Silicon Futures for December 2024 (with Karl Freund and Jim McGregor)

Jim McGregor of TIRIAS Research and Karl Freund of Cambrian-AI Research joined me to recap December 2024, another action-packed month in the world of semiconductors and accelerated and non-accelerated computing on the neXt Curve reThink Podcast series, Silicon Futures. The trio also shares their highlights from 2024 in a retrospective of the things matter this year.

We parse through the key announcements semiconductor industry and AI headlines of December 2024 and their year-end thoughts:

➡️ We share our thoughts on Pat Gelsinger's departure from Intel (2:18)
➡️ Marvell, AWS and the other GenAI supercomputing market (14.40)
➡️ The impact of the hyperscaler custom chip game on AI (16:35)
➡️ The objective of economical GenAI (19:07)
➡️ The end market red flags of GenAI and the adoption/diffusion dilemma (22:50)
➡️ Jim gives his report on the Arm vs. Qualcomm trial (27:30)
➡️ Does the CPU matter in GenAI supercomputing? (36:50)
➡️ What does the trio expect in the industry and AI in 2025? (38:11)
➡️ Jim throws a monkey wrench into the works at the last minute (45:04)

Hit both Leonard, Jim, and Karl 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.

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.

Happy New Year!

Chapters

• 2:18 We share our thoughts on Pat Gelsinger's departure from Intel
• 14:40 Marvell, AWS and the other GenAI supercomputing market
• 16:35 The impact of the hyperscaler custom chip game on AI
• 19:07 The objective of economical GenAI
• 22:50 The end market red flags of GenAI and the adoption/diffusion dilemma
• 27:30 Jim gives his report on the Arm vs. Qualcomm trial
• 36:50 Does the CPU matter in GenAI supercomputing?
• 38:11 What does the trio expect in the industry and AI in 2025?
• 45:04 Jim throws a monkey wrench into the works at the last minute

Transcript

0:06

Next curve.

Leonard Lee: 0:09

Hey everybody. Uh, this is a Leonard Lee, executive Analyst at Next Curve, and welcome to this next Curve Rethink podcast episode where we break down the latest tech and industry events and happenings into the insights that matter. And in this Silicon Futures episode, we will be talking about. 2024 and a month of December. This is our wrap up episode. And yes, I am joined by the XPU and that's what the small X accelerated call front of Cambrian hyphen AI research and the astronomical Jim McGregor of the famed. Curious research. So gentlemen, welcome. Welcome. How are you doing? Good. Thank you. Good. Happy new year. Happy new year. And Jim you look really purple there, my friend.

Jim McGregor: 1:10

That's the goal. Yeah. Are you trying to make it a pie there? Uh, it is definitely a branding thing. We are the first.

Karl Freund: 1:17

You get my, I don't have my brand shirt on. I got my brand doggy. I got Daisy. Does that count? Yeah, I'll take the

Leonard Lee: 1:27

brand dog any day of the week. You know, nice, fluffy at them, you know, it keeps you company. So yeah, everybody this is the last installment of. Silicon futures for 2024, because yes, we are at the end of the year and we're going to cover what was big. In December, there was a lot of stuff that happened in December. A lot of very pivotal things. I think I would say a crazy way to end the year. And then we'll also provide a broader highlights. John will share the thoughts on what were the things that mattered in 2024 as a whole. So, why don't we get started? And geez, if we're to think about December 2024, we'd have to think about Intel, right? Oh

Jim McGregor: 2:16

yeah, just the departure of Pat Gelsinger and throwing Intel's future and their roadmap in up in the air, depending on, when we have a new CEO and what that new CEO wants to do to what remains despite some of their, setbacks, what remains still one of the most powerful companies in our industry. I mean, that's just huge. And the

Karl Freund: 2:41

role they'll play in the U. S. economy, technology economy with the foundry business is vitally important to the U S global competitiveness and potentially resiliency did something untoward happened in Taiwan. So, I was really surprised. I would have liked, frankly. I can say it. I would have rather Pat fired the board. So the board asking Pat to leave or Pat deciding to leave. We don't know which, but I thought Gelsinger had the right strategy and he was executing and it takes time to turn a ship that big around. And the board just got a I guess they got impatient.

Jim McGregor: 3:19

No, I think there was some investor push there too But still I completely agree with you carl I think you know all of us that were watching and have been watching the industry for a long time knew that it was going To take up to five years for their product turnaround and they're well into that You know with lunar week and with zeon six and everything else that's going on and the foundry strategy was always a 10 year strategy, especially to become a major You Competitive foundry. And they executed on those new process nodes. They're entering production at least with their products on 18a. We don't expect customer products until the middle part of the year. So they are hitting those targets.

3:59

Yeah.

