything you're doing. And then more fundamentally, and you know this because we've both been around this stuff for so long, I defy you to name a large cap tech company in the software development tools business. And there's a reason for that. Selling to developers is a really crappy business. So the notion that I should therefore extrapolate a giant multi trillion dollar market on the back of it of huge inference traffic flowing from developers is wrongheaded on at least to at least two bases.

B (36:52)

New Year, new me. Cute. But how about New Year, new money? With Experian, you can actually take control of your finances, check your FICO score, find ways to save and get matched with credit card offers, giving you time to power through those New Year's goals. You know you're going to crush start the year off right. Download the Experian app Based on FICO Score 8 model offers an approval. Not guaranteed guaranteed eligibility requirements and terms apply subject to credit check, which may impact your credit scores. Offers not available in all states. See experian.com for details.

A (37:20)

Experian Shopping is hard, right? But I found a better way. Stitch Fix Online Personal styling makes it easy. I just give my stylist my size, style and budget preferences. I order boxes when I want and how I want. No subscription required. And he sends just for me pieces plus outfit recommendations and style styling tips. I keep what works and send back the rest. It's so easy. Make style easy. Get started today@stitchfix.com Spotify that's stitchfix.com Spotify all right. So Paul, how is this going to end? And I will sort of provide context of in the end of 1999 I was very aware that it was probably a bubble.

C (38:05)

Yeah sir.

A (38:06)

And what I couldn't do in my my head was figure out how we were going to get from this moment of total euphoria where a business plan could raise whatever they needed to do. They would go out and spend all the money to this wreckage that everybody was talking about. I couldn't figure it out. Then it became obvious. The financial interest rates went up, the financial markets closed. Suddenly there was no money that everybody could spend anymore. And then all of these different things that were supporting the fundamentals just collapsed and we had the wreckage. So what happened happens here.

C (38:36)

So I always fall back. And there's an old line from a member of the Council of Economic Advisors during the Nixon era. I think it was Herb Stein who said that things that can't go on must stop. Which is like a ridiculously tried statement, but actually applies in all of these contexts and in particular in this one. It applies because there's so many ways in which it can fail. It can fail because we have a moment where debt markets are no longer as fluid in terms of providing financing for the more speculative data center providers. It can fail on that basis. It can fail because we have have 100 IPOs this year or over the next two years that dilute the risk capital from retail investors and suck all the air out of the frontier companies. So the way I look at this is a it must end because that's what happens with all of these capex frenzies. They always end the same way, which is to say by an oversupply something like 50% of the rail lines ever built were abandoned. Similar was true with canal, similar with true with rural electrification. Most of the dot com fiber sat fallow for a better part of a decade, that will be true here as well. So knowing that, trying to time it and say, oh, you know, it's going to end in six months because of this and this and this, all I can say is we're closer to the end than the beginning. Could it go on for two more years? Sure, it could go on for two more years. And is it possible that in the middle of all of this, we have a kind. I had someone argue to me the other day that we'll have like, the uber of tokens emerge that'll be using, you know, trillions of tokens. And all of the arguments about declining, you know, collapsing token prices and that these frontier companies will never get enough appetite don't matter, because there'll be autonomous robots exploring the world and consuming. Consuming it on the basis of tokens, and they'll be streaming it all back, and that'll require quintillions of tokens flowing back and forth. And I'm like, you're like that old joke about there's like an economist and a physicist and somebody else who gets stuck on a desert island, and the physicist says, I'm going to drop. And they want to open up a can of food. And the physicist, I'll drop it out of a tree. And I forget what the chemist says, but the economist says, I'll just assume a can opener. I'm like, dude, you just assumed a can opener. You just assumed away the problem. This is a really bad way to approach these kinds of moments. So the way I look at it is, roughly speaking, we're in this moment. It always turns out the same way in these capex frenzies. And it'll end with declining roi, or at least an appetite for finding out what the actual ROI is on this stuff. That's already begun to happen. As I was saying with the XMag and Mag 7 stuff and the sentiment change in the fall, and even look at the credit default swap numbers for some of the proxies, like, for example, Core Weave or look at Oracle, and you'll seeing all of this, the sentiment's already begun to change. And so there'll be a quick flip of it to retail that'll soak up a lot of the risk capital, and that gets you to maybe 18 months from. From now at most.

