원문 보기
We have an S1 filed by xAI. Of course, this is one of the two key labs that we're watching so closely, a huge company in the private markets that has plans to go public and will probably do so. Let's say within the next 3-4 weeks. In the last third quarter, we reported $16.8 billion in revenue. It was the slowdown in income growth that generally alarmed markets.
So, well , the interesting graph that we just looked at, of course, is the one that I may have mentioned, Jim. They obviously get a lot more money. Yes, that's exactly what he says. But they also spend significantly more, which is also worth remembering. As for the cost of computing power, think about how much they pay SpaceX alone , which we just talked about.
You know, SpaceX is talking about getting $50 billion for every gigawatt of computing power. So, while the revenue they are able to generate is increasing, the question is , Carl, what are their costs of all this? Yes, I mean, when we looked at risk factors, we talked about how many elements would be embedded in those factors.
The first concerns the possibility of a lack of computing power. True. This is the main thing. Yes. And I think that's actually the main focus of Elon Musk's amazing conference last week, where he talks about this incredible expansion of his computing power. There is a page now, and I'm sorry we're looking at this in parts, but it's better than nothing.
On page 137 it says: operational efficiency contributes to margin expansion. The margin is growing rapidly. And so when you think about the calculations, you start thinking: Okay, how many people, how many companies are really going to be left? And I think that was also Elon Musk's point. But this company is definitely not going anywhere, and our roadmap begins with Musk predicting a shake-up in the cloud technology market, confirmation of SpaceX's $40 billion deal for AI chips, plus Meta's announcement of a share buyback worth up to $95 billion .
The amount of money that needs to be raised at this particular moment is extraordinary. And I don't know if we will have enough resources. I don't know, David, if there are enough of them to handle what everyone needs. I think we will have to think about capital and access to it as a crucial factor for companies that are clearly competing with each other to create AI and satisfy their huge appetite for computing power.
So, just this morning we mentioned that Meta is going to issue $22.5 billion worth of common stock. In total, Meta is going to raise $35 billion very quickly. In addition, they plan to market new non-voting Class C shares, which they will issue next year. This will be a market placement, so not necessarily all at once, but they will be able to sell them for up to 60 billion depending on demand.
So, that's 95 billion. True. And finally, you have 40 billion. The SpaceX deal is a bit more complicated, involving a special project company and Apollo as a lessor buying the chips and GPUs, but it's essentially a loan to SpaceX. I think it will be on their balance sheet as a lease, but since it's a loan for SpaceX, they're going to the market, and that's 40 billion that will go towards purchasing GPUs.
Of course, we are waiting for other companies to go public or tell us about their IPO plans , like Anthropic. And they will also go to the market for money. And then there were these comments from Musk, guys, which were very interesting over the weekend. Is this self-serving, or is it just the way things are?
Both, right? Well, it's Musk. Selfish, but at the same time there is some truth in it, right? He's just saying that—and he's saying this here—it's all going to depend on access to capital, debt markets, and capital markets in general. And then he goes on to tell Carl about the new clouds, that he sees deep shortages and consolidation.
Maybe he's not talking about the biggest players. Maybe he's referring to some smaller companies. One of the richest companies on Earth just did something rich companies almost never do. Meta plans to sell its own shares worth almost a hundred billion dollars. Why? To pay for AI. And the company that just bought them for three billion is Nvidia, the same one that sells Meta chips.
If this makes you a little nervous—that's fine. Stay with me. Over the next few minutes, I'll explain what this deal is , why Meta needs the money, and what it says about the current AI race. And near the end, I'll show you one number that completely changes the face of this deal. Most people will miss it.
Let's start with the main thing. On Tuesday, Nvidia announced a three billion dollar investment in the sale of Meta shares. In simple terms: Meta issues new shares and Nvidia buys some of them. But it's not just about buying stocks. The companies are also expanding their partnership around a product called Muse.
So, what is Muse? Muse is a personal AI agent from Meta. It was launched on September 8th, and it is not a chatbot. It really does the work for you. Meta claims it can send emails, book trips, fill out forms, and even pay for purchases. You just tell him what you need, as if you were writing to a friend, and he does the task.
