[0:00]I think circular financing should be illegal. I don't think it is an honest way of doing business. It helps establish, it's funny, this era is sold on this liberal democratization. But what it actually is is helping retrench the largest players, except the largest players gambled hard on this, thinking that it would magically turn into something profitable, and it won't. And they, we have the two load bearing fail sons of Anthropic and Open AI, who must grow. Because if they don't grow, Amazon, Google, and Microsoft have wasted their capex. The core business of all of these companies is slowing down. The only thing they have to patch it up is AI, and the only way to patch that up is to feed Open AI and Anthropic, the circle is complete.
[0:52]Joining me on the tech report is writer of where your edits and the host of the better off flying podcast Ed Citron. Thanks for joining me as always. Thanks for having me. So there's two big things I want to talk about this week. There's the Nvidia VP of applied deep learning who admitted that AI is more expensive than a human worker, which obviously throws massive questions on the whole, the whole idea of the AI builder. But I want to start with the news that Open AI failed to meet its end of year 2025 user targets as well as, I think is at least a few monthly revenue targets. Open AI called this, I think it was ridiculous and clickbait, even though it's true. What, what does, what does this mean for Open AI? Here's the thing, the reporter of the Wall Street Journal who did that story, a guy called Bubergen, uh, he is not a clickbait guy. Buber is not that kind of guy, he doesn't do stories like that, and his stories are written with a, a calm countenance, they are not fluffy at all and indeed, I would say there are times where he pulls punches. Not as an insult, but he's a fair reporter, so this is a real story, and it's a real problem. Because the specific thing that Sarah Freyer said was that she was afraid that revenue growth was not there and that it would not keep up with their compute commitments and they would not be able to afford them, which is a pretty big deal when Open AI represents 70% of Microsoft's AI revenue. And so, I think the magic is gone. I think that it's kind of slowing down. To be clear, if Open AI wasn't losing what I probably five to 10 bucks for every dollar they made, it would be an incredibly high revenue company, two billion a month in revenue, if you believe that, which I have questions on. Point is, it's partly I think Anthropic and others, but I think it's also partly not everything grows forever. Not everything can keep growing, and Open AI, I think the number is, they are going to make 600, they claim, they claim, they're going to make 673 billion dollars in revenue, and they're going to have 852 billion dollars in burn by the end of 2030. They can't afford to have revenue slow. So there's been a charm offensive with Sam Altman going online and saying he was drunk. That's a real thing, which is one of the saddest things I think I've seen in my life. He was like, yeah, I've been drinking. Ooh, go weird uncle energy. They've I've never really credited him with uncle energy, but he's got it's very strange, divorced uncle energy. But he's going out, doing the charm offensive on Twitter, trying to take advantage of the fact that Anthropic keeps crapping their pants. Their service keep going down, uh, they're having they're spending more money, spending more tokens, using Opuss 4.7, there's issues with Claude code still. But nevertheless, Open AI is in a bind because it's not just that growth is slowing. It's that they cannot afford it to. They have what I think I read recently that they've agreed to 10 gigawatts of compute capacity, well not that really any of it's built, but that's a hundred to 140 billion dollars of revenue a year for those theoretical compute providers. Open AI can't afford that. Most companies can't afford that. I think Microsoft's Opex is like 150 billion. What are we doing here? So Open AI also has another fun story. It's leaked to the information this week, this Germain. They believe that 80% of their $20 a month chat GPT plus subscribers are going away. So they're just losing 80% of their largest business line this year, but the good news is is they're going to make it up with 109 million chat GPT go subscribers at five bucks to eight bucks a month with add with ad support. Now I just want to be clear, that's 109 million new users by the end of this year. I don't think that happens, and that story got, it really, I saw every coverage of that. Focus on one thing, which was that, oh, well, Open AI's going to have 109 million new users, not Open AI's largest business line is falling apart. And these are Open AI's own projections as well, that should be cause for alarm. It isn't. The tech industry, the tech