Silicon Futures for February 2025: NVIDIA, Arm, Dragonwing (with Jim McGregor)

Jim McGregor of Tirias Research joined Leonard Lee of neXt Curve to recap February 2025, another action-packed month in the world of semiconductors and accelerated and non-accelerated computing on the neXt Curve reThink Podcast series, Silicon Futures. Jim and Leonard share their thoughts on the state of NVIDIA, Arm's new edge AI play, and Qualcomm's Dragonwing.

This episode covers:

➡️ NVIDIA's Q4 2025 Earnings & Outlook - The bellweather of AI (2:31)

➡️ Intel dismantling rumors regarding Broadcom & TSMC  (6.00)

➡️ Arm announces new IoT Platform and Cortex A320 (10:03)

➡️ Qualcomm unveils their new Dragooning brand for industrial(11:21)

➡️ MWC 2025, Embedded World, NVIDIA GTC, OFC, IBM and more! (16:50)

Hit both Leonard 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.

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 for the tech and industry insights that matter.

Transcript

0:06

Welcome

Leonard Lee: 0:10

everyone to this episode of NextCurve Rethink Podcast. Um, and you know what we do here? We break down the latest tech and industry events and happenings into the insights that matter. And I'm Leonard Lee, Executive Analyst at NextCurve. And in this Silicon Futures episode, we're going to be talking about The happenings of February. yeah, I'm no doubt. We're going to be talking about NVIDIA because I need to get the take from Jim McGregor the highly accelerated Jim McGregor of Tyria's research. I think you sound like I'm stoned or something. I know Yeah, I don't know. You could be I used to Carl too. So I'm just you know, recycling Recycling titles but, yeah, we're gonna try to talk about NVIDIA as little as possible, but I'm afraid that's not going to be possible. But I have my good friend, Jim McGregor here. Hey, Jim, how's it going?

Jim McGregor: 1:11

How you doing?

Leonard Lee: 1:12

You have a very peaceful background.

Jim McGregor: 1:14

actually you should see the other one I spent a week up in Alaska watching the Aurora.

Leonard Lee: 1:19

Oh yeah, that's right. well, you're going to have to tell me about it when we're in Barcelona next week, or at least no, actually on Saturday. hopefully you had a, I had a great time. I had some, it

Jim McGregor: 1:31

was spectacular. Yeah, definitely worth it.

Leonard Lee: 1:35

Okay. Well, we don't have a lot of time, so we got to get on with it. But before we do, uh, please remember to like share and react and comment on this episode. Also subscribe here on YouTube and buzzsprout to listen to us on your favorite podcast platform opinions and statements made by Jim are entirely his own and don't reflect those of next curve or myself, so. We're, doing this We have to get the legal disclaimer. Ah, yeah. Come on, man. Don't give me a hard time here. But, yeah, we're doing this to provide an open forum for discussion and debate on things, related to technology and happenings in the semiconductor and AI world. in this installment we're going to be kicking things off with NVIDIA because we both just got off of a series of calls. post, earnings. Jim, what's your take on, what did you think? we had the earnings calls and several, pre briefings and briefings, post earning call. what was your take on things?

Jim McGregor: 2:35

Well, first off Nvidia is still charging ahead. we had the whole deep sea scare and wall street kind of got it wrong. Thinking that, Oh, well, it's going to drive down the amount of GPUs. No demand actually drives demand for GPUs. So, no, they're in a very good position, they've got a full stack solution from the chips to the boards to the systems to the software. They are cranking, their earnings are great, they even beat estimates, they're projecting, A great Q1, which is actually usually when we're down a bit in our industry. So, I mean, there were a couple of soft spots in gaming on their earnings, but pre launch of their new Blackwell platforms.

3:16

So you

Jim McGregor: 3:16

kind of expect that a little bit, but other than that,, they're now making money or they're now increasing revenue once again in, robotics and automotive, which they kind of lumped together, but most of that's automotive. and they're doing great in the data center, and they're still preferred solution. And we're 2 weeks out from, GTC where, we're going to see some new products from them. I have no doubt about that.

