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

Good morning, everyone. Thank you, and welcome to this day 2 of our BofA Global Technology Conference. I'm Vivek Arya. I cover semiconductor/semi-cap equipment. I'm absolutely delighted to have Dan McNamara, the Senior Vice President and General Motors -- General Manager of AMD's server business unit. Really delighted to have Dan. And since -- I know Dan has been with AMD for about 1.5 years or so. And since he's kind of -- it's our first visit with him in this format, we would love, Dan, if you could just give us a quick summary of your background, your career experiences and just what made you join AMD.

Well, sure, Vivek. Thank you for having me. This is really exciting to be talking to you all today, and I appreciate everyone joining on the webcast. So as you've mentioned, I joined AMD about 18 months ago. So for the previous 15 years or so leading up to my joining AMD, it was really centered on Altera and then Intel. So for about -- from 2004 to 2015, I had senior leadership roles across application engineering and then the business units at Altera. And then when the acquisition of Altera happened with Intel, I joined Intel and led the actual group, Altera, called Programmable Solutions Group at Intel. And I spent almost 4 years there. Took on other leadership responsibilities at Intel as I was there. So -- but ended up there for about 4 years. When I think about joining AMD, and as you might guess, you form your own thesis on some of your career moves. And when I think back to 2 years ago or so talking to AMD, I was really struck by 3 different things. Really, the first one was their vision of driving and changing the compute landscape within the data center. And it was centered around high-performance computing and delivering the best TCO, with also a further vision around how does heterogeneous computing really evolve. And then I looked at the technology, and I saw scalable chiplet architectures. I saw advanced process technology with TSMC, and I saw really advanced packaging road map, which was very exciting. But most importantly, it was really about the team. The team -- the leadership team was very, very focused on not only execution, but listening to our customers and delivering to what we said and constantly raising the bar. So I was really compelled to join AMD and talking with Lisa and Mark Papermaster and a number of the leadership team. If I fast-forward back to today now, we believe we are -- we have clear leadership with products across the data center from server CPUs. And we really are doing this at what I would call the most opportune time because we are in the midst of this mega cycle of compute from the edge to the cloud. So general-purpose computing is really exploding across every single hop in that line, and we really are excited about having that product leadership and driving and engaging with our customers today.

Got it. Excellent. Thank you, Dan, for that introduction. So it's interesting that your kind of entry into AMD has sort of coincided with AMD's very remarkable market share gain and traction in the market, right, on the server side. Give us a flavor of the discussions, if you will, with your -- the cloud customers, the enterprise customers, the OEMs. How have you observed these discussions essentially change over time as AMD has launched -- has been on this kind of consistent road map, right? Give us a flavor of what these discussions have been.

Yes, Vivek. A lot has changed probably over the last 2 years in terms of the discussions with customers. When I talk to customers today, we have moved clearly from an alternative in the data center to a clear strategic and trusted partner to most of the hyperscalers and the enterprise, and really have become more of a clear part of their road map. So we're engaged in growing our footprint, but also growing and talking about very future out in terms of 3 to 5 years, what are we building? So it's really changed quite a bit. And if you think about it, we've really hit an inflection point here in Q1. When you look at the results from Q1, we doubled our revenue year-on-year. We're driving, we believe, over 400 public instances in the cloud by the end of the year. We're driving with Milan over 100 new platforms. We've doubled the number of solutions and appliances. So we really believe that we've hit this inflection point that is really going to continue to accelerate our growth. And we've also -- when you look at some of the supercomputing wins, we've had -- we've announced in the past some of the large supercomputing wins like Frontier and El Capitan. But just recently, we announced the Pawsey supercomputer in Australia and the EuroHPC LUMI in Finland, which are great new wins for AMD. So we continue to really drive that traction, but we're also evolving our strategy. And since I've been here, I've been very focused on -- while we bring a tremendous amount of value, performance and TCO value, it's about accelerating that time to value for our customers. So we've really been focused on building out that ecosystem and not only just in platforms but across the ISV and OSV space and really accelerating that time for customers to actually put AMD in their fleet. So we're a trusted solution today, but how do we accelerate that? So we're very focused on that. We're growing our head count in terms of customer enablement and engineering. And it is really, really going well. And as I said, the customers now see us as a 3- to 5-year partner and are really talking about a number of different things, including how does the data center evolve over the next 5 years?