Leonard Lee: 4:00

Yeah, I mean, most part, but I think it was always a really aggressive plan. Right. And it required perfect execution. But then also, I think it also needed patience and support. And I think Pat lost it. Toward the end, and whether it's bad politics, which I think it is just simply because of some of the toxicity that's leaked out as, things transpired with his departure. I mean, it just doesn't look right, but I agree with you. It's like. Wait a minute. 2025 is going to be the pivotal year where we really see the fruits of this, whole strategy, at least start to ramp up and prove its viability and its promised impact. Right? And along the way, I quite honestly, I thought there's going to be some hiccups. Right, definitely hiccups especially when you consider the broader macro and broader industry trends and how the landscape has shifted, especially with generative AI and the data center. Segment. So there's a lot of complexities that were introduced a lot of threats that had to be responded to because, you know, with Lunar Lake, they had to push that stuff up as quickly as possible. There were a lot of things that put pressure on the roadmap and I think smart adjustments were underappreciated. But definitely Q2, right? That particular earnings result was definitely a big shock. And I think It is also sort of an outcome of having split out the financials between product and foundry.

Jim McGregor: 5:46

And also hindsight is 2020. I mean, nobody knew, that the PC market and other segments of the market, we're going to slow down significantly after COVID not to mention, I think if you had that hindsight, we probably would have seen larger cuts in terms of operating expenses earlier on. And that really hurt not making those cuts. And, that's also part of Pat. Pat is really focused on keeping the talent and keeping the people and trying to keep them focused. So, hindsight's 20, 20. I still think he had the right strategy. I still think he was the right person for the job. I think it's going to be a challenge to find somebody that is willing to take the job going forward.

Karl Freund: 6:26

Yeah. I mean, there's lots of rumors of who that might be, but I suspect all the superstars that are on that list would not be interested. Yeah, it's a, it's a high risk proposition. And I don't know what you would change. The board must not have liked something. They may, maybe they had a fallout over spitting out salary services.

Jim McGregor: 6:45

I don't know. Well, I know that they, I know they had a fallout over cuts. Board members wanted deeper cuts earlier on. And in hindsight that would have been good, but however I still think that it's going to be a challenge. I think there might be a opportunity here though. Intel is still, I don't see anybody just acquiring them outright because they're so huge and it really takes time to run a founder. However, There might be some opportunity for some strategic acquisitions on the Intel side or a merger with somebody that might be a better fit. Yeah.

Leonard Lee: 7:22

So my thoughts and I love your reactions to this bullet point of things that I'm going to share with you. I know that there's a lot of interest in finding an external quote, unquote, permanent CEO. MJ is in there right now splitting her splitting the temporary CEO role with Dave, right? But why do they need to look externally? And what's wrong with MJ?

Jim McGregor: 7:48

To appease the board and well, really to appease investors, um, and you have to remember that there are so many parties involved in this. It's not just Intel. It's customers. It's supply chain. It's the governments. especially the EU and the U. S., are investing a lot into Intel going forward, so they have a vested interest. So, I think there's so many outside interests in what happens at Intel, that you almost have to bring somebody in from the outside that is going to have a new perspective to really appease all those parties.

Karl Freund: 8:22

Yeah, I, I think you're right on. I think if you've got somebody from inside, that would imply minimal change. And that's not obviously not what investors want, certainly not what the board wants. So I think they will have to go outside. And I think finding somebody good enough to do that job. Is this willing to take that job? I think it's gonna be tough. It's gonna be a tough sell. Right? I mean, I told him though.

Leonard Lee: 8:50

Yeah, but I think we've spent the earlier part of this particular segment arguing that things shouldn't necessarily change. And we give the road map time, right? And so that's why I'm thinking they need to number 1, they need a customer for our I. F. And the customer is Intel. The person who made the call into whether or not they're going to go internal for Lunar Lake or external, uh, TSMC. At some point it had to be M. J. She's the customer. Go customer

Jim McGregor: 9:21

first. That customer was made two years ago. Yeah, that decision was made two years ago. It's really hard to look at that and use that as a judgment call because they had, Pat was really just gearing up that IDM 2. 0 strategy when Lunar Lake was beginning work, and they had to make that decision. Or reaching a point where they had to make that decision, I should say. That's the hard thing. Is there so many things in timing? You gotta remember that. When Pat inherited Intel, or took over Intel, there were already a lot of things in the works. And some were good, some were bad. He had to weed through that, but it still takes two to three years for those decisions to really start having an impact. So we just really started to see the Pat Gelsinger era. Yeah,

Karl Freund: 10:06

exactly. And so just when his projects are coming to light and actually looking pretty good. They decided to throw the baby out with the bathwater. And if I look at it and say, well, if there's one mistake from a product standpoint that he made, it might have been the Gaudi transition over to a GPU based architecture. That's a very tough transition to make. And Gaudi three sales and Gaudi two sales so that the impact of that, nobody wants to buy a dead end product. it's actually a very good product. And given the supply constraints that the industry is suffering under, you would think, or apparently Intel thought. Then people would look past that and say, don't worry, they'll take care of us. But I don't think they have the trust of the customers to say, don't worry, we'll take care of you in this transition to a GPU based architecture.

Jim McGregor: 10:54

Yeah. Even announcing that transition pretty much shot them in the foot.

Karl Freund: 10:59

Yeah, they shouldn't have announced it, but they should have called it accounting for yeah. That's what I told him anyway, but they didn't.

