Silicon Futures - Analyst takes on Intel 14A, TSMC blowout, America's AI Action Plan, and more!

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

In the month of July, Mario Morales, GVP of Semiconductors and more at IDC, joins Leonard, Karl, and Jim talk about their key takes on Intel's earning and their announcements regarding the prospects of 14A and 18A in determining the future of Intel Foundry, and a number of other hot topics and news items running up to the end of July 2025.

Topics that mattered in the AI and semiconductor universe in July of 2025:

• Intel's earnings results and the future of the company, Intel Foundry, and 14A.
• TSMC's blowout quarter and what it means for the industry at large.
• The Trump Administration's America's AI Action Plan - what does it mean?
• The state of China's AI industry and future.
• Micron and their SOCAMM deal with NVIDIA - implications.
• Consolidation and new industry structure - GlobalFoundries acquisition of MIPS, Synopsys acquisition of Ansys.
• Why Apple's AI strategy is a dog. ABN (Anything But NVIDIA).

Hit Leonard, Karl, Mario, 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 Mario and his research at IDC at www.idc.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

Next curve.

Leonard Lee: 0:09

Hey everyone. Welcome to this next Curve Rethink podcast episode where we break down the latest tech and industry event and happenings in the world, uh, semiconductors and Carl's favorite topic, AI into the insights that matter. I'm Leonard Lee, executive Analyst at Next Curve. And in this, um, Silicon Futures episode, we will be talking about all kinds of stuff that's been happening in the month of July. We're still in July, but the headlines still keep coming at us. no doubt. We're gonna try to touch on A SML. The tariffs on the eu, probably talk about TSMC blowout quarter. There was that, America's AI action plan, micron socom. you know, Carl wants to talk about China. The list goes on and on and on, and it includes Intel and the. The fate of Foundry and 14 a I bet you a million bucks. One of you guys wants to talk about that. So those are the things that we're gonna be covering. And I'm joined by the always peak loaded call Fre of Cambrian hyphen AI research and, the overclocked and running hot that rhyme. Jim, Jim McGregor of famed cheerious research and oh my gosh. Special, special guest, and hopefully permanent resident of Silicon Futures. Mario Morales of IDC. And if you don't know who Mario is, then you know nothing about the semiconductor industry. Gentlemen, how's it going?

Jim McGregor: 1:44

good. But since none of my, topics were on that list. Do I get a million bucks from you?

Leonard Lee: 1:52

Really? Oh, I'm, so, I'm really interested in, oh, I am just teasing. I'm teasing. Want there buddy? Alright. Okay, so, hey guys, let's get this started. I mean, where do we want to start? There's just so much stuff. Mario,

Karl Freund: 2:06

the latest news is Intel. You wanna start there?

Leonard Lee: 2:08

So who wants to make the first comment? Mario, come on.

Mario Morales: 2:12

Intel. What can we say? It's still. a lot of work to be done. You heard the numbers. At a top line level, their numbers looked, pretty solid, right? But when you start digging deep into the margins, what's most concerning to me is the debt structure, right? The company now has over$50 billion worth of debt. Their foundry business is losing probably over$10 billion this year. That's up from last year. So that's a considerable concern. And also the unfortunate part of all of this is that the lack of execution of the company has, is now gonna cost 25,000 people. they're not only their jobs, but their livelihoods. And so a lot of times the employees are the ones that bear the brunt of all of this. But it really comes down to the management teams, in the organization that just made bad decisions. And one of'em was excessively investing in an industry that they needed to. really understand more. and it's very, the foundry industry is a very different industry. It's not about manufacturing alone. it's a service industry. And then you have to be able to manage hundreds of customers at a time, a very complex supply chain. And these are things that Intel is still gradually learning. Right. And it's gonna take some time. I think people are, were spooked by the point, and it, it makes sense, right? But they were spooked by the point that Intel talked about the potential of not moving, 14 angstrom forward, especially if they can't find the customer. And I'm more concerned about what happens to 18. A i, I know that's gonna be the strategic focal point for the company, but it's still an area where I think we need to see more success before we at least hail it as a first turnaround for the company.

Leonard Lee: 3:58

Yeah. No, that's a great comment. And I've always thought that it was important for them to. make traction internally, with their foundry services, right? The internal customer being, key. And as you guys know, that's still a work in progress. But gentlemen, any thoughts? I'm sure you have tons, but top of mind, top insights.

Karl Freund: 4:21

I'd say Intel's problem is not, they made bad decisions. They did, they made easy decisions, They weren't willing to make the hard decisions. What are you not going to do? what are you gonna double down on? now lip Bhutan is saying, yeah, we're gonna double down on ai, which is the right thing to do. AI's becoming pervasive. but they've tried it three times, right? They tried it with many Core, then they tried it with Nirvana, then they tried with Habana Labs. Now they can come back to their own GPU. so their fourth architecture, you gotta wonder, does the market have a patience for that? we'll see. They've already said that their first generation of that, was not gonna be competitive, so they canceled it. It'll be a development platform, so that pushes it way out. And meanwhile, NVIDIA's just taking off like a rocket and AMD's run as fast as they can to catch up a little bit. But, I'm not sure there's room for a third.

