This is the issue with every bubble, every single financial crisis, everything. Until something goes wrong, nothing is wrong. Everyone thinks a quarter, two quarters in the future, and they're like numbers going up right now. Each quarter's success is contingent not on Nvidia's ability to make great chips, but on whether their various financial victims will be able to raise debt.
So, the limitations on these models are the same as they've always been. What hasn't been limited is the amount of money they can spend kind of forcing it, buying data, creating data, paying data companies. And that's where the limitations run. And that's where the limitations will be when the bubble actually pops.
Hi, I'm Isaac and joining me on the tech report today is writer of Where's Your Ed At and the host of the Better Offline podcast, Ed Zitron. Welcome back.
What's up? So, there's a few things I want to talk about this week, including Anthropic floating the idea that the total AI market is going to be $30 trillion.
But first, we should definitely look at Nvidia, who posted their revenue and earnings this week, and their CFO saying that the backlog across major cloud providers has climbed to $2 trillion.
It's no secret that spending is accelerating amongst hyperscalers, especially. And so it's kind of no surprise that the Nvidia CEO Jensen Huang himself has said that we're now in a golden age.
But how much of this is real? Obviously, you've got the circular financing going on on the one hand and then there's things like letters of intent kind of being spoken about as if they're done deals.
So let's start with that. We've talked about the 500 billion before, but that deal is not real. Like nothing ever happened. Never took place. No money was existing. It's all of us. I must be clear. Nvidia's latest earnings, that was real money. Congratulations, Nvidia, on making that money. However, making that money is also a technical term. So, there's this term called accounts receivable. It means stuff that has been shipped to the customer that we've yet to be paid for, like an IOU, but it's official. There's nothing illegal about it. In fact, they're doing the accounting right. I want to be really clear, Nvidia is not doing anything weird really. So 70% of their accounts receivable this quarter was from five customers. 16% of their entire revenue was from one customer. And for the last 6 months, 44% of their revenue is from three customers. So the money is real. But in a regular business, you want diversified income. You want to have so that one customer can't drop out and then take a chunk of its revenue with him. So with Nvidia, this is the thing. This is the very frustrating part. The people that email me, people that throw rocks at me when I go outside, they try and hit me with the hose when I walk past the house, that kind of thing. They're screaming, 'Nvidia is real, Ed. Number go up.' The problem is that for the number to keep going up, Nvidia needs to keep getting customers. Except the customers, there's only a few of them. And those customers need a bunch of debt. Debt right now and interest rates are going to go up. Debt is becoming more expensive. Coreweave just had to pay 9.2%, 2% on their debt or something ridiculous like that, like higher and higher and higher. At the same time, Nvidia has raised prices by 17%. I have actually heard from sources who deal with the resellers that it's more like 22 to 40% in some cases. The point I'm making is this is the issue with every bubble, every single financial crisis, everything. Until something goes wrong, nothing is wrong. Everyone thinks a quarter, two quarters in the future and they're like numbers going up right now. So Nvidia said next year they would make 70% more revenue than this year. So consensus estimates have us about, and this is an annoying note everyone, I'm sorry. Nvidia is currently in fiscal year 2027. Their fiscal years run February 1st of the year before to January 31st of the year of. So fiscal year 2027 just began on February 1st. Annoying but necessary. So fiscal year 2027, which we're in right now, 396 billion is the consensus estimate for revenue. They are saying they'll make 70% more. So about 674 billion in fiscal year 28, the beginning of February 2027. That's going to be really difficult, like really difficult, not impossible, but what I'm dis- the difficulty is I want to really put this simply because it is kind of technical and annoying, but you have, call them five customers: Microsoft, Google, Amazon, Oracle, SpaceX. Let's just, we don't know, they don't specify, maybe one of them is Coreweave. Of those, Amazon and Google are both cash flow negative right now and they have to keep raising debt. And this is, they have to raise debt right now when before the chips were more expensive, before construction was more expensive, before debt was more expensive. SpaceX already like the rung, maybe a rung or two above junk but still investment grade. So they have to raise debt but their debt will be more expensive and the GPU is more expensive. It's