Cerebras IPO in focus as Nvidia earnings loom

Epistrophy Capital Research Chief Market Strategist and The Drill Down Podcast host Cory Johnson joins Market Catalysts to discuss the Cerebras IPO and key factors to watch in Nvidia’s (NVDA) upcoming earnings report.

Cerebras, such an interesting story. I talked to the CEO Andrew Feldman yesterday and we talked about, you know, the

Now he's a billionaire. He was happy.

He's worth yes, he said it's a pretty good day. A $3.2 billion dollars is reportedly how much he was worth after yesterday, um making these big chips that their argument is are better suited to AI, can do can uh reduce latency, make things work faster. Um, but I I guess I'm curious first of all, what do you think of Cerus specifically and then I'm curious what we think of like what it shows about the bigger environment.

So, I've been on this theme for a little while in the chip world that Nvidia's and I emailed you about this idea, but the Nvidia's obscene gross margins has made a lot of money for Nvidia. Nvidia, I saw this headline yesterday that Nvidia is now worth more than silver. The second most valuable thing in the world. Gold is more valuable than what

It's down a little bit today, but, you know.

Yeah. Nvidia. but uh down 4% maybe on after some China news didn't turn out to be so good, I think. Okay. So, uh, but Nvidia's margins are crazy high. I mean, they were high when they were 60% and now they're closer to 75%. And while that has allowed the company to make boat loads of money when everyone else in AI is bleeding money or a lot of companies and I are bleeding money. Um, it's left room for other solutions. So it's let other companies develop training chips uh like Tranium the and the TP from from Amazon TPUs at Google. Um AMD's

Yeah. So, it's

doing some business uh with Oracle and with others. Um, and and of course uh Intel's out there. But, um with with with its own offerings. But what's going to be interesting is what happens in inference. Because you might not need the chips that you needed for the Blackwell might have been great for inference. It was cheaper and everyone would have just stuck with that. But it's so expensive that lots of other companies have found room to be creative. Cerus one of them, Potronic company I'm invested in as well.

But Cerus's chips are also expensive, if not more expensive.

Yeah, but they do so much more and they do so much more on an inference level. So, they can manage so much more information so quickly. and it's a it's a brilliant design where they've got the memory kind of built into the chip. They've managed power and heat in ways that weren't thought to be possible. And they're just crazy honking and big. Did you guys talk about that much yesterday?

Yes, he held one up while we were talking, but we did talk about dinner plate or iPad or whoever it was. Or or as our our Dan Hally likened it to a pizza pie. Like you've got most of the chip makers that just have a slice or two, whereas

Well, even even the Blackwell's huge. I mean, you know

It is because they put different chips together though. They don't make it as a single chip to begin with.

Right. You buy the you buy the Chiplet. what they call. But but in fact it is a very large chip. Uh comes out of the die even bigger just the basic chip in itself and many layers. But um, you know, it's an interest I I I pitched this magazine story once.

Yes.

Recently to a friend who runs a

Which was what?

Which magazine? No, what was story big getting. I thought it'd be clever magazine story to show what the Intel chips, the CPUs looked used to look like and now the Nvidia chips. Now the Cerus thing and then they could fold out a big page. Nobody nobody listens to me, Julie. Except for you. Um, I think that uh, uh, what we're see from Cerus is that Nvidia left room in the market for different solutions for inference. When Nvidia came out of the gate about three or four years ago, suddenly dominating AI, putting up prints that were doubling revenues year after year. Other companies said, where are we going to be five years from now? Let's make chips for them. And this's what Cerus is, that's what Potron is. That's what some of the Intel chips are going to be. I could see AMD focusing on this as well. And so we'll see what the dominant chip company is in five years. You might not be Nvidia anymore.

Well, the the other question I have too is and I keep asking everybody this today. Right now, we are in a moment where no one cares how much this costs, meaning the

Absolutely true.

When do we get to that point of more price sensitivity where this like the the sort of like fever dream of compute at any cost starts to fade.

