Subscribe
Share
Share
Embed
A podcast about the business and market of semiconductors
Ben Bajarin (00:01.212)
Hello, welcome to another episode of The Circuit. I am Ben Beharren.
Jay Goldberg (00:08.398)
Greetings, internet. I am Jay Goldberg.
Ben Bajarin (00:11.772)
So while we don't want to spend the entire episode talking about this news bit that feels like it's been dragged out, drugged out, dragged out in the news about Sam Altman's desires to raise what was first, I don't know, whatever, $50 million or something. And then everybody was like, bro, you're going to need a whole lot more than $50 million or billion dollars, then became a bigger number and is now.
$7 trillion, which yeah, so, okay. But despite that, like there's, people keep talking about this, which is noted, right? I'm not sure. I mean, he can't not know people in this industry to be like, you crazy, bro? Or you got a good handle on the semiconductor industry or?
or whatnot, but obviously he feels pain points for open air. He's beholden to Nvidia and he'd be like, give me a million more H100s. Nvidia is like, dude, we can't make those and it's gonna be years before we even get anything. So let's just start there. We know it's been talked about. It seems silly. So be it. It's been talked about before, but here we are today in a world where everybody's talking about.
the mighty Sam Altman trying to raise $7 trillion for something in semiconductors.
Jay Goldberg (01:47.95)
So you realize it's been like 24 hours. Like the journal broke the story 24 hours ago, 36 hours ago. I don't know, something. It feels like an eternity. But yeah, so the journal is saying that Sam Altman wants to raise five to seven trillion with a T dollars to semiconductor. He's gonna make it a verb. He's gonna do the semiconductor things. And like, I don't think, I've talked to a bunch of people about this.
Nobody quite knows what to make of it. I know people who have worked very closely with large Middle Eastern sovereign wealth funds to do other semiconductor foundry stuff. They haven't heard anything from him. So it's unclear to me who he's talking to. I know a bunch of people who would be obvious candidates to support this, and nobody seems to have heard a word from him. I strongly suspect.
that this is something, some form of public negotiation. I think that's part of what's going on where I can imagine the conversation went like this. Sam and OpenAI went to Nvidia and said, hey, can we buy some of your H100s? And Jensen said, sure, it will cost $11 billion.
Ben Bajarin (03:05.902)
$11 billion.
Jay Goldberg (03:07.694)
And you can get them when we feel like it. And so they went back to NVIDIA said, no, that's not going to work for us. We need it sooner and cheaper. We're just going to design our own chip. And so then NVIDIA said, that's cool. By the way, did you see this? We have a custom silicon business. Reuters just put out a story about that, like NVIDIA has a custom silicon business. We'll help you design that chip. And if you don't work with us or one of the others,
you're not going to get any capacity at TSMC until, you know, even, you know, 2028. And, and so then, then he leaked the stories like, oh, we're just gonna, we're going to build a fab. And videos like, cool, bro, good luck. And it just escalates, right? It's, it's just going back and forth escalating. And now it's like, sure. It's, you did I say billion? No, I meant trillions. I'm going to raise trillions of dollars.
Ben Bajarin (03:54.012)
It is...
Jay Goldberg (04:06.254)
And I, so I think that's what's going on. It's like they're negotiating. People are leaking these stories in public as part of the negotiation. I don't, you know, just the, like who has $7 trillion to do this? Like that is double China's U S dollar reserves, total reserves.
Ben Bajarin (04:24.348)
I mean, not only who has $7 trillion, who believes they're going to make that money back and then some on these efforts, right? Because the worst part about the semiconductor industry is when inevitably cyclicality hits and you are slowing fab capacity and burning cash with not a full foundry.
Jay Goldberg (04:52.398)
Yeah, I mean.
Ben Bajarin (04:52.892)
There's a reason why they're very conservative with their capex and they don't overbuild supply.
Jay Goldberg (04:59.822)
Right, right, and I actually look at it more than just the capital. Where are you gonna find the people to do this? Like you're gonna go build a TSMC alternative? Who are you gonna hire to do that? Like.
Ben Bajarin (05:18.524)
Yeah, I mean, I think the issue is like, it's now spun out, like he wants to partner with someone like TSMC. And interestingly, like Intel's be down for this. If anybody wanted to give them money to help fund factories going forward, like that he wants to bring the money and somebody else runs it. Again, though, that's just by itself, not a great investment business. Like you're going to raise venture or sovereign fund government money. That's not the best business. So I don't...
