Transcription
I think that that's actually going to be the question that finishes the AI bubble. Every time you look for the supposed gold mine, you find someone just handing someone money instead of the gold coming out. I don't think gigawatts are getting built. I think we're talking the data center buildout is in the hundreds of megawatts a year. I think it's so much smaller than people think because I can't find data centers that are actually opening. The market's no longer listening to subtlety or even reality. So, they're waiting for something that is completely and utterly impossible to ignore. I don't know what it will be, but considering the demand doesn't exist, I think it's kind of inevitable.
On the tech report with me today is author of Where's Your Edat and the host of the Better Offline podcast, Ed Zitran. Thanks for joining me as always.
Thanks for having me. Goldman and Sachs released their tracking trillions report which outlines a damning assessment of the AI buildout. It's it projects could cost I think about 7.6 trillion depending on which end of the spectrum you go to by 2031. This is all despite what it points out is AI is costing the majority of businesses much more than it is saving. How did Open AI convince everyone that AI was going to be the next era defining technology?
Think of it like this. When OpenAI and Anthropic, but to a much larger extent, OpenAI started, they were treated like rich kids. They were never challenged to be efficient or sustainable or profitable in any way. And Microsoft built out all of OpenAI's infrastructure, much like Amazon and Google built it out for Anthropic. They were as a result effectively all the demand. But the big thing is is that people focus way too much on how much money is being fed from the hyperscalers to these companies, which is correct to do. But the big thing is is I calculate this. It's like $300 billion was invested just in OpenAI and anthropic servers. So the entire AI industry saw that and said, "Wow, those those cool cats, they need this much compute, right? That must mean that everyone else will need that much compute." Except I've spent most of the last week looking for any evidence of anyone who is renting anything close to that much compute capacity. And I found one company, XAI, who built their own data center, the Colossus 1, which they have now rented to Anthropic in totality, which suggests that and that's a big AI lab as well. That's a giant AI lab theoretically speaking. And what it suggests is that when Deep Seek happened in January 2025, that was a warning to the entire market. It wasn't just about China saying, "Oh, we can do things with less GPUs." And they use 50,000 H100s to be clear. What it actually was saying was no one needs that many GPUs. Not just, it's not just about DeepSync. It's everyone. It's not even about China. It's about the fact that the majority of companies, even pretty big ones like Perplexity, only did a couple hundred, couple thousand GPUs at most. And Nvidia has said last year that they had shipped six million GPUs. The demand does not exist. But because the market saw OpenAI and Anthropic taking up all of this data center sprawl, they said, "Well, well, that must mean that there will be another anthropic, another Open AI." But that's the thing. The only reason those two companies existed was they got to act like rich kids running up credit card bills in their mommy and daddy's cards. They were they were allowed to spend as much as they wanted. In the early days of OpenAI, middle of 2024, the information had a great story about this. They were being charged at cost rates for a A100 GPUs. And that's the thing, these companies could never have grown to their size without hyperscala bank rolls and hyperscala subsidies and welfare. So what we've ended up with is this illusion of demand. And people will laugh at this and say, "Oh, illusion of demand. Look, number go up in the stock market." But I've looked I've looked everywhere. There really aren't any signs of major customers for GPU compute. And all these data centers say that, oh, we got pre-bookked tenants and all that. Sure you do, Cupcake. Yeah, I bet you do. They live in Canada by any chance. And it's really weird. It's really weird. And I feel kind of crazy after it, but hours and hours of looking for customers for an industry that's meant to be the biggest thing ever. Should come up with one, but the only ones I can find are OpenAI, Anthropic, Jane Street, which is a hedge fund that is invested in Corewave, Microsoft for OpenAI, Google for OpenAI and Anthropic. Actually, Microsoft's also giving compute to to Anthropic as well, Amazon for Anthropic, and of course, Meta. Now Meta just has AI psychosis. Like Mark Zuckerberg has AI psychosis. He's just a He's building a CEO chatbot for himself. He He's just I think at this point Meta is either doing something very weird with their capex or he's just a mad kid. It's a Meta is like Versailles at this point. Probably smells the same too.
