AI transcript
0:00:02 Artificial Intelligence is just the latest
0:00:05 in the wave of technologies to increase the demand
0:00:08 of high-performance chips, but it’s not the only.
0:00:10 The last decade was filled with the proliferation
0:00:14 of cloud, gaming, social networks, IoT devices, and more,
0:00:18 which led to some pretty incredible growth stories.
0:00:19 Many are familiar with NVIDIAs,
0:00:23 as it skyrocketed to the top market cap stock in the world
0:00:26 earlier this year, and as of this recording,
0:00:29 sitting at over $3 trillion.
0:00:31 But let’s not forget other semiconductor companies,
0:00:34 like TSMC, ASML, or Broadcom,
0:00:36 each roughly growing by an order of magnitude
0:00:38 over the last decade.
0:00:42 But today, you’ll get to hear about a 50X growth story
0:00:45 over the same time period from Lisa Su,
0:00:47 the incredible woman who’s been at the helm of AMD
0:00:49 throughout that run.
0:00:52 For now, I’ll pass it over to A16Z growth general partner,
0:00:55 Sarah Wang, to properly introduce this episode.
0:00:59 – As a reminder, the content here
0:01:00 is for informational purposes only.
0:01:03 Should not be taken as legal, business, tax,
0:01:05 or investment advice, or be used to evaluate
0:01:07 any investment or security, and is not directed
0:01:11 at any investors or potential investors in any A16Z fund.
0:01:13 Please note that A16Z and its affiliates
0:01:15 may also maintain investments in the companies
0:01:17 discussed in this podcast.
0:01:19 For more details, including a link to our investments,
0:01:22 please see a16z.com/disclosures.
0:01:25 (upbeat music)
0:01:28 (upbeat music)
0:01:31 – Hey guys, I’m Sarah Wang,
0:01:33 general partner on the A16Z growth team.
0:01:36 Welcome back to our AI revolution series,
0:01:37 where we talk to industry leaders
0:01:40 about how they’re harnessing the power of generative AI
0:01:43 and steering their companies through the next platform shift.
0:01:45 We have a very special guest this episode,
0:01:48 Lisa Su, chair and CEO of AMD.
0:01:51 Lisa is one of the most impressive CEOs in history.
0:01:53 Since she took over the helm of AMD,
0:01:56 the value of the company has grown over 50 times
0:01:59 to a market cap of 250 billion
0:02:01 as of the recording of this podcast.
0:02:03 Even more importantly, Lisa and AMD
0:02:06 are democratizing the benefits of gen AI.
0:02:09 Thanks not only to AMD’s top notch chip design,
0:02:11 but also through an open ecosystem
0:02:13 that allows developers to build AI tools
0:02:16 across a huge range of use cases.
0:02:19 It’s no surprise that Lisa is at the helm of this effort.
0:02:21 She’s been an innovator in high performance compute
0:02:23 across her entire career.
0:02:26 Starting from Texas Instruments to IBM and FreeScale
0:02:28 before joining AMD.
0:02:30 This pod is an especially exciting one
0:02:32 because Lisa is joined by none other
0:02:35 than A16Z operating partner, Bob Swan,
0:02:37 who was most recently the CEO of Intel,
0:02:40 one of AMD’s fiercest rivals.
0:02:43 It’s a huge treat to sit in on Bob and Lisa talking shop
0:02:46 and reminiscing about their time together in the industry.
0:02:47 In this wide ranging conversation,
0:02:50 Lisa and Bob cover the state of the art in AI compute,
0:02:52 the chip supply chain,
0:02:54 the role of ecosystem partnerships,
0:02:57 and where Lisa thinks AI is evolving from here.
0:02:59 Let’s get started.
0:03:01 (upbeat music)
0:03:08 – It’s great to have you here
0:03:11 and thanks again for spending some time with us.
0:03:12 – Thank you so much, Bob.
0:03:13 It’s great to be here with you.
0:03:14 – Cheers.
0:03:15 Let’s jump in on the Inquisition.
0:03:16 – All right.
0:03:20 – So 12 years at AMD,
0:03:23 10 as the CEO,
0:03:26 tell us a little bit about your career journey
0:03:28 and how you got to AMD, if you will.
0:03:29 – I grew up as an engineer.
0:03:31 Engineer at heart went to a school
0:03:33 and semiconductor devices
0:03:38 and really did the majority of my early career at IBM
0:03:41 doing R&D around devices.
0:03:44 And then as you think about sort of fun things to do
0:03:45 in the world,
0:03:47 I was always fascinated with the idea
0:03:50 that the work that you do in chips
0:03:52 is such that you can influence
0:03:54 really the way so many things.
0:03:56 Like technology is so important.
0:03:58 And so I just loved being at the forefront
0:04:01 of high performance computing and computing all these years.
0:04:04 And that brought me to a free skills semiconductor
0:04:06 for five years where I was CTO for a while
0:04:08 and then to AMD 12 years ago.
0:04:12 Like I used to say when I would tell people what do you do?
0:04:14 Well, I build semiconductor chips
0:04:15 and people are like, well, what’s that?
0:04:17 Like, why should I care about that?
0:04:19 Is that important?
0:04:23 And now everybody knows what semiconductors are
0:04:24 and why they’re so important
0:04:26 and why they power everything in our lives.
0:04:30 And so that’s what’s fun about the industry that we’re in
0:04:32 and the fact that you’re able to do things
0:04:33 that actually matter in the world.
0:04:34 – Yeah.
0:04:37 If you think about when you first started
0:04:41 the role of compute in the grand scheme of things
0:04:46 relative to today was relatively narrow.
0:04:47 – That’s exactly right, Bob.
0:04:51 I think when you think about even the idea of PCs
0:04:53 and people using personal computers
0:04:54 and everyone needing a computer
0:04:57 and then everyone needing sort of a smartphone
0:05:01 and then everyone needing big cloud data centers
0:05:03 and now everyone needing AI.
0:05:05 I do think it has been an evolution
0:05:09 of how semiconductors and sort of the power of chips
0:05:12 have really infiltrated every aspect
0:05:14 of the business world, our personal lives
0:05:16 and for the good, right?
0:05:18 – We’re all much better because we have all this technology.
0:05:21 – And it’s almost like what would we possibly do
0:05:24 if we didn’t have all this technology behind our desk,
0:05:29 in our hands, in our car, wherever compute is happening
0:05:30 everywhere through all sorts of devices.
0:05:34 And then along comes this thing called AI
0:05:37 and it’s like compute, it’s everywhere.
0:05:41 Can you just talk about how you see AI,
0:05:43 it’s relative importance.
0:05:46 And then over the longer-term horizon,
0:05:47 where is this gonna take us
0:05:49 and where are you and AMD gonna take us?
0:05:53 – Yeah, I think as you think about all of the various
0:05:56 large technology discontinuities
0:05:58 that we’ve seen over the last 30 years,
0:06:00 they’ve all been super important.
0:06:02 They start small and they really influence
0:06:04 every way we experience technology.
0:06:06 I think AI is probably the most important one.
0:06:11 I’d like to say over the last 30 years, 40 years, 50 years
0:06:14 because it’s something more than just technology.
0:06:17 I mean, it really is, you know, AI becomes the ability
0:06:21 for us all to become smarter, more productive,
0:06:24 really utilize the incredible data that’s out there
0:06:25 to help us move forward.
0:06:29 So I really see AI, we’re just at the very beginning
0:06:32 of the AI arc and it’s an opportunity for us
0:06:35 to take technology to yet a different level.
0:06:39 And for us at AMD, my belief is AI is gonna be everywhere
0:06:40 and every product that we build,
0:06:43 but importantly, it’s at the foundation
0:06:46 of what enables all of these great applications.
0:06:49 And so, yes, we’re all building AI compute these days.
0:06:51 We’re trying to build it as fast as possible
0:06:54 so that we can have all of those smart developers
0:06:57 really take advantage of the technology.
0:06:58 – And as you said, it’s fascinating
0:07:02 because in some ways, AI has been around
0:07:05 for such a long period of time.
0:07:08 And while new technologies and innovations
0:07:10 have a tendency to start slow,
0:07:14 this one has moved pretty fast.
0:07:16 – You’re right, it was always around
0:07:18 and it was always something that we thought
0:07:20 had a lot of potential.
0:07:23 But frankly, AI before generative AI
0:07:25 was somewhat hard to use.
0:07:28 And so it took experts to really unlock the technology.
0:07:32 I think the chat GPT moment as we all remember it
0:07:35 was the moment that AI became easy.
0:07:40 We could all talk to our computers and ask it questions.
0:07:41 And yes, it is nowhere near as perfect.