It's all up in the air. Like we're just completely speculating at this point from this narrative that's gone. And I agree with you to the, with the negotiation point, which I do think is interesting. At the same time, you can't for a second believe that Jensen and Nvidia finds you a credible threat to TSMC, to them, to any other factory. Like he'd just be like, laugh it off. Like I still got you, dude. I got you.
where I want you. You are beholden to me. I don't care what you say. There's no way around this. You are mine. You are mine.
Jay Goldberg (06:21.134)
Cool, cool.
Cool story, bro. Cool story. Yeah, I think, I mean, look, for $7 trillion, you could buy 35 Intels. Buy Intel 35 times, right? So I think, I don't think it's, we're not exactly speculating. I think what we're coming to the conclusion of is this is not serious. It is part of some other communications effort.
that likely involves a negotiation between OpenAI, NVIDIA, and TSMC. And we are not going to see any of these companies, we're not going to see a $7 trillion raise to reinvent the wheel of the entire industry. But I mean, part of me though, like I don't want to, he's a smart guy, he's very capable. Part of me wants to poke a little fun though, because this to me is the ultimate software,
Ben Bajarin (06:59.804)
Yeah.
Ben Bajarin (07:10.94)
Yeah.
Jay Goldberg (07:23.342)
person saying, oh yeah, hardware is easy. And I'm sure many of the people listening to this, that sentiment will resonate. The software people seem to think that, oh yeah, it's easy. Well, you just throw some money at it. We'll solve the hardware problem. It's not that easy.
Ben Bajarin (07:30.044)
Yeah, right.
Jay Goldberg (07:48.878)
And I think what is interesting though is just sort of, it is as a thought exercise, it's interesting to think about like, if you were to have all that money, what would you spend it on? Like, this is a complex industry. How would you sort all that out?
Ben Bajarin (08:06.716)
Right. Right. Well, and I, and I, again, I come back to right. The, the business side of this, he made the point, right. And I think the fundamental, I guess, tweet or statement, right. That he said is that we need more energy. Yes. Right. So maybe, maybe, maybe raise money and build a nuclear power plant for a day. I don't know. Right. I'm just saying, right. Okay. That makes sense. Right. You could recoup those costs as expensive as it is. Energy is expensive and you may very well run out of it.
trying to run these data centers. Yes, we need more capacity. Okay, we agree. That is a problem, right? At the same time, Intel's ramping up to try to help capacity and Samsung. So two vested parties trying to monetize and make a business of it are going in that direction. So you look at it to the economic side of what you deploy that capital, it's not sunk costs. You need to make that back. It needs to be in...
in something that's valuable and has economic upside. And it's a really short list. And again, the infrastructure side of Semi's is a very mature process for what it does. I mean, and we recognize we are up against supply constraints, but again, that's because no company is going to overspend supply because they know that...
Eventually those are going to slow down, right? Those machines will slow down. It will be cyclical. We are in a supply constraint now. It will end. Like at some point in time, we will not be where we are today. And by the time this guy, even, I mean, again, I think we agree it's not happening. By the time somebody raises that amount of money, builds factories, fills them out, we're probably going to have plenty of capacity to meet the demand for these. So it's...
It's just weird. Anyway, weird across the board, but that sort of leads us to, I think the question again of the moment is you do have a lot of these companies realizing that their growth upside, and I'm going to say this was a software because I think this is a really interesting point that, you know, we know where we're at compute wise for Silicon and we know that, you know, we just can't make.
Ben Bajarin (10:31.772)
as many as we want, but in a really weird way. And I would, again, love to know what historical parallels we have to this. We're slowing down the software industry, right? We're slowing down software innovation around AI because they can't get enough compute. So it's just super interesting that these two things go together. And so you could see why if you're a software company, you're like, I need to move faster. I want to make more money. If I could get more capacity and more compute, I could make more money. I could scale my software business. I could get more customers.
I could push the boundaries of my innovation. Like we get it, we get it. But we are slowing down all of this effort, right, in software, which I think is interesting. So you see the point that they're up against. Everybody's up against this. It's not unique to OpenAI, but it's shining this really interesting light on this. The software industry is just going to unfortunately not be able to move at its normal pace because we are up against compute constraints.
