저자: ChamathPalihapitiya · 외국인
요약 보기
Groq 창업자 발언과 엔비디아의 추론(inference) 시장 위협을 논하며, 엔비디아가 HBM·광케이블 등 공급망을 선점해 구조적 혁신 딜레마에 놓였다고 주장. 아마존 실적(capex 81% 증가, 마진 확대)과 애플 가이던스 하향에 따른 하락을 다루고, 블랙웰 출하 지연은 일시적 문제이며 엔비디아 구조는 변하지 않았다고 봄. 현재 주가는 선행 PER 약 30배로 5년 중앙값 수준이라 비싸지 않다는 입장.
다듬은 글
you know HBM in their design decisions, or these specific optical cables, or a whole bunch of other things, you're going to be in a really difficult spot, because Nvidia bought them all.
You went to another country. Where'd you go this week?
I was in Paris, France.
What were you doing there?
There was an important conference, and Jonathan Ross, the founder of Groq, and I kicked it off with a fireside chat for about 45 minutes at the beginning, just to get everybody amped up and excited.
Cool. How did it go?
Yeah, it was really good. One thing that stood out to me was a soundbite from Jonathan. I think he did a very good job explaining where Nvidia is really strong, particularly in learning, and where Nvidia is trying to build their business, which is inference. But the problem is the features almost compete with each other between learning and inference, and I thought that was a great clarification.
The second thing he mentioned was that if he were to reveal his roadmap, he'll have 50% of the available inference compute by the end of next year.
Wow, based on all the needs of OpenAI and everyone else, that's impressive, and it has a smaller physical footprint, right?
Exactly, and it's because they can do higher tokens per second with less power across all the major models, making it meaningfully cheaper as well. That was a pretty amazing soundbite. His whole explanation was great.
So I'm trying to get a copy of the talk. It was recorded, and I hope we can post it. Anyone interested in studying Nvidia should listen to what he has to say, because the nuances between us and them are really profound. The way he described them was stark, which was cool, and it was great to be in Paris, frankly.
How does Nvidia respond to this movement, or to the notion of an entirely different chip category focused purely on inference? Do they have inference-focused chips in development, or a separate roadmap they'll need to prioritize?
I think it's really hard for them. It's a structural innovator's dilemma.
Why is that?
Because they've made architectural decisions around things like HBM, which is high-bandwidth memory, or specific kinds of high-throughput optical cables. They got up the global supply and cornered those markets as a structural part of their design. It's very strategic and legal, but if you design around that and go through an entirely orthogonal process, you're not subject to any of those supply chain constraints. I think he explained that well. I didn't fully understand it before, but I did after this.
This is important for the hundreds of chip startups and the venture investors pouring money into these companies, who need to know these differences. If you're investing in a company that depends on HBM for design decisions, or specific optical cables, or other constrained resources, you're going to be in a tough spot, because Nvidia has bought them all. That was really interesting to me.
Another thing Nvidia does brilliantly is what Intel did in the '80s and '90s when they wanted to choke CPU competition. They defined a metric everyone started to care about, like clock speed. If you remember, Intel pushed Moore's law, transistor density, and clock speed as metrics of value for many consumers. We didn't know better, so we went through the Pentium upgrade cycle thinking it was necessary, but clock speed doesn't impact speed the way users experience it with software.
It's similar in AI, and we talked about that too. Nvidia is an incredibly important company. They're performing brilliantly, running the same plays Intel did to create dominance, but there are important nuances around power, supply chain decisions, and technology decisions that some might view as unsustainable. I'd be in that camp. The more I understand it, the more I think it was an important conversation, and hopefully we can post it.
Why did we start off with Amazon? What did you think of what you saw there?
I thought it was interesting. Cloud actually missed estimates, but we're seeing the stock move higher.
Yeah, Amazon had a few blemishes, and after yesterday you might think everything would get punished, but not Amazon, apparently. I think the profitability was the big surprise of the quarter, with a lot of it on the bottom line, and that gives investors comfort. The sales numbers were close, so it's no surprise the stock is up. It hasn't been a top performer among the Magnificent Seven tech stocks, so there's some room for it to grow. We've seen capex increase 81%, and at the same time we saw margin expansion. That's tough to pull off.
It really is, and the capex number is important. Now that we're past the big capex spenders, you see all the cloud spenders focusing on capex and spending large sums. Jassy's comments on generative AI and their focus on it mean that story will continue, which is good news for Nvidia, which will be the last of the Magnificent Seven to report.
Let's discuss Apple quickly. Those shares are down just over 1% right now, mostly due to guidance concerns. There were other positive points in the report. According to Deepwater's Gene Munster, the guidance suggested 4 to 6% revenue growth, but the Street expected nearly 7%. That explains the stock's 1% drop, even with good news on iPhone sales and a percent increase. Apple has had a strong run this year due to iPhone 16, and both were somewhat underwhelming in functionality, though iPhone sales were decent. That was better than feared, from my perspective. Since the stock hasn't performed well recently, a 1% decline is relatively mild, considering there's still more to come with the iPhone 16.
True, but given its weight in the market, it's significant. Let's talk about Apple's forward P/E at 31 times. Is this dip a sign that investors are paying more attention to forward P/E valuations? Does this hint at what we might see for the rest of earnings season?
That's what we saw yesterday. It's typical when the market nears all-time highs and P/E multiples hit the mid-20s: you're priced for perfection, and when results miss expectations even slightly, the market reacts, as we saw yesterday. None of the news was terrible across the board, just a bit less than expected. In a market based on strong growth and earnings, that can prompt repositioning.
Comments 0