I want to go back to the AI revenues. You you briefly mentioned Microsoft boasting 37 billion, Amazon boasting about 15 billion, but again also as you mentioned they spent about $300 billion to get there. How much longer can investors applaud what are massive losses as as if they're profits?
Well, the other thing about that, and you're completely correct, the the infrastructure costs were bad, but out of that 37 billion annualized revenue from Microsoft, based on my own reporting of OpenAI's spend on Azure from last year, I think OpenAI represents 70 to 80% of that revenue, and the rest is Microsoft 365 and uh GitHub Copilot, and that's GitHub Copilot before they changed to tokenbased billing. So in reality most of that revenue is open AI over with Amazon 15 billion in annualized revenue on my reporting from last year about their AWS spend 80 or more% of that is anthropic. So investors are completely disconnected from reality at this point. They will in they will buy into stocks without a second thought. They will still buy companies like iron uh Neocloud uh AI compute company. They will still buy that despite them losing losing more money and making less money in the last quarter after delaying their earnings for an hour. An hour. I've never seen it in my life. The thing is investors are herd animals right now. The herd all the penguins are squawking and saying this is the right thing. This is the right AI is the biggest thing. We just need to wait. The agentic era is nearly here. Even though it was meant to be last year, but it's here now. Maybe next year. Maybe maybe 2028. get back to you on that one. So, investors are going to keep doing this until someone moves, until Microsoft drops capex, until Google drops capex, meta,
or a very bad signal comes out of OpenAI. And you've had a few of them right now. The Broadcom chips deal that was meant to be 10 GW by the end of 2029. Turns out that actually that deal has not consummated and requires Microsoft to buy 40% of the chips up front. And Microsoft doesn't want to. So, not really sure how OpenAI is going to build their custom chips. That's the kind of thing that should shake the markets, but Broadcom stock's still looking all right. So, it's not going to be that, but it's going to be something where the money doesn't come through. Earlier this morning, SoftBank, they be trying to raise a margin loan to fund OpenAI. They've dropped it from I think they were going to raise 15 billion, but now it's 6 billion. It's so strange. The hysteria will continue until someone big enough freaks out the market completely. A a kind of signal that just is impossible to ignore. Someone doesn't get paid by open AI. Someone doesn't get paid by anthropic. A round doesn't close because it can't be something subtle. The market's no longer listening to subtlety or even reality. So, they're waiting for something that is completely and utterly impossible to ignore. I don't know what it will be, but considering the demand doesn't exist, I think it's kind of inevitable.
James Cavella, Goldman's head of global equity research kind of agrees with what we and certainly you have been saying this whole time, what you were just saying now as well, that the the driving factor behind behind all of this is not not rational investment, but insecurity, if not just outright fear. How much more cash can be burned until the alarm bells grow loud enough to start waking people up?
