Subscribe
Share
Share
Embed
A podcast about the business and market of semiconductors
Ben Bajarin (00:01.243)
Hello listeners, welcome to another episode of The Circuit. I am Ben Beharon.
Jay Goldberg (00:08.078)
Greetings, internet. I am Jay Goldberg and glad to be here.
Ben Bajarin (00:12.379)
So this week was super interesting. We had teased it out last week, but on the same day, two events happened. One, which was, I don't know, Jay, many people were saying like the earnings shot heard around the world, the state of our economy hinging on Nvidia's earnings, but you kind of have, which I guess I'll say it this way, right? On the same day,
We had who is the, not only the, the most unprecedented valued semiconductor company in history, in terms of value, we have the Bellwether, the trendsetter, the king of semiconductors in Nvidia on the same day that the prior most valued semiconductor company, the prior king leader and dominant company in Intel had a, a foundry event. So there's the, uh,
the two events, the challenger and the current champion, which was a role reversed a decade or so ago. So yeah, so let's dive into it. I was at IFS Connect, which is now, that was the name of the event. IFS was Intel Foundry Services. It is now Intel Foundry. That is the name.
I'll lob out a couple takeaways and then I'll let you ramble. So the first is we have always asked this question, right? You have asked this question and I've tried to say, well, they're serious. And I don't know how much you believed me, but there was the, are they serious about Foundry this time? So I've been optimistic that they are from conversations I've had with management. I think a very, very clear,
Yes, based on several things that they did yesterday, a gigantic ecosystem surrounding around them. Rene Haas was there, our friend from ARM talking about their collaborations with the ARM ecosystem. Basically the deep collaboration that they've done so that Intel can make any ARM chip that has already made it TSMC there. You had Synopsys, you had Cadence, you had Ansys, you had a whole host of chip design companies, EDA tools, say they're ready for
Ben Bajarin (02:41.531)
18 a right. And this is really the focus of foundry will be people making on 18. Now that doesn't preclude them from 20 a Intel three, Intel four, Intel seven, et cetera. There will be ways they can work with companies that want to do those somewhat mid modern notes, but, but really 18 a is it because, um, and in, in case anybody's unclear with this naming scheme, let me just,
give you the analogy that makes sense. So, uh, 20 a is roughly two nanometer. So 2 .2, 0, 2 .02 nanometer. 18 a is 1 .8 nanometer. If we're again using common terminologies. And then they announced the subsequent suppress, uh, successor to 18 a, which is 14 a 1 .4 nanometer. So the number gives you essentially enter a decimal point where it's at. So most people.
are looking for our 1 .8. That's what they're trying to build on with 18A, which would again suggest that's capacity competitive to TSMC 2 nanometer. I think that's how, how, how I would look at that. So, so they've brought the tools they've, they talked like a foundry, which I think again, was kind of an important takeaway, right? They said all the right things that you would say if we were listening to a TSMC, Samsung and or global foundries event in terms of product, product.
roadmap, um, the, all the tools and ecosystems, what's available to customers in terms of tape outs, testing, et cetera, et cetera, et cetera, et cetera. So I think that's why they did this event. It was the foundry coming out party and it was to sound like a foundry. So everybody knows that undeniably they are, they are serious about this. So that was kind of a first, a first step. I think that's good. You know, again, we have said as many.
Oftentimes, this is still a long journey. We want to see what happens on 18 and 25 a, but, but that coming out party, I think again, they, they feel very set up their menu is there. Any customer that's made with TSMC or Samsung has everything they need to do this. They just need to keep winning those customers. They announced Microsoft as a customer. They announced an up, up of.
Ben Bajarin (05:03.419)
I don't remember the amount, but now $15 billion of, of lifetime value, um, dollar value to, uh, to the foundry. We don't know what that means in annually. That's lifetime. It's up slightly. So again, they are getting those customers, but bottom line, big, big picture, the key top line point, they sound like a foundry now it's operating like a foundry. It seems to be very unbundled from Intel product.
Um, well, again, we'll see proof in a pudding, but it looks like they address some very hard questions that we asked behind the scenes around that. Um, so I take that as a positive first step. If this was truly the day that Intel Foundry was born, I think they've got the right foundation to move forward from here.
Jay Goldberg (05:53.582)
Okay, okay, so I did not attend in person. I listened in, and so I missed some of the body language.
But I would say that I accept that it's a good first step, coming out debutante moment. But I walked away a little bit underwhelmed. And I'm not just being picky here. Let me walk through my things that sort of left me unexcited. First off, they only announced Microsoft as a customer.
It's one customer and that's good. But like for months, we've all been debating like who's going to be all the speculation, who their first customer is going to be, who, you know, where it's going to be. And they, they announced Microsoft, but it wasn't like a big fanfare. It wasn't like, Hey, this is the first customer. I mean, it's true. That's, you know, they should have had like confetti launched from the stage and balloons falling from the rafters. Welcome. It was just like, Hey, Microsoft's here. Like, and yeah, Microsoft execs got up and said nice things, but it wasn't like, it didn't feel exciting.
