Silicon Futures for May 2026 - Qualcomm hyperscale AI, Cerebras IPO, Huawei 1.4 nm

Silicon Futures is a neXt Curve reThink Podcast series on AI and semiconductor tech and the industry topics that matter.

This month, the AI narrative has rediscovered its mojo with hyperscalers continuing to double down on their CapExc on AI infrastructure, while realizing unrealized "other income" from their investments in AI startups and labs as the likes of Anthropic, OpenAI, and xAI prepare to go public. We didn't cover it in our discussion, but yes, Elon Musk lost his case against OpenAI. Too bad. 

Regardless, another action packed month in May 2026. 

In this episode, Leonard, Karl and Jim talk about some of the top headlines from May of 2026. 

➡️ Qualcomm's mystery hyperscaler customer?
➡️ The geopolitical risk of playing the AI game 
➡️ Huawei teases "1.4 nm" chip and novel architecture 
➡️ Cerebras finally IPOs! 
➡️ NVIDIA's Q1 FY27 results and new reporting structure 
➡️ Memory vendors are now $trillion market cap 
➡️ Will Apple use Intel Foundry for Apple Silicon?
➡️ The state of Intel Foundry business 
➡️ Alex Katouzian joins Intel to head up Client Computing Physical AI 
➡️ AMD proving they are more than AI accelerators 
➡️ Is the AI infrastructure boom driving IoT revival? 
➡️ AMD, Arm, NVIDIA, Intel, and the CPU boom
➡️ Dell's pivot toward warm-water cooling 
➡️ Dell leans into AI workstation! 
➡️ NVIDIA investments in Taiwanese AI startups 
➡️ "Other income" unrealized gains of AI investments 
➡️ All eyes on COMPUTEX 2026

Hit Leonard, Karl, and Jim up on LinkedIn and take part in their industry and tech insights.

Check out Jim and his research at Tirias Research at www.tiriasresearch.com.
Check out Karl and his research at Cambrian AI Research LLC at www.cambrian-ai.com. Check out Karl's Substack at: https://substack.com/@karlfreund429026

Please subscribe to our podcast which will be featured on the neXt Curve YouTube Channel. Check out the audio version on BuzzSprout or find us on your favorite Podcast platform.

Also, subscribe to the neXt Curve research portal at www.next-curve.com and our Substack (https://substack.com/@nextcurve) for the tech and industry insights that matter.

NOTE: The transcript is AI-generated and will contain errors.

DISCLAIMER: This podcast is for informational purposes only.

Transcript

Copy Transcript Copied

Karl Freund 0:06

Next curve

Leonard Lee 0:09

Welcome everyone to, uh, this episode of Next Curve's Rethink podcast, where we break down the latest tech and industry events and happenings in the world of semiconductors, and Carl's favorite topic, AI, AI, AI, and AI, into the insights that matter. I'm Leonard Lee, executive analyst at Next Curve, and I'm joined by the illustrious Carl Freund of Cambrian AI Research. And we also have agentically modified and augmented Jim McGregor of the famed and infamous, a- and, fantabulous, right? That's another good word to describe- Terius Research. Yes, Terius Research. Welcome, gentlemen.

Jim McGregor 0:57

Th- this is not the Jim McGregor you're looking for. No, just kidding.

Leonard Lee 1:01

Yeah, you're in a purple metaverse. In this, episode, of Silicon Futures, we're gonna be talking about the headlines for the month of May, which has been completely insane. And before we get started, remember to like, share, react, and comment on this episode. Also, subscribe here on YouTube and Buzzsprout to listen to us on your favorite podcast platform. Opinions and statement by my wonderful guests here are their own and don't reflect mine or those of Next Curve. We're doing this, for informational purposes only, to provide an open forum for discussion and debate on all things, AI and silicon. So with that, wow, g- guys, it's just an- I mean, this is, like, r- really nutty. It's getting out of-

Karl Freund 1:46

You say that every month.

Leonard Lee 1:47

I know.

Karl Freund 1:48

That's the new norm.

Leonard Lee 1:51

I'm gonna kick us off here with, um, this, announcement by Qualcomm. I'd love to get your reactions to this. There's this report, from Bloomberg, that mystery, uh, hyperscaler that, Qualcomm mentioned on their earnings call this month, is ByteDance.

Karl Freund 2:09

Mm-hmm.

Leonard Lee 2:10

looks like there is an alleged, uh, deal for, quote-unquote, "millions of ASICs", uh, to support, uh, AI requirements for their social platform, so this is largely inference, right? agentic AI. Uh, so what do you guys think? I- is this-

Karl Freund 2:27

I think it's fi- finally happened. I mean, Qualcomm's been in AI on the device for a long time. eight years in production, I think, maybe nine.

Leonard Lee 2:37

Oh, yeah.