0:07:42 I mean, we have so much work to do.
0:07:44 We’re still very early.
0:07:48 But the fact that we can make technology now so accessible,
0:07:53 I think is what makes this generative AI arc so interesting.
0:07:55 And it’s what’s accelerated the adoption.
0:07:58 – But for AMD, you’ve always been
0:08:02 a high performance compute company.
0:08:06 As you think about the intersection of the company
0:08:08 and what it’s meant for the industry
0:08:11 and then the overlay of AI,
0:08:16 do you see any commonalities with the internet mobility,
0:08:20 just the commonalities that positions you so well
0:08:22 to capture this opportunity?
0:08:25 – Yeah, well, if I just go back a little bit,
0:08:28 when I first took over as CEO of AMD,
0:08:30 it was like 10 years ago now,
0:08:32 it was really a moment where we were like,
0:08:34 what should we be when we grow up?
0:08:36 And if you remember back in those days,
0:08:40 this like 2014, the whole craze was around mobile phones
0:08:43 or tablets, like everybody was into that kind of thing.
0:08:45 And my board even asked me, well,
0:08:48 at least AMD can’t not be in tablets, right?
0:08:52 And I said, well, I’m actually not sure that’s our specialty.
0:08:55 Our specialty is around high performance computing.
0:08:58 We build big things, as one says,
0:09:00 everyone has to know what they’re best at.
0:09:01 And that’s what we’re best at.
0:09:04 We’re best at building large complex microprocessors
0:09:08 or GPUs or with our acquisition of Xilinx,
0:09:10 adaptive and embedded computing.
0:09:13 And when you look going forward,
0:09:16 you see that high performance computing
0:09:19 is really important in the industry in so many places.
0:09:22 And it is at the heart of what makes AI possible.
0:09:24 ‘Cause if you think about what makes AI possible,
0:09:27 it’s the ability to train these models
0:09:29 with hundreds of billions of parameters,
0:09:32 trillion parameters, so that they become ultra smart.
0:09:34 And then we can ask it all these questions
0:09:36 and it gets most of them right.
0:09:38 You need high performance computing at the heart of that.
0:09:41 So it is a great feeling to be in…
0:09:42 – Good place to be.
0:09:43 – Yeah, well, it’s a great feeling to be in a place
0:09:46 where you know that the technology that you’re building
0:09:49 can really push the envelope
0:09:51 on what can be done in the industry.
0:09:53 – And do you see over time,
0:09:58 will it always be the highest performance chip
0:10:00 that is gonna be the differentiator,
0:10:04 or is it gonna evolve for the different workloads
0:10:08 and the different multimodalities in the AI world?
0:10:10 – Yeah, it’s a great point and a great question.
0:10:13 I am actually a believer in you need the right compute
0:10:15 for each form factor, for each application.
0:10:18 So right now, there’s a lot of energy
0:10:22 around building the largest language models
0:10:25 and really this large GPUs
0:10:27 that are being used for training and inference.
0:10:29 But I do see if you look whether you’re at the edge
0:10:32 with embedded applications, industrial applications,
0:10:35 automotive applications, medical applications,
0:10:38 or you’re even at the client level in your PC or your phone,
0:10:40 you’re gonna need different types of AI.
0:10:42 And so you’ll have different engines for that.
0:10:43 And that’s good.
0:10:45 I mean, that’s what spurs all of the innovation
0:10:47 that’s happening around the industry.
0:10:50 – The other thing that I found maybe most fascinating
0:10:54 about the semiconductor industry is the ecosystem
0:10:56 and the importance of the different players.
0:11:00 As you think about development of product,
0:11:04 how do you deal with both what you need to do,
0:11:07 but also the interdependencies with the ecosystem
0:11:10 on delivering something requires a bunch
0:11:11 of different players?
0:11:13 How do you think about that in the context
0:11:18 of product development, but to get products to market?
0:11:20 – Well, I don’t think there’s any one company
0:11:21 that can do it all.
0:11:24 I mean, at the end of the day, we all have specialties
0:11:25 and they’re things that we’re good at,
0:11:28 but the opportunity to closely collaborate
0:11:30 and partner is so important.
0:11:33 We’re a big believer in open ecosystems
0:11:34 and industry standards.
0:11:38 And the idea that, hey, I’m building these great processors,
0:11:41 they should connect to other people’s networking
0:11:43 and we should be able to interoperate together.
0:11:46 The software ecosystem is super important too.
0:11:48 Developers shouldn’t have to develop
0:11:49 for one company’s hardware.
0:11:52 Developers should be able to develop what they need
0:11:54 and be able to use what’s the best hardware underneath.
0:11:57 So I think that’s part of the evolution
0:12:00 of an open ecosystem so that we can get
0:12:02 the best innovation out there.
0:12:04 – So in this constant evolution,
0:12:07 and we’ve seen this debate over time,
0:12:11 closed garden, interoperable,
0:12:14 what is gonna be the predominant winner
0:12:17 if there is such a thing in an AI world
0:12:21 on open interoperable technologies and interfaces?
0:12:23 – I’m a big believer in open ecosystems.
0:12:26 Interoperable is really important.
0:12:29 Closed walls usually end up being a problem.
0:12:32 If you look at sort of the technology arcs of time
0:12:35 and in this world where technology is moving so fast
0:12:37 and whether it’s a new model
0:12:40 or a new hardware technology or new capability,
0:12:43 you wanna make sure that it’s interoperable.
0:12:45 – Along the same lines, you and other players
0:12:47 in the industry recently announced
0:12:51 the new ultra accelerator link
0:12:53 and the ethernet link standards.
0:12:56 Is that an example of how you think about opening,
0:12:58 how you engage with the ecosystem?
0:13:00 – Yeah, I think it’s a great example.
0:13:03 We’re all looking at when you think about these large AI
0:13:06 clusters that you need in the future,
0:13:08 networking is such an important piece of it,
0:13:11 but you do want this choice as to
0:13:12 what hardware are you connecting,
0:13:15 what’s the processor, what’s the networking fabric,
0:13:16 what’s the overall system architecture.
0:13:18 So the ultra accelerator link
0:13:21 and ultra ethernet consortium are great examples
0:13:24 where competitors and peers can come together
0:13:27 and say, you know what, we’re gonna adopt open standards
0:13:30 and we’re each gonna innovate on top of that.
0:13:32 And those are two great examples
0:13:36 and it includes many companies that do compete,
0:13:37 but also can cooperate.
0:13:38 And if you think about that,
0:13:41 that’s exactly what an open ecosystem is supposed to be.
0:13:44 – And you talked about competitors,
0:13:46 industry players coming together.
0:13:50 The demand for compute over the last several years
0:13:52 has been maybe unprecedented,
0:13:56 just in terms of its pace and its distribution
0:14:01 and the ability to ramp up supply
0:14:04 to meet these incredible demands
0:14:07 when you throw in cycle times to put more capacity
0:14:12 in COVID supply chain disruptions.
0:14:16 Have these disruptions or challenges on the supply side?
0:14:20 Have they impeded your ability to move faster in some areas?
0:14:22 And what have you learned from this
0:14:25 that will make the next supply constraint
0:14:27 be a bit smoother for the industry?
0:14:30 – Yeah, when you look over the last four or five years,
0:14:32 probably the largest disruption
0:14:36 to the semiconductor supply chain was really around COVID.
0:14:40 It was the moment where everyone needed
0:14:42 more semiconductors at the same time,
0:14:43 which kind of wasn’t expected.
0:14:46 Usually what happens in the semiconductor market
0:14:49 is you’ll have one market up and one market down.
0:14:51 Mobile may be really hot,
0:14:54 but infrastructure will be down or vice versa.
0:14:56 What we saw in COVID was basically every market
0:14:59 at the same time had this concentrated effect.
0:15:01 And the semiconductor supply chain
0:15:04 is actually really good at meeting demand.
0:15:06 Actually, we usually overshoot.
0:15:08 As you know, Bob, that happens from time to time,
0:15:09 but it takes time, right?
0:15:13 It takes 18 months, 24 months to really put that all on board.
0:15:15 And so I think the industry as a whole
0:15:19 has done a good job at bringing more supply on board.
0:15:21 The more recent thing as it relates to AI
0:15:24 where it’s super hard to get GPUs,
0:15:26 that truly is, again,
0:15:29 nobody forecasted what generative AI would need.
0:15:31 And so it has taken some time
0:15:35 to really build all of this advanced packaging capacity
0:15:37 and high bandwidth memory capacity.
0:15:40 But again, the semiconductor supply chain is good at that.
0:15:42 And we just have to get a little bit better
0:15:45 at forecasting what long-term demands are.