Leonard Lee: 3:41

So, yeah, yeah they have to keep the, they have to keep the train moving, right? They got to keep this feverish pace going. I beg to differ with you as well as Carl, because Carl was on. Yesterday, in terms of the deep seek impact, I think it hasn't been studied very well. Everyone keeps if you're focusing on our 1, you're missing the point, you there's a lot of unknowns related to V3 that have not been explored, or maybe people just don't want to. See it the way that they should see it, but I think the jury is still out and the results don't bake in, the deep seek impact because I think that's going to be organic. It's not going to be hype based or speculative. I think it's going to be a material. Thing that's going to be, impacted by what I call token dumping, which is actually a reality, right? The price of tokens, not the cost that being something different is collapsing because of deep seek or one. and so, yeah, let's see how that plays out because, I think this is the beauty of what we do here. In this collaboration is we have, different points of view and we converge on things that we've distilled into what we think is a grounded perspective. Right? I mean, I think, obviously your view is that your view is grounded and I, so do I, but I think the interesting. synthesis that happens is when we have these debates and take two diversion points of view and then distill into something that's even better.

Jim McGregor: 5:10

I think the only thing we can really take, effectively from deep seek is the fact that, it points to a new trend. We're gonna see a lot of deep seeks pop up. We're gonna see a lot of companies pop up. whether they're using their own data, somebody else's data, we're going to have all kinds of providers, of models out there in the market. And yeah, it's more competition, it's going to drive down prices, but I think that's the only thing we can really take from DeepSeek at this point in time is the fact that this is just the beginning. We're going to see an explosion of different models and service providers and everything else. This is not just a big guy's game. It's not just going to be open AI. It's not just going to be Microsoft or Google or anybody else. There's going to be a ton of solutions.

Leonard Lee: 5:54

Yeah, yeah. And that's disruption though. I mean, that's like book disruption, what's happening. And so, yeah, one of the things that I was really curious about, and I'd like to get your reaction to this is it seems like this whole notion of post training, is becoming really, let's say, divergent, or there's a lot of contradictions there in terms of themes, you know, one of the things that Jensen mentioned on the call yesterday. Or the three scaling, observations, right? I refuse to call them laws, by the way. It's just observations. it's weird because his position is that, there's going to be more compute dedicated to post training when I think R1, okay, and I'm not talking about V3. R1 with the distillation approaches and sort of these novel things that they did on with fine tuning, that connotes something. I think a little bit opposite of what, or sort of a headwind dynamic versus what Jensen was suggesting on the call. I don't know. What do you think?

Jim McGregor: 7:05

Well, I think, I don't think the future is necessarily a single, open model or a single model. I think it is a mixture of experts where you're going to be using a lot of models that are, and I think that's where that post training comes in. The post training comes in to where you're optimizing it either for a particular application, a particular workload, a particular use model, whatever. Sometimes even for a particular company based on their information, or something like that. So I think that's where he's referring to is the fact that a lot of people are going to take these open models, and, whether it's deep seek, whether it's a llama, whether it's whatever, they're going to take these and do a lot more post training on them to shrink them down to optimize everything else. So I think that's where he's coming from. And no, it doesn't take the 100, 000 GPU clusters to do that, but it does drive more cycles. It does drive a lot more use. And quite honestly, it's a good sign because it drives AI closer to applications.

Leonard Lee: 8:06

Yeah. And I think that's where the impact of, this token dumping, that's happening right now. And that's probably a pretty mean term. maybe I should soften it a bit, but it. In my view, it is kind of like a, a dumping deck.

Jim McGregor: 8:23

It's kind of dumping, you know, they're trying to undercut everyone's price. they're doing it even below their cost.

Leonard Lee: 8:30

Yeah, well, we'll see about that. Right. Or who knows, they might be giving it for free, but see, that's the other thing. I think the monetization question is a big one. I know, like in the Nvidia circles, they like to put revenue ahead of the equation. Demand, does not equal monetization unless you're charging for that demand, right? You have

Jim McGregor: 8:52

to have a business model

Leonard Lee: 8:53

And that, I think, is huge. It's a huge problem that I think is pretty persistent and, hasn't been resolved or abridged in the past, two years.