And so I kind of wondered about that. Like nobody seemed that excited about it, including Intel. Like this is your first customer. Like take their first dollar, you know, pin frame that first dollar they send you, you know, hang that up behind the cash register. There's still rumors that MediaTek and Amazon are out there in the works. Like, you know, the MediaTek people were there, said nice things about it. So yeah, there are other customers out there that they can't announce, right. But like,
There just wasn't much on the customer side.
Ben Bajarin (07:35.163)
Yeah.
Jay Goldberg (07:36.238)
A second concern I had was they didn't announce CHIPS Act funding. And I get that that is completely beyond their control. They're dealing with the federal government. It's a lot of money. But again, that was missing. And third, they didn't give out the financials. And so last week we did an episode looking at their financials. And we knew at that point that they weren't going to be releasing financials today, this week.
Ben Bajarin (07:41.435)
Yeah.
Ben Bajarin (07:56.537)
Hmm.
Jay Goldberg (08:05.262)
But all of that, it's so there are three big missing things. And it left me wondering why they did the event now. Why not have it two months from now? If you know the government is going to be slow, so get everything signed and finalized with the government. And then ask the government, you don't put out the press release until two weeks from now or whenever, like in the future. So you're going to have that. Same thing with customers. Like, why don't you have any more customers? Talk about the customers. That's what matters. Wait.
months and see who else you can convince to be on stage with you. Having three CEOs on stage saying, hey, we're Intel customers, would have been a huge deal. And third, put out the numbers. Let us see them. Because the numbers are going to be tough. We all know that. Let us mix that in with all the good news. And so, yes, at one level, I'm just kind of quibbling here. I don't want to be nitpicking. But it does worry me a little bit that this is the old Intel.
Ben Bajarin (08:37.103)
Yeah.
Jay Goldberg (09:02.414)
This is Intel relying on its old bad habits where they assume everyone's going to care because it's Intel and they're saying something, so everyone's going to pay attention. And they can't take that for granted, right? Not just Intel as a company, but especially Intel Foundry. And so the mismatch between whatever was going on at the executive level and the actual mechanics of a marketing event were absent in my mind. It could have just been much more...
coordinated and I think they would have carried a big punch and it makes me wonder like How how coherent how concerted their effort is going to be to really get this thing up and running? so Those are my big things right now I'll agree that like they they did have I mean it was the ecosystem side was pretty important They brought come talking about ASIC services. They had arms synopsis all those big important Partners lined up and I don't want to underplay that that's very important, but that's that's kind of like
Ben Bajarin (09:41.795)
Yeah.
Jay Goldberg (10:01.71)
you know, your mechanic, your car wash and your dealership teaming together to talk about what great things like you still need the customer, like who's going to buy the car. So under, I was underwhelmed.
Ben Bajarin (10:11.931)
Yeah. Yep.
Yeah. Yeah. I'll be really curious to see. I mean, this is probably first of many. You know, my guess is they'll, they'll do a founder your vent every 12 ish, 18 months to really talk about momentum. Um, you know, we know the finances are coming. I think we just got to get over this hump right, right now. Like you said, the customer one, I think it's going to be really tricky. Um, you know, we, I had some conversations with management, you know, behind the scenes and I, and I knew this because TSMC doesn't release customers either, but.
You know who they have because the supply chain is leaky. So essentially, you know, anybody who has priority status at TSMC that's talking to Intel doesn't want to give that away because they don't want to give up priority status to the degree that anybody that has priority statuses, which is a question, but, but regardless, um, the good thing is over the course of the next few years, whether Intel announces customers or not. And I do hope that they have.
customers on stage. I mean, when we saw when Renee came on, he sort of did the whole like everybody was kind of like, Oh, you know, arms, arms on stage with, with with with Intel, like, that's just, you know, that's crazy. Who would have thought, right? And and they were like, it's kind of like this moment. And he and Renee used the whole, like going back to when Steve Jobs, you know, said, hey, or brought brought iTunes to Windows. And he was like, hell's frozen over was like the marketing term.
But the real though to me, hell's frozen over moment would be when Jensen's on stage with them at a Foundry event. Everybody who knows his history with Intel, the weight of that baggage that goes with Jensen and Intel, that's that moment where you're like, oh, holy crap. That's a very different, if it ever happened, but regardless. Yeah, the supply chain, I guarantee will leak those customers whether they did it or not.
Jay Goldberg (11:44.558)
Hehehehe
Ben Bajarin (12:06.395)
So we'll know whether anybody says, but I agree with you. I do hope they can have customers on stage saying, this is why we're excited about Intel Process Technology, et cetera, et cetera, et cetera.
Jay Goldberg (12:18.926)
Yeah, I -I -they need that and they just need a - like -like I want -like I -I - Go ahead. Go ahead.
Ben Bajarin (12:21.019)
Okay, point two. Let's go into the other point.