Jay Goldberg (11:28.43)
Yeah, I think that's a very valid point, right? It is one of those moments in history when there was hardware constraints, commerce. And we don't see those a lot. I mean, I guess we do. Like we used to see that in PC land and in smartphones. In smartphones, in phones, we used to see it all the time. Like every time there'd be a, when 3G first launched, the biggest constraint on 3G adoption was there were no 3G phones.
Ben Bajarin (11:44.54)
very rare.
Ben Bajarin (11:48.54)
Yeah, that long time ago. Great.
Jay Goldberg (11:57.774)
because there were no 3G chips. And we've seen Windows cycles like that, too, where some new version of Windows comes out. But to your point, that hasn't happened in a long time. Qualcomm and Microsoft figured those problems out a long time ago. But this AI stuff, machine learning is moving so fast and it's so new that it is where old problems are reappearing.
Ben Bajarin (12:02.3)
Yeah, right, right.
Ben Bajarin (12:32.732)
Sorry, I'll pause it. If I freeze up, just make something up and keep talking. Cause for some reason the video is dropping. So if I don't say anything, just just mutter. Okay. On to this point. So, but what's curious though is like, so I've been thinking about this too. Again, within the theme of compute constraints. We know that AI software, everything that everyone wants to do and in grading this into their...
Jay Goldberg (12:39.246)
Yeah. Okay.
Ben Bajarin (13:01.616)
back in software systems, into their businesses, processes, whatever, right? We're in a business cycle. Consumers aren't there yet, but by the time consumers come in, again, even more compute. So we have this AI compute and software cycle that we're hitting. And so you see where we're up against constraints -wise there. And then on the flip side, it's intriguing because everybody knows, right? Vision Pro has been on my face, on my mind, almost literally on...
on my mind, outside of my mind. But what's fascinating about that is again, like you could very quickly push the constraints of this machine within a very compelling vision for the future of computing. And so what's odd to me is kind of we have these two different cycles. Like we're on the cusp of a very long, I don't know how long it's gonna take, five, six, seven years, where this concept of spatial computing and mixed reality.
is going to continue to get momentum. Like the cat's out of the bag now. It's going be small volumes for Apple, you know, Quest 4 plus Quest Pro, whatever probably sells in the millions of units, small volumes. But that's a different cycle in terms of compute paradigm. AI though, kind of same thing. It is also a compute paradigm, but it's a lot more ready today because of compute. So I'm intrigued by... One's a giant snowball that's been rolling. One's a snowball that's starting to roll.
But again, both are very, very compute constrained, like in terms of where they're capable of. And so I don't, that's what I was thinking about too, from a historical present also, like when's the last time if ever we had a simultaneous cycle of compute demand? Like not even, like two different ones. One's a platform, visual interface, computing software development in terms of that. And then the other one being driven by intense compute first, everything around AI.
being ingrained into existing software. Like it's just, it's just interesting that we've kind of got two things that are snowballing around the same time. And I, and I rack my brain is kind of the last time mega cycles have happened at the same time.
Jay Goldberg (15:14.926)
Yeah, nothing jumps to mind to me. It feels like we're...
We're on the 49ers time scales here, right? The last time the 49ers were in the Super Bowl was 29 years ago. We are recording this on Friday, the February 9th, two days before the Super Bowl. So I have to throw in the obligatory 49ers reference. It's probably, we haven't had probably seen anything like this since the dawn of the internet in 2000, right? There's a lot.
Ben Bajarin (15:50.428)
because the internet and PC's hit at the same time? Is that the -
Jay Goldberg (15:53.358)
Well, I think the internet and like just the consumer internet happened at the same time as the sort of the infrastructure behind the internet. Those are two very like how we were the kinds of things we were doing with websites exploded just at the moment when cost of fiber optics and IP switching fell to the point where it has to be explosion. I think that was probably sort of the early 2000s where you had the
Ben Bajarin (16:02.652)
Mmm.
Jay Goldberg (16:22.152)
massive internet build out at the same time that there was this lots of experimentation around the software that was going to run on top of it, both happening simultaneously. That's the best analogy I can think of.
Ben Bajarin (16:30.446)
Yeah? Yeah?
Jay Goldberg (16:36.398)
So not quite 29 years ago, but pretty close.