Leonard Lee: 11:05

Right, but what would you think would have to change? I mean, there's a lot of talk about, hey, we need to bring somebody from the outside to institute some kind of change. What change are we talking about? What would it spend

Karl Freund: 11:20

spending out? I, uh. Yeah, but I don't think you for two to

Jim McGregor: 11:24

three years. I, I, you can't do that until it's at least profitable or at least breaking it. Right,

Karl Freund: 11:29

right. So you wouldn't get what you want in terms of selling it. You wouldn't get the kind of price you want for IF until it can demonstrate that they have turned the corner and they are, you know, going to be a rocket ship.

Leonard Lee: 11:43

But that just sounds like a disaster, right? If that were the strategy was to let spin off, I mean, if this whole thing is about spinning off. At this moment, it would be a disaster,

Jim McGregor: 11:56

right? I would agree. I think it would be a disaster right now. I think no matter what, you have to have somebody that's going to come in with a vision, a vision for what Intel is. And that's what Pat had. He had a vision for Intel. Unfortunately, they didn't have the time he needed to really execute that vision. This kind of goes back to something I've been saying for quite some time now, uh, there's, everyone's looking at AI and AI is a thing and Intel kind of has missed the boat so far on AI, at least on the data center side so far.

12:26

Oh yeah, yeah.

Jim McGregor: 12:26

So to look at that and say, okay, well, Intel needs to be an AI company. What does that mean? Companies are looking at how they use AI to make themselves more efficient, and they're also looking at how AI changes their products and services. But very few companies are taking that third step and saying, how's AI going to change my business? What does my business look like once AI enters the market over the next decade? I asked that of Jensen Hong at NVIDIA. And he looked at me dead in the face and said, Jim, I have no idea. Yeah. Yeah. And then that's actually a good answer. Cause he's always looking a lot of companies aren't looking. So just to say that you need to be in AI is not enough. You need to be able to look up and say, what is my company going to be in this AI era? And it may be completely different. It may be more services. We don't know. We really don't know. I think the automotive market segment is a good example of that. The traditional value chains being blown apart when you start looking at autonomy, some of the tier one or some of the OEMs are developing their own technology. They're going to be offering cars as a service, they're even looking at insurance. All of a sudden that whole value chain is kind of shot. You even have some of the major tier ones like Magnet International that's actually designing and producing cars for the OEMs. So, it's really going to be difficult to really look at the next 10 years and not question what your company's going to look like.

Karl Freund: 13:53

I'd like to see Intel ramp up their services with the goal being now in generate rep for high profit revenue, like Marvell and Broadcom are enjoying for AI customized AI chips but also to help people get the most out of IF. So if you combine those two thoughts, I can see a very nice business or potentially an acquisition that would get them into that space is, as a really net new revenue for them.

Jim McGregor: 14:22

And with the leadership in process technology and their leadership in packaging technology, I think they'd have a huge benefit in doing those customized chips. That's always been part of the plan, but once again, they haven't reached that part yet. Yeah. Yeah,

Leonard Lee: 14:38

and it is interesting that with Marvell Industry Analyst Day, what we saw is the, and Jim, you and I, I think, is it Jim or no, Carl, you and I had a segment on custom AI chips, right? And how that's probably going to be that next opportunity. And it's the impression that you got when you were there at the Marvell industry analyst day, right? Let's be honest, the reason why everyone's going bonkers about AI is because a generative AI and 1 of the interesting things that, um. What we're seeing is this divided generative AI supercomputing universe. 1 side is for operational AI. This is where you're seeing Amazon and meta and Google and the like. Building their own custom systems for their custom enterprise operational needs. Right? Because these are the companies that are going to be deploying generative AI and its derivatives at some large scale. Right? And it's probably going to be mostly for like recommender engines or enhancing them or some of this agentic stuff that meta is talking about. And then you have the other side, which is. The stuff that is catered to model building where the GPU with the software defined layer and all the thing benefits that Jensen talks about of being programmable and extensible that's like a kid is at the supercomputing level to a different market, which is model building. And some of the experimental stuff that's happening around the enterprise. Yeah. And one of the things that impressed me in the latter part of this year is this emergence of that operational generative AI supercomputing side of the story.

Karl Freund: 16:33

Well, I think the, in a similar vein, one of the things that impressed me in 2024 was how well. The cloud service providers have improved their ability to make competitive silicon.

16:46

Yeah,

Karl Freund: 16:47

so what that means, and if you look at Trillium from Google, if you look at Trillium 2 from AWS announced in

Leonard Lee: 16:55

December, right?