Karl Freund 2:37

and they dabbled in the data center. It got some ODM traction with Dell and HP, with the Qualcomm AI 100. But it was really underpowered for today's AI. Mm-hmm. it was designed for yesterday's AI. Now Qualcomm has what we all believe they will announce probably next week at Computex the next generation of that, which is rack scale. Mm-hmm. if you wanna start fresh with new AI models for agentic AI and a lot of CPU power, that seems to be where they're heading. Mm-hmm. And ByteDance fits

Jim McGregor 3:17

Well, and they showed off their first rack scale solution at, Mobile World Congress. it's a huge rack. it's a stan- it is a standard Really tall, right but it's really tall. It's standard than, what I would consider a normal rack bait is. It is an industry standard still. So no, it, it's gonna be interesting. first off, they, they've kinda hinted that they're talking to more than one hyperscaler or w- more than one- Yeah large customer. So, uh, I would expect that this is probably the first in many announcements, and they planned on making their first announcement about their first customer at, their AI inve- their investor day, which is later in June. So this is kind of, uh, uh, leaking, the first announcement. But I'm also a little concerned because, this also kinda puts them on the government radar. they could very easily all of a sudden be pulled into some of those geopolitics that Nvidia and AMD have been pulled into. Mm. I'm hoping that's not the case, 'cause I really hope that, we get rid of some of those geopolitical issues. but I think it's a great I- if it is true, I think it's a great win for Qualcomm.

Karl Freund 4:20

Yeah. Did they announce where this would be stood up?

Jim McGregor 4:24

They have not, and that's

Karl Freund 4:26

the interesting part. I was wondering if maybe they're gonna do, like Singapore or something like that, that would allow them to, perhaps sidestep government intervention. So I

Jim McGregor 4:34

The government already opened up the prospect of selling H200s and, AMD products That's true but the, the Chinese government has said, "Absolutely not, we don't want them in the country."

Leonard Lee 4:45

Yeah.

Jim McGregor 4:45

Yeah.

Leonard Lee 4:46

And I think a lot of the policies are geared more toward, concerns around, foundation model training versus inference, right? True. And, and the systems- are different. And, and, you know, if you recall, when they first started the bans, and NVIDIA introduced they couldn't package HBM into them, right? And so the memory content is different. I think the whole, let's call it security control policies, might have to evolve, Mm-hmm if inference is as much of a concern as foundation model training. And, if you read some reports, y- H100 is great for model training, and so where does that put the older generation, GPUs, right? And those can fall under the radar, or at least the current one, right? Uh, from a policy perspective. So yeah, it's interesting. I guess the bigger question for me is what do they mean by ASIC? And, is it NPU or, is it gonna be a custom job? that was the thing that I thought was, a head scratcher from the announcement.

Jim McGregor 5:53

Qualcomm has indicated that they are interested in doing custom- Custom CPUs and custom accelerators. And that they, they hope to be doing both. So y- it could be one or it could be- One of those one or both. We'll have to wait and see.

Leonard Lee 6:06

so yeah, it's, that was an interesting announcement. And then, of course, I think the other one that caused quite a stir this week was Huawei's 1.4 nanometer Kirin chip, or at least it's been characterized as 1.4 nanometer in the headlines, but that's, that's a claim that they might be able to achieve an equivalent, if you will, in, in 2031 or something like that. Jim, what were your impressions of that?

Jim McGregor 6:35

I take all of that with a grain of salt at this point in time. I still think that they're gonna be limited on how far they can go with their current lithography technology. and they're still gonna be several years behind at best, even with that, and not having, especially not having EUV and High-NA. So it's gonna be a challenge. But with that said, they are a very innovative company. They continue to find ways to be competitive, and they always have. So, you know, even if they go to electron beam or some other technology, I'm sure that they will come up with something that's gonna at least fit their needs and make them competitive in the marketplace. And you have to remember that Huawei's not gonna be selling these chips to the global market. They're gonna be selling it in China, and they're gonna be using it themselves, selling it in China to other vendors, and a select other countries. it's not necessarily a global competitor to what we already have from the Western companies.

Karl Freund 7:28

The government's made it a little bit easier for them and their compatriots to, have a competitive platform in a country where there's really no outside competition coming in, right? Mm-hmm. And the bar is lowered, however. Sound- sounds like they are making, good progress- Yeah and will satisfy the needs of domestic Chinese-based hyperscalers.

Leonard Lee 7:52

I think a lot of the media is latching onto this 1.4 n- nanometer, data point that was shared by the head of HiSilicon. but it's this idea that you're not looking at geometric scaling, you're looking at time scaling, right? That's

Jim McGregor 8:11

exactly it.

Leonard Lee 8:12

So this whole tau- We

Jim McGregor 8:13

stopped geometric scaling, or we stopped tying the numbers to geometric scaling about a decade ago.

Leonard Lee 8:19

Yeah.

