Possible: Lisa Su on semiconductors, compute, and big bets

0:00Starting a business comes with its share of ups and downs, which is why staying true to your vision is essential. a non-negotiable for Romeo and Milka Bregali, Capital One business customers and co-owners of Ra's plant-based restaurant in New York. Romeo and Milka took a leap of faith when starting their own restaurant, gutting an empty space and building it from the ground up. Every pipe, every wall, every detail. But building from scratch came with a heavy financial burden, which is when they turned to their Capital One business card. With the flexibility of the card's no preset spending limit, they were able to spend more and earn more rewards while bringing their vision to life.

0:36Today, Raz's success is proof that with passion and the right support, it's possible to make your dreams a reality. Learn more at CapitalOne.com slash business cards. If you've spent any time building AI products or leading technical teams, you know this. Transformation doesn't fail because of ideas. It fails because teams can't move together. Enter Atlassian's Teamwork Collection. It has planning in JIRA, documentation in Confluence, video updates in Loom, and now AI agents in Rovo, which connects the dots across your work so nothing gets lost. It's one AI-powered teamwork platform designed for how modern teams actually build.

1:21Learn more at Atlassian.com slash TeamChanger. That's A-T-L-A-S-S-I-A-N dot com slash TeamChanger. So when we started, people were wondering, well, what is AMD going to be when they grow up? And, you know, our driving force at the time was to decide what we could be best at. And I really believed that we could be the best in high performance computing. Back in the day when I would sit with our technical fellows and we would make that big decision of, hey, are we going to bet the company's roadmap on chiplets? And we said, yes.

2:07Hi, I'm Reid Hoffman. And I'm Aria Finger. We want to know how together we can use technology like AI to help us shape the best possible future. We ask technologists, ambitious builders, and deep thinkers to help us sketch out the brightest version of the future. And we learn what it'll take to get there. This is possible.

2:33If AI is the engine of the future, then computer chips are its crude oil, vital, contested, and entangled in geopolitics. The next frontier of intelligence depends on the hardware we design, the energy we can harness, and the infrastructure we can scale. Joining us today is a visionary CEO who understands this perhaps better than anyone, Dr. Lisa Su. Lisa is chair and chief executive officer of the U.S. semiconductor company Advanced Micro Devices. When Lisa took the helm in 2014, AMD was on the brink of bankruptcy. Her commitment to semiconductors and high-performance computing set forth what is arguably one of the most impressive turnarounds in Silicon Valley history.

3:23Over the last decade, AMD has launched game-changing new server processors, and its market cap has risen from around$2 billion to nearly$300 billion. Today, half of the world's top 10 supercomputers use AMD chips. The company is a global powerhouse. And Lisa, a former engineer with three degrees from MIT, is regularly featured on lists of the world's most powerful people in AI and in business. She was Time Magazine's 2024 CEO of the Year. We sat down with Lisa to talk about CHIPS, the AI arms race, and what it will take to power us into this next stage of our AI future, from the global importance of AI to its implications for human medicine.

4:08Here's our conversation with Dr. Lisa Su.

4:20Lisa, I'm so glad this has worked out. Welcome to Possible. I've been looking forward to this for months. Well, thank you, Reid. It's great to be here and great to be here with you and Aria. So you've described yourself as someone who has always liked hardware and building things. So tell us more. Where does this kind of curiosity and drive come from? Yeah. So, I mean, just a little bit of background. So I was born in Taiwan and I integrated to the United States when I was three. And so my dad was really into math and science and those kinds of things. And so when I was a kid, I just found it really interesting to understand how things worked.

4:56my younger brother had a remote controlled car that would go you know sort of up and down our hallway and and like one day it stopped working and I was like huh like why would it stop working and so you know as a curious kid you unscrew it and you realize that there's a loose wire and like if you connect the wire it starts working again and that kind of just gave me this feeling of like it's nice to work on things where you know there's cause and effect right like you you do something and all of a sudden you can make something work or you can do something that sort of is unexpected. And that's kind of what got me into hardware.