Ben Bajarin (16:41.564)
Cause I, when I think about that, I'm, it, it, it weirds me out because I'm like, like, this is a whole new world of evolution of technological evolution, right? Having these things hit at the same time. And I'm just not sure. I keep asking myself like, can we handle it? Like, can the industry handle it? Can the market handle it? Like there's a lot, a lot going on with kind of two parallel mega cycles that are only related in that they are.
compute intense, like just drastically. We don't have the compute for where these things wants to go. Like, in fact, I think this point is super interesting because this does in a way also apply to AI. That Casey Neistat's video on Apple Vision Pro, like he was just skateboarding around New York, going through the subway, like living it up, wearing Vision Pro. But the way he framed it, I thought was very, very good because he basically said it's like somebody tried to...
like came back from the future, tried to build the future with today's limitations of technology, like today's technology. And in a lot of ways, you could sort of say the same thing about AI. Like we're seeing the vision, like where it can go. Like we get the concept, where might a fully, you know, artificial general intelligence be or, or what it can do. And it takes over workflows and it works for us. And, you know, even how AI or robotics goes into full autonomy of cars or anything, right? That's.
You see that future, but we're bound by today's technologies. Like it's just a, it's a fundamental limiter. And it's just interesting, right? Cause we have all these conversations about how hard it is and how expensive it is to make chips at the leading edge. Um, how people just can't get enough. Like we can't sell enough for these cycles. There isn't enough capacity. And even that what's being made, the use cases or the desire for that software use cases already outstrips component or supply for the future chips. That'll be a year from now.
Like the years from now chips can't even probably surpass or, or exude the use cases we want to do today. Like it's just, we want to run really fast on all of these, but like we can't, we're bound by physics and science and the complexity. And it's just, I mean, I don't know. I just, it just feels like we haven't had something with that kind of perspective where you just, you want to run, you see the future, but you just can't do it. Like, cause you know, like you said, the internet one.
Ben Bajarin (19:06.012)
I like the first time we ever came across anything like that. Like you're just figuring it out. You're like, I don't know. Like a lot of people were really, really wrong. These two cycles, it feels like we're so familiar with technology. We can kind of get that sci -fi future and you could see it. And we just can't run fast enough because we're burdened by, you know, the, the limitations of, of compute supply, physics, science, technology, et cetera.
Jay Goldberg (19:31.406)
I think there's another wave of this that's important too, is we're also undergoing some pretty significant changes in how the chips we work with are designed and manufactured. There's really big changes happening. We're only a few years into EUV. We have all this new packaging, we have backside power and new architectures, all the new FETs.
Ben Bajarin (19:42.94)
Mm. Mm -hmm.
Jay Goldberg (20:00.142)
There's some really significant changes that are all sort of coming right on top of each other in the next few years, which I think are in part put in place to relieve the harder challenges. But it's another very large technical hurdle that's sort of stuck in our face right now with some pretty significant changes.
Ben Bajarin (20:20.124)
Hmm. You know, I like, no, I was gonna say I like, I like this framing exactly what you just pointed out, which we are on the cusp of something that's coming that could feel like a big leap in innovation, perhaps that we haven't before, which maybe alleviates some of these things over a shorter time horizon. I think that's an interesting.
Jay Goldberg (20:21.742)
I will say, though, that go ahead.
Ben Bajarin (20:48.988)
Like it's time for a breakthrough is kind of my point. Like we've had breakthroughs in software, obviously relative to AI, there was multiple breakthroughs in software with transformers and whatnot, but we need, we kind of need a new breakthrough. Like we haven't, maybe in Silicon, maybe design, I don't know what it is, but it kind of feels like, maybe it doesn't happen. And I'm wishful thinking it would be nice, but I kind of like this, like we're in a moment where we could use a breakthrough.
Jay Goldberg (21:15.822)
I look at it a little differently. I would say we've already had the technical breakthroughs, but it's much more about the commercialization and the productization of some of these things. Because I think that's the big challenge with a lot of these new manufacturing processes and semis. We figured out the technical stuff, but actually getting them manufacturable at scale is the challenge in front of us. That's one. I think we see the same thing in AI.
Okay, we have all these interesting new transformer models. What are we gonna do with them? So I see it much more on the productization side as opposed to the technical breakthrough side. I think we've had a lot of the technical breakthroughs and they're more out there. Like I'm not saying we're done with that, but to me it's much more about the practicalities of making these into commercial scalable businesses that we're stuck up against.