Jim McGregor 8:19

So-

Leonard Lee 8:19

And this is what I think is really interesting. I, and 'cause, a lot of people still think that Huawei just steal stuff, they're not innovative, they're not... they're copying everyone else. I still hear this all the time. No, they are a very innovative company I think it's extremely dangerous, yeah, extremely dangerous to assume that these guys are a bunch of idiots. they're not. And, they're also not constrained to, legacy architectures, which-

Karl Freund 8:45

Mm-hmm

Leonard Lee 8:45

you could argue, the current semiconductor industry is stuck with a legacy right now. As you mentioned, geometric scaling has been, is pretty much, um... Now you're essentially stacking anyways, right? And so these guys are doing, 3D stacking and some novel, I don't know if you wanna characterize it a networking or interconnect, to, address tau, right? reducing the signaling, latency, to get more performance, right? They're looking at the problem more holistically and finding an alternative, approach that ecosystem will support. But the question is, how long will it take for them to actually scale this thing economically and in terms of production as well, right?

Jim McGregor 9:35

you- That's the big question you have to remember that there is a lot of innovation in semiconductors still to be had. we traditionally had three main pillars of innovation- Mm and those were the transistor design, which we've gone, from 2D planar to 3D FinFET to,, Gate All Around, technology now. there's materials technology, which continues to change. They're basically- Yeah using the entire periodic table to come up with- Yeah new solutions, and it varies by generation to generation. Matter of fact, that's one of the things that kind of tripped Intel up starting about the 14 nanometer generation. but you also have the lithography technology, and that was the geometric scaling, really tying it to the gate length, which we stopped doing that. But it still imp- we are still scaling it, and it still impacts your design constraints in terms of how you design, features within the chip. And that is- Yeah it is still scaling. I wanna make that clear. It is, we are still scaling out with each generation, just not the same rate that we used to be. Yeah. And now we have a fourth one, which you mentioned, and that is packaging technology, where we're stacking- Yeah die. and we're doing things with the substrate. We're doing things on the base wafer. We're doing, all kinds of very innovative things that are really changing, Moore's Law from millimeters squared to millimeters cubed. To your point, Huawei is a very innovative company. They've always been highly competitive. And you have to remember that it's not just Huawei. It's a Chinese ecosystem. They have government support. Yeah. They've got, semiconductor foundries with SMIC and others over there. SMIC, right, yeah. So it is an entire ecosystem that's supporting this. It's not just Huawei.

Leonard Lee 11:08

Yeah. And it, it's surprising that they're gonna be, targeting, chips, based on this process, for the P90 smartphone- Mm-hmm which I thought maybe they would go for something less demanding than a smartphone SOC. But yeah, yeah,

Jim McGregor 11:27

Smartphones are always on the bleeding edge- Mm-hmm along with the, PC processors now, the data center processors, and the- Yeah AI accelerators. All of those are pretty much on the bleeding edge at this point in time.

Leonard Lee 11:38

So, Cerebras.

Karl Freund 11:41

Yeah. It's, it's

Leonard Lee 11:43

ideal Oh, you got to have something to say about

Jim McGregor 11:45

Oh, yes.

Karl Freund 11:45

Well, of course, I've been a big fan of Cerebras since- First off, congrats.

Leonard Lee 11:49

Huh?

Karl Freund 11:50

First off, congrats. Yeah. Exactly. Congrats to the whole team, and their customers. They've really, grown enough that the public market was very receptive to this IPO. Mm. Stock skyrocketed. it's fallen back somewhat, since then. It's up a little bit more today. But bottom line is th- they're the only guys in town that have wafer scale computing.

Leonard Lee 12:13

Yeah.

Karl Freund 12:13

And there's, there's good things and bad things about wafer scale. I think most people get the good stuff. you, you- That's big eliminate all the switching and all the rack-based architectures to get a really fast, solution to market. But there's a couple of issues you have to deal with, and it's interesting to see how well Cerebras are dealing with those issues. So one of the issues is not enough memory, okay? It's all SRAM, kind of like Groq. But instead of Groq, it's like Groq, right? so you get a lot more SRAM per mo- for the model to run on die. Yeah, I guess it's still die on wafers. But once you go beyond the size of the memory, you need to either offload the weights to a streaming device, primarily for training, and that's called MemoryX. also you have to go off chip more efficiently, If you think about, I- IO is a, peripheral problem, right? And pun intended. you have to use the shore line of the chip to send wires out for IO. Well, what about all the chips that are inside the wafer? Well, there is no shoreline. Zero.

Jim McGregor 13:25

Right.

Karl Freund 13:25

And so a lot of people complain that, "Ah, Cerebras gonna have a problem scaling wafer to wafer, just because they don't have enough IO bandwidth." Well, that's absolutely true, but there are ways around it.

Leonard Lee 13:40

Mm.

Karl Freund 13:40

and Cerebras has done a pretty good job of getting around it. In fact, last week, they came out and said, "Hey, look, we're running Kimi 2, 2.6." Well, that's a trillion parameter model.

Leonard Lee 13:52

Mm.

Karl Freund 13:52

and they're doing it 1,000 tokens per second- in enterprise trials right now. Well, that's about six and a half times faster than the fastest GPU cloud.

Leonard Lee 14:03

Yeah.

Karl Freund 14:04

And it runs Claude Op- Opus at 10X

Leonard Lee 14:07

faster.