Karl Freund: 16:57

Yeah, announced in December, you say, well, okay, let me think about this. So all these guys are building these massive supercomputers for AI that nobody else can afford to build, by the way except for potentially governments. But in terms of commercial enterprises, nobody else can afford to build these models. that means the market. For semiconductors being sold into AI building applications is really the cloud service provider. So now your cloud service provider can say, well, okay, I've got NVIDIA. That's great. That's what the market wants. I'll keep buying tons of NVIDIA. I've got my own chip, which I kind of like. In fact, I'm telling all my engineers to use it and optimize for it. So that's good. Why do I need a third chip? And if you go down that route, then you say, well, is there room for AMD and Intel back to this? Is there really room there? And I'm starting to think I haven't concluded, but I'm starting to think the answer is no. And that may be one of the reasons why AMD's excitement around the MI 300, 325, 350 has ebbed a little bit, because the market's going away from them, the market's not really attracted. If you want to build your own. You could do that, right? With Marvell and Broadcom and Tenstorm, right? You could build your own with chiplet technology that maybe will solve your need at a better price point than acquiring semiconductors from somebody else. So I don't know.

Jim McGregor: 18:34

I also remember that a MD makes a very good business doing custom chips

Karl Freund: 18:39

for other markets. Yeah. Yeah. And I think that they're likely to move heavily into the automotive space, uh, which would be a natural place for them to go with their strengths and cpu, their strengthened GPUs, strengthen and FPGAs. Boy, that would just be a rocket ship for them. They probably got a better story to tell there than they're telling.

Jim McGregor: 19:01

Well, I think we're still in this trend of, or I would say the beginning of making AI more economical. So everything is still so expensive, whether you're using a custom chip or whether you're using an off the shelf chip, it's still very expensive to do AI. We're improving efficiency by leaps and bounds, but it's nowhere near where it needs to be to really be able to make it be able to roll it or roll it out across enterprises, much less make it adaptable to consumer applications. I think we're still a couple of years away before we really reach that efficiency point. And both Jensen and Pat had made statements that the new unit of compute is the data center. And I completely agree with that. You have to start Developing from the building level on down to improve efficiency, and there's a thousand different things you can do from, the type of infrastructure you put in from the network to the cooling systems to the rack design, NVIDIA's and VL 72 was a huge leap forward this year, and they opened it up through open compute so that anybody can actually use that design and modify that design. Matter of fact, Matt has already done it. So it's a huge leap forward. Well,

Karl Freund: 20:19

meanwhile, the software is demanding more performance, which is kind of surprising. I thought, I thought by the end of this year, we'd start to see the efficiency curve kick in. But open AI's of three is. Everything I'm reading, I'm like, I don't know how you conclude this is not artificial general intelligence for certain segments, not for everything. Maybe it's semi general intelligence, but it's really hard technology and it requires 10 times amount of compute.

Jim McGregor: 20:48

Yes.

Karl Freund: 20:49

And yeah. Then chat GPT is using today. Yeah, more compute.

Jim McGregor: 20:54

That's probably the most impressive thing I think for 2024 is the race by the hyperscalers or by these major AI companies just to give stuff away. And the reason is because they want to be the de facto standard. They want to have the models that you use to recreate your models, or to use in your application. Whether it's the huge models, hundreds of millions or billions or trillions of parameters. Or it's the, couple billion parameter models that you can fit on pretty much anything. But they are racing forward. I mean, Meta. What can you say about Meta with Llama? They're racing forward just as fast as OpenAI is.

Karl Freund: 21:35

And IBM with their latest granite is exactly a small model that outperforms large models. Yeah. And of course, that's gonna fuel a lot of opportunity for IBM consulting go in there and help people modify and optimize for those for specific business purposes.

21:51

Yeah.

Karl Freund: 21:52

So I think you're right, Jim. And you come back to the silicon aspect of it and say, okay, where's all this going to run?

Jim McGregor: 21:57

First tier and second tier. Huge question. It's a huge question. And in 2024 was the year of the AI PC, but we don't really have a lot of things that run on the PC from an AI perspective. It's mostly in the cloud. But I mean,

Leonard Lee: 22:12

guys, I have to be honest with you. Those are red flags, right? You do have all these massive amounts of investments being made and I think two in probably three buckets, right? Number one, the experimental model building supercomputing. The other is the hyperscaler operational stuff for some ridiculous workloads. That most companies don't have. I don't know how much of that will be able to be scaled down like you're suggesting and brought to the edge. And then there's the edge stuff, right? The things that are related to PCs and even smaller smartphones, right? And there are a lot of red flags. Quite honestly, because if you look above the stack monetization of models themselves, very, very difficult. I don't think anyone's cracked the code on that number two, even at the services level. So you go up a layer monetization of services is actually very challenging with Satya Nadella on a podcast recently, really asking the question, who are those service providers that. then X number of it was actually a pretty small number of daily active users. And he's miffed by it. He himself and so there's a lot of red flags. Then there's still that big question of how quickly can the. Return part of the ROI equation for the ecosystem be accelerated because the POCs are not proving out, exponential value that will pay for all the investments happening in technology and infrastructure. I think, as analysts, those are the things that we hopefully can bring insight to so that these, the ecosystem can start anticipating the problems to be solved because these are pretty big problems to solve.