Ben Bajarin (12:29.115)
So, yeah, so say let's, let me jump into point two and then have you, have you give me your, what you're not going to tell me, but I know is prepared remark to this statement. Okay. So I think it's become more clear to me and I toured Intel's advanced packaging facility in the end of last year or fall of last year. But I think now I firmly kind of understand that they are, I don't want to say betting the foundry farm, but.
to a huge degree, they are leaning very heavily on the bias that the next 10 years, the future of chip design will be complex systems of chips in a package. So not systems on a chip, so not SOCs, but systems of chips in a package. So chiplets with advanced packaging, yielding very, very large packages, not dies, but packages.
on and challenging the, uh, can traditional constraints of design supply chain substrates, et cetera. So, so they are, that's why they, they use this language. We are positioning ourselves as a systems foundry for the AI era. So again, their bias is. As we move to AI, you're going to need a completely different way to make these chips. You're going to architect them differently. You're going to build them differently. And so they're, they're betting that, that, that that's the case. And I do know, like I, they are.
well ahead in advanced packaging and advanced chiplet infrastructures than TSMC. So they have that highly differentiated advantage for any customers who want to use them. Again, you got to make a pretty big jump to move from TSMC to Intel, although they're the only packaging, advanced packaging facility that can dual source, meaning you could make your wafers at TSMC, bring them to Intel and they will patch them together as long as you design with EMIB.
in mind and some of their, their bonding techniques. But all that aside, my point is that's their bias. Now they could be wrong. I don't know. I mean, we don't know where the future goes. Is the future monolithic is the future, uh, disaggregated design and chiplets. I mean, I tend to think there's a lot of logic to that, but, and then here's the point I'll make. And then you go, the analogy that then I think comes becomes interesting is that TSMC won the mobile era.
Ben Bajarin (14:54.203)
They won monolithic. They were the default choice for you to go and design monolithic chips for on their way first. Intel, I think is trying to say, we are trying to set ourselves up for this next era by being that choice. If you want to go and make these advanced systems on a chip, et cetera, et cetera, et cetera, we will make those investments at a rate far more than TSMC has. They'll go through those hard learnings. They'll, they'll.
do what they need to do in advanced packaging, et cetera. TSMC can shift CapEx, but again, very highly still focused on monolithic in their volume. So that's the analogy. Could Intel be in this next 10 year era of advanced AI different chips, the right play the same way that TSMC was in mobile? So that's the thesis, that's their bias, et cetera, et cetera, et cetera, et cetera. And...
wherever TSMC falls into that as they again continue to do advanced packaging is the question, but I think it's, I think it's a worthy, a worthy bias, I guess, analogy. They got to make a bet somewhere. That's where they've made it.
Jay Goldberg (16:08.59)
I didn't want to go to the prom. It's not that I couldn't get a date. It's that I didn't want to go. That's my response to that. We called this a debutante coming out party. They didn't want to go to the last one. It wasn't that they weren't ready. They couldn't get their act together. They just didn't want to go. So I don't quite accept the analogy.
Ben Bajarin (16:13.819)
Sir, sir, sir.
Jay Goldberg (16:36.078)
I get what they're making and I do agree that packaging is very important. And sure, they're ahead of TSMC on packaging. We can debate how far ahead and what that means. But TSMC for starters is no slouch in packaging. The biggest problem in the world right now is Nvidia can't get enough of their packaging. But it's good. So that's one issue. Two, like Intel didn't...
lose, wasn't immobile because they weren't trying, they lost it because they just couldn't execute. And what is supposed to give me confidence that this time around we'll get it right?
Now, in fairness, Pat is a very different CEO than the ones they had in the past. And they've certainly shown a lot of signs of improving their technical chops. But I'm not sure that it's very early to make that analogy. And I don't think it's quite saying what they think it's saying. I'll tell you, my biggest concern overall, and this harkens back to what I was saying a few minutes ago, is about their misallocated marketing for the event.
My big concern is that you have all the execs, you Pat and Stu and a bunch of other people from the Foundry side, the top level smart people doing the right things, saying the right things. The problem with Intel for years though has been the middle, the culture of the whole company. And I get that the execs want to make this transition and packaging is going to be really important and they're going to sell that hard and push that hard and systems of chips.
That's the big thing. Can they actually deliver that on the ground through whatever number 10 layers of Intel culture and management that is resistant to change and has been resistant to change? Pat's not a good job of shifting the organization, but it's a big shift. And I am just, from what I've seen so far, I'm not convinced that we're anywhere close to that process, which makes me.
Jay Goldberg (18:51.18)
you know, a little hesitant to get fully on board with their new mantra.
