Try Free

Lisa Su on Nvidia, China and the American AI Stack — TomorrowX Summit

iConnections November 19, 2026 24m 4,226 words
▶ Watch original video

About this transcript: This is a full AI-generated transcript of Lisa Su on Nvidia, China and the American AI Stack — TomorrowX Summit from iConnections, published November 19, 2026. The transcript contains 4,226 words with timestamps and was generated using Whisper AI.

"- Are we in an AI bubble? Are people over-investing? I truly believe that this technology is the most transformative technology that we've seen over the last 50 years. We're not seeing something that looks cyclical. We're seeing something that looks generational. One engineer over the weekend with..."

[00:00:00] Speaker 1: - Are we in an AI bubble? Are people over-investing? I truly believe that this technology is the most transformative technology that we've seen over the last 50 years. We're not seeing something that looks cyclical. We're seeing something that looks generational. One engineer over the weekend with Claude was able to get it up and running. I'm not gonna say over-promise, but to promise and exceed expectations. [00:00:24] Dom: - Lisa, I asked a lot of people for advice before I did this, and the best advice I got was, just treat this like one of our quarterly earnings callbacks. So that's the spirit of which we'll do this conversation. And I know there's a few elephants in the room. I promise we'll address those. - Really, are we? - I'm gonna try, at the very least. But before we do that, I actually wanna go down memory lane. So when Lisa and I first met, AMD was roughly a $5 million market cap, $5 a share, a billion shares outstanding. I was the associate analyst on our small cap value team. And now we're north of $500 a share, approaching a trillion dollars of market cap. I mean, it has been a wonderful journey. So first off, thank you on behalf of our shareholders. But when I take a step back, I'd love to kind of know, what were those important initial decisions you made? And what lessons kind of still shape AMD today? [00:01:34] Speaker 1: - Yeah, absolutely, Dom. So first of all, welcome everyone to Austin. This is my hometown. So I hope you guys like it. And Dom, thank you for the tremendous partnership with T. Rowe over all these years. And I guess what I would like to say is, you know, when I think about the journey of AMD, the journey of AMD is really about like pushing the bleeding edge of performance. And, you know, we really believed that high performance computing would be the most important part of the semiconductor market. It's the hardest part. It's where you're always having to innovate. You always have to take risks. And you have to make, you know, a lot of right calls. And so our principles are really around, you know, building great products, spending a lot of time with our top customers. You know, these are like the key folks in the world, whether you're talking about the hyperscalers or the frontier model labs, knowing where the workloads are going and then making sure, you know, we're designing the technology underneath of that. And I think we made some really good bets. You know, this idea that Moore's law would be changing, slowing down. We had to think about a different way of putting high performance computing together. Back in the day, you know, people didn't understand what a quote unquote chiplet was. It's the idea that, you know, to make the largest computers in the world, you'd have to break it up into smaller pieces. And, you know, fast forward, whatever, 10 plus years. And here we are like high performance computing is the center of the universe. You know, so much of the world is focused on technology and AI and how do we feed all of that compute requirements and needs and, you know, we love where we are today. [00:03:15] Dom: I mean, that modular nature of the chip that gives you so much flexibility, right? And I think we've seen it recently too. Look, AMD is now clearly a data center first company, right? And, you know, we have a bunch of investors here. How should we think about how the growth profile has changed now because of that and how the business mix has changed, right? I always think about growth and margin, right? And that interplay between the two. [00:03:41] Speaker 1: Yeah. So, you know, going back to sort of the trends that we see overall, I mean, we've looked at this market and it's like an incredibly exciting market. You know, talk about a $2 trillion TAM over the next five years. And so much of that is concentrated in the data center because we do need more compute and we do need more intelligence. And, you know, our business has really, you know, gone that way. I think when, you know, when I first started, data center was a tiny piece of our business. In '25, we were less than 50% data center. You know, this year, we're going to finish the year probably, you know, greater than 60% of our business being data center. We're going to more than double our data center business this year. We just, you know, had our third quarter earnings call where we talked a little bit about what 27 and beyond look like. And we see just tremendous demand for more compute. So the overall company is going to grow over 80% year on year, which is a big number. [00:04:39] Dom: I'm hoping it's a big number, big number, big number. [00:04:41] Speaker 1: And our data center business will be over 80% of AMD. So the world is a different place. I think it's exactly kind of how we thought it was going to work out. But if anything, I think it's faster. I think it's more intense. I think it's more dynamic. But the foundation is, you know, technology has never been more important to, you know, every enterprise, every country, every application. And we're so fortunate and, you know, excited about being in a place where we're delivering some of the greatest technologies out there. [00:05:16] Dom: You know, everyone always says under promise and over deliver. But I'm sure, Lisa, you've heard me say, "No, over promise and over deliver." And you've done that. Look, I think there's a lot of