And people like it. It became one of the most popular apps in the first week after launch. Mark Zuckerberg says the goal is to give billions of people personal superintelligence. That's the dream. But here's a point that should alert you. Why is Meta selling shares at all? Meta is making a fortune. Usually she doesn't need to ask anyone for money.
But AI is extremely expensive. Meta predicts that data center spending will reach $135–145 billion this year. And when the bills get so huge, even Meta has to look for funding. On Monday, Meta announced a share sale worth up to $97.5 billion. Part of this is a program of gradually selling shares on the market over a period of time.
The company has so far sold $12.5 billion in preferred shares and $25 billion in common shares at $625 apiece. And Meta is not alone. Alphabet and Intel held similar fundraisers earlier this year. Think about it. These are some of the richest companies in history, and they raise funds like startups.
This shows exactly how expensive this AI race has become. Okay, now I suggest you play a game. Why is Nvidia, a company that "sells shovels," buying $3 billion in shares of its largest customer ? Is this A—just a profitable investment? In —a way to "bind" Meta as a client? Is C— something much bigger?
Choose one option. Record your choice. I'll tell you what I think in a minute. First, let's see what Meta gets. Under the agreement, Meta plans to expand the use of Nvidia equipment in its data centers. This includes Vera Rubin, Nvidia's next-generation AI platform, and Vera processors to power its AI agents.
The companies will also work together to make Muse models learn faster and perform faster. And yes, they even released an image of the Muse mascot— a character named Jolly dressed as Jensen Huang. Even with a leather jacket. Jensen says AI agents are a new paradigm for computing. Software that understands your goals, thinks, uses tools, and does the work for you .
So Nvidia is n't just betting on Meta, it's betting on agents. But wait, is there a real demand for such enormous computing power? Here we move on to the third heading— GPU rental prices. Here's a short analogy. When everyone is looking for hotel rooms in the same city for the same weekend, prices skyrocket.
With GPU everything works the same way. Research firm SemiAnalysis found that the price of renting an Nvidia H100 chip on a one-year contract increased by almost 40% from October to March—from $1.70 to $2.35 per hour. And this continues. Earlier this month, cloud company Jarvis Labs raised its hourly price for renting an H100 on demand from $2.69 to $3.49.
Reason? Global memory shortage for GPUs . When the prices for renting chips that are several years old are constantly increasing, it is a sign of a shortage of computing power. Part of the recent growth is due to the cost of memory, but the problem is really serious. This brings us back to Nvidia's last quarter: $96.2 billion in revenue over 3 months.
And now—the number I promised, the one that changes everything. Nvidia invests $3 billion in Meta . Meta plans to spend between $135 billion and $145 billion on AI and data centers this year. 3 billion is only about 2% of that amount, and a significant portion of Meta's spending will likely go to Nvidia, which dominates the market for AI chips, networking equipment, and systems.
So think about this. Nvidia is investing several billion, Meta is committing to deploying even more of Nvidia's infrastructure. That's why I choose option C, something much larger. It's not really an investment, it's a flywheel. Nvidia funds its customer, the customer builds more AI, and that AI runs on more Nvidia chips.
To be fair , critics call such deals "circular." Funding companies that buy Nvidia chips could artificially inflate demand, and that 's a real risk to watch out for. If Muse fails or AI spending slows, this whole cycle could spin in reverse. But now, with rising rental prices and a shortage of capacity, the demand looks quite real.
So let's close these circles. Why is Meta selling shares? Because AI is so expensive that even Meta needs outside money. What is Muse? This is Meta's personal AI agent, and Nvidia is helping to make it faster. Is this demand real? GPU rental prices indicate that there is still a shortage of computing power.
And what number changes everything? Nvidia's $3 billion is a pittance compared to Meta's spending on AI, and a significant portion of that money is likely going back to Nvidia. sentence. Nvidia is n't just selling shovels for the gold rush, it's paying miners to dig more actively. If this helped you understand the point, click " subscribe".
I'm breaking down every major step in AI, and next time I'll show you which other tech giant could start selling off its shares and why this could be a warning.
댓글 0