Jim McGregor: 9:06

Just at the start of the internet, you know, all these other companies came out and most of them didn't have a business plan to make money. Um, there's always a fallout. There's a correction. and we end up with something that's actually very effective in the end.

Leonard Lee: 9:20

Yeah. Well, come on, Jim. No, I, I saw a lot of business plans back then. They just weren't that good

Jim McGregor: 9:28

They weren't business plan to make money. They weren't. Everyone was just jumping in because they wanted a foothold. I mean, let's face it. Meta's giving stuff away as fast as they can. With no business model to make money off of Llama. They just want to be the de facto.

Leonard Lee: 9:45

But I think that's where this whole question of operational AI is interesting. Because I think Meta and some of the hyperscalers are different. in the value proposition, let's say of generative as an operational tool or technology versus for an enterprise, where you're trying to

Jim McGregor: 10:05

take money off it through their own businesses.

Leonard Lee: 10:08

Yeah, I mean, it'll be difficult to, but this would be like, let's say, maybe raising the competitive parody bar, versus actually creating new value. Right? the actual revenues might just be. The same, you're just either, you could be fabricating eyeballs, which then raises the question of what, how legitimate is this revenue that's being generated? Is it delivering real economic value versus just, let's say, printing tokens, I'm saying, so I think those are the things that are actually interestingly Those topics are interestingly starting to converge or at least be expressed in the discussion. So, I think it's a really exciting time to be talking about, generative AI, but then also, NVIDIA's role and how things are shifting.

Jim McGregor: 10:56

And 1 of the things I think, and I've said this on previous podcasts. Is I think every company out there needs to not just ask how they can use AI, but how is AI going to change their business model?

11:07

Yeah, yeah.

Jim McGregor: 11:07

And very few companies are asking that, even tech companies. Ah, I give Jensen credit because when I asked it of him last year, he looked at me and says, Jim, I wish I knew. I don't know where NVIDIA is going to be in 10 years. He's just kind of going with the flow and going after opportunities as they arise. And that's pretty much what you have to do.

Leonard Lee: 11:26

Yeah, yeah. Okay. Well, anything else you want to talk about regarding NVIDIA? I thought you'd have a lot more to say or did I just crowd you out?

Jim McGregor: 11:36

Well, no, you took it down a particular avenue. Now, NVIDIA is making a big, where NVIDIA also has a strong presence and we're starting to see other companies trying to enter that and use the same terminology and they term it physical AI. And that's really robotics and autonomous machines. And, while the rest of the industry is focused on agentic AI right now, just trying to create agents for everyone to use, they're focused on that next step, and that is really marrying AI to robotics. It was funny because, we've all dreamed about the house robots since the fifties with the Jetsons and, and all that stuff, but, this may actually be a time either whether it's this year or by the end of the decade, we're going to see robotics take a leap beyond just, I think the manufacturing floor to being practical solutions and even humanoid solutions. I was impressed with some of the ones I saw at CES. I was like, wow. When they're walking up to me and shaking my hand, I was like, okay, I'm impressed.

Leonard Lee: 12:36

you can underestimate how difficult that is. You can't, right?

Jim McGregor: 12:42

Yes, you can't. It's just a simple thing of twisting off a bottle cap, shaking a hand. Anything that requires some level of sensitivity, complex motion, is very challenging. But, we took a major step with AI in teaching computers how to learn. that translates more to robotics than any other application.

Leonard Lee: 13:04

Yeah. It's interesting. that handshaking robot, I think the name of the company is, is it UB name escapes me.

Jim McGregor: 13:13

There were three of them.

Leonard Lee: 13:14

Yeah. But there was one that didn't walk up to you. It didn't have like a head. It had like this, this strange donut type of thing as a head. And it walked up to you and it would shake your hand, right? And so I took a video of it posted on, on YouTube and one of the comments was, so what? And so all everything you just said, I mean, obviously some people don't appreciate and maybe don't understand how difficult it is to make something like that happen. But yes, simple things like that are tough.