all more expensive, all more and more. So next year they're going to have to sell 70% more probably to these same companies and they're going to have to sell more and more chips at a higher price when debt's more expensive. That is going to be difficult. Not going to say impossible, but it will mean we're going to enter the Jensenverse. I think we're really going to, the circular financing is going to become such a perfect circle that James May Keenan or what have, James Keenan is going to come out and sing about Judas or something because the problem is that data centers are also taking a while to build. The power's hard to get. Local authorities are pushing back. I think Illinois, Arizona, and one other I can't remember are cutting tax breaks. Pennsylvania and Texas are kind of moving against data centers. They're getting rid of all the fast-tracking. So we have all of these forces pushing against Nvidia which is why Nvidia is doing the most insane circular financing thing ever. So right now everything looks great. It looks perfect. Number go up. Number go up so high. Number so big. Everyone see number go up. And I have had such emails this week like aha, like various forms of goblin in my email just being like, ah got you, Nvidia number go up so big. A week ago these people were squealing. Now they're like, 'Ah, number go up.' And the problem is that these people are going to assume everything's fine right up until someone can't raise debt. And that really is the ultimate pale horse here. It's going to be Coreweave, Nebius, and I haven't even talked about Australia yet. Shahin and Fermos, which are two Australian NeoClouds that Nvidia has become the biggest customer of. What a surprise. And I think a really simple way of putting this is a normal healthy diverse industry does not act in this way. This is not what normal looks like. And I think the defense that people will make is, well this is the new industrial revolution. It's the new industrial revolution, Ed. We need to do weird stuff to make the industrial revolution happen. And the problem is that people are overly reliant on history as a coping mechanism to the point that they will ignore reality.
I mean to your point, warranted or not, the promise of this elusive AI payoff that is kind of supposedly going to be coming I guess in 2030.
Yeah. Like you say with Nvidia's earnings there is clearly the demand for that hardware right now but and I appreciate that this is kind of a tricky question because an exact number would be kind of impossible but how much of this demand involves Nvidia promising money to its own customers?
Well it depends on the customer you're talking about. So I brought up Shahin. So Shahin AI, the reason I bring them up is they were the first customer of this new nebulous Nvidia revenue share program where Nvidia invests in the company, helps them raise debt, they build the data centers at some point in the future that we don't know. And then when that happens, the GPUs rent out at a certain rate and then you add in interest, appreciation, amortization, all that crap. And then on top of that, whatever's above that, Nvidia gets. Now, Colette Kress the CFO said on the call that this was very real and Anissa Gardizi at the Wall Street Journal reported that the revenue share deal was dead. So, who knows. But at the point, Shahin and Fermos are two Australian NeoClouds that Nvidia has invested in. Nvidia will take money from them maybe as a revenue share. We don't know. So, those two only exist because of Nvidia. Shahin's last revenue, or its depending on how you feel about talking about the company, their quarterly revenue is $1.8 million, but Nvidia just signed a $4.9 billion deal with them. That's kind of what I'm getting at that there are just these bordering on fake companies that they're real. There's a bank account and a person that runs them and they're going to build a data center, but they just existed. Fermos, I was told by a source, was near death until Nvidia invested in them. Another one is Poolside. Now Poolside isn't really a customer. Well, they are. They have been trying to build data centers and put Nvidia GPUs in them. Nvidia now, okay, the CEO of Poolside said this was not an acqui-hire. All that happened was Nvidia gave them $6 billion, hired most of their staff, gave them another billion dollars in investment, and now all of them are going to be working on Nvidia things possibly at Nvidia's building. Not an acqui-hire though. Stop calling it that. So again, a potential customer of Nvidia that is mostly handling Nvidia's money. Nvidia invested $30 billion in OpenAI. OpenAI now doing a somewhat gigawatt, we don't really know, large data center in Ohio. Nvidia is backstopping $105 billion of that if it gets built, but only if it gets built. The Jensenverse, we just live in it. So the thing is when it comes to like SpaceX, for example, that's the biggest one of the circular financing. They did this deal with Valor Equity Partners earlier in the year where Nvidia invested $2 billion, some private credit dynamics.