Karl Freund 14:08

So it's like, wow, they seem to have solved the problem. So I called Andrew Feldman, the CEO, and said, "How'd you do this?" And they said, "Well, you know, we got some smart engineers, and we figured out a way around the problems of not having enough memory." And what they did is they take each layer of the network and they parallelize at that level. So the inner layer communications aren't that demanding. So That gets them around their IO problem. And since they're just having one wafer, excuse me, one layer of the neural network on a wafer, they get around the memory problem. So it's pretty impressive. Ve- very impressive

Jim McGregor 14:46

And you have to remember, there's still a lot of innovation going on. First off, it proves that there's not one size fits all for AI because- Yeah the models are different, the applications are different, the model sizes are different. So I mean, that's gonna lead to a lot of different architectures, especially for inference processing. Yeah. But also the fact that, we're seeing a lot more innovation, and this is interesting 'cause a lot of it was all around the models and, going to- especially to- towards transfo- going from RNNs to CNNs to transformers, blah, blah, blah. Now we're seeing a lot of innovation in algorithms to run those models. Simple things that they can do and how they run those models that can speed them up two, three, four, even 10X faster. Matter of fact, even working with some of the startups we've seen recently, and NVIDIA, some of the stuff that they're doing. They... Just through that software layer innovation, they're being able to do a lot more than they ever have before, and I would suspect a lot of that's also coming into play with Cerebras as well. Mm-hmm. So I I think that there's still a lot of innovation to be had there, and it's not just tied to the hardware. A lot of it's still in the software.

Leonard Lee 15:49

When you look at- the size of the models that are being used in, let's say, edge AI type scenarios or inference in general, the smaller models. Mm-hmm. you have the whole, tiny ML or tiny AI or smaller AI movement going on. And so there's that as well that I think changes, the technical requirement, situation, for these accelerators, right? The demand profile is, always changing or the requirements are always, shifting, toward, more for less actually. you get more out of less compute, thanks to innovations on, model compression and, the performance improvements that we're seeing with smaller models, for inference, a lot of the requirements are domain or application specific anyways, right? you don't need a world model to, do certain, specific edge functions or, to support, you know, certain edge applications. And then also from a safety perspective, you're seeing a lot of isolation anyway, so it's nicer if you have a capable small model that you can deploy into a container and use, the least amount of resources in ex- executing that, that intelligence, if you will. So-

Karl Freund 17:09

Mm-hmm.

Leonard Lee 17:10

No, that's cool. That's good insight.

Karl Freund 17:12

And I think it's also important to remember that they're not just one customer shop anymore, right? Yeah, yeah. G42. They have a large deal outside of G42 with AWS and with, OpenAI. Well, that's pretty good.

Leonard Lee 17:27

Yeah.

Karl Freund 17:28

mean, if you're gonna pick two customers, those would probably be two of the ones you'd wanna go after.

Leonard Lee 17:32

hyperscalers are th- are the thing, right? I mean, if we're- Yeah gonna talk about Nvidia's results that came out, the top hyperscalers are still 50% of their data center business- Yep right? If not even, if you consider the Neocloud's hyperscalers, which they really are, it's even more, right? they put hyperscalers in this weird bucket called- A- ICE or something, A-C-I-E?

Karl Freund 17:58

ACIE is, is, is for the h- neo- neo clouds and enterprise- Yeah and for, Industrial, right? In- industrial, right. A- as well as c- as well as, the, you know, country-specific op- Sovereign stuff data centers being set up The sovereignty sovereign data centers. That's also in the I- I- AICE- Yeah bucket. But they, it's, it, the thing is that's growing so fast now. Mm-hmm. It's, the, Nvidia's, I think, doing a good job of segmenting the revenue so they can show things that they, that counter the narrative against them, which is they're just selling to five guys. they're saying, "No, no, no, no, we're, we're selling to major, data centers for sovereign AI, for enterprise AI, and for the, neo clouds." Yeah. And that's growing really fast, growing faster than the, than the hyperscalers. And that, the, to me, that was- Yeah the highlight of the earnings call.

Jim McGregor 18:53

Yeah well, it's growing faster for Nvidia. I don't know that it's growing faster overall. I mean, still, the vast majority of the, the money, the CapEx that's going into spending for AI data centers is still by the hyperscalers.

Karl Freund 19:04

Oh, yeah.

Jim McGregor 19:05

it does point out that, the hyperscalers, if nothing else, they point to a key trend, and I think that's one of the key things people focused on while, Nvidia's facing more competition with, ASICs and everything else, as well as the hyperscalers doing their own silicon. There's not a single hyperscaler out there that's using one solution. So I mean- Yeah. Yeah I think it's really important to note that they're still scaling up Nvidia, they're still scaling up, AMD. Now they're using Cerebras, they're using their own silicon solutions. They're scaling multiple solutions for different AI platforms.

Leonard Lee 19:37

Yeah. Mm-hmm. Mm-hmm. Yeah. And, the other thing to recognize here with the hyperscaler bucket that they have is most of the growth is actually networking, so it's- Yeah 3X.

Karl Freund 19:48

Infiniband was up 4X.