It's hard to say because what we're they are now hyperscaler. No, I don't mean hyperscalers. I mean four companies Meta, Google, Amazon, and Microsoft are going to spend about uh 800 $900 billion this year, a trillion dollars next year. At some point, what's going to be for the problem for them is if growth shifts in the next few quarters. if growth just trends downwards. Right now, growth, especially for Microsoft, is being heavily pumped by OpenAI's voracious compute demands. They're spending $50 billion on compute this year. Now, is that profitable? No. Are the margins good? No. Can OpenAI afford it longterm? No. But as long as the cash keeps coming in, Microsoft sitting pretty. That being said, I hope the rest of their business lines don't show slow down because if they do, that's the kind of thing that is going to start pissing off the markets. Now, with Meta, they're already kind of on thin ice. The last earnings didn't go brilliantly for them. They showed that I think they showed their first drop in users since they started sharing them. It's very strange what's going over there. But Amazon, I also think it I think Amazon and Meta are the weakest links because Amazon's AI revenues are pathetic. I mean, come on, you're feeding yourself money. You can't feed yourself more money than that. And I think that another underdisussed point here that might get in the way is right now is a little challenge for the listeners. Go and try and find a completed data center that was announced in 2023. Just go and try and find one. I tried I went through all of them. I Microsoft claimed they brought up a gigawatt of capacity in the last quarter and a gigawatt before that. I went and looked and I've looked and I've seen every single announced Microsoft data center project. Every single one. Gone and looked at all of them. None of them are complete. Some of them aren't even under construction. Some of them already just started construction again in March of this year despite being announced in 2023 2024. I think a narrative that could set in that could be real trouble for these companies if it if it becomes blatantly obvious that data centers aren't getting built. And the reason is is that these companies have been saying they've built all these things and they're spending all this capex. If the market suddenly believes that actually these things are just sitting in warehouses, everything kind of falls apart from there. And I think that's more likely than people think because going back to the Colossus data center, that was a 300 megawatt data center. So pretty big but not the biggest. But there's meant to be even sighteline said 5 GW was built last year. Hm. 300 megawws should be small potatoes. Why did they have to buy data center capacity from Elon Musk? Especially because those are older GPUs, predominantly H100, H200s. You got some GB200s in there, but nevertheless, it suggests that new capacity was not coming online from Amazon, from Google, from Microsoft in the next quarter or two. If that's the case, we have to start asking where these GPUs are. And I think what's more likely is that data center capacity is even lower than even Sighteline says. And I like Sighteline Climate. I think they do good research. I have to wonder if more than a few hundred megawatts are coming online every quarter. I don't know how you prove this, but if it gets proven to the market, even in some messy form, I think that that's when the rubber meets the road, so to speak, because the reason that everyone's tolerating this capex beckonalia is because the theoretical thing is, oh, it's not built yet, but when it's built, oh, baby, then the money will come. But if the narrative becomes it isn't going to get built, well then it starts then the narrative becomes why are we buying all these GPUs? And I think that that's actually going to be the question that finishes the AI bubble. Nvidia believes they have a trillion dollar in GPU sales through the end of 2027. Where are they going, Jensen? Where are they? Are some going to China? We know they are. Maybe they're go maybe they're going in a Microsoft data center but where because I reached out to Microsoft and tried to get comment on these data centers. They said they get back to me and haven't it's been three days must have fallen in a well or something. The fact that the companies can't just say yeah we built a data center here is very strange and I think it suggests that something stranger is going on and if I'm right I don't think anything's getting built. I truly I think I I trust Sighteline a great deal as I knock my microphone over. Um I trust them a great deal, but I don't think gigawatts are getting built. I think we're talking the data center buildout is in the hundreds of megawws a year. And perhaps I'm wrong. Perhaps it's a gigawatt or two, but I think it's so much smaller than people think cuz I can't find data centers that are actually opening.
What you're talking about there's very concrete results. It's very well maybe it's not easy to measure like you say but eventually it's going to it's going to be known we're going to be able to see the the proof in the pudding but I guess my point is how much longer is on this this fuse for this time bomb it really it could be tomorrow it could be a very smart trader I know he's been ahead of multiple financial crisis said to me the other day he said um it could be a few days it could be a few months because it is narrative dram of hummus has been closed for months like the markets don't act that way. The markets are on an absolute crazy run right now. And they're like, "Oh, nothing's problematic." There was an FT uh article that quoted a an investor saying, "It doesn't really matter whether it's opened or closed," which is horrible, but true. At least in the markets, the price of oil doesn't even at least in the Brent crude oil index doesn't seem to be connected to reality. So what will shift it will be either an egregious event that I can't even predict like a big company dying which I don't know who that would be or or it will be something narrative based it will be a report that says something like only a gigawatt of data center capacity got built. I really think the data centers themselves are the biggest thing that will shift things because like I said the narrative is well we're being very patient with the buildout because when the buildout is complete then the money will come in. It won't but like that's that's what they believe. So then it becomes, well, if they're not getting built, then you just spent all this money to fill out the equivalent of like several hundred warehouses full of GPUs that are just sitting in boxes, I guess. Cool. Great job. And that is the cataclysm that faces Nvidia because Nvidia will beaten race. I'm 100% sure next quarter, I'm sure. I'm sure the next earnings will look even bigger and huger than ever and everyone will say, "Oh, Ed, oh, you're worried about nothing." What I think is far more worrying is if Nvidia is pre-selling GPUs at scale and hyperscalers are taking ownership of them. This means that at some point it's going to become economically unreasonable to even install them. If you have Blackwell GPUs and in a year's time you haven't installed them, but Vera Rubin is out, why would you install them? They're less efficient. They're not as powerful. Kind of a waste of money. You only have, especially with data center capacity taking so long to build, if anyone does any write downs or selloffs, anyone dumps a bunch of capacity, that would be bad. But really is if a narrative takes hold that data centers are not being built, which I think is the case. That's what's that's what begins a series of dominoes falling. Messy metaphor, but you get my point.
