AI Data Centers Are the Most Speculative Bubble in History w/ Ed Zitron

0:00Hello, everybody, once again from Newsweek headquarters at the top of the World Trade Center in New York City. I am Carlo Versano, and this is your weekly edition of The 1600. This country was stone cold dead. My guest today is fan favorite, love him, hate him, mostly love him, Ed Zitron. Ed Zitron's back in the cut here with us in the studio to talk all things AI. You can think of nothing more American than having competition in the large language model arena. And I think regular people can see that AI does not deliver what people promise. Generative AI does not have a business model. Mark Zuckerberg wants us to talk about meta in terms of super intelligence so that we don't think about the fact that he's making boring, also-ran large language models that do basically the same things as everyone else.

0:44Realizing that everybody made a huge mistake. Everyone has invested in this thing for really effectively no reason. Ed is the CEO at Easy Primary Research. He's author of the Substack, Where's Your Ed At?, and host of the excellent podcast, Better Offline. Ed has been talking for years about how the artificial intelligence industry, as we know it, is really built on a house of cards. And lately, a lot more people have been listening. Ed has been getting a lot of big profiles in the so-called mainstream press and the legacy media lately. He's become a mainstay. One of my favorite outlets, the Drudge Report, which is pretty much as good a sign as any that you've made it in the media, Ed.

1:23yeah it's the it's so strange but he just puts me up there every time i see your face up there i'm like my boy ed it's so strange as well because i was very like even like when i first moved to america in 2005 when i was at penn state i remember the judge report being like oh this but this website like my father knows what the judge report is and then the first time it happened it's happened like five times the first time happened i was like oh this is a one-off thing now i'm like Does he just – does he – he doesn't read me. He doesn't link to my newsletter. He just sees my YouTube. He's like, I've got to get that guy.

1:56I've got to look at this. I've got to get this guy on. I've got to put this guy on the Drudge Report. The man drives a lot of traffic still all these years later. He absolutely rocks. You've been doing a lot of writing and crunching of numbers on your newsletter, which is a must read, by the way, if you're not reading it. It tells you everything you need to know about this moment in technology. Your most recent piece is titled Don't Look Up, and it's basically a – I mean it's an opus. These pieces you write are a lot. I don't know how much time you're spending on them. Presumably a lot. A day.

2:24I mean, it's real heavy work. The piece I'm talking about, you're basically connecting the different like jigsaw puzzles here on how these big tech giants, the Googles, the Amazons, the metas of the world are increasingly tied to the fates of these two LLM companies, OpenAI and Anthropic. So can you just lay out in layman's terms why that is a concern? so right now so bloomberg reported this the other day and i also found analyst letters to cooperate it bloomberg reported that 70 of microsoft's ai revenues in their last fiscal year which just ended at the end of june their strange fiscal year 2026 was 70 of their ai revenue was open ai and analyst letters from ubs wells fargo and barclays said that the same is the case with Google and with Amazon Web Services.

3:12So they have now spent over a trillion dollars in capital expenditures pretty much to make money from two companies that they have to constantly fund. Right. This year alone, Amazon sent$50 billion to OpenAI,$5 billion to Anthropic. Google sent $10 billion to Anthropic. I don't remember if Microsoft was involved in the earlier round with Oaks. They're paying these LLM companies and then they're also making their money from the LLM companies this is the circular financing exact problem and so there's two problems there one it's inherently circular they're feeding money to themselves it's actually three two it's they've spent a trillion dollars to make in microsoft's case their total ai revenue for 2020 fiscal year 26 was 34.3 billion dollars which sounds like a lot until you realize they spent 261 billion dollars of capex to get that right not a lot of roi there no but also when you remove open ai share which is 24.1 billion dollars you're left with 10.3 billion dollars worth of revenue for every single other ai thing that microsoft has without exception microsoft is the second largest infrastructure provider of compute in the world behind amazon they are also the largest seller of software in the world and they are making single digit billions of dollars off of ai outside of open ai now people say 10 billion dollars is a double digit yes but there's two different businesses connecting the math is hard and people will say well billions of dollars is good not when you're spending telling your microsoft yeah also not when you're spending 260 billion but your point about microsoft is correct because it's microsoft is a company where i think intelligent cloud which is one segment of their revenue is like 30 something billion or 27 billion a quarter so like a single digit or a 10 billion dollar annual revenue segment is enough to get Steve Barber to fastball you with a with a neuron chair and the other problem is is one of the notes I found from this guy called Stephen Zhu over at UBS from late June was that apparently 48 percent of Google Cloud's entire revenue not their AI revenue all so 48 percent of that is going to be open AI and Anthropic next year that is insanely bad because first of all it's incredibly centralized around two unprofitable unsustainable companies who've proven at no point that they could become anything else.

5:28And it's the fact that if Google doesn't make the money from them, then Google's growth will not match the analyst expectations that drive Wall Street's interest in stocks. And the thing is, this is the problem across the board. Microsoft's, it was in, don't look up. Microsoft's revenue based on analyst consensus, which is just all the financial analysts, what they all believe, they think that like 70 % of their growth in the next few years is going to come from cloud. And I think 33 % of that is OpenAI and Anthropic. So everything is on the back of these two companies spending not just a lot of money, but more and more and more every single year.