All right. Absolutely. And Dan, I'm glad you brought up, right, that road map because I think that's been one of the more attractive aspects of the business. And talk to us about the journey from the second gen to the third gen. So from Rome to Milan, what kind of additional workloads, right, what new applications you have been adding to the mix? And then importantly, what investors, right, kind of care about, how does Milan stack up, right, apples-to-apples, no pun intended, against your competitor's Ice Lake server?

Yes, Vivek, this is really the key to our success. So our road map journey is quite interesting and exciting. If you think about it, you go back to June of 2017, our reentry into the data center market with Naples. And with Naples, it was a 14-nanometer introduction and we really introduced this chiplet approach. And then you go to August of 2019 when we introduced Rome. And Rome brought just a pretty giant leap in terms of performance, right? From top of stack of 64 cores, hardware-based security, PCI Express Gen 4, which was the only solution out there for quite a while and then 8 channels of 3,200 megahertz of DDR4. So we really brought that leadership to market with Rome. And Rome was very successful in the cloud, across high-performance computing and even in the enterprise. So we saw a really nice leap there. And then with Milan that we launched in March of this year, we had 3 key objectives. And the first one was maintain that perf and TCO, right? So continue to raise the bar in performance and TCO value for the high-end infrastructure as a service market, for high-end -- high-performance computing and supercomputing, even some of the high-end enterprise. But the second one, which was really critical for us, was bring more value to the enterprise. And the enterprise is a really good story for us today with Milan and Rome, to a certain degree, but mostly with Milan because we see the enterprise evolving. And what I mean by that, with digital modernization in the enterprise, we're -- it's moving to a hyperconverged infrastructure, which means the deployments on-prem are really starting to mimic the cloud, so these hybrid multi-cloud environments. So what we saw was, "Okay, our traditional value prop that was strong for the cloud, which is that high performance, that density, that TCO, is really playing itself well into the enterprise now with hyperconverged infrastructure." So we saw that there was a big opportunity there. And we also wanted to bring single-thread performance leadership. So the second one was really about engaging that enterprise and really expanding the footprint there. And the third piece of our objective with Milan was really to expand that solution set and that time to value. And as I mentioned, we had almost 2.5x the number of solutions and appliances with the Milan launch than we did with Rome. So those are the 3 objectives. But what's really exciting about Milan is what we brought to market. So we delivered 25% better performance per watt and a 19% instruction per clock increase against Rome. So gen to gen, really important. We currently hold -- and it grows every day, but we hold over 220 world records across energy processing, floating point processing, Java applications, database, pretty much the wide gamut of applications that you are tested and you can actually hold a record for it. And as I mentioned, we doubled the solution set. So we really feel good about where we are with Milan. It's poised and delivering across all 3 of the major segments we're focusing on, which is the cloud, the enterprise and then high-performance computing. So the journey is really exciting. And then let's fast forward a little bit or jump to Genoa. Genoa, we're on track with 5-nanometer for 2022. And we believe between Rome as a value play going forward and then Milan with our hard performance and then bringing on Genoa in 2022, we feel like our lineup stacks up extremely well against all of our competition.

Got it. No, I'm glad you brought that up, Dan. And I know it's being reflected in some of the market share data that we see when we look at AMD's market share gain in Q1, right, of 400-plus basis points. It's been kind of an acceleration versus the annual 200 to 300 basis point expansion that we have seen. So we are starting to see that being reflected in the reported market share. On the Genoa question, I know there have been some media reports about when it's supposed to come out, right? Could you give us maybe a little bit of a teaser, a preview of what we should expect in '22? And I ask that because there have also been other reports about Intel's -- what they claim as their kind of 7-nanometer, right, or in their version, 10-nanometer Sapphire Rapids product, which will ramp in 2022. So Genoa on 5 is coming out next year. That says conceptually, you will be 1 node ahead of Intel in 2022, which is why I think the Genoa launch could have some added importance.

Yes. So we're on track for 2022. We're not giving explicit months or quarters at this point.

Right.