Ben Bajarin (18:56.795)
Yeah. So I think there's, there's two things. I think the reality is I firmly believe if Qualcomm or MediaTek came to them and said, we want wafer scale production of monolithic designs on 18A, they could do that. The reality is they're not going to do that because they've got what they need in TSMC. So regardless of whether Intel can, which I believe they can, they say they can and commentary from said, go Qualcomm's of the world and MediaTek's.
would suggest that sure, that's an option. I just don't think they're going to. I don't think those wafer scale customers are going to leave TSMC with this point in mind, as long as A, they're still a preferred TSMC customer. And did I say A or one? Either way, B, they design monolithic. Intel can do it a hundred percent, but so can TSMC. That's not the differentiator in my opinion. The differentiator is these advanced designs.
these complex designs. So whoever, whenever that is that we have exhausted the capabilities of monolithic and that will happen at some point, Apple, et cetera, everybody will do that. TSMC will have to be in a position to service those customers in the same way that Intel is positioning themselves. That will happen at some point, but there's a lot of people who want to do that today. Obviously Intel as that client with Granite Rapids and what's coming.
crazy amounts of sophisticated technology going into those things. Ridiculous compute. At all add a point that I appreciated that came out of this because I have pushed back on Intel saying Moore's Law is not dead for the many years that they've said that, because Pat came back and said that day one. My beef with it has always been that you are not acknowledging the economic part of Moore's Law, that it should become economically more
viable to do this, right? Or at least as an initial investment comes, it should subsidize. They went out of their way to make a point. I have to point this out because it's funny, right? It was for the last, when Pat came back, it was Moore's Law is not dead. Then on Wednesday, it was the economics of Moore's Law are back. And I was like, well, why weren't they back four years ago? Like this was the situation, but they're back now. Okay.
Ben Bajarin (21:22.907)
Let's just unwind that for a second because that'll happen. I do believe that this part is true, that chiplets do bring back an economic advantage back to manufacturing. It's at least efficient, if not beneficial, if you look at a whole host of the deep tech analysts that have been doing transistor costs, density designs, et cetera. People believe that. So there's an economic part of this that comes back that I don't think we fully wrestled with.
as a potential option for Intel to become attractive from a cost standpoint, as well as more of these things mature. So again, we won't know that for a couple of years, but they did go out of their way in some of our backroom conversations to show how the economics play out toward bringing back Moore's laws economic part, but it's not with monolithic. I think that's the point. Monolithic designs are not economically friendly.
Chippets become much more so as the as the disaggregated designs. You're only using, you know, you're using many wafers for many different chips across the system. So I appreciated that. Cause like I said, that's been my sticking point. When you say Moore's law is not dead, you're not addressing the economic point. And now all of a sudden the economics of Moore's law are back because of it. So, so, so I appreciate that, that element of.
wherever we go from here. But anyway, point is, I don't think Intel's differentiated monolithic. That's my point. That's why to your answer, it doesn't make as much sense there. So fine, they're just not going to get those customers. Their hope is they get everybody once they, or they get a lot of business once they start thinking about all these different types of designs that TSMC is not actually currently investing in doing at scale like Intel is.
Jay Goldberg (23:17.39)
You had me there almost to the end. But because I was basically I was going to say we can stop recording. I agree with everything you said. I know I think I think one it is it is a very smart strategy on Intel's part to play up technical innovation. That's that's how they're going to survive this. If they can really deliver on all those things, packaging, backside power, get all around, whatever all that stuff. Go down the list.
Ben Bajarin (23:27.259)
Yeah.
Jay Goldberg (23:44.046)
That's how they're going to get pushed through this. They're going to have to lead with that. That's the real competitive advantage. And I think they've held onto that talent. They still have immense capabilities. So that's the right strategy. I think I only quibble with or I question TSMC's response to this because TSMC is not a lightweight when it comes to this stuff. And it is an interesting point to say that they are fairly committed to all.
Ben Bajarin (24:04.091)
I think it's fair, that's fair.
Jay Goldberg (24:12.974)
not old technologies, but like the monolithic, what you call the monolithic path, right? But I don't think that limits their ability to innovate as well. They're working a lot of these things too. It's very hard, but know, TSMC has at this point, a lot of credibility when it comes to technical innovation. They come about it very differently. It's not going to be as dramatic as Intel is going, you know, five nodes in four years. It's a huge jump, which is very exciting.
Ben Bajarin (24:15.387)
They have to be. Yep.
Ben Bajarin (24:34.555)
Yep. Yep.
Jay Goldberg (24:41.816)
TSMC doesn't have to move at that pace. They just have to increment steadily. But you know, it's hard. So maybe it takes them a little bit longer. I think it will be very interesting to see how this plays out sort of in the last couple of years of this decade, 27, 28, when these things are actually production worthy and in the market.
Ben Bajarin (24:45.113)
Agreed.
Agreed.
Ben Bajarin (25:03.579)
Yeah. Yep.
Jay Goldberg (25:05.614)
And I think then that gets into sort of customer dynamics, which are interesting, right? Because you sort of touched on this. None of the big, fabulous companies are going to be in a hurry to raise their hand and say, we're an Intel customer because that risks jeopardizing their relationship with TSMC, right? AMD and Nvidia in particular, who are almost entirely dependent on TSMC have very little to gain by saying, oh, we're going to go with Intel.
because they already get the best pricing from, well, the second best pricing after Apple, right? And so who's gonna like really step up? And, you know, there's a little bit of irony in this because for years, Intel treated the channel this way, saying, oh, you wanna look at an ARM server CPU? Well, we're gonna cut your allocation for the next part. So channel control is a real thing, Intel is aware of it. There's a great irony in that now they're facing this from the other side when they're trying to break into Foundry.