Leonard Lee 19:50

Infiniband That's scary. Yeah. Yeah. Infiniband- Yeah. And, um-

Jim McGregor 19:54

Can, can you imagine what Nvidia would be if they were a memory vendor? And, and, and, and I, and I kid you not when I say that. I mean, that's, that's one of the other major things in May that we have to talk about, and the fact that now all three of the major memory vendors are in the trillion, $1 trillion- Trillion dollar

Karl Freund 20:12

cap

Jim McGregor 20:13

mark- yeah, market cap, you know, gang That's crazy And they all hit it in the same month. Yeah. So, I mean, that's huge. I mean- Yeah Micron, SK Hynix, and Samsung are all, you know, above a trillion dollar market cap now, and because of memory. And as, as we look at memory, it's funny 'cause, middle of last year we were thinking, oh, there might be some oversupply and everything else, and then just things went crazy. And now we're looking at 2028 or beyond before- Mm-hmm we ever catch up with memory demand.

Leonard Lee 20:42

Yeah. Um, yeah. A- and I think, and I always like to point out that valuation's one thing, revenues are another thing. it's-

Jim McGregor 20:52

But it points to demand, and the demand is key. And that's what's really driving it.

Karl Freund 20:56

Yeah exactly, Jim.

Leonard Lee 20:57

Or the perception

Karl Freund 20:58

of demand- That, that's producing a longer- is driving the valuation it's producing a longer cycle. Yeah. So instead of a three-year cycle that memory vendors have typically operated in- Yeah two to three years, now we don't know how long this cycle's gonna be. Some people- Yeah are saying it could be five-year cycle. Yeah. And so that reduces some of the risk on their stock prices- Right as, as well as risk of an impending i- implosion. And I tell you, the, financial analysts keep raising targets on these guys.

Jim McGregor 21:27

Oh, yeah.

Karl Freund 21:28

They're all expected to basically double Yeah, and then- Again, and they've already tripled

Leonard Lee 21:34

Yeah, and I think that's also creating risk outside of the data center or the AI- Absolutely infrastructure segments, and that's where, we saw that announcement, I think, or I don't know if it's a rumor, Apple talking to or, inking some deal or coming close to inking a, an agreement with Intel for their foundry services, right? There's no indication on, which Apple Silicon series, product would be the first candidate. But, I think that's where even the likes of Apple, which historically, or at least for the last couple of, a decade and a half, has really, dictated the, the conversation around, advanced, manufacturing. Now they're in a position wh- or, or in a situation where they might have to diversify, right? Go back to, the days before they committed to TSMC.

Jim McGregor 22:30

Well, I would warn you- Yeah that, you have to remember that they made a similar commitment to GlobalFoundries a couple years ago and pulled out of it. So, Mm I take everything that comes out of Apple, especially in terms of new suppliers, with a grain of salt, because I've seen them just tank people and leave people high and dry- Mm-hmm with the crystal display, with the Sapphire crystal displays- Yeah with the, GlobalFoundries with the foundry capacity. Apple can be your best and worst customer all at the same time.

Leonard Lee 22:57

Oh, yeah. Yeah. but y- y- given the situation, don't you think that the circumstances are different for them? I mean, they're getting crowded out a, a, a bit, right? Yeah, but TSMC- If you look at TSMC

Jim McGregor 23:08

results- is adding a lot of capacity too. So, I mean, TSMC is adding a lot of capacity. They've got several fabs- Mm one at least, already in operation in Arizona, another one going up. Matter of fact, we know that, NVIDIA's gonna be producing products here in the States as well at that facility. Amkor is in, in, adding additional packaging capacity, as well as TSMC in Arizona. So, I mean, there's a lot of capacity going in place. And, I-- First off, I think it, it would behoove, Apple to definitely diversify into other foundries. It's not a cheap process. It's a very- Yeah complicated process to be able to d- support multiple foundries, but I think it's in their best interest to do that in the long run. But we'll have to wait and see. I wouldn't count TSMC out by any means, and if Apple decides that they just wanna continue on TSMC, they can, and I believe that they would if it comes down to it. So it'll be interesting to watch. On the other hand, though- Yeah Intel is doing fairly well. we expect to see by the end of this year some announcements around 18A, and/or 14A customers. Yeah and we already know that their packaging group is doing exceptionally well. They noted north of a billion dollars in revenue just from advanced packaging.

Leonard Lee 24:20

Yeah.

Jim McGregor 24:21

and that's out of, Penang, Malaysia, and Rio Rancho, New Mexico. So that side of their foundry is already cranking, and they won't break it out, but I believe they're actually already making money through packaging.

Leonard Lee 24:32

And, and there's a lot of news that came out of Intel this month, right? Mm-hmm. One of which w- is, Alex Katouzian is the new- Yeah head of, yeah, consumer- I was- consumer and physical AI, right? Yes. Yeah. Consumer computing and, or client computing and physical AI. He's gonna kill me for getting that wrong.

Jim McGregor 24:54

I'm gonna tell him you got it wrong.

Leonard Lee 24:56

Oh my God, no.