Ben Bajarin (22:11.9)
Well, like, so, okay, so I guess I'm saying like, what, what you, you laid a good foundation, right? We've got new EUV machines just being rolled out. Designers are going to look at that and say, Oh, look what I can do. Like something new. We got ribbon FET or the FETs, like you said, so nano sheets. Broad using an industry term, backside power using an industry term, kind of all aligning. And so I guess what I'm sort of saying is I want, I wonder.
If you're in videos of the world, you're AMD, Intel, you know, Qualcomm Apple, et cetera. When, when, when all of that stuff is hit, which is again, an unfortunate, not a year or two from now, but maybe three or four years from now when it's mature, maybe something crazy happens. Like maybe some whole new way to handle compute or handle AI or a new architecture breakthrough. Like I just, I wonder if we're, we're, we need it so badly that that.
creates the right mix of goods where somebody super interesting comes up with something new and we kind of have a new invention, if you will, in one of those vectors.
Jay Goldberg (23:21.614)
So we need AI to solve our AI bottleneck. There we go.
Ben Bajarin (23:24.732)
No, I don't see it. If all AI is going to do though, all AI is going to do is it's going to solve, because this is just me always worried about the matrix. It's just going to solve its own self -serving problem. And at the end of the day, all it's done is that's an elaborate plan to enslave us in the, in the guise of innovation.
Jay Goldberg (23:45.838)
Thanks for listening, everybody. We're going to become a Luddite podcast now.
Ben Bajarin (23:46.348)
No one should have to think about the reasons that Ben wants to build a bunker in his yard for the future. But yeah, I mean, like I said, I'm intrigued by the mix that we're up against, just because again, going back to this point, this reason that Sam is doing this, whether it's negotiation, it's a very big problem. He is outlining.
Jay Goldberg (23:56.558)
I just, I want a Faraday Gauge.
Ben Bajarin (24:15.452)
a significant, like we're up against a really big challenge. And so I just wonder if out of this pressure, out of this heat to wiggle our way through this, something gets created that wasn't there before is kind of my point, right? And I don't know. I just, you could argue that the primordial soup of silicon stuff is there to have something new emerge.
I don't know, I just feel like this is such a big pain point, that was my point. Like this is a real pain point for a lot of people. It gets a huge pain point. And I just wonder, right, are we stuck in it? That's it, just gotta evolve, or might a new invention break through.
Jay Goldberg (25:00.206)
Well, so I think that lends itself to.
Sam Altman's $5 trillion, throwing $5 trillion at the problem, right? Where, you know, I'm joking, but.
Ben Bajarin (25:15.1)
the most expensive innovation breakthrough in history.
Jay Goldberg (25:17.646)
Just like throw money at it because we need all these technical breakthroughs. I think I I'm sticking with my position where it's, it's not so much the technical breakthroughs. It is just the much more mundane day to day. Let's get this to work. Lots of, lots of smart engineers and product managers figuring out exactly which problems to solve and in what order to solve them and commercializing them. Cause I think we've, we've, we have a pretty good track record. We being the broad industry or humanity has a pretty good.
track record of steady rate of incremental improvements to push forward semiconductor design and capabilities. And I mean, we know what needs to get done, right? But like, it's going to take, it's going to take years. Like high NAEUV is not going to be commercially viable for what, five, six years, right? And it's not going to be needed, right? It's not going to be necessary.
But I would say that's a pretty big technical break. I you look at what it actually involved to get that done is just a staggering amount of technical innovation. We have that. What we need is the right mix of getting it to work economically. And before we get to that, we have lots of other plain vanilla problems with regular EUV. I mean, TSMC is not getting on the high end EUV bandwagon for a reason because they see...
Presumably, they see a good roadmap for their existing products. And so I think that's a, it's less exciting. It's less about the technical breakthroughs and more about just everybody doing their job.
Ben Bajarin (26:56.956)
Yeah. Economic efficiencies. I I mean, you're totally right. I did think the TSMC saying, you know, they're not, what did they say? They're not going to jump on it until like 1 .5 or something like that nanometer. Whereas Intel's, Intel's trying to do it for probably, I mean, I would guess similar, right? It's not coming to 18a. So 1 .8, 1 .5 nanometer for Intel also, but Intel wants to be viewed as a leader and probably the undis, I think they want to be viewed as the undisputed process leader. We're going to argue that forever.
but they want to be viewed as the undisputed leader in process technology to get that back. And they clearly feel like they need to absorb that capex of EUV to get there. You're totally right. TSMC doesn't.