Jim McGregor: 24:05

Yeah, but we've seen this happen time and time again with the tech industry. Come on. I mean, we saw it with the, uh, bust around the turn of the the century. We saw it with 3d TV, man, that was a bust. We saw it with the internet where, you had a whole bunch of companies rushing in and no business models to make money initially. So I think it's just the nature of innovation in our industry. Everyone rushes ahead. And you just have to see, what filters out. Eventually, yeah, eventually a couple of the major players stay up there. A couple of the major players fall off the map. Yeah. And some of'em, if they're lucky, just kind of find a new middle ground. There's always new companies that rise at the top. They to Google expect.

Leonard Lee: 24:51

Right. They become the Google'cause. Everyone thought that Yahoo was gonna be the king of the Hill for.

Jim McGregor: 24:57

Yeah, but where was Google when it came to social networking? They're in a lab. Yeah, exactly. So, I mean, every company. Even Google has had its ups and downs with some of these transitions. So, it's going to be interesting. Come on, how long are we going to wait for an Apple car? Oh, forever. You asked.

Leonard Lee: 25:20

Okay. And you used the bad word bust. Come on, Jim. Can't say bus.

Jim McGregor: 25:25

It, it, it happens. A couple people have mentioned these waves or these super cycles. Jensen did it. Victor Payne did it when he was talking at Hot chips saying, listen, we have several cycles we're gonna go through with ai and I, I really agree with this, this kind of. Mindset. First it's really about the data center and then it's about the enterprise and then it's about the consumer. And now those overlap, but it's going to take, that's probably a decade to look at that period at least, but so it's going to break down.

Leonard Lee: 25:55

I want to highlight something. Um, yeah, I actually reared its head in. Smartphones in terms of commercial application, broad commercial application before any anything else. Right? And nobody noticed.

Jim McGregor: 26:09

Yeah, I know. Yeah, it was used for a kind of computer photography. It's been used for battery management. It's been used for. So, yeah, it's been used for several generations in smartphones to improve the user experience, but it wasn't really necessarily Something that the consumer latched on to.

Karl Freund: 26:27

Nor is it something the consumer would pay for, right? It was Samsung versus Apple versus all the other manufacturers. And They're going to make money whether you use that AI agent or on that device or not, that's very different than what we're seeing today with the data center, where people are spending a ton of money, hundreds of billions of dollars in the next couple of years. To outfit the data center with AI capabilities now. So the question becomes, okay, how is that paid for? Where's the monetization of all those investments? And how's that going to play out? And I don't think we know yet. I think that's why I think we're still in the picks and shovels. Phase of growing AI and that's why NVIDIA has done so well as that's why AMD and now, especially Intel are struggling because nobody wants to pick some shovels.

Leonard Lee: 27:23

Now, hey, let's move on to the other big thing that happened in December, which is this. This thing that happened in Delaware and I don't know, Jim, there was something in Delaware,

Jim McGregor: 27:35

something happened in Delaware. I don't, I don't think you're referring to the, you're referring to the long anticipated, it took two years to litigate and finally get to court the arm versus Qualcomm trial arm, the IP company sued its largest licensee Qualcomm over the acquisition of Nuvia. And the use of NUVIA technology. And I'll be honest with you, I read through the pre court proceedings, and everything else leading up to this. What was presented in court didn't match that. And then the questions that the jury had to answer were completely unexpected from that. And I was just, it was so confusing. But really if I can simplify it a little bit, what it really came down to was, Nuvia was a startup. They were given what is a golden ticket by ARM, and that is an architecture license agreement where they can develop their own micro architecture and use the ARM instruction set architecture on top of it and go after the server market. And Pretty much everyone knew it's a startup. There's a good potential they're going to get acquired. Well, they did get acquired. They got acquired by Qualcomm. Although Qualcomm really didn't want to use the technology for servers, they wanted to use it to really go after the traditional markets, which are smartphones and going into automotive and XR and all these other markets they're trying to expand into. And really, competing against what Apple and others have done with their architecture licenses. Now the question comes in of when obviously Qualcomm already had an architecture license and a very Broad one that pretty much covered anything they wanted to do because they've developed their own CPUs in the past And it extends out fully extended to 2033 with very reasonable royalty rates Well Nubia being a startup They said, listen, if you help us out a little bit, you'll get more royalties on the back end. Well, it turns out they didn't really help them out too much on the front end. They did pay quite an extensive license fee and they were going to pay higher royalty rates. ARM wanted Qualcomm to assume those royalty rates. Qualcomm said, we don't need to, we have our own architecture license. And then ARM said, well, you're assuming ARM IP. And they're like, well, not really. We're assuming a micro architecture. And that's really where the debate came in, is what technology was Nuvia's, what technology was ARM's, and what's the responsibility of Qualcomm? There were three answers that the jury was given, or three questions the jury was given. One, did Nuvia violate their architectural license agreement with ARM? The second question was, Did Qualcomm violate the Nubia agreement, which is kind of ludicrous, since they were never a party to that agreement. And they weren't even allowed to see the agreement until they acquired the company. And third this is the countersuit by Qualcomm, Are the products that Qualcomm has developed using both their technology and using the Nuvia technology. Covered under the Qualcomm ALA architectural license agreement? Well, the jury couldn't decide on whether Nuvia violated or not. That was the first question. And I'm not surprised because Nuvia added some terminology in, ARM added some terminology into an agreement that got really confusing. And we never saw the entire agreement, even though I was in the courtroom, they picked and choose different parts of it. And it was obvious this was a bad contract. It was not very well worded. There was a lot of confusion around it. So if the parties couldn't agree to what it meant, I'm not surprised the jury couldn't agree to what it meant. The second question, did Qualcomm violate the Nubia agreement? No. And did Qualcomm's technology or new products covered under their ALA? Yes. So for the most part, it was a win for Qualcomm and also some of the things, and this is the scary part about going to trial with another company is you always know things are going to come out. You don't expect or that, that are proprietary. And mud's going to come out. Well, the mud was kind of on the arm side, it was pretty clear that they had a strategy to try to push Qualcomm to higher royalty rates, not to mention the rest of the industry. Qualcomm was kind of, you know, best practices, you know, and Qualcomm, has been accused of being a patent bully in the past. they operated, with good intentions and there was nothing to indicate that they didn't operate with the best of intentions throughout the trial. So it was a little bit of, mud on the face of arm and they have to deal with that, especially as they're interacting with their customers going forward.

