The Transcripts, Transcripts
Advanced Micro Devices, Inc. (AMD) Presents at 51st Annual J.P. Morgan Global Technology, Media, and Communications Conference (Transcript)
Advanced Micro Devices, Inc. (NASDAQ:AMD) 51st Annual J.P. Morgan Global Technology, Media, and Communications Conference May 23, 2023 3:15 PM ET
Company Participants
Jean Hu – EVP, CFO & Treasurer
Ruth Cotter – SVP, Marketing, HR & IR
Conference Call Participants
Harlan Sur – JPMorgan Chase & Co.
Harlan Sur
Okay. Let’s go ahead and get started. Good afternoon. Welcome to JPMorgan’s 31st Annual Technology Media and Communications Conference. My name is Harlan Sur, semiconductor and semiconductor Capital Equipment analyst for the firm.
I’m very pleased to have Jean Hu, Executive Vice President and Chief Financial Officer. And Ruth Cotter, Senior VP of Marketing, HR, and Investor Relations at Advanced Micro Devices here with us today. Jean and Ruth, thank you very much for joining us this afternoon.
Jean Hu
Thank you, Harlan.
Question-and-Answer Session
Q – Harlan Sur
Let me start it off with you, Jean. You’ve been with the AMD team now for four months. Came to AMD with a great track record of execution. Just wanted to get your thoughts on what you’ve observed so far, the AMD team, the execution, and more importantly, the growth opportunities.
Jean Hu
First of all, thank you, Harlan, for inviting us, and good afternoon, everyone. Thanks for the question. I’ve been really fortunate to work at some of the best semiconductor companies. And I’m super thrilled to be part of the AMD team at this inflection point for high-performance computing. The last four months has been very exciting as part of the team. I continue to be very impressed by the whole team’s strong focus in developing innovative products and also execution. That’s just super impressive.
I would say one of the most exciting aspects during the last four months, actually is to see incredible market opportunities, especially the opportunities related to generative AI, a large language model. Literally, it’s just unfolding in front of us. That’s very exciting. I think AMD has a broad portfolio. The company has been investing in both the CPU, GPU for the last decade. And the last year, with the acquisition of Xilinx and Pensando, the business added not only CPU, GPU but DPU adaptive compute and adaptive SoC.
So, when you look at the portfolio, probably it’s one of the best positions of the company to address the future opportunities ahead of us. Financially, the company has built a strong business model with strong financial flexibility, so we can continue to invest to really aggressively double down on the AI front and address the opportunity ahead of us. So overall, really, really exciting, a lot of work to do.
Harlan Sur
Perfect. Well, I appreciate that, and very insightful. Let me — I’m going to start off with the first few questions, sort of near to midterm, get that out of the way, and then let’s focus — I really do want to focus on kind of the mid-to longer-term strategy and traction of the team. So, server shipment TAM forecasts for this year have continued to come down, right? The latest view is that server shipments probably down about mid-single digits this year.
The AMD team has clearly been under shipping this dynamic in the first half of the year. You’ve talked about weak enterprise, weak China, mixed trends in cloud, but you anticipate a better second half of the year.
In fact, you expect to grow 40%, 50%-plus second half versus first half implied within your full-year guidance. What dynamics are you tracking customer programs, ramp of your next-generation Genoa, Bergamo platforms? All of that, that gives you sort of confidence on a significantly better second half of the year?
Jean Hu
Yes. I think in the first half of the year, we do see a mixed-demand environment. Some of our cloud customers, they are going through inventory correction and optimization. And also, enterprise is seeing weakness just because of macro uncertainties. But if you look at our Q1, because of those factors, it does impact our Q1 revenue. We guided Q2 just up modestly quarter-over-quarter. Overall, the first half our revenue run rate is quite low.
We do think from that much lower run rate, the second half, we’ll see revenue growth and a step-up on the revenue side. There are a couple of drivers, and Ruth, you can add this. First, I think, is from Genoa, right? It was introduced last November. We have been working with not only cloud customers and enterprise customers to qualify a much broad set of workloads.
So, when you think about, Genoa has been ramping, but the second half, we do see broader workload adoption, which will help us. And also, we do add much more value with this generation for the product versus supply generation.
Secondly, Bergamo, you mentioned, we’re going to see the initial ramp of Bergamo because it’s actually tailored for cloud-native applications. We do think there’s a demand for that kind of workload and applications. And then Milan continued to see strong demand actually because we continue to have cost efficiency, total TCO. We do think Milan continue to be really a good product for us. And last, I guess, is MI300. We did say in Q4, we’re going to see the initial revenue ramp from a supercomputer side and some of the initial AI workload, too.
So, when you look at it all together, we think we have a very strong product portfolio in server market. And we have been gaining market this year for the last four years, so we do think it will continue to drive the leadership and continue to gain share.
Harlan Sur
Does the transition — Genoa, Bergamo, they’re pretty big inflection points for the team. During these periods of inflection, do you get better — is that partly what drives the confidence in the second half is because customers have been qualifying it, developing software, they really want to exploit the TCO benefits? They want these servers in their data centers. Is that partly — and so they’re giving you better, more clear visibility in terms of deployments? Is that also kind of what drives the confidence?
