원문 보기
Before the break, I told you in detail about the terrible sell-off in technology stocks last week. Although AI data center stocks recovered a bit today, this whole situation is questionable, right? And I think this group will have a hard time finding balance for now. Especially given this uncertainty hanging over us.
So what really needs to change before we can start buying stocks on days with some margin of safety? And we know I say "certain" because "bottom hunting" is never safe. First of all, it sounds simplistic, but you know what? The graphics need to get better. AI data center stocks peaked in January and then recovered until last week, when they were crushed again.
This creates some pretty ugly graphs. When you look at some of these groups represented by ETFs that were at the center of the AI rally earlier this year , I'm thinking of the VanEck Semiconductor ETF or that newer Roundhill Memory ETF with the apt ticker DRAM. They look like they've formed double peaks, and DRAM is actually DRAM, DRAM.
This is one of the worst patterns, a double top. It almost always leads to lower prices. So they need to break this double top pattern. This means we need a good rally, better than today's, that will prove that the path to higher highs and higher lows is still possible. Worse still, those without double tops have the most pronounced head and shoulders patterns, which some consider even more negative, where the only way out is further decline.
For experienced chartists, these stocks have no bottom. You know, if you're an optimist, the good news is that despite the bad situation last week, we didn't break through the lows of late July. It was then that we saw a crescendo of AI data center sell-offs when a highly leveraged hedge fund was finally liquidated due to an assessment of the situation.
Today's rebound, however, I would call it a good start. How do you like that? After Friday's panic, which we now call a " get out now" sell-off, the Nasdaq Composite and Nasdaq 100 indexes fell below their 50-day moving averages. They haven't lost touch with these averages, and today's rebound helped us get closer to key levels, but it would be very helpful if the Nasdaq indices could recover above the 50- day moving average, ideally by the end of the week.
Beyond the charts, which means many of you will turn off the show now because you are only interested in the charts, but I believe the fundamentals still matter. The next big test for the AI data center complex will be the hyperscaler earnings reports. So, SpaceX, have you ever heard of Adam Jonas from Morgan Stanley, he writes an article about SpaceX every day.
SpaceX will give us some insight when it reports tomorrow night. I will definitely read it. I think he is the most important of all. This will be grand. Meta Platforms, with fewer moving parts, could be even more useful when it reports on Wednesday night. After that, we'll hear reports from Alphabet and Microsoft next Wednesday, and Amazon will report the following evening.
Yes, we have real giants. What should we hear from hyperscalers? Well, if you hope for a revival of the AI data center cohort, these companies need to maintain their gigantic capital expenditures. That's right, we should support it, not reduce it. You see, if they start signaling that they are preparing to reduce their investments, the whole group will take a hit.
Why? Well, because, you know, we have to see that they believe it. Um, we have what we want to hear. How about this? We see the prospect of generating hundreds of billions of dollars in revenue. It would be unwise to cut costs, and we don't need to , because 2027 will be the year of a bountiful AI harvest.
And then my hyperscalers, please write this down. This is really good. Additionally, we want to see hyperscalers continue to make serious money from data centers. We started to see it in some of them. I like the great growth figures of cloud infrastructure from Alphabet, Microsoft and Amazon. SpaceX has demonstrated this because it can build data centers and then rent the computing power to others for significant amounts of money.
This is the best example of exactly how much money can be made when you create what we call computing. The situation with Meta is less clear, as it doesn't have a cloud business, but perhaps it's moving in that direction, or perhaps it will use that computing for Muse. This is a personal AI agent platform that is developing so incredibly well.
In short, we're about to go through a high-stakes test in tech earnings reports, and if hyperscalers flinch on AI spending, the entire data center structure could...but as long as they don't flinch, I think we can get through this, especially if they show strong numbers on cloud infrastructure. Third, and perhaps most importantly, I don't think the AI data center segment will be able to reach a sustainable bottom before the midterm elections.
This means we are in limbo until people go to the polls in two weeks, two weeks from tomorrow. We know that data centers have become politically toxic even in states that previously supported them. So far, Democrats have emerged as the party more opposed to data centers, and polls suggest they are likely to do quite well.
Very high chances of taking the House of Representatives , good chances of taking the Senate. However, I'm not too worried about Washington. I think the main problem is the visuals. No politician wants to support data centers before the elections because they don't want to drag their party to the bottom.
But that becomes much less of a problem on November 4th . At this point, I think investors just want certainty about what Washington will look like early next year. Although who knows whether we will have certainty after the elections. It's worth wondering if the Justice Department will sue to block the seat from going from Republican to Democrat.
Don't laugh. This can happen. Perhaps we will see another final fall in this group after the results are announced. But outside of an election campaign, state and local leaders usually like to attract new business. We just need to get through Election Day, and regardless of the results, our leaders won't be so afraid of public backlash, at least until 2028.
Data centers may be unpopular, but local elected officials who want to lower taxes can do so by bargaining hard with hyperscalers who want to locate data centers in their cities . Everyone loves lower taxes. They love them more than they care about big buildings with machines that may or may not eliminate jobs in five years.
Finally, the IPO of Anthropic, one of the leading private AI labs, is looming. Ideally, we want to see Anthropic go public and trade successfully after opening, but at this stage I do n't even want to be that greedy. I just want the deal to happen. Before the break, I mentioned that Anthropic's pre-roadshow last week helped sell out, as a lot of people were disappointed with the numbers.