6:07And if that doesn't happen, Wall Street's expectations will not be met. It also suggests that the underlying businesses of Google, Amazon and Microsoft are slowing down dramatically. And so all of these big expectations around these earnings and all of the good earnings season they just saw were basically anthropic and open ai paying them right with money that they kind of gave them so why aren't these stocks taking nose i mean i guess we have seen a lot of volatility in the tech sector rate recently but these stocks are still ripping right so none of these companies disclose their ai revenues that's the biggest thing also the mainstream media is just taking the and died on this one they feel they haven't really bothered because what happens is the world is dictated by the value of stocks and since 2023 honestly every single time there's been any revenue growth people have said oh it's ai not because of them saying it's ai in fact in 2023 at one point cfo of microsoft said oh yeah only one percent of one segment segment is coming from ai which is a teeny tiny amount i read a business insider article it was like ai is paying off and And it's because modern journalism is vulnerable to corporate, not espionage, shenanigans.

7:18Sure. It's the right words. Well, journalists are lazy and we don't understand this stuff by definition, right? It's not that people don't understand it. It's that they don't want to push back. They don't want to look. Because I understand the rationale. And the rationale is Microsoft, Google, Amazon wouldn't spend hundreds of billions of dollars on something that led to most of their revenue coming from unprofitable companies. They would never do this. It must be somewhere else. They must be growing because of this. what's actually happening is across the board they've been raising prices they've been yeah training their customers microsoft's the worst of them but google's done it as well also all of them other what not amazon in this case but microsoft and google especially have taken their existing software packages and just shoved ai into everything oh it's it's unusable and even doing that yeah microsoft can only get 10.3 billion a year and that's everything that's anyone else renting ai gpus that's anyone else paying for models and that's anyone else buying microsoft 365 copilot it isn't working but when we take a step back and if you're a journalist who's like oh yeah well ai must be working because they'd never waste this money right you just look at any growth and you go that's ai right now in this year the story slightly changed where now it is ai in fact ai is kind of propping up the growth except when i say ai i mean anthropic and open ai right them spending lots of money on compute which they can't afford to do which is why both of them combined have raised over 217 billion dollars in the last six months it's great it's actually crazy but the thing is the corporations know they can get away with it they know that there's no pressure from analysts or the media to say hey what are your actual ai revenues what is your actual customer diversity and analysts what's crazy to me is analysts know analysts are writing notes about this all the time i've i found three different one from wells fargo one from barclays one from ubs actually multiple from ubs it's like yeah you know customer concentration is mostly just these two companies but number go up right number i was gonna say this is just number going up and to the right so all good next year google is expected according to ubs to make 124 billion dollars just from open ai and anthropic that's crazy amount of money man that's that's so much money and that's also if that doesn't happen growth expectations to google cloud are not going to be met.

9:37Well, and that's the sort of thing, I mean, everybody's, I think everybody sort of agrees, or a lot of people would agree that we're in some sort of bubble right now, whether it's, you know, similar to the dot-com bubble or something new. All it's going to take is one piss poor earnings report or one bad revenue forecast for that, for investors to really look at this and be like, are people like Ed Wright here, is this thing a house of sand, right well it's that and also there's two things going on yes i think a bad earning season where they miss on guidance or they miss on estimates that will be bad for them i don't think that will snap it but i think it will start the collapse also i think amazon amazon definitely guided down for q3 this that they were going to expect lower meta did as well i think it's that and it's also the other thing is anthropic and open ai cannot just spend a lot of money they have to spend a lot more money and they have to keep spending it it's the thing that people don't really understand it's not just hey they're running out of money it's google microsoft amazon need them to spend more and more money which requires them to have the money to pay them it requires the three companies to have built the data center capacity and have them operational to make the money and then it requires them to keep doing all of this in perpetuity right open ai uh Anissa Cardizzi over at Wall Street Journal reported this a few weeks ago.

11:00OpenAI projects to spend$750 billion on compute through 2030. That's less than three and a half years away. And where's the money going to come from? How's this going to work? No one seems to be concerned about this. Because when you tell them this, I'm talking serious financial journalists, serious people, they go, oh, they'll just raise it from the market. Right. Imagine a regular person saying this. I'm going to get a$60 million house. And they're like, yeah, sure. How are you going to pay for it? Well, we'll figure it out. Oh, I'll figure it out later. No, you're getting lost out of the bank.

11:29Right. Okay. There's a lot there. I want to come back to OpenAI and Anthropic, both of which are going to go public one of these days. That's another question. Theoretically. That's the question I want to ask you. But first, I want to talk about everybody's favorite tech CEO, Mark Zuckerberg. Oh, yeah. Saving money when you shop at Menards. We have an excellent in-stock selection of Whirlpool appliances. We carry great brands at the best prices. ready to take home today. Whether you want new laundry appliances or need upgrades for your kitchen, you'll find appliances that'll make your life easier.

12:04View our selection of Whirlpool appliance options today on Menards.com. Plus, check out our weekly flyer and find plenty of great deals happening this week. Save big money at Menards. One of the best, one of my all-time favorite guys. Zuckerberg's been in the news lately for this manifesto of sorts that he wrote that's called the quote, the future is for everyone. And it takes about 6 ,500 words, which I read some of. I skimmed it. But basically you get the gist, right? He's laying out this sort of utopian vision for our AI future, which just so happens to align perfectly with Meta's business interests.

