Monologue: OpenAI Could Kill SoftBank and Oracle

0:03 Media. What's up? It's better offline and I'm your host ed Zetron It's your better offline monologue for the week, folks. Has been a long few months, and I've taken the vast majority of this week after recovery, after writing about h two hundred thousand words in the last few months, not even including my upcoming book, The Hater's Guide to Silicon Valley, coming out in twenty twenty seven. If you want to support my work, I'm currently doing a week long sale for the premium newsletter ten bucks off the annual rate for life. That'll be in the notes. Please do.

0:43 If you feel like supporting me, do that. New's learned. The podcast are now my by principal source of income. It's happened. The era of smiles has begun. Join me and receive the bounties of words every week except this one, and I take it the week off. So let's begin, And I wanted to start by saying how you the listener might help me. First of all, my signal is Ezitron dot seventy six. The things I'm looking for right now that would help me are anthropics S one or any and all documentation of its cloud spend and financial condition.

1:11 Especially if you can get me something from Amazon Web Services or Google Cloud that shows me how much this company is spending. That'd be very very helpful to me, as would any and all information about the AI revenues of any major tech company that aren't public knowledge, and any and all information about the cloud spend of open Ai, Anthropic Perplexity, or any other major AI startup, and of course any of the financials of any AI startup. You can get me the same ghosts with token spend at any corporation. Please use signal, Please take photos of any documents.

1:41 Do not send me PDFs. Do not send me documents for your safety, m for mine, and please please please do not send me stuff that's already public. This isn't because I'm chiding you. I'm just saying people get excited that they want to help and they send me stuff that's already out there. I've probably already seen it. I'm a freak. I'm looking at this stuff all day. All right, folks, let's roll to the actar monologue. So earlier in the week, I put out a fifteen thousand word piece called the Open Ai Bubble, a massive guide to how open ai is the cause, enabler, and justification for the entire AI bubble.

2:11 Writ large and how everything kind of collapses when it dies. I wrote it because I've heard a lot of people say it's an open ai bubble, not an AI bubble over the last year, as a way of waving off potential detractors suggesting that open ai could collapse and everything else would be left behind him. Fine, you're living at a bubble yourself if you think this way, that's a crazy way to live one's life. And please, please please stop emailing me about the bailout. I already I've gone over the summits anyway. In truth, open ai is the largest consumer of AI compute accounts for over eight hundred billion dollars of remaining performance obligations across hyperscalars and neo clouds.

2:45 The people that rent how ai GPUs and basically just exist a funnel Nvidia money, and likely over one hundred billion dollars the future revenues of Broadcom and Nvidia themselves. I go into detail about it in the piece, but I kind of wanted to go over it a little today in today's monologue two. Open ai is also a material risk to both soft Bank and Oracle. If open Ai can't go public, soft Bank is going to face a massive liquidity crisis. It has forty five point nine billion dollars in debt and matures within the next twelve months, and they'll either have to refinance it or pay it.

3:18 Except that's going to be much more difficult because they've jettisoned most of its valuable stocks in t mobile and in video, all while pushing up against the limits of what it can borrow against its holdings in the chip maker arm which it bought I think think twenty twenty three for thirty two billion dollars. It went public at fifty billion dollars, but the holdings are worth one hundreds of billions of dollars. Except that's kind of a problem, which I'll get to in a minute. But soft Bank will never truly go bankrupt. That doesn't mean it's not a problem if open ai fails, because it will be put in a historic crisis, one worse than the dot com bubble, if where Masayoshi's done to bet everything like GeoCities,

3:55 and even the horrors of twenty twenty two and twenty twenty three when the Vision Funds had lost is of what thirty two billion dollars. Back then, soft Bank had plenty of other valuable stock and ways to raise debt. They had the Ala Barber stock, They had their holdings in Betfair and other things like that, except most of that is gone now. And now S and P Global is downgraded its outlook to negative, which means they think they might actually downgrade its credit in the future. And just to be clear, they specify that the reason that S and P did that was open ai. They said the words, they literally said it was the case. Except now, what the hell is soft Bank meant to do? It doesn't have anything else to flog really, other than it's

4:35 arm stock. And the problem there is that soft Bank is the largest holder of armstock in the world, which makes it difficult for it to liquidate too much without spooking the market. Think of it like this. If the largest holder of a stock says, oh, I'm just going to start selling it off for liquidity, it officially stops being about the stock in question, but the company's selling it.