Ben Bajarin (25:38.523)
Absolutely. Absolutely.
Jay Goldberg (26:03.854)
But there's signs of movement, right? The fact that MediaTek is anywhere near Intel on this is, I think, really telling because MediaTek and TSMC have deep, deep bonds that go back 40 years, 30 years. Like those two companies are... Right? Right. And so I think though, it's interesting because MediaTek, it used to be... Well, it still is. It's a top 10, top 5 customer for...
Ben Bajarin (26:19.003)
And they're a top priority. Yeah, they're still a top priority. Yep.
Jay Goldberg (26:33.294)
TSMC, but they're way behind the top three. There's a big gap. And that's because that side of the market has moved much faster than MediaTek's side of the market. And so for MediaTek to be sitting here saying, hey, wait a second, we're a little bit unhappy with TSMC, maybe not unhappy, but at least unhappy enough to signal that they want to look at IFS is telling. And so how those customers and those channel relationships play out,
Ben Bajarin (26:34.875)
Five.
Ben Bajarin (26:38.651)
Yeah. Yep.
Ben Bajarin (26:55.227)
Yeah. Yep. Yep.
Jay Goldberg (27:01.39)
is going to be an important dynamic, but none of that will happen until we actually get production for probably 18a underway for external customers, which is what, two years away, Nastel? So.
Ben Bajarin (27:13.627)
Yes.
Yeah. 25, 26. Yeah.
Jay Goldberg (27:17.838)
Right. And so my, so my guess is, my guess is that none of the major wafer scale customers sign up flagship leading edge parts, leading edge, not the right word, sort of major, major parts. None of those go to IFS for 18A. I think 14A is starting to sound like where the real fight will be. Right. And if you see, like, if you see like a Qualcomm Snapdragon 800 series or whatever they call it then.
Ben Bajarin (27:45.179)
Yeah. Yeah.
Jay Goldberg (27:45.806)
in 2027 going to IFS, that's what they should be targeting, right? A big, big volume platform. And we just won't know that for, I mean, we won't even be able to, nobody will even be able to assess it for another year or two. So we got a ways to go.
Ben Bajarin (27:50.779)
Yep. Yep.
Ben Bajarin (28:02.139)
Yeah. Yeah. So let me, let me lob one other thought thought point out. And this is going to anger some people who listen to this that, um, are, are, are, are very, very, uh, in love with Apple semiconductors. So the day that this changes for TSMC is the day that Apple moves away from monolithic. Apple is the single greatest reason why they are maintaining
leading edge adoption in terms of wafer commitments. Because right, if you are a foundry, you need a customer who can guarantee you scale. That's Apple more so than anybody else. I didn't that's Apple. The question I have. So, so, okay. So let me back up. So the, the way to think about this is Apple funds, TSMCs, leading edge and any new innovations, right? They were, I think one of the first to adopt co -hosts for ultra, um,
uh, ultra bonding of, of two chips. Apple does not make chiplets, but you could argue maybe that is maybe that as it regardless, Apple would not say they make chiplets. Um, but Apple is, is to TSMC what Intel is to now Intel Foundry. So Intel is again, betting the farm not on monolithic. They are betting the farm on, on chiplets. So is there a risk that their commitment to Apple, if, if we believe.
Over the course of the next 10 years, we move away from disaggregated design from monolithic to chiplets. I don't know about mobile, but whatever. Is there commitment and deep, deep relationship to Apple in any way a hindrance of them adopting and advancing all of these areas that Intel is putting a lot more equity into?
Jay Goldberg (29:59.246)
That's a good argument. I think the flip side of that, though, is when does Apple start to rethink its architecture?
Ben Bajarin (30:10.189)
Agreed. I don't know, but.
Jay Goldberg (30:10.478)
And then say, oh yeah, TSMC, for the next A series, we need you to do something funky.
Ben Bajarin (30:19.803)
That's, and that's the day this changes. But, but, but what if Apple's stubborn? What if they're like, you know what, we don't need to, because we don't have to abide by the same economics as everybody else because we're vertical. Like you could, you could argue that Apple could slow this down and that TSMC will be in a pickle given how much they meant. I mean, I'm completely speculating. I would love to see Apple adopt chiplets tomorrow. I think it would be fascinating and exciting and the kind of crap that they could do in things like vision pro high -end desktops and mobile devices would be amazing. But.
They don't even seem to be on that track. Like they seem to firmly be comfortable with monolithic over the long haul. So I agree with you. That's the, that's why I said the day this changes for TSMC and all of a sudden we're like, look at the, look at the wind for advanced packaging is the day that Apple moves that forward. I'm just slightly concerned, I guess that, that that does become a bit of a burden if, if the market moves that way, their commitment to, to Apple.
Anyway, I'm just throwing that out there as a lobbed out thought experiment.