0:15:50 – I do remember exiting 2019, entering 2020,
0:15:54 where that normal cyclical nature of the industry,
0:15:59 I think we’re all looking at expecting 2020 to come down.
0:16:03 And then COVID hitting for a short period of time,
0:16:05 it looked like things were going further south.
0:16:08 And all of a sudden, to your point,
0:16:09 everybody needed supply.
0:16:13 And in many ways, while there was issues materializing,
0:16:15 the way the ecosystem comes together
0:16:17 in the semiconductor industry
0:16:19 and the sophistication of the supply chain,
0:16:22 despite the challenges, is pretty impressive.
0:16:22 – That’s exactly right.
0:16:25 And I think we’ve all gotten smarter and better as a result.
0:16:30 I think this idea of, hey, let’s try to just eke out
0:16:31 every last penny in the supply chain
0:16:33 has gone a little bit of ways
0:16:37 to let’s build resilience into the supply chain.
0:16:38 When governments are asking,
0:16:40 do you have enough semiconductors?
0:16:42 I think that gives us permission
0:16:45 to really think more broadly about resiliency.
0:16:48 – And we’ll talk about resiliency and government asking.
0:16:53 I think it was roughly two years ago to the day
0:16:55 there was this thing called the CHIPS Act.
0:16:59 And for our audience, the government signs into law
0:17:04 a bill authorizing $280 billion to help in the design
0:17:11 and the manufacturing in the US of semiconductors.
0:17:16 And then $53 billion of that was authorized to be spent.
0:17:18 I know it’s still a work in process,
0:17:20 but as you think about the CHIPS Act
0:17:23 and some of the challenges from the last couple of years,
0:17:26 how do you see that helping resiliency
0:17:28 of supply chain going forward?
0:17:31 – I have to say I’m a big supporter of the CHIPS Act.
0:17:34 I never would have thought that five years ago,
0:17:37 semiconductors would be high enough priority
0:17:41 in the US government’s view of what needed
0:17:43 clear industrial policy.
0:17:45 Some people say, hey, it’s not enough
0:17:46 or does it make a difference?
0:17:48 I think it’s made a huge difference.
0:17:51 It’s made a huge difference because what it’s really done
0:17:56 is it’s put at the top of the priority list, resiliency
0:17:58 and semiconductor, both manufacturing
0:18:01 as well as research and development in the United States.
0:18:04 And of course, there’s much, much more work to do.
0:18:06 As you said, it’s a work in progress, but it’s a good thing.
0:18:07 It’s a good thing for the industry
0:18:09 that there’s a focus here.
0:18:12 I’m actually particularly excited about some of the work
0:18:14 that’s being done on the R&D side
0:18:17 because I think there’s a whole opportunity
0:18:20 to really train the next generation of leaders
0:18:24 who will lead the semiconductor research and development
0:18:26 as well as future capabilities.
0:18:27 So yeah, I think it’s a great thing.
0:18:29 And yes, it’s still early days.
0:18:32 And we need to make sure that every dollar is spent,
0:18:34 is spent for good reasons, and that we
0:18:37 get the return on investment on the other side.
0:18:40 But it’s a clear indication of how important semiconductors
0:18:44 are to the US and really to the global economy.
0:18:45 I couldn’t agree more.
0:18:50 The dynamics of– we’ve talked about the ecosystem working
0:18:52 together and the importance of the ecosystem.
0:18:54 To throw the government in there,
0:18:56 obviously it creates challenges.
0:18:59 But industry and government working together
0:19:01 to solve really big problems, I think,
0:19:04 is a real necessity in some areas.
0:19:07 And this is the one where I’m thrilled about the CHIPS Act
0:19:10 itself, but the deployment and the proof points,
0:19:11 I think, are still in front of us.
0:19:13 So it’ll be an exciting time.
0:19:17 And I hope that the challenges around resiliency will be–
0:19:19 Remembered?
0:19:21 Remembered or rearview mirror.
0:19:22 Yes.
0:19:23 Yeah, exactly.
0:19:25 Remembered is the best way to frame it.
0:19:29 At a time when innovation is happening all the time,
0:19:33 you still are a relatively long cycle development time frame.
0:19:35 Yeah, very long cycle, really.
0:19:39 How do you guys deal with long cycle development
0:19:43 with short cycle innovation and what inherent challenges
0:19:46 or opportunities that creates for you in the industry?
0:19:49 Yeah, I think the most important thing in our world,
0:19:52 especially in hardware, is one needs
0:19:54 to try to have a crystal ball.
0:19:57 You’re never going to predict the future entirely,
0:20:00 but you do need to be able to say, hey, these
0:20:02 are the disruptions that are coming up.
0:20:05 These are the things that we need to pay attention to.
0:20:08 Probably the best example that I can think of–
0:20:11 and this is one where we had a lot of debate internally–
0:20:14 is, what’s the future of Moore’s law?
0:20:15 That’s been debated just a little bit.
0:20:16 I remember those debates.
0:20:19 And by the way, I’m a believer in Moore’s law
0:20:22 has been extended so many times because people are super smart
0:20:25 and able to come up with different ways of extending
0:20:28 the same principle of more transistors, more capability
0:20:30 every couple of years.
0:20:32 But for example, that’s like advanced packaging.
0:20:35 And when do you go to 2 and 1/2D and 3D packaging?
0:20:39 And for us, we use this technology called chiplets.
0:20:40 We didn’t know.
0:20:43 We didn’t know at the time when we were making that decision,
0:20:45 was it going to be the right bet?
0:20:47 But we knew that we had to make that bet.
0:20:49 And you really don’t figure that out until three to five years
0:20:50 out.
0:20:52 So your question about how do you know?
0:20:54 You don’t know, but you try to make sure
0:20:55 that you’re betting in the right direction.
0:20:59 And then you have to be agile enough to adjust accordingly.
0:21:02 And that’s what this whole world of high performance
0:21:04 computing is about.
0:21:06 You talked about the right bets.
0:21:09 And you guys have had incredible success
0:21:12 on making the right bets.
0:21:14 What is the balance between how you
0:21:18 learn from your customers about the right bets to make,
0:21:21 but also how you lead your customers
0:21:22 given the development cycles?
0:21:25 How do you strike that balance at AMD?
0:21:29 Yeah, our top two priorities that I tell the company
0:21:30 all the time.
0:21:32 The first one is your tech company.
0:21:35 Our job is every day to wake up and build great products.
0:21:39 But we do that through having very deep customer relationships.
0:21:41 Because I really do believe that they go hand in hand.
0:21:44 Our customers are some of the largest,
0:21:47 whether it’s cloud manufacturers or OEMs or enterprises
0:21:49 in the world, that they see the problems
0:21:51 that they’re trying to solve.
0:21:54 And that’s where it’s most beneficial is talking
0:21:57 to our customers about, hey, what problems are you having?
0:22:00 What are you trying to solve two, three, four years out?
0:22:02 And then our technologists can really
0:22:04 come up with ideas for how to solve those problems.
0:22:08 So it’s not like it’s one for one where we listen
0:22:09 to everything people say.
0:22:11 But we do listen a lot.
0:22:14 Because that tells us that we’re working on the right things.
0:22:16 Because whatever we do, you want to ensure
0:22:18 that the technology you’re building
0:22:21 is something that will solve somebody’s problem.
0:22:23 Hyperscalar market.
0:22:25 Tremendous progress over the last several years.
0:22:26 Congrats.
0:22:30 Correlation, learnings from winning in hyperscalar
0:22:32 with this rapid growth from AI.
0:22:34 Is there learnings that you’ve been
0:22:37 able to extract from what it takes to win in one?
0:22:40 And then how do you translate to win it in AI?
0:22:44 So when we started in the hyperscalar market
0:22:46 with our first generation products,
0:22:49 our Zen product portfolio, I think
0:22:53 we were about maybe 1% share of the server market.
0:22:57 And actually, the whole idea of having deep partnerships
0:23:00 with customers is really, we needed
0:23:03 to be able to say that, hey, it’s all about roadmap.
0:23:06 Yes, it’s great, the product you have today.
0:23:08 But it’s all about, can you keep a sustained
0:23:12 level of constant innovation many generations out?
0:23:15 And I think we have made a lot of progress
0:23:16 in the hyperscalars.
0:23:19 I love the relationships that we have across the top brands,
0:23:23 whether it’s the Microsoft or Amazon, or Google, or Oracle,
0:23:24 or Meta.
0:23:26 It’s always about, how do we innovate together?
0:23:30 I think the AI arc is very similar in the sense
0:23:33 that these are big bets that the hyperscalars are
0:23:36 making on who their technology partners are going to be.
0:23:38 And we want to help them accomplish that.
