저자: LarryFink외 · 외국인
요약 보기
AI 데이터센터 투자 붐은 지속될 것이며 수요가 공급을 초과하는 메모리(HBM 등) 공급 부족이 이어진다는 주장. 중국 저가 AI 모델이 수요를 줄이기보다 확대하며 인프라(GPU·메모리) 수요는 계속 증가한다는 논리. 인텔은 과거 급등 후 추가 상승 여력이 제한적이라 신규 매수는 비추천(보유·관망), 3~5년 관점에선 데이터센터 관련주 보유 권고. AMD는 AI 수주 확대로 긍정적. 메모리·반도체 장비 등 진입장벽이 높다고 평가.
다듬은 글
Data centers finance the purchase of chips. As I said before, I think we're going to have markets, you know, investing in computing power. We will have a futures market. This will be the next revolution in finance, and it is happening because of the strength of the United States capital markets, which have the ability to finance these new technologies.
So the United States and other places in the world can be leaders in new technologies. You know, you mentioned 50 years. So I'm curious. I mean, have you ever seen anything like this, Larry? We have companies that were the most profitable in the world that decided -- bad word -- decided to no longer have free cash flow, to no longer return capital to shareholders, because they have to participate in this mad race to stay ahead of the AI future .
When I hear myself say that, there are some disturbing aspects, right? I mean, I spoke to some leaders this week and last week in this sector. Their biggest concern right now is that supply is not keeping up with demand, and you can see that reflected in the stock prices of memory manufacturers. Okay, we have a demand for memory that exceeds supply.
Of course, they were able to significantly increase prices. I don't know how stable it is; maybe in 3-4 years we will have enough supply. But, you know, and what we see as a big investor in data centers, the demand for computing power is not slowing down; it grows faster. The problem we have as a country: we are not investing fast enough, so I have the opposite opinion.
So we even...we need to do even more, I mean , a trillion dollars in corporate capital spending is not enough . I think it's not just about that. We don't invest. Profits are almost twice as high, strong forecast. We're still seeing a sell-off here. What happened? It's not the best day for the tech sector as a whole, but I think it reinforces the point I was going to make today: Intel's story is mostly over.
This action seems to have exhausted itself . I wouldn't start trading Intel stock now. There are other things I like more. The thing about Intel is that it had a great year, as you mentioned, for the stock, but the returns over the last three years have been pretty marginal. In fact, there was no growth in profits.
When you see a stock move like that , especially in a company with a history like Intel, which has been in Silicon Valley for decades and has been a giant for many years. I think that the main big money has already been earned. I would be very reluctant to start trading these stocks today. Maybe if you could buy them at an extremely low price and you really believe in them in the long run, but that's the maximum—hold.
I wouldn't say it's anything more than " holding," but rather even "selling." Sean, what do you think about what Michael just said? Has Intel's potential run out, or are their numbers sufficient to maintain some optimism? Well, I think in the short term I would probably agree. I mean, especially when you see that the stocks have gone up so much, right?
You know, trees don't grow to the sky, as they say. But if we expand the time horizon, we have to remember that the plant here in the US is not yet operational, but it will eventually be operational. I think the costs of AI are serious. We own many chipmaker stocks in various ETFs, but we actually own everything in a data center: from chips and hardware to software, networking, and cooling systems.
We call this the “big spenders vs. big beneficiaries” strategy. Those who spend are hyperscalers. They are not slowing down, and the beneficiaries are everyone who provides data center equipment, and Intel is part of that. So, depending on your time horizon, if you own these stocks, you should probably continue to hold them.
I own them personally, and we own them in our firm. Not sure I would open a large position now. If they had sagged a little, I would probably have done it, but if you look at the 3–5 year perspective , I think you can get a higher than market return. I don't think you'll triple your money in the next 5 years, but I think there's still potential for stocks to grow over that period.