5:31And when I was a young college student looking for a job, I had a chance to work in a semiconductor lab. And I think the rest is history. Well, the next time my kids break their remote control cars, I'm going to be like, if Lisa Sue can fix it, just get in there, open it up. And I have a feeling that that experience probably led to my next question, which is, I've heard you say that some people said you were crazy to go into chips instead of hardware. So can you take us back to that time and that decision making? And did it feel crazy to you? Well, so, you know, back to, it seems like, you know, ages and ages ago when people would say, you are interested in semiconductors.

6:14And they would be like, what are semiconductors? Like people didn't really know what semiconductors were. And I would say, hey, I work on chips. And then people would say, what kind of chips? Like you mean like potato chips? And I'm like, no, no, no, no, no, no. I mean, semiconductor chips. And so, yeah, I mean, look, you know, 30 years ago, now I'm dating myself. People didn't talk about semiconductors. People didn't think about what they were. And for me, you know, I got my bachelor's, master's and PhD at MIT in electrical engineering. And what was really important is that you work on projects and you work, you know, I had a research thesis and all of that.

6:49And so I worked in a semiconductor lab. I built devices. They were like one micron devices, which seemed like, you know, eons ago when you were talking about one micron. And now everyone's talking about like, how do we go to two nanometer? So a lot has changed in that timeframe, but I thought it was amazing that you could build devices on a chip, you know, the size of like a dime or, you know, a quarter, and it could do things like computing. And that's kind of what got me into semiconductors. And yes, it was a time when semiconductors were not interesting. They were not sexy. Nobody talked about them.

7:24But for me, they were kind of interesting because I could see the product of my work actually turn out to make something run, like a computer, which was kind of very exciting for me. Absolutely. And I think you're right. It was like the time before chips were cool. And so we're going to get to the present, but from your early days as an engineer to today. Can you talk to us about those career chapters? You know, I would say that when I started, I was an engineer's engineer. So, you know, I was a device physics person. I really enjoyed dressing up in a bunny suit and going into the lab and, you know, just imagine me in a bunny suit.

8:01I love it. Going into the lab. Hi, Google Gemini here to provide some context. A bunny suit, more formally known as a cleanroom suit, is a full-body garment worn by engineers and technicians in highly controlled environments called cleanrooms. The name comes from the fact that the one-piece suit, with its integrated hood, can make the person wearing it look a bit like a cartoon rabbit. I would do experiments on my chips. I would go and test my chips. It was very exciting. I spent a lot of the early part of my career at IBM, and I think IBM is just an amazing company. I think as you've looked at the contours of IBM and now under Arvind Krishna, who's doing just a fantastic job, you could learn everything at IBM.

8:45You know, I went in as a 25 year old and it was 300 ,000 people. And you're thinking like, how can little old me make a big difference in a company of that size? But you realize that at the end of the day, the beauty of engineering is it actually matters what you do. It matters less what your title is. It matters less how many years you've been with at the company. and it matters more what ideas that you have. And so I had a chance to really work on some exciting projects. At that point in time, I worked on some of the first copper interconnects when we were integrating copper into semiconductor labs.

9:17And I kind of grew up at IBM. So I say like, you know, every two years, I got to do something different. So I was, you know, first a device physics person, then I became sort of a manager of a product team. And then I started learning something about business and sort of how do you bring products to market. And I think that brings me along to where I am today, which is, I like to say, doing probably the most fun thing I can imagine doing, which is being at the leading edge of technology in an industry that makes the world a better place. Today, massive conversations around AI, AI hardware. We tend to hear about two major powerhouses, AMD and NVIDIA.

9:55For people who aren't necessarily up to speed on hardware, let's hear direct. What are the ways to understand AMD and what its role in this is and how it's different than NVIDIA? Yeah, absolutely. So, Reid, if I can just take a little bit of a detour. You know, AMD is a company with a very rich history. We've been around for 55, 56 years. We've always been in the computing world. and it's always been around pushing the envelope on computing. So if you ask, what do we do when we wake up in the morning? It's about, you know, how do we push the envelope on high performance computing? And it turns out today that high performance computing is all about AI and how we can get AI infrastructure to a point that, you know, more people can use AI in their businesses and in their daily lives.

