Transcription
We have this incredibly bloated financial sector. It's a phony crisis that they don't have to be cuts. We know how to get out of a great depression. You spend money. If we take a look at the typical worker, they're not doing fine.
Hi, I'm Dean Baker, your host for Mostly Economics. Today we have with us Ed Zitron. Uh who's the CEO of ECPR and probably the the the best known critic of of AI, the biggest uh I advocate the right word, but the biggest warner, let's just say, of a bubble in uh in the market. So we're going to talk about AI, the bubble, and where it's going to go. So thanks for joining us, Ed.
>> Thanks to you for having me.
>> So let me start. We'll get right to it. Um, you know, I I I haven't looked today, but uh, last I looked, uh, the market cap of Nvidia was around 6 trillion. So, should I buy?
>> That's the thing. We're in this kind of another era of irrational exuberance because maybe, but today, as we speak, don't know when this will go out. It's Nvidia's earnings and the tech Super Bowl and bunch of people who don't really care about tech, including the people who cover Nvidia, will speak about Nvidia and say, "Wow, what an amazing company." They're probably going to beat and raise. They're probably going to be analyst estimates. They'll raise again. And the most worrying thing is is that every time this happens, the worse it gets because I have done the research. I've spent hours and hours and hours doing this. There is no way in hell that we have installed even half of the GPUs sold in 2025 in and nor do I think we will for several years. It's not that there's it's not anything to do with demand. It's the fact that data centers take two to four years to build even smaller ones. So Nvidia is in this irrationality sphere where it kind of always gets bit up. It always has the beat and raise cuz if it doesn't well the whole market relies on it. So right now what they're at 220 something which is really weird because there's so many signs that something is wrong but the markets are now fundamentally disconnected from fundamentals and there's obviously something strange going on there. So it's really hard to say when to stop buying Nvidia. I don't give stock advice, but it's like it's this weird situation where nothing about the company really makes sense when you think about it on very practical terms.
>> Well, let let me go on a couple things here. First off, I'm totally sympathetic to the saying you can't say when to stop. And I, you know, I'm old enough, I think I a few years on you, but to remember the tech bubble. And, you know, I was talking about it and going, "This looks really bad. This looks really bad." And, you know, for my part, not any great sum, but I had a 401k. pass 401k and I took it out in March of 98 cuz I go this is nuts. This is nuts. Well, it kept going up for two years. I mean I wish I could have said I took it out in March of 2000 but you know at the end of the day you know it was the right call but it was two years too early.
>> At least she didn't take out March 2002.
>> Yeah. No, exactly. So at the end of the day I end up being very happy. Although yes, you could always do better, you know. But anyhow, so so I'm totally Yeah. I don't, you know, I I I didn't expect you were going to tell me, give me a date when the video is going to turn, you know. So, yes. But if I can ask you to explain a little bit because I I kind of understand what you're saying, but I I think a lot of uh a lot of people we might have watching this when you're saying that they haven't installed uh the what what they've sold. So, if you could clarify,
>> okay,
>> you know, what exactly is going on there?
>> Happy to. So, a data center, an AI data center. You hear the if you hear someone say we're building an AI data center that's a gigawatt in capacity that has never happened. They have been started. None of them have ever been finished. The closest we've got is Stargate Abene which is this OpenAI Oracle mess out in Texas. Two buildings out of eight. It's been 2 years. AI data centers take forever. They take 18 to 24 months. Nvidia's Blackwell GPUs really started shipping at the beginning of 2025 because there was a heat related issue that delayed them. So you got the situation where Jensen Hong claims 3 million Blackwell Nvidia GPUs shipped. Now that is about 4 GW of capacity. Just stick with me. There is no way in hell that much capacity has come online. In fact, no one has a good handle on how much capacity has come online. Now, the reason I'm so sure of this is I spent hours and hours and hours finding things that had broken ground and then trying to find any proof they'd opened. I found four of them that had opened and they were teeny tiny ones. They were like 42 36 megawatt. So almost every data center you've ever heard of is not finished. They're