4:54 The value of arm will turn into a bet on whether soft Bank can afford to pay its own bills. It's already taking out a bunch of margin loans on the stock and further lands will be difficult to raise for the very same reason. While it would be likely rescued by the Bank of Japan and other local financial institutions, there has never been a more dangerous time for Masayoshi's son.

5:14 Check out my hater's guide to SoftBank. I know it's a premium thing, but there's a deal going on. It's a beautiful deal. I stopped doing the Trump impression now and get onto one of his friends, Oracle Now. In Oracles case, its entire future rides and open AI's ability to pay it about seventy billion dollars a year in annual revenue, and that rides on Oracle's ability to build about seven point one gigawats or so of data sene capacity in the next two or three years. I'm not kidding, it's genuinely that's that's what they need to do. Oracle and open Ai have a three hundred billion dollar contract that's meant to have started on June first, twenty twenty six, which said revenue coming from the Stargate data sentence, the

5:53 Oracle is building for Clammy Sammy. That's right, mister Ortmand Clamuel. We don't like him. The problem, of course, is the seven point one gigawats of capacity needs to be built, and barely four hundred megawats of it actually exist. And that's in stockg Abilene, a project that broke ground in the middle of twenty twenty four and was meant to be done either the end of last year or a couple of weeks ago. And everything I hear from the people inside Oracle is that they keep firing all the people who know other fucking business works. It's a terrible thing for them. Oh and another thing is that Oracle literally warned in its anal report that it may or may not get paid by open AI, and that said

6:33 nonpayment would potentially also mean that it couldn't leasse the GPU capacity to anyone else, as in no one else really exists. They could buy it, and if they did manage to sell it to them be a massive loss. Is that good? Also? Tell me is it good that S and P Global also downgraded Oracle just one step above junk grade? They're barely investment grade credit wise for now for all, an angel status aren't incoming. Maybe on that means another monologue you don't really care. Let's get to the funny stuff though. Larry Ellison, CEO of sorry Chairman of Oracle CEO is currently two different mister Bean type characters. He's also used three hundred and forty six million of his one point one six billion Oracle shares

7:15 for an estimated twenty one billion dollars of margin loans personal ones, just to be clear. Otherwise, he has about ten billion dollars in cash and fifteen billion dollars in Tesla shares. Why do you care about that? Well, this is material because he also has to contra up about twenty billion dollars in cash actual cash to push through the paramount Warner Brothers deal by September thirtieth. Otherwise, he gets find about five hundred and six hundred million dollars a quarter, and to do so, regardless of whether it happens, whether it's stopped by then, or whether it's meant to go through, he's likely going to have to take out further margin loans on his Oracle stock, And depending on how said stock is doing, getting that loan might be

7:52 quite problematic. I think had just said problematic, but I'm just going to keep on going. So to explain A margin loan works by offering up a certain amount of shares is collateral a little like a mortgage. So if you are taken out a twenty five percent loan to value loan for one billion dollars, you'd offer up about two hundred and fifty million dollars in stock, maybe a little more depending on how volatile it is. When the value of said stock drops below a certain amount, you get forced to do something called a margin core, while a margin call happens to our main, where you're forced to either proffer up more cash to cover the shortfall, or, far more likely, in Ellison's case, give up more Oracle stock.

8:41 The problem that Ellison faces is the very same one that SoftBank has with Arm. He owns forty percent of Oracle shares, which means that liquidating them would be very difficult. If he started selling them off, the market would say, oh, Christ, is he going to sell his forty percent holdings in Oracle? Sell?

8:58 Sell it all? The Piss index is crashing. And to make that as worse, he's collateralized over fifty billion dollars worth of Oracle shares the ones I mentioned earlier, which means that if Oracle stock drops below, say I don't know sixty dollars a share, He'll be put in a vicious cycle where he'll either have to hand over more and more shares to cover the shortfall, or worse, still have to sell them to keep up. Hey, hey, stop laughing, stop it, stop laughing. Mister Ellison is in real trouble.