10:40And so what we do is really all kinds of computing. Like I like to say billions of people touch AMD hardware every single day. And you touch our hardware in the cloud. So if you're running any of the basic services, probably on the podcast right now, today, we're running through some AMD hardware in the cloud. Many of the most important services that we have, whether you're talking about Teams or you're talking about Zoom or you're talking about Instagram or you're talking about if you call an Uber or any of those things, somehow go through AMD hardware. We're also in things like Sony PlayStation or Microsoft Xbox.

11:17You know, those are also AMD hardware. And what I'd like to say is, you know, our goal is to make sure that we have the right compute at the right time. You know, your comment about what separates hardware companies. We are actually here to provide the entire solution capability, Reed. It's not just about hardware, but it's about how do we accomplish things, whether you're trying to process as much data as possible in the data center or you're trying to have the best client experiences in a PC or in another device. We have the right computer for the right application. And I think that's very core to our mission in life.

11:52So I love that you have described AMD's mission as solving the world's most important problems and challenges. And like Reid and I, you are an AI optimist who's excited about it. So if you could have AI help solve just one of the world's most important problems, what would it be and what do you think the impact could be? Well, first of all, I feel like I'm talking to the experts here, you know, certainly Reid being very prolific on this topic. I'm optimistic about AI for many, many things. And I view AI as really a multiplicative factor to what humans can do, right? I mean, the fact is, you know, humans are amazing.

12:28You know, we have made so much progress over this period of time, but now we can use computing to kind of add an X factor. Perhaps, Aria, I could say, look, there's so many things that AI can do for businesses. We're certainly using it throughout AMD for lots and lots of things. We also look at it as what it can do for personal productivity. But probably the place that I'm most excited about is what AI can do for healthcare and medicine, really for quality of life. Right. Because, you know, I think about, you know, the things like money can't buy, like money can't buy health. But if computing can help improve health outcomes and, you know, AI is such a big part of that.

13:10I think it's just amazing. I mean, it goes back to this notion of, man, we're working in like the most fun industry, you know, on the face of the planet. We'll get back to AMD because obviously very important part of exactly, as you mentioned, the infrastructure of everything that's happened in compute, you know, in the cloud and many other places. You know, I actually have a couple of Microsoft devices that are powered by AMD. But on the healthcare, I do think this is actually one of the things that people don't quite realize about how much of an implication there would be. Everything from like a medical assistant that's 24-7 available on your smartphone for your kid, your parent, your sibling, your partner, et cetera, to obviously what I'm doing with Manus AI, with Siddhartha Mukherjee, which is how do we create a drug discovery factory using AI to accelerate the cure of cancers?

13:58What are some of the things that people should understand as kind of the near and medium term possible futures with health care? What are some of the things you're seeing from your perch at the center of this with AMD? And why should people be excited about this? I think, Reid, it's a great point. I think it's all of the above to some of the examples that you've given, right? So very near-term examples. I think, you know, AI can actually improve health care outcomes for people like today. And when you think about the diversity of healthcare that exists, even just across this country, whether you're at the best hospital versus you're in some rural community where you don't have the best doctors and you don't have the best capabilities, to be able to equalize some of those healthcare outcomes with the usage of AI.

14:44You know, one of the things that I found most interesting about medicine, just kind of spending some time in it, is it's actually more art than it is science, right? Early diagnosis of things like cancers or other things really help improve healthcare outcomes so much. And you can't count on the same quality of service everywhere you go, but being able to use AI as an assistant. The best wording that I had was a conversation I had once with Eric Corbett when he talked about like, you know, what do you expect of a AI medical assistant? Like you wouldn't say that you would want your AI capability to make all of the decisions because we wouldn't trust that capability.