12:41Right. So he's joining this tradition of our most repellent, unlikable, unpersonable oligarchs sort of trying to spin this technology they're building as the answer to all of our problems. Altman and Amadai from Anthropic, the big ones. What was your takeaway from Zuckerberg sort of joining the AI booster crowd here? First of all, 6 ,500-word essay about AI. That's my job. What the hell, Mark? It's not as good as your work. Second of all, this f***ing piece, it was just this meany-mouthed s*** about super intelligence for everyone. And it does this and it'll do that. And it's nothing to do with what they're building.

13:19It's absolutely nothing to do with it. What even is meta-building? more llms more llms like everyone else is everyone's doing the same thing right some people do it faster some people do it better some are expensive some are not for the most part everyone just pays for anthropic and open ai meta has been an also ran since the beginning the reason i've not mentioned meta in the beginning is they don't have an ai strategy yeah just every few months mark zuckerberg moves moves people around desks around yeah they check he changes the titles on them and they're like fires a bunch of people yeah fires people pays way too much for some ai scientist goes all right now go right and then not just meta's culture is in the toilet as well people that are absolutely miserable i have friends there it's just same i've heard the same thing from people who work miserable but what he's trying to do is trying to make ai seem more than it actually is the same thing that wario amaday does it's the same thing that altman does they want you to talk about super intelligence they want to talk about oh utopian things so that you stop looking at the llms they're building which don't really do that which don't they don't and they don't have well if we're thinking about the idyllic form of ai it is this thing that is makes your life easier and is autonomous and right unprompted you it just knows and people will say well you need to feed it there no i don't i'm not building this product i don't have to i don't have i shouldn't have to do this for you it's artificial intelligence it's meant to be intelligent so to this day llms remain this thing where you have to tweak it and product and harness it and change the model and change the prompt and do this thing the way they describe ai is it's just this thing that knows everything about you does everything for you is private but also all-knowing and anticipates your need that's what software is meant to be software is meant to be something that you use that works the same way yeah a tool but also what we think of as ai is something that knows everything we're doing and is great and just works we don't think about it it just works lms are the direct opposite it's yeah every ai fan is like well actually yeah you it's not working for you because you need to do this and this in this case refers to a thousand word guide right well have you used this have you tried hermes if you try to try claw code you try to claw code you try to chat gpt co-work you try microsoft co-work and it's just like what are you talking this this i can't even keep these things of straight anymore i i've used and poked with them and the thing is with them is they can do more but they cannot do more consistently and it's unclear what they're actually good at honestly to this day people can't really describe what large language models do they're just like yeah they can generate summaries and all this some of them burn a bunch of compute to do more than that we have this odd hacking thing that's going around what's that the the models that are hacking oh they're hacking each other and what's funny is they're like wow these models are escaping the sandbox and they're doing this and this and it's like did you prompt them to do that did you well that's it strikes me is it that whole thing about how like oh we have to protect it these models are so powerful we can't give them out to the to the public that strikes me as just pr spin it is and also people are like well we can't let dangerous models into the hands of the wrong people right we already have they're called anthropic and open ai we've already given or allowed this technology to be used without restraint and regulation by malcontents and they are called dario amade and sam altman these people are not responsible the fact that they didn't just put the kibosh and training these things in this way the fact that they very clearly trained them to do some of these things and they're like oh it got out of the sandbox bro and what actually happened was they didn't set the sandbox up properly right it's just every one of these stories is so frustrating and again none of this is truly autonomous none of it's truly intelligent clearly didn't listen to the boundaries they gave it i don't know why i'm even anthropomorphizing it the software did not work as planned the software did not work as they programmed it and so we're in this weird stage where mark zockberg's like damn i'll go get me some of that ai right whatever that means right i'll give 14 billion dollars to alexander wong from scale ai and then he'll come in and nothing will change or at least some models that are slightly better and we will spend 180 billion dollars in capex and then and the thing is he's writing six and a half thousand words you know he's never said how much money he makes from ai he's never sent how that's going to get that and that's because he knows the media will slurp up the slot they'll they'll eat it up they won't question and there's some people who push but elizabeth lapardo at the verge wrote a great piece about it some people have pushed back but mainstream just like copy paste yeah casey newton had a good had a good sort of takedown of the of the Zuckerberg piece in his sub stack, or maybe it's not a sub stack, his newsletter.