With that money we've given you, you will give it back to us. That puts you in messy legal situations. Winstar famously sued what remained of Lucent Technologies and won. The bankruptcy courts won and ripped that away from Lucent. So that never goes well. What Jensen does is Jensen either does an equity investment which makes the banks go, 'Oh, you've got money. We can loan you money. You have collateral. You have cash. That's totally fine.' Or Nvidia becomes the customer. This is what Morgan Stanley calls balance sheet as a service. But I actually would go further. I think it's vendor financing by proxy. It's just the, it's like all of the AI things. It's exploiting these little if not loopholes, they're intellectual loopholes in everything. Because taking a step back, if you were buying a house and your parents gave you the money for the down payment, the bank might say yes, but the bank is going to ask your parents for real details to say this is non-recourse, right? You're not going to expect this money back. This is not another loan. They would also, if all of your income was coming in from people with your second name and you were claiming that your income was from your family, they would say no. They would say this is not distinct income. However, if all of your income is coming in from Nvidia and you're using debt based on that contract from Nvidia to buy things from Nvidia, the banks say great because Nvidia is a business. It's a business with good credit. And I must be clear, Nvidia is a real company with real money. Other than that accounts receivable thing, they are shipping tens of billions of dollars worth of stuff without getting paid and then there was this worrying little note at the bottom that said for some investment grade customers they are extending payment terms from 90 days to a year. Now you may think investment grade, right, that means Microsoft, it means Amazon, don't you worry friend, uh-uh, that includes SpaceX. And because this didn't say investment grade credit. It didn't say credit of investment. It just said investment grade. It could refer to Nebius or Coreweave who have both raised investment grade debt connected to contracts from Meta. So this is all these little ways of doing effectively the same thing as vendor financing without ever having to actually finance anything. And it's really bad. And it's one of those things where it will go totally fine as long as Nvidia keeps being a cash generating machine. Where it will go wrong will be if one of these customers dies or one of these, sorry, dies isn't even necessary. If one of these customers can't raise debt and next year I think they want to raise like half a trillion dollars worth of debt on top of the rest of the debt that the rest of the world wants to raise for AI chips. It's just genuinely bonkers.
So, if you were running the SEC, what kind of stuff is not on Nvidia's earning reports that should be?
So, the thing is the SEC would say that you wouldn't have to do these things if I was running it. I would actually bring the, here is the most obvious regulation you bring. If you have a customer over 10%, you have to name them unless it is national security. I can understand that. But if you have customer A, if you have a customer that's 16% of your revenue, it is in the best interest of the public and investors to know who that is. If you mention investment grade clients that are getting three months to a year payment terms, you have to name them. Really simple. I don't actually think it's a particularly unreasonable request because all we're saying is, 'Hey, who are they?' You don't even need to tell us what it is they're buying or when it's shipping or anything. Just tell us who they are because there is a meaningful difference between we've extended 90-day payment terms to Microsoft. Microsoft has the cash. They'll be fine. They haven't even really had to take out debt in the AI bubble. Microsoft's great. If it's Oracle, I'm a little bit more worried. Okay. Well, Crazy Larry's having a bit of a rough time right now. If it's Coreweave, I'm calling a cardiac surgeon. Like, but that's the thing. These are, and of course the AI boosters will read all of this and use this as evidence that the great prophecy has come true. Actually, Nvidia's fine. Look, investment grade. You've jingled the keys in the right way. The problem is they better hope they're right because if that 16% customer is SpaceX, for example, SpaceX does not have great credit. SpaceX loses billions of dollars. And also, SpaceX will need more debt and their debt trades pretty poorly. If it's Coreweave, Coreweave's had to raise, I really should go and look this up. Coreweave's had to raise billions of dollars this year. Billions and billions. I think over 10 billion dollars this year. I'd have to look exactly, but they've had to raise billions of dollars every year. They're going to have to raise billions of dollars next year and the year after that and the year after that. They'll be dead by then. But this is a huge issue and the obfuscation of the money is really bad because as I've said, number go up right now. Everyone think Nvidia is so big, so huge, so massive. Jensen muscle so big. Jensen money so big. Why worry me? Jensen money big. But it's because, and I can understand it from just a consumer perspective. You go 96 something billion dollars. Hell yeah. Why would I worry about that? They make 96 billion. There is a big difference between a company that makes $96 billion from hundreds of customers or even 20 customers with one that makes it from 10 or five or three or one. And so it means that one quarter's success and each quarter's success is contingent not on Nvidia's ability to make great chips or be great salespeople or provide great service, but on whether their various financial victims will be able to raise debt. And it's just frustrating because it's been a frustrating, I'm not going to lie, it's been a frustrating week seeing people say this because it's like yeah until it happens you will assume it is impossible.
Let's move on to some even more frustrating news arguably that Anthropic thinks the total addressable market is going to possibly be $30 trillion according to the Wall Street Journal. And I mean of course people inside Anthropic would be saying that for very obvious reasons not least the $2 trillion IPO target but what assumptions are being made for that? I mean prediction is maybe generous but that statement and then how realistic or unrealistic are those predictions.