being built but they're not being finished because construction is difficult. Putting aside the fact that the power isn't available. There's a shortage of electric steel, shortage of transformers, shortage of talon, shortage of like materials that are now more expensive because of tariffs. Putting all that aside as well, it's just very tough to build these things. The fact they're building them in Texas with all the weather problems is also a problem. So on in very practical terms, Nvidia is clearly selling GPUs years in advance, but they're not disclosing anything about it. They're trying to avoid talking about the fact that I think this is just an assumption of mine that they're doing something called bill and hold which you might remember from the com bubble where they say great here's a they sell $50,000 GPUs that we're going to sell a few billion of those oh you don't have the data center ready no problem mate we'll put them in a warehouse for you'll have to sign this NDA so that no one knows that that's happening but that's what you need to do because of the irrational exuberance people are also desperate to get ahead. They're desperate to put as many of these GPUs in their own stock, even if they're years ahead of them. There are economic problems with that as well, which is, okay, so in 2 years, you're going to have a 2-year-old chip and these chips are replaced on a yearly cycle. Ver Rubin is coming this year sometime. That isn't great. So, we're in this perpetual cycle of eventually standing up years old chips. And there are some Blackwell chips going online. There are some, but not that many. Maybe a couple hundred thousand, maybe a million max. But that still means the vast majority of them are sitting in warehouses. Nvidia has not disclosed this. And Nvidia beats and raises every time. And if they're doing this by pre-elling GPUs, you've got two worrisome things. One, will the companies buying them still be around by the time they need to pay for them or by the time that they need to pick them up? Two, can they cancel them? Because that's exactly what happened with Cisco, Lucent, Nortell, all sorts of equipment manufacturers during the com bubble. Selex were canceling. Everyone was canceling because why would they need this? And people were intentionally buying more than they needed knowing that they could cancel those orders. There's something up here.
>> Yeah. No, that makes a lot of sense. And you know, in terms of the the buyers now, as I understand, they also have lost circular finance that very often Nvidia is,
>> you know, basically paying for the for their own product. They're they're lending the money to uh to the companies that are buying up their their product.
>> Oh, it's even dumber than that what they do. So, Cororeweave Core is a great example called a NeoCloud. A Neocloud is a fancy way of saying we're a company that buys GPUs and puts them in data centers. Sometimes we build them, sometimes we lease them. Coreweave, they were invested in by Nvidia. Nvidia is one of their largest customers. Nvidia propped up their IPO. Nvidia has invested at least $2 billion in postIPO equity. They've done this with Lambda, Iron, Google has done this with Terowolf and Cipher. I mean, this should be illegal, just to be clear. Like, this 100% should be illegal. There's not a no, we need regulation on this. But yes, Nvidia has spent tens of I mean, they invested $30 billion in OpenAI. Nvidia is keeping itself alive by feeding the money in there. And I wish we had better members of the media to say, "Hey, a healthy industry doesn't need this." Instead, we have some of the most brainless, corrupt, craven sellside and buyside analysts who are just going, "Oh, this is the big it's a super cycle. It's a super cycle. Look at all the stuff that's being sold." Not real not actually bothering to look at the other end and see if the revenues are there, which they are most decidedly not outside of OpenAI and Anthropic, who I think are doing dodgy accounting.
Well, it sounds a lot like uh going back to the days of the dotcom bubble and also the uh the uh housing bubble. I mean, you had, you know, again, it was uh you know, both of them I was going this is crazy. This is crazy. The housing bubble was more going it's crazy crazy because it was just at least to me very very visible. I mean it's and and you know people were just saying you know ungodly when I'm saying people I mean like reporters at major news outlets were just saying ungodly stupid things. I don't know how many times I saw people say, "Well, you know, housing's not like like a dot stock because you could always live in your house." And you're going, "Yeah, but how many people could afford to live in a house that they paid 400,000 for that's now worth 200,000?" I mean, like, what the hell does that mean?