Karl Freund 24:57

For those in the

Leonard Lee 24:57

audience- Please don't. He works out who might

Karl Freund 24:59

know who Alex-

Leonard Lee 24:59

he works out Alex- He's very buff.

Karl Freund 25:01

He's very buff.

Leonard Lee 25:03

So.

Karl Freund 25:03

for those who don't know, Al- Alex has been with Qualcomm forever.

Leonard Lee 25:08

Oh,

Karl Freund 25:08

yeah. And, getting Intel capturing his, his talent and leadership s- skills, was a real coup for Intel.

Leonard Lee 25:17

Yeah.

Karl Freund 25:18

I mean, if you look at the Qualcomm team and you had to say who's got the most, talent-

Leonard Lee 25:24

Mm

Karl Freund 25:24

on the team, I'd say, I've always thought it would be Alex, and sure enough, there he goes.

Leonard Lee 25:29

Yeah. No, he's a great guy. all the best to him. he's, got big shoes to... He, well, he has big f- shoes to fill. maybe that's not the way of putting it. he has, a very interesting challenge ahead of

Jim McGregor 25:41

him. he's walking into

Leonard Lee 25:42

a very

Jim McGregor 25:43

challenging environment And

Leonard Lee 25:44

so let's give a little love to AMD. Anything on the AMD front? Anything else you guys wanna talk about?

Jim McGregor 25:49

Just the fact that, they, uh, we're not talking financial, we're not financial analysts here, but- Yeah when they came out with their Q1 results, or their Q4 results, you know- Who? they kinda, AMD, they kinda said- Oh, okay. Yeah that, we're worried about memory and everything else, and I- Yeah kept looking at this and saying, "Guys, I'm not worried about AMD. I think they're cranking." And, after seeing the Q1 results, obviously they- they've kinda revised that. They are cranking. They're doing very well. And the thing about AMD, and I- I think that this kinda gets lost in the a- in the whole AI discussion, a lot of times, it's not just about selling AI accelerators. They are doing well.

Karl Freund 26:27

Yeah.

Jim McGregor 26:27

They are making headway with the MI300 series, and a lot of interest in the MI400 series. But they are also very, very competitive, and I would argue, and continue to gain market share in server processors. So they are no longer, I think, the second child in that race anymore. they are doing exceptionally well, and Epic has favored in a lot of those AI platforms. and they continue to do well. They've got great partnerships with, with TSMC and GlobalFoundries. They're doing very well in, especially in desktop PCs. they're doing well in embedded segments. You gotta remember that they invested heavily w- in all these- Yeah industrial embedded applications with Xilinx, and they continue to execute very, very well.

Leonard Lee 27:11

Yeah. And you know what? speaking of embedded, one of the things that I'm noticing is the IoT for AI infrastructure is starting to turn pretty hot. Last year when we were at Cadence LIVE- Mm we heard a lot about digital twinning. NVIDIA had this whole collaboration with Cadence, and I'm sure they had it with Synopsys as well, about, instrumenting, and modeling AI factories or, AI infrastructure data centers, right? But, you can only do that if you have, um, sensor devices and embedded, technology out there providing that perception edge around that data center or for that data center. And so I, I don't know if you're noticing, some movement there, but it's interesting to see that, some of the, the non-processor, 'cause you mentioned, the accelerator. We- we're seeing some of the IoT-ish, chip guys- Mm-hmm out there doing pretty well now, you know?

Jim McGregor 28:12

a- absolutely. I, I think we've s- we've turned the corner on that whole supply glut that we had, you know- Yeah coming off of, COVID and everything else.

Leonard Lee 28:19

stopped, yeah.

Jim McGregor 28:20

And we've always said this about IoT, and it extends, to your point, to the data center. Sensors are extremely important. Mm. wireless communications are extremely important. RF, and, analog solutions are important, obviously, as part of that. And then you also have power. Power. Just managing power is becoming critical in just about every environment. Mm. So it is, a... We, we definitely see the market turning and all of those guys doing well. Matter of fact, I still think that sensors is the underrated area, and I think that we're gonna see a lot of... We've seen some transitions. NXP sold off some of their sensor, st- some of their sensor products and technology. we've seen On Semiconductor buying, we've seen TI buying, we've seen other companies buying, or ST, sorry, ST buying. They bought the NXP stuff. it's gonna be interesting. I think we're gonna see a couple of, major sensor vendors emerge out of this over the next decade.

Leonard Lee 29:16

yeah. And then just going really quickly back to AMD, one of the things that I sensed in, and have sensed in their reporting for the past few quarters is, CPU and traditional data center stuff, i- is, really a key driver- Mm not only of profit, but, a significant driver of revenue as well. They put CPU in front of GPU. and the odd thing is also, during Arm's, earning call, Rene, I think, he mentioned th- their CPU pipeline has doubled since, Arm Everywhere. and then we have NVIDIA. What did NVIDIA... NVIDIA announced, or, I think Colette mentioned that, uh, CPUs will be, what, 20, is it 20 billion?