In a similar vein, there was a report in the Financial Times I saw, which talked about banks offloading different AI related debts, sometimes at discount rates just to get people to buy it. I mean, given what we were just talking about, what what does that tell you?
It tells me two things. One, it means that banks are getting very, very, very nervous. Banks are getting really concerned about the event. I mean, the FTPs talked about them choking on the debt. That's just to be clear, I ain't no business investment genius, but you don't want to be choking on something, at least not in this case. And that's the thing that alone is worrying because it suggests that their hunger for data centers is actually not that much bigger than it is right now. It means that they are now going to not make as many of these loans. The second part of this is the real problem though because the only way that Nvidia keeps growing is if data centers get built. Well, if data centers are actually getting built that is but if even if they are if because these things are so expensive upfront. What happens if the debt don't you? No debt means no data centers. No data centers means no GPU sales. Nvidia number go down number go down damn bad and I think that people need to realize that we've got what 113 gaw just made up numbers at this point of supposed capacity being built by the end of 2030 or what have you.
For that to be possible we need nearly unlimited debt. We need trillions and trillions and trillions dollars more debt. And we need it to never ever stop. There cannot be anything that gets in the way between the money people and the GPUs. Nvidia relies on it. The AI bubble relies on it. Sam Alman, Dario Amade, they rely on it, too. That's the thing. The if they're choking on the debt right now, buddy, buddy, you How do the hot dog people do it when they're doing the eating competition? You got to do that. You got to got to find a way to take take more debt. Because that's the thing people the one of the classic defenses here is oh there's tons of money sitting on the sideline. That doesn't mean it's all going to AI even though it needs to. So this entire industry pretty much only makes sense with unlimited debt and unlimited resources and also very fast data center buildouts that just aren't happening right now. The markets have not been convinced of reality. All it will take is that is that to truly set in. And I think it will happen. But good lord, they're already they're already choking on debt. They're already there. That also makes me ask, when are they going to be paid off for the debt? If they're selling it at a discount, that means others don't think it's super valuable either. And this is when private credit comes in. There's a trillion dollars of private credit money in insurance companies now. retirement funds, your 401ks, your pensions are going into AI data centers now. And this is where the real calamitous stuff starts. And I deeply worry about it because as bearish as I am about this, as much as my job is talking about how bearish I am, the actual knock-on effects of this are quite scary. Vanguard heavily bought into Nvidia stock. uh private credit private uh you've got private um sorry asset managers like uh Athe uh Apollo for example with their Athen retirement life insurance side they're investing in AI data centers Blackstone doing the same thing they're putting pension funies in AI data centers that is really bad especially because for those deals to make sense well you have to build the bloody things they have to get finished and then someone has to pay for them I don't know who's going to do that I can't find proof that anyone wants to for things to like stay on track with with the funding there's going to have to be at least another another few rounds of hundred billion dollars for anthropic hundred billion dollars for open AI and it kind of begs the question of if people are already saying maybe we don't want to be getting into this now how many more rounds do these companies have before people start running for the hills and that's kind of my point it's on top of the trillion or so that we need in debt just to do the data centers that everyone's promised. We also need I mean uh hyperscala backlogs the revenue backlogs I'm going some with this I promise uh 748 billion dollars of Microsoft Google and Amazon's backlog of all of their revenue upcoming is OpenAI and Anthropic both those companies lose tens of billions of dollars a year which means that they're going to need to raise I'm going to say another $400 billion and if you look at the last funding round for Anthropic it is like godamn we are the world. It's like live aid but for old old men like Dario Ammedday 37 investors I think it was 37 we're running up against the limits of capital of deployment of capital and Anthropic is currently raising a $50 billion round. Sure they're probably going to close it. People are feeling really stupid, but $50 billion just isn't going to cut it. They're going to need at least another hundred billion in the next year. The more they grow, the more they lose. And I have so many questions about their $45 billion of annualized revenue. I