저자: JeremySiegel · 외국인
요약 보기
결론: AI 반도체 수요 확대는 지속되며 단기 시장 방향은 상승, 다만 금리·경쟁·자금조달 리스크로 분산투자 필요. 시장 전망: Siegel은 시장 방향이 상승이며 최근 몇 주간 방향 변화는 거의 없었다고 봄. AI 성장 열차는 둔화되지 않는다고 주장. 장기(2029~2030)에는 성장 둔화가 예상되나 여전히 성장 환경이라고 봄. 에너지: 러시아 제재 완화는 유감이라고 언급했으나, 디젤 선물 20센트 하락은 단기 효과가 있다고 평가. 근본 해결책 아님. 반도체장비: Rolf(반도체장비 애널리스트) 전망으로 WFE 투자 2026년 약 30% 증가, 2025~2028년 시장 두 배 확대 예상. ASML 지표도 우호적. 불확실성은 2029~2030년 사이클과 최종 AI 수요. 엔비디아/AMD/브로드컴: 엔비디아 2027년까지 매출 70~100% 성장 전망, AMD 매출 두 배 전망, 브로드컴 유사 성장률. 단기 전망은 안정적이라 판단. CoreWeave: 어제 15%, 오늘 5% 이상 하락. 수주잔고는 크고 컴퓨팅 수요가 공급 초과. 금리 상승 시 자금조달 부담 증가, 소형 네오클라우드는 2~3년 내 통합 예상. 붕괴보다는 통합 시나리오. 엔비디아 리스크: 2029~2030년 성장 둔화 및 AMD, 브로드컴, 미디어텍에 점유율 상실 가능성 언급. 메모리/마이크론: 마이크론은 기대 EPS의 6배 미만에 거래, AMD·인텔은 약 14배 이상. 중국 CXMT 경쟁 우려는 존재하나 HBM 등 고급 메모리에서는 격차가 크다고 봄. 장기공급계약(2030년까지 26건, 일부 take-or-pay)이 사이클 특성을 바꿀 수 있다는 논리. 다만 이론 단계. 채권/헤지: 25년 만에 보는 높은 실질금리로 채권이 헤지 수단이 될 수 있으며, 소외 종목(laggards)도 헤지 가능성 언급. 시장 변동성 확대 시 분산 필요. 리스크: AI 수요 둔화, 데이터센터 계획 축소 가능성, 금리 상승에 따른 자금조달 어려움.
다듬은 글
In the coming weeks, let's jump right into our keynote guest, Wharton School Professor Jeremy Siegel, professor of finance and, of course, chief economist at WisdomTree. Nice to see you. Let's start with the news. I want to hear your reaction. Of course, any positive information in the energy sector will be well received by the market, regardless of whether it is a play for the midterm elections or something else.
The market doesn't particularly care if only oil and gasoline prices decrease. Yeah, honestly, I don't like that it looks like Trump made a deal with the devil. By which I meant Putin. And politically I don't support it. However, I saw that the price of diesel fuel in the futures market dropped by 20 cents per gallon, which helps by 4-5%.
A representative on a previous show noted that this is about 2-3 days of diesel supplies. Of course, the volumes must increase. This is not a permanent solution at all. It's something like a short-term Band-Aid. And you know, reducing or lifting sanctions against Russia for its invasion of Ukraine, I believe, is a very unfortunate consequence.
Yes, Paul Sankey, of course , along with Brian Sullivan, are looking at numbers that point to an overall insignificant picture. Ahem. Nevertheless. Nevertheless, positive things are happening, progress in the field of AI continues and is not going anywhere. And the prospects for economic growth thanks to the AI revolution this week are as real as last week and the week before that.
So, really very, very little has changed in the last couple of weeks in terms of market direction. The market direction is upward, and nothing is stopping it. Yes, we had a bit of a panic yesterday. Is it fair to call OpenAI's earnings report just such a panic? Yeah, and I think Jim Cramer was talking about how they were looking at the wrong earnings, and that shows sensitivity.
Listen, for any growth stock, any bad news is immediately taken into account. So, listen, there will always be rumors. I mean, you know, it all started two years ago with DeepSeek, and throughout that there was, you know, a situational awareness foundation. There will be, there will be some obstacles along the way, but this car, or perhaps the train, called AI, is not slowing down.
