원문 보기
The Wall Street Journal, uh, got a big scoop, uh, last night, saying they saw a new draft of the IPO prospectus. Of course, we received the first round of this with Reuters a couple of weeks ago. Now the Journal says they have seen a new, updated draft. I want to review some of these figures because, again, they keep growing.
In the first half of 2026, Anthropic's revenue was reportedly $16.3 billion, up from $ 1.1 billion in the first half of 2025. That's huge growth. Half of a spending on computing and infrastructure of almost nine billion dollars compared to 2.3 billion in the same period last year. A net loss of 80.7 billion, although most of that is related to certain accounting with the convertible shares and convertible notes they have issued.
And then, perhaps the most striking figure of all, is that the company could be on track to spend $750 billion in the coming years to buy the computing power they so desperately crave. Yes. Can we break it down for a minute? Let's pause for a moment . Okay. Well, the Journal reviewed it. $111 billion for Google.
This is over a period of seven to ten years. $111 billion for Google. $110 billion for Amazon Web Services. $31 billion for Microsoft. Here's the asterisk with those three. Regardless of actual use. These are non-cancellable contracts. With or without that ability, they are paying for it. That's the gist of that story.
And there's more. Ah, but wait, there's more. $86 billion to lease Nvidia chips from Nscale and Lambda. These contracts extend until 2033 and 2034 respectively. $160 billion to lease Google chips in its deal with Broadcom. And then, you know, there are more negotiations to further increase part of the spending.
And that's how they get those 750 billion dollars. Right now , basically, an additional $150 billion has been committed. Can you think of another company that has these kinds of outlays? As I said, you already know how much it costs. Perhaps OpenAI. Perhaps OpenAI. Clear. We talked about how I talked about how in the automotive world building a factory costs five billion dollars.
And that's a lot of money. Money that is frightening in the world. Five billion is, as I said, our coffee bill; Our coffee or our luxury coffee beans represent five billion. Alright. Clear. Five billion dollars is no longer dismissed. Calm down. This seems like a lot. And this is an irritated Frost.
It's an irritated Frost. Furthermore, regarding your point about the $111 billion for Google and $110 billion for AWS , this S-1 says that two customers, just two, represent approximately one- fifth of the revenue of one of the halves. Yes. Yes. If those customers, whoever they are— probably the hyperscalers of the hyperscalers—stay strong, that's great.
But if there is any oscillation or pullback in capital or computing spending by any hyperscaler, what do you think will happen? Suddenly, your income will drop drastically, which will be a real problem. And then, of course, there are all the risk factors that the Journal talks about. There has been a lot of reporting about them, but they talk about security concerns, they talk about the opposition to data centers that we have seen.
And, by the way, on that front, the president intervened this morning with his own suggestion. That's how it is. For data center builders , who shouldn't limit themselves to offering to build schools or whatever in areas; They should offer a dividend, they should offer a direct payment, which of course is a favorite tactic of Trump's own to offer that kind of payment, not that it is fulfilled, but that is what, you know, he.
Well, this is, you know, if the Republicans win the election, I'll give every American $5,000 in a check, which would add something like $1.6 trillion to the budget. Ah, you know, what are a few billion? What are a few billion among friends? It makes sense that he, you know, would suggest—who knows if anyone will accept his suggestion.
But you have to find ways to gain acceptance. Yes, exactly. That points to the risk that Anthropic highlights at this point; Yes, you are absolutely right about that. We also received other interesting comments. I believe it was yesterday when Ken Griffin, who runs the Citadel hedge fund, one of the largest and most successful hedge funds in the world, spoke at a Bloomberg forum.
In recent months, he was perhaps the man behind, or at least partially behind, the situational awareness explosion of that hedge fund run by former OpenAI researcher Leopold Aschenbrenner. Griffin said that AI is almost certainly a bubble. Yes, it's a bubble. However, every time you have a major technical advance like this, a technological advance, sorry, a bubble is produced .
And as we've been saying on the show, there will be winners and there will be losers . But what usually emerges from the back of these bubbles is transformative, world-changing technology , and the companies that survive and end up at the top end up doing enormously well. Well, as he says, he wishes he had kept his Amazon stock at that time.
However, there are hundreds of companies that went bankrupt or whose shares have not yet recovered, decades after the dot-com crisis. Nobody is saying that Anthropic is a technology company that has done well. Okay. One quick thing he mentioned is that he described the AI sell-off as investors doing their homework , right?
Okay, we've heard that too. And he said that the deleveraging of hedge funds would be the tenth explanation in response to this bubble talk. It's like, yeah, you were involved in that deleveraging quite closely with one of our dear friends, as you mentioned. Leopold is one of Leopold's stories. That's how it is.
This is one of the stories we 've heard. It doesn't necessarily mean there has to be some kind of market downturn. But there will almost certainly be a downgrade at some point once things calm down and we get past some of this IPO hype and investors start saying, "Well, what are these companies really worth?
" Once we see real numbers. If you listen to the Godfather of AI, the demand is that there's nothing to be done, that everything will be alright. I'm talking, of course, about Jensen Huang from Nvidia. [ __ ] Bose just spoke to him after the GTC presentation he gave in Berlin. By the way, do you know the name of the place where he spoke?
I love it. It was called Tempodrom. The Tempodrom. I do n't know if I'm saying it right in Berlin. Anyway, let's listen to what he said, as requested. On demand it's simply incredible. Core Weave is more or less the same. Nebius is the same. And Scale is the same. I think that's the challenge, of course, and this is something like what you were trying to achieve from the moment the computers are shipped until the moment they start generating profits.
It could take many months . And the amount of free float, because we're talking about a fairly significant investment, is quite large. That's why we need to help these companies reduce it. And I believe that this reduction in the supply chain, which makes it easier for them to implement these computer systems, will make a big difference.
And this relates to what we discussed yesterday regarding the Business Insider report on Amazon. Yes. There is a kind of inequality between the moment they receive the chips and the moment they can deploy them in a data center and recognize that revenue. Yes. But I think what it really doesn't address are other issues that Gartner , and I believe Bain, have pointed out.
There is not only a gap in inequality in the supply chain. It also exists, even if what you say you're seeing is high demand. There is a gap between spending on all of this and the generation and recognition of revenue in the end market. Correct. And that's because he talks about one part, but not the other part of it all.
Correct. Because you have to assume that if you're going to spend all that money, and let's say you're someone who buys Nvidia chips worth $10 million , but you need 6 months to install them. You are assuming that during those 6 months with that money spent on your balance sheet you will recover it later, that the income will appear.
And this brings us back to the fact that it's not necessarily that it won't appear because it's very possible that it will, but does it happen quickly enough? Do you get the return quickly enough? Some of these reports say that the magnitude will never be one or never. Something technological has to happen for it to get there.
Regarding innovation, that's true. So that it gets there. And then, quickly, Nvidia started talking about the company's IPO, right? How, reportedly, they are going to start backing them with a lot of money to get that IPO off the ground. Once again, the debate on circular finance has resurfaced. What if the chips are hardwired like Nvidia chips, etc.?
And then I also just saw an Australian data center company called NextDC that's having trouble building in the community because they simply don't want to, and they might be considering issuing a force majeure notice. Ah, another one.
Comments 0