questions right now on, you know, there's this tremendous infrastructure build, like you said, trillions of dollars. And people are actually starting to question the ROI, right? How can you continue to spend north of 100% of your cash flow from operations on CapEx? What does that return on invested capital look like for your customers? This is something that you have a really close view to. So how do you get comfortable with the ROI for your customers? [00:05:58] Speaker 1: Yeah, I think this is a really important point. I think we get asked this all the time. You know, are we in an AI bubble? Are people over investing? What's going on? And I guess, you know, with my seat and the conversations that we have with our customers, what's really clear is we truly believe, I truly believe that, you know, this technology is the most transformative technology that we've seen over the last 50 years. And the power of AI when applied correctly is such that it makes sense to invest. It makes sense to lean in now. Now, of course, you have to be, you know, thoughtful and that's, you know, that's part of there will be winners and losers in this. But the winners are those that see the trend, that have built the balance sheet and the capability to invest that realize that these investments are three to five year investments. So, you know, I think I saw David on just before me. It's like, well, we can't predict three to six months that well. But when you look, you know, over the longer period of time, you say, of course, you're going to need more compute. And the key is, you know, for us to build the right technology, work with the best partners and ensure that we're looking at, you know, the overall supply chain and the overall set of investments. So, you know, the best thing that we can do is to ensure that every part of the supply chain and every part of the value chain is investing at the right amount, because you certainly don't want to be limited by underinvestment, but you also don't want to overinvest. And that's where we've had to really step back and say, this is a much, much bigger AI ecosystem than, you know, a traditional semiconductor market, and we're managing it that way. [00:07:35] Dom: You know, Gavin always talks about the iron law of semiconductor cycles, right? It is the conditions of the supply tightness that create the glut over time, right? Customers over order, you know, EMS companies over order, and the tail of the whip effect, right? We all know how it is. You know, seen a few of these cycles now. How do you think this one's going to play out? Because this one is a little different than others. [00:08:03] Speaker 1: I think this one is a little different. I think what you're seeing is, you know, typically you do see the semiconductor market has been known to be a cyclical market. And when demand is high, we build a lot, and then you satisfy that demand, and then there's a little bit of a digestion period. But we are not seeing that in this today. We're not seeing something that looks cyclical. We're seeing something that looks generational in terms of the growth that's required because, you know, AI is eventually making every company, every enterprise more productive. And so you're willing to invest in that productivity so that you build better products. And it's like a, it's a vicious positive cycle in that. Now, that being the case, we are always looking at ensuring that the investment is a coincident with the return on investment. And so that's a very regular conversation that we have. But this cycle does feel quite different because of the productivity capabilities that you can get from these investments. [00:09:07] Dom: All right, here we go. There's been a lot of news in the industry recently. Just a little bit. Just a little bit. Your main competitor, NVIDIA, has been top, top, top of the news. Just look, what's your perspective on, and where I would really love to hear is how you think about it from an AMD perspective. [00:09:29] Speaker 1: Yeah, so first of all, NVIDIA is a phenomenal company. You know, Jensen has been a tremendous leader for the industry. Colette's a great leader. There is a lot going on. What I try to do from an AMD standpoint and from an industry standpoint is, look, we have to separate a little bit from the noise. You know, the noise will work itself out in time. But on the, in taking a step back, what we see is, look, from an AMD standpoint, we are in a place where we have just tremendous opportunity. That's tremendous opportunity to be part of this infrastructure build out, to work collectively with our partners in terms of the roadmap going forward, to, I'm not going to say over promise, but to promise and exceed expectations in terms of delivery. Sounds good. Those things must happen. I think there's another piece that's also really important that I'd like to, you know, kind of express is that this industry is really important. Yeah. And, you know, a lot of the dialogue, if you look through, you know, leading up to the midterm elections and right now is, hey, is the technology industry in touch with the rest of America? Yeah. Are we thinking about the things that are important? And I want to say that we absolutely are. And we need to do more of that. And we need to be better at ensuring that, you know, as the technology build out happens, that, you know, everyone who's impacted by it feels the goodness of it. Like, I'm a big believer in technology has a way of tremendously improving people's lives. And I think some of that has been lost in the noise. So, you know, we look at these data center concerns that are out there. I think this is a great opportunity for dialogue around those key topics. And, you know, we need to make sure to talk about that as much as, you know, some of the other things. [00:11:19] Dom: Yeah, that makes sense. The other piece that's tied up in all this is the geopolitical strategic nature of semis, right? I mean, I think about the life cycle of the industry. We had these boom-bust cycles, and then we had immense consolidation, right? This is when I was getting started at 15, 16, 