9:25 Larry Ellison could be in a real pickle. Stop it. I just kid and fuck Larry Ellison anyway. The worst part is that open ai doesn't even have to collapse for this to happen. The AI bubble isn't based on anyone's actual revenues, productivity or ROI. No Hyperscala actually reports their AI revenues, Oracle kind of bundles them within OCI their infrastructure bucket, and outside of open Ai and Anthropic the industry is actually incredibly small revenue wise, very loud and annoying, though the markets have conflated akapex bubble as in GPUs and the surrounding gear for data centers with a thriving AI industry, which is a nice way of saying that the current slate of prices for Microsoft, Google, Amazon,

10:08 Meta and Oracle and their current AI focused bumps are based on vibes, and mania vibes can break mania subsides. All it takes for things to go pair shaped is for the market to believe, to become convinced that Oracle won't get paid by open Ai for them to start dumping the stock. All they have to believe is that all of that capex, all of that negative cash flow is just for nothing, is for a company that either won't exist or won't be able to actually afford all of that data center capacity. Every time I say this out loud, I feel crazy. The biggest myth though about Oracle is that it's this healthy, profitable business and that this AI thing is just yet more good stuff that they've added on top, when in fact, every other business

10:56 segment other than renting out aigpus is either plat code or in active decline, meaning that Oracle's entire future rides and whether it can monetize its fin doom relationship with Nvidia and Jensen Wong and really Oracle's story is fundamentally insane. Right now, It's spending three hundred and forty billion dollars or more on AI data center capacity for one client Open AI in the hopes that open Ai can bring it seventy billion dollars a year in revenue for a contract that began a month ago, except most of the capacity is barely under construction, and because data centers take eighteen to thirty six months to build, it's unlikely that more than a single gigawatt will be ready before the beginning of twenty twenty eight, if at all, building this

11:40 capacity is costly. An Oracle's cash flow has been negative twenty billion dollars or worse for the last two quarters, with plans to spend another ninety billion dollars in the next fiscal year, which just begun in June. The twenty two thousand people that it laid off in the past six months were based on discussions with many sources load bearing parts of it's infrastructure, throwing parts of the business into active chaos and genuinely causing problems with renewals for its Oracle database products that are the cornerstone of how Larry Ellison's heart beats. I don't know what to say.

12:14 I don't know. I am just a guy. Why am I the one that found this? It's sitting in broad daylight. Maybe people just don't do mathematics. I don't know, but this is this is quite bad. And the point I'm making, though, just to simplify, is that for me to be wrong, open ai will have to bring in hundreds of billions of dollars a year in revenue for themselves by twenty three, which will require it to become either profitable massively profitable too like not just like a little bit of a bit the profitability I mean twenty thirty forty billion dollars a year plus or just continually raise money. I'm talking hundreds of billions of dollars a year, and no, no government bailout will cover the shortfall. The government's talking about

12:58 five percent share be forty two billion dollars. That won't even cover their fucking influence costs if they lasted twenty twenty seven. It's a joke. Oh and by the way, even if open ai does that, Oracle will have to complete the single most ambitious construction project of all time, building enough power to power multiple cities, erecting giant data centers in multiple states, make them fully operational with far less people than the shortage of both talent and the core materials like electrical grade steel and also do so on a timeline that doesn't put it in breach of contract with open AI. Would it shock you if I told you that this is considered radical thinking? Would you be surprised if I told you that many journalists simply

13:42 shrug when you tell them this, and they go, they'll work it out, they'll work it out, It'll all work out, it'll work out. Don't worry about it. Ed, you're crazy head Ed. Why are you outside my office? Why are you dressed as a dug? Why do you keep saying I'm the deterioration dug. Quack quack quack, the market's going to hell? And the answer is that's a bit I'm working on that I'll go to later. In any case, I am kidding, and I do need to say. The future really is one strewn with chaos and carcasses, and I'm not sure the world is actually ready for it, thanks to a captured media industry that refuses to do the messy work of reconciling with the projections that range from unrealistic to fantastical. When the time comes, people are

14:23 going to be asking why we didn't see this coming? And as I've said before, it's because nobody wanted to fucking look anyway, Thanks as ever for listening. Things are going to get real interesting over the next few months. I just got me a Bloomberg terminal, and it's a source of unfathomable power that I believe will take my coverage to the next level. If you're on there, shoot me a message, shoot me a goot on Ploop. That's another social network I'm on. And I really do, genuinely, deeply and meaningfully appreciate the many wonderful messages and emails I receive every day, and I'm so grateful to have such incredible fans and followers. You're all amazing, like I hear from so many of you, and you're very thoughtful

15:00 and smart, and you pick this stuff up very quickly. I love you all, and I assume your bellow we love you chef in response, See you next week.