17:57Although he's married to like an anthropic guy, so I don't know how much to trust what he writes. I really rate him. But to me, again, it's just sort of like a layman in this stuff. The thing about Zuckerberg is whatever he, you cannot believe a word this guy says about anything, right? Zuckerberg was, you know, public enemy number one when he was spinning the same yarn about how Facebook and then Instagram were going to connect the world, bring us all world peace when actually i mean they're just products to like make teenage girls anorexic and they are deliberately programmed right to get engagement and growth over everything else jeff horowitz's broken code is a fantastic book about this he's actually in that movie he's played by jeremy allen white uh but nevertheless the internal machinations of meta exist only to make more engagement they don't know how to build product anymore if they do they've he was talking about the metaverse a couple years ago yeah remember that happened to that what the hell was that what the fuck it what's insane to me is the media just like memory hold that yeah they bring up and go like wasn't that weird it's like no what was weird was how you fell for it what was weird was when you were like yeah we're all gonna live we're all gonna live in the metaverse everybody metaverse wow like you know our our floating heads and it was crazy what i liked about the metaverse was it was either describing online video games that already existed or something that cannot exist and also required like extra sensory stuff and just what was crazy was he also did a video that was making egregious promises and it's like i thought that companies couldn't just lie like that but these guys can't we don't regulate tech right we don't give a f**k but we just let them go but yeah this damp freak he always is just he's gabbing and gabbing and gabbing about all all and sandra but he never gets around to like how are we going to do this what's it going to look like because an ai product that actually worked would work it wouldn't require a bunch of prompts wouldn't require training it wouldn't require us to become ai first it would just work that's what ai is meant to mean instead we have this expensive annoying energy intensive thing that requires bunches of data centers like poison black neighborhoods and we have to hear these guys warble on like oh 6500 words mark last year he did another one called personal super intelligence it was basically the same thing because what they want you to do like i said is think of anything else other than what they're building yeah just just give us credit for what we haven't done yet please god well i think in the other guys like altman and ammo die wario ammo die as you call them um they they backtracked a little bit off of what they were doing which is just casually talking about how this technology is going to like end the world yeah and put everybody out of a job so that's a start i guess from a pr perspective uh right i mean whatever i I mean, like, talking to people you can't trust a word out of their mouth, Sam Altman, the other, what was it, last year or earlier in this year when he was like, yeah, you know, it takes a lot of energy to train an AI model, but it takes even more energy to train a child.

20:53Yeah. And he's just like, you're a sociopath. Right, bro. Shut the f***. Like, it just. No, he talks about, like, outsourcing parenting to chat GPT. So that's something that we want to do. The other day, the tweet he did where it was like, yeah, great use for chat GPTs. You feed your calendar into it and you have a podcast about your kids' hobbies. It's like, there's actually a totally free version. You could talk to your children. Where your children are present in their life. But far be it for me to order you around, Mr. Oldman. But I personally would just see my wonderful son I love and enjoy his life as he grows versus feeding my calendar into an LLM and then listening to a podcast ostensibly in front of him.

21:33Yeah. Well, he had another one that was like, I don't know how people even parented before ChatGPT. It's like, I do. I mean, every human, every parent in the history of humanity pulled it off. I don't even. This is the thing. Every so often he opens his mouth and he says something that is just so insane. Well, speaking of that, so I want to you mentioned this super briefly. What is the log line? What is the one word, one sentence definition of what super intelligence is or supposed to be? That's the thing. There isn't one. What they are describing when they say our official general intelligence, AGI, is a conscious, autonomous computer software that can do anything.

22:11Could do anything a human can. Yes, and that changes, and it's actually smarter than a human. Just a side point as well. What do you call a conscious being that you trap and bend to your will? I call her the slave. That's a slave. We're describing slave. I keep saying that because I really want people to realize that that is what they are discussing. Digital slaves for everybody. A digital slave, yeah. but they again super intelligence is a completely fictional concept everything that they get even close to discussing they're like there you go they brought up recursive self improvement the other day that's a fun one this is the idea that these LLMs are making themselves better by themselves I love it as well because they're just like f*** how are we going to make these things better what if someone else what if the AI did the AI for us It's like because they are all part, they want to reach the apex of capitalism, which is just they just get money for no reason.

23:06Like they own a thing. They already basically have. Right. I mean, does Facebook produce anything? I mean, Facebook at least provides a service of sorts. But these people want to be like, we don't want to innovate. We don't want to build anything. We just want to have a thing that does the work, both the work of pleasing the customer and building the product for us. And then we want to own it. And we want to just sit there and cast checks. But obviously this thing that's super autonomous could never replace the CEO. They never quite get to that level. But yeah, again, it's much like Zuckerberg's blog.

23:36It's just another way to distract from what LLMs do and don't do. Well, it's clearly become the sort of buzzword here. So Zuckerberg writing in the Wall Street Journal a couple of weeks ago, the defining question of our age isn't whether superintelligence will exist, but who will have access to it. Will it be centralized and restricted to a few institutions, or will it be a tool that empowers everyone? I guess just like Instagram has a tool that empowers everybody. What does that even mean, Mark? Well, so hold that thought. Mickey, can we play this thought of Sam Altman talking about super intelligence here?

24:07Well, we need to explain the world. Number one, we are close to creating a genie that can grant any wish. Number two, we are going to make sure that our first wishes broadly benefit humanity and that we kind of get the world to a place where a lot more people get to have a lot more wishes. wishes and number three the space of what you can wish for is incredibly big and creative and it'll be quite exciting to like figure that out with real sort of human values and preferences but but i think there's a fourth thing too which is exactly that you start making these wishes the computer grants them and then you're like i didn't think that was going to work yeah what now it's a weird feeling so he so he's not talking about as a gene as a genie in a bottle First question, what the f*** are you talking about, Sam?