Karl Freund: 32:19

It basically reinforces the impression that Arm is simply struggling to be competitive. And it took Nuvia engineers, they actually in court talked about how the Nuvia engineers were trying to help the Arm engineers improve their architecture. It's innovative or die. And if arm doesn't figure out how to innovate and you're not going to innovate by spending all this money. Suing your largest customer and had they won. Can you imagine how ridiculous that would be? Oh, yeah. Okay. We're gonna shut this down because it's too good.

Jim McGregor: 32:54

All right, and it would have basically killed the AIPC, based on Windows on ARM. Right, so there goes

Karl Freund: 32:59

opportunity for ARM, goes out the window.

Jim McGregor: 33:01

So

Karl Freund: 33:01

it's just a poorly conceived strategy and poorly executed.

Jim McGregor: 33:07

Well, and it really brought to light one of the key issues. ARM has kind of a hole in its business model. It offers two types of licenses. A TLA license, where somebody uses their standard course, technology license agreement. And then they have the ALA, where somebody can develop their own stuff and just make it ARM compatible. And problem is they always knew that there was a potential for competition there. If somebody's licensing your IP and you're hard cores, you're going to get a higher license royalty fee. If all of a sudden they decide like Apple did, which is probably the best example to go to their own custom cores, that license rates going to go down. And you got to remember that the companies that they're selling to typically have higher R& D budgets and much, much higher, I mean, multiple times higher revenue than Arm itself. So it's a challenge. And I think that's one reason why it would have been great if Arm was part of a larger entity, whether it was NVIDIA, which that deal didn't go through or somebody else, SoftBank hasn't really, you know, Effectively leveraged it, unfortunately, but it really exposed something that's been developing for the past decade. And that is that ARM created a business model that's going to compete against itself with that ALA. And going forward, we've seen this, especially in the server environment where most of the major people going after servers with the ARM architecture have used, have developed their own micro architecture. And we've seen this a couple of times where they basically said, listen, we needed something more customized and we could develop it better than arm could.

Karl Freund: 34:38

Meanwhile, the attention and dollars have all shifted to AI arm has really struggled to provide a decent product. Technology. Yeah, they got GPUs, but nobody's using them for AI. So are they going to deselect AI? I think you deselect AI, you're dead. There's no future without AI. So where's that going to come from?

Leonard Lee: 35:01

well, they're in road or at least their channel or in tapping into the opportunities on the CPU core side of things. Right? I mean, it's the, and Neo verse and these other frameworks that are being leveraged by NVIDIA and others. That's really where that linkage. Is how you qualify that in terms of, a business and what that has in terms of implications on the top and bottom line. Those are things that, if you guys have insights on that, then, you know, I think that'd be really interesting thing to analyze and understand, Yeah, on the accelerator side, definitely it will show up on the map, right? But, for every two Blackwell, there's a Grace sitting in the middle, right? So,

Jim McGregor: 35:49

but also remember that NVIDIA itself has also developed its own custom ARM compatible. Process CPUs in the past, not, it's not out of the realm to think that if they have a lot of success with grace, that they wouldn't do the same thing, especially if they get to a point where they think they can optimize it.

Leonard Lee: 36:09

Well, haven't they? You know, great. A lot of people don't talk about grace because her and Blackwell have stolen the headlines. But I mean, think about the implications of grace itself, right?

Karl Freund: 36:23

CPU, it's the only CPU in the world that speaks NVLink. And that's, that's all it has to do to fit in with NVIDIA's strategy. Now, could they do that with their own ARM architecture? Absolutely. Could they do it with RISC V? You bet they could. Well, we haven't, we haven't seen how this movie ends.

Jim McGregor: 36:41

Well, and we've seen from all of the major players in the data center segment say that, listen, the CPU does matter in AI workloads. If you pair the right CPU, you improve the efficiency of the CPU, you're going to get better AI performance because some of the tasks you're gonna be running on the CPU, the CPU's gonna make the entire system more efficient. Even the messaging we've seen from Intel and we've seen from AMD plays well for race as well.