Jim McGregor: 5:16

I would kinda like an intel where it was when Paul Alini took over, they were behind the curve on 64 bit, a MD and Microsoft had taken the lead on that. and they were facing stiffer competition. They're very much in that same position now and failing in some of their product areas. So they're in the same position now and Lipan is following a very similar playbook that Paul Delaney did. Scale everything back to the core competencies and the core products. Intel gave up on the communications sector and a lot of other sectors, automotive, which they've given up on again. and a lot of segments that, they aren't profitable in, they don't have a market leadership position in. from a survival strategy, I think that's great. it does have long-term implications on the growth opportunities, but let's face it, Intel does have to secure the markets it's in before it does anything else. So I completely agree with that, and it has to be a player in ai. If it's not, then they become one of these technology providers that fades into the background during one of these major market transitions, one of these inflection points. so they need to overcome that. big concern, some of the statements coming out of Intel, Bhutan's going to approve every design win, or every chip design before it goes to tape out. that's a huge challenge. But, obviously, Intel needs more structure and needs to be going, focusing on profitability and focusing on the products where it needs to win. So hopefully we see that going forward. I

Mario Morales: 6:49

would agree, Jim, I think that you, you can't compare it to Paul Otto's, uh, time. I think the difference though is that there isn't the top line growth that Paul had, right? So if you look at the tenure of Paul, they were still able to manage a considerable amount of growth because the server part of their business really began to grow in that space, and Zion did so well. Yes. and so that helped them kind of mask some of the weaknesses that they were experiencing, other parts of the business, not being in mobile and so forth. Now, today, it's completely different. You've missed the largest opportunity. even though you're showing little pieces of growth in the market, you're not growing at 20% the way the market is, are growing even faster because you're not In ai. And that's the challenge now, right? As you're pruning more and more, you're still trying to find that discipline that you need. you're missing the top line growth. and that's what I don't see within Intel now, is that top line growth is gonna be a challenge that's going to eat into their resources that they have today.

Karl Freund: 7:49

I think that's, is absolutely a key point. if you look at, the IDC, server market revenue tracker, you know, it was shocking. And it doesn't get the attention perhaps that it used to, but, it showed that the, X 86 servers are growing at 60% in fourth quarter last year, while non X 86 servers, everything Intel's not focused on increased 260%. It's just phenomenal. A big, big chunk of that is going to arm and a big chunk of that is going to Nvidia Grace, And, there's a chunk of that going to, other platforms like I, IBM, m LinuxONE and others. But Grace has got probably, we don't have the numbers behind that. You have to remember those are

Jim McGregor: 8:37

different perspectives though. Totally.

Mario Morales: 8:40

to a certain point, Carl, I agree that there's faster growth to be had that they're missing out on. But a lot of the growth from Marm has not been from Merchant Silicon. it's been all captive, right. This is why, and Pier was sold off and they just could not create a. Merchant business most of the other companies, including Broadcom and Qualcomm's initial take into the space, they all could not grow merchant business. And I think a lot of the business for ARM will be internal. It'll be very custom. the only exception will be Nvidia because Nvidia has been able to pull in grace with, uh, Blackwell. And so that's allowed them some penetration if you look at Grace from the beginning, it was mostly focused on supercomputers. That's where they've had. Some traction, right? But now it's broadening because the ramp of Blackwell has been so fast but I still believe that NVIDIA and AI servers still needs X 86. You're still seeing the traction. That's where I think Intel had a bright spot where they talked about some of the wins that they've had there. And I think that will continue. I think ARM will become more interesting as we start thinking about sovereign AI and we start thinking about players like Open AI and SoftBank and all these other portfolio companies that SoftBank has. They're all trying to maneuver and start trying to think about how do I use the asset of ARM within my portfolio of companies? And I think that's where I think it's gonna get a lot more interesting. And of course, Amazon. Google and Microsoft, they're all using more and more arm as well, so we'll see how that plays out over time.

Jim McGregor: 10:08

Regarding Intel though, you have to remember that there's three core competencies of that company. One is the IA processor architecture, one is r and d, which without Intel investing in our industry, we wouldn't be anywhere where we are today. They have been a huge, huge, everything from process technology to materials technology to, transistor design. We wouldn't be where we are today without Intel. So the industry needs intel from that perspective. And the third one is their actual process, technology and packaging technology, their manufacturing, arm. They're definitely a leader in packaging and they're actually making money in the Foundry business doing packaging for third parties. the struggle is becoming a foundry. we've seen other companies struggle with that. Global Foundry struggled at that for over, over a decade. Samsung is currently struggling with Foundry, so, they need to find their fit for the Foundry segment. Um, 18 a. Mario mentioned that, it definitely will be used internally. Yeah. unfortunately, and, and Intel admits this, they were kind of late in locking down 18, the PDK for 18 A, which kind of missed some consumer, uh, uh, customer targets. 14 a they're trying to lock down a lot earlier, and they're trying to come out with different SKUs, for mobile and low power type applications. So there's still a lot of work that has to be done there, but Intel has to execute on all three of those core competencies and the industry needs it to. you also have to remember from the Foundry perspective, there are a lot of stakeholders. the EU government has invested, in Intel. The US government has invested Intel. Customers have invested in Intel for Foundry. They need to execute on Foundry more than anything else.

Leonard Lee: 11:50

Well, yeah. and products as well, because the latest news coming out is, there's headlines coming out about Intel looking to spin out and, or spin off. the networking edge group, right? So. there's a lot of pieces that continue to move within Intel, but I want to go back to the comment that, Carl made earlier. and the need for an AI strategy, I think that continues to be the biggest gap. We talked about this at. Vision, right when we're at that bar and we recorded our little thing. What bar? I don't

Jim McGregor: 12:25

remember.