And Michael, let's discuss some of the potential avenues where we see this growth. The data center business grew 59% last quarter. Looking at this number , is Intel proving or trying to prove that it is becoming a significant player in AI infrastructure, or is one quarter simply not enough? Well, one quarter is n't enough, but I agree that in the long run it's probably fine.
But the scope of AI is very broad. So it's not just chips, but also cooling companies like Vertiv and stuff like that. Another company I really like in the AI data center space is Comfort Systems, HVAC. Who would have thought that HVAC would be part of the AI game? If you had told me this three or four years ago, I would have laughed at you.
So this is a good option for investment. As for data center chips, I think AMD is a stock I've mentioned a few times. in your programs, and I still like them, as the company scores one big win after another in the world of AI. Oracle, Anthropic, Open AI, Microsoft —they received orders from all of these companies.
So, essentially, all the big players in AI seem to be making their way to AMD's door. Sean, when you look at expenses, you know, that seems to be the focus of this reporting week, and maybe the entire season, and the same will be true for those companies that we haven't heard of yet. But we saw similar reactions from Alphabet and Tesla.
We saw a thesis from these companies : "we need to spend now, and we will prove the results later." when you look at Alphabet's reaction, and I guess we can add Tesla here. Should they be viewed differently than Intel, or should this “spend now, earn later” thesis be viewed in the same vein? Well, a few years ago we created an ETF.
Ticker —TRFK. It actually owns everything that Vertiv— that's the name a previous guest mentioned. Um, you know, over the past few weeks, investors have become increasingly concerned that a wave of cheaper AI models could weaken demand for hardware from the companies that are fueling the AI revolution .
But not everyone agrees with this conclusion. In fact, more and more investors believe the opposite could happen. Their argument is surprisingly simple. Cheaper AI does not destroy demand. He expands it. Every major technological revolution has followed a similar pattern. Computers were getting cheaper, and more and more people were buying them.
Internet access became cheaper, billions of people went online , cloud computing reduced the cost of launching applications, and software developed rapidly in every industry. Artificial intelligence is following exactly the same path. The emergence of more affordable AI models does not necessarily reduce the need for computing infrastructure.
Instead, it allows a much larger number of companies to implement AI that previously could not justify such costs. This is why recent competition from China may actually strengthen the long-term investment potential of AI infrastructure, rather than weaken it. Chinese developers continue to release increasingly sophisticated open-source models, making advanced AI accessible to a wider audience.
However, these models still depend on a huge amount of computing power. Every AI answer still requires GPUs, and every large model still requires modern memory. Each output request consumes data center resources, and while the model may be getting cheaper, the infrastructure behind it is not going anywhere, and this is an important limitation.
The level of hardware for AI is becoming increasingly competitive. The infrastructure layer remains extremely difficult to replicate. Building advanced AI models is a complex task. Creating the semiconductor ecosystem necessary for these models to work is even more difficult. Companies like Nvidia, Micron, Broadcom, Western Digital, and others occupy positions that cannot be easily replaced overnight.
Barriers to entry remain extremely high . Take memory for example. High-speed memory has become one of the most valuable components of modern AI systems. Demand has grown so rapidly that suppliers are signing multi-year agreements with customers simply to guarantee future supply volumes. This is a significant change from the previous semiconductor cycle.
Historically, memory has been considered one of the most volatile businesses in the industry. Today it is becoming one of the strategic ones. As AI systems continue to grow, they require significantly more memory, wider contextual windows, more complex reasoning, longer conversations, and enterprise-scale deployment.
Each of these use cases consumes additional storage capacity, and there is no easy replacement for this. Industry analysts increasingly believe that memory will remain one of the biggest constraints to AI expansion in the coming years. This explains why companies continue to talk about a shortage rather than an oversupply.
This also explains why memory prices remain unusually high. Of course, higher prices ultimately stimulate additional investment . New plants will be built , capacities will increase, competition will intensify, but semiconductor production operates within multi-year schedules.
Comments 0