But you're right. Genoa will be a 5-nanometer product. We will still have Milan that arguably will compete with Sapphire also. So we're going to have a nice one-two punch in the 2022 time frame with a leading 7-nanometer and a leading 5-nanometer product. So we feel very good about the stack-up here coming in 2022 versus the competition.

Got it. Next thing, Dan, I wanted to explore was just how the mix of customers has evolved, cloud versus enterprise versus high-performance computing. And kind of one level more deeper into the cloud side, how much of the usage is for public cloud versus internal applications? Because to me, internal applications is kind of the real test, right? It's a hyperscaler saying that, look, I have enough confidence in the product to use it for my internal applications. This is not something that I'm just putting on the shelf and just enabling an instance, right, for somebody else to take advantage of that when I use it internally that -- that to me demonstrates kind of real confidence. So talk to us about how the mix of that customer has evolved for you.

The cloud is a very exciting opportunity for AMD, and we're doing quite well. And I think you -- we hit it really on the head in terms of what's happening in the cloud. So public infrastructure as a service, there's public records out there of counting -- you can count up the number of instances. And I think we're growing pretty rapidly in public instances. And that's an important business for us. It really is. I think the big point that you just made is really what we call the internal properties. So those -- whether it's software as a service or platform as a service, entities that the top cloud providers bring to market are really critical. And when you think about our value proposition in the cloud, it's density, it's 64-core top of stack, it's the performance, and then it's the TCO. Our cloud partners want to drive as much density as many VMs in a rack as possible to drive their own TCO model, and that's why we win in the cloud. But when you shift to internal properties, it's a really direct model for the GMs of those businesses. So when you meet with a GM of an internal property, they're measured just like everybody on a P&L. So when they see the Milan product or the Rome product and they see the TCO that they can get on their solution set, it really is compelling. So we see a big growth within the internal properties as well as the public infrastructure as a service. But the internal properties, I believe, are a true measure of the value. And the reason being is, is we have no control over how instances are sold by the cloud providers. But what we do have control on is the value we bring and the perf per dollar, if you will, that we bring to the customer. And we do see these internal properties really jumping on the value prop of AMD going forward. So it's pretty exciting model within cloud -- across all of the hyperscalers in terms of when we discuss both the public cloud and the internal properties.

Got it. Do you have a rough mix, Dan? We would love to know if you -- I know that when I look at AMD's overall market share, right, as we get from Mercury or other third-party data, it's about 10%, 11%, but that obviously includes parts of the market that AMD doesn't play in, such as on the comp side. So it's about 10%, 11-ish percent. I assume on the data center side, it's -- and the cloud side, it's obviously much higher than that. But within that, could you give us some flavor for internal versus public cloud?

Vivek, it's hard to pinpoint because when a hyperscaler buys product, they don't tell you. But what I can tell you is when you look at the -- we are much more represented within internal properties than we are from a public cloud instance standpoint. I would say pretty dramatically higher percentage than the -- what shows up on a public instance. But it's very hard to give you a pinpoint number because we don't -- when the orders come through the ODMs and things like that, you really just don't see the breakdown.

Got it. I understand. The next thing, Dan, from just a competitive aspect is that Intel has said many times that, look, one of our advantages is the ability to customize product that I remember this isn't a new phenomenon. In fact, when we have listened to presentations from the data center group for the last, I think, almost 7 to 10 years, that half of their SKUs have been kind of customized. What does customizing mean, right? And is AMD able to customize? So give us some sense for, is customization an important part of dealing with cloud customers? And if yes, what kind of customization aspects does AMD bring to the table?

So customization for us is a big part of the cloud business. And I'll go back to what I talked about at the start of this, the chiplet architecture and the scalability of that architecture feeds really well into customization. So we can customize across total power, number of cores, base and boost frequency and a number of other areas. So we do -- when we look at the cloud business, we have a large percentage of custom SKUs. And again, I think our architecture and ability to customize these is really why the cloud hyperscalers are jumping on this. And it's -- you would think that everyone wants to jump on the top of the stack, but they don't, right? There's all sorts of different architectural areas where they look at where they would maybe tweak a TDP power envelope and reduce core count to really fit into their envelope. And I mentioned earlier, it's all about driving the absolute most density in a rack. So we are very, very flexible there. Our architecture lends itself to be very flexible there, and we continue to do that. And I would say that from Rome to Milan, we're actually doing more custom SKUs in the cloud than we -- and we expect that to continue.