>> Yeah, there's um the equivalent of that, it's not like for like, but there are many similarities. The equivalent with that is there right now there are data center capacity constraints because as I said, they're not being built. People are conflating that and saying, "Wow, it's hard to get a GPU right now to rent for my AI services." They're conflating that with the idea that there's unlimited demand. That they're saying because it's because it's hard and expensive to get compute right now, that means that there's unlimited demand versus the fact that we just have a constraint problem. The other thing is is these data centers, they're building them, they're building them, and they're saying, "Well, we have guaranteed tenants. We have tenants who have agreed to take this up whenever it gets built." Two problems with that. Number one, most of these contracts have an out. If you miss a milestone, if you don't make it by this date, we no longer have to take this. Microsoft has one of these with Nebus, a big Neocloud. I think it's if they don't hit certain production milestones, Microsoft can walk away. The other thing is as well, great. Wow, you built this house for these people that said they'll buy it. Sorry, messed that one up. You built this data center for an AI startup. Brilliant. Cool. They're going to use it. Quick question. What happens if they run out of money? What happens if they can't pay you? What happens if nothing happens? What happens if Cuz Oracle right now, for example, is building 7.1 gawatt of data center capacity for OpenAI company that cannot afford to pay them. And Oracle also can't afford to build the data centers. But everyone's saying, "Wow, holy, holy crap. Oracle is going to get $300 billion over five years. Wow. Wowzer. That's like more than their annual revenue every year. Wow. This is going to be crazy. One little problem, OpenAI doesn't have the money. Other problem, Oracle doesn't have the money to build the data center. So, I have to get a bunch of debt. And I keep talking to I talked to economists, I talked to journalists, I talked to academics, and they all say, "Well, Oracle will get paid." I'm like, "How? Open AAI loses billions of dollars a year. Well, they'll raise more money." is your only answer. And it literally is, well, they'll just raise money from someone. It's rich kid logic. It's straight up rich kid logic. It's just, well, daddy's credit cards never been declined before. And on the more macro scale data data center thing, when you actually look at the data centers and you look at how many are being built, there's apparently like over a 100 gawatts of capacity that's been announced. Only 15 gawatt actually under construction. that accounts for I think it's like 100200 billion of annual AI revenue AI compute revenue. Yeah, when you remove anthropic and open AI, there's maybe a billion of AI compute revenue max. The customer base doesn't exist. But when you try and talk to people about this, when you try and just have a genuinely rational conversation, they get so angry. They get very defensive. They get very much like, well, what do you mean? Nvidia is selling all this stuff. Corning just signed a deal with Nvidia. Like, how do you mean? Oh, Marvel. You've got um Micron, they're all doing crazy. AMD, they're all going crazy. The parabolic sales. What do you mean there's no market for this? And the answer is because the AI companies need to rent the GPUs and the AI companies are all unprofitable. How do they pay for them? And no one really has a cogent answer for this other than nu.
Let me ask you a little bit about that end product here because this is you know I mentioned when we were before we came on about China and I you know again you're way way more into this than I am but you know I I read some of the press and it looks to me like you know China's AI I'm talking end product here is way cheaper like a fifth a tenth the price of US AI and again just going on what I've seen in some articles in the press it sounds like you know China's just eating our lunch when you know you look at the rest of the world and apparently even to some extent in Silicon Valley which I mean to me this seems important because at the end of the day you need someone to buy the stuff and if the Chinese AI is a fifth a tenth the price who's going to buy the stuff that you know our guys are selling am I missing something here or is you know
>> I mean the thing you're missing is that the end product isn't that impressive because try fun game to play for anyone here go and try and find first of all revenues for AI very difficult. You hear you get annualized revenues which is a classic.com bubble scam where you do month times 12. But another thing to try is try and find productivity benefits of AI. In fact, just try and find what someone's doing with them. You'll get a bunch of people saying, "Wow, I'm doing all my coding with this." Then go and look at the 10 Q 10K. See if you can find any kind of economic benefit. You won't be able to. And they'll say, "Well, we've been automating this thing. Goldman Sachs signed a deal with Claw with Anthropic for Claude." and you're like, "Great. What does that mean?" No one will tell you. The problem is is that no one has really worked out yet what this thing can do. Everyone's got very excited about it. No one's really sat down and be like, "Wow, okay, what is the actual productivity and economic benefit of this?" And there really isn't one. You have engineers claiming they're writing no code anymore. What does that mean? Like just I don't know if I agree with that being unilateral, but take that face. Okay, what does that mean? because you're not replacing the engine is because you can't trust this stuff and it's not getting better. It's not improving to the point when you can leave it. It's never going to. Okay, great. So, that getting back to the China side, the reason that everyone's a bit nervous is because no one can really find the productivity benefits in a tangible sense. The Chinese models being cheaper is scary. Kimmy K2, I think, is one of the models from over there. Wow. Yeah, that model's pretty good and it can do some other things. It's not as good. You can run it on device. That's competitive. But again, what does this result in economically? Revenue, any kind of like tech, no one can really tell you. The one thing you can do, however, is go and look at the Chinese companies that actually discuss the AI revenues. Miniax losing over $200 million a quarter. Zepoo, I think, doing the same thing. Uh Alibaba doesn't seem to be gaining anything off of AI other than very large capex charges. When you actually go and look at the numbers, the numbers are bad. The thing that they're everyone in America is scared of with China. They're going to say, "Oh, well, they'll they'll have access to powerful AI. They'll have access to powerful Nvidia GPUs. They already got them. They already smuggled them over. There's a There is a short seller report Culper. They say that they think that Nvidia has been selling directly to China using intermediaries. China already has access to all of this. China is already there. What's happened? I don't know. I really don't. There are people I get emails from every day who are like, "Well, I think you're underelling the fact that people are doing AI coding now. Oh, I was able to pull together a report and the report was accurate." And I'm always I'm I'm always like, "Okay, really? Is that it, though? This is Everyone's talking about this all the time. And the best you can say is I got a maybe accurate report and I did some coding for an indeterminate economic output for it is not clear what this we're not getting a flood of good news software. We're getting a lot of slop but we're not getting it's not like any of these companies are better. It's not like you can point to any of these AI pill companies and say wow Spotify is so much better. Wow into it. They're really their AI stuff's making it better. In fact across the board they're pretty bad.