Leonard Lee: 37:11

Yeah. And so like we've been saying throughout the course of this year, it's about the system and you know, just having a myopic view on, let's say the accelerator versus interconnect versus networking in front and backend and ignoring the CPU. All of a sudden, you just don't see the big picture, which is, I mean, it's about data center this year. The unit of compute we're talking about here is in the data center data

Karl Freund: 37:39

center,

Leonard Lee: 37:40

and that's

Karl Freund: 37:40

why Andy acquired ZT systems. To give them the skills need to build their data centers and pin sendo. Yeah,

Leonard Lee: 37:50

yeah. Yeah. Wow. What a crazy year, huh? It is a crazy

Jim McGregor: 37:54

year. Yeah. So, yeah. And I think Carl will completely agree with this statement. When it comes to ai, we're still learning how to learn, so it's still changing.

Karl Freund: 38:05

Yeah, I think they're looking to 2025. We'll see the focus shift more towards the debate of artificial general intelligence. Oh, 3 really predicts that by offering something that's. Pretty close. I mean, it's as good as PhDs in mathematics. It's interesting, there's a, there's an article today that came out. Gary Marcus came out with a bet 10 to one bet that will see a GI by 2027. He says, no. And the other guy, I can't remember his name, I apologize. An ex, open AI guy says, yes, we will. And the way they defined it was 10 tests. You'll have to pass. Three of those are ridiculous. Like, okay, you gotta write a novel that is a Pulitzer Prize candidate. You've gotta write a screenplay that is a Oscar level screenplay. And let's see, what was the other one you have to write? That's not artificial general intelligence. That's artificial super intelligence, right? I don't know any. Nobel laureates. It doesn't mean that doesn't mean the people I know are stupid. It's just they're not super intelligent.

Jim McGregor: 39:14

Yeah, I'm I try to stay out of that debate just because. Yeah. Defining what it is, is hard. But do I think in performing specific tasks, it will be as good or better than humans? I think we're there on a couple and I think we're going to, especially 2025, I agree with you. I think there's going to be so many of those tasks that, AI can do as good or better than a human that you have to pay attention to it. You have to utilize it where you can. It's going to be a critical technology. And once again, the transformation of AI is very similar to, I tell people take the impact of the PC and the impact of the internet and combine that, and you're still not close to the changes you're going to see with AI.

Karl Freund: 39:59

Yeah, I think you're right. I think you're right. It's 2025 is going to be a fun year to watch.

Jim McGregor: 40:04

I would agree. I think 2025 is the year of the agentic AI, agentic, where we see more and more usable Agents that really obscure. And it's funny because, you know, first we had the internet and then we had apps that obscured the internet. Now we have AI that obscures the apps, obscures the data, even obscures the hardware platform. So, you know, we're moving to another level of obscurity. And I really think that agentic AI gets us there. And I think we're going to see that this year. I also think we're gonna see start of that, really that third wave or that third supercycle where we start seeing AI creep into more and more of those edge devices. Now, when I say ai, I'm not just saying Edge, I'm, I don't, a lot of those applications, they're never going to use, generative AI generat.

Karl Freund: 40:54

Yeah. But

Jim McGregor: 40:55

traditionally, I still going to be critical for a lot of those applications. And I think that's, I think we're going to see much more of that. So I'm excited, not just about what we're going to see at CES, but I'm excited about things we're going to see at Embedded World and Mobile World Congress and some of these other events throughout the year where it's a much broader scope. CES is that Eureka Park is probably going to be the most exciting place to be. Yeah.

Leonard Lee: 41:20

Yeah. Hopefully we won't see too many. Animatronic devices hooked up to a generative. I accidentally smacking people. Um, that that's kind of like, well, I saw a

Jim McGregor: 41:35

good sign a couple of years ago. I kept running into Stevie Wonder everywhere. I went around Eureka Park. So I. Kind of felt like he was following me, even though I know that's not feasible. Um, I do think robotics is a key part of that, especially the humanoid robot and we saw the humanoid robots really start coming out mass in 2024. But they're not really that functional yet, though, I think 2025, we're going to start seeing some really practical applications. Yes, but