just don't believe I I think that they're pre I think they're pre-selling API credits and counting them within the month, but we'll find out in the future. Thing is, you're right. They're going to need hundreds of billions of dollars more. data centers is going to need good trillion dollars more. The money's there, I guess, but at some point there's going to need to be some payoff. And I think you're going to see in the next 3 to 6 months that's going to happen. Something could happen quicker. If the market just gets convinced that something's wrong, yeah, something bad might happen. But now I'm hearing a lot of questionable things about Open AI. There's a lot of market sentiment that open AI isn't quite as strong as people think. And you'll hear people say now, it's one of the funniest ones that we're in an open AI bubble, not an AI bubble. Well, well, here's the problem. when open AI takes up so much of Microsoft's capacity and represents 70 plus% of their revenue and in AI and they represent what more than 70% of Coree's revenue and uh Microsoft's whole deal with Nebius is to provide compute to open AI and most of the most of the consumer revenue and AI is open AI dies and everything falls apart but people still haven't quite worked that out. I think what's most interesting though is that even now even where I am today and where a lot of people have reached it's still fringe to say this stuff to say oh yeah the company that expects to lo that needs $800 and something billion dollars through 2030 I don't think they're going to make it. You say that to people and they still look at you like you got three heads and I think that's just a testament to how illogical things aren't right now. It isn't just the market either. It isn't just the public stocks. It is, I think, an investor community and a society driven completely mad by the existence of social media. I think social media has created I was just talking to an investor about this the other day. I think social media has created this insanity when it comes to investment and investment bubbles because you look like the.com bubble would have been so much worse if we'd have had Twitter during the com bubble. You'd have had people threatening to kill me for not liking JDS unif and Enron. You'd have people going to war because you weren't big fans of Lucent Telecom. And I mean, sorry, Lucent Technologies. And it's just none of this would be possible without the chaos and psychosis of social media pushing it up, without the societal pressure in the media as well to say we must adopt AI. AI is the biggest thing. And I've said it before, but the narratives came pre-baked. Oh, they're spending too much money. Well, Amazon Web Services cost a lot of money. It didn't. Cost $52 billion over the course of 2002 through 2017. It was profitable at the end. Uh it's just like Uber. Incorrect. Uber is a completely different business. It's just like the com bubble. Actually, it isn't. the dotcom bubble was they're saying, "Oh, well, it's better than that though because the hyperscalers, they're all got beautiful free cash flow except the Financial Times has found that in the next few years their free cash flow is going to drop down to $4 billion from the 20s and 30s. People were ready to accept this bubble and social media and Sam Alman honestly and well Sam Alman used social media. Darama day used mainstream media to spread these noxious things about AI. And because the market was desperate cuz we're out of big ideas, big hyperrowth ideas like the smartphone and the internet, they're ready to eat it up. The problem is is that the contagion is now everywhere. The Korean stock market has hit new highs. pension funds, retirement funds, insurance funds invested in data centers reliant upon the yield from the payments on those debts from unprofitable data center operators whose customers are unprofitable AI startups. This is also a sign that we need new regulations. We need to make circular financing arrangements illegal. I think we should make it illegal to disclose contracted revenue that can't be collected in the next 12 months. I genuinely think that there needs to be rulings around major counterparties like Nvidia not being able to invest in customers like Iron or Nebus or Coreweave or Lambda. By the way, uh they get this is this is what drives me insane. Iron who is a former Bitcoin miner who became a cryp sorry former cryptocurrency miner who became an AI compute company yesterday announced uh that the the Nvidia is now their their largest customer which is what Nvidia did with Corewave as well and they also did that with Lambda as well. This should be illegal. This is and also just putting aside the illegality of it all does this sound like a healthy industry? Nvidia has said in their last earnings that they've through 2032 they are spending $27 billion on cloud compute on renting back their own GPUs. Does this sound like an industry with real demand? If we're apparently in capacity constraints, how the hell is their capacity left for Nvidia to buy? The answer is cuz no one's buying it other than OpenAI and Anthropic and their weird tendrils and of course crazy Mark Zuckerberg. Wacky Mark.
