The Four Horsemen of the AI Apocalypse with Ed Zitron

The Compound and Friends

In episode 257 of The Compound and Friends, hosts Josh Brown and Michael Batnick welcome Ed Zitron to discuss the 'ultra-bear' case for AI, questioning

Key takeaways

• AI revenue is not being disclosed by major tech companies, creating an illusion of success despite lack of actual financial performance.
• Hyperscalers like Microsoft, Google, and Amazon are financially dependent on unprofitable AI startups for cloud revenue projections, which may be unsustainable.

Main topics

• AI economic sustainability
• Nvidia's growth and market dynamics

Notable quotes

"I couldn't find any revenues. And every time I was like, okay, great, but how much does it make? And everyone was like, I couldn't possibly say."

Conclusion

Ed Zitron presents a stark warning about the AI bubble, emphasizing that without transparency and

Transcript preview

Speaker 3 (0:00) Like I ran a PR firm until like last year and I was like doing that as kind of a contract labor thing and then I started writing this in 2020 and I just kept going and going and going and going and then initially I was writing about like management theory stuff and how much I liked remote work and how evil the articles were about it. Then I wrote about crypto and I wrote about Elon and by the end of those two things I was so thoroughly depressed with both I was just like f***. What am I going to write about? Speaker 1 (0:24) Well, with the topics or with writing? With the Speaker 3 (0:26) topics. The Speaker 1 (0:26) topics were so Speaker 3 (0:27) depressing. Like every week following what f***ing Elon Musk does. And then the metaverse happened. That was kind of fun to write about, but that was before that. But then OpenAI got like, sorry, Sam Altman got fired from OpenAI. And I saw on Twitter that there were journalists who were like, oh my God, I hope they bring him back. They were like talking about him like a rock star. It was like, oh, then everyone's, there was someone like, oh, I'm crying. because how beautiful it is that everyone's showing support for Sam Altmer. Josh cried. Journalists. Yeah, journalists. It's f***ing crazy. So I saw this and I just, I didn't even judge them for it. I was like, fine, whatever. But I was like, this is weird. Speaker 2 (1:05) Why Speaker 3 (1:05) are you acting? Like, I was like, what does this mean? Why are you acting this way? And so I kept looking really intently and I just immediately saw this thing where everyone was saying, AARCH is the biggest, most hugest, best thing ever. It's inevitable, blah, blah, blah. But I couldn't find any revenues. And every time I was like, okay, great, but how much does it make? And everyone was like, I couldn't possibly say. How much does Microsoft, they're spending billions of dollars on CapEx. How much does this make? Can't tell you. No one wants to tell you. Speaker 2 (1:33) But Speaker 3 (1:33) you kept having these articles every earnings season saying, Microsoft's AI bet pays off. Google's AI bet pays off. Even though they never mentioned the AI revenue. And I just kind of taught myself economics from there. Like a lot of this, like everything's self-taught. Which to the chagrin of my many haters. It's just like stuff I learned. Speaker 2 (1:51) But Speaker 3 (1:52) you can learn a lot by reading books and such. Okay. Speaker 2 (1:55) So you, all right. So you're, I guess it's a sub stack or? Ghost. Okay. So what you're doing is being read everywhere. People are sharing it and bulls on AI are reading it because I think, so my view of the market is most professional investors are intelligent enough to listen to both sides. Speaker 3 (2:18) And Speaker 2 (2:18) they don't have to agree with everything they read in order to want to keep reading it. I don't think that's the same as true for retail investors, but we're not going to concern ourselves with that. I think professionals do like to hear both sides, especially if they have a lot of money on the line. That's actually been my experience talking to some of the banks and the institutions. Even the ones Speaker 3 (2:37) who are very pro-I. I agree with that. Speaker 2 (2:39) All right, we're the same way. So we've been doing this show. We tape every Thursday. Going back to the summer of 2021. And for five years, the majority of our guests are bullish. And so far, so good for the most part. Yeah. Okay. We have had people who are bearish, but they are non-specifically bearish. So we've had people like Jeremy Grantham, who's become famous for, he takes issue with the perma bear label, but he's almost always bearish. So we've had people like that on the show. I always find that fascinating, but we haven't really had specifically an AI bear. Somebody that like specific, you know, most of the people that come on portfolio managers or their analysts covering the space and you know, all the incentives to see the brighter side. So this will be an interesting test of all of the things that we've done on the show to hear the other side. So we're really excited about it. I'm pumped. Okay. Awesome. Speaker 1 (3:36) Do bulls think you're an idiot or do they generally