Got it. Next thing, Dan, I wanted to explore was your presence in the enterprise. Because when I think back to the last time AMD had very high market share in servers, right, back in 2006, right, at that time, AMD's market share, I believe, was 25%, 26%. There was no cloud back then. So AMD had all this share with just a very, very fragmented enterprise base. This go-around, I think the transformation has kind of first started on the cloud side. But talk to us about this enterprise market, right, which is another 30% or 40% of the market where you are perhaps less represented today? What are you doing to expand your position in the enterprise market? When should we start to hear the success that you have had on the cloud side being replicated on the enterprise side?

Yes. The enterprise is extremely important for us. And as I mentioned, that was one of the key tenets of the Milan launch. Here is why I think we will continue to accelerate server and it will -- I'm sorry, enterprise, and it will continue to -- or start growing at the same clip as our overall server business. So first and foremost, I talked about the road map and the execution cycle of the road map. And what I think about with our Fortune 1000 enterprise CIOs, it's all about trust. So not only are we bringing performance and TCO benefit on a gen-to-gen basis, but we're building their trust. And we've been very, very focused out talking to Fortune 500, Fortune 1000 CIOs to talk about what is it that they are looking for from us. And again, that is growing, that trust is there. They are now talking to us about really what are their goals over the next 3 to 5 years. So that's really positive. So that's job 1 is, I think, building that trust. Secondly, as I mentioned, we really are in this modernization phase with the enterprise, whether it's VDI clients or just a hybrid multi-cloud environment. We see that our traditional value prop that has won a lot of traction in the cloud and HPC feeds very well into where the market is going. So we have a lot of solution sets around hyperconverged infrastructure, and that's really positive. And then again, the third piece is what I talked about earlier is we're bringing more and more solutions and more and more manpower to the table to enable and fine-tune and try and accelerate the bringing of AMD into their fleets on prem. So we really see that this is starting to move. Today, we're a bit underrepresented, but we believe that over the rest of this year and into 2022, that is changing. And we do believe that enterprise will grow at the regular clip that we're seeing with the overall server business. So it's quite exciting to do that because when you win in the enterprise, they're staying power because you win their trust. It's not just a transaction, it's trust. And so we're excited about those 3 things, and those 3 key value props are driving our growth in the enterprise space.

Got it. And then the final leg of the stool is kind of on the supercomputing side, and I feel that today's supercomputer is kind of tomorrow's cloud server. So if you can start the journey with kind of the most demanding, right, performance-intensive customers, that actually helps to kind of lay the groundwork, right, for the other parts, right, close to cloud and then to enterprise and so forth. So talk to us about what has enabled AMD to be successful with supercomputing customers? And I have seen a lot of announcements. When will they translate into revenue that you can talk about in terms of actual dollars and market share?

Yes. So high-performance computing is supercomputing. They are a very good bellwether in terms of how your products stack up. We've always felt like if you win there, those architectural decisions are really critical. And if you win there, we believe that we've got a very, very solid product stack. So we're continuing to win there. What's most exciting about some of the supercomputing wins and high-performance computing today is we're starting to win more with CPU and GPU, right, with our Radeon Instinct. So it's not just a Rome or Milan story anymore, it's Radeon Instinct. And we tie that together with Infinity Fabric, with cache coherency. And we believe that, that is a very compelling performance value proposition. And then what we're doing is we're tying that together with ROCm software stack with compilers, libraries and then going to have sort of this containerized hub of applications. So we believe that we're building this solid CPU plus GPU software stack solution for high-performance computing that will transcend, just like you said, into other segments. So it's an exciting time for us there. We are announcing Frontier, which is the Oak Ridge National Lab's supercomputer. That's coming online later this year, and we're going to announce our next generation of Radeon Instinct. So that's an exciting time for us. And then further, El Capitan in 2023. But the learnings from these big supercomputing wins and the architectural wins we've seen has really transcended and kind of moved into the cloud and even into some of the enterprise. So we see a nice uptick here. We believe we are very, very well positioned there. And again, as we talked about earlier, with Genoa coming online in 2022, the Rome to Milan to Genoa, and then that tight tie with GPUs, we really feel that we're positioned extremely well going forward for the supercomputing space.