So, you have no concerns about this trade, as we are not about that train. I mean, I think it's very...Look, there are risks. That is, there are great prospects and great risks. I don't think I've ever seen the spectrum of what can happen in the economy be as wide as it is today, which means you still have to diversify.
And I start looking at bonds and I say that at these rates, especially these real rates that I haven't seen in 25 years, it could be a hedge, as could, of course, the lagging stocks, the so-called securities. If there's a break in the train's movement, and, you know, it could be anything. You know, suddenly calculations become less necessary.
The breakthrough means we don't need to build those data centers. I mean, you can list them one by one. You need balance on the other side. Um, although of course hedging is always painful when the trend continues, as it has been for the past few months. Rolfe. Yes, hello, Marley. And thank you for inviting me.
Well, you know, in general, there's still a lot of optimism in the semiconductor equipment space . Most people I spoke to expect spending on WFE, i.e. wafer and fab equipment, to increase by about 30% this year, next year, and even through 2028. So, we see a market that is set to double from 2025 to 2028.
So optimism about the near future remains very strong. As for the end of the decade, 2029-2030, expectations are still a bit uncertain, you know, what this cycle will look like at the end of the decade. But in the near term, everything looks very favorable. And this has also been confirmed by ASML over the past few days.
So we are looking at favorable indicators . We see signs that AI development is accelerating until at least 2028. Where do you hear any warnings, Rolf? Well, um, particularly, uh, particularly when it comes to estimating what the demand will be, the ultimate demand for AI at the end of the decade. There are still questions here.
This is also what we've been seeing in recent days, uh, with, for example, CoreWeave. Um, there is skepticism about whether this development can and will be financed and whether the ultimate demand will meet the high expectations that the market currently has. So, there are certainly doubts, but in the short term, we should not forget that Nvidia is forecasting revenue growth of 70-100% by 2027.
AMD expects to double its revenues. Broadcom has very similar growth rates. So the short-term outlook looks safe, stable, but after that, at the end of the decade, yes, it remains a bit of a question mark. And let's talk about CoreWeave, because it's obviously an exciting movement. I mean, yesterday's 15 % drop.
Today we see another move down by over 5%, dragging many other neo-cloud companies along. When you look at it, I mean CoreWeave has a significant backlog. When analyzing CoreWeave, how do you weigh the company's order book against the financing and actual construction of the infrastructure needed to fund and execute those contracts?
Yes. Ultimately, I believe that for large neo- clouds like CoreWeave, given the limitations we continue to see in compute capacity, the demand for them far exceeds the supply. So I expect that an integral part of the AI infrastructure chain, like CoreWeave, will continue to do well. This does not mean that financing new projects will not be a difficult task for them.
Obviously, as rates rise, it becomes increasingly difficult to justify building these data centers unless you have a very strong balance sheet. Now CoreWeave is backed by Nvidia, you know, capital investment, financial support, GPU supply agreements. So I expect them to continue to do well , but some of the smaller neo-clouds may be under threat in the coming years.
At the same time, all these companies have assets, tangible assets. They have GPUs, power supply agreements, data center buildings. So I don't expect this space to collapse, but I do think we're in for some consolidation over the next few years. So, not a collapse, but some consolidation and perhaps even a slowdown.
This is exactly the conversation I've been having all morning. If the cost of borrowing remains high, could this cause some slowdown in expansion while demand for AI holds? You mentioned Nvidia, do you see a point by the end of the decade where this will potentially negatively impact Nvidia and semiconductor equipment suppliers?
This is a very likely scenario. I wouldn't call that a reasonable base case scenario. You know, my base case scenario is continued strong growth, almost doubling capital spending in 2027. Another year of strong growth in 2028. We've already heard Nvidia and Broadcom talking about 2028. But 2029-2030—it 's reasonable to assume a slowdown in growth, but I still expect a growing environment .
And yes, Nvidia, they will also see a slowdown in their growth. Not only because of the market slowdown, but also because Nvidia is starting to lose market share to companies like AMD, Broadcom, MediaTek, etc. Let's talk about memory and how it fits into this equation, because the memory sector has been hit hard this week, but it has become an important part of this story.