Leonard Lee: 12:25

Yeah. And one of the things that really concerns me is this continued lack of, vision for where they want to go. There was mention of, maybe. Intercepting the market trend toward, agentic ai. That was one of the things that Lupu,

Mario Morales: 12:43

yeah, when they mentioned that is really them admitting that they can't compete in training. Right. And so they have to go to right inferencing to be able to at least capture some mind share and eventually some commercial opportunities, in ai. So this is a pretty big pivot for them, but I think it's the right pivot now, just because inferencing is exploding, not just in the data center, but also at the edge. Right.

Leonard Lee: 13:08

But here's the thing, Mario, was that ever really a pivot for them? Because this is what we heard also from Pat, back at Vision last year. And so I don't think this is new thinking. I think it's just curious that almost a year and a half. After those early conversation about, with Intel about where, what is your AI strategy, that they still don't have something jelled.

Mario Morales: 13:32

Yeah. The difference though is that when Pat said it, the AI inference market was not really, formulated. Right. I think when you look at it today, you're definitely seeing the growth and there's a big difference there. That's one part of it. And, and I was gonna just point out that, that. Everyone's had an inference strategy. it used to be that Intel used to do a lot of the inferencing in the data center. 90% of, was really geared towards Zon, right? Zon had some capability to support inferencing, but that market just became, so much bigger and faster that, the GPU took over. And I just think for them, they're, betting on the fact that not everyone's gonna just focus on frontier models at the end points, right? Or in enterprises that there's gonna be a happy medium or the market will segment a lot more. And if that's the case, it does open up the room for other alternatives to go after that space. And one of the key strengths I think Intel has is their position in enterprise. I think that this is something that they need to leverage more and they just need to have the portfolio to be able to go after that.

Leonard Lee: 14:38

Yeah. Okay. any last thoughts on Intel?

Karl Freund: 14:43

One, one quick one. the fact they're giving up training could be. That'd be exactly what they need to do. because what most of Nvidia SA sales in the data center are chips that can be used for training and inference. And that's a strength today. But having a solution that's designed for inference opens up new degrees of freedom they didn't have before. So, for example, they can look at the memory architecture and come up with a much lower cost approach for inference using different kinds of memory than HBM that would allow them to have a significant cost advantage if they can deliver adequate performance for distributed inference, right? Things like agent ai.

Leonard Lee: 15:22

But that's already happening. We've already, Jim, you and I, we were at Compex. We see that happening and Intel is sort of in the game, but they're a little bit, I don't know, they're tracking, but they're not ahead. And they, I think they've missed the opportunity to be ahead. But you know what, now that you mentioned, memory, alternative memory of Micron and socom, anybody want to talk about that? I, I thought that was a pretty big deal, right? I think,

Mario Morales: 15:50

I mean, I look at that space. First of all, we're expecting the bit growth for HPM to be in the 96 to a hundred percent growth range this year. So a lot of that is clearly coming from the data center. And the announcement that Micron made with their technology was, something that is more application specific. And it's really meant to drive, their own differentiation in that space because today they're behind SK High, next, they're behind Samsung and trying to get their socket wins with some of the AI accelerator companies and also the ASIC companies. So this is a move that Microcar is making in order to do that. I think Micron has very good technology at the process level, but they, they've struggled a little bit in getting their technology solutions qualified, and we're hoping that this approach might help them accelerate into the space, especially to capitalize on the growth that we see for the DRAM portion of the market.

Jim McGregor: 16:45

Actually they, on HBM, because of Samsung's, limitations or struggles with HBM three E, they vaulted into the number two spot from a very, very distant third in less than 12 months, mostly on the heels of, their NVIDIA design wins, uh, with the Hopper series. And they already supposedly have designed wins with the Reuben series, with their HBM four, and they've got design wins with HBM three on the new a MD uh, MI three 50. So. I actually think microns in a pretty good position. I mean, they went from way, way back. They're focused on the entire data pipeline. They're focused on the memory. They're focused, well, the main memory, they're focused on the HBM memory. They're focused on the storage. they've got a pretty good high performance solution. that scales, I think anywhere from enterprise all the way up to, AI applications.

Karl Freund: 17:37

Yeah, I think they're passing, Samsung was not what the market expected, and whether that's because of good execution on their part or bad execution on Samsung's part is, open for debate. But they're definitely in a better position now than they were 18 months ago.

Mario Morales: 17:51

Well, I would give them also, I can't not talk about memory and not bring up SK Hynek because I think they've executed extremely well. They've been pretty much locked in and on allocation One thing that I haven't really talked to or posted about this was that in the first quarter of this year, SK Hynek actually surpassed Samsung from a revenue standpoint. This is the first time we've seen that in dram. So that's how significant the growth has been for the market. So, I just wouldn't have thought that that could happen, but that just goes to show you that SK Hynek is definitely still outpacing these guys, despite the success that Micro has had.

Jim McGregor: 18:30

I'd also keep an eye on the Chinese vendors. they are, you know, obviously big players in China, but we may see those emerge, outside of China and other regions, over the next year or two.

Leonard Lee: 18:42

China. I heard China. Carl, you wanted to talk about China.

Karl Freund: 18:48

Yeah. You know, there's been a lot of discussion about China, and the reopening of marketplace to a MD and, and especially to Nvidia age twenties by the taco movement. I did some research on China, and they actually have about half of the global, AI software developers now, that's a significant competitive advantage for China moving forward and for the US to simply say, now we're not gonna, we're not gonna participate in that market. And he hands the future to China. Mm-hmm. So, I'm not often one to agree with Trump's policies, but in this case, I think he, I think Jensen Wong convinced him. Probably with help from others like Lisa Sue, that, that you can't just take America's biggest competitive advantage globally across all technology sectors, which is AI and GPUs, and say, no, we're not gonna sell that to what could become, one of the largest markets for ai. Yeah. just walking away from that made no sense at all. and allowing selective products to go into that market will, help amortize the cost of developing these expensive chips, with the additional revenue from China. So I think it was a good move.