Jean Hu
I think in general, when you look at our EPYC server platform, we do work with our customers closely, especially cloud customers. You literally have to go through the qualification process. You will have a lot of discussion on which workload they use and their ramp plan. Of course, the macroeconomic condition always can change. But overall, we do have a lot of visibilities how customers think about the ramping. So, from that angle, the engagement of our team with customers are very deep.
Ruth Cotter
I think in addition, as we look at approaching 30% market share in server market, we have gained critical momentum, to your point, with customers. And because of that, we get deeper and more longevity into their road maps, which allows us to make sure that we approach our data center efforts in a very customer-centric way as we look to sort of solve their problems. And on the CPU side, that strategy has played off very well. And to Jean’s point, we look to replicate that now beginning with MI250, which is on the CPU side and market today, and MI300, which will be launched before the end of the year.
Harlan Sur
Perfect. And then on the client side, similar question that I just asked you but now more focused on the PC business. What metrics are you monitoring, tracking that gives the team confidence that PC client business bottomed in the March quarter and is set for growth through the remainder of this year?
Jean Hu
Yes. PC market, as many of you know, has been going through one of the worst down cycles since probably 1995. AMD, we have been undershipping the consumption for the last three quarters, actually very significantly. So, we do monitor the sell-through data carefully, and we do think there’s an inventory digestion continuously, but the stability of the sell-through definitely is there.
If you look at our own business, we continue to believe, for Q1, it will be the bottom of the business because of the inventory digestion because the second half, you do see some seasonality pickup. Given what I said, I do think right now, our planning scenario is a steady improvement, right, and the second half will get help from seasonality.
Harlan Sur
Lisa talked about seeing some positive signs out of the China. I believe it was enterprise markets, right? It’s been a slow demand recovery, right? China has emerged from the COVID lockdown. Is the team also seeing signs out of China, positive signs out of China in your PC client, and embedded businesses as well? And is any of your second-half growth assumptions dependent upon a meaningful recovery in China?
Jean Hu
We do see some stabilization. I would not say pickup, I would say, stabilization in the PC market and the server market. We probably have more engagement and discussion but not pick up yet. And our assumption for the second half really did not incorporate whether China is going to go through recovery.
Harlan Sur
Perfect. So, let’s talk a little bit more about sort of the medium to longer-term trends. I do want to start off with AI. Obviously, the big focus by the market has been on this AI arms race amongst the cloud and hyperscale titans. And what we do know is that all of them are spending on compute, networking, storage, infrastructure silicon, right? You have your data center and AI tech event, right, on June 13. So, I’m hoping that we’ll get more insights then.
But the team has a strong portfolio, as you mentioned, to take advantage of these opportunities, CPU, GPU, FPGA, ASIC capabilities, right? And as it relates to these new transformer base, large language models, right, that are driving things like OpenAI’s ChatGPT platform, you guys are focused on this opportunity with your instinct-based GPUs, right? Your MI300, very innovative, integrated CPU, GPU platform. What does the engagement and revenue opportunity with your cloud and hyperscalers look like for the MI300 or the instinct-based GPU platforms over the next, call it, 12 to 24 months?
Jean Hu
Yes. Ruth, maybe you want to kick off to talk about the June 13 events and the whole thing. I can definitely add.
Ruth Cotter
Sure. Thanks, Jean. So, to your point, Harlan, on the 13th of June, we’re hosting a data center and AI event. We’ll focus on our CPU product momentum with the next product in the family to come as Bergamo, as Jean talked about, which is sort of cloud-optimized high-performance, low power. So, we’ll give more details on that. And we’ll also frame our AI strategy and vision to make sure that expectations are clear.
And then all of that will be leading up to the MI300 product launch, which will be later in the year. We said that will be in the second half of the year. So, we just want to make sure the strategy and vision is appropriately set. And then we’ll also give some details on software side, software engagement. That’s obviously a topic of interest to complement the hardware work that we have been doing.
And then on the MI300 front, we’re very pleased with the customer engagement and momentum. And we had said that engagements are at 3x today, what they might have been even at the beginning of the year as you think about this AI inflection point in the market. And we’re also happy with how MI250 has been progressing with Microsoft, in particular, as the sort of publicly announced customer for that product. And that’s sort of setting things up very nicely for MI300.
Harlan Sur
Perfect. And then I had asked this question on the conference call because there is another segment of the market going after these AI processors and accelerators. And that’s the cloud and hyperscaler, your customers opting to, at some point in their portfolio, add some custom ASIC capabilities, right? And if you look at what it takes to be successful in the custom ASIC market, right, expertise in designing this huge system-on-a-chip, billions of transistor chips, strong compute, networking, IP, right, and a track record of execution, right?
And the AMD team has that with your semi-custom business. You’ve been a big player in the gaming market with your semi-custom solutions. You have all of the right building blocks. And so, has the team — does the team think about potentially entering the ASIC market and servicing your cloud customers?