>> So you're way more negative on this stuff than I am. I thought I was negative. I you know in my experience I've had very little. It was just direct experience. I a few times had it you like get I remember my biggest success was I had to write a ref reference section for me. So I was writing a paper and I had to you know reference sections pain in the ass you know to you know and I I'm a bad typer so I make a lot of mistakes. So I
>> like formatting
>> excuse me
>> like formatting the reference section.
>> Well well no and finding the sources. So I said you know you know particular economist and you know find the papers he did on a particular topic and put in a reference section and it did that for me and I had to go through there because there were a few that I wouldn't use and you know whatever but you know it probably saved me two hours three hours of work you know and probably did a better job because I make as I say I make mistakes so you know anyhow I'm not going to say that this changes the world but I was happy
>> but that's but that's actually a really good point. So if what you were describing was an industry that was selling itself as a modest software play, I wouldn't argue with you. If you were say if this was a if they were saying this is a 30 billion TAM market using specialist compute capacity, if every 3 months we didn't have to do a a Super Bowl party for Nvidia's earnings. If this was just Yeah, it was able it's kind of bit rough but I was able to do my references with it. I assume you check them all also just in case of hallucinations.
>> Then I'd be like fine what you were describing is a better Google search
>> or a more more easy to use Google search or what have you. Google search has been a mess since uh 2019. I think that was when things went wrong there. But nevertheless, if you were describing a modest software play and people were talking about this in a normal sense, I'd be like, "Yeah, sure. I find that inoffensive." That's not what's happening, though. people are people would hear that people would use that story that you just told which is perfectly reason nothing wrong with that I mean the fact you checked it is good formatting is annoying this is the kind of thing that computers are meant to fix
>> yeah yeah
>> but I'm sure I could have found software that would do that too. prei I don't know what
>> maybe you couldn't have maybe this is something MLM are good at formatting a page that's fine I find that inoffensive the thing is people would take that example and they would extrapolate from there and say, "Well, next time in like a year or two, you'll just do that and it'll write the whole paper and it'll check the references and everything will be perfect and then it will publish for you and it'll do it all of these insane tendrils out from a relatively pedestrian feature, like a useful one, but a pedestrian feature. I also, by the way, I find the research of these things pretty damn poor. I think the only thing with LLMs that really impresses me and impresses is kind of a little bit of an overstatement is that they are better ingesting what I'm searching for than regular Google search that they can understand if I say find me this this this and this. It's the natural language processing is better. Again though that's not how this is being sold. This is being sold as the change the change of everything. Everything will be different. IC the very nature of society will change. You have people ostensibly smart people who are talking about this like we are mere years from artificial general intelligence an entirely fictional concept of a conscious compute.
>> Yeah. No, I I I see that and I I Yeah, I'm sympathetic to that. Also, you know, again, I I'm struck because I I have all these I don't if argument is the right word, but there's all these people saying, "What are we going to do with all the people being laid off because of AI?"