Mario Morales: 20:05

Well, I don't wanna give a, a MD and Nvidia all the credit. I think part of what's happening here is that, we've been seeing for the last 90 days the US trying to negotiate with the rest of the world, including China. And it hasn't gone anywhere. It's been stalled, right? So you have to start conceding certain things. And one of the immediate ones was this issue here, because you stop A A MD and Nvidia and others from growing, you're stunting the growth, not just of these companies, but the whole entire industry because AI is the only thing really driving the market today. If you, if you. Take that out. Uh, you got automotive and industrial and consumer, they're all declining at this point, so you don't wanna fracture and really stunt the growth of a very important market segment for the world. So I think this is still gonna be a volatile space. I think China, as I completely agree with you, Carl, the, the, not only the amount of developers, but how quickly they're moving too. I mean, their large language models and their smaller models are really, really good. and you're seeing them deploy more on clients than you are in the data center. But they're happening, and, and we cannot forget the fact that China drives a very large amount of data creation. Right? You think about everybody that's on. Every type of platform in China and there's so much data being created, that's what I think is gonna fuel the capabilities that China will continue to have. And, and you can't underestimate that. So it, I, I don't know if the, is, if it's foresight by the administration. I think they were starting to be put in a corner and they need to decide to show something in terms of, um, progress with, especially after the April announcements that shocked the markets And, and now we're still in a stalled position. But I still worry about the volatility of, of tariffs really still coming out. What, whatever the numbers end up being and that I think will hurt the market not helping.

Jim McGregor: 21:58

China's important for the sell out or the sell through, but it's also important in terms of the rest of the ecosystem. You gotta remember that, it's important in terms of establishing standards, global standards, especially as we start looking at six G, which that effort just kicked off a couple months ago. It's important in terms of rare earth materials where they, they control a large percentage of the vast majority of rare earth materials at this point in time. It's important for engineering resources, for, uh, assembly and test capabilities, both at the semiconductor and at the, system level. we spent decades creating a global, ecosystem for the electronics industry and trying to pull a piece or a major piece out of that puzzle, just makes everything collapse.

Mario Morales: 22:44

Yeah. The interconnectedness of our supply chain. Exactly. I think that you can't just pull these things apart, and I completely agree.

Leonard Lee: 22:52

Yeah. And you know, I think the other thing is, is that, even beyond a lot of those, which I think are surface level, under the hood, there's the Chinese secret sauce that I don't think the industry really has figured out yet, since the advent of deep seek in particular V three. And so the Chinese language specific optimizations that have ha happened under the hood and in all likelihood what they're gonna be doing with Huawei, with their AI supercomputing platforms and what they will be collaborating with Huawei on in terms of optimizing their, uh, software layers, in a way that optimizes for the Chinese language. And I think the language itself is highly discounted. simply because a lot of people who aren't Chinese, don't know Chinese, and you know, we all have to remember what these large language models, one of the most. fundamental, ingredients or basic pillars of this is language. And this is one, one of the reasons why we, even at GTC, one of the research items that Bill Daley put on his agenda this year was language. And I think there's been a bias that English is somehow the optimal language for ai. that may simply not be the case. And so I think, this is gonna be a really interesting, market, obviously how things play out and whether or not this is, where we are today is gonna play out in terms of a, market opportunity that, the Nvidia a MD and others thought it would be. I, I, time will tell, but I think there's factors that have not yet played out that will play out in the next six months when we'll see what kind of traction, that translates into in terms of, the AI business, for. US players in China. So, and I think this is a good segue

Mario Morales: 24:49

to the next topic that you had brought about, which was the meeting from this past Wednesday on the new, AI plan for the US and establishing and maintaining that leadership. I think when I look at that, there's two components to this. One is making sure that. We continue to democratize our capabilities, our technology around the world. But the other component to this that we're not talking enough about, but again, it is just a few days ago, but it's really the energy component to this Right. re bringing these two things together I think was very important. And, I think that the tech guys know this already and I think that the administration is sort of figuring this out and they see also an opportunity to push their energy, uh, agenda as well. But I think those two things are gonna go hand in hand moving forward. And it's a good step and I'm hoping that the technology companies are gonna be the ones that are really pushing this forward and not necessarily the administration. Because for me, every time I see the public sector involved in any of this, it just slows down innovation, right? It doesn't really help. It. I, that's a mess. There's been, yeah, in the past you could say That the public sector really helped us get to the moon and, bring a lot of, innovation in aviation and these different industries and sectors. But today, the AI space and the semiconductor space, it moves very, very fast. And I just don't think that policy is moving fast enough to support it in a way that you can continue to innovate. So I see those are, things that have to be tackled.

Jim McGregor: 26:17

Yeah. You can't expect, people that can't even, spell the technology to be able to regulate the technology. And yes, that is a direct shot.

Karl Freund: 26:25

Yeah. I had direct shot. Well deserved. No, I agree with you, Mario. I think the best thing that the public sector can do is get outta the way

Jim McGregor: 26:33

Exactly

Karl Freund: 26:34

Allow the technology to develop and flourish and make. Lots of millionaires in Silicon Valley.

Mario Morales: 26:42

billionaires here.

