00:00 Dan
There’s a lot of older hedge fund leaders and and and financial services people that will say it’s never different. It’s always going to be the same and this is going to be another bubble that bursts.
00:09 Dan
I genuinely don’t think looking at 1999 and comparing the quality of the companies leading this AI revolution.
00:20 Dan
You didn’t have Nvidia, Google, Microsoft, Meta. You didn’t have companies with those quality, the earning quality, the revenue quality, the mos that those businesses have, leading the charge. Now, having said that, I always think it’s important to and you and I probably talked about this.
00:40 Dan
there will be crappy companies like the all birds thing that get kind of pulled into this whole AI, you know, exuberance and people go, oh, it’s a bubble. There’s the example. But when you look at Micron, trading at what about eight times forward earnings, look at Nvidia, depending on the estimate, it’s somewhere between like 17 and 25
00:58 Dan
forward multiple. These are great companies with deep cash flow moats that have a long trajectory and even if it’s cyclical, Josh,
01:06 Dan
this cycle is nothing like the historic consumer PC memory cycles. This is going to last for five, six years at least until we get this agentic, uh, you know, AI revolution.
01:17 Josh
Do I keep it, do I should I just keep it as simple, Dan? Some some of the smart tech investors on the show, they’ll say, listen, I I just keep it simple. Compute demand continues to outstrip supply.
01:27 Dan
Well, that is probably the simplest way to look at it right now is there is absolutely no signal that we are going to be able to meet demand. People keep saying like 27, 28 for memory, our research shows it’s probably closer to 2030.
01:36 Dan
And again, every phase of this revolution, Michael Barry’s been saying for like three years every day now that it’s a bubble. At what point do you just get to say something every day until you’re eventually right? I mean look, we’ll see some drawdowns like we saw today, we’ll see some pullbacks like we did today, but the insatiable demand, we’re just beginning to see. 20% of businesses are using AI at any sort of scale, 20% and only about 2% of consumers are paying for AI so far.
01:54 Dan
I keep saying we are really early, Josh. So we’re barely seeing the other side the demand side revenue.
02:00 Dan
And this build out is existential. These these CEO’s, Zuck and and and Satya, they understand they have to invest right now because being wrong by not having enough infrastructure is going to be way more risky to their long term than overinvesting.
02:16 Josh
Dan Hylig, bringing you in. What do you make of this red hot move in in chips?
02:21 Dan Hylig
I think it’s partly, uh, perhaps a kind of recognition that CPUs have been, well, I mean, it’s different chip, right? CPUs are are becoming more important to the AI trade generally. You know, uh, I mean, you look at what uh all these companies are doing. They’re pushing agentic AI and you know, in order to run agentic AI.
02:37 Josh
Look at that move in Intel.
02:38 Dan Hylig
Right. Yeah, which is, you know, I mean, you you got to I mean that’s it was 400 something percent in like a year.
02:44 Josh
Yeah, it was up 250% last I checked this year.
02:47 Dan Hylig
Yeah, yeah, yeah, year date, yeah, yeah, yeah. So I mean, you know, I think a lot of that is is perhaps this recognition of of agentic AI, you know, just becoming more popular or being told that it will become more popular and needing CPUs for that. You have the likes of, you know, uh Intel obviously in there.
03:03 Dan Hylig
Nvidia has its own arm chips as well. They’re they’re uh shipping servers to uh meta that are just running their CPUs. AMD is in that game as well. Qualcomm is pushing further into that game as well. Uh I I do think uh on the memory side, you know, as Stan’s saying, uh you know, God, I hope the 2030 thing is wrong because I need to get video games and that’s it’s like the worst thing in the world having the video games, but we’ll talk about that another day.
03:26 Dan Hylig
Uh you know, I think that those are going to continue to be an issue as they because you need more memory for all of these types of of of models. And so when you’re running different models, you’re running different agentic uh applications, you’re naturally going to need more servers that run more memory.
03:45 Dan Hylig
And so it’s not just memory, it’s also storage because you need a huge amount of storage on the side to continually pull from. And you don’t want to have to constantly pull from that memory. So you have uh that that storage rather, so you have a large chunk of memory to hold it there. So it’s, you know, it’s, it’s one has to go into the other. That’s just how a basic computer works. Uh, and so as they continue to build out, as they continue to run more applications, as they continue to build more models, you’re going to need more and more memory and more and more storage.
04:10 Dan Hylig
And I hate that we call RAM memory. It’s just RAM. We’ll just call it that. You know, I deal with memory whatever, but sorry, it just annoys me.
04:17 Dan
But but it it’s it’s interesting, but right now any compute that can be built, you kind of ran the list. Intel hasn’t really built more or shipped more yet. Pricing power, demand, they’re selling old inventory at extraordinary margins.
04:29 Dan
There’s a lot of enthusiasm about its future. I was I was pounding the table at 20 bucks. So now everybody, you know, hello. But like you look at I could’t
04:36 Josh
find someone by the way. It very tough to find anybody out on the table. I mean how many people, how many people were you saying were taking, that thing’s a bust.
04:42 Dan
No people people told me it was crazy. I my thesis was always about national security. I always thought eventually the US would come around to needing to make more chips. So I got the rest of it right kind of by accident.
04:54 Dan
The CPU thing, everybody missed. Everybody missed how big the agent move would be, uh and what it’s done for arm, AMD, Intel. You mentioned Qualcomm, very interesting one, like they’re going to do CPUs, they’re going to do AI accelerators, they’re going to do and and what I was getting at there Josh was every chip that can be made. You see the enthusiasm for Cerebras, by the way, putting memory on the chip instead of HBM. Everyone’s really excited about that. And it’s not zero sum. So, anytime you’re up here and someone, you know, you want to ask the person, well, what does AMD success mean to Nvidia?
05:22 Dan
It means nothing. Nvidia will keep selling everything it can build and so will AMD. So will Qualcomm, so will Intel. We’re a few years out from having any chance of creating any sort of uh of level in the market where we can actually start to see the best products start to shake out meaningfully. Right now, it’s really about, do you have capacity? Can you deliver? Can you stand up uh gigawatts? Uh you know, can you energize? Can you plug it in? Can you serve it? Um it’s going to get sold.