media, the business media, the finance media, completely captured, refuses to bother even thinking about things for two seconds. Open AI may just explode one day, or more slow down and just crash into something, it really depends on your metaphor for the day. So, in reaction to the news of Open AI missing its targets, stocks dropped, I think we saw with Nvidia, Microsoft, Broadcom, Corweave, Oracle, and of course, Softbank, all down. And I think that kind of goes some way to demonstrating the insane levels of interdependence and circular financing that's all going on at all at play here. How, how big of a fallout is there if Open AI misses its next targets as well? Well, I mean, Broadcom, they said they were going to build 10 gigawatts of data centers with Broadcom by the end of 2029. By the way, they're yet to make an order, as of the latest that Broadcom have said, they haven't made an order. They are a large client of Corweave, and the thing is, it's not just Open AI that is as well. 67% of Corweave's earnings in 2025 came from Microsoft, who was selling the compute to Open AI. Kind of double booking revenue, not really. FCC if we had one should look into that, uh, I mean, Microsoft, Open AI makes up Microsoft claimed to have 37 billion dollars of annualized revenue in AI. I reckon 24 billion plus of that is Open AI's compute. So any connected counterpart to Open AI is in trouble if Open AI can't pay its bills. The other thing is with Open AI's revenue is a lot of it is in revenue backlogs for data centers that aren't built yet. Unclear when that will happen. But yeah, the knock-on effects of Open AI's slowing growth are that you've got, I mean, at least half a trillion dollars, maybe more, worth of backlog of remaining performance obligations across Microsoft, Amazon, uh, maybe a little bit of Google, and definitely a ton of Oracle, like most of Oracle's backlog, I think they have 500 something, at least 350 of that is Open AI. These are, to be clear, remaining performance obligations, I think are a fake thing. I don't think that companies should be allowed to present them without them, actually, the regulation should be they could only do 12-month horizons. They can't do beyond that, because otherwise you have RPOs that are just into Infinity. We'll have talking dogs and flying cars by the time that Open AI's paying for any of Stargate if they're not dead. So yeah, the knock-on effects are that Open AI is a load bearing fail son. They are a company that if they do not survive, then, I mean, a bunch of revenue goes away in AI. And it's all part of the illusory demand story. There's a capacity crisis because Open AI and Anthropic are taking up most of the world's compute. And so there's not really, it it's not so much that, oh, there's so much demand waiting in the wings. It's that what demand exists has a very, very small amount of compute it can actually use because of the two large Value Fail sons, Open AI and Anthropic, who I believe are just most of compute spend at this point. I don't think the rest exists. I actually before coming on tried to look around and find anyone who was spending more than 50 million dollars a year on AI compute that wasn't Anthropic, Open AI, or one of the hyper scalers. And the answer is there isn't anyone. Even Curser, who Elon Musk may or may not be buying with SpaceX, it'll either be 60 billion dollars acquired or 10 billion dollars break up. They're never paying. Open AI, I think X AI just kills Curser eventually, but I even then they will get, they're the largest independent AI company, I think, outside of Open AI and Anthropic. They were going to rent 10,000 GPUs from Elon Musk. That's still only like 40 million dollars of revenue. If the largest AI company that isn't an AI lab, only needs about 10,000 GPUs to train something, where is the rest of this revenue coming from? And I truly think it's just hyper scalers feeding themselves cash and going, look, look, look we money, money go up, number go up, or hyper scalers making their own systems to just pretend they're competing. It's very strange. You mentioned that Anthropic and Open AI are using a lot of the compute of of different companies or sort of the world's compute that's kind of on sale at the moment. Do, do you know, do you have some more specifics on what those figures are? So, okay, I worked it out that right now Open AI represents 70% of Microsoft's AI revenue, if not more, about 24 billion of annualized revenue. I think that Anthropic may represent 80% or more of the 15 billion in annualized revenue that Amazon has. So, I mean, Project Rania, which is meant to be 2.2 gigawatts out in Indiana for Amazon, um, that's entirely Anthropic. They have 500 megawatts of capacity right now. Epoch AI estimates that Amazon has about two gigawatts of capacity. I think Anthropic's three quarters of that by now.