Karl Freund: 26:43

billionaires. Yeah. There

Mario Morales: 26:44

are areas where they can actually. Bring in that will help I, supporting universities, which we're going against that today. Yeah. Also, being able to support r and d efforts. Right. These RD efforts are really an area where semiconductor companies don't cut back in that space, but if they can get some augmentation or some support in that space, that I think has better influence than trying to force people to ship their products from being built in Taiwan to bring'em in into the us. I mean, we're working with customers now that they're dealing with this pressure from the US administration and they're being forced to reconsider where they're building existing products, which makes no sense because in some cases it could be 20 or 40% more cost add to you if you move it from one region to another, and that's just not acceptable. It doesn't keep these companies competitive, and then ultimately it reduces their ability to invest back into our industry.

Jim McGregor: 27:41

And they have to really break down the roadblocks., In terms of building new data centers, getting them up and running, leveraging, different types of energy resources, especially allowing SMRs small, nuclear, modular reactors. And, you know, we can't afford, especially if we're gonna maintain our leadership with the growth and demand and with the continued innovation, we can't maintain our leadership. If we can't maintain the capacity, we need the compute capacity. we can't afford to wait 2, 3, 4, even sometimes five years for a new data center. They need to be able to be up and running within 12 months.

Mario Morales: 28:17

Yeah, I think one concern I see with all of this though, is that we are starting to destroy some of the regulation that we put in place that actually is good for our environment, is good for our planet. Being able to, drive down, chemical absorption in our environments is important. The types of materials that we use to build are important. And we had passed laws about five to six years ago that were gonna help remove some of those materials, but they're materials that just don't go away. Right. and now we're removing that and I see that as a big concern. we're sort of chasing profits and scale at the expense of climate. Right. you saw what we saw recently in Texas and what's happening in Florida, but really across the country, the world's. Atmosphere and weather patterns are completely changing and becoming completely unpredictable. Right? And a lot of that has to do with some of these chemicals that we've put in our environment over the years that we were trying to remove now. But I think that's gonna all go away. These policies that we're implementing, especially in the us, are against everything against our planet. Right? I agree with you.

Leonard Lee: 29:24

There has to be a balance. But here's the interesting thing. I think everyone has just latched onto these narratives. the narrative that in order to win an ai, you need to build out massively all these energy hungry data centers. Right. This is the assumption. Yeah.

Jim McGregor: 29:44

I don't think that's just it. I mean, we need capacity, but at the same time, you gotta realize our industry working to improve efficiency. Just going from, the hopper generation to the Blackwell generation, we're looking at increase efficiency of 32 x. I mean, that's huge. That's huge. And that's unprecedented in our industry. So I would argue against that. It's not just about capacity. it's interesting'cause Pat Gelsinger, two years ago at Direct Connect said, you know, listen, we need to kind of. Densify computing. To make it more efficient, we have to put the processors, the accelerators, the memory, the networking all closer together. And that may eventually be in a single module. Yeah. And that module may be three to five kilowatts. That was two years ago. This year. I'm talking to people and they're like, Jim, yeah. That's gonna be 30 kilowatt 20, 20 to 30 kilowatts. and we're, I mean, we're already specking out, racks that are a megawatt, a megawatt for a single rack. Right. But, um, but those racks are a thousand times more efficient than the previous generation.

Leonard Lee: 30:48

Right. But for, what is it for model training? What is it that we're doing? Yes. All of this.

Mario Morales: 30:56

it's for model training. It is for some of these new workloads that we're beginning to see be commercialized. But I think ultimately we're still in this middle of this race, right? Where everyone's chasing to get to a GI and if that happens in the next three to five, yeah. But what I'm saying, if it happens in the next three to five years, at that point in time, then things start changing in terms of how we use capacity differently, how we parse it out differently. I think that right now everybody's just focused on. We can't be left behind, so we're gonna invest, invest, invest. Until we get to that point where whether it's open AI or maybe it's meta or someone else gets to that, a GI point, then we start seeing a different kind of a landscape. Right?

Jim McGregor: 31:40

Right. but it's not all training. I mean, we've already passed the point where inference processing, surpasses training in most data centers or most applications today. and you gotta remember that whether you're using an ASIC or whether you're using A GPU, especially the GPU, it's programmable. Those resources can be parsed up into handling multiple different models, handling hundreds, thousands of different, users all at the same time. So, I mean, we really are improving efficiency across the board,

Leonard Lee: 32:10

right. At the system level. And then the token production level. The big open question still is to what? Economic value, right? That's still a big question out there. So again, I'm just going back to, drawing, folks attention to what are some of the assumptions are supporting this narrative that we need to create an energy crisis in order to win ai. And it's just, I'm just leaving it out there. obviously it's great for folks who sell, the picks and shovels and build out these data centers, but, you know, going back to a lot of the concerns that Mario brought up, to what end and for what purpose ultimately to create,

Karl Freund: 32:51

didn't the Google earnings. Report. I view that as the first nail in the coffin of the concern, that AI is simply not going to produce financial returns because it certainly is producing really attractive financial returns at Google. Is that just an outlier, you think, Leonard, or are we gonna see more in next week?