0:23:41 So it is about putting out great technology,
0:23:44 but also being very consistent in execution
0:23:46 and offering a long-term roadmap.
0:23:51 The progress you’ve made on that less than 1% market share
0:23:54 pre-Zen is unbelievable.
0:24:00 I remember those less than 1% days, not fondly, either.
0:24:01 Just so you know.
0:24:02 It’s a tough market, though.
0:24:03 It’s a tough market, as we know.
0:24:05 But we must earn it every day.
0:24:07 So I’m very cognizant of that.
0:24:09 Well, that’s what keeps you ahead of the game
0:24:10 and progressing forward.
0:24:13 So many years ago, before your arrival,
0:24:16 you were not a fabulous company, but posts
0:24:20 a spin-out of what’s now global foundries.
0:24:23 You are dependent on the ecosystem,
0:24:25 the manufacturing ecosystem.
0:24:29 Can you talk a little bit about the challenge of not only
0:24:33 integrating tightly with your customers and hyperscalars,
0:24:35 but also the need to integrate tightly
0:24:37 with the fab players as well?
0:24:39 Yeah, absolutely.
0:24:42 So it was the right answer at the time for AMD.
0:24:44 It was before my time.
0:24:48 But to separate the manufacturing operations
0:24:50 from the design operation, we just
0:24:53 didn’t have the volume, the capex, the business model
0:24:55 to make that work.
0:24:57 Now, what it is today is we get to focus
0:24:59 on what we’re good at, which is design.
0:25:01 And that is what we are focused on.
0:25:04 However, we do have to be very tightly partnered
0:25:06 with our manufacturing partners.
0:25:09 TSMC is our main manufacturing partner
0:25:11 for advanced node technologies.
0:25:14 We’re plotting out far beyond the next few years.
0:25:17 We’re really looking into the five-plus-year time
0:25:18 frame of what we need to do.
0:25:20 And it is something that you learn.
0:25:22 You learn how to partner well.
0:25:27 And you learn how to really get advice on these other areas,
0:25:28 like where’s technology going?
0:25:30 And how do we optimize our designs?
0:25:32 But yes, I think that’s part of the ecosystem now.
0:25:35 And it’s even more complicated because it’s not just about silicon.
0:25:37 It’s about packaging and really how
0:25:42 do we put these chips together for very complex, multi-node,
0:25:44 multi-chip type things.
0:25:47 And recently, I mean, you talk about the integration
0:25:52 and how it’s not just about chips anymore.
0:25:54 But M&A has been a really important part
0:25:57 of your strategic agenda in many ways.
0:26:00 And you’ve done some incredible acquisitions
0:26:02 at incredible times.
0:26:06 ZT systems, maybe just talk a little bit about how important
0:26:09 M&A has been for you, and then illuminate a little bit
0:26:13 how you see the role ZT systems will play in the evolution
0:26:15 of solving customers’ problems.
0:26:16 Yeah, absolutely.
0:26:20 We’ve used M&A to really round out our portfolio.
0:26:23 So if you look at over the last five or six years,
0:26:26 we’ve probably acquired about six companies or so, some small,
0:26:27 some larger.
0:26:31 Xilinx was the largest semiconductor acquisition.
0:26:33 I think it’s still the largest semiconductor acquisition.
0:26:37 And that was bringing in the FPGA and adaptive computing
0:26:41 portfolio into AMD, which really brought in our portfolio.
0:26:43 We announced the acquisition of ZT systems.
0:26:45 And we’re talking a little bit about AI
0:26:47 and how fast AI is moving.
0:26:49 What we’ve seen certainly going forward
0:26:51 is it’s not just about the silicon.
0:26:53 The silicon is important, and we’re
0:26:56 pushing every ounce of getting more computing technology
0:26:58 on the silicon in the package.
0:26:59 The software is incredibly important,
0:27:03 so being able to get just enough AI software people
0:27:06 so that we can help our customers and partners utilize
0:27:07 our technology.
0:27:10 But we’re also finding that the integration of hardware,
0:27:13 software, and then really systems is critical.
0:27:18 Because now you’re building these very large clusters
0:27:21 of high performance computing, CPUs and GPUs,
0:27:23 and everything from how do you connect them
0:27:26 from a networking standpoint, a thermal standpoint,
0:27:29 just a reliability standpoint is so important
0:27:31 to actually make it productive.
0:27:32 That’s what ZT systems is.
0:27:35 So it’s a third leg of our stool, if you think hardware,
0:27:36 software, now solutions.
0:27:37 So yeah, I’m very excited about it.
0:27:40 It’s really an expansion of the problem
0:27:44 that we’ve been solving around how do we enable our customers
0:27:45 with the best high performance compute,
0:27:48 and that now extends into the system.
0:27:50 As a student of what’s going on in the industry,
0:27:53 and you guys in particular,
0:27:58 whether it’s organic development or M&A or partnerships,
0:28:00 each step you make always seems to be skating
0:28:02 to where the puck is going,
0:28:04 as opposed to necessarily where it is.
0:28:09 So a lot of the audience is an early stage startup land.
0:28:13 Can you talk a little bit about how you see the role
0:28:16 of startups in semiconductor broadly,
0:28:20 but more AI specifically is the CEO of a large company,
0:28:23 how you see the role of startups in the industry?
0:28:25 – There’s so many good ideas out there,
0:28:29 and the beauty of a startup is you can get a good idea,
0:28:32 and you get some backing from great venture capitalists
0:28:37 like yourself, and you can really innovate and experiment
0:28:40 and learn on those ideas so fast,
0:28:42 and that’s really, really valuable.
0:28:44 I’m really enjoying the work that we’re doing with startups.
0:28:47 We’ve decided to become much more active
0:28:49 in how we’re working on this.
0:28:51 One is we wanna help many of these companies.
0:28:54 So by the way, if anybody needs GPUs,
0:28:55 we’d love to work with you.
0:28:56 – Did everybody catch that?
0:28:58 Did anybody need GPUs?
0:29:00 – Small advertisement.
0:29:03 – Yeah, yeah, I got it, it’s okay.
0:29:04 – But I think the role of startups,
0:29:07 especially right now, has never been stronger.
0:29:10 Cutting edge innovation, experimentation, really,
0:29:12 what I’ve seen, and maybe you’ve seen it as well, Bob,
0:29:14 is I think even large enterprises
0:29:17 who typically used to be,
0:29:18 let’s call it much more conservative,
0:29:21 and working with startups are also becoming much more open
0:29:25 because again, this is back to the disruption I talked about.
0:29:27 Nobody wants to be behind in AI,
0:29:30 and so they want and need people with good ideas
0:29:33 to help them implement in this complex world,
0:29:35 and if it’s a startup, that’s great.
0:29:38 And we’ve learned a ton from startups, actually,
0:29:41 and the rate and pace and speed
0:29:44 at which people are moving is fantastic.
0:29:48 – I mean, in some ways, given the evolution of the ecosystem,
0:29:51 the barriers to enter semi over time
0:29:53 have been relatively large
0:29:58 because you have to find who’s gonna make my product for me,
0:30:01 and the capital you raise,
0:30:04 if it has to go to build your own server farm
0:30:06 or your own fab,
0:30:10 the lack of innovation takes place in the startup ecosystem,
0:30:12 but with the hyperscores and the role they play
0:30:15 to make getting started much simpler
0:30:19 with the world-class foundry capabilities that exist,
0:30:23 and we love interacting with you and being a part of that.
0:30:24 I can’t thank you enough for doing this.
0:30:27 It’s been such a treat to chat with you,
0:30:31 and congratulations on what you guys are doing at AMD.
0:30:33 I admire your leadership
0:30:34 and the role you’ve played in the industry.
0:30:36 Thanks so much for spending time with us.
0:30:37 – Thank you so much, Bob.
0:30:39 It’s a real pleasure and really appreciate
0:30:40 all the collaboration.
0:30:41 – Cheers.
0:30:44 (upbeat music)
0:30:46 (upbeat music)
0:30:49 (upbeat music)
0:30:53 (upbeat music)
0:30:56 (upbeat music)
0:30:59 (gentle music)
0:00:05 in the wave of technologies to increase the demand
0:00:08 of high-performance chips, but it’s not the only.
0:00:10 The last decade was filled with the proliferation
0:00:14 of cloud, gaming, social networks, IoT devices, and more,
0:00:18 which led to some pretty incredible growth stories.
0:00:19 Many are familiar with NVIDIAs,
0:00:23 as it skyrocketed to the top market cap stock in the world
0:00:26 earlier this year, and as of this recording,
0:00:29 sitting at over $3 trillion.