like, like. what you're saying to them. So Speaker 3 (3:40) it's got to a weird point. So if we're mixing in every kind of bull here, so I imagine a lot of them are retail bags. They are driven, like, because I used to be in PR, they're like, he's just a PR boy. He's just a PR guy. He shouldn't talk about it. Stop talking to him. He could say something as if financial experts have never got anything wrong before. There are some of, like, what's weird is like, there are some of them who are quite bullish on semi-cadets, like, who follow me on Twitter, like Bubble Boy, who's really great and very, very smart on semis. I'm not sure he agrees with me on everything, but very kind and friendly. And like, there are lots of them who are. Some of them are vile and horrible, but... So say many such cases. And so the thing is with the bulls is the ones who are like, I love this tech. I have some crazy dream or what it will be, but there's a bubble. I respect those. If you can like love a, I still think there are massive environmental and social problems that they're not thinking of, but at the very least you're living in reality. I think the ones who are like anthropic and open AI are going to grow to $284 billion in revenue. That's what OpenAI is projecting in 2030. The people who read that and they're like, yeah, sounds good to me. Those people are not living in reality. And I don't need a degree of Speaker 2 (4:52) fun. Speaker 3 (4:52) Because of the number? Speaker 2 (4:53) Yeah. Speaker 3 (4:54) OpenAI is going to become bigger than Meta in three and a half years, if you believe their projections. And they're going to do that while spending more than twice of Meta's OPEX. I think Microsoft's OPEX is like $150 billion. And OpenAI is going to spend $200 billion. something billion or more on compute in 2030. Speaker 2 (5:12) To get to $380 billion in revenue. $284 billion. $280 billion. So Speaker 1 (5:16) wait, Speaker 2 (5:16) are Speaker 1 (5:16) you outright bearish or you think the balls are smoking crack? Or am I saying the same thing? Possibly the same thing. I Speaker 3 (5:21) think the... Speaker 3 (5:23) After 2022, when everything got really grim in tech, I don't know if you remember, like NVIDIA had a flat year, fiscal 23, really rough time, and everyone was kind of blowing hard. Microsoft, Google, Amazon, Meta, cranked up prices, changed ad auction stuff. Amazon actually made a surprisingly healthy ad business. They had to find ways to grow revenue. And also chat GPT happened. So they went, oh, we'll buy a bunch of GPUs. And the market immediately was like, all of this revenue growth is coming from AI. And because the companies never disclose their AI revenues and journalists don't have the teeth to bother them, or analysts especially, people were like, oh, it's AI. Every time they spend this money, it's AI. And this was really good. It helped their stocks pop crazy style. One fair criticism of my work is if you traded off of my work, which I've never said to do, I do not give financial advice. Yeah, you probably... Well, but if you Speaker 2 (6:12) end up being right about this... they're going to put a gun to your head and say, start a hedge fund. I mean, oh Christ, Speaker 3 (6:17) I don't even know how to do that. But I'm just telling you that. It sounds fun. But the point is, I like writing a lot. I really enjoy actually getting into it. It's genuinely intellectually fascinating. But the point I'm making is, up until about 2025, it was mostly just buying GPUs and hoping OpenAI and Anthropic grew into it. And also hoping they'd become profitable. But then what happened is the hyperscalers have now become... financially dependent on the growth for OpenAI and Anthropic. Analyst expectations from like UBS, Barclays, and Wells Fargo have $440 billion of cloud revenue across Google, Amazon, and Microsoft coming just from OpenAI and Anthropic, two unprofitable startups who need to constantly raise money. And also, the big companies need to build the data sets to make the money. So there's all of these very improbable, if not impossible things that need to occur. And also, OpenAI and Anthropic need to have 10 times the demand they have right now. It's not even, however you may feel about these companies, they are not big enough. They are not generating enough cash right now to even get close to covering their $1.1 or more trillion dollars in commitments. And on top of that, they're taking up 90 % of AI infrastructure. This is creating an illusory demand signal because right now it's difficult to get GPUs. People are saying, oh. Oh, it's because there's so much demand for AI. There's a big Barry Bonds asterisk at the top of that. There's so much demand from AI from two companies, pretty much. Meta as well, but they're not doing anything. They're just rolling in their filth. So what's happening is everyone's going, oh, there's tons of demand for AI. I'm going to build a bunch of data centers. Even though when you, so in fiscal year 26, Microsoft, they made, according to Bloomberg, about $34.33 billion on AI. $24.1 billion of that is open AI. which means that open air Microsoft software and GPU rental services are single