[10:20]I don't think that people realize how much, but on a revenue basis, yeah, Anthropic is 80% of Amazon's AI revenue. Andy Jesse just talked about AI revenue for the first time at the beginning of April. He shouldn't have, that was a mistake. But these companies take up so much space, their services are extremely inefficient and dependent and demanding on GPU infrastructure. So, and the thing is, if there were other big clients, we'd know who they were.
[10:56]Because right now the, the story is, oh, number keep going up, everything look good. But when you look in the numbers, they never really talk specifically about AI, and every time they do, it gets kind of muddled. It's, it's like, oh, yeah, AI helped boost this in a way we're not going to break down.
[11:21]The thing is is that at some point, there is going to be enough compute for the current demand, and I'm not sure whether much more than that exists. Like, because, like I said, if there were other big clients, we'd know about them, but go and look at the big compute deals signed in the last two years. Microsoft, Meta, Microsoft, Meta, Open AI, Anthropic, Microsoft, Meta, and the occasional Google, Terrawolf, I believe was Google. But that one for Anthropic, I believe. It's always, it all roads lead back to two fail sons, Anthropic and Open AI. And the thing is, it's really easy to look at this picture and say, wow, look at the demand, look at all the GPU demand there is, no one can find GPUs. Yeah, because someone's taken them all up. If you're on a if you're on a bus and one very large man is taking up four seats, do you think that there's crazy demand for bus services, or do you just think one person's taken up half the bloody bus? And that's the thing with these companies, and that's actually the great lie about this whole thing. Is that the the crazy demand for AI never seems to manifest in somebody being able to give me a number that proves there's crazy demand. What it is is actually that these companies, Google, just funded another 10 billion dollars to Anthropic this week, with up to 30 billion dollars more. Amazon just invested another five billion dollars, I think another 15 billion dollars on the way. And Amazon invested 15 billion dollars in Open AI, they're going to do another 35 billion dollars if they get AGI or IPO. Microsoft invested 13 plus billion dollars in Open AI. It's just handing money, it's either hyper scalers handing money to Open AI and Anthropic who can hand it back, or hyper scalers handing money to Jensen Hong and Nvidia. And at some point, this falls apart. But my real question is, how many data centers are actually getting built? Because we've been over the fact that only five gigawatts of the ones meant to become come online this year are actually under construction, but I actually cannot find much proof of data centers coming online. Lots of stories about them not coming online. The stories about like say, Stargate Abilene operational in September of last year, one building out of eight. Uh, Project Rania, wow, it's completely operational. No, it's not. It's meant to be 2.2 gigawatts of capacity, so only 500 of it up.
[14:15]The actual it it but this is the AI bubble. The AI bubble is big headline that says number so huge, big thing happen. When you look in the actual body, it's thing happened in future. Number not so big, number you think won't big, actually pretty small or we're not going to say it. And the funniest thing is on Google's earnings this week, my faith things happen. They said, oh yeah, we have users so many users of our TPUs. And they named three companies. I forget all three of them. Boston Dynamics, uh, thinking machines, I can't remember the third one. And then they had to go to the information afterwards and say, actually, only Boston Dynamics is using TPUs. So there's crazy demand for these things. If they were if the demand was so crazy, you'd be able to come up with three people that use them. And that's the thing. That's the thing. No, I think only Anthropic is using them. And we don't Google are smart, they didn't disclose their AI revenues, because they know I would go through them or someone else would. But that's the thing, whenever you ask for proof here, whenever you're like, hey, so this is, you've been spending what Microsoft's over 300 billion in capex, Amazon over about 300 billion, Google in the 200s, meta in the 200s, maybe more, forgive me for not having the numbers in front of me. And when you ask them to explain what AI is doing, you're talking to the Riddler. It it's just like, riddle me this, what if a number went up, but another one went down, and the growth was so big, and you're just like, I don't know what you're talking about. Satchin Nadella of Microsoft got asked on the earnings call, so, how are you seeing the return on investment? 200 word response. No numbers, just like, yes, usage is up and usage is good, and AI is so big and so huge. But because we have no SEC, because we have analysts with rebar in their skulls, and we have a captured business and tech media, we are left in the situation where these companies are going to blow so much money. Amazon's nearly at negative free cash flow. It's unbelievable, and people will say, oh, it's just like Amazon Web Services. Wrong-o. Between 2002 and 2017, Amazon spent 52 billion dollars, and that's normalized for inflation, on Amazon Web Services, and that's all of their capex. That's all of it, including amazon.com. The store, there is no comparison here. The only comparison here is between functional businesses and the dysfunctional businesses of AI. So, if we go back to Open AI's missing its targets, it's you mentioned earlier that to to justify their 600 billion dollar commit compute spending, they need to basically double their revenue every year from 2028 to get to 280 billion by 2030. Yes. But considering they just missed their targets, what kind of growth are we going to have to see for them to even get close to closing that gap?
