AI adoption is early, but power and scaling remain constraints

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

1. 9:04

   reality because I own all these stocks. But it seems to me I completely accept what you're saying about the demand. There's no question the demand is there. There's no question the CapEx is there. It seems to me there are two potential pitfalls to the whole AI trade, and I like your feedback from it. So the first one is just, you're building all these data centers, are you going to have power? Now, there are two data centers in Santa Clara, California, which were on the drawing board in 2019, that are fully functional, and they only have one problem. They can't plug in because it's just the utility out in California can't supply the power.

2. 9:48

   So it takes, as you all know, it takes time to build all this stuff. So I think there's a legitimate argument to... be said that some of this stuff may take longer than people think, and that would cause a delay. That doesn't mean that the AI trade is wrong. It's just that maybe it gets pushed out, we have a correction, et cetera. I think the more interesting critique is the argument that you find on Twitter, which was very much like a minority opinion, and now is gaining some credence, which is that continuously scaling LLM models is becoming now diminishing returns.

3. 10:26

   Now, there was a guy, Gary Marcus, who was like the only person on Twitter that I could even find that had this argument. So I just read him not knowing anything. Okay, this guy's just saying something nobody else is saying. I gotta read this guy. A couple of weeks ago, one of the co-founders of OpenAI took exactly the same position. So I'd be curious what you're hearing because if it turns out that you're getting diminishing returns now and scaling LLM models, then maybe you're going to see maybe six to nine months from now orders start to pull back because the thing that eats the chips are these LLM models. I'd be curious what you're hearing. Look, 3% of US companies have gone down the AI path today. In Europe, zero. They're basically trying to build blockbuster videos. Asia, ex-China, it's less than 1%.

4. 11:16

   Middle East are starting. Sovereigns are starting. So just to take a step back, like where we are, how early days we are in the AI build-out. I would say, like, from the Palantirs to the Mongos to the Snowflakes to everything we see on hyperscalers, look, use cases are accelerating on the enterprise. So even though, like, that whole argument on LM, my view is, like, that is just tip of the sphere relative to where this is going as an enterprise consumer trade.

5. 11:44

   Next few years, it's all about enterprise. But my also view is, like, humanoid robotics is going to be real in the next three or four years. Autonomous is going to be... We're not talking just in our lifetime. We're talking about the next three, four, five years. So my view on the data center perspective is there's more data centers under construction today than active data centers in the United States.

6. 12:05

   I do agree with you. The biggest constraint is energy. So I'm a huge believer in nuclear energy and where we need to take that. By the way, even if you're a believer, which I am a believer too, you don't just wave a magic wand to build a nuclear power plant. I mean, it takes time. But given China, given where they are from an energy... perspective, it is an arms race what we need to do. Well, the language we've been using, Steve, is, you know, so much of this maybe 2.0 or 3.0 AI trade might be the war beneath our feet, right?

7. 12:35

   All the resources and the commodities and the metals and the rare earths and the energy that's necessary to do what Dan believes is coming to fruition here. So, you know, I look at maybe what was maybe the more important or one of the more important developments of the last four or five months is this turn in commodities, right? This may not be a cyclical signal. This may very well be a reflection of Ai2.0 of the raw material necessary to make this happen.

8. 13:01

   I mean, look around the world, whether it's GE Vernova in U.S., whether it's Siemens Energy in Europe, whether it's Mitsubishi Heavy in Japan, right? You could kind of go geography by geography and say this power theme really transcends national borders here. This is a really, really prevalent theme. Yeah. No, those are great thoughts.