>> Well, I'm saying I'm not convinced we're going to see mass layoffs from AI. I mean, we, you know, I I I've done a lot of work in labor economics. We should be worried about people losing their jobs generally, but, you know, it's not it's, you know, I I I really doubt we're going to have to worry about mass layoffs with AI. But, you know, if we want to worry about people losing their jobs, that's a good thing to worry about. But I don't think
>> Yeah, the job loss story is really disgraceful as well because Oxford Economics, if they had a if if they had a soul, they would be going out there and correcting the record of how often their studies are misqued. I read a thing in Axios last year that said an Oxford economics paper proved that AI job loss was happening. When you go and look in the paper, there is a single line where they say we should we saw some evidence of some job loss related to AI and provided no data as to what that was. I think that modern economics has failed. I think economic research has failed. I think many economists have failed. Justin Wolfers, who I genuinely admire otherwise, has become fully AI pill. He is saying, oh, he he is talking about how, oh, we're going to have to replace we're going to have to get ready. This is the next industrial revolution. It's like, my man, this is still a chatbot. My fancy chatbot tendrils, but it's still a chatbot.
You know, I'll speak up for my profession because I I do try and, you know, not I read everything on AI that economists produce, but you know, there's a range, you know, people at least I consider more serious macroeconomists. There's a range of estimates of its impact on productivity and at the low end you have people like Darren Asamigloo who's around a tenth of a percent a year which is not trivial. I mean, because if you take that over a decade or something it adds to something but it's not lifechanging. At the high end I've seen you know again among the people I consider serious it's about a half percentage point a year which is is a big deal but again it's not lifechanging. So if you go back to the uh the '9s tech boom um the uptick in productivity growth was about a percentage point a year.
>> Yeah.
>> So if we get a half percentage point, you know, to my mind that's a great story. I mean that, you know, principle we're getting wealthier, people get higher wages, shorter work time, you know, all sorts of good things. But it's not, you know, it's less than what we had in the late 90s and, you know, going back to ancient history now. It's much less than what we had in the long post-war boom from, you know, 45 to 73. So yeah,
>> you know, those strike me as plausible numbers. I mean, you know, that's, you know, not not a huge range. And, you know, taking the high end, again, people I'm considering serious. I'm sure I could find people much higher numbers. Yeah. But, you know, I don't I don't think if someone says it's a half percentage point a year and ends up being 210 of a percentage point a year, I don't know if I'd go to town on them. Now, if if in fact, you know, as you're saying, and I'd agree that we have a serious bubble, I think that's, you know, missing that is going to be a really big deal cuz, you know, again, I I, you know, I I'm old enough to have been through the history and it's easy enough to recover it. I mean, we have the predictions, you know, in both, you know, 2001 when the tech bubble burst and then in 20089, you know, the economy collapsed. I mean we saw a radically different story in terms of you know growth and more importantly you know employment jobs than what was projected before it burst and you know basically no one was projecting it burst um you know and that that that was a serious failing.
Let me posit a really simple thing. What if there's a 0% effect of it? What if what if there's no productivity benefits? Like that's what that's what I'm kind of getting at. What if there's none?
>> Because
>> you know again that that would be the bubble story. I mean, you know, so, so if we go, let's say that, you know, Aston is too too optimistic at a tenth of a percent. I'd have to go back to see what his exact number was, but let's say he's saying a tenth of a percentage, but hopefully I'm not maligning him, you know, exaggerating, but you know, so let's say it ends up being zero rather than a 10th, that really doesn't matter much. But what will matter is if everyone's banking on Nvidia's stock, Nvidia's earnings actually justifying a $6 trillion cap and you know all the others. You know the story better than I do. That's going to be a very bad story because we're going to see you know serious collapse in asset values and that's that's driving the well both parts are driving the economy as you say you know the effort to build data centers even if they just sit idle you know you're creating a huge amount of demand in the economy you know it's it's very visible I mean you could see it in you know I'm macroeconomist I could see it in the data it's very clear that's that's a sector driving growth and then also you know people spend based on their stock wealth so everyone's running around you know buying a second home, a new car, you know, whatever, because they just made 50,000 on stock market and their Nvidia stock. Now, if their Nvidia stock loses 90% of its value, they're going to stop buying those cars.
>> Wait, are you suggesting that people are buying Nvidia stock previously and selling it and then feeding that money back into the economy?
>> Selling it. They might just buy based on the fact that they have they they're they're sitting there saying, "Look, I you know, I had uh I spent $100,000 on my Nvidia stock and now it's you know what it go up like 20fold, you know." So,
>> well, here's here's kind of the thing I'm getting at. So, there are two things we're talking about here. I think that there is a very worrying trend of retail piling into Nvidia and related stocks. I think re I actually think retail trading needs to be paired back heavily. I don't think I don't think regular people should have access to options. I think it's inherently
>> good with that one.