Jean Hu
Yes. Harlan, thanks for the question. I think from an AMD perspective, the way we think about is we address high-performance compute across all different — either it’s x86 or ARM or customer silicon or potentially ASIC. I think the key thing for us is to think about what we can provide to customers with the best TCO. And if you look at the approach and the strategy we have been going after is on the x86 side. With the Genoa family, we actually broadened our offering from not just Genoa. We have Genoa-X focused on the technical workload.
We have Genoa focused on the low-cost telco market, and we have a Bergamo, we talked about which literally is tailored for native cloud applications. That actually offers better efficiency and performance than some of the ARM merchant product supposed to come to the market. But overall, we are very open. We have a deep engagement with the customers. If customers want to do customer silicon or wants to do ASIC, we actually have all the capability and IP. As you said, the team also has a strong track record to execute on customer programs. So, we are quite open for all the solutions and definitely just want to provide the customer best TCO.
Ruth Cotter
And the track record that we’ve built through our semi-custom business with game consoles to date and the complexity and precious IP that’s been involved in that has stood well to sort of continuing to engage with customers. But whether they decide to go with a custom offering or not, the building blocks of IP that we have amassed in the company and how we’ve put those to work is very attractive to our customer base, and it allows them a lot of flexibility. And as the company has scaled more recently over the last several years, it’s also proving to be a win in terms of our strategy as we look to continue to gain share pervasively.
Harlan Sur
Yes. Now you can add adaptive compute to the ASIC IP portfolio as well, right, add some programmable capabilities to any custom solutions that you do. On the data center and server markets, you guys continue to gain server CPU share from your competitor. I think by our estimates, you gained about 10 percentage points of share in 2022 to about 22%, 23%. You drove 25%, 27% share in Q1. And we think you’re on track to capture 5 percentage point to 7 percentage points of share, somewhere around 28% to 30% share for this year.
Obviously, technology performance lead over your competitor with your EPYC family has driven this strong share momentum. Clearly going to sustain that momentum this year. You’ve got 5th Gen EPYC CPUs slated for next year, which would ramp around the same time as your competitor’s next-gen CPUs, and they hope to reach competitive parity with the AMD team by 2025. So first, are you still on track with your Turin in architecture in 2024? And then how do you see the competitive dynamics potentially changing if your competitor is successful at executing to their 2024, 2025 road maps?
Jean Hu
Yes. First, Turin is absolutely on track, and we are actually very pleased with the initial results. I think maybe what we need to do is take a step back to look at AMD’s journey over the last six years to really build the leadership position in server market. If you look at the last six years, the company has consistently executed from Gen 1 to Gen 4. Each generation, literally, we delivered, as we said, with the regular cadence, right? And each generation, we increased the differentiation, increased the performance.
If you look at Milan, which we introduced more than two years ago, that actually helped us to gain tremendous market this year. And even today, Milan continue to perform extremely well against competitor’s newly introduced product. And then last November, we introduced the Genoa generation. Not only we learned from the prior three generation continue to improve the differentiation and performance. Frankly, it’s matched from the performance perspective, efficiency, security, every aspect of the Genoa product. And also, we expanded into the family, right, with the four different flavors of Genoa. So overall, we continue to drive the leadership.
I think to your question, when you think about the history over the last six years of our journey of technology leadership and the share gain, it’s not one thing or two things. It’s about architecture, design, IP, packaging technology. And working with a foundry partner, TSMC, closely not only to just use their foundry but optimize the process technology, optimize the packaging technology, co-design, co-optimization, and above all, it’s consistent execution. Of course, I was not here, but the team has executed, did exactly what they said six years ago with a regular cadence. So that combination of success formula, I think we’re going to continue, and that’s how we can continue to drive leadership. We always assume it’s a very competitive market.
Harlan Sur
Perfect. And then my next question, and Ruth, you touched upon this a little bit is as we move into this period of accelerated compute, we continue to hear that software, software framework is becoming a huge differentiator, right? And your competitors that compete with you in the more accelerated compute AI-focused markets tend to lead with their software frameworks. And so, I’m wondering, what is the AMD team doing here to close the gap on software, AI frameworks, and just overall accelerated compute ecosystem development?
Ruth Cotter
So, I think our history and history of innovation has been grounded in hardware traditionally. And we’ve also obviously done a lot on the GPU side as it relates to software. Now moving into the data center with GPUs, there’s been a lot of focus on software, to your question. We’ve done a nice job addressing the external workloads and frameworks that are required to be supported all the large ones such as TensorFlow, PyTorch, et cetera, and staying abreast of what may be next to come as part of that effort and focus.
We’ve been deploying incremental resources in that area consistently for the last several years to ensure that we can keep a pace with that development. And now the focus is, as it relates to internal workload, several of the hyperscalers’ own software capability, making sure that, that is running well on our hardware. MI250 has been very helpful there with several providers as we’ve looked to porting their in-house software onto our hardware in preparation then for MI300.
So, we feel well positioned. And there is some more work to do in the internal workload space. As we continue to add more customers to our portfolio, that work will continue. But we’ve had some good momentum. So, we feel well positioned now when we’re showing up in front of customers today as compared to how it might have been 18 months ago.
Harlan Sur
Let’s switch to talk