Karl Freund 30:06

20 billion.

Leonard Lee 30:08

Yeah.

Karl Freund 30:08

Yeah

Leonard Lee 30:09

what do you th-

Karl Freund 30:10

what do you

Leonard Lee 30:10

think?

Karl Freund 30:11

I think NVIDIA's kind of underrated by many people in terms of the strength of their CPU business. If you look at the Gartner, server report, it shows Arm growing fantastically, and 80% of that growth is roughly is NVIDIA. and so While NVIDIA's two biggest competitors are both x86, Intel and AMD, the pole position for Arm servers is wide open, and it looks to me like NVIDIA's winning it.

Leonard Lee 30:44

Yeah.

Jim McGregor 30:45

I would agree on the Arm side, but you also have to remember a lot of those Arm CPUs, are displacing what used to be other risk-based CPUs for storage- Yeah for networking, for other platforms within the data center. So, but it is interesting as you start looking at the numbers for, the likes of NVIDIA, AMD, Intel, et cetera. It is data center and then everything else.

Leonard Lee 31:09

Yeah. I think everything else is only, like, 7% of NVIDIA's business. It's, like, insane.

Jim McGregor 31:15

Yeah, gaming is now everything else.

Karl Freund 31:17

Everything

Leonard Lee 31:17

else. Who would've thought

Jim McGregor 31:19

that?

Leonard Lee 31:19

Yeah, they didn't even bother talking about everything else. It's like, well, we have two, two parts of our business, hyperscaler and this ACIE thing, right? I mean, it's amazing how much things have changed in just one year. I, um, tuned into Dell Technologies World. One of the things that came out of there that I thought was really interesting is they're pivoting toward warm water cooling, Lenovo was one of the f- two or three pioneers in this particular area, and one of the things that's really interesting about warm water cooling, it uses facility water, so y- it reduces requirement for, chillers, right? and that has pretty huge implications in how you design your data centers, which is interesting because now we have f- you know, generation over generation, the data centers themselves are on a advancement or architectural, diversification curve.

Jim McGregor 32:19

Well, and that, that was a big focus of Open, Open Compute last year. Yeah.

Leonard Lee 32:23

Yeah

Jim McGregor 32:23

I just don't see the immersion taking off, and the warm water is much more appealing. And quite honestly, the directed chip is still a much more appealing solution than I think immersion ever will be. Yeah. But yeah, it is... You're right. and that's an area where you've got companies like Vertiv, and Flex- Yeah and LiquidCool, and you know, that continue to innovate around there.

Leonard Lee 32:43

Yeah. And the other thing, um, AI workstation. Yeah. More and more people talking about it. And Jim, remember you and I, when we were at Computex, was it last year? Mm-hmm. We did our little bit, and we talked about how AI workstation is probably gonna be that beachhead for enterprise AI. Yeah. And lo and behold, we you know, we have OpenClou that got everyone excited about putting, agents locally on device. And, yeah, there was a lot of talk about that. And then I think, the, the work that Cisco's doing in helping to fortify some of these open, agentic frameworks on device i- is pretty, pretty interesting. I'm tracking that pretty closely, and I'm gonna be at Cisco Live next week, so I'm gonna be delving into that. And Jim, you still owe us That, readout from your team on OpenClaw Yes, we

Jim McGregor 33:40

need, we need to get Damian

Leonard Lee 33:41

on here- And NemoClaw

Jim McGregor 33:42

because he's-

Leonard Lee 33:42

Yeah

Jim McGregor 33:42

He's been working on, first off, the, especially the different types of workstations, from a full-on workstation down to the mini workstations. And it really is going, the workstation market's going through a renaissance with AI because it is so much more efficient, or I should say, economical- Cost efficient,

Leonard Lee 33:59

yeah, economical

Jim McGregor 34:00

economical to actually be working on those than sending everything to the cloud all the time. Yeah. So that's going through a renaissance, and then OpenClaw just exploded everything, where everyone's, looking for new ways to create personalized AI agents, sometimes in a very dangerous fashion. But it is- Yeah it is just exploding. It is amazing- Yeah what's going on. So we should really have, Damian- We sh- and/or Kevin on one of these calls to talk about- We should what's going on.

Karl Freund 34:26

It'll be interesting to see if Nvidia makes that a highlight next week at Computex.

Jim McGregor 34:31

it will be

Karl Freund 34:32

I think it will. And Jensen was this morning announced, significant increase in his investments in, in, in Taiwanese companies. and I think a lot of that's gonna be around the PC. Who has the best platform for running OpenClaw?

Leonard Lee 34:48

Yeah.