>> Well, oh no, good luck. No, I agree to that. The toothpaste is
>> Yeah, the toothpaste out the tube at this point, but that's a very bad thing. That is one thing where I think you've got 401ks, pensions, well, pensions are also in data centers through private credit, which is another problem. But then there's the other one, which is the actual AI LLM productivity thing. I don't think people realize how little actual productivity changes are happening as a result of this. There are we are not seeing really new companies being built by small teams. That was the big sales. Oh, we're going to have people who run their run their own companies just one person in AI. Not really happening. Maybe in very small scale, but not really happening. People are saying, well, they're going to be able to lay people off and replace them with AI. Again, not really happening. Hasn't happened anywhere. They're claiming AI. They're doing AI washing, which is when they say, "Oh, it's AI." When they're actually just trying to make the net income look a little bit better. And then there is this thing where they say, "Well, it will enhance the team." Again, not really any evidence that's happening either. And you keep I could keep going down this list, but the actual productivity benefits are are really not there. And I mean that that in a very defined sense. They the one productivity thing we're seeing is that there's what data center construction's overtaking commercial real estate at this point. So very bad by the way the constru I think people really need to realize there's going to be a construction bubble that bursts as well and that will have a massive horrible blue collar effect specialized and unspecialized talent as well. It's going to be really horrible there. But then there's the far simpler one which is when this era ends when AI bursts and a we have run this we've run away from this maybe there's dregs of these companies left big tech doesn't have another idea that's why they did this they don't have any other hyperrowth ideas they don't have a new iPhone they don't have a new cloud computing they don't really have anything else they wanted to replace coders or enhance coders to do more to propagate software what have you. But they don't they haven't really got anything else. This is why, by the way, they're really desperate to try and sell this SAS apocalypse story. You've heard about this one.
>> Soft SAS software as a service. So they're trying to claim if you watch CNBC, if you want to punish yourself, watch CNBC.
>> There is this thing, the SAS apocalypse that's saying, okay, AI is going to replace software. And it's this is the dumbest idea in the world. What they're saying is, okay, because you can write code with AI, cuz whether or not it's functional is kind of up for question. Because you can build software with AI, you can replace Salesforce, Inet, Crowd Strike, whatever software you pay for with your own software that you that you build. I've seen very very smart people otherwise say this, and I need to be clear, this is an insane idea. It's like building a big Lego set and saying, "Wow, I could build a house. I don't need to I don't need to rent anymore. I'll just go outside and build a house." It's that simple. Software as it's sold is not just sold as the software itself. It's not just the IP. It's the data. It's the structure. It's the support. It's the keeping it bug-free and it's the ability to accommodate more staff as you grow. There are massive infrastructural sides of software. But anyway, this is the only thing they have to sell and they've let this propagate because the market was ready to accept this. The market was saying, well, software stocks haven't been growing and revenues have been slowing in software. It must be AI.
>> Now, the reason I've jumped over to this is it gets back to the idea of no more hyperrowth. That is because the software industry has a dark secret that they want to hide. We are at the end of hyperrowth. We are not go Google, Microsoft, Meta, Meta, um, Amazon, they don't have another iPhone. They don't have another cloud computing. It's not quantum either. They don't have a next big thing. Adobe, terrible company, software company. They don't have a next big thing. None of them have a next big thing. And their growth has been slowing, slowing and slowing. 2018 was really the watershed moment where things started to slow down. People don't realize that anytime you read the AI software story, you are reading someone trying to hide that software is dying. That the software industry, well, dying is probably the wrong word. Software is no longer this perpetual growth machine. AI is not changing that. AI is not making it worse either. They all thought that adding AI would actually fix the problem. No, it's kind of punted it a little bit. It's led to stock getting cut down. I mean, every time Anthropic claims they've launched a plug-in, someone stock dumps. So, this is all a very messy way to say that AI bubble apocalypse, all of that is a symptom of a greater problem, which is tech is slowing down. Tech has been slowing down for a while. Why else would we be talking about goddamn semiconductors so much? Why are we talking but what what do you like, oh, we're going to Why are you pretending you care about high bandwidth RAM? You don't know. You know what that meant two years ago. Stop talking about it now. You know, it's kind of like all the people who suddenly became straight of straight of Hormuz experts. Suddenly suddenly everyone's a semiconductors expert. Everyone's trying to get away from thinking about the very dark very sad thing which is the markets are dependent on software. They need software to keep growing. It's not. And it's not coming back. Like we need to we are in a postgrowth software economy. They'll keep growing here and there. It's not over yet, but it's coming. It's coming fast. It's almost the inverse of the dot bubble where people
>> Let me bring you to something that it was actually the first time I I heard you on the radio talk about this. Facebook and
>> and uh I I heard you a few years back and you were totally trashing Facebook about how it's going to go out of business. I forget your timeline, but you know, but that was within well maybe a couple years after they started Meta. And I guess they've basically written that off, which I
>> Yes, the metaverse.