Jay Goldberg (31:24.736)
I think it's a really good point. But I think the where do we test it? So on the PC side, I think we I think there's clear pressure now on on PC CPU makers to move towards chiplets. Right. And is there already Intel's coming soon, which makes me wonder when does the M series from Mac have to go down that path? While while you were talking, I looked up a
a die shot of the A series application processor for mobile. And I don't think it's something that lends itself readily towards chiplets. I could be wrong. Maybe I'm missing something here. But mostly what's, there's a lot of heavy digital logic on there, which really needs to be big advantage being monolithic. As opposed to a CPU, which has a lot more other things, IO,
Ben Bajarin (32:08.301)
Agreed. No, I would agree with you.
Ben Bajarin (32:14.715)
Yep. Yep. Yep.
Jay Goldberg (32:23.566)
memory stuff, where the different pieces can go on different processes more readily. So the A -series PCs, though, yeah, we're saying the same thing. PCs are going to push it in that direction. And let me throw out another one for you, because another major CPU customer for TSMC is going to be Intel. So does
Ben Bajarin (32:27.963)
Yep. Yep.
PCs though, I think you could argue, yeah.
Ben Bajarin (32:48.923)
Yes, is already, is already, yes.
Jay Goldberg (32:51.246)
So it would be interesting to see if Intel, as it pushes more in the chiplets, how heavily it's going to lean on TSMC to do that, because that risks sort of funding their biggest competitor. I...
Ben Bajarin (33:01.883)
Yes, yes. I love this observation. I think everybody, we should track this. It's such a good observation.
Jay Goldberg (33:09.806)
That's right. I don't know all the Intel roadmap lakes. I get them all confused. But on that roadmap, I'm pretty sure they have at least the next two, which are going out TSMC, are chiplet -based.
And so I think that is going to help push TSMC in the right direction because maybe Apple doesn't move, but then some combination of AMD, Intel and Nvidia is going to push TSMC to get down this path. And absolutely, I was going to say absolutely Apple would make it happen much faster, but I think you add up the PC side, they get there anyway.
Ben Bajarin (33:40.123)
Yes. Yeah. So the caveat.
Ben Bajarin (33:48.771)
Agreed. 100%.
100%. I think the caveat to the Intel point is that they will still do the packaging at Intel. They're just having the, the, the, the wafers made by TSMC. And that's, again, that's a whole can of worms that we have to address. Like, do people want to do that? Like, do they want to, would they prefer if TSMC had all the same advanced packaging, uh, uh, tools, menus, um, options that, that, that Intel has, which I do believe are is.
highly differentiated towards Intel's favor. But let's just assume that DSMC has those. Why would you leave? Like you would prefer to single source at your one foundry, right? So there's a real competitive analysis to this that's gonna get done that I'm excited about. But you're right, right? I think AMD and Nvidia in particular can move them in the right direction faster. But when you look at, and I've read all the different supply chain notes relative to CoWOS and where they're going, when you look at how much they're investing in that,
compared to Intel, there will be a gap. I don't know how long that gap is and if the timing overlaps at all, but there's going to be a gap. And so that's why, that's the only reason I make this point that like, if we're trying to create the absolute bull case for Intel Foundry, it's that this bet pays off. They're well positioned because they have outspent TSMC and capabilities for advanced packaging and advanced systems manufacturing. And those timelines align correctly.
When said companies are like, you don't have what I need and Intel does. Now again, I don't know if that happens. I'm just presenting the ultimate bull scenario for, for Foundry.
Jay Goldberg (35:34.446)
Yeah, that makes sense. That's the bull case.
Ben Bajarin (35:35.931)
And we can end Foundry on that. Let's move to NVIDIA. All right. This was a very clean quarter, to be honest with you. So much weird stuff happened. One of the things I think is fascinating about NVIDIA, and I understand why. I don't like it, but I understand why. But there are way more non -institutional investors. So retail.
Jay Goldberg (35:39.758)
Yep. Yep.
Ben Bajarin (36:05.371)
and and traders in Nvidia than any other semiconductor stock I've ever seen in history. It's super weird. Because of that, I had no idea what would happen. Like, what if they didn't beat enough? What if they didn't raise enough? Would the stock drop? And I think you saw right there was like some option calls that I think someone was saying was going to go between 150, maybe 200.
a billion in swings of, uh, of dollar amount based on these calls. It's like people had no idea yet. They had a moderate. I mean, again, I'm not going to say $2 billion is, is not a great beat, but you know, it's a moderate beat and a moderate raise again, two ish billion dollars. And at the end of the day, the stock was up like 8%, 10%. It jumped to the next day, but it felt clean. There wasn't an overreaction in any way. So I don't know what to make of that. What, what did you.
make of the, again, clean quarter, nothing shocking, really fairly moderate reaction, like it didn't break the world one way or the other.
Jay Goldberg (37:17.55)
I remember in the distant past, four years ago, when there were maybe, before times, when maybe there were ancient mists of yore, when there were maybe a thousand people who cared about an Nvidia quarter. And now suddenly this quarter, the fate of Western civilization rested on this one.
Ben Bajarin (37:22.555)
the before times.
Ben Bajarin (37:29.147)
Yes, yes. Agreed. Maybe a thousand people.
Ben Bajarin (37:39.131)
Our whole economy, the United States' fate hung on this earnings call. Ugh.