A (41:12)

And who's going to get hurt?

C (41:15)

So that's an interesting question. So the weird thing in this particular moment is that it's really metastasized, to use the, you know, the oncology term. It's spread across the entire marketplace. This little, this cancerous cell, if you will. So you know, there are insurers who are providers of capital with respect to data centers that are going to be heard. And then on the other side you have the obvious candidates. Right. I think, I think OpenAI is probably in the weakest position with respect to being one of the people who survives the great shakeout. With respect to frontier models, I think obviously Google and Microsoft are in much better positions. Anthropic is to a larger degree because I think they're more focused on the enterprise market, which is probably a better place to be than trying to farm engagement and convince people that your model loves you and wants to take you home if you're in the more retail side of the chat market. So these are some of the people I think who are sort of obvious winners and losers. I think in the, in the longer run, Nvidia isn't a winner here at all. I mean they're kind of a mini computer provider as personal computers emerged, if you want to think about it in historical computing terms, they're very monolithic, expensive, power hungry chips. And people misunderstand the role of Nvidia chips here. I use a highway toll booth analogy here. So Nvidia is the toll booth provider that's trying to chew through this line of people who are coming across a bridge and all the lanes are funneling into the toll booth. But if I have 100 LA down to 2 or 3, it doesn't matter how fast I make the toll booth provider. So Nvidia. So you're seeing this already where Nvidia will announce that it's got, you know, I have petaflops of performance and people are just like, yeah, don't really care. And the reason is because that's no longer the gating factor. The world has changed from 2023, the gating factor are the number of lanes, if you will, feeding the toll booth. And that's memory. And that's why companies like SK, Hynix and SanDisk and others are doing so well just because it's already changing the nature as we move towards an inference world about where the blockages are. So I don't think in the long run Nvidia, which kind of stumbled into this in weird sorts of ways because this just happened to be really well attuned to what they were doing. So I don't think they're a long term winner here either. And for some of those reasons that increasingly it's about memory storage and high Bandwidth, memory in particular, not about whether or not my toll booth runs faster than yours.

A (43:29)

And so from what you're saying, it goes way beyond just, just the enormous, incredibly cash risk rich tech companies like Microsoft and Facebook and Google that are spending hundreds of billions of dollars collectively. Amazon too. Because I think for a while people did feel like, hey, you know, the money's not coming from individual investors, it's coming from these companies that make $40 billion a year. They don't care, they can lose it. And we do know they can lose it. You look at Mark Zuckerberg burned tens of billions of dollars a year on the Metaverse, which was a hallucinated nation, nobody cared. Ultimately he said, ah, shut it down, we'll write off 80 billion or whatever it is. So I assume that that's what Microsoft will do, just write it off and nobody cares. The stock will go up because they finally got the bad news out. But what you're saying is it goes way, way beyond that.

C (44:17)

Oh, way beyond that.

A (44:17)

Into the economy. Okay, right.

C (44:19)

And even in more actually somewhat sadder ways, which is I often talk to regional economic development people who are offered the opportunity in air quotes to have data centers built in their region. And so these you think about, I'm a small town, I'm a regional economic development official, I bid on the Hyundai battery factory, I bid on the assembly plant for Honda, didn't get any of those. I've got 40%, 30% unemployment. And along comes, fill in the blank company who wants to build a data center. And they make this great story about how these are the factories of the new industrial revolution. And all I need from you is a tax abatement and subsidized water, subsidized power, and I'll be there for 20 years with high quality tenants and it'll create permanent jobs. But when you do the math, almost inevitably on these things, the cost on a per job basis is usually in the millions, given the subsidies inherent in constructing these things. And yet they're being built all over the country in this kind of speculative fervor. And so there'll be a lot of regional economies hurt by this that end up with these white elephants, data centers that are either underused or unused, that were built at the peak of the frenzy of building out the cycle. And we're done for good reasons. Regional economic development officials have an impossible job. And when given this and told that this will be in used for decades, how do you say no to that? And so these are. So the damage goes much more broad than that. Which is why I think it's important to get away from saying, oh, you know, Microsoft's good for it. This is a mistake.

