Advanced Micro Devices, Inc. (AMD) Earnings Call Transcript & Summary

That's really helpful detail. And I think people do underestimate the magic of the Infinity Fabric and everything else that, that's unlocked. So if we just get into the straight node, Moore's Law node war discussion, how are you thinking about the advantage -- well, if we back up a second, talk about the ramp you've had at 7-nanometer to date. And does anything change given your primary competitors' latest hiccup on their manufacturing side? Does that change anything strategically for AMD?

Yes. Look, I think it's -- we're very focused. We're very focused on executing our strategy and running our play. And we always assume that our competitor is going to do extremely well or our competitors are going to do extremely well. So I never want to assume that they're going to run into a manufacturing, market issue or any other issue. And so we cut our road maps together with the best technology that we can possibly get access to and the best architecture that our engineers can dream up. When -- now because we did that, we could drive to 7-nanometer in a very fast way. And again, accessing it to begin with, with relatively small chips just using many output. And that's driven our migration to 7-nanometer that -- at a very fast clip. I think it's the fastest ramp we've had on a new node. And again, accesses that advanced technology in a prudent way. You should expect that we're going to continue to take that sort of approach going forward, ensuring that we've got the architecture and the topology to get us access to the best possible node at any given point in time. And so I think you'll see us be very aggressive.

So last question on the manufacturing side. Given you guys raised guidance on your last call, which seems like a long time ago already back in July, that this year is turning out to be much better than feared. How is the supply situation? Do you have sufficient supply to do everything you want to do on the 7-nanometer side?

Yes. Look, I think supply is -- and I think we've commented, supply is sufficient. But it's tight. It places a premium on planning for ourselves as well as planning for our customers. But we think we've got the supply chain lined up from wafers to substrates and everything in between to hit our guidance.

So let's pivot over to the server CPU market specifically and some of the market share side of things. I think AMD deserves a huge congratulations of hitting up to the 10% or double-digit market share in the June quarter that you guys had promised. I know you're not going to give whatever the new market share target is. I think the last event that I actually physically attended within the industry was your analyst meeting back in early March, and no such number was given at that time. But I did want to just get your thoughts on the market share side of things. It took you about 6 quarters or 1.5 years to go from the 5% to the 10% market share. How do you think the slope of that share gain curve looks going forward? Would you be able to double the share again in the same amount of time? Is it more a much fairer way to look at it that adding the incremental 5 points of share rather than thinking of it as a doubling would happen over that time? Would it happen faster or slower? And why?

Yes. So I think -- I don't want to get into the mode of sending new benchmarks along the way with fine granularity. I think all we've said about share looking forward is we certainly aspire over time to exceed our historical high-water mark, which was 26%, 27% unit share. And certainly, over time, I think we've got the road map to do so. But we're not going to put a slope on that line or further intermediate milestones. What we're focused on, though, is taking the momentum that we've built with the ecosystem and our partners and customers with the first generation of EPYC Naples and now the second-generation and continuing to build on that. So Naples for us was getting -- quite frankly, it was getting that first 5% was getting to having people used to AMD back into -- in the market was having customers believe that we could deliver a high-performance component that we were doing what we said. And to that end, Rome was hugely important, the second-generation EPYC processor that we introduced last August. It's hugely important, not just because it's a great product. And really with 64 cores offers literally double the performance of our principal competitors' offerings at that point in time, still just about the same. Not only offering literally double the performance, the other critical part of Rome is it's exactly what we said we were going to do. And that predictability of execution gives customers confidence that they can invest in putting AMD into their infrastructure. They can invest in optimizing for AMD microarchitecture and that we're going to be around. And so Milan, along the same lines, Milan is critically important because it's a further demonstration that the road map I stood on stage and in front of everybody and their brother's uncle in 2017 that we're delivering -- 3 years later, we're delivering like a metronome. And that confidence that we can be trusted, that we will do what we say we're going to do, and you can count on us, coupled with the high-performance that we're offering in Rome, and we're going to build on in Milan is, I think, driving rapid adoption for us across cloud and enterprise. And we think we've got a clear path to continue growing this business aggressively.

So yes, I agree that trust is something that's key in delivering the road map with the products as promised on time is something that all these customers need to see to be able to adopt the product. And thankfully, that's exactly what you're doing. You also mentioned about -- I'll put it -- I'll describe it a little differently than you, but an expansion of the type of customers that you're addressing. The initial ramp to 5%, I think, was largely cloud-driven, less of the enterprise side of things. Talk about the broadening of the market that you are addressing, what your customer profile looked like in the first 5% share, what it looks like at 10% share, and how you think it's going to look as you go beyond that level?