0:00:31 But let’s not forget other semiconductor companies,
0:00:34 like TSMC, ASML, or Broadcom,
0:00:36 each roughly growing by an order of magnitude
0:00:38 over the last decade.
0:00:42 But today, you’ll get to hear about a 50X growth story
0:00:45 over the same time period from Lisa Su,
0:00:47 the incredible woman who’s been at the helm of AMD
0:00:49 throughout that run.
0:00:52 For now, I’ll pass it over to A16Z growth general partner,
0:00:55 Sarah Wang, to properly introduce this episode.
0:00:59 – As a reminder, the content here
0:01:00 is for informational purposes only.
0:01:03 Should not be taken as legal, business, tax,
0:01:05 or investment advice, or be used to evaluate
0:01:07 any investment or security, and is not directed
0:01:11 at any investors or potential investors in any A16Z fund.
0:01:13 Please note that A16Z and its affiliates
0:01:15 may also maintain investments in the companies
0:01:17 discussed in this podcast.
0:01:19 For more details, including a link to our investments,
0:01:22 please see a16z.com/disclosures.
0:01:25 (upbeat music)
0:01:28 (upbeat music)
0:01:31 – Hey guys, I’m Sarah Wang,
0:01:33 general partner on the A16Z growth team.
0:01:36 Welcome back to our AI revolution series,
0:01:37 where we talk to industry leaders
0:01:40 about how they’re harnessing the power of generative AI
0:01:43 and steering their companies through the next platform shift.
0:01:45 We have a very special guest this episode,
0:01:48 Lisa Su, chair and CEO of AMD.
0:01:51 Lisa is one of the most impressive CEOs in history.
0:01:53 Since she took over the helm of AMD,
0:01:56 the value of the company has grown over 50 times
0:01:59 to a market cap of 250 billion
0:02:01 as of the recording of this podcast.
0:02:03 Even more importantly, Lisa and AMD
0:02:06 are democratizing the benefits of gen AI.
0:02:09 Thanks not only to AMD’s top notch chip design,
0:02:11 but also through an open ecosystem
0:02:13 that allows developers to build AI tools
0:02:16 across a huge range of use cases.
0:02:19 It’s no surprise that Lisa is at the helm of this effort.
0:02:21 She’s been an innovator in high performance compute
0:02:23 across her entire career.
0:02:26 Starting from Texas Instruments to IBM and FreeScale
0:02:28 before joining AMD.
0:02:30 This pod is an especially exciting one
0:02:32 because Lisa is joined by none other
0:02:35 than A16Z operating partner, Bob Swan,
0:02:37 who was most recently the CEO of Intel,
0:02:40 one of AMD’s fiercest rivals.
0:02:43 It’s a huge treat to sit in on Bob and Lisa talking shop
0:02:46 and reminiscing about their time together in the industry.
0:02:47 In this wide ranging conversation,
0:02:50 Lisa and Bob cover the state of the art in AI compute,
0:02:52 the chip supply chain,
0:02:54 the role of ecosystem partnerships,
0:02:57 and where Lisa thinks AI is evolving from here.
0:02:59 Let’s get started.
0:03:01 (upbeat music)
0:03:08 – It’s great to have you here
0:03:11 and thanks again for spending some time with us.
0:03:12 – Thank you so much, Bob.
0:03:13 It’s great to be here with you.
0:03:14 – Cheers.
0:03:15 Let’s jump in on the Inquisition.
0:03:16 – All right.
0:03:20 – So 12 years at AMD,
0:03:23 10 as the CEO,
0:03:26 tell us a little bit about your career journey
0:03:28 and how you got to AMD, if you will.
0:03:29 – I grew up as an engineer.
0:03:31 Engineer at heart went to a school
0:03:33 and semiconductor devices
0:03:38 and really did the majority of my early career at IBM
0:03:41 doing R&D around devices.
0:03:44 And then as you think about sort of fun things to do
0:03:45 in the world,
0:03:47 I was always fascinated with the idea
0:03:50 that the work that you do in chips
0:03:52 is such that you can influence
0:03:54 really the way so many things.
0:03:56 Like technology is so important.
0:03:58 And so I just loved being at the forefront
0:04:01 of high performance computing and computing all these years.
0:04:04 And that brought me to a free skills semiconductor
0:04:06 for five years where I was CTO for a while
0:04:08 and then to AMD 12 years ago.
0:04:12 Like I used to say when I would tell people what do you do?
0:04:14 Well, I build semiconductor chips
0:04:15 and people are like, well, what’s that?
0:04:17 Like, why should I care about that?
0:04:19 Is that important?
0:04:23 And now everybody knows what semiconductors are
0:04:24 and why they’re so important
0:04:26 and why they power everything in our lives.
0:04:30 And so that’s what’s fun about the industry that we’re in
0:04:32 and the fact that you’re able to do things
0:04:33 that actually matter in the world.
0:04:34 – Yeah.
0:04:37 If you think about when you first started
0:04:41 the role of compute in the grand scheme of things
0:04:46 relative to today was relatively narrow.
0:04:47 – That’s exactly right, Bob.
0:04:51 I think when you think about even the idea of PCs
0:04:53 and people using personal computers
0:04:54 and everyone needing a computer
0:04:57 and then everyone needing sort of a smartphone
0:05:01 and then everyone needing big cloud data centers
0:05:03 and now everyone needing AI.
0:05:05 I do think it has been an evolution
0:05:09 of how semiconductors and sort of the power of chips
0:05:12 have really infiltrated every aspect
0:05:14 of the business world, our personal lives
0:05:16 and for the good, right?
0:05:18 – We’re all much better because we have all this technology.
0:05:21 – And it’s almost like what would we possibly do
0:05:24 if we didn’t have all this technology behind our desk,
0:05:29 in our hands, in our car, wherever compute is happening
0:05:30 everywhere through all sorts of devices.
0:05:34 And then along comes this thing called AI
0:05:37 and it’s like compute, it’s everywhere.
0:05:41 Can you just talk about how you see AI,
0:05:43 it’s relative importance.
0:05:46 And then over the longer-term horizon,
0:05:47 where is this gonna take us
0:05:49 and where are you and AMD gonna take us?
0:05:53 – Yeah, I think as you think about all of the various
0:05:56 large technology discontinuities
0:05:58 that we’ve seen over the last 30 years,
0:06:00 they’ve all been super important.
0:06:02 They start small and they really influence
0:06:04 every way we experience technology.
0:06:06 I think AI is probably the most important one.
0:06:11 I’d like to say over the last 30 years, 40 years, 50 years
0:06:14 because it’s something more than just technology.
0:06:17 I mean, it really is, you know, AI becomes the ability
0:06:21 for us all to become smarter, more productive,
0:06:24 really utilize the incredible data that’s out there
0:06:25 to help us move forward.
0:06:29 So I really see AI, we’re just at the very beginning
0:06:32 of the AI arc and it’s an opportunity for us
0:06:35 to take technology to yet a different level.
0:06:39 And for us at AMD, my belief is AI is gonna be everywhere
0:06:40 and every product that we build,
0:06:43 but importantly, it’s at the foundation
0:06:46 of what enables all of these great applications.
0:06:49 And so, yes, we’re all building AI compute these days.
0:06:51 We’re trying to build it as fast as possible
0:06:54 so that we can have all of those smart developers
0:06:57 really take advantage of the technology.
0:06:58 – And as you said, it’s fascinating
0:07:02 because in some ways, AI has been around
0:07:05 for such a long period of time.
0:07:08 And while new technologies and innovations
0:07:10 have a tendency to start slow,
0:07:14 this one has moved pretty fast.
0:07:16 – You’re right, it was always around
0:07:18 and it was always something that we thought
0:07:20 had a lot of potential.
0:07:23 But frankly, AI before generative AI
0:07:25 was somewhat hard to use.
0:07:28 And so it took experts to really unlock the technology.
0:07:32 I think the chat GPT moment as we all remember it
0:07:35 was the moment that AI became easy.
0:07:40 We could all talk to our computers and ask it questions.
0:07:41 And yes, it is nowhere near as perfect.
0:07:42 I mean, we have so much work to do.
0:07:44 We’re still very early.
0:07:48 But the fact that we can make technology now so accessible,
0:07:53 I think is what makes this generative AI arc so interesting.
0:07:55 And it’s what’s accelerated the adoption.
0:07:58 – But for AMD, you’ve always been
0:08:02 a high performance compute company.
0:08:06 As you think about the intersection of the company
0:08:08 and what it’s meant for the industry
0:08:11 and then the overlay of AI,
0:08:16 do you see any commonalities with the internet mobility,
0:08:20 just the commonalities that positions you so well
0:08:22 to capture this opportunity?