Got it. One other aspect, Dan, I would love your insights on is just this entry of ARM-based competitors in the market. Actually, just last week, we had a chance to host Ampere Computing, right, and Renée James, who you might remember from the Intel days, right? We have heard of NVIDIA with their launch of Grace. Amazon, right, with their Graviton is doing work with ARM. How do you see ARM as a threat in the data center both kind of in the near and long term? And how would you contrast kind of the threat from ARM on one side, but the ability to kind of take share against x86 competitor on the other side? Like how would you balance these 2 forces?

Yes. Vivek, this is a really interesting and very germane question to today. So what I would say is the evolution of the data center is happening right now. And we -- what we see is it's moving to more what is the right architecture and solutions for the right workload. It's no longer just a broad, general-purpose compute space where you're just throwing CPU cycles at it. And I think that's why you're seeing ARM come online. ARM has a good custom - typically a good customized approach to certain different workloads. And they've seen some traction. And AMD has built ARM-based products. So we don't see that as much of a threat and we see it -- ARM as a partner, as Lisa has talked about in the past. But what we do see is bringing together all the key components for some of the cloud vendors, which is the best CPUs, the best GPUs, over time, hopefully, the best FPGAs and then ASIC custom blocks. And we're driving this vision of heterogeneous compute where you're really solving point solutions with some of these. So we see -- ARM is going to -- they've had some success, and they probably continue -- will continue to do that. We believe our lineup across all of these different heterogeneous components will really uniquely position us for the future here. And we've been working on this vision for quite a while. And when you tie in our packaging capability, that we really believe that, that combined solution set with an overarching software stack actually will win out and solve a lot of the problems or challenges that our hyperscaler customers are looking at. So not as much of a threat as much as an evolution of the market and the solution sets that the customers are looking for.

Got it. Maybe, Dan, let's bring it kind of specifically to what we are seeing in 2021, so kind of in the near- to intermediate-term time frame. So AMD has established a very strong growth objective for the year, over 50% growth, and your business is supposed to grow much, much faster than that. And meanwhile, there are all these supply constraints. So give us some assurance that, first of all, you have the supply needed to grow your business, right, that AMD has the supply it needs to grow the overall business and any other puts and takes. Like, is there more upside? Like, are you limited by supply? Or are you limited by demand? So give us a flavor for what more do we have to look forward to in this near to intermediate term?

Yes. So the supply conversation is very dynamic. And I think Lisa has said it pretty well over multiple times over the last couple of months. Really, the demand across virtually every segment has exceeded our expectations. And from a server standpoint, though, we feel pretty comfortable. We're poised for growth this year. We believe we've got the supply to do that. We are a focus for the company because of the data center. But when you think about AMD, we have many very high-growing business -- fast-growing businesses, right? Whether it's the client or the semi-custom business or the server business, all of these businesses are growing aggressively. We do believe that we've secured enough supply to support the growth. And we also do believe that our relationships from across the supply ecosystem are very, very strong and growing. So we're doing a lot of work there. Our supply teams have done a tremendous amount of really great work this year, and we'll continue to do that in the outer years. So we feel good about the supply. It's a very dynamic situation, though. So we're keeping very, very close eyes on it. But we do have enough to support the growth for '21, and we do believe that improves throughout the year and into 2022. So we feel pretty good about it. However, it is a very, very tight, watchful situation that we need to keep focused on.

Got it. So part B of that is, can we -- do you see any signs for the business to actually accelerate in the second half? And if that were to happen, would it come more because of units or ASPs or enterprise coming online? Like what are the prospects for your business growing much faster in the second half?