Jay Goldberg (37:47.118)
I mean, my conclusion published in my post was like, Nvidia, the company is doing great. Right. They're doing great. Nvidia, the stock scares me a lot. And we dealt with this, what was it? Like nine months ago, and there are all those crazy conspiracy theories about what, you know, and I was on that weird Twitter call and people were just like, it was just full on Looney Tunes. Those people are still there.
Ben Bajarin (37:55.003)
Yep, great. Yep.
Ben Bajarin (38:01.453)
It's wild. It's wild.
Ben Bajarin (38:09.051)
Oh God.
Oof.
Jay Goldberg (38:16.398)
there's an immense amount of retail interest in the stock. And I get it. I mean, like the revenue earnings increase in Nvidia in the last 12 months is something no one's ever seen before. And we'll probably not see it again in our careers. I mean, just amazing, amazing shift. And so it's going to attract a lot of interest. I worry, though, that a huge amount of that interest does not have any idea what's going on. And I know that...
Ben Bajarin (38:17.115)
Yeah.
Ben Bajarin (38:25.563)
Absolutely.
Ben Bajarin (38:31.963)
Yeah. Yep.
Ben Bajarin (38:43.259)
Agreed.
Jay Goldberg (38:44.27)
I know that the average retail investor is severely outclassed in understanding what's going to happen with Nvidia stock. Maybe if you're a retail investor, you have as good a sense of what GameStop is doing, because you can go to a bunch of malls and look yourself. Nobody knows what's going on with Nvidia and their supply chain. There are hedge funds who are paying huge amounts of money to track it all down and figure it all out. Retail investors cannot.
Ben Bajarin (38:52.251)
Absolutely.
Jay Goldberg (39:14.062)
cannot compete with that. And it's a dangerous stock for retail investors. But like I said, to some degree, that doesn't matter. The company's doing well. So I don't want to fault them. It's not Nvidia's fault. The only thing I would add is that Nvidia has always had a hard time predicting its quarters. I was talking to a friend of mine who's in the markets, and he looked at the history of Nvidia over the last 25 years as a public company.
Ben Bajarin (39:22.395)
There, yeah. Yep.
Jay Goldberg (39:44.11)
And he says there's been three times in that history when the stock has drawn down over 50 % in a quarter. Right? I mean, massive, this is massive swings, big company. I don't think it's going to fall 50 % on its next earnings call, but I like it's just, it's priced so tightly. Like if, you know, if, if Jensen like hiccups on his next earnings call or trips over a word or two and stock could fall 10 % just because everything is so, so raw and tense around it.
Ben Bajarin (40:01.371)
Yeah. Yeah.
Ben Bajarin (40:10.363)
Yep. Yep. Yep. So, so a financial point, and then I want to make a market point. Um, so the, the reality is, you know, again, I think it's reasonable to assume. And I had some talks with some of the folks on the, on the, on the sell side from, um, anyway, on the sell side, who I looked, we, I looked at their models and.
know, they're projecting pretty solid, you know, double digit year over year revenue growth through the next couple of years. I think it's reasonable to assume they are going to grow. Now, how much are they going to grow? That's the question, right? They're going to grow. The issue I have is exactly what you said is that I'm again concerned that there are outsized expectations in volume from those group that's in this that can create those swings so that if they come back and one quarter's only up,
8%. I don't know if it is. I'm making this up. I'm just saying versus the 15 % that, that they're at for whatever reason that people are just going to freak out, right? Because the growth is not the rate that they're assuming. And I say that because my biggest worry about all of this is can Nvidia make enough products to even meet those revenue guides and expectations when we're up against capacity constraints. Now they did say, so here's the two market points. They did say,
Supply is easing, great. However, GTC is coming in May and I've heard a product may get announced, which is showing that their product cadence has shrunk. So if product next, whatever it is, is coming and it has drastic performance upgrades, there's gonna be lines out the door for that one as well that they then again can't go and make. And so.
You have this weird swing of product cadence, performance improvements, demand in that supply chain. Even if H -100s get better, people are going to be like, well, I want the H -200 or the B, you know, whatever it's going to be called. And, and that's the, the volume driver that again, they can't go and make. And did they over make H -100s? Are people still okay? Probably supply is easing. Great. But those two things are my, my broader concern. And then one other market point.
Ben Bajarin (42:31.963)
They made a point that I think is important where they said 40 % of revenues were for inference. Now we have had this debate as with others, there's a huge conversation on is our GPU is the best place for inference. We know that the best place for training Nvidia will dominate the training era for AI a hundred percent. Are they the best place for inference? Meaning you just want to inference these things. And so it's, it seems that people are buying these things and or using them for inference, which
I don't know if that's sustainable, but at least for the time being, puts a little bit of confidence, I guess, that yes, the market will also adopt and use these for training.
Jay Goldberg (43:18.158)
OK, that's a lot. And we're running up on time. But let me break that into three pieces. One, I agree. Their product cadence is, on the one hand, impressive. H100 last year, H200 is coming out shipping this quarter. Next month, everyone assumes they're going to launch B100, which is the next iteration of the H100. That comes out, gets announced in a month, and then ships early next year.