Jay Goldberg (27:40.43)
I actually think that lends credence to my point that it's an economic argument because I would argue, and I'm doing a lot of this based on what Dylan Patel wrote, the point of high NAEUV is that it is, it can do most, you don't need it to do a lot of the things we already can do. And if you try to use it for, you know, for a, let's call it 18A,
It's uneconomical. And probably whatever comes after 18, I think it's 14A, the next process for Intel. Like it's not, they're not going to need it, whatever it's called, they're not going to need it then either. They are buying these, this equipment though, because like you said, they want to say, Hey, we're the process leaders. And I actually think that's a, that is a little bit of a blind alley for them because they don't really need to be process leaders. They need to just.
Ben Bajarin (28:16.668)
We don't.
Ben Bajarin (28:37.916)
They don't compete. That's exactly right.
Jay Goldberg (28:38.478)
They just need to be really good at it. And so I have to worry that they're making a slightly uneconomical decision. It's going to cost them more to use those systems than if they used to use regular low NAEUV systems. And again, to my point is, it's not about the big flashy breakthroughs. It's about just do your day job. Everybody do their day job.
Ben Bajarin (29:00.444)
Yeah, just, just, just be competitive. I mean, that's the thing. Like I, after Intel's earnings, like we had a, I had a series of conversations with some folks on the street and you just came back to like, you know, look at the end of the day, they just need to be exactly what you said, competitive. So in a position to compete. So just right there in the vicinity of TSMC in terms of process quality, assure the market that they have separated church and state.
and have a predictable roadmap going forward. Like there will not be future delays, right? And if that happens, they're in a great position for several big wafer scale customers to say, I'll jump to you because I will not get leading edge from TSMC ever, ever. You could be a third or fourth tier customer with TSMC. You're not getting in the top, you know, in the top two years of leading edge. So there's a huge demand for that. And so all it has to do is be there, right? And so,
You're right. Like it's interesting point, right? Because the question is when, when would IFS be able to monetize itself? And if they overspend in this case on something like high, you know, an EUV, that's just going to mean it's going to take that much longer. Right. Cause what is that? What did somebody say the other day? It's a 380, 350 something million dollar machine. Yes, they bought one of them, but still, right. It's, it's, it's many. It's not like it's going to crank out that many more wafers from an efficiency standpoint.
It's going to be bleeding edge, but it's costly, right? So it just means that these bets need to pay off for Intel over a length of time. Like the turnaround story economically is in three, four years from now. Like it's 10 years, right? Or longer before we know if they pulled us off from an economically viable perspective.
Jay Goldberg (30:48.334)
Yeah, and you know, we've been working on this IFS financial model and...
like, you know, trying to figure out like, what do Intel's Foundry Services economics look like? They said they're going to start breaking those out soon. And so I was trying to, we're trying to get a head start on that and look at those numbers and our preliminary numbers will probably see soon, but our preliminary numbers are, they're going to lose a lot of money. They've been losing a lot of money. They're going to lose a lot of money this year and next year and probably the year after. And these are, these are big numbers. I mean, not.
trillions of dollars big, but certainly billions of dollars big. And I think in that context, maybe not worrying about being process leaders is as important as just making wafers very efficiently.
Ben Bajarin (31:40.154)
Efficiently and again at scale, right? I mean, this is again, we agree with many in the industry whose assessments are that chip prices are rising. That doesn't necessarily mean that volumes aren't going up in some pockets, but yes, chip prices are rising. So sure, you're going to make more high ASP products with costs more, but you got to have way for scale customers. That's at the end of the day, they need to be getting way for scale customers. And you know, that's kind of the spotlight on.
where they're at, right? Where they get to, but you know, again, I just, I think it's interesting to think about this from the perspective of, if all goes well for Intel and maybe Samsung, there's also over the next couple of years. I mean, again, if somebody who's listening to this has a different perspective, I'd definitely share in, cause I want to hear it. I'm just not sure we're going to be supply constrained forever, right? Demand will balance.
and compute cycles will mature. When PCs were ramping, when the internet was ramping and infrastructure for networking and telco, it took a few years, but then it balanced itself out. Stuff came back to normal. I talked to imagine that's going to happen again.
