원문 보기
server centers around the world. And then they help manage digital traffic so everything runs faster and more reliably. But this is not a new business. I consistently recommended Akamai back in 2005 when the show started. And the company showed excellent results until the financial crisis struck. Recently, they have focused more on cybersecurity and data collection, which is one of the reasons I continue to recommend Cloudflare.
In general, the thesis about content delivery networks is that these companies thrive when internet traffic increases. And the emergence of AI agents has significantly increased this traffic. I believe this is the main reason for the success of these promotions. That's why they "light up ", but every story is very different.
Let's start with Cloudflare. NET, which is the largest among content delivery networks . Its shares have risen by over 2200% since going public in 2019. We've invited CEO Matthew Prince on the show many times, and he's been very apt in predicting what the advent of AI means for the rest of the internet.
We last spoke with him in June. He said that traffic from AI agents has already exceeded traffic from humans. And it was a great prediction. This is the essence of agents. If you are comparing prices on something, you may only check a few sites, but an AI agent will visit thousands of pages to find the best deal for you.
So, there's a lot more web traffic now, and someone like Cloudflare needs to help us deal with that. This is one reason why the stock continues to perform so fantastically with a steady acceleration in earnings growth over the past year or so. Next up—Akamai. Late last year, they announced a new initiative, Akamai Connected Cloud, which is essentially a network of many smaller data centers spread across the world that help these AI models process data.
Matthew believes we are moving from a period of training AI models to a world where we actually use these models at scale. And this requires significantly more infrastructure. In early August, we invited Akamai CEO Dr. Tom Layton to the show for the first time in over five years. I'll let him explain what his company is trying to do.
We are unique in that we can support AI infrastructure needs from core to edge, so no matter what your task is, what agent or model you are working with, if you need to train it— we can probably do it at the core. But the real use, the everyday use, is the output: when the model is called with a request, it tells the robot to do something, and you want to do it close to the robot or close to the user so that it's fast, scalable, and efficient.
Since I gave that interview, I've been thinking a lot about this Akamai plan. First, it makes perfect sense that distributed data centers are needed, especially as we move into things like physical AI, because that requires a lot of bandwidth and you want the infrastructure to be as close to the users as possible.
This reduces latency, increases speed. Furthermore, recently we have seen how huge central AI data centers are becoming politically toxic . It seems to me that Akamai's idea of building smaller data centers might be easier to promote, don't you think? This is not a year for investment for Akamai. They have to spend a lot of what they have on acquiring and building out infrastructure for their cloud mining, which will take a significant hit to profits.
But management says customer interest is huge, and they expect business to start growing in the fourth quarter before really exploding in 2027. By the way, last Thursday Akamai announced a big win with Anthropic—a contract worth $11.6 billion over seven years. Currently, Anthropic needs additional capacity and is willing to pay for it.
That, by the way, amounts to $1.66 billion per year, which is pretty impressive considering it's more than a third of what Akamai expects to earn this year. Initially, stocks rose sharply in response, as one would expect. But now they are losing all these gains, and even more. Well, to me, this looks like a buying opportunity.
Finally, let's talk a little about Fastly. This is not a company I have been spending much time with lately. I remember Fastly as a fleeting favorite of the pandemic era. The stock once skyrocketed, largely because it was a content delivery network for TikTok, which was rapidly growing in popularity at the time .
But then Fastly lost most of TikTok's business, and the stock fell. For years, it looked like a losing stock that had lost momentum. They only started making a profit last year. Recently, however , stocks have been showing incredible growth . 159% growth year-on- year thanks to impressive financial performance.
If that wasn't enough to catch Wall Street's attention, Fastly came into the spotlight again last week. First, the company held an investor day last Tuesday, where management explained how Fastly is transforming from a regular content delivery network (CDN) into a much more interesting player in edge cloud computing, fueled by the rapid growth of web traffic thanks to all these AI technologies.
Second , it happened right when everyone realized the power of Meta's Muse platform, which really gave the stock a boost. Ultimately, the rise of giant AI has made all three of these content delivery networks big winners. So did you miss this or can the stock still be Nico, wow. I mean, we could say what a crazy week for Lumen and, frankly, for the entire AI sector, but it's been a crazy year, especially for you.
So, you said you're sold out for the next few years. You, you're like those Hermes Birkin bags. No one can take what you produce. Can you tell me the moment, the exact date on the calendar, when you will start to have more inventory? Well, Liz, first of all, I want to congratulate you on your new show.
Very, very good, and I can't imagine anyone better to run this format. Um, first of all , you know, in terms of demand, we see that demand is increasing every day . So your question is relevant. When will we finally catch up? And it seems like we're running in circles to some extent . We're building, we're adding a lot more capacity, but every day that we build, our customers seem to want even more.
So as you mentioned in the introduction, we see a lot of volatility in the markets overall, but when we look at our demand profile, it is strong and getting stronger every day. So, we are doing everything we can to catch up, but so far we don't see an end to this situation. There is no end in sight.