Karl Freund 34:49

It's probably made- I mean, that's- It's probably Nvidia

Jim McGregor 34:51

Yeah, and you should, we should probably highlight that and the fact that Nvidia, now that's basically printing money, should It has become one of the largest VCs in the tech industry. Yeah. Mm. Matter of fact, I've got a tracker now, and, like at least once every two weeks, there's a new announcement of, about an investment or partnership, not always the financial- Yeah details, but, something Nvidia is doing with somebody, and it is, it's ridiculous. I think I've got over 60 names on there already, and that's just over the past three years. It is insane. and they are, they have... I think Jensen's always had, a preference towards the Taiwanese companies and working with the Taiwanese companies and raising up the Taiwan,

Leonard Lee 35:32

Yeah

Jim McGregor 35:32

visibility and, tech ecosystem. But even beyond that, Nvidia is investing across the

Leonard Lee 35:39

spectrum. Yeah. I think he's-

Jim McGregor 35:40

You know, from networking to EDA tools to, models to, you name it. it is incredible the amount of investment that this company is pouring back into the industry, Yeah through its success.

Leonard Lee 35:53

Yeah. and it's other income

Jim McGregor 35:57

Yes, in its other income

Leonard Lee 35:58

Which is interesting. We're seeing a lot of these players, report other income and, these are their investments. Th- they're- Mm-hmm they're reporting their unrealized, um, gains on these, investments that they're making, in startups and such. So yeah, that's a... That was another interesting thing, thing that came up in this, last, earnings season for the tech guys. I think it was Google and AWS that reported really big numbers. So, but then, um, yeah, anything else? Want any things that you wanna get off your chest and share with our audience?

Jim McGregor 36:36

I would just say that- I think we covered

Leonard Lee 36:37

a

Jim McGregor 36:37

lot already,

Leonard Lee 36:37

but-

Jim McGregor 36:38

myself and Damien will both be at Computex, so we'll be reporting live from there. Awesome. Damien will also be at, Automate, Industrial, Robotics Automation Conference in Chicago, in June. And then I will be at the Qualcomm, Investor Day also in June. So a busy month. And Kevin will also be at N- Nebius Inflection, so it's gonna be a very busy

Leonard Lee 37:04

month for us, yes. Really? Okay, cool. Yeah, I'll be there. I have eight events next month, so it's- Eight? Eight.

Jim McGregor 37:13

Wow.

Karl Freund 37:14

Only four weeks, so you're doubling down.

Leonard Lee 37:18

I don't know what happened.

Jim McGregor 37:19

Well, and there are a lot more events- Yeah on our calendar. That's just the ones that we've committed to.

Leonard Lee 37:24

Yeah. Carl, any last mentions or sharing,

Karl Freund 37:29

I think we should all watch what happens in Taipei. I think Computex... a lot of companies I work with are saying, "Wait till Computex. We'll tell you all about it."

Leonard Lee 37:40

Mm-hmm.

Karl Freund 37:40

I really do think that, it will be a big event for all the Nvidia wannabes to come up with new technologies, new announcements, new customers. So, I think it's- Not gonna miss out bears watching. Bears watching.

Jim McGregor 37:53

and this is important because Computex traditionally was The PC show. The largest PC show in the world. now it is really transformed, and it's funny 'cause at one time they tried to tran- transform it into mobile, and that really didn't work well. But now there is a whole section, especially one whole floor that's dedicated to embedded computing. So industrial- Yeah automation. Yeah robotics, those types of applications, and we're seeing a lot more, especially since AI is so dominant in the data center, we're seeing more innovations and announcements and everything around the data center and AI- Yeah at Computex as well.

Leonard Lee 38:25

Yeah. and I think you're right, Carl. This might be an AI PC, AI workstation year for Computex.

Karl Freund 38:32

Yeah.

Leonard Lee 38:33

Yeah. I

Karl Freund 38:33

think so.

Leonard Lee 38:34

Yeah.

Karl Freund 38:34

find out till next week, yeah?

Leonard Lee 38:35

Yeah, we will. So gentlemen, let's call it an episode. What do you say?

Karl Freund 38:41

Yeah,

Leonard Lee 38:42

it's an episode. Carl's gotta go, so yeah.

Karl Freund 38:43

go.

Leonard Lee 38:44

Thanks everyone for, tuning in. We really, appreciate your viewership, and, thank you, gentlemen, for sharing your knowledge and, your, insights.

Jim McGregor 38:52

I didn't get to use the slap icon, but we'll use it next week. I'll save it- Yeah for then.

Leonard Lee 38:57

did you invent it? Did you actually vibe code that?

Jim McGregor 39:02

I did, but we'll save it for the right time.

Leonard Lee 39:03

Oh, okay. All right. Send it to me, okay? Send it to me in a zip file. But hey, everyone, make sure to reach out and follow Carl Freund @CambrianAIResearch www.cambrian-ai.com. He's also on Substack and Forbes. also LinkedIn. So- Connect with him, tap into his research. It's great stuff. And also reach out to and follow Jim McGregor and the Terius Research Team, who are, the foremost authorities on, everything that has a chip in it, at www.teriusresearch.com. And, also please subscribe to our podcast, which will be featured on the Next Curve YouTube channel. Check out the audio version on Buzzsprouts or find us on your favorite podcast platform. Also, subscribe to Next Curve Research Portal at www.next-curve.com for the tech and industry insights that matter, and we'll see you next month. Take care, gentlemen. Cheers.

Karl Freund 40:15

Safe travels.