A (45:44)

All right, let's assume that this vision of the future is right, that there's going to be this big crash, wreckage, the press is going to pounce, all these books are going to be written about how stupid everybody was and how could they believe these things and so forth. And, and yet from that there will be green shoots and then hopefully a massive AI powered economy will be built just the way the Internet was, just the way railroads were. And you have been very optimistic about that and we've been painting you into.

C (46:13)

This, you know, oh, I'm going to.

A (46:15)

Be the curmudgeon and I'm going to tell you that all this stuff that looks so great, it actually sucks and people are being stupid and all that stuff. So tell us the positive vision of the single.

C (46:24)

Oh, yeah, absolutely. This is literally the single most important technological change in my lifetime, and probably one of the most important, probably dating back to maybe like air conditioning or something else in terms of really changing the way people work and live. Think, if you think about it in those terms, it's like antibiotics in healthcare, it's like air conditioning and urban life. It's like that. It's this fabric that really radically changes the nature and cost of everything we do. So in that sense, hence turning from what's going to happen to these random fools who are building giant data centers in locations that it can't justify, that's kind of irrelevant in the longer run. In the longer run, what's going to be much more important is that cognition loosely is going to become free. That's what we're really. So the atomic units of cognition in an AI sense are these tokens. Tokens. Prices are collapsing. They will be ubiquitous everywhere and essentially costless. So they'll underlie almost everything you do. That will have immense job transformation consequences and it could cause huge disruption in the white collar workforce. But on the other hand, the opportunities inside of science, inside of research, inside of medicine, inside of all these areas that were previously very difficult to plumb because the complexity and the interactions inside of it made it hard. All you have to do is look at the cost of drug discovery, the cost of launching almost anything in medicine, but across a host of these fields, suddenly all of these open up because the data and cognition essentially became costless and free. And one of the ways my partner and I like to look at the world in terms of our own investing is what happens if something that was previously scarce and expensive becomes free? And that was true about bandwidth, that was true about CPUs and processing, it's been true about physical distance back in the days of railroads, it was true about power back in the days of overall electrification. And it's true now in the sense of cognition. What happens if cognition becomes free and again something that was expensive and scarce becomes cheap and ubiquitous? It's lousy for people in grad school. I will concede that right away because these are again predicated on cognition remaining scarce and expensive. But it's tremendous society wide in terms of unlocking. It's a cheat code for unlocking cognition for a huge population of people who might otherwise never have access to all of the tools that were otherwise locked up in seven years of grad school or locked up inside of years of working in a particular domain. I talked to my plumber the other day and he often has his phone sitting on his shoulder filming things as he does them and getting feedback on a continual basis from a vision model. He uses ChatGPT for this. It's not built for that, but it's a general purpose technology that's providing costless cognition and that's incredibly valuable.

A (48:58)

And so what about all the folks in grad school and the job transition? I mean there have been some very dire predictions off and out of Silicon Valley about oh, nobody's going to have a job anymore and so forth, which would lead to societal collapse and so forth. What do people do?