So the 30 trillion number is just make up a number. I would be honest. I think you could probably, they should have said a hundred.
I just, why not? Why go so low?
So let's talk to Steve Eisman who's the bloke, the bloke who played Mark Baum. So Mark Baum, Steve Carell's character in The Big Short, that bloke who that's who it's based on, and he was saying that it's the middle point between the entire US economy and SpaceX's 28 trillion dollar TAM. It's just like they chose that number. They're claiming it's based on if all labor was automated. So, putting my if I ran the SEC hat on again, yeah, I'd make that illegal. I would actually make, I think IPO prospectuses need to change. I don't think they should be able to say TAM further than two years. And people say, 'Oh, that's not fair. How do you value a stock?' But shut up. Just shut up. Just there is no value in a company saying $30 trillion total addressable market. That's not, what does that tell anyone? All it exists to do is mislead investors. All it does is add obfuscation to the company. And that two trillion market cap that they want. Maybe they get it, maybe they don't. To get that they will have, I mean they could just, because there's so much fluff around this and people are so rock hard for Anthropic, but they will be raising a hundred billion dollars which won't be enough to sustain them longer than like a year and it will just, it will be the last money out I think because Clammy Sammy, poor Clamuel is really going to have a lot of trouble taking his dog public at this point. But Anthropic, let me take a step back. Actually, the reason that Anthropic is doing this so fast and why they must get it out now is they need to, what we're coming to. They want it out in like September, October so that they don't have to close the books on Q3 because Q2 was really the peak of token maxing when everyone was, if they just moved everyone onto the enterprise contracts where they have to pay the actual cost of tokens. You had the token maxing era. You had the slowdown, but the full slowdown has not happened because token maxing was not just here. I'm plugging my credit card into Dario Amodei's face. It's I'm going to agree to spend $50 million. I'm going to agree to spend $100 million. Now, this is one of the problems with annualized run rates with multiplying a period of time by 12 or 13 because they were very likely annualizing token spend, which is not a recurring expense. So, if you're spending $1,000 on tokens in a week, that's not a recurring expense. That's not like spending 200 a month. So Anthropic wants this thing out the door so that they don't have to show that, they don't have to show everyone. I'm not saying people have stopped paying Anthropic. I'm saying that people are not paying Anthropic as much. People are moderating their token spend. Reuters just had a story about how they're lowering their token spend using open source models. So Anthropic needs this dog out now. They could get 100 billion. I'm not saying it's impossible. I think all those investors are very unhappy. Maybe it pops 15, 20, 25% on the first day and everyone goes like, 'Oh, we win.' But I think what's actually, I think there's a chance that those numbers make people really upset. I also think the people are going to be willing to look past them just to enjoy the fluff of AI. But the thing is long-term, Anthropic, they need hundreds of billions more and the public market scrutiny is going to be rough. I also question whether there'll be like, what quality is their debt going to be like. It will have to be junk or high yield or whatever you call it, like the lowest junk because this is going to be an unprofitable company. One prediction I'll give you is I wonder, maybe a prediction, guess, whatever. I wonder if they're going to try and capitalize training expenses or if they're going to do some really non-GAAP, so non-generally accepted accounting principles things where it's like, yeah, our GAAP margins are negative, blah blah, but our non-GAAP margins when you remove training, you just take out this big cost, they're actually positive. Checkmate, atheists. And I think that we're going to see a lot of that. The arguments around this thing, I'm going to get the worst emails in the world. I'm going to get the most annoying emails in the world. Comes out after Labor Day in America. So, we got a couple weeks before this crap. But they are claiming that $2 trillion valuation based on fairy dust, based on theoretical growth. And I mean obviously their last valuation I think was 900 and something billion.
Even that was a bit high. But this is like, okay so you want double that. Has your revenue doubled? Is your revenue doubling? Are you going to be at 100 billion ARR, sorry annualized run rate not recurring revenue, by the end of the year? Are you going to be there by IPO? Why are you using run rate? You're a real company. You should be talking in dollars and cents.
And I actually think that it could go public. It could have a good run initially. But I think that the public scrutiny is going to be brutal on this company. It's going to be really nasty. And the same way right now all the Twitter psychopaths, everything good and bad happens with Anthropic. They freak out, they scream, they cry, they laugh, whatever. They all have various forms of bipolar disorder when talking about Anthropic. I think the stock itself is going to be so aggressively volatile that trading it is going to be like throwing knives and catching them at the same time.