How do you look at this, a memory in the future at the same time period? Do you see this, as I keep hearing from many people, as the biggest potential bottleneck in the overall development of AI? . Something strange is happening in the stock market. The demand for AI is growing. Big tech companies are spending billions of dollars, and chipmakers are reporting strong business results.
But shares of semiconductor companies are falling. Micron is under pressure, and Nvidia faces market doubts. Investors are concerned about new competition from China. So, here's the main question. Is the AI boom slowing down, or are investors overly concerned? Let's figure it out. First, let's talk about Broadcom CEO, Hock Tan.
He recently warned that many investors don't fully understand the companies they invest in. Some investors expect too much, and when things don't go according to plan, they start selling. But here's what's interesting. According to market analyst Gil Luria, many AI companies have reported better results, stronger demand, and better prospects.
This applies to chip manufacturers, chip equipment companies, and cloud providers. So the problem may not be weak business results. The problem may be that investors are unsure how long this growth will last . And that brings us to Micron. The biggest concern is competition from China. Chinese chipmaker CXMT is working on improving its memory chips.
Recent reports about their new technology have raised concerns that China could produce cheaper chips and rapidly ramp up production. Market experts noted that China's share of global DRAM revenue has increased from approximately 7.2% to 9.5% over the past two quarters. This is a change that investors cannot ignore.
But there is an important detail here. Not all memory chips are the same. Basic memory chips and the advanced memory used in AI data centers are very different. lso known as HBM. These chips must process large amounts of data at very high speeds. According to Gil Luria, Chinese companies are still much more focused on the lower end of the memory market.
They have not yet reached the same positions in the field of advanced memory for AI. There is another important point. Micron sells its products to major American technology companies, including Microsoft, Amazon, Google, Apple and Nvidia. Restrictions on the purchase of certain Chinese technologies make it difficult for Chinese suppliers to replace Micron in this relationship.
This does not mean that China does not pose a threat. Chinese companies are improving their technologies. But investors need to understand where this competition matters most. A new memory chip doesn't automatically mean Micron will lose its biggest customer. The real question is whether Chinese companies can produce these chips on a large scale, maintain quality, and compete at the cutting-edge of the market.
Now let's look at why Micron shares have become so interesting. According to market discussions, Micron shares were trading at less than six times expected earnings per share. Meanwhile, AMD and Intel were trading at a much higher multiple of earnings— around 14 or more. This is a huge gap. Why is the market valuing Micron so low?
The answer lies in the history of the memory business. In the past, memory prices moved cyclically. When demand was high, prices rose, companies earned more, then manufacturers increased production, and supply increased. Prices were falling and profits were under pressure. Because of this history, investors often assume that Micron's high profits won't last long.
But the AI boom could change this model. Modern AI data centers require vast amounts of advanced memory. And large technology companies need stable supplies to build their systems. They cannot always wait for prices to fall or additional production volumes to appear. This makes long-term supply agreements more important.
Neil believes this could change how investors value Micron. If profits become more stable, the market may stop perceiving the company as a regular memory manufacturing business. But this is still just a theory that needs to be proven. Micron should continue to deliver strong results. And here's another detail worth paying attention to.
Market discussions highlighted 26 long-term agreements covering revenue through 2030. Some of these deals have been described as "take or pay" contracts. Simply put, these contracts may require customers to pay for agreed-upon purchases under certain conditions. This could give Micron a better understanding of future demand.
It also helps customers secure the memory they need for their data centers. Building AI infrastructure takes time, and companies need memory, power, cooling, and specialized hardware. They can't always replace these components quickly. This gives reliable suppliers an important role. However, long-term contracts do not eliminate all risks.
Investors should continue to monitor demand, pricing, and competition. But these agreements help explain why the current memory cycle may be different from previous ones. Now let's talk about the broader semiconductor market. Chips and stocks often rise and fall together. When investors are nervous, they may sell several shares of semiconductor companies at once.
This can hurt companies even if their individual business results remain strong . South Korea plays a major role in the current sell-off. SK Hynix is one of the world's largest memory chip manufacturers. The movement of its shares could affect how investors perceive the entire memory industry.
Comments 0