Jay Goldberg (32:57.422)
Yeah, I mean, that's sort of, I mean, that's what happened in the internet, right? We overbuilt fiber around the world in the 2000s, we got way ahead of demand. But then the end result was by the end of the 2010s, we had plentiful, cheap internet because the price of connectivity had dropped so dramatically. People had overspent. A lot of those companies went out of bankrupt. A lot of investors lost a lot of money. But for
and consumer was probably in that benefit. We were able to share in that. I don't know if that is going to happen this time around. Because again, it's a very different set of stakes. Economically, we're very constrained. Everyone's being very disciplined. And I think that's been the story of semiconductors for the last 10, 20 years is very, very disciplined capacity expansions.
very, very disciplined products. We see that in memory, where we've gone from dozens of memory makers down to basically four. And those guys all operate very, very cognizant of their others, and they don't raise capacity too much for fear of bombing prices. You look at the whole semiconductor chain. I think we talked about it last time. Stacey Ragsend at Bernstein put out a note saying, more than 100 % of
the gains in semiconductor revenue over the last five years have come from price increases, fewer units, higher prices. That's been the story all along. And I think certainly at the leading edge, that's going to continue for the next, I don't know how long, several years. You made that point earlier that NVIDIA is so capacity constrained that it will not fulfill its orders for H100 by the time its next product, the B100.
Ben Bajarin (34:48.796)
Yeah. Yeah. Yep. Yep.
Jay Goldberg (34:49.198)
launches. And that's a condition that the hardware industry learned the hard way back in the 80s, like Osborne's law, where you're really careful to not announce a new product until the old one is sold out. And we don't need to do that anymore, it looks like, at least in advanced semiconductors. But at the same time, I think it's really important, and certainly all this talk about
Ben Bajarin (35:03.292)
Yep. Yep.
Jay Goldberg (35:17.774)
spending trillions to increase fab capacity has got me thinking about what's happening at the other end of the market with mature trailing edge processes, where we're in a very, very different world and a very, very different part of the cycle, where we have way too much capacity already, and there is a lot more coming on stream, mostly in China, but plenty of other places too. I think, you know,
the market for cheap semiconductors is going to get much, much worse in the next few years. And I think in that situation, it's probably good for consumers. It'll mean that we get a lot more electronics in our cars, even our cheaper lower end cars. Those gains will be shared by consumers. I don't think we're going to necessarily see that at the high end for a long time. We'll see.
Samsung is still out there. Maybe Samsung does something crazy. And then we have Intel, Samsung, and TSMC all get to advanced processes in parity in 2030. But that's the time frame. That's the earliest I think we can get to real over capacity and advanced processes. You need Intel and Samsung to speed up things.
Ben Bajarin (36:29.276)
Yep, agreed. Well, interesting. So we, uh,
More to the story as it evolves. To your point about Nvidia, I resonate because they're about to blow the doors off revenue year over year, and they could have made more money if they could have made more chips. Do you know what I mean? It's one of those things where you're gonna look back, if my point is right, and we slow down and everybody comes back to Earth where you're just like, man, I could have crushed it.
Jay Goldberg (36:56.782)
Mm -hmm.
Ben Bajarin (37:06.37)
But she did crush it, but I could have crushed it more if we could have made more chips. It's the tragedy that was we could have done so much better. You did really good, but that's the tale of a lot of people. They could make more money if they could make more product. And it is that it is.
Jay Goldberg (37:25.806)
Yeah. You know, maybe we're all better off listening to my good friend Jerry Maguire, right? Fewer clients, less money. That's what we all need.
Ben Bajarin (37:36.476)
fewer clients, less money, higher quality engagement services, et cetera. All right, well, we got a big couple of weeks coming up in February. We got NVIDIA's earnings, we got Intel's IFS Day, or we're gonna have to talk about all of that. So stay tuned for, that's right, and Mobile World Congress, end of the month. So we've got lots to talk about. We're gearing up for some good guests.
Jay Goldberg (37:53.006)
And we got Mobile World Congress. And we got Mobile World Congress coming up at the end of the month.
Ben Bajarin (38:05.052)
If anybody has anybody that they're like, hey, you guys should have a guest, hit us up, let us know, give us ideas. We got some good executives lined up as well. So until then, thanks for listening to the circuit. We will catch you next time.
Jay Goldberg (38:17.55)
Thank you everybody for listening. Tell your friends. And if you want to meet up at MWC, hit me up. You can find me at Jay Goldberg on Twitter. My DMs are open. Thank you.