24:56What do you mean by that? That's the thing. These people who interview Sam Altman are the most... Well, he never does real media. Exactly. He's terrified. He claimed that his PR people claimed he was too busy. Yeah. Yeah, too busy for this. But here's the thing. What I like about this is they've just given up at this point. They're like, what is it? It's a f***ing genie. It's a genie. It's a genie, I guess. You ever see Aladdin? It's like that. Yeah, you know it's like Robin Williams. He's going to do a Jack Nicholson impression. It's going to be great. You love it. don't let Jafar use it though I don't know what he means because it doesn't mean anything it's just he is and this is just a sign that the media is completely captured because any journalist and I've seen Jealousy interview let him babble on like this should just be like hey man quick question what does that mean just real not even aggressive just like what do you mean by that what does granting a wish mean because it doesn't mean anything he's just like genie it's a genie it's magic but the thing is you know that we're towards the end of a hype cycle when people start mentioning magic when people are like yeah it's magic he's magic he's magic deep you know this uh there's a genie in there there are famous stories about how bad genies go as well there are way more stories about like well it's supposed to the whole proverb or whatever you want to call it is like yeah the genie escapes the the body you can't get the genie back in the box like the genes in history and i'm saying that wrong oh like historically some of them have been quite evil right like they they have exactly bad ideas now he would claim well this is why we have ai alignment this is a malevolent genie it's just also like benevolent genie i should say yeah i mean probably if open but they aren't the other thing is they're nowhere near anything like that they're nowhere near they want to talk about that because if they have to talk about today what can it do right software is pretty good at tech support if you if you have a really specific the only thing i've ever used ai for is hey this random one program is broken here is a 4 000 word log dump it yeah what's going on and it will find out what's wrong that's awesome that's not worth a trillion dollars that's not worth a hundred billion dollars that's worth that's a 10 billion dollar company but it costs them it costs them hundreds of billions of dollars to get there and it's like wow you've slightly improved tech support cool right well the other thing i mean in our industry at least the thing is like, oh, it's going to become a better writer.

27:12I mean, my boss a year ago was like, AI will be a better writer than you, Carlo, within a year. I know. I was like, thanks. But, I mean, it is. And frankly, AI writing now is, you could clock it immediately. Yeah, it's not good. It's crap. It's garbage. But the other thing is as well is people always say like, oh, it's good at research. No, it is not. It really is not good at research. So I've got a Bloomberg terminal. And another useful AI thing, again, very low level, is that when you get information out of Bloomberg Terminal, it generates something called BQL, which is like a Python-style coding language.

27:46And instead of having to do that, it generates the Python, runs the thing, and you can directly check it. You can actually check it, which I do every time. However, you can't ask it too long of a question because then things start hallucinating, right? Still, to this day, they hallucinate. And the only way to catch it is if you're really paying attention, which is why I'm very... I watch The Little Bastard closely, and I don't use the LLM stuff very much. And the thing is, we are a trillion dollars in. We're more than a trillion dollars in. Is this the base you've got? And AI boost is like, aha, we found a use case, Mr.

28:20It's like, ah, checkmate, QED. And it's like, mate, you've been promising me a magic genie for four years. Well, I've been told, yeah, I've been told this technology is going to make it so I can work less, maybe. Maybe I don't have to work 40 hours a week. Is that happening? No. actually meta's coo cto andrew bosworth andrew bosworth he was like um he was saying yeah if you save time with ai uh you're not using it for vacation you should use it for more work especially with how meta is right now and famously boz many many years ago in 2016 i think it was sent a letter round to people at facebook saying that everything was justified under growth even a terror attack or a kid being bullied this real thing was reported by buzzfeed news yeah i remember that yeah yeah this is this again where's the techniques being like hey this sociopathic sounding guy is saying that no one should have vacation that they lay people off arbitrarily so to get right back around to the meta conversation that is what matter is and also ran trying to work out what's going on fudging things together and honestly altman and amaday are not so different they at right now anthropic is going around trying to set up its ipo and reassure the story from bloomberg or maybe the journal was yeah they're trying to reassure investors about the threat of open source and it's like i love to reassure my investors for an ipo and they're like but don't worry we're going to move into biology okay that's the new we're going to move into biology to deal with concerns over ai anthropic is legitimately claiming they're going to make new drugs and i just i the fact that every news outlet reported that was just without just being like shut up shut up right liar they have a robot yeah anthropic has a robotics so i didn't know that they don't make robots but i know that because they just hired from open ai open ai's robots supposed to go work well they're all poaching each other i was so sick i want to do like robotics and prop and get like 500 000 a year in s in um in rsu's to do a job where it's just like and this is probably not going to happen but like i need to be paid so i can read academic paper a skunk works job in ai seems like about oh that's a great great gig this is going to piss some people off i think a lot of the jobs there are adult summer camp yeah i think they sit around running compute experiments being like what if we did this what we did this what we did this what we did this what do we do most of the jobs in big tech i mean i worked briefly for a time i could say because i didn't didn't sign an mda when they fired me i worked briefly at robin hood that company nobody was doing anything that's so cool it's like nobody was working that's so it was yeah that's so sick it was awesome i was like this is the best job ever and then i got canned because i didn't do anything but like just copy my boss yeah right um okay and i want to i want to i don't mean to beat up on on zuckerberg actually i do i i find him to be one i find him to be a truly evil person large chunk of white yes no he's a he he i mean i've said this before i would work at philip morris before i worked at facebook yeah rtx oh yeah no absolutely yeah and that's all the arms manufacturers i can i would make raytheon tomahawk missiles before i before right um no so in this in this opus zuckerberg he he goes out of his way to mention this data center this 50 billion dollar meta data center that they're building in richland parish louisiana uh when this thing is complete it will use more energy than the city of new orleans great um but this data center, I mention it because it's become an example that they're trotting out to show that these things are good for their communities, right?