17. Everybody was buying everybody. Every day you'd wake up on a Monday, there'd be a new deal. And, I mean, semis are really the lifeblood of American technology, right? And as a result, of course, that creates geopolitical tensions with China. And just how do you think about how AMD has navigated those? And it looks like restrictions are continuing to increase. So, how do you plan to continue to navigate those going forward? [00:12:09] Speaker 1: Yeah. Well, I think it's natural. You know, semiconductors are now such a piece of economic prosperity, of national security. So, you know, we have been going through this for the last few years. I think the key with any type of export controls or, you know, semiconductor policies, ensuring that we're in the room working with the administration, working with congressional officials as they're, you know, coming up with these rules. The idea meaning we want to ensure that, you know, first and foremost, we want to protect national security. So, you know, the highest performing chips are going to be restricted. And we get that, and we know that, and we want to be part of that. I think that it goes to say, though, there are a lot of positive things about our technology that we do want to export, and we want to keep that in perspective as well. So, as it relates to some of the new restrictions that came out this week, you know, I think we've looked over them, I think we anticipated a good amount of them. So, given where the demand curve is right now, I think we're still in a place where even with restrictions, we will see demand still substantially higher than supply. But more importantly, we're also working on ensuring that, you know, the world builds on the American AI stack. I think that's really, really important. The Commerce Department just recently released some, you know, export packages, and, you know, AMD had a good number of them that were approved in that, because we do have great partnerships around the world. And, you know, America leads in, you know, AI and semiconductors today, and we want to continue to do that, you know, going forward. [00:13:48] Dom: I'm going a little off script here, Lisa, but you mentioned the AI stack, and I think about what you've built with Rock'em, right, and how Rock'em has so changed AMD's industry positioning. And, you know, Anthropic really did you a favor on this roadshow. They talked about how they got you guys up to speed in three days, something that would usually take, you know, nine to 12 months. Maybe you want to kind of expand on that. [00:14:14] Speaker 1: Yeah, well, I mean, this is actually a great real world example of, you know, the power of AI and what was happening. There has been, you know, this, you know, sort of belief that it's actually very hard to use different hardware architectures, you know, between NVIDIA's architecture, AMD's architecture, you know, other architectures that are out there. And, you know, we've spent a lot of resources on that. We've acquired, you know, like 10 software companies. We've built up that team. But what we've seen over the past six or nine months is the frontier models have gotten really, really good at engineering problems. And especially in this area of software and when you're doing kernel optimizations and things like that, it's gotten exceptionally good. And we're continuing to make it better through, you know, deep collaboration with all of the frontier labs. And so, yes, you know, Anthropic mentioned when they were on our stage and they mentioned as part of their recent comments that, you know, literally they took their model, you know, one engineer over the weekend with Claude was able to get it up and running on AMD and it ran really, really well. And that's an example of, yeah, we want technology to be that easy to use. And that's, again, the power of, you know, AI in action. [00:15:34] Dom: And it's funny to see the through points throughout history, right? So much of the Rock'em team, so much of the software comes from the Xilinx engineering team, right? And, you know, you always talk about the right compute for the right workload, right? And I think about now, AMD really has all the pieces, right? We have CPUs, we have GPUs, we have FPGAs, we can do custom silicon if the customer wants it. How do you think about that broad portfolio? [00:16:00] Speaker 1: Well, I guess if I want to say one thing, you know, the market tends to want to be very quick to say, well, hey, this is the next best thing, like everyone's going to do it. Or this is the next great model. And I think what you've seen over the last, you know, hopefully 12, 18, 24 months is there is no one thing that's the best. You actually need the best of a number of things. So you need the right compute, you need the right hardware, you need the right model. If you talk to enterprise CIOs today, all of them are thinking about how do I build choice into my ecosystem so I can have the right hardware at the right model, the right price, so that I can optimize those things. And that's really the foundation of AMD. Like we've always believed you need CPUs, you need GPUs, you need FPGAs, you're going to need ASICs, you're going to need them to run a software stack across all of that. Sometimes you're going to build custom chips, sometimes you're going to build a standard products. But the foundational IP is very, very similar. Like you need the best architectures, you need the right packaging technology, and that's what we've invested in. So I'm a firm believer in having the right compute for the right workload. And as we go forward, I think you're going to see much more of that. [00:17:23] Dom: There's two pieces of this that I'm most excited about. So one is on the CPU side, right? I mean, I think agentic computing and particularly things like computer use and tool use is just so CPU intensive. And I think that core Epic franchise can really shine. And the other, I really think this is the year of instinct and the GPU series. And I remember our call last year and I said, I think this is the year. And you said, dumb, you have to be a little