Yes. No, look, you're absolutely right. When we went -- when we put the road map together, you can't eat an elephant in 1 bite. You can't reenter a market like this with -- all at once. And so what we tried to do was we took the characteristics of the Zen core, the first Zen core that we had, which was a good core, and it was a very efficient sort of at lower power levels. So it couldn't necessarily reach the same frequency as our competitors' part in a power and constrained fashion, but it was more efficient than our competitors' product in a power constrained environment, a high core count environment. That drove our strategy for that initial product to design something that is great at throughput. It was great at -- let's -- in scenarios where you've got 32 cores operating within a power constraint, 200-, 225-watt envelope. And it was very efficient and very competitive there. What workloads correspond to that? Cloud. Cloud workloads, at scale workloads, certain HPC workloads. And so that's -- and certain virtualization, probably cloud workloads, that's where we saw traction in the first generation. There are also customers that, quite frankly, are more very technically savvy and more tolerant, perhaps with a little risk. And so that's where we focused Naples, and we said this in the introduction, and that's where we had the most success. With Rome, we opened up -- Rome is a much higher performance part, both in terms of it doubles the number of cores. It uses a 7-nanometer process of cores, and it has the Zen 2 engine in it. That opened up the aperture of the workloads we could compete in from, say, maybe 50% to 60%. And with Naples, those cloud workloads is at scale workloads. With Rome, we designed it to have coverage of, say, about 80% of the workloads, including many in the enterprise side. Now as it turns out, that design point was against where we thought our competitor was going to be. And they weren't quite where we assumed they would be. And so I think the coverage on our own is even better than we thought. And so many more enterprise workloads are a great fit on Rome than we originally planned for. And so the expansion on Rome has been from going beyond cloud and certain memory-focused HPC to really every HPC workload and almost every enterprise workload in addition to the cloud. So seeing a lot of traction there, and we've also put a lot more time into the ISVs that are relevant to the enterprise side. The strategy with Milan was to get the Zen 3 core and Milan out there to really have 100% coverage, no excuses, no compromise really to have leadership performance in virtually every workload from enterprise to cloud, from new software-defined telco and storage, down to IoT gateways. And we think we're on track to achieve that. And we think that's really going to drive the business going forward. So I think when you see Milan come out, we think we've got complete coverage in the market.

So I put that together, it seems like you've built the trust up, doing what you said on time, you're expanding the customer base, like you said. Now Milan has no excuses, talks to -- can address all the markets, all the workloads, et cetera. So I'll give one more attempt to maybe put some words in your mouth, but would you be disappointed if the slope of the share gains was slower going forward? It seems like you have 2 big positives, big tailwinds that you didn't have when you're going from 0% to 5% or from 5% to 10%. So with those tailwinds there, maybe you -- just smile if you agree.

Well, look, I want to drive this business as hard as we possibly can. We've got a great, great product road map to do it.

Perfect. So let's talk about the competition side of things. How does pricing enter this equation? Unfortunately, I'm old enough to remember the last time when AMD did hit that kind of upper 20s market share. And the response at that point from your primary competitor was for a period of time to get very aggressive on the pricing side. And that kind of damaged the market temporarily for both, but really hurt AMD somewhat structurally. There's 1,000 things that are different this time. But how do you think pricing enters this equation? And are you seeing any change in the behavior on that front?

Look, I think we've always anticipated. There's a lot of scar tissue here at AMD of competing with Intel for many years, and they're a fantastic company, great competitor. So we've always expected the environment to be extremely competitive. And I would say that it's about what we expect. But the thing to keep in mind about the data center market is, and I've said this a number of times, it's all about performance. The -- one way to think about it is the CPU is about, say, 25% of the cost of the system, 20%, 25% of the cost of the system. But unlike its power, you've got perhaps sheet metal, power supplies, motherboards, network adapters, memory, storage, et cetera, around it. And so if you think about it from that point of view, a simple way to think about it is, if your performance given that the cost of the component is 25% of the cost in the system, if your performance is, say, 75% of the competitors', you could give your part away and it doesn't matter. You still can't win the total cost of ownership at the system level. So for us, the principal focus is all about performance. How do we deliver compelling performance that translates into system-level TCO for the end customer because that's what the informed customer is going to do -- is going to make their buying decision on the basis of that, that system-level TCO, which is only very weakly correlated to the price of the CPU component. So that's our focus. And I think we've architected -- now I will say one other thing that we've, again, architected the solution with the chiplet approach to be far more cost-effective than perhaps following a monolithic road map with very large chips to reach the upper ends of the performance spectrum. And so I think we're in -- we're seeing about what we expected. Our focus is on the system-level TCO, and I think we've got the performance to win those fights. And we've, I think, very thoughtfully architected our solution to be cost effective.