Leonard Lee: 33:17

You always have to unpack that. You can't just say, because of ai, their business grew X, Y, Z, the, it's important to do attribution. And that was one of the important things, that was very difficult to do when I was in consulting. We always like to build these business cases and case studies about how a certain technology did. Contribute to x, y, z, business metric improvements. And that's very difficult to do. You can't just generalize a particular performance in a quarter and entirely attributed to a certain technology. It's very, very difficult. In fact, today we don't rightly know how much impact, quote unquote, whatever AI is on any of these companies. Right. It's actually pre, opaque. Yeah. But that,

Jim McGregor: 34:00

that's like saying, in the early 1980s, what impact is the PC going to have financially? It's had a huge impact's. It changed everything. It changed how we think of jobs. It changed how we think about society. It changed how we learn, how we play. It changed everything. Yeah. Um, AI is in that, especially the more,

Mario Morales: 34:18

the more we went mobile. Right? Yeah. It really changed how we use our devices. I see your point. Leonard, you, you've always been more of a pessimist and I appreciate that because we need some, some, some folks to be pragmatic, but at the same time pragmatic. Yeah. I think, is that the better, the better word

Leonard Lee: 34:34

for you? Too much optimism. Yeah. Yeah. And you know, too many people looking at the picks and shovels in the go. Well, I'll

Mario Morales: 34:40

give you some nuggets that support your position. For me, it is a little bit alarming that Google can show such a surge in business or Microsoft or even Amazon. And yet these companies continue to lay off. Resources. just this year so far, we've seen almost a hundred thousand people in tech lose their positions. And to me that's alarming because we're supposed to be in this era where we're growing and we're trying to identify more and more markets on top of what we're investing in, but yet we're getting rid and seeing attrition across a lot of these. Leader companies that we have in our ecosystem. So I, that goes to your point, to a certain degree that they're still trying to figure out beyond the internal, automation and operational scale that they can get with ai, how do they commercialize that even further to some of their customers where they now can bring it directly to their customer and not just to benefit internal? Right. So, yeah, yeah, yeah. There's some of that that has to balance itself out, but I, I do agree that some of those things are a little bit alarming. especially when you see a company like Microsoft or Google that has done very well, but yet they're still getting rid of resources that I think are really talented resources that any other industry would love to have.

Leonard Lee: 35:58

Yeah. Awesome. See, Jim, we're not all about Kumbaya on this show. You know that, right? Yeah, I know that. But you're wrong. And I'm right.

Jim McGregor: 36:08

really matter.

Leonard Lee: 36:09

Yeah,

Jim McGregor: 36:09

yeah. I'll let you tell, tell you what, we'll come back to this conversation in five years and we'll see where we are. I guarantee you we will have the business model.

Karl Freund: 36:20

And I think in five years, Mario's concern about, the layoffs will just magnify the, there's been a lot of CEOs in the last three or four weeks admitting to the fact that AI has been, allow them to significantly reduce their workforce. You know, and I don't think we as a society have begun to, adequately deal with that reality. I've been telling

Jim McGregor: 36:44

my best friend that's a radiologist he needs to retire, because quite honestly, ai, just if, and a lot of the, especially in the us, a lot of the healthcare companies are hiding by HIPAA laws, which, they can share information, they just can't share the customer the particular personal information. But just sharing all that information and all those resources about all the diagnoses, all the scans, all the images, all the information that's out there at all this, these US healthcare providers, I guarantee you, we wouldn't need a radiologist tomorrow.

Karl Freund: 37:14

Yeah, and we probably won't need industry Analyst. Analyst, unfortunately. So luckily I'm close enough to retirement.

Mario Morales: 37:19

Well, we'll, we'll, we'll need Leonard. We'll need, but, but I think, I mean, this goes back to immigration, right? Yeah. And the immigration policies that we see today. So all these jobs that we're gonna lose in tech, where it's gonna be a lot of openings for agriculture and service and retail oriented jobs. So hopefully, right. By that point in time, they'll start paying, six figures for these smart tech pickers.

Karl Freund: 37:43

Yeah. I don't see orders becoming strawberry pickers. Yeah. I don't know.

Jim McGregor: 37:47

They,

Karl Freund: 37:47

they

Mario Morales: 37:48

could

Jim McGregor: 37:48

be good at it. I still think you're gonna see a lot of tech jobs created. It's just, a lot of it's gonna come from startups. A lot of it's gonna be in different areas. We may not need the people doing hardcore programming, but we still need people that are looking at the data and, programming the AI models and making the AI models more efficient. We still need a lot of talent. It's just, it's changing. Same way the PC changed a lot of industries. It's shifted from manual tasks to automated tasks, but we still grew a lot of these industries.

Mario Morales: 38:17

Jensen has said it well, Jensen said that, you know, if you're not doing ai, that's the, that's when you risk losing your job. It's not AI itself. Right. I think that there's automation that definitely will displace. But it will also create new ways of us doing work that we haven't thought about yet. so yeah. Well, I, I wanna be hopeful about that because if we're not, I mean, right now it's more of a cost decision. You're seeing Microsoft make moves to cut heads, but they're applying for more H one B visas and stuff. So things like that has to be addressed. That's

Leonard Lee: 38:51

the interesting thing is, so here goes back to the. Earlier topic about attribution. A lot of that in many people's mind is that they're attributing it to AI as being the reason for the headcount reduction. But you just mentioned something really important. Now you're bringing in cheaper, labor to replace maybe older labor or maybe even outsourcing. And so this is where I'm saying we have to be very, very careful. And whereas analysts, we need to have diligence and not just face value. Take a press release headline and think that's, that's basically what's happening. and that was my only point, guys. and keep in mind, I'm pretty deep in the end market stuff, all the way up to software stack. I'm not seeing the kind of revolutionary adoption you guys are. You guys are talking about, in fact, like even at, uh. DTW they're very cautious guys. Mm-hmm. And so this assumption that you're going to see generative AI orchestrating and managing a telco network autonomously, we're, we're very far away from that.