31:49Because the local teachers here in Richland, Paris, Louisiana, are getting$50 ,000 cash bonuses, basically as bribes from Meta laundered through the tax code. That's some values coming out. Well, I was going to say, well and good. Perfectly happy to see these companies paying off the local municipalities rather than just sort of like bulldozing right through them. However, in this particular case, they did that first, right? So the New York Times reported this particular data center project came together with all these NDAs,$10 billion in tax breaks for Meta, of course. Oh, I know they got the tax breaks.

32:20I should have assumed. There's so many of these deals to follow, but that's the thing. Yeah, okay, you should have given the teachers$5 million each. Well, right, and you should have made every teacher a robber baron. Exactly, and no public comment when they're actually doing the deal to build this thing. There's also a story now about this Google data center in Columbia County, Georgia, where the city has said property taxes will eventually be eliminated. I don't believe that. It's always eventually, isn't it? It's always eventually, right? But it's always in the future. So if you put aside whether handouts for boomers and public teacher unions are the best way of spending all of this additional tax revenue, what is the NIMBY argument for why it's still not a good trade-off?

33:02They don't create jobs. It's a myth. Even in the construction, this is my favorite myth to break. So people are like, well, there's all the construction jobs. those construction jobs are specialist jobs flown in right to the state right in many ways the days that is being built especially in the south are the west looting the south because those it's specialist talent specialist electricians h2x i talked these are not guys they're taken from the home depot parking lot no no it's so work harder um but these like fruit i talked to people in abilene yeah at abilene texas where they're building the open a data center very slowly this is hyperion now hyperion's in louisiana that's the meta one okay abilene is open yeah an oracle and in all the cases it's specialist talent flown in a ton of noise for neighbors and people like well it raises it raises property values what are you talking about no it doesn't but most importantly it does not create jobs because a data center when complete is a very small labor it's ideally they barely have the lights on in some of them right and they're again specialist talent your local joe schmo is not getting a job wiping the floors and open ai right they're not they need security alone they're not going to do that so what these things do is they take up space they are a lot of space a ton of space they're basically this is actually a good point to make really simple thing a data center when people describe it they think big building they should actually think small town both in the power drawer and the amount of power and gas and also just sheer space these things take up and the noise they make and the noise they make and the construction is bad the noise when they're finished is less yeah there's like a hum right it's horrifying and there's just constant 24 hours because the only way to get these things done late because they're all behind schedule is to have constant construction and then when they're done great they fly in more people who now live in the local neighborhood and what the boosters are saying is well look in the places where they're building these things revenues up at restaurants and hotels and stuff it's like yeah because people are visiting right it's basically a miniature tourist economy and then once it's done all those people all those will disappear and they will disappear with every dollar that every dollar they have these people are not locals they're not going to stay and they have no intention of adding any value once they're built if they get built that's the other thing about data centers a lot of these things are not being finished they take forever to build stargate abilene that thing is meant to be a 1.2 gigawatt data center so that's the power about 840 megawatts which is the actual it that works that thing broke ground in june 2024 maybe july it is now the year of our lord august 2026 and this thing has only got three operational buildings and i have been unable to through multiple sources confirming if a fourth is even operational or when it will be there are eight of the buildings there's eight of them and it was meant to be done at the end of last year the beginning of this year in the middle of this year and now i have multiple people saying me to me q1 q2 2027 at best and again media just ignoring this because there are no one gigawatt data centers none of them are getting built they're getting built in chunks but they're not finishing them and that's because the power and the labor and everything is so difficult because these are like building small towns they are insanely intricate ambitious infrastructure projects oracle is apparently building 7.1 gigawatts these things this is the new deal on steroids except only one horrible man benefits larry ellison crustiest of them all and what's crazy is can't complain about the media too much but i'm going to anyway this is happening in real time this is you can see these things are not getting built very fast but because there's such money behind it and the biggest smartest companies in the world are behind it people just look the other way they're just like well i they'll work it out if the best we've got is they're going to work it out we are screwed it's amazing because i mean there's there's a skyscraper right out my window here that i've been watching them put up in less than a year this thing is going to be done in less than a year a skyscraper in in manhattan and you're telling me it's taking three years for these things that are essentially warehouses and they're not even done they're not warehouses that's the other thing a skyscraper assuming it's condominiums or even just office space they are you have to cool them you have to power them or less but they are fairly even with a new build that's complex they're still you know what you're doing you have an architect perhaps it's good perhaps it's bad but a data center is depending on the gpus the amount of load that will be on them they are they require specialist h-rack just for a set of 72 gpus specialist liquid cooling which means you need actual liquid that can get you need piping you need specialist cooling for outside of them you need power that is you have separate uninterruptible power supplies for everything you have networking that needs to be locked down to the t so you need specialist networking you need all of these things and then you need to do that i mean technically stargate abilene's eight buildings but each one is four data halls so you've just got this multiplicative thing and then by the time you get this done you find out the gp is in there are two years well i remember you saying that last time you were here i was like that's a whole another problem is the heart and the hardware cycle on these things right thing is the depreciation question is actually interesting because it's unclear whether these things are dying in three years or not we actually don't like like maybe they're not like perhaps that's something i can conceive but at the same time he's come i've heard from enough people in the know who are like these things die in two years right so they could just be rotating them they could be rotating them through nvidia we don't know again these things were obfuscated but nevertheless what we do know is the Blackwell GPUs that started selling the end of 2024 are not all installed.