We do believe the business will continue to accelerate through the year. And I think Lisa had guided recently a high teens percentage of revenue on the larger revenue number. And we do believe in the outer years, as we talked about in the Financial Analyst Day, about 30%. So we feel very good about those projections. This year, we will see a couple of things happen. We will see ASPs remain pretty strong. We believe that the value prop and the performance per dollar with Milan is very strong. So we'll see that at the high-performance leadership point. We also see the mix changing. We will see, as we talked about earlier, the enterprise come more online with all of those platforms getting through the system and getting those solutions we believe that enterprise will help accelerate in the second half. So we feel very, very good about the prospects for our growth this year. The other thing I would say is we have very high visibility across both the cloud and hyperscalers, the high-performance computing space and even the enterprise now. And the enterprise is a little bit dicier here because it's not as predictable. But we do have very, very high visibility. We're having conversations well past the next 2 quarters. So we have this really nice visibility. So we feel very, very good about our position. We feel good about the growth. We will see a mix -- the mix shift a bit. We are a little bit underrepresented enterprise, but we see that will change over the course of this year.

Got it. How is the -- this conceptually the pipeline for next year shaping up, right? Because I imagine that all these chip shortages and just how tightly balanced the system is -- or actually not balanced, but how tightly the system is being managed, I assume that at least your cloud customers would want to have some assurance of how they will line up instances, right, how they will line up capacity. So how are your conceptually discussions going on to fill up next year's pipeline?

We are definitely in those conversations with customers in terms of what does -- what could the growth be at any one customer. So we don't have like a top line number that we could talk to, but the reality is each of our customers are looking at how much can we grow together in terms of units. So we're kind of working -- that's a time -- that's a very timely question because it's really a conversation that we're having through the course of June and July to really look at what does this -- what can this be in 2022? But we don't have a whole lot of estimates about it right now for next year.

Understand. One other concern that investors have brought up is that, look, there is a new team at Intel, right? They have new products coming out this year. What role can pricing play in the market share dynamic, right? Do you expect them to suddenly start slashing prices to win back the market share? What role is pricing playing in the discussion in terms of how customers are allocating market share?

So let me talk about the competitive landscape just briefly, and then I'll talk to the pricing question. But first, I would say is it's a very highly competitive segment today within the data center, and we expect that. And we're ready for the challenge in the marketplace, right? We're a fairly competitive group also. But where we focus is what I talked about earlier. If we continue to deliver that performance per dollar increase gen-on-gen, and we continue to deliver to what we say we're going to do, we believe that price is not really a part of the conversations because we are bringing so much value on a gen-to-gen basis. And the customers are seeing a nice jump in performance per their dollar spent. That's our focus. And then for us, from a just competitive -- overall competitive standpoint, we're going to continue to play our own game. We have a clear road map. We listen to our customers. We are very tightly engaged with our customers, and we're going to stay that way. And we're going to continuously evolve our road map. We're going to expand our road map as we go forward. And as I talked about earlier, the evolution of heterogenous computing is going to be front and center for us going forward. So we feel like pricing is always going to be there, but it's not really the focus. If we continue to bring the value and play our game and stick to what we want to do from a strategy standpoint, I think we're going to be okay.

Excellent. And Dan, in the minute or so we have left, you have mentioned the word heterogenous computing a few times. Maybe just describe for us what that means from an AMD perspective? And why is it important, right, to have that kind of vision for the data center?

Yes. It's actually an exciting evolution, if you will. And when you think of AMD, we're really the only one today in volume production with both CPU and GPUs across multiple segments. As we go forward and if -- as we go forward, we're going to add FPGAs to it and custom -- more custom blocks. So if you think about being able to offer all of the functional blocks, along with a software stack to solve virtually any of the problems that our customers are trying to solve, we believe that we're uniquely well positioned for this heterogenous compute landscape, if you will. So we're pretty excited about that. And we're in conversations today with customers about how do you actually partition hardware and software to solve for the right workload. And that's really the key, Vivek. It's really about when you see some of the hyperscalers going out and building custom devices, it's really because they're solving really important problems that they have within their data center. And we believe we have all the assets going forward to solve virtually any problem that a hyperscaler has. So we're pretty excited about that going forward, and that will be a conversation over the next couple of years that we'll be having.

Excellent. Great. With that, we are at the end of our time. Dan, thank you so much. Lovely to see you again, and thank you so much for taking the time and sharing your insights. Really appreciate it. And thanks, everyone, for joining.

Thank you, Vivek.

Okay. All the best.