It does make me wonder if if so much weight is going back onto that circular. It does make me wonder if we're going back onto the circular financing stuff. If so much weight is on the back of open AI and anthropic, could we maybe see companies like an Amazon or or Microsoft essentially propping up these companies forever or maybe even eating them and taking them over in some way?
I think so. You're already seeing them propped up. you uh lo two weeks ago, Google put $10 billion with an option to put $30 billion in. Amazon put $5 billion to put 20 may put 20 more into Anthropic. But those numbers are actually quite indicative of the limits of their appetite because it was $10 billion from Google and up to $40 billion. If they truly could prop them up, they would prop them up more because Google, to be clear, invested at the old valuation 380 billion for anthropic. So they could have really made out like bandits, unless of course they don't think that there's a reason to do that, unless they're not sure that their shareholders will respond well to $40 billion going out the door. I do however think that there is a compelling argument that Microsoft absorbs open AAI at the end. I also my chaos pick which I think is unlikely but not impossible is they merge anthropic and open AI. That's a cute idea. XAI Anthropic also throwing that out there. There are so many funny ideas. says they're already living in his house. But nevertheless, I think that they are already propping them up and I think that they are going to continue doing so, but in dribbs and drabs. But the thing is these this is actually the big problem that they've made. Anthropic and Open AAI have now grown too big. Not too big to fail. They can fail all they want. Trust me, there's nothing stopping them from failing if they want to or need to. bum. They are now too big for an Amazon, Google, and Microsoft to prop up. They're just too large. They they require too much money. They I there's just they need hundreds of billions of dollars for large entities that even though hyperscalers are very big and have a lot of money, they can still only do 10, 20, 30 billion dollars. And even then, their shareholders are going to look at them really angry like and say, "This better pay off." Right now, shareholders are stupid and they're going, "Yeah, AI wee." But that's the thing. I also think that, by the way, another thing, another sign that we're nearing the top is if Oracle and Anthropic sign a compute deal. That's that's how you know that's how you know things are going bad. But yeah, I think in the end we could see a merger. We could see an acquisition. We could see just a collapse. Microsoft still own still owns OpenAI's IP. They could just kill him. They just be like, "Yeah, off you go, Sammy. No more. No more Sam for me. Thanks." What a delight that would be. But I think that it will be really weird and probably involve some strange antirust action. Like America's half asleep antirust people will wake up just completely bolt up like David Saxs will smash through the wall like the Kool-Aid man and start saying weird things about China. I think you could probably hear a sovereign AI thing where the US government talks about it. But the problem is is that even with acquiring them, you are also acquiring all their customers and all of their customers compute demands. And right now it's very clear that those compute demands are costing Anthropic tens of billions of dollars. They've had about $ 108 billion committed to them in the last 5 months. That's insane. Anthropic and OpenAI are basically the same company now. They they got both got deals with Core Wave. They both need hundreds of billions of dollars. They both have no path to profitability. And the CEOs are both extremely annoying. They're the same company. So really, it's just a question of who runs out, who what breaks first. Is it the narrative or the business? Maybe a way to sum it up is that the AI the AI boom or the AI bubble is fundamentally just a revenue model for Nvidia. Hyperscalers are losing money. Model developers are losing even more money. But it kind of begs the question of what what are hyperscalers going to do if they don't start seeing a payoff soon and kind of what does that mean for the AI bubble as a whole? The circularity is quite beautiful. I think what they thought they were doing in