15:24But boy, if you had AI just equalize, call it a very strong capability to give you a second opinion, you know, right on the spot. Like that's just a very easy example right now. to the much further out examples like, you know, the one that you talked about, which is, you know, drug discovery. I mean, like to help take what might take 10 years of experimentation with sort of normal traditional computation to reduce that to weeks and months so that we can really accelerate the rate and pace of innovation in this area. I think those are all really interesting outcomes. And perhaps one that I'm personally passionate about is I've experienced like in the health care system, you actually see a lot of cases where we have specialists in medicine.

16:08But when you think about the human body, the human body is not a set of organs, right? It's not, you know, sort of your kidney specialist and your heart specialist and your it actually is an integrated system. And to find people who could integrate all those systems is actually pretty hard. And I actually think that that's where AI can help us quite a bit in terms of integrating knowledge across a vast amount of data and a vast amount of experience to give us the best knowledge possible. And actually, one particular follow up that I think is really important that everyone pay attention to in your answer is today already, it's a great second opinion.

16:43So if you're countering anything that's serious, you check it as a second opinion. By the way, if the two don't match, then go to a third, right, as a way of doing it. And that's already an amplification for what we have today. I thought that was a really important thing for people to understand the AI moment that's already today. Yes, yes. And the other thing I think we find, Reid and Aria, and you've probably experienced this as well, is it's getting better practically on, you know, like a weekly basis, right? So the AI that you might have experienced six months ago will be different than the AI you will experience six months from now because there's this constant learning in process.

17:23And that's exactly what we would want for, you know, like I said, all of these applications, but particularly in the medical field. So, Lisa, something tells me that you're not a person who yells from the rooftops about your successes. But I would be remiss if I didn't mention because the turnaround that you have had at AMD is like truly incredible. You went from revenue of$4 billion to more than$23 billion in your time as CEO when you joined in 2014. And you saw shares go from$2 to$200. And so you can't have that amazing turnaround without making some risky decisions, making some bets, betting the company.

18:01Were there times that you thought if this bet doesn't pay off, like I'm betting the company here or I'm taking a risk that's necessary, but it's still a big risk in that 10 year trajectory? Well, 10 years sounds like a long time, Aria, but actually in tech, it's not that long. Right. It really is a few product cycles. And look, I, you know, I've said this before and I'll say this again. AMD is like a dream job for me because I'm getting to work on the most exciting technology in the industry with amazing people, amazing customers, partners, all of that stuff. And so when we started, we were probably about$4 billion in revenue.

18:42And people were wondering, well, what is AMD going to be when they grow up? And for me, I think the most important thing for us to do was to decide what we wanted to be best at. Because I really do believe for tech companies, the key is you have to be best at something. It is, you know, there's a whole bunch of other things that are important. You need the right balance sheet and you need the right, you know, all of those things. But, you know, our driving force at the time was to decide what we could be best at. And I really believed my CTO, Mark Papermaster, has been my partner on this journey.

19:15We believe that we could be the best in high performance computing. And at the time, high performance computing wasn't as exciting as it is today because AI wasn't as big as it is today and the cloud was, you know, kind of on a different path. And so for high performance computing, CPUs, GPUs, the right compute at the right time, that was what we were going to be best at. And the world was also changing. Reed knows this very well, right? Whenever there are technology inflection points, it actually brings opportunities. And so we were in a place where, you know, the world in semiconductors had for 30 years really was driven by Moore's law and the idea that every two years or every 18 months, you can double the productivity.

19:55And so it was all about scaling, scaling, scaling, scaling. but scaling was changing and we saw that change. And yes, we made a few bets. We made the bet that scaling would slow down and we had to innovate in different ways. And perhaps one of the largest innovations is we were the first to really put high-performance computing in sort of breaking it up into what we would call chiplets such that you could take advantage of putting lots and lots of chips together to become a higher-performance computer. And if you look at every high-performance CPU today, if you look at future generation AI GPUs, they're all going to use this technology.