digit billion dollar businesses. And they have 260 plus billion dollars in capex in this. It's a disaster. And it's not something that gets easily fixed. I don't think it gets fixed at all. And on top of it, AI GPUs are pretty specialist. They're useful for like AI and data analytics and 3D modeling, which is a very small business. There's no dot-com bubble thing after this. The electricity will be just as expensive. if Speaker 2 (8:32) not more. Oh, wait, you want to roll? I want to give people the chance to close out their long positions and we're going to introduce the show properly and then we're going to dive right back in. Who's doing my countdown? Miss Nicole? All Speaker 1 (8:46) right. Speaker 1 (9:02) This message is brought to you by Fidelity Investments. When timing is everything, you need powerful tools and research that can meet you in the moment. With the all-new Fidelity Trader Plus platform, your charts and preferences show up consistently, synced up across all your devices, so you can act fast whenever and wherever you're trading. You can save Speaker 2 (9:20) and order on your desktop at home, get a mobile alert when you're at work, and complete the trade in the Fidelity app without starting over. And with the downloadable Fidelity Trader Plus desktop platform, you have more control with multi-monitor views, enhanced tools, and customization options, and integrated screen sharing with Fidelity trading specialists. Speaker 1 (9:42) Try Fidelity's most powerful trading platform yet at fidelity.com slash trader plus. Fidelity Investments and The Compound are not affiliated. Views, opinions, products, services, and strategies discussed are not endorsed or promoted by Fidelity Investments. Fidelity Brokerage Services, LLC. Member NYSE Speaker 2 (10:00) SIPC. All right. Ladies and gentlemen, welcome to the world's greatest investing podcast. I say that without a trace of irony. I really mean it. Today, you're in for a very special treat. We are having a conversation that we have not really had here before. What is the ultra bear case on AI? Why are people clapping? It's crazy. It's your retirement on the line. What's going on? Guys, calm down. All right. Joining us today, Ed Zitron. Ed is the founder and CEO of EZPR, a research and media analysis firm he launched in 2013. Speaker 2 (10:40) He also hosts the Webby Award-winning Better Offline podcast with iHeartRadio and Cool Zone Media. Ed writes a newsletter called Where's Your Ed At? that he launched in 2020, where his 2023 Rot Economy essay, I'm going to get into that, arguing tech firms have chased growth at the expense of innovation and users has gone viral. Today, he's one of the loudest and most cited AI skeptics. focused on the unit economics of OpenAI, Anthropic, and the hyperscaler CapEx build-out. Ed, thank you for being here. We appreciate it. Thanks for having me. All right. All right. So we have to start with NVIDIA. Yeah. I feel like they're proving everything that you've said to be not correct. Speaker 3 (11:31) I know, Speaker 2 (11:32) but just— No, no, no, no. Hear me out. They just came out. Here's what they did last night. And Michael will jump into the numbers, and then we're going to give you a chance to comment on it. But they just said for the first time ever, we're actually going to give a full year's worth of guidance. They were going quarter by quarter. They actually gave guidance through fiscal 2028, which is not calendar 28. It's February 2027 off in the woods. So they're saying like, we're going to grow revenue by about 70%. We've got the sales locked in for GPUs and Vera CPUs and et cetera, et cetera. We know that like 90 % of their business is selling to data centers. A huge chunk of that is selling to the same four or five customers, hyperscalers. They are talking about broadening that out. They're talking more about now automation, robotics, but the data center business is the business. Okay. So. Is he going to be completely wrong in his own forecast or will they make their number? But that's part of the problem because of who the buyers are and what their motivations are. Like what, let's give people the nuanced take on exactly what you're saying, because NVIDIA really is at the heart of the entirety of the AI story. I just want to be clear as well. NVIDIA made a bunch of money. Speaker 3 (12:49) Like that's, I'm not questioning that. That'd be crazy. Of course. However. 16 % of their latest quarter revenue was one customer. 