0:08:25 – Yeah, well, if I just go back a little bit,
0:08:28 when I first took over as CEO of AMD,
0:08:30 it was like 10 years ago now,
0:08:32 it was really a moment where we were like,
0:08:34 what should we be when we grow up?
0:08:36 And if you remember back in those days,
0:08:40 this like 2014, the whole craze was around mobile phones
0:08:43 or tablets, like everybody was into that kind of thing.
0:08:45 And my board even asked me, well,
0:08:48 at least AMD can’t not be in tablets, right?
0:08:52 And I said, well, I’m actually not sure that’s our specialty.
0:08:55 Our specialty is around high performance computing.
0:08:58 We build big things, as one says,
0:09:00 everyone has to know what they’re best at.
0:09:01 And that’s what we’re best at.
0:09:04 We’re best at building large complex microprocessors
0:09:08 or GPUs or with our acquisition of Xilinx,
0:09:10 adaptive and embedded computing.
0:09:13 And when you look going forward,
0:09:16 you see that high performance computing
0:09:19 is really important in the industry in so many places.
0:09:22 And it is at the heart of what makes AI possible.
0:09:24 ‘Cause if you think about what makes AI possible,
0:09:27 it’s the ability to train these models
0:09:29 with hundreds of billions of parameters,
0:09:32 trillion parameters, so that they become ultra smart.
0:09:34 And then we can ask it all these questions
0:09:36 and it gets most of them right.
0:09:38 You need high performance computing at the heart of that.
0:09:41 So it is a great feeling to be in…
0:09:42 – Good place to be.
0:09:43 – Yeah, well, it’s a great feeling to be in a place
0:09:46 where you know that the technology that you’re building
0:09:49 can really push the envelope
0:09:51 on what can be done in the industry.
0:09:53 – And do you see over time,
0:09:58 will it always be the highest performance chip
0:10:00 that is gonna be the differentiator,
0:10:04 or is it gonna evolve for the different workloads
0:10:08 and the different multimodalities in the AI world?
0:10:10 – Yeah, it’s a great point and a great question.
0:10:13 I am actually a believer in you need the right compute
0:10:15 for each form factor, for each application.
0:10:18 So right now, there’s a lot of energy
0:10:22 around building the largest language models
0:10:25 and really this large GPUs
0:10:27 that are being used for training and inference.
0:10:29 But I do see if you look whether you’re at the edge
0:10:32 with embedded applications, industrial applications,
0:10:35 automotive applications, medical applications,
0:10:38 or you’re even at the client level in your PC or your phone,
0:10:40 you’re gonna need different types of AI.
0:10:42 And so you’ll have different engines for that.
0:10:43 And that’s good.
0:10:45 I mean, that’s what spurs all of the innovation
0:10:47 that’s happening around the industry.
0:10:50 – The other thing that I found maybe most fascinating
0:10:54 about the semiconductor industry is the ecosystem
0:10:56 and the importance of the different players.
0:11:00 As you think about development of product,
0:11:04 how do you deal with both what you need to do,
0:11:07 but also the interdependencies with the ecosystem
0:11:10 on delivering something requires a bunch
0:11:11 of different players?
0:11:13 How do you think about that in the context
0:11:18 of product development, but to get products to market?
0:11:20 – Well, I don’t think there’s any one company
0:11:21 that can do it all.
0:11:24 I mean, at the end of the day, we all have specialties
0:11:25 and they’re things that we’re good at,
0:11:28 but the opportunity to closely collaborate
0:11:30 and partner is so important.
0:11:33 We’re a big believer in open ecosystems
0:11:34 and industry standards.
0:11:38 And the idea that, hey, I’m building these great processors,
0:11:41 they should connect to other people’s networking
0:11:43 and we should be able to interoperate together.
0:11:46 The software ecosystem is super important too.
0:11:48 Developers shouldn’t have to develop
0:11:49 for one company’s hardware.
0:11:52 Developers should be able to develop what they need
0:11:54 and be able to use what’s the best hardware underneath.
0:11:57 So I think that’s part of the evolution
0:12:00 of an open ecosystem so that we can get
0:12:02 the best innovation out there.
0:12:04 – So in this constant evolution,
0:12:07 and we’ve seen this debate over time,
0:12:11 closed garden, interoperable,
0:12:14 what is gonna be the predominant winner
0:12:17 if there is such a thing in an AI world
0:12:21 on open interoperable technologies and interfaces?
0:12:23 – I’m a big believer in open ecosystems.
0:12:26 Interoperable is really important.
0:12:29 Closed walls usually end up being a problem.
0:12:32 If you look at sort of the technology arcs of time
0:12:35 and in this world where technology is moving so fast
0:12:37 and whether it’s a new model
0:12:40 or a new hardware technology or new capability,
0:12:43 you wanna make sure that it’s interoperable.
0:12:45 – Along the same lines, you and other players
0:12:47 in the industry recently announced
0:12:51 the new ultra accelerator link
0:12:53 and the ethernet link standards.
0:12:56 Is that an example of how you think about opening,
0:12:58 how you engage with the ecosystem?
0:13:00 – Yeah, I think it’s a great example.
0:13:03 We’re all looking at when you think about these large AI
0:13:06 clusters that you need in the future,
0:13:08 networking is such an important piece of it,
0:13:11 but you do want this choice as to
0:13:12 what hardware are you connecting,
0:13:15 what’s the processor, what’s the networking fabric,
0:13:16 what’s the overall system architecture.
0:13:18 So the ultra accelerator link
0:13:21 and ultra ethernet consortium are great examples
0:13:24 where competitors and peers can come together
0:13:27 and say, you know what, we’re gonna adopt open standards
0:13:30 and we’re each gonna innovate on top of that.
0:13:32 And those are two great examples
0:13:36 and it includes many companies that do compete,
0:13:37 but also can cooperate.
0:13:38 And if you think about that,
0:13:41 that’s exactly what an open ecosystem is supposed to be.
0:13:44 – And you talked about competitors,
0:13:46 industry players coming together.
0:13:50 The demand for compute over the last several years
0:13:52 has been maybe unprecedented,
0:13:56 just in terms of its pace and its distribution
0:14:01 and the ability to ramp up supply
0:14:04 to meet these incredible demands
0:14:07 when you throw in cycle times to put more capacity
0:14:12 in COVID supply chain disruptions.
0:14:16 Have these disruptions or challenges on the supply side?
0:14:20 Have they impeded your ability to move faster in some areas?
0:14:22 And what have you learned from this
0:14:25 that will make the next supply constraint
0:14:27 be a bit smoother for the industry?
0:14:30 – Yeah, when you look over the last four or five years,
0:14:32 probably the largest disruption
0:14:36 to the semiconductor supply chain was really around COVID.
0:14:40 It was the moment where everyone needed
0:14:42 more semiconductors at the same time,
0:14:43 which kind of wasn’t expected.
0:14:46 Usually what happens in the semiconductor market
0:14:49 is you’ll have one market up and one market down.
0:14:51 Mobile may be really hot,
0:14:54 but infrastructure will be down or vice versa.
0:14:56 What we saw in COVID was basically every market
0:14:59 at the same time had this concentrated effect.
0:15:01 And the semiconductor supply chain
0:15:04 is actually really good at meeting demand.
0:15:06 Actually, we usually overshoot.
0:15:08 As you know, Bob, that happens from time to time,
0:15:09 but it takes time, right?
0:15:13 It takes 18 months, 24 months to really put that all on board.
0:15:15 And so I think the industry as a whole
0:15:19 has done a good job at bringing more supply on board.
0:15:21 The more recent thing as it relates to AI
0:15:24 where it’s super hard to get GPUs,
0:15:26 that truly is, again,
0:15:29 nobody forecasted what generative AI would need.
0:15:31 And so it has taken some time
0:15:35 to really build all of this advanced packaging capacity
0:15:37 and high bandwidth memory capacity.
0:15:40 But again, the semiconductor supply chain is good at that.
0:15:42 And we just have to get a little bit better
0:15:45 at forecasting what long-term demands are.
0:15:50 – I do remember exiting 2019, entering 2020,
0:15:54 where that normal cyclical nature of the industry,
0:15:59 I think we’re all looking at expecting 2020 to come down.
0:16:03 And then COVID hitting for a short period of time,
0:16:05 it looked like things were going further south.
0:16:08 And all of a sudden, to your point,
0:16:09 everybody needed supply.
0:16:13 And in many ways, while there was issues materializing,
0:16:15 the way the ecosystem comes together
0:16:17 in the semiconductor industry
0:16:19 and the sophistication of the supply chain,
0:16:22 despite the challenges, is pretty impressive.
0:16:22 – That’s exactly right.