39:59

And

Leonard Lee: 39:59

so again, the question is what kind of demand, what kind of end market demand other than CAT videos and pornography are going to fuel this thing is still the big question. and you know, that is the reason why I take more of this alternative stance when we guys, we get together and you know what, this is beautiful. I, this, I think this is the most important part of our program here and what we're trying to do. and, you know, just, bring our different perspectives to the table and then hopefully our audience walks away with something a hell of a lot smarter than what. Any, one of us might be, um, opining about. So, any final thing because this is, this is running pretty long. It's really good, Mario. It's awesome having you on, man, because it's, I think you make it. Thank

Mario Morales: 40:47

I'm glad you invited me finally.

Leonard Lee: 40:50

Oh, come on, man. You know, it was always a open door for you, man. I know, I know Jim is your favorite, but, you know, TSMC No, no, no. Yeah, I'll replace him with you any day. Wow. Look at that. You know, I, I keep saying

Jim McGregor: 41:04

we, we, we need a slap button so I can, I can slap letters when I.

Mario Morales: 41:10

Yeah, right now Js can you slap'em for me please? Yeah. Alright. So TS TSMC C Come on, we gotta talk about tsmc. you know, not surprised by their earnings announcement. They continue to drive faster growth than the industry as a whole. So I said this in a couple, interviews I've done, but you know, the whole foundry industry is growing 17 to 18%. TSMC will grow about 30% and they already control about 67, 60 8% of the market. Right? Yeah. So very dominant position. I think there are a couple generations ahead of the competition in terms of, you know, Samsung does, has talked about two nanometer, even Intel's doing 18 A. But the reality is that these notes don't matter until customers. On board, right? And until customers are shipping product commercially and they're just doing it right, over 53% of, of their business is coming from what they classify as HPC, which is CPU and AI compute and so forth. So I see tremendous growth for them. I don't see anybody else touching what they're doing at this point. And there's such a big important weight into this industry, including what's happening in the us. I think that the administration's tried to put pressure on TSMC, but I think they hold more cards that they haven't used yet. And it's good to see investing around the world.

Leonard Lee: 42:32

I love how Jim

Mario Morales: 42:32

is

Leonard Lee: 42:33

proactively trying to ruin the show.

Jim McGregor: 42:36

My, my camera fell off. Sorry about that. We'll edit that out. We'll edit that out

Leonard Lee: 42:44

then

Jim McGregor: 42:44

won't,

Leonard Lee: 42:46

he's gonna have to live with that.

Jim McGregor: 42:48

Hey, am went from being gone upside down the right side up. Respect, man. It's a Friday.

Leonard Lee: 42:57

Yeah. Carl, any thoughts? Let's go around the table on TSMC.

Karl Freund: 43:01

Yeah. No, I think Mario nailed it. I think the, what they've, what they. Not just on this quarter, but what they've done over the last five years has been truly phenomenal. And, if anybody deserves to lead the industry right now, it's TSMC. I think the only, the only issue will be, potential for, geopolitical conflict that would damage their ability to continue to deliver. other than that, they've got Blue Sky ahead, green Lights, blue Sky, and Pink Floyd. It's just, they're just nailing it.

Mario Morales: 43:33

I think over the next three to four years, there's gonna be two companies that are$200 billion semiconductor companies. They'll be one of them. And we don't know the other ones. I don't see anything slowing that down. unless we go into a World War or something like that. That's the only way.

Leonard Lee: 43:48

Well, you know, it's, Jim and I talked about this, not that it's gonna lead to World War, but the militarized, the rapid militarization of, or, of the technology. Right. and it's happening. You know, you see defense talking about AI much more than before, and it's an interesting topic.

Mario Morales: 44:05

you'll have to tackle that next time.

Leonard Lee: 44:07

But you know, the defense industry, it's an industry, right? Yeah, it is. Talk about, and it's a reality that that is where AI is going, going to be applied. And the question is how, and, undoubtedly, as the defense industry, pursues the workable application of ai, which actually interestingly. In the, AI action plan. They mentioned this AI is really not at the state where you're gonna throw it out into the middle of a battlefield or in, frontline operations because it's just not reliable enough. Right. But we are, we are

Karl Freund: 44:41

seeing that. I mean, I disagree. I mean, drones, drones are now the primary, vehicles for military operations. Yeah. Wish say that too. They're all, they're all, they've all got AI chips on

Jim McGregor: 44:51

if you can think about it, I can guarantee you the defense department's been working on it for over a decade. I used to work for Gel Dynamics and some of the technology we worked on still hasn't seen the light of day, but it was far in advance. Anything that, you could imagine. not to mention, even facial recognition, and, voice identification, stuff like that. That's actually been in use for more than 20 years.

Leonard Lee: 45:15

Jim, Jim can't wait until we have the episode dedicated to,

Mario Morales: 45:18

Yeah, we're gonna probably have to sign something to waive our rights away if we start talking about military. But I agree with Carl the drone wars that we've seen. Have actually brought a lot more visibility to the potential of applications being used in military. Sure. I think, we can't forget the worldwide surveillance that we're seeing now using social media platforms as a way to gather more and more information on Oh yeah. On citizens. And look at what is doing today in the us they have access to databases that companies have been building for years. And that's how they're able to identify and locate folks. And, and I'm not talking about the random things that they're doing, but some of those things are all using technology that we've had for a long time. And you start putting AI on top of some of this technology, it just gets a lot more, intrusive. And we'll see where that goes. I hope that we don't get to that point, but I'm afraid by. The leadership that we see around the world these days, that that's the way things are going

Leonard Lee: 46:21

well, there's conflict. So it's naturally going to create demand for those applications. I mean, that's the thing. but, well, I think it's

Mario Morales: 46:28

not just conflict. I also think that, Convenience always convinces consumers and citizens to give up their privacy. And we've seen that with our emails. We see that with our phones and, and with even social media today But anyway, that's a different topic. I'm, I'll, I'll, I'll stick to semi connectors.