44 % of their first half of fiscal 2027 was three customers. Five customers make up 70 % of their accounts. Can I jump in right there? They're good Speaker 1 (13:05) customers. Speaker 3 (13:06) Whoa, whoa. Can I jump in right there? Are they? Speaker 2 (13:07) Because they don't name them. Well, we know who they are. Do we? Well, we might not know who's 16 and who's 12, but we know the batch Speaker 3 (13:15) of companies. Yeah, but Speaker 2 (13:15) if one of them is Oracle. But hear me out. That's their immediate customer. Those customers have millions of customers. And that's what this, in other words, you don't have a hyperscaler spending on NVIDIA chips to serve itself, except in the case of Meta. Right. They're serving Fortune 500 businesses, governments, sovereigns, et cetera. Those are their customers. But Speaker 3 (13:42) I just, the point I made about Microsoft's AI revenue is where this comes in. So fiscal 26, which just ended. $34.33 billion of AI revenue. $24.1 billion of that is open AI. Speaker 2 (13:53) That means that selling GPUs to anyone else other than Speaker 3 (13:57) their large fail son is a single-digit billion business. Microsoft is the apex predator of software sales. They have hundreds of thousands of resellers. They have tens of thousands of salespeople. They can't scrape together more than single-digit billion dollars of sales for Copilot and all the stuff. And this is on top of the fact that GitHub Copilot, one of their only successful AI products, has now gone token-based billing, which means people before were able to spend $40 a month. spend $5,000 of tokens. Now the token-based billing is gone. Gone. That business is dead. Speaker 2 (14:27) And Speaker 3 (14:28) on top of all of that, wow, $260 billion in CapEx and all of that Fortune 500 demand, like whatever it may be, is what, $10 billion a year? Satya Nadella Speaker 2 (14:37) would say, give us a couple of years. Why do I have to give, sorry. But the Speaker 3 (14:42) point is, is like, we've given him a few years and this is the best he's got for us. We have a, OpenAI is a liability now. Like that's, they are material enough that Microsoft actually has, I think, what's like 27 % of them? It's actually a deeply Speaker 2 (14:57) load-bearing company. They have to consolidate their financials onto their own income statement. And the Speaker 3 (15:01) other thing is, is Microsoft's only getting paid if OpenAI can keep raising money. And OpenAI, it has a shrinking amount of people to raise it from. Their last round, $122 billion. Only $12 billion of that came from venture capital and the very same asset managers who were funding data centers. $110 billion was NVIDIA, $30 billion, SoftBank, $30 billion, which put them into financial dire straits, and they're still there, and Amazon, $50 billion. That's not going to happen multiple times. If NVIDIA invests again, they said they wouldn't. They literally said the words. If they invest again, it's bad times ahead. So Microsoft's growth is dependent on open AI and now anthropic spending. Outside of that... They don't have significant demand. This isn't even an opinion. This is a Bloomberg article that spelled this out. And if you go and look at Speaker 1 (15:50) - Speaker 3 (15:50) Was it a Bloomberg opinion? No, it was Brody Ford, I think. But it was, and it was also based on the actual earnings documents. Microsoft Speaker 2 (15:57) doesn't have the demand for its own AI native products from its own customers. Speaker 3 (16:02) At the Speaker 2 (16:02) scale that they would need to Speaker 3 (16:03) justify, Speaker 2 (16:04) like, I'm Speaker 3 (16:04) not saying there's zero demand. They have found- But this is also after them harassing their customers and massive sales. I must be clear. Microsoft sales teams are, there's only one level higher of more annoying, and that's Oracle. And Oracle is a whole other problem. But the point is, is if this was a case where... 30 % of their revenue was open AI. I'd be like, all right, maybe my book was kind of falling apart. 70 % of their AI revenue being open AI is existentially bad. They have $250 billion of Azure spend committed Speaker 2 (16:35) to. Speaker 3 (16:36) That's so much. Speaker 2 (16:37) Microsoft was $331 billion, I think, last fiscal year. But now they've divorced themselves from the exclusivity agreement with open AI. Which only makes it worse because now OpenAI needs to spend Speaker 3 (16:49) $138 billion over eight years with Amazon, $20 billion with Cerebrus, $22.4 billion with CoreWave, $300 billion plus with Oracle. They're spending, this is the greatest thing, Stephen Jew from UBS, he had an analyst note, they're spending like $12.5 billion a year at Google. I didn't know this until very recently. It's weird that it's not being reported very well. But the thing is, is OpenAI, Sam Altman has one real talent, and that's signing his name. My man loves signing contracts. And the thing is, when the problem that everyone has right now is you can only rethink a quarter, two quarters in the future. You can't, because even like, oh, it's going to happen in 2028. NVIDIA will get to 600 and something, 674 billion, I guess it would be based on consensus for fiscal 28. So NVIDIA will get there because right now NVIDIA makes lots of money. It's not a problem now. To get there, NVIDIA is going to have to basically take three to five customers. they're going to have to find way more debt, like way more, because the price of debt is increasing, but also NVIDIA just bumped up prices 17%. The cost of memory is skyrocketing. And on top of that, you have a shortage of talent, shortage of electrical grade steel. There are all sorts of things stacked against NVIDIA here. They could pull it off, but I don't know how they're going. The actual money that needs to get raised is astronomical. Speaker 1 (18:10) So Daniel, chart five. Ed, if