[17:42]Based on what I I've been coming up with last week, it's about, they have to 10 X their revenue from here. That's that's the long and short of it. They have to 10 X their revenue, because this year they're looking to make about 25 billion dollars, they actually have to more than 10 X, because they need 284 billion dollars in 2030 and to be profitable. That's the other thing they say happens that year. I don't know. Uh yeah, they would have to 10 X their revenue, it's that simple. And yeah, they can't. I just that's not happening. I don't think they're alive in two years. I think they're dead as disco. And I think disco actually based on YouTube is quite alive compared to Open AI, but the thing is, no one, it's weird how few people want to talk about this because when you start thinking about to your question, the slowdown of Open AI, the rest of it begins to feel increasingly more irrational. Because you're like, okay, the only way that Oracle can justify their 384 billion, sorry, 348 billion in capex that they'll spend on all of the Stargate data centers is if Open AI can pay them 75 billion dollars a year. Open AI has to grow. Has to, it's no longer a, they should grow. It would be interest, they need to grow, they must grow now, because 250 billion dollars of incremental compute on Microsoft Azure, 138 billion dollars over eight years on Amazon Web Services, indeterminate amounts of money on Google Cloud, uh, 22.4 billion dollars on Corweave over five years and 20 billion dollars over three years with Cerebras. So that's a lot of money, and that's more money than they have. Even their 122 billion, probably raised 60 at most. Yeah, they can't afford that. And what gets a bit scarier is when you go and look at how the money's been raised as well. Softbank, Softbank has to do a bizarre IPO of an unknown Robotics company, they're trying to do a 100 billion dollar IPO there. They're doing a 40 billion dollar bridge loan to fund this. Even if all of this works out, this is how much Softbank has had to do now. Open AI needs them to do this three or four more times, maybe more. Amazon had to split up the investment, 15 billion up front, 35 billion dollars if you can do other things. This is how difficult it's going to be if not more, when they keep growing. They need that to happen. They could go to the bond and the debt market, but I can't find a company in history that has actually raised more than 50 billion dollars on an annual basis. And those were real companies. I'm talking about the aged form of GE Capital before Sa and stuff like that. I'm talking about Microsoft, even Microsoft's not done that much. Oracle has, look what's happening to Oracle's stock. I think that people are very flippant about how bad things are economically, and I think they are looking for any excuse to ignore the big blinking, blaring warning signs that we see every month. And the thing is, Open AI cannot afford to slow down. Neither can Anthropic. Anthropic raised what, cumulative, I think it would be another 60, 70 billion dollars between Amazon and Google. They're apparently working on another funding round for another 50 billion dollars, at 900 billion dollar round. Anthropic also has made 10 gigawatts of compute commitments across Amazon and Google. It's this doesn't look good. There's not enough money to to substantiate all of these fail sons. How are we possibly going to keep them alive? I also think that we need a regulatory body, if one existed, to ban this kind of thing. I think circular financing should be illegal. I don't think it is an honest way of doing business. It helps establish. It's funny, this era is sold on this level of democratization. But what it actually is is helping retrench the largest players, except the largest players gambled hard on this, thinking that it would magically turn into something profitable, and it won't. And they, we have the two load bearing fail sons of Anthropic and Open AI, who must grow. Because if they don't grow, Amazon, Google, and Microsoft have wasted their capex.