Jim McGregor: 13:47

And when you think about how manual certain processes still are, like, agricultural, especially when it comes to, picking fruit or something like that's very delicate, we're in a state where robots could actually physically do that now,

13:59

and

Jim McGregor: 13:59

we could overcome, and that's an area where we have a huge area of labor shortage. not to mention health care, where we have a labor shortage, there are a lot of very practical applications where having a humanoid or humanoid type robot, or at least having those kind of capabilities, is going to play a huge, a huge factor. So, I think it's. Incredible to watch.

Leonard Lee: 14:23

Yeah. and you know, I wouldn't give the folks doing a gentic AI stuff a hard time though, because I'm trying to renew my global, global entry and it's going to take 14 months. And so I think a gentic AI would be really nice. No,

Jim McGregor: 14:40

no. I'm just telling you now what's going to happen, especially if you already have it, is you're going to get a message in about a month that says, Oh, well, you're already pre approved.

14:51

yeah, anyway,

Jim McGregor: 14:53

it's going to come it's happened to myself and everyone else.

Leonard Lee: 14:57

yeah. Okay. so anything else on Nvidia? Because I want to pick your brains on other things because a lot of stuff happened in February and I'm sure. Let's move on. Let's move on.

Jim McGregor: 15:07

always gonna be top of stack,

Leonard Lee: 15:08

Yeah, yeah. Okay. the one thing that I want you to, respond to or share your reactions to slash thoughts, this stuff about the Nvidia Intel split, right? And Broadcom, Broadcom and, TSMC. The whole deal that's going on right now, or the rumors

Jim McGregor: 15:27

first off, that won't go through. It will not go through because you've got two foreign entities that the U. S. government's never going to approve, especially when they have investments there and Intel's designated as one of those critical suppliers. So that won't go through. and I quite honestly, I think the. Even the proposal of doing it this time is still a foolish mistake. I think you need intel especially as it's where it's developing its foundry services And getting its product strategy back on back on track. You still need it as a combined entity at this point in time. I think separate your you're, you're not only, hamstringing the entities that used to be Intel, but also the companies that acquire them. And I think that's the biggest challenge.

Leonard Lee: 16:11

Yeah. Yeah. And quite simply, in my view, you are, you're killing off the leadership dynamic, and who don't see that, you're doing the US industry, semiconductor industry and the objectives of, achieving us. Technological leadership in manufacturing, you're undermining it. And unfortunately, there's a lot of folks out there who are advocating for it. And I think it's highly misguided.

16:42

yeah,

Leonard Lee: 16:42

yeah. Oh, no, I'm dangerously slow because I, the thing is that we can't forget that Intel has a huge ecosystem. There are a lot of different industries. Telco, we're going to, we're going to MWC next week. Telco, industry, you want to talk about O RAN, Open RAN, right? A lot of it is x86. Intel bound, right? And you start messing with that. You mess with a lot of strategic technologies and themes that, at the moment cannot be hobbled. If you do that, just because you want to. Make some quick buck on, banking fees or whatever M and a transaction fees, you're really doing a disservice. I mean, there's a bigger picture here. That really needs to be recognized. it's just something that really bothers me, Jim. So I'm glad that you and I are on the same page with this and I'm sure that Carl is as well.

Jim McGregor: 17:40

And the biggest thing you have to consider is Intel's value as a whole. the entire industry in terms of R and D, they have the largest R and D budget of, I know, and they continue to invest in quite honestly without them. the entire industry wouldn't be where it is today. and a lot of that is still tied together, extends from deep product technologies like, laser based, optics too. silicon optics and to, neuromorphic computing to, manufacturing to semi inert to process technologies and, transistor designs and everything else. Losing that right now. well, first off, you can't really separate the two right now. And I think that it'd be a huge loss. Unfortunately, wall street has always undervalued. I think R and D.