38:43NVIDIA, like, data centers are taking so long to build that they have not got all of the Blackwells, two-year-old chips in there. So by the time you got that in there, you've got to replace it with Vera Rubin. Except after the first generation of Vera Rubin used the same slots. Second generation requires an entire new rack. So your old, busted, nasty data center that you schlepped to build is actually not going to happen. Yeah, it's not going to happen. They're doing one megawatt racks now which is insane just it doesn't make any sense but nevertheless in two years don't think i'll make it that long it's going to be a completely different kind of rack type of gpu buy it again and it's also more expensive because the cost of memory has gone up which is screwing everyone when you say it all like this it doesn't sound very good i'm surprised people aren't more worried about this well that's the thing right so i'm very skeptical skeptical of the data center I mean, again, I have sort of like a simplistic view here as a layman, which is just that I don't trust these tech companies to ever do what's in the best interest of the public.

39:44Like, why would they? And they never have shown that they have before. So why would you believe them now? And also, these local governments are all on the take and the federal government is all on the take. So you can't trust any of these people. However, to steel man this – the data centers are good actually argument. first of all starting to feel a little bit like a moral panic with the um these things are taking up all the water and they're and they're polluting the environments i mean maybe that's true i have seen evidence of that true but there's part of it that's starting to feel like the new fracking or the new 5g build out like these are these are the root of all evil so karen howe who wrote empire of ai she made a she made a numerical error she overstated it the right wing damn they are right wing for the most part people that love data and it's like aha a single thing actually one of the problems is that it's not that they take up a lot of water is that they process it out and they process out all the nutrients to be used in the local ecosystems that's bad right the water thing is its whole other kettle of fish i'm not super well ridden it i don't need to be because all of these things are having such trouble building power that they're full of gas turbines and because these gas turbines are extremely because there's a seven or eight year backlog they're the older ones so they're belching gas right into the air elon musk status and is poisoning black neighborhoods these things are surrounded by these caterpillar generators they're burning tons of gas it's obscene and they are horrible environmental blights right and anyone trying to get around that i have deep suspicions of because of one thing on top of the poisoning of black neighborhoods and poisoning of regular neighborhoods which is why do you like these things so much what do you think and actually very simple question why do we need more data centers anthropic and open ai don't seem to be not taking money from anyone anthropic has troubles with stability but they're not driving costs down when they build more things why do they need more why do we need more i can get into the the 70 thing is actually a good point as well because the 70 of ai revenue coming from these two companies also begs the question of why do we the rest of these data centers but really why do we need more data centers and the vile centrist argument is well we need innovation in our communities again gun to your head explain to me what a data center does and why we need it here right because it's not content delivery cdns like akamai they're all like we have that infrastructure the delivery of ai services is not slowed you need regional support you need like an azure region so where the gpus are so you can get it to a certain amount of space, but you don't need to dot it everywhere.

42:16The only reason we haven't dot it everywhere is speculative investment. The demand for AI data centers is single digit billions of dollars when you remove Anthropoc and OpenAI. The reason that people are saying, well, actually it's good when we should have more of them, is they don't care to know. They don't care to think because they cannot tell you why we need another data center. So the constraint problem is people like, well, we have all this demand that isn't being met. No, we have two companies taking up all the space and what dregs left over other people can have when you remove those two companies every ai data center or inference company so providing the out the power of like open source models what have you they make like a couple hundred million a year each and it's like that does not we're not building several hundred gigawatts of data capacity to make that like a few hundred million that won't even buy all the gpus so everyone talks about data centers as this thing that we need it's essential we must have it we must build board because if we don't china i guess or if we don't build them we won't be in the future and this is 100 the tech industry's plan is to shove these things through and to prey on a media that doesn't give enough of a to actually check their facts and find out why we need them and actually ask dario amadette ask samlman ask mark zuckerberg why do you need this because mark Zuckerberg is currently making negative dollars on air.

43:39Like, oh, well, Mew Spark 1.2, don't care. Well, what about the argument, the more sort of just like realist argument that these are basically the only things propping up the American economy right now, right? I mean, you look at the latest jobs report. The only industries that are adding jobs right now are manufacturing and construction. Like this CapEx infrastructure boom is single-handedly keeping the stock market alive and probably the economy alive. if the only way I could keep alive was shooting a bazooka at pandas, would you want me to keep shooting the pandas at the bazooka? No, look, I've said that.