patient. And then now this is the year, you know, 2026 is the year of the GPU. So maybe we talk about those agentic workloads, particularly for, you know, computer use and CPUs, but then also the rise of the MI series. [00:18:05] Speaker 1: Yeah. So this is, again, an example of how quickly things change and how dynamic the market is. So if you think about it, you know, we started this whole AI conversation with chat GPT and chat bots, and you asked questions and it gave you answers, they seemed good. And then, you know, you quickly have migrated. And this year has really been, in my opinion, the year of agents. Yeah. And the idea that the technology can get good enough that it can complete a full task, not just ask a question, but actually give it a task, whether it's, you know, book your plane tickets or it's, you know, helped me find flaws in my software and that whole range of things. [00:18:46] Dom: Write your AMD earnings note, write your post earnings note. I've been using it a lot. Have you been, okay, all right, great. [00:18:53] Speaker 1: And what that really needs is that needs all kinds of compute. So CPUs now become super important. You know, the CPU market used to be a market that would grow, you know, maybe single digits. You know, it was a $20 billion market, $30 billion market. You know, we're now expecting that over the next four or five years, it will be a $220 billion market. So it's incredible growth. Yeah, wow. And it's because, you know, these agents actually need somebody to orchestrate the various workloads. And that's what CPUs do. So I really like our position in the CPU market today. We just launched, you know, we call all of our products, you know, Italian cities. So we've gone from Genoa to Turin to now Venice. We just launched extreme capability, like best performance, best performance per watt, best performance per dollar. The set of workloads are evolving. So we really like that market. And then you say, you know, the year of instinct, we just started shipping our newest MI-455 and Helios racks in the third quarter. They perform extremely well. They perform extremely well. We're sitting in, you know, OpenAI, Anthropic, Meta, Microsoft, Oracle data centers. They're bringing those up. And my point being that the technology cycle is not getting slower. You know, if anything, the technology cycle is getting faster. It really gives us the motivation to push the envelope because you get so much benefit each cycle in terms of performance per dollar. Like what we're trying to provide for the market is, you know, better AI at lower cost, more broadly available. Yeah. I mean, those are the three pieces. And I think we're doing that. [00:20:45] Dom: Definitely. You know, the theme of the conference beyond democratizing finance is this concept of how do we make AI better for, you know, ordinary Americans? And just where do you see the biggest opportunities for AI to improve American lives over the next three years? And what is AMD doing to make that happen? [00:21:04] Speaker 1: Yeah. I'm really passionate about this. I think this is something that we all have to work on, which is showing the power of the technology. I'm exceptionally, you know, I grew up like an engineer scientist, always love to solve problems. And my biggest hope for AI is that we show the typical American how much AI can improve their lives through scientific discovery, through health care, through better outcomes, because that's something that is very near and dear to all of our hearts as we as we think about how you interact with technology. And from an AMD standpoint, you know, we are big believers in partnering and developing compute for these types of workloads. So we have developed significant compute. You know, this is this week is also the supercomputing conference. Yeah. So there are lots of new announcements around supercomputing. But we continue to push the envelope in that area with the national labs. And so, yeah, I hope a couple years from now, every one of us has a story as to how AI has improved the outcome for your family, because I think those are stories that can be told over and over again. And we should be very proud that we're in an industry that can make, you know, those types of outcomes happen. [00:22:21] Dom: Okay, I'm going to do what I do in our callbacks and sneak a final one in, Lisa. And I know we're at the end here. You know, we've talked about CPUs. We've talked about GPUs, FPGAs, and kind of an odd one to end on. But it's always something that intrigued me when you did Xilinx. You kind of, you got so many talented people. You know, that industrial market's kind of been an inventory correction for a while now. Just how are you thinking about that business going forward? [00:22:50] Speaker 1: I think the great thing that we've learned about this market is you do need all the different pieces of compute. We acquired Xilinx back, I think, in 2022. It's a great deal. It was a very large semiconductor acquisition at the time. Some people said, you know, Lisa, are you sure you can integrate something of that size given where AMD was? And I think what we have found is a real formula for one plus one is greater than three. Because what we've taken is a franchise that had 6,000 plus customers in lots of different places, including aerospace and defense, industrial, test and measurement, equipment, all of these things. And we've integrated the full AMD roadmap with that. And so, yeah, we're seeing tremendous growth in that part of the business as well. It is back to the motto of you need all kinds of compute. Because in those data center environments where you have CPUs and GPUs, you also have lots of FPGAs, you have lots of NICs, you have lots of controllers. And, you know, we want to make sure that all of that gets built out together. [00:23:51] Dom: And now you have all the right pieces. Well, thank you so much, Lisa. Thank you, everyone, for your time. [00:23:55] Speaker 1: Wonderful. Enjoy the conference.

Transcribe Any Video or Podcast — Free

Paste a URL and get a full AI-powered transcript in minutes. Try ScribeHawk →