Okay, that sounds great, but I need to dig deeper into the procurement obligations you have from your customers. Are these forecasts and allocations that clients can opt out of? We made it much harder for customers to opt out. You know, as you've been covering our story for most of last year, we have to invest well ahead of the cycle.
So, one of the things we do is we have our own factories. So, you just mentioned TSMC a minute ago, we are kind of the TSMC of the laser chip space. And when we invest, we invest three years before we can see demand. So, to do that, we want to spread the risks a little bit and shift some of them to our customers.
Our purchasing commitments are largely non- cancellable, so customers cannot quickly back out, and this gives us some confidence as we continue to invest in this huge surge in demand. Got it. So, you tie them to yourself. But the biggest news of the week in the sector, aside from the Chinese memory chip maker's breakout we just talked about, was that rising interest rates hit AI companies that had borrowed billions.
I mean, Treasury yields have risen from about 3.96% in March for the 10-year note to over 5% recently. Your debt is small compared to your size, but your customers—that's a different story, Mike. Have you seen any customers slow down, postpone, or request to reconsider their orders since the rates started going up?
We didn't notice that . Actually, you know, it's interesting. Google, Alphabet's management , spoke just a few days ago and talked about their capital spending growth on an annualized basis. Yes. But for the most part, I think we'll see hyperscalers reporting increased capital spending this cycle. And of course, it agrees.
Before we get started, subscribe to the channel for more news and analysis about Nvidia, AI, and the stock market. Nvidia shares are under pressure. Investors are concerned about debt, data center costs and whether the AI boom can continue to grow at its current pace. But new numbers from SpaceX raise another question.
What if the demand in the AI industry never ends? What if they run out of the capacity and computing resources needed to meet this demand? SpaceX reported a sharp increase in revenue. Elon Musk has outlined an aggressive plan to expand AI computing, and large cloud companies are still struggling to get enough capacity.
At the same time, reports that GPUs are waiting for electricity have sparked new discussions about Nvidia's future. So let's take a look at what's really going on and what these latest numbers might mean. And why the next round of earnings reports from the biggest tech companies is so important. SpaceX figures that attracted the attention of investors.
SpaceX reported revenue of about $14.6 billion for the quarter, up 161% from a year earlier. And that came in above the roughly $12.7 billion analysts had expected. Both the space business and the Starlink connectivity business performed better than expected. But the greatest interest was aroused by the direction of intelligent computing and artificial intelligence, which the company now calls the " superintelligence" segment.
This segment reported revenue of about $7.56 billion . SpaceX also reported adjusted EBITDA of approximately $8.8 billion , a significant increase from the prior year. This number is important because SpaceX is no longer just a story about rockets and satellites. The company is also building a large-scale computing business, providing AI infrastructure to customers who need access to powerful chips.
And this directly implicates Nvidia in the matter . According to the discussion, SpaceX plans to continue building its AI infrastructure on the Nvidia platform. This creates another potential source of demand for Nvidia hardware. But the most striking comment came when Elon Musk was discussing the company's future plans.
He said SpaceX aims to reach about $100 billion in annual recurring revenue by the end of the year. He also outlined much more ambitious plans for the end of 2027, suggesting that annual recurring revenue could reach between $500 billion and $1 trillion. These are extremely ambitious indicators that should be seen as management forecasts, not guarantees.
However, the scale of the plans demonstrates how aggressively SpaceX wants to expand its computing business. The company is preparing to increase its computing power, planning to add about 10 gigawatts of additional capacity in 2027. Musk also discussed a more ambitious scenario involving 20 gigawatts of additional power and cooling, while acknowledging the difficulty of implementing it.
Why is this important to Nvidia? Because every large-scale expansion in AI computing requires more than just software. You need processors, memory, networks, power, cooling, and the infrastructure that brings it all together . If SpaceX pulls off its plan, it could become a much bigger customer for the AI hardware industry.
And it can have implications that go far beyond a single financial statement. The real bottleneck may not be the chips. One of the most important moments of the reporting conference was the discussion of power supply and cooling. Musk suggested that the main limitation is not just getting the chips.
It's about getting enough power and the necessary hardware to run these chips on a large scale. Think about what happens when a company builds a large data center. You can't just install thousands of GPUs and start making money right away. The building needs sufficient electricity, the cooling system must be ready, and the network equipment must be installed.
Memory and servers must work together. The entire system needs to be tested and connected to the network. If even one important component is missing, part of the computing power may be idle. And that brings us to a story that has attracted a lot of attention. Reports indicate that Amazon was receiving Nvidia systems even before some facilities had enough power to operate them.
The report described GPUs that were ready to use but could not be turned on. Some call them "dark" GPUs. At first glance, this sounds like a warning sign. Why would a company spend billions of dollars on equipment that can't be used immediately? But there is an important distinction here.
Comments 0