Jim McGregor: 46:45

Mar Mario, living out in the middle of the woods in the mountains in eastern Arizona. It doesn't look so bad anymore, does it? No, it doesn't.

Mario Morales: 46:54

I might throw some dollars over there. Build me a little shack up there.

Leonard Lee: 47:00

But, real, really quickly, going back to TSMC, I, one of the things I think has been really interesting over the last, couple of quarters is, that HPC is now. a larger portion of revenue than mobile. And going back, it

Mario Morales: 47:14

has been, it has been since, probably late 2022, that transition happened, against mobile and then it went

Leonard Lee: 47:21

back up and then it, and now I think it's, it's been sustainable. Yeah. Sustain. Yeah. I mean now it's like 60% that's like, yeah. Right. Versus now I think it's like 37% or much smaller number. and it goes back to, Carl's point that, and why I think it's important. For, I, I think we'll all agree it's important for Intel to figure out if they're gonna be relevant, figure out what their HPC game is going to be, whether it's pc, but you know, obviously the, outsized growth of HPC due to AI infrastructure, compute, and probably, advanced packaging and all the other stuff is probably, even more critical now for Intel, despite where Intel is today. And so, and again, the fact that they don't have a well jelled, strategy at this point in the game, and Jim, you always say it in the semiconductor industry, whatever comes to market you planned, you know, two, three years before that is my biggest concern.

Mario Morales: 48:28

Yeah, I mean, you hear it from folks like Lisa Sue. She'll talk about when we made bets on Zen, it was for the next five to 10 years. And I think the bets that Intel's making now, or that they have to make is for the next five to 10 years. But for me, I'm looking closely at the sovereign AI investments, the announcements like Stargate or Chris, uh, celestial that you're seeing in Japan. There's so many different types of, public sector programs. And for me, what's notably interesting is, is that there are certain big names that are not in those announcements, right? you're not seeing some of the hyperscalers in that announcement with exception to Microsoft, and you're not seeing Intel. that's interesting. So they have to get back to that point where they not only take advantage of their position on the enterprise, but even some of these public sector areas, which I think will drive more AI infrastructure. They need to be in those accounts. Yeah, in those opportunities. That's probably

Leonard Lee: 49:24

where the big customer is that they're looking for, for 14 a, right? Yeah. That has to start now, right? So yes. Last word for to Jim. Because Mario, I think, sucked most of the oxygen out of the room.

Mario Morales: 49:38

Wow. This, I guess he wants me the first and last time to be on.

Jim McGregor: 49:43

No, no,

Leonard Lee: 49:44

no. Dude. You know

Jim McGregor: 49:45

I love you, man. Giant sucking sound. Okay, we'll, we'll attribute that to Mario from now on

Mario Morales: 49:55

the last. Jim. that sounded bad in so many ways. I don't even know why you say

Jim McGregor: 50:00

no.

Mario Morales: 50:04

The camera again.

50:07

Drop kidding.

Jim McGregor: 50:10

Drop my microphone this time. Right. Last word on what? actually, it's interesting because say that you're all wrong. No, actually, I would say that, some of the things we see going on right now, in the industry, some of the most interesting things are around partnerships and acquisitions. We saw Global Foundries acquire MIPS over the past month. We saw, synopsis complete their acquisition, uh, of Ansys. I think a lot of what we're going to see over the next couple years, and this is I, I think a topic for a future podcast is how the dynamics of our industry change. Because I think the traditional value chain, not just because of ai, but because of innovation is, is getting turned upside down.

Leonard Lee: 50:57

Yeah. Carl, you and I talked about that at Cadence Live.

Karl Freund: 51:01

Yeah, yeah, yeah. It'd be interesting to see, Tim, I agree with you and, and I think it'd be interesting to see what Apple does, right? Because so far Apple has been victim of their own. Anybody but Nvidia they what I call their a BN strategy. Yeah. How you gonna do ai? I don't know, but we're gonna use everybody. But Nvidia, we, we should. They're paying the price for that now. They're paying the price for that now. And they've been, they've been nowhere. Siri still doesn't need ai? It's just like, really? Are you kidding me? You either hired the wrong people or you gave'em the wrong tools, or both,

Leonard Lee: 51:36

or, or you didn't give them$200 million, yeah, that these are all great, topics for us to talk about, in future episodes. But, that's all the time we have. We gotta cut it off here because this is the longest episode ever. Mario, great to have you on, man. fantastic, Jim. Wonderful as always. Carl, love you to death and so thanks everyone for tuning in. Hopefully you found the episode, enlightening and, insightful please contact these gentlemen, they're not part of Next Curve, so everything that they say is entirely their own thinking and opining. But you can, reach Carl on LinkedIn as well as Jim, as well as Mario. So connect with them there. But you can reach, Carl at www.cambrianaiuh.comandjimatteusresearch.com and Mario at IDC. Remember to like, share a comment, share your perspective and reactions to this episode, and subscribe to the Rethink Podcast here on YouTube and Buzz Prout. And take us on the road, your jog and listen to us on your favorite podcast platform. Remember to, also subscribe to Next Curve research portal@www.next curve.com for the tech and industry insights that matter. We will see you again next month and hopefully we'll have all four of us on again because this is really great. This is great. Really fun. Hang out with you guys and just, arguing. All right, time. Drop the

Mario Morales: 53:18

mic, Jim. Alright gentlemen, thank. Enjoy. Bye.