They previously said they wanted to go public in November, although now we're hearing rumors about December. Perfectly. Will you guys do it already? We need to see that someone is making or will soon be making money on all these data center investments. Of course, hyperscalers make great profits from their cloud infrastructure businesses, but Anthropic and OpenAI generate a significant portion of their revenue on these platforms.
Microsoft's new cloud revenues depend on OpenAI. The growth of Amazon and Google's cloud services depends on Anthropic. So, we have to see at the very end of the cost chain that this can be a sustainably profitable business worth investing in. Nvidia's CEO himself says that U.S. export rules have reduced his company's share of the market for AI data center chips in China from a near monopoly to zero.
To zero. Then why did China's three largest tech companies reportedly place orders with Nvidia worth over $35 billion anyway? By the end of this video, you'll learn what this order really is , why Nvidia is giving away money to startups in Germany and France, and who's really paying for the AI boom.
The answer given by Nvidia's CEO in an interview is not what most people expect. I'll get to that number in a few minutes, and I want you to try to guess it before I tell you. Let's start with orders. Reuters reports that ByteDance, Alibaba, and Tencent have ordered more than 3 million Nvidia RTX Pro 5500 graphics cards priced at around $11,400 each.
If you do the math, you get a potential value of over $35 billion. Among all Chinese customers, orders have exceeded 6 million cards, according to two sources. But wait, here's the same "but": it's not Nvidia's best chip for artificial intelligence. The B300 cannot be sold to Chinese buyers due to US export controls, and the H200 requires licensing on a case-by-case basis .
The RTX Pro 5500 is less powerful than both. This is one of the most powerful Nvidia chips that Chinese buyers can legally get. Imagine a restaurant where the best dishes are not on the menu, so people order whatever is left. And even that is not guaranteed. Nvidia has told customers that it cannot guarantee these orders will be fulfilled , and one source claims that partners may only receive 5-10% of what they requested.
So, here's the question. If you can't get what you want and you can't get what you ordered, what does that say about demand? Keep that thought in mind, because a story emerged that same week that seems to be the opposite. Yesterday in Berlin, at the “Invest in Germany” summit, Jensen Huang stood next to Chancellor Friedrich Merz, and Nvidia announced planned investments in four German startups.
Agile Robots, which manufactures robots for factories. Avelios Medical, software for hospitals. Mirelo, audio for video generated by artificial intelligence. And Reflex Aerospace, which builds satellites. Nvidia didn't say how much money it's putting into each of them. This is part of the obligations to Europe worth almost $10 billion .
A week earlier in Paris, the company did the same with four French startups. Four French, four German. So why would one of the most valuable companies in the world write checks to young companies? Remember the times of the gold rush. If you sell shovels, you should know your miners. Some observers believe that backing early-stage startups embeds Nvidia chips and software into their core, creating long-term customers.
But there's a " but" here: others ask whether European startups will be able to grow on Nvidia chips without giving profits to Nvidia itself. No one knows this yet. OK. Let's go back to that, because I watched an interview where the CEO of Nvidia explained exactly who these startups are for his business, and it genuinely surprised me.
The interviewer asked a difficult question. Will AI labs be able to double or triple their revenue each year to recoup trillions of dollars in data center spending ? The interviewer also cited an estimate of 43 gigawatts of new AI computing capacity next year— more than exists today. The director's response was something like this: people misunderstand what is happening.
He said that Nvidia has about 80% of the AI infrastructure revenue market today and that its business is not just a few labs. Now I want you to play a game. You hear about famous AI labs in every headline. What share of Nvidia's computing power do you think they account for? Choose a number. Remember it.
Don't scroll any further, because this is probably not what the headlines make you think. That's what he said. About a quarter of Nvidia's computing power, perhaps up to 30%. Not quite a third, according to him, let's say a quarter. Another third, he said, is basic calculations. Data processing, physical modeling, non-AI scientific work , which is growing by 30–40% per year.
And the last third is accounted for by thousands of startups that rent Nvidia computers. He called it next-generation cloud computing and said it was a big mistake to look at just a couple of companies and think they were the ones funding the infrastructure. Now I have to be objective. These are approximate numbers from the person who stands to gain the most from this story.
He kept repeating: "Let's call it that." Take this as his perspective, not an audit. And his claim only works if these thousands of startups eventually start making real money. But now the German and French checks make sense. If a third of your business comes from startups renting your chips, then supporting the next batch of startups is like planting gardens whose fruits you hope to buy.
This is reasonable, and it is also worth monitoring, because an orchard only makes a profit if the trees grow. So, let me summarize. 35 billion is the potential value of orders for a less powerful card that meets export requirements, which, according to one source, Nvidia will only be able to fulfill partially.
German and French investments are planned, but their amounts are not disclosed. And the number I asked you to guess is about a quarter, maybe up to 30%, which means: Nvidia CEO says that famous labs are not everything. If you only remember one sentence, let this be it. Nvidia's argument is not that a few giants are buying chips.
And the real test is that everyone is lined up, and whether this long line will be able to pay. And none of this is financial advice. If you want more such analyses, subscribe. And the next time, the same CEO will say that the AI industry will not exist without safety and security. And his plan involves treating the new AI like a toddler in a playpen.
Comments 0