C (49:13)

So I think people will do what they've always done, which is they recognize that the thing that I've been training myself to do in a longer is a thing. And that's unfortunate and it's often sad and it's often multi generationally consequential for those people and I take nothing away from that. After the Industrial Revolution, the UK got richer on GDP per capita basis, but individuals in general were poorer for almost 30 years. So you wave that away at your peril because think about that in the US terms, the US gets richer and most people get poorer. Are people going to say up with AI and I love everything that's going on, or are they going to say let's burn it all down? So we have to be very careful to, to say even though there are gains coming, there have to be mechanisms for making it easier for people to gain access to this. And I'm not a believer that you do that by training people to become AI impresarios. I think you do that in part by Just dodging this altogether. You say, why aren't you in trades and vocational schools? Why aren't you in things that are less in the bullseye of what's going on right now and more importantly have huge numbers of openings. I had a water commissioner in a relatively small town. The other tell me he had more than 100. No, he had 25 openings, none of which paid more than 100, paid less than $100,000 for people who were going for technicians vocational jobs within his town. And he's like, I can't get anyone to apply for any of these. And this is the problem is that we've got this huge misallocation of people towards occupations that are shrinking and even going to go away and away from things that have large numbers of openings, stable incomes and a huge appetite for people to do them. And for whatever reason societ, we've turned up our noses at those things. And I think that's where everything has to change is recognizing that that's where the opportunities are. Not in the old world of saying if I just train long enough I can eventually become a managing director at Goldman Sachs.

A (50:59)

I'm going to ask you to end with a very optimistic picture of what humanity and progress and everything else looks like in the2030s. So if I understand you correctly, what you think is happening is that basically things that it took human beings five to ten years in grad school to learn are now going to become available to everybody at their fingertips and that will create lots of opportunities in medicine and so forth. But paint that picture for us. What does that mean? Does that mean, as some optimists are saying, that suddenly GDP growth is going to radically accelerate and we're going to cure all diseases and so forth? What does it actually mean?

C (51:34)

Yeah, I don't think it means that. So I think there's a lot of relatively low hanging fruit in terms of things in materials and medicine and drugs and so on that'll suddenly take turn out to be tractable like solvable problems that weren't previously solvable. Because we can deploy these tools that are very good at analyzing this giant amounts of data and finding trends and patterns and helping us develop things that deal with our problems. So I think we'll see a step function improvement in those things. But at the exact same time there are other countervailing forces that I think will hold back economic growth. The biggest one is demographics. Most developed economies are shrinking in and so population is going to be one of the factors. And there's been some good research on this that while say, AI related productivity improvements might take us out over the next, say 10 years, maybe add maybe a little bit, maybe 2% to productivity growth, on the other side, we have a shrinking and aging population which will probably subtract one to one and a half percent from that. So the gains are there to be had, but in some sense they're being offset by changes in demography in the underlying makeup of the U.S. population. And that's not just true in the U.S. it's true in Japan, it's true in Western Europe and so on. So, but that doesn't make this less important, it makes it even more important because we're trying to compensate for changes in the nature of the workforce and the population of the United States. And there's some great data showing that there's huge advantages to jumping to some technologies earlier. Like for example, Japan has one of the highest levels of robots per capita in the world. And that's in part because it's one of the oldest countries in the world. So weirdly, an aging society, Tom, sometimes leaps to new technologies that help solve the problems faster than others would be. So I think a US in the early 2000 and 30s will be one that's much more robot intensive for that reason, as a compensation for the changing demography of the country, that rather than having mass immigration or whatever else, we'll have an increasingly number of robots. And I think you're seeing the very beginnings of that right now. And it's scary to people, but if you think it through, it's actually hugely important because unless you're willing to have like 10 babies over the next few years, it's not going to change the underlying problem in terms of the nature of the US workforce. So I think that's incredibly important, that this is an offsetting force to huge headwinds that we're already in place and we're not going to be able to do anything about. So I'm really optimistic about some of the developments coming, but I'm also optimistic that it'll help offset the demographic headwinds that might have otherwise been just dire for the US over the next 10 years.

A (54:02)

Paul, it's always fascinating and a pleasure to talk to you. Thank you so much for joining us.

C (54:07)

Yeah, thanks, Henry. It's great.