Leonard Lee: 18:28

And I think that's shooting and shooting the industry in the foot. My personal opinion. okay, so good. Hey, we agreed on something. I think we agree on a lot more than we'd like to admit, but, oh, the other thing, you probably wanna talk about ARM and their new announcement, right? You and I are on the call. No. Which, well, when is this? You? the Edge, edge AI iot? Yes. Yeah, yes. Well,

Jim McGregor: 18:58

they are putting together a platform, more competitive. It's what really amazes me about the announcement that they've made right now is the fact that they're bringing down. They basically broken the barrier between their A class, which is the application CPU cores and the M class, which are the embedded microcontroller course. So they've broken that barrier to where they overlap now, so it now comes down to a system design option of, you need, do you need on board memory, like a microcontroller, what do you need, what's your power budget, it really gets down to where it's broken the barrier between an MCU and an MPU. And I think that's the biggest part of the announcement for me, to come out with the A320, which is the most power efficient A class core they've ever come out with, which can still do high level applications and operating systems. And they're also announcing, obviously a kind of revision of Ethos, their AI core, which is mostly for embedded applications. they're still targeting their core market, which is at an IOT segment, and I think they, they really. it's funny because you might say they're creating more complexities for designers to figure out what they want to use.

20:19

Yeah,

Jim McGregor: 20:19

but this really, I think, breaks down that barrier between the microcontroller and the processor. And I think that's really good for a lot of designers.

Leonard Lee: 20:28

Yeah, and, it might just be, ARM tuning into sort of this trend where you do have, power efficient. AI compute, if you will, or edge AI, now, making its way further out toward the very edge of the edges out there. And, I think that's also another thing that, Qualcomm's recognized for quite some time, or at least that's a core part of their strategy, which leads us to dragon wing. Right. What did you think about that? They, we're inspired by curious research purple. Well,

Jim McGregor: 21:06

first off, I've been encouraging them to break off a separate brands for different markets for over five years. ever since they introduced their automotive product line, I think they need a little bit more brand diversification to separate that out And sticking with at least dragon in the title, I think helps create that recognition and they really need to do that. they've always had the philosophy of mobile first And that's fine. And that's good because it really shows our commitment to low power, and focus on, performance efficiency. However, especially as they've gotten into these other markets, they do have to invest not just in that low power, but in specially technologies for each different product area, whether IOT or networking or, even data center. So I think that helps establish the fact that yes, this isn't we're not just a smartphone company. We are focused on the markets and we do recognize that, both our investment in R& D and our investment in our brand should should recognize that.

Leonard Lee: 22:09

Yeah, and that point on R and D, I think that's 1 of the, I mean, the leadership that they have in a lot of different categories in as well as wireless connectivity. I think those are the things that really positioned them. Well, and I think what we're probably going to see, and this might take time because you, as you know, these brownfield IOT environments, across different industries tend to stick around for a really long time, right? what I'm going to be interested in seeing is how, the industry, you know, players like Qualcomm, ARM, NXP, others, can take this edge AI theme and then change the thinking about, what these. Edge, IOT, industrial systems look like, right? Because fundamentally there has to be a re architecting of, away from what you might consider legacy to this new 1 where, the infrastructure is going to look different. the system deployments are going to look different and the way that you. Place, compute and data is going to be different and that's kind of a heavy lift. You know what I'm saying? So how all of these guys maneuver into this future vision they have of, a transformed, industry is going to be really interesting. It's going to be a tough journey, I think, for a lot of these guys.

Jim McGregor: 23:36

AI was kind of awkward for the IOT and embedded segments because it's kind of like bolt on. And that's what a lot of companies, especially startups were kind of proposing. Okay, you've got your microcontroller, you got your microprocessor unit, you bolt on this accelerator, blah, blah, blah. It doesn't work when you're dealing with form factor, cost, power, efficiency issues. You need a holistic solution. So you have to start with AI in mind. Yeah. Qualcomm, for example, is probably the best company we know of at heterogeneous computing.

Leonard Lee: 24:06

Yeah.