Microsoft, Google, Amazon was they were building their future customers that would after a few years their kids would leave the house and keep paying them. Okay? Eventually the kids will stop paying rent and I think they thought that these companies would become real businesses at some point. They never did. In fact anthrop we forget that Anthropic's revenue growth was very sudden and aggressive which means their compute demands were very aggressive too. And I think that it is an AI bubble made up of feeding money to Nvidia. But you need to go back to 2019 and remember Microsoft when Nvidia acquired Melanox which was I think an Israeli uh networking company which allowed them to do so when you do a big AI cluster you don't just plug them in together you need to use high-speed networking infiniband is Nvidia's tech for that when they acquired Melanox very soon after Sachinadella and Nvidia sorry Sachinadella and Jensen Huang announced that they were doing some vague partnership and suddenly open AAI got a billion dollars worth of compute credits Everyone's involved in the scam. Big big shout out to just Daario. He's a reporter who basically was the only person who talked about this. So, from the beginning, there's been kind of a kind of a no scams maybe too much a business plan to find ways to feed money to Nvidia because by feeding money to Nvidia, Microsoft justified a bunch of capex because CEOs only know how to hire people, fire people, and spend money. which gave them a chance to say, "Look, we've built a brand new cloud computing business line and we own all of it. We have the biggest company in it because we feed the money to them to pay us." And initially, this was a great plan. They would feed their money to themselves. Their revenue would increase on paper and because their other businesses were so profitable, didn't really notice the fact that they were basically eating cardboard that they were just feeding themselves money and losing money the whole time. Anthropic came along and it was a great exa Amazon and Google they were like hell yeah we could do the same thing we can justify spending all this money and we can justify feeding ourselves money and because the markets run by babies and the tech and business media are fully captured we know that no one will ask how much actual money are you making from these companies the idea I think was that the system involved yeah we'll just feed money to Nvidia feed money to ourselves and then at some point something magical will happen where these companies just feed us money and we own them completely because OpenAI and Anthropic do not own their infrastructure. They own nothing. They are asset light burnheavy insane kind of company. So Microsoft, Google, Amazon believed that they'd invented their both their new business line justified massive capex cuz they built the biggest companies in the world. And then what was obviously going to happen was they would become profitable and they would keep feeding money to Nvidia. Nvidia would keep hyping the AI bubble and then profit would come out. Except the profit never never came out. Never came out at all. So yeah, at this point it's basically feeding money to Nvidia. I should also say it's pretty. The Neocloud stuff is really bad. Iron, Nebius, Cipher Mining. Cipher Mining is a great one because they're not in that's a TPU one for Google. Cipher Mining and Terowolf, again, two former crypto companies turn into AI companies. Um, Google has had to backs stop them for billions.
Every time you look for the supposed gold mine, you find someone just handing someone money instead of the gold coming out. I don't know what convinces the market of reality because reality isn't doing that on its own. And I think that it has to be Nvidia missing. It has to be my three horsemen of a data center in progress being cancelled, a data center that wasn't built yet being cancelled, and then a fully operational data center dying. We need to see signs that data centers don't work because it all comes down to the data centers because it certainly isn't about the businesses built on using them.
Well, Ed Zitron, thanks for taking the time.
Thanks for having me. If you enjoyed today's episode and you want to hear more of the tech report, please consider liking and subscribing. Also, you can get episodes of the tech report wherever you get your podcasts.