>> Yeah. Which it's it's if it weren't for all the AI stuff, it'd probably be getting a lot more attention. I mean, would they spend like $70 billion that I mean, that's a lot of money. I know it's
>> Where did it go? Where did it go? Where'd all the money go? People are really cavalier about the fact that Meta lost $70 billion. Like that in a functional SEC, we'd have a fundamental investigation of this. That's a lot of money to just go nowhere. And I don't think it was in the pervert glasses, the Ray-B bands. I don't think that's where it went. But Meta is interesting because I think they could die in the next decade. Their advertising business is falling apart. They are the move to the metaverse was exactly what I'm talking about. They know that their ad platform cannot grow forever without well I mean they continue to screw over their customers and make the product worse. They know they know they're in post growth. So they had to come up with something and then they had the single worst idea I've ever heard. I mean the meta the metaverse was one of the worst ideas but it was the first time again sell side byside journalists immediately like oh the metaverse we're all going to live in the metaverse. We're going to do this. What are you talking about? You sound like you got hit in the head. You sound like you have a gas leak. What do you mean? What do you mean? What do you mean? We're We're going to all go to work with a VI headset. Are you insane? Are you a crazy person? What happened to you? Did you please check your
>> No, no, that's pretty much my reaction. I wasn't writing about it, but I was just, you know, more casually. I was going like, who the hell is going to want this?
Well, I mean, act um Microsoft bought Activision Blizzard in that era and they claimed it was a metaverse thing and it was just like what it that was the first sign that something was up. We we as a society have lost our grip on fundamentals. We have lost our grip and people say, "Oh, the market can stay irrational longer than you can stay so blah blah blah blah blah." That statement is a way of endorsing bad behavior. That is a way that people get away from thinking about and talking about bad things and bad behavior. It's a lazy, cowardly thing to say because it allows you to think, well, it's okay. The world is full of scammers and grifters and lies, misleading statements. It's good actually. It's because the markets are bad, but the markets will eventually turn good one day. And it frustrates me because I'm relatively new to all this. In the last few years, I've really got into the the deeper economic side. And the more you look, the more obvious it becomes. There's a great book called Financial Shenanigans. Is it Shillet? I think it it fantastic book. You should look it up there. It is straight up a long bullet point list of all the different ways that companies manipulate their their shareholders and the dodgy things that are legal but accountancy their accounts and see tricks, financial shenanigans. And I read it and I went, "Okay, Nvidia, Oracle, Nvidia, Oracle, Meta, Microsoft, Amazon, Google, like throughout you could just every major tech company was doing all of them. We're we're in a decaying market. We're in a market that is pumped up by outright scammers. And those scammers are the chief financial officers of major companies. Microsoft claimed that they in each of the last two quarters brought 1 gawatt of data center capacity online. I could find I could not find any data center that they broke ground on since 2023 that's actually finished. There's something very wrong and something very sick about our markets and it's worse than what happened with the housing market cuz the housing market at least created housing. Data centers have no other use unless we're going to have laser tag everywhere. Unless we're going to turn them all into Walmarts. I don't think that's going to happen. But GPUs, the the classic post.com bubble comparison, people says, "Well, the dark fiber could get turned on." Here's the problem with this one. Dark fiber useful. It was dark. You light it up. It was still useful for getting the internet places. The interchange stuff was put in. It was expensive up front, but it didn't cost a ton of money to stand up and make useful. With data centers, if these if this construction is not finished, it will be just as expensive, if not more, to finish it. those GPUs, the Nvidia GPUs, those aren't going to be cheaper in the future to run. There is no way the costs come down. Nobody has an answer there. People will say specialized comput uh specialized uh silicon like AS6 they call them. It's just a chip specialized for one thing yet. Not happening. Just hasn't happened. Broadcom's been trying for years for Google. Now they're trying for Open AI. Not happening. There is the like a black world GPU is going to be expensive in the same to the same level in 2030 as it is today. There's no economic shift that's going to happen that's going to make any of this cheaper in the way the.com bubble did. And in the com bubble again, you could go out and buy a