And then we already have the code name, the X or the R100, which is announced maybe late this year, early next year. I actually asked them this on the call, like, aren't you guys worried about Osborne affecting yourself? Like you announced a new product and everyone stops buying the old one. But that doesn't apply to them because their backlog for the H100 is still so big that by the time they're actually shipping the B100, they may not have satisfied all demand. Now, that's a contentious point. Everyone on the street is debating how long the lead times really are for H100. I don't know. I don't really care.
The point is, they're pretty confident that demand is strong enough that they can keep moving it forward and that there'll still be H100 customers even as B100 starts shipping. I won't get into their explanation for it, but there's some credibility there. But whatever, there's a little tension there. And we don't know exactly how well the B100 will do, so it has to be much better. I mean, that's the other downside is if B100 is only a little bit better than H100, then that whole mix.
Ben Bajarin (44:40.973)
Agreed. Right. Agreed. Yep.
Jay Goldberg (44:44.43)
But they've got a pretty good execution track record. Let's assume that it keeps going forward. Then it comes down to demand, which at this point, I think it's pretty solid. Again, for the company, for the stock, who knows? But for the company, I think demand is OK. The second point, though, is sort of around long -term expectations. And here, I think they have a little bit of an investor relations conundrum. And no shade on Simona or her team.
Like they're, they're a little bit stuck because I think one of the critical weaknesses over history with Nvidia is that Jensen really likes to talk big picture and full credit to him. Like he's, he's right about AI and he's been talking about it for years. So having that big vision is important. It's gotten them where they are. The, the, the wrinkle comes in the communication about it, because when you ask him a question, especially on a call, he likes to talk about those big picture things.
And that sort of infuses through what they have to talk about in their cons messaging. And so one of the things I picked up on is they're talking about the TAM, the total addressable market for data center, Silicon is going to double over the next five years from 250 billion to 500 billion. Right. I posted on this today. You can go through the math, but what they're basically saying is the market's going to double over the next five years. And they strongly imply that they're going to capture most of that.
Ben Bajarin (45:59.451)
Yeah.
Jay Goldberg (46:13.582)
And that's the big picture, which is, those are really big numbers, right? You sort of, you work out the math. I was very conservative and I got like, you get to some crazy big earnings and revenue numbers.
Ben Bajarin (46:20.827)
Yes, right.
Ben Bajarin (46:26.907)
It's a big, it's a big, it's a big assumption.
Jay Goldberg (46:30.382)
It's a big assumption. And so talking about it now is just ratcheting up those expectations. That's my concern. You have all these people with very heightened expectations, and they're just double the time and through kerosene on that. So there's a little, that is, like I said, an investor relations dilemma they're going to have to, that will at some point come back to bite them. But the business is good. The last point on inference.
What they said was 40 % of their data center revenue is going towards inference. My take is that it's just early days. Like the inference build out is, you know, by anybody's math, not just their doubling math, but by any math, like the inference is going to be a big, big market. We just don't know yet. We're probably in first inning of that inference build out. And I think they're seeing a lot more competition on that front.
Ben Bajarin (47:07.291)
Mm. Mm.
Jay Goldberg (47:24.174)
Certainly the big companies, the AMDs and Intel and Qualcomm are all talking of inference really heavily. There are also a lot of interesting startups. I think we probably have a future episode on this. We have companies like Etched and Grok with a Q. Grok came out with some numbers this week and everyone's debating. They look like they have a pretty good inference part for certain use cases. So there's a lot more competition there. So I don't think we know. I think there's a lot of...
Ben Bajarin (47:24.507)
Yes, we'll see yes, yes.
Jay Goldberg (47:55.214)
pros and cons towards GPUs as inference. I think Nvidia has a pretty strong story there. They have the software ties that they're going to lean very, very heavily on. And for a lot of customers, if you have Nvidia GPUs for your training, it's very easy to just use Nvidia GPUs for inference. And I think that I wouldn't, I mean, again, technically it may not be as performant as some specialized ASIC for that, but the software workload is a
Ben Bajarin (48:22.491)
Right.
Jay Goldberg (48:24.782)
pretty significant barrier to entry or competitive advantage towards Nvidia. So I...
I'm, I'm, just don't know yet. Right. And I think Nvidia has a good story here. They're not going to get a hundred percent of the inference market like they do with training, but like big swings in outcome for everybody. If they're 50 % share or 20 % share of inference, those are, those are radically different worlds we live in.
Ben Bajarin (48:43.291)
Absolutely.
Ben Bajarin (48:49.659)
Yeah. Yep.
Yeah, yeah. Agreed. All right, let's wrap there. We are up against our longest, almost longest episode ever. Hopefully everybody appreciates the rich conversation. More to discuss, obviously next week will be Mobile World Commingers happenings, learnings, discussions from the bars as Jay twists people's ears about the goodness. So with that, we appreciate everybody's time and we will talk to you next week.
Jay Goldberg (48:59.66)
Well...
Jay Goldberg (49:22.294)
Adiós, gracias, vaya con Dios, I'll see you all in Barcelona. Thank you.