Leonard Lee: 42:06

I have seen several actually hit people. So, yes, um, uh, you know, make progress this year. Did any of them hit you? Leonard, no, no, I know enough to stay away. At least I got to improve my programming skills beyond slap. Yes, obviously, the guardrails weren't in place with those demos. So here's my take. I don't know. You guys see, I always get nervous sharing my thoughts with you guys because you always call me a hater or some crap like that. But anyway, it's going to be a desperate year. Like, what we're talking about earlier all of this investment needs to find and market return. And I think it's going to have a lot slower than people anticipated. And definitely for the expectations that were set. 2 years ago, the challenge for 2025 is going to be doing exactly what you, you mentioned, Jim is helping enterprise figure out how to find a value, helping leaders to apply that filter that will get them to value. But I think this is the more important thing. It's not about intelligence. It's about how these things are tools, because one of the things that historically we've seen with AI in general is that it is an augmenting technology is not it typically is not core, but we're now led to believe that is core, despite what people have admonished us about that. That this is the core technology. And I think. That's 1 of the things that is going to be really driven home in 2025 is that it is not core. It is indeed probably less than a copilot. It's something that is a tool, and you need to figure out the fit for purpose scenarios where it exhibits reliable and sustainable value. And I don't think people have figured out that equation yet. They don't even know about the equation yet. They certainly haven't solved that problem for their business. And until they do that, I think, Jim we're not going to see companies figure out how AI is going to transform their business and how it's going to reshape things. I think, until these filters are really developed, which is going to be tough because it's going to be a dynamic filter, like you said, it's constantly changing. Right. But the big challenge is how do I avoid making bad investments? And this stuff that are not going to improve my business, but is going to hydrate some applications at a and service providers a certain moment that are going to prove to be like, what you said part of that bust. The cycle, right? The super cycle. I think I'm hopeful just because I think these are great problems for analysts to help the industry understand and accelerate the path to value. And so I'm looking forward to working with you guys in 2025.

Jim McGregor: 45:03

I wanna throw a monkey wrench in two yours, because I'm very, no, I just, I I I have to, you know, it, it's me, Jim, uh, well first off, I'm still very optimistic about where we're going and the innovation level we're seeing, but I'm gonna put one caveat out there. Please do Rare, rare earth materials. I think that is the biggest concern. It goes all the way back to our political situation right now that we have. I'm almost surprised that China hasn't put more restrictions on rare earth materials sooner because it is a wrench that they can turn that we can't get through. And if anybody lived through the early 2000s in our industry, they remember the impact of the shortage of tantalum. Tantalum capacitors, just little capacitors, really froze up our market for quite a while because we didn't have enough tantalum. And the same thing can happen with the rest of the industry if we don't have rare earth materials.

Leonard Lee: 46:01

Yeah, there's a simple answer to that. The long game and that's what should concern everyone.

Jim McGregor: 46:06

Carl, what do you think?

Karl Freund: 46:09

Yeah, I'm more optimistic on the return on investment and debate. You and I've had quite a few times,

Leonard Lee: 46:18

which I, which I love, by the way, I appreciate

Karl Freund: 46:21

great. It's great. I just look at what IBM has been able to do with their customers. Their clients and the kind of return on investment they've been able to make. That's not all with generative AI, but it is using AI technology to solve real business problems and give a return on investment today. And so I think that gives us a solid foundation for 2025. And as we add more generative AI applications as portability continues to improve and the intelligence of our intelligent friends continues to improve. I, I think we have a very bright future and yeah there's issues. I think the rare earth out rare earth. The issue that Jim brought up is an important one for us to solve as an industry and as a country.

Leonard Lee: 47:09

Well, this is all I know at the intersect of our three perspectives here, there's a better perspective. So that's what is a hopeful thing for me going into 2025. And so with that, gentlemen, thank you so much. I hope you have a new year. Happy new year. Holiday season, happy new year. Look forward to continuing our collaborations into 2025 and why don't you both take a moment to share with the audience, how they can get in touch with you and the new year and where you would like for them to Soak up all this wonderful insight and perspective and knowledge about the industry. Where should they go? Jim, why don't you start?

Jim McGregor: 47:51

Okay, definitely reach out to myself and my colleagues at teresa research dot com. Our. Emails are easy. It's just first name at turiusresearch. com. I'm a Jim at turiusresearch. com. And, definitely look out for the multitude including, what we're doing here in X Curve, but other channels that we leverage and we work with, including EETimes, Forbes. com, ECT News, Microelectronics in Taiwan. A multitude of other media channels that we'll be in partnering with throughout 2025.

Karl Freund: 48:29

And likewise, you can reach me at carl. freud. cameron ai. com. Visit our website to see our latest thinking on what's going on in the AI space. And likewise, through Forbes and EE Times. And if you have a question, give us a call.

Jim McGregor: 48:47

And right here, right here and right here, although we have to add something in here for 2025. We have to add a, we have to add sound effects. Okay. Alright. So we don't agree with somebody with something that's being said. We can do a slap or we can do a Boeing or we can do something. We need

Karl Freund: 49:07

framers

Leonard Lee: 49:10

Yes. To figure that out. Maybe that'll become a Zoom feature. Right? You'll have to, there you go. Yeah. You know, um, virtual, yeah. Gentlemen, appreciate you both so much. And this is a great way of wrapping up the year and having both of you share your perspective with the next curve audience. And so everyone, thanks for tuning in. And remember to like, share, comment on this episode and subscribe to the Rethink podcast here on YouTube and on Buzzsprout there, you can get the audio version of our podcast and take us on the road. On your jog and you can access us through your favorite podcast platform and also check out and connect with these gentlemen as they've mentioned how they would like for you to engage with them and subscribe to the next curve research portal at www dot next dash curve dot com for the tech and industry. Insights that matter. Happy new year. Happy

50:17

new year.