server, some micros systems, what have you server from someone, a dead startup, and a bunch of Aeron chairs, and you could start up your own thing. You can't do that with an Nvidia GPU. You need Rexace in a data center with power. You need somebody that can actually set it up. You need someone who can monitor it because especially the latest generation ones require specialized bespoke cooling. And the more you have of them, the more of that specialized cooling and specialized talent you need to actually use it. Michael Bur, who's famous for his do his um housing bubble stuff, has gone completely tits up on this. He is full AI be uh bull now. He's pro-anthropic, pro all of them. And I think what it is is that people are very good at numbers and statements. They can go, okay, people aren't paying their credit cards, that's bad. People aren't paying their mortgages, that's bad. Michael Bur example, wow, these big blocks of mortgages, these these uh these big big bulky financial instruments full of crap mortgages that look good on the when you look from a far away point, you look in, the mortgage is bad. That's bad. I should short the housing market because most mortgages are going to go bad. Something about AI does something to the trader and economist brain. They think h this isn't this isn't a bubble because AI could dot dot dot and then they say a bunch of insane stuff just a bunch of crazy things. They'll be like yeah and all software will be automated and uh every job will be automated. Everything will just be controlling AI agents of course. And it's just it's just something about the concept of AI, which is really a marketing term. AI is really just marketing. It doesn't really mean anything in the end. Something about that and extrapolating from the point of what automation theoretically could do drives people insane and makes people make these assumptions and jumps and leaps of logic. Insane leaps of logic by the way that say okay well from this point we are going to have the unlimited automation that will change everything in every industry infin infinitely. They c they it's almost like because there's a possibility maybe even if there's no real evidence that something might happen in three years. We can ignore everything bad now. Every bad every worrying economic thing like OpenAI burning 20 3040 billion a year anthropic doing the same thing. We can ignore that. We can ignore the fact that most data center debt deals will go unpaid because their project financing which requires revenue which won't be there. That every AI startup is unprofitable. People will ignore all of that. Really smart people, people that should know better, many of them will ignore all of that because Dario Amade or Demis Hassabis or Sam Alman or Jensen Hong or one of these other goofballs says, "Well, it will actually change everything in the future at some point." It's so strange. I I I don't even think the com bubble compares. I mean, there are obviously parallels, but I don't think it compares because even when you go back to the dotcom bubble era, I read I've I wasn't I was younger back then. I'm only 40.
>> When you go back and read the journalism of the time, there was a ton of people saying crazy stuff, but not as many. And there were a fair amount of people going, "Hey, wait, wait. These businesses, these businesses are losing so much money and they're making $1 million a quarter. That's bad. Excite just bought Blue Mountain gift cards for like a billion dollars. That's an insane deal. Why would they do that? People were still saying that and they were saying it years before it burst. We don't I have I've genuinely been attacked in the media and on social media for saying the things I'm saying. Like there is an actual cultlike atmosphere from journalists, by the way. Not just like regular freaks on Twitter. I mean actual jouos who are like how dare you like this person he's obviously got he's got some he's got some agenda. No, I I don't have I I don't have an agenda other than the fact that I think that people are being misled.
>> Yeah. Well,
>> it's quite a story. Uh I guess we'll see whether it ends up being worse than the dot bubble. Um I mean the one thing I
>> That's what I fear. You know, I I I keep saying I hope it happens s you know, sooner that collapse sooner and people going, "Oh, you want bad things?" I go, "No, it's it's just going to be worse. I mean, if it's not real, if it turns out to be real, wonderful. But I don't see any story there. And I'd rather see it collapse tomorrow than, you know, go on for another 6 months, year, two years." So,
>> yeah, that's exactly how I feel. It's I want this to end sooner rather than later because it's going to suck. But the longer it takes, the worse it's going to get. Yeah. Yeah. All right. Well, I'm going to have you back here 6 months, a year down the road, and either it'll be a lot worse or maybe we'll be picking up the pieces. So,
>> I I really look forward to it.
>> Thanks a lot for joining us.
>> Thanks for having me.
>> If you like what you just saw, please help us out by liking us and subscribing. Thank you.