When there is capacity. and there isn't capacity. I don't know when that's going be. And it's no one knows when that's going to be and I've I spent uh the day yesterday at Oracle in Nashville, Oracle's headquarters in Nashville now. I spent the day with a bunch of executives there in an off the record briefing, but I learned a ton. And one of the things I can tell you and I hear this from every company I talk to at every stage of AI is they cannot build it fast enough, literally, they cannot build it fast enough. For every single thing they can build, they've already got a sale ready and they can sell more if they can build more. And so the whole the race is to build it fast. And so that's what you see. And and this is different than the cycles we used to see. The cycles we used to see in compute used to be governed by um the the evolving pace that used to seem really fast of a new Intel chip and a new Windows operating system. And that is out the window. This is about building for a future of AI that we don't even know what it is. Agenic AI is driving growth at rates that were seeming impossible just two years ago when AI growth rates were jaw dropping. And we don't even know where this is going to go.

What's so fascinating to me about this stage, if as you frame it like that is, you see anecdotal evidence out there that the data centers that are out there are not running at full capacity.

At least because it seems like everybody is building for the eventual, right? But that right now, it's not necessarily being used.

Yeah, I I think the use cases and how data centers work is evolving. So let me explain what I mean by that. Some use cases are very GPU intensive and don't really need a ton of memory at a given moment. Others might need a lot of memory at a certain moment when it's doing a lot of of of uh uh big compute problems and not really using the GPU as much. And so you build a um a data center for max capacity. But you don't run it at the max all the time. You don't run it at max power, you don't run max systems. And we see changes uh ongoing literally right now. There are designs of how to use data centers in ways that can shift loads within the data center to use the things that the user needs, the business needs at the moments that it needs them and free up other resources otherwise. But the but building for max capacity and not running at max capacity, you know, when you're when you're driving an F1 car, you're not at full tilt in the turns. And so the data center's not being used to capacity. That's a big part of that.

Um, what were your other takeaways from that uh Oracle visit?

Oh, I have so many. Uh, I've got so much to write about that. Um, that this isn't slowing down. And that that the growth rates coming out of Oracle, I'm biased on this one. I I own the stock. I like the people at Oracle. Um which not never not everybody likes people at Oracle. I love the people at Oracle. Um, they are growing at breakneck paces and they have been a breakneck pace. They've been growing faster than anyone else in the data center business for a couple years in a row. But they've they've got this real advantage that you wouldn't think is an advantage. They came into this a little bit later than some of their competitors. And so they don't have the technical debt. You know, Amazon has to support every single possible use that they've ever supported in the past. So they have hundreds of different products they have to support at AWS. Oracle doesn't have that problem. They've got the best and the brightest and the newest stuff. And so they can run things faster and cheaper and they're winning business in that regard. It's interesting to see. My Amazon sales people tell me, I was texting with an Amazon sales person just this morning. and he's like, oh, let me get this straight. So we they have less to offer and customers want that. But I think when you tell customers what you can give them and you can give them compute. We see Oracle making great sales there and doing a real big business.

Okay, but all the other stuff that Amazon is supporting, they're making money from all that stuff too.

Everybody's making money on this. Yeah.

So, I mean, so that's that's not a terrible.

It's a good business to be in when you can charge whatever you want.

couple more things because we have a lot to get through with you. Um, I do want to come back around to Nvidia and what you're going to be watching next week in those numbers, especially given the run that Nvidia has been on again, right? It was so interesting this year to see Nvidia sort of lagging everything else in semiconductors, right? And then like rocketing up and the stock.

Not the business.

Not the business. Yeah. The business has been doing what it's been doing, which is well. Um, but, you know, coming back, rocketing back to a new record high. It's pulling back a little bit today as we talked about. But like, what are you going to be watching for to see if it continues to support that enthusiasm in the market?

I want to hear things about how they're incorporating Gro and incorporating new technologies. I want to hear about the use of inference. Inference is going to be so much bigger than training, something that I didn't really understand a couple years ago, but w