[23:59]How do you think the 90 odd thousand employee or tech employees who were laid off this year under the guise of AI efficiencies gains are feeling seeing the news that the person selling the picks and shovels is saying, is saying, well, actually, it's it's kind of easier, it's kind of cheaper to dig by hand?
[24:32]I think that the tech industry has massively, massively overplayed its hand. And also, I think they do not even realize how much they have pissed off everyone. This is a vulgar, wasteful, aimless, directionless technology. It is unreliable at its at its core. It is expensive in a way that nothing has ever been expensive before. There is no comparison.com bubble, nothing like it. The Sun Microsystems, uh, servers you needed was 64 grand, but you didn't need anywhere near as many of them. People are being fired on a completely fictional basis. They're being, people at Meta, I have heard, are being fired, who are performers. People who are very good, they're arbitrarily laying people off now. I think the tech industry is radicalizing people against the tech industry. They're radicalizing customers by making the products worse, by putting AI in everything, by taking up massive amounts of land with loud ass data centers that sound like a goddamn Dalek. They are firing people by saying, yeah, we found a technology that's better than you. Except the technology sucks and it doesn't replace anyone. At best, it makes people more more efficient in a way that no one's no one could categorize. Also, they're spending more money than ever before, while they're laying you off. They're laying you off, they're saying we cannot afford you, but we can afford another data center that will get built in three, four, five years, maybe, and uh will lose us money the moment we turn it on. That is more important than you, a person. Yeah, I think most people are going to end up hating the tech industry by this point. I think that by the end of the AI bubble, I think those people are going to be angry. And I don't think that the tech industry has any idea about it. I think the business idiots at the top, the Satian Nadella's of the world, they are so disconnected from real people's problems that they don't even know how to solve them anymore. Because how would you know what's going on? Do you think Satya, when do you think the last time that Satya Nadella talked to a janitor? When do you think he last talked to like a regular person? You think he walks into a shop? You think Mark Zuckerberg has gone and bought a sandwich? You think he's going and picked up a sausage roll? You think that any of these people have regular experience? You think Elon Musk goes to the public swimming pool? You think that any of these people have regular experiences anymore? When did any of them do any work last? They go to and from lunch. They wake up at 5:00 in the morning. All of them say, we wake up at 5:00 in the morning. To do what? To spend other people's money and waste it? That's the thing. When you have so many companies run by people who are so rich and so disconnected, of course, they're not going to actually realize this stuff sucks. I paraphrase a Twitter post, but it's like being kicked in the head by a horse every day. They don't experience reality. So they're going to keep doing this. And I mean, this is kind of what you're seeing across the board with these CEOs who are spending as much as an employee costs, while firing real employees because AI is so good. Because they are not the ones using the product. They're not the actual people doing work, and thus they are going to create things that are what they consider connected to reality, but it's a reality that they themselves have crafted, and can only last as long as, well, the market's are stupid. And yeah, the market's love being irrational, but look at how quickly they freaked out with Open AI. Look how quickly they got spooked by that. And that was a minor story. Wait until something big happens. Wait until something really shakes them, because they might be looking for a reason to run. They might be looking for a reason to run high. I mean, but they're also looking for a reason to doubt themselves, because even the most lead poisoned hedge fund kind of sees something's wrong now. And it's weird, because right now we're in the hysterical era where number go up so big. Eventually, that's not going to be enough. Eventually we're going to need to see real ROI from AI, and it's completely insane that we are three years in and we're still solving the riddle of the Sphinx. Every time we ask someone, hey, does this make more money than you spend?
[30:19]Well, Ed Citron, 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.