Jim McGregor: 24:06

Elements that they can, different compute elements they can put on a chip. It's phenomenal and make it really powerful. Synaptics is doing a good job with some of their new products. NXP, Microchip, Renesas, ST, are all good at doing certain things. But definitely the generation of products we have to see have to be designed. from AI from the start and they have to have this heterogeneous. Architecture to be able to do it. More importantly, though, I think we need kind of open platforms to be able to develop the solutions on because seeing also is that everyone has their own tools. Everyone has, you know, their own platforms and you have to optimize for the platform. That doesn't work for developers. You really need OpenStandard. Synaptics is one that's tried to create, they've tried to offer theirs as an OpenStandard. And there's other companies trying to do that too. But there's not one out there at this point in time. So it's going to be, I think it's still going to be a challenge for a while for the embedded market. But you're starting to see the lights go on. That's the good news. I think the most important thing to talk about is the fact that all the things we have coming up. Several major events coming up where we're going to see a ton of announcements Some have already come out this week prior to mobile world congress and embedded world, but there's going to be a lot more coming We have mobile world congress in barcelona. We have Embedded world in nuremberg the following week. We have gtc In san jose, there is an IBM event that there's OFC Intellivision and, MediaTek event all the week after that. It's a crazy time.

Leonard Lee: 25:47

Yeah, it is.

Jim McGregor: 25:48

It's the semiconductor spring.

Leonard Lee: 25:53

I haven't had a break. You can't live with that 100. Yeah. Yeah. yeah. And, I'll be joining you in many of those. And especially next week at MWC, right? So, yeah, it's going to be interesting. AI RAN probably going to be a big topic. let's see how that shapes up. I think it's still really early for that stuff. but, yeah, a lot of my focus there from, a chip standpoint. Or a semiconductor standpoint is going to gravitate around, accelerators for, Problems to solve today, right? Not like a re architecting of the RAN or, that stuff I think, is going to take some time, but what are some of the things that you're going to be looking forward to next week?

Jim McGregor: 26:38

a lot of discussion around O RAN and AI RAN, as well as 5G advanced, 6G. I think we're going to see more around some of the. other type of mobile applications and mobile devices. Yeah, there's, I did note that there's a number of robotics companies that are going to be there. So it's kind of a mixture of a lot of different things. I think there's going to be a big focus though, on that core network and the, wireless network architecture going forward.

Leonard Lee: 27:06

Yeah, you mentioned core network that that 1 is really important, especially, the transition to stand alone. Right? which is progressed a lot slower than the industry has hoped, but I think the, the timing is actually really good now, before it probably it didn't make sense for a lot of operators. And so, yeah, we'll have to definitely get together. either in Barcelona, if we have time, because, it's going to be just, a mad schedule. for me, I'm sure you as well, but definitely let's do a recap when we, get back, at the least. Right? If not, we'll hang out there and try to do something live in Barcelona over some, Rio, Wine and, Sangria

Jim McGregor: 27:53

and tapas.

Leonard Lee: 27:54

Sangria and tapas.

Jim McGregor: 27:56

What country are you going to? I'm going to Spain. Sangria and tapas. Come on.

Leonard Lee: 28:03

it looks like you have Spain in the background there. It's beautiful.

Jim McGregor: 28:07

It's a very similar climate.

Leonard Lee: 28:09

Yeah, yeah. That's my backyard. That's your backyard? Yeah, that's my backyard. That's incredible. That's crazy.

Jim McGregor: 28:19

I live on 120 acres and border 640 acres of open state land.

Leonard Lee: 28:24

Wow. Wow. you're a genuine Ben Cartwright. Kind of remember that. Yes. I'm old. Yeah, you watch the show too little Joe So, hey, let's wrap it up, I got to get on another call in a little bit so hey Jim Thanks for jumping on sharing your perspective I think it was a really good conversation. I'll be at a short one. but Hey, everyone, thanks for tuning in. Remember to follow, next curve and curious research, follow the podcast here, which we call a Silicon futures on, YouTube, as well as on buzzsprout, take us on your hike, your daily commute and keep in touch and stay tuned into the tech and industry. Insights that matter until next time, Jim, we'll see you actually. I'm going to see you in Barcelona. So have a safe travel and look forward to catching up over dinner. You too. Thank you very much. All right, take care of everyone. Cheers.