20:36But it is always about making bets. And yes, Aria, there were those meetings back in the day when I would sit with our technical fellows and we would make that big decision of, hey, are we going to bet the company's roadmap on chiplets? And we said, yes. It's trust, but verify, right? We had a lot of learning. I mean, I can say there were products that worked really, really well, and there are products that worked not so well, but we learned a ton from the process of that. In everything that I've read about you, you seem very patient. You're saying, hey, we have a three to five year plan. You know, this to your point, like a decade is nothing.

21:15How do you get everyone else to come on board for that when there's so much noise and there's fads and there's we should be going left because our company is going right. Like, how do you get everyone else on board for that longer term vision? Well, I think, Aria, you have to have conviction. Yeah, there are all kinds of fads. I mean, we see all kinds of fads in even in tech, right? When you think about what people went through, like when we started, there was a lot of conversation on, hey, Lisa, you know, AMD should be making smartphone chips. And then when you look at where things have evolved, you actually realize that companies have sort of fundamental DNA in terms of what they're good at.

21:53Like, we are just really good at building big computers. Like, that's what we think about every single day. You know, we've been working on AI for the last, you know, seven or eight years, I think. You know, AI didn't just show up with ChatGPT. AI was around for a long period of time. The industry didn't quite figure out how to make AI accessible enough. And I think that's the difference with large language models is now quite accessible to a much broader population across a much broader set of applications. And it does require making big long-term bets. I do like to say that the decisions that we're making today, we will know three to five years from now whether we've made the right decisions.

22:32And that's okay. That's actually good. The rate and pace of AI, though, is different. And I'm certainly curious to see how you feel about it, Reid, but certainly in some of the stuff that you've talked about. I think the rate and pace of AI is faster than any other thing that I have seen in my career. I think it's because you are allowed to experiment much more. Like hardware is on three-year cycles. Software can be on three-week cycles. And as a result, although we provide infrastructure for this, the tremendous amount of innovation on the application layer and on the usage models, I think, is what makes AI quite different and actually quite exciting.

23:14By the way, I think it's not only going to be three week, I think it's getting the three day and potentially even three hours. I mean, I think part of your chiplet thing is exactly right. The AI revolution is how do we harness large scale compute together with large scale data, together with teams putting that together. And that's the kind of revolution that we're in. And obviously it becomes learning machines, learning compute machines. So, you know, you said the word, the early innings of a 10-year AI super cycle. Give some more color to that. What are the things that you're seeing from the central vantage point of AMD coming, things that we need to navigate, et cetera?

23:52Yeah. So I spend an enormous amount of time with, you know, sort of customers and partners. So just understanding, you know, what they're seeing, really understanding where the use cases are going. And I guess what I would say is as much progress has been made over the last two years, AI is like now part of our every day, every hour vocabulary, right? Everybody talks about it. Whether you're talking about, you know, forget tech. I mean, of course, tech, we talk about it all the time. But you think about investors, you think about people on the street, you think about everybody uses AI in their life.

24:26In my mind, it's still so early because the technology is good, but is not yet great. Right. I think it's gotten a lot better, but it's still like we can see the potential is nowhere near where it could be. And, you know, I view that across all application bases. Right. So if you think about, you know, sort of the evolution from large language models now to the usage of agents and we're all excited in companies with the idea that we're going to augment our workforce with agents. that can do a lot more, let's call it collective tasks so that they're not single task-oriented things. But we're still at the very early innings of implementing these things.

25:07Where we see a lot of promise is just in building chips, for example. We think we can build much better, much faster, much more reliable chips over time. We can apply AI to our design elements. We can apply it to our test elements. We can apply it to our quality elements. And we're just at the beginning phases of doing that. And the cases that we see today that are probably most interesting is we're using AI to write a ton of our software. One of the things that people would say about AMD is, hey, we've been a great hardware company, but there's a lot of software infrastructure on AI that's been written in CUDA and needs to be, let's call it, translated to AMD.