0:16:25 And I think we’ve all gotten smarter and better as a result.
0:16:30 I think this idea of, hey, let’s try to just eke out
0:16:31 every last penny in the supply chain
0:16:33 has gone a little bit of ways
0:16:37 to let’s build resilience into the supply chain.
0:16:38 When governments are asking,
0:16:40 do you have enough semiconductors?
0:16:42 I think that gives us permission
0:16:45 to really think more broadly about resiliency.
0:16:48 – And we’ll talk about resiliency and government asking.
0:16:53 I think it was roughly two years ago to the day
0:16:55 there was this thing called the CHIPS Act.
0:16:59 And for our audience, the government signs into law
0:17:04 a bill authorizing $280 billion to help in the design
0:17:11 and the manufacturing in the US of semiconductors.
0:17:16 And then $53 billion of that was authorized to be spent.
0:17:18 I know it’s still a work in process,
0:17:20 but as you think about the CHIPS Act
0:17:23 and some of the challenges from the last couple of years,
0:17:26 how do you see that helping resiliency
0:17:28 of supply chain going forward?
0:17:31 – I have to say I’m a big supporter of the CHIPS Act.
0:17:34 I never would have thought that five years ago,
0:17:37 semiconductors would be high enough priority
0:17:41 in the US government’s view of what needed
0:17:43 clear industrial policy.
0:17:45 Some people say, hey, it’s not enough
0:17:46 or does it make a difference?
0:17:48 I think it’s made a huge difference.
0:17:51 It’s made a huge difference because what it’s really done
0:17:56 is it’s put at the top of the priority list, resiliency
0:17:58 and semiconductor, both manufacturing
0:18:01 as well as research and development in the United States.
0:18:04 And of course, there’s much, much more work to do.
0:18:06 As you said, it’s a work in progress, but it’s a good thing.
0:18:07 It’s a good thing for the industry
0:18:09 that there’s a focus here.
0:18:12 I’m actually particularly excited about some of the work
0:18:14 that’s being done on the R&D side
0:18:17 because I think there’s a whole opportunity
0:18:20 to really train the next generation of leaders
0:18:24 who will lead the semiconductor research and development
0:18:26 as well as future capabilities.
0:18:27 So yeah, I think it’s a great thing.
0:18:29 And yes, it’s still early days.
0:18:32 And we need to make sure that every dollar is spent,
0:18:34 is spent for good reasons, and that we
0:18:37 get the return on investment on the other side.
0:18:40 But it’s a clear indication of how important semiconductors
0:18:44 are to the US and really to the global economy.
0:18:45 I couldn’t agree more.
0:18:50 The dynamics of– we’ve talked about the ecosystem working
0:18:52 together and the importance of the ecosystem.
0:18:54 To throw the government in there,
0:18:56 obviously it creates challenges.
0:18:59 But industry and government working together
0:19:01 to solve really big problems, I think,
0:19:04 is a real necessity in some areas.
0:19:07 And this is the one where I’m thrilled about the CHIPS Act
0:19:10 itself, but the deployment and the proof points,
0:19:11 I think, are still in front of us.
0:19:13 So it’ll be an exciting time.
0:19:17 And I hope that the challenges around resiliency will be–
0:19:19 Remembered?
0:19:21 Remembered or rearview mirror.
0:19:22 Yes.
0:19:23 Yeah, exactly.
0:19:25 Remembered is the best way to frame it.
0:19:29 At a time when innovation is happening all the time,
0:19:33 you still are a relatively long cycle development time frame.
0:19:35 Yeah, very long cycle, really.
0:19:39 How do you guys deal with long cycle development
0:19:43 with short cycle innovation and what inherent challenges
0:19:46 or opportunities that creates for you in the industry?
0:19:49 Yeah, I think the most important thing in our world,
0:19:52 especially in hardware, is one needs
0:19:54 to try to have a crystal ball.
0:19:57 You’re never going to predict the future entirely,
0:20:00 but you do need to be able to say, hey, these
0:20:02 are the disruptions that are coming up.
0:20:05 These are the things that we need to pay attention to.
0:20:08 Probably the best example that I can think of–
0:20:11 and this is one where we had a lot of debate internally–
0:20:14 is, what’s the future of Moore’s law?
0:20:15 That’s been debated just a little bit.
0:20:16 I remember those debates.
0:20:19 And by the way, I’m a believer in Moore’s law
0:20:22 has been extended so many times because people are super smart
0:20:25 and able to come up with different ways of extending
0:20:28 the same principle of more transistors, more capability
0:20:30 every couple of years.
0:20:32 But for example, that’s like advanced packaging.
0:20:35 And when do you go to 2 and 1/2D and 3D packaging?
0:20:39 And for us, we use this technology called chiplets.
0:20:40 We didn’t know.
0:20:43 We didn’t know at the time when we were making that decision,
0:20:45 was it going to be the right bet?
0:20:47 But we knew that we had to make that bet.
0:20:49 And you really don’t figure that out until three to five years
0:20:50 out.
0:20:52 So your question about how do you know?
0:20:54 You don’t know, but you try to make sure
0:20:55 that you’re betting in the right direction.
0:20:59 And then you have to be agile enough to adjust accordingly.
0:21:02 And that’s what this whole world of high performance
0:21:04 computing is about.
0:21:06 You talked about the right bets.
0:21:09 And you guys have had incredible success
0:21:12 on making the right bets.
0:21:14 What is the balance between how you
0:21:18 learn from your customers about the right bets to make,
0:21:21 but also how you lead your customers
0:21:22 given the development cycles?
0:21:25 How do you strike that balance at AMD?
0:21:29 Yeah, our top two priorities that I tell the company
0:21:30 all the time.
0:21:32 The first one is your tech company.
0:21:35 Our job is every day to wake up and build great products.
0:21:39 But we do that through having very deep customer relationships.
0:21:41 Because I really do believe that they go hand in hand.
0:21:44 Our customers are some of the largest,
0:21:47 whether it’s cloud manufacturers or OEMs or enterprises
0:21:49 in the world, that they see the problems
0:21:51 that they’re trying to solve.
0:21:54 And that’s where it’s most beneficial is talking
0:21:57 to our customers about, hey, what problems are you having?
0:22:00 What are you trying to solve two, three, four years out?
0:22:02 And then our technologists can really
0:22:04 come up with ideas for how to solve those problems.
0:22:08 So it’s not like it’s one for one where we listen
0:22:09 to everything people say.
0:22:11 But we do listen a lot.
0:22:14 Because that tells us that we’re working on the right things.
0:22:16 Because whatever we do, you want to ensure
0:22:18 that the technology you’re building
0:22:21 is something that will solve somebody’s problem.
0:22:23 Hyperscalar market.
0:22:25 Tremendous progress over the last several years.
0:22:26 Congrats.
0:22:30 Correlation, learnings from winning in hyperscalar
0:22:32 with this rapid growth from AI.
0:22:34 Is there learnings that you’ve been
0:22:37 able to extract from what it takes to win in one?
0:22:40 And then how do you translate to win it in AI?
0:22:44 So when we started in the hyperscalar market
0:22:46 with our first generation products,
0:22:49 our Zen product portfolio, I think
0:22:53 we were about maybe 1% share of the server market.
0:22:57 And actually, the whole idea of having deep partnerships
0:23:00 with customers is really, we needed
0:23:03 to be able to say that, hey, it’s all about roadmap.
0:23:06 Yes, it’s great, the product you have today.
0:23:08 But it’s all about, can you keep a sustained
0:23:12 level of constant innovation many generations out?
0:23:15 And I think we have made a lot of progress
0:23:16 in the hyperscalars.
0:23:19 I love the relationships that we have across the top brands,
0:23:23 whether it’s the Microsoft or Amazon, or Google, or Oracle,
0:23:24 or Meta.
0:23:26 It’s always about, how do we innovate together?
0:23:30 I think the AI arc is very similar in the sense
0:23:33 that these are big bets that the hyperscalars are
0:23:36 making on who their technology partners are going to be.
0:23:38 And we want to help them accomplish that.
0:23:41 So it is about putting out great technology,
0:23:44 but also being very consistent in execution
0:23:46 and offering a long-term roadmap.
0:23:51 The progress you’ve made on that less than 1% market share
0:23:54 pre-Zen is unbelievable.
0:24:00 I remember those less than 1% days, not fondly, either.
0:24:01 Just so you know.
0:24:02 It’s a tough market, though.
0:24:03 It’s a tough market, as we know.
0:24:05 But we must earn it every day.
0:24:07 So I’m very cognizant of that.
0:24:09 Well, that’s what keeps you ahead of the game
0:24:10 and progressing forward.