somebody said to you two years ago that this is going to happen, you would be saying the same things that you are right now about how implausible it would be. So what we're looking at for people that are listening is the trailing 12-month net income for NVIDIA just screamed past Apple. Speaker 3 (18:28) Uh-huh. But how much of that is equity gains, though? Speaker 1 (18:31) What does that mean? Speaker 3 (18:32) As in their investments in Anthropic, OpenAI. They've invested in like CoreWeave, Nebius, Iron. They have a bunch of equity investments that help bumping it up. To be clear, I was wrong back in 2024. And why? Because I was f***ing naive. I was like, the market's a sensible place. Microsoft and Google, Amazon, Meta wouldn't spend hundreds of billions of dollars for no goddamn reason. Wait, so you were wrong Speaker 1 (18:52) because the market's stupid? I was wrong because I was stupid. Speaker 3 (18:55) I was wrong because I was naive about how the world works and I've had to take a hard- You didn't think they would take Speaker 2 (19:00) it as far as Speaker 3 (19:00) they've taken it. Exactly. And Speaker 1 (19:01) I didn't think that the debt system would support it. So yeah, I- So what do you think the market is getting wrong right now? Because I think everyone knows that open AI is a potential problem. Yeah. Look at Oracle stock. It's down 70%, still on the mat. Even as software stocks are bouncing, Oracle has caught no bounce. So what do you think the market is misunderstanding that you're saying? The Speaker 3 (19:21) thing I said about the centralization of data center demand. OpenAI, I think, is like 60 % of all AI data center demand. Without OpenAI, there is not another OpenAI-sized spender. And the only reason there is an anthropic OpenAI-sized spender on AI compute is because of the availability of venture capital dollars. So without that, you can't, like, the demand isn't there. The market isn't seeing that because it hasn't happened yet. The quarter, two quarters think they're doing where they're like, well, the money's still coming in. Because what, I kind of said this already, but the amount that Anthropic and OpenAI Speaker 2 (19:57) has to spend has to increase by like 10x. Well, their revenues are increasing and those are real revenue dollars. And they're both going public in the next, call it six months. Speaker 1 (20:08) Wait, you just made a face. Anthropic's doing 60 billion. Is that not real? No, they're not. Run rate. Go, go, go. Run Speaker 2 (20:13) rate. Okay. This is Speaker 3 (20:15) the biggest scam of them all. What? ARR run rate as opposed to last quarter. That's the first part of the scam. So ARR used to mean annual recurring revenue, referring to if I have 10, 100 million contracts, I've got like a billion dollars a year. Easy peasy. That's annual recurring revenue. Wow. How amazing. That's SaaS. SaaS. I probably f***ed up the math there, I realize, but whatever. Moving on. And how they're booking the revenue you probably don't like. Well, now they're calling it run rate. And run rate kind of… Anthropic has never defined this. Even Bloomberg in the 65 billion story didn't define it. It can mean four weeks times 12 or four weeks times 13. The problem is, is Anthropic and OpenAI are both annualizing token spend, which is not a recur- it's not a software subscription. It's not recurring. A customer could spend a thousand this month and then not spend- and then spend 50 next month, especially if they're moving to open Speaker 2 (21:04) source models. Sorry, that's 100 % true, but- every publicly traded corporation that says anything to Wall Street about its spend, the only thing you ever hear is, we already ran out of our compute spend and need to spend more. Yeah, that's because Anthropic and OpenAI are taking up Speaker 3 (21:23) most Speaker 2 (21:24) of the Speaker 3 (21:24) compute infrastructure, so there's not enough. There is more demand than there is right now. The question is, how much more? Because I don't think that, I mean, I kind of estimated in a newsletter. week ago, I think there's about $22 billion of compute demand. Sightline Climate said back in February that it was 190 gigawatts in planning. That's about $1.5 to $3 trillion a year in compute demand they'd need. There is a complete economic Speaker 2 (21:52) mismatch, and it's stark. But you don't like ARR run rate because it's just a... extrapolation of a it's a snapshot of a period in time and then they plug it out to like a year yeah so Speaker 3 (22:05) it's we also don't know what the period we don't even know what it means they never define it anthropic has used it several times in their own announcements they never define it it could refer to hey we just released a model and everyone's trying it hey we just did a new feature and everyone's buying a new subscription oh on this day we lost a bunch of subscriptions so we'll pull the period over here so we look bigger it is not a trustworthy measure of a business. And the fact it gets accepted is an insult to investors' intelligence. So Speaker 1 (22:33) who's the idiot here? Is it the CEOs and the boards for all of this CapEx? Is it the equity investors?