44:16But that's my point. It's like, oh, a bad thing's keeping everything propped up. So we just keep doing the bad thing. Right. We keep losing it. We keep doing this. And that's the thing. If your argument is this kind of nihilist, not yours, but if the argument is this nihilistic thing of, well, if we don't do this, then the economy will fall apart. we're losing anyway because it's going to fall apart anyway. Because if the best you've got is not, this is actually good and we need it, and it's just we need this because we've got nothing else, which is the problem I've actually put in the piece as well, then there is no argument for AI other than this is a very large con, which it is.

44:54It 100 % has just become a shell game of circular financing, of deceit, and of actually malevolent actions from these companies. I want to ask you a little bit about these IPOs, which we just hit on briefly. So this is your third appearance on our show, which has been great. We love having you. Since we did last have you, you've had a couple of viral appearances in the business press, CNBC and Bloomberg, where you laid out your bare case on AI and particularly OpenAI and Anthropic, which you argue should not be allowed to go public. Take a look here. Why do you think OpenAI and Anthropic shouldn't be allowed to go public?

45:31Because they are lossy companies. They do not have a path to profitability. Anthropic, they leaked, or however, that they were profitable. They were profitable in two months when Elon Musk, they gave them a discount on their compute through SpaceX. Clearly planned leaks. Clearly planned financial engineering. No normal profitable company does things like this. OpenAI and Anthropic are both horribly unprofitable, and they are not getting better. The information reported a few weeks ago, OpenAI had a negative 122 % non-gap operating margin in the last quarter. But again, just because a company doesn't make money doesn't mean it shouldn't go public.

46:12Just because a company is not profitable doesn't mean it shouldn't go public. Let me reframe this then. I don't think these companies will survive. And I think that investors are being, they're doing exit liquidity for venture capital. And I think that these are dangerous, unstable companies that have faced basically no real scrutiny for most of the years of their investment. You have journalists and analysts who are actively cheering them on. This is a dangerous precedent and it's a dangerous thing for retail investors who are being conned by this. What was the industry's reaction to your statement on Bloomberg that these companies should not be allowed to IPO?

46:48They got very upset. They were like, well, they should be allowed to IPO. The free market. And it's like, no, they shouldn't be allowed to IPO because they're a risk to investors because they are companies that cannot exist. They will run out of money. And also the idea is to take them public and try and dump them onto investors. Right. And even then that won't work. There's not enough money. Right. Make the retail investors the sort of like exit liquidity for these guys. Exactly. And then try and raise debt from the public markets for the worst time in history to do so. And also, just to be clear, OpenAI plans to spend$750 billion in the next three and a half years.

47:21where do you think that money's coming from venture capital it's not got enough money that's more think venture capital in 2025 was 600 and something billion um debt markets 750 billion will be more than the combined debt that hyperscalers are raising in the next two years so like whatever also i think i might be part of the problem so open ai i published there since we last see each other i published their auditive financials i got them i worked with the ft to verify them and what was interesting with that was when they went out was watching people go like actually this is good it's good they lost 20.9 billion dollars they're actually profitable if you look at it like this and what they did was they moved stuff out of cost of revenue in sales and marketing and the oh you don't learn that in business school you don't learn that it's just it's so annoying but i published that and then a week later the new york times reported that they're delaying their ipo right i don't know if i don't know if it was me because the specific thing it was around that time spacex went public and also the times mike isaac i think reported is great um the times said that one of the reasons was their advisors were like yeah you're not going to get a trillion dollar valuation i must be clear how bad that is because open ai's last private valuation was 865 billion so you can't get a 30 premium ipo for what's meant to be the messiah of tech that's terrifying that's actually really bad and i also this is my personal thing i have to wonder if the s1 so the thing they felt got public may have been different to the numbers i published maybe maybe it looked a little different i don't know i'm quantificating i will say my numbers are correct they're really correct i'm watching the ai industry wriggle was quite pleasant they they were so upset they were so mad at me the first the immediate thing i saw them just be like fake fake news fake and it's like nah mate i got this is why i worked with the ft on this one so that you you can do that the funniest thing though was that open ai someone i assume is a source unnamed i truly don't know uh was like actually they didn't lose 20.9 billion they lost 8 billion and it's just like shut the what the are you talking about man and it's just like you can do this wanky outside of the public markets you can't really do that exactly and the thing is with anthropic is they're definitely unprofitable too and i think it i so if i had to guess anthropic wants to go public in september or october they're either just going to push through with the economics ever and just be like we're going to grow to seven bazillion dollars in the next five years and the ose will clap or they're going to move training expenses into capital expenditures and be like ah we're profitable it's like yeah i'd be profitable too if I removed all my costs.

50:02But that is the trick of the AI industry. And the thing is, though, even if these dogs of companies can make it public, it isn't enough. They raise$50 billion each. It isn't enough. They're going to need hundreds of billions of dollars in the next few years. And if they fail to raise that money, Microsoft, Google, and Amazon, their growth goes away because their growth is mostly driven by Anthropik and OpenAI, buy and compute. So it's just this weird situation where it's not clear how it works out but it's not like they're magically going to fix their economics and there was this funny story a few months ago where wall street journal reported yeah anthropic had a profitable quarter and it happened to be a quarter where elon musk because they rent a compute from elon musk reduced the cost of their computer for two months it's just like oh okay when you remove my costs i'm profitable and it's just so it's so silly as Well, because the Wall Street Journal published that.