0:24:13 So many years ago, before your arrival,
0:24:16 you were not a fabulous company, but posts
0:24:20 a spin-out of what’s now global foundries.
0:24:23 You are dependent on the ecosystem,
0:24:25 the manufacturing ecosystem.
0:24:29 Can you talk a little bit about the challenge of not only
0:24:33 integrating tightly with your customers and hyperscalars,
0:24:35 but also the need to integrate tightly
0:24:37 with the fab players as well?
0:24:39 Yeah, absolutely.
0:24:42 So it was the right answer at the time for AMD.
0:24:44 It was before my time.
0:24:48 But to separate the manufacturing operations
0:24:50 from the design operation, we just
0:24:53 didn’t have the volume, the capex, the business model
0:24:55 to make that work.
0:24:57 Now, what it is today is we get to focus
0:24:59 on what we’re good at, which is design.
0:25:01 And that is what we are focused on.
0:25:04 However, we do have to be very tightly partnered
0:25:06 with our manufacturing partners.
0:25:09 TSMC is our main manufacturing partner
0:25:11 for advanced node technologies.
0:25:14 We’re plotting out far beyond the next few years.
0:25:17 We’re really looking into the five-plus-year time
0:25:18 frame of what we need to do.
0:25:20 And it is something that you learn.
0:25:22 You learn how to partner well.
0:25:27 And you learn how to really get advice on these other areas,
0:25:28 like where’s technology going?
0:25:30 And how do we optimize our designs?
0:25:32 But yes, I think that’s part of the ecosystem now.
0:25:35 And it’s even more complicated because it’s not just about silicon.
0:25:37 It’s about packaging and really how
0:25:42 do we put these chips together for very complex, multi-node,
0:25:44 multi-chip type things.
0:25:47 And recently, I mean, you talk about the integration
0:25:52 and how it’s not just about chips anymore.
0:25:54 But M&A has been a really important part
0:25:57 of your strategic agenda in many ways.
0:26:00 And you’ve done some incredible acquisitions
0:26:02 at incredible times.
0:26:06 ZT systems, maybe just talk a little bit about how important
0:26:09 M&A has been for you, and then illuminate a little bit
0:26:13 how you see the role ZT systems will play in the evolution
0:26:15 of solving customers’ problems.
0:26:16 Yeah, absolutely.
0:26:20 We’ve used M&A to really round out our portfolio.
0:26:23 So if you look at over the last five or six years,
0:26:26 we’ve probably acquired about six companies or so, some small,
0:26:27 some larger.
0:26:31 Xilinx was the largest semiconductor acquisition.
0:26:33 I think it’s still the largest semiconductor acquisition.
0:26:37 And that was bringing in the FPGA and adaptive computing
0:26:41 portfolio into AMD, which really brought in our portfolio.
0:26:43 We announced the acquisition of ZT systems.
0:26:45 And we’re talking a little bit about AI
0:26:47 and how fast AI is moving.
0:26:49 What we’ve seen certainly going forward
0:26:51 is it’s not just about the silicon.
0:26:53 The silicon is important, and we’re
0:26:56 pushing every ounce of getting more computing technology
0:26:58 on the silicon in the package.
0:26:59 The software is incredibly important,
0:27:03 so being able to get just enough AI software people
0:27:06 so that we can help our customers and partners utilize
0:27:07 our technology.
0:27:10 But we’re also finding that the integration of hardware,
0:27:13 software, and then really systems is critical.
0:27:18 Because now you’re building these very large clusters
0:27:21 of high performance computing, CPUs and GPUs,
0:27:23 and everything from how do you connect them
0:27:26 from a networking standpoint, a thermal standpoint,
0:27:29 just a reliability standpoint is so important
0:27:31 to actually make it productive.
0:27:32 That’s what ZT systems is.
0:27:35 So it’s a third leg of our stool, if you think hardware,
0:27:36 software, now solutions.
0:27:37 So yeah, I’m very excited about it.
0:27:40 It’s really an expansion of the problem
0:27:44 that we’ve been solving around how do we enable our customers
0:27:45 with the best high performance compute,
0:27:48 and that now extends into the system.
0:27:50 As a student of what’s going on in the industry,
0:27:53 and you guys in particular,
0:27:58 whether it’s organic development or M&A or partnerships,
0:28:00 each step you make always seems to be skating
0:28:02 to where the puck is going,
0:28:04 as opposed to necessarily where it is.
0:28:09 So a lot of the audience is an early stage startup land.
0:28:13 Can you talk a little bit about how you see the role
0:28:16 of startups in semiconductor broadly,
0:28:20 but more AI specifically is the CEO of a large company,
0:28:23 how you see the role of startups in the industry?
0:28:25 – There’s so many good ideas out there,
0:28:29 and the beauty of a startup is you can get a good idea,
0:28:32 and you get some backing from great venture capitalists
0:28:37 like yourself, and you can really innovate and experiment
0:28:40 and learn on those ideas so fast,
0:28:42 and that’s really, really valuable.
0:28:44 I’m really enjoying the work that we’re doing with startups.
0:28:47 We’ve decided to become much more active
0:28:49 in how we’re working on this.
0:28:51 One is we wanna help many of these companies.
0:28:54 So by the way, if anybody needs GPUs,
0:28:55 we’d love to work with you.
0:28:56 – Did everybody catch that?
0:28:58 Did anybody need GPUs?
0:29:00 – Small advertisement.
0:29:03 – Yeah, yeah, I got it, it’s okay.
0:29:04 – But I think the role of startups,
0:29:07 especially right now, has never been stronger.
0:29:10 Cutting edge innovation, experimentation, really,
0:29:12 what I’ve seen, and maybe you’ve seen it as well, Bob,
0:29:14 is I think even large enterprises
0:29:17 who typically used to be,
0:29:18 let’s call it much more conservative,
0:29:21 and working with startups are also becoming much more open
0:29:25 because again, this is back to the disruption I talked about.
0:29:27 Nobody wants to be behind in AI,
0:29:30 and so they want and need people with good ideas
0:29:33 to help them implement in this complex world,
0:29:35 and if it’s a startup, that’s great.
0:29:38 And we’ve learned a ton from startups, actually,
0:29:41 and the rate and pace and speed
0:29:44 at which people are moving is fantastic.
0:29:48 – I mean, in some ways, given the evolution of the ecosystem,
0:29:51 the barriers to enter semi over time
0:29:53 have been relatively large
0:29:58 because you have to find who’s gonna make my product for me,
0:30:01 and the capital you raise,
0:30:04 if it has to go to build your own server farm
0:30:06 or your own fab,
0:30:10 the lack of innovation takes place in the startup ecosystem,
0:30:12 but with the hyperscores and the role they play
0:30:15 to make getting started much simpler
0:30:19 with the world-class foundry capabilities that exist,
0:30:23 and we love interacting with you and being a part of that.
0:30:24 I can’t thank you enough for doing this.
0:30:27 It’s been such a treat to chat with you,
0:30:31 and congratulations on what you guys are doing at AMD.
0:30:33 I admire your leadership
0:30:34 and the role you’ve played in the industry.
0:30:36 Thanks so much for spending time with us.
0:30:37 – Thank you so much, Bob.
0:30:39 It’s a real pleasure and really appreciate
0:30:40 all the collaboration.
0:30:41 – Cheers.
0:30:44 (upbeat music)
0:30:46 (upbeat music)
0:30:49 (upbeat music)
0:30:53 (upbeat music)
0:30:56 (upbeat music)
0:30:59 (gentle music)
Lisa Su has transformed AMD into a global leader in AI and high-performance computing.
In this episode of the AI Revolution (AIR) series , Bob Swan, a16z Operating Partner and former CEO of Intel, sits down with Lisa Su, CEO of AMD, to discuss how her leadership has propelled AMD’s growth and positioned the company at the forefront of AI innovation.
They explore AMD’s pivotal role in democratizing the benefits of gen AI, the evolution of AI computing, and the importance of open ecosystems and partnerships in driving technological breakthroughs.
Resources:
Find Lisa on X: https://x.com/lisasu
Find Bob on X: https://x.com/bobswan
Stay Updated:
Let us know what you think: https://ratethispodcast.com/a16z
Find a16z on Twitter: https://twitter.com/a16z
Find a16z on LinkedIn: https://www.linkedin.com/company/a16z
Subscribe on your favorite podcast app: https://a16z.simplecast.com/
Follow our host: https://twitter.com/stephsmithio
Please note that the content here is for informational purposes only; should NOT be taken as legal, business, tax, or investment advice or be used to evaluate any investment or security; and is not directed at any investors or potential investors in any a16z fund. a16z and its affiliates may maintain investments in the companies discussed. For more details please see a16z.com/disclosures.