So the other competitive dynamic I wanted to hit on was more ARM-based CPUs in the server market. It's long been rumored as a viable threat against x86 in aggregate. You guys, back in time, even dabbled a little bit of some of the heterogeneity that ARM might be able to bring to it. Recently, some of the hyperscale cloud guys are doing some internal efforts on that front. And then obviously, as of yesterday or the day before, you also have the turbocharging of the server CPU on the ARM side as part of NVIDIA's logic in attempting to buy ARM itself. So how do you view the competitive threat either internally or from merchant silicon providers from ARM-based CPUs in the server market?

Well, look, I think, first off, I think we've always said that our focus is on high-performance CPUs and GPUs and offering a compelling, competing -- compelling competitive road map. And again, referenced my comment of a moment ago, I think that the performance is the critical thing. If you've got a high-performance solution, you're going to win the TCO battle at the system level. So maintaining that high-performance road map is critically important. Secondarily, there is the issue, of course, of migrating workloads from one instruction set architecture to another. I would maintain that generally, that hasn't happened. If you take a look at the history of the computing industry, it's very difficult to displace a computing architecture unless you have a new workload coming in. Because the thing that is, theoretically, the most fungible in the data center, the software is actually the thing that's most adamant. It's hardest to change, it's slowest to change. And so for existing workloads, it's a huge barrier to somebody migrating over to a new instruction set architecture. That said, look, I think there is -- there has been for a decade interest in ARM in the data center. I think that we see a lot of tire kicking. We've seen it for a decade. And different companies make their own independent decisions. Our focus is make sure that we are always providing extremely high performance, the most competitive performance CPU cores and increasing the GPU and the combination together. And I think that so long as we're doing that, we've got a great competitive environment ahead of us, and we expect to see x86 is still the dominant instruction set architecture on the core.

So we really have a couple of minutes left. So I want to hit on a topic you just mentioned about the partnership between the CPU and the GPU. Obviously, GPUs as accelerators have been a huge market. Your biggest competitor on that side has grown that market from a couple of hundred million dollars to multiple billions of dollars per year in business. Talk about AMD's aspirations in that market? And when do you think that can be a needle mover within your company as a whole?

Yes. No, I think we definitely see GPUs as a critical driver of growth in the data center for us going forward. And we've thought that way for a while. We just tried to be very thoughtful about what we're -- any company has to make resource decisions, and I'd much rather to do a great job at a few things than a crappy job at one thing too many. But that said, I think we've been very thoughtful in preparing for GPUs being a bigger part of our data center portfolio for some time. When we think about GPUs in the data center, we really think about it in 3 different primary areas. One is essentially VDI or remote rendering, rendering in the data center, be it for cloud gaming or be it for a desktop or workstation replacement. We think that's an important growth area for us. We're pleased and proud to be part of several major cloud gaming initiatives. We're pleased and proud to be part of some great cloud VDI offerings as well on the productivity side. And given COVID, we expect to see quite a bit more of that -- well, not just COVID. COVID is in after effect of what I've heard people refer to as the next normal of increased work from home, no matter what. We expect to see that VDI interest in market continue to grow. The second piece is really traditional HPC, traditional high-performance computing for scientific research, medical research for product design. And then lastly, of course, and certainly far from last in terms of importance, machine intelligence, artificial intelligence, machine learning. We see all 3 of those as being critical growth drivers for us going forward. With all of them, the software is the key aspect. And so we focused on a couple of things. First off, there's no point in trying to duplicate what my competitor has done, trying to run and catch up, running their exact same play is a fool's errand. And so what we've done is we've really worked to partner with others on development of an open-source software framework for -- or software frameworks for both machine intelligence and HPC. And so our rack and stack, for example, Radeon Open Compute is a critical part that allows our customers and us to innovate together and to drive some pretty compelling solutions. The second piece of this is accelerators for HPC or machine intelligence, which is still hard to use. And so part of it is continuing to work on frameworks, continuing to make it easier for developers to embrace this technology. But the other key part that we've talked about that's really led to our exascale wins was adding coherency. So the CPU and the GPUs instead of being apart connected by a thin I/O link are connected via high-performance coherent interconnect, where, very importantly, for the software developer, they no longer have to explicitly manage the pool of memory down on the accelerator, be it a GPU or anything else. We think that's a huge disruption and it unlocks a lot of additional accelerator capabilities. It makes it much easier to program. And I think that was key to our exascale wins. So putting those things together, we're very confident in growing that market quite a bit.

Great. Well, Forrest, time went really fast on us. Thank you very much for taking the time. It's great chatting with you on this. Congratulations about the great growth in your business, and we look forward to monitoring it going forward, whatever the slope may be. So everybody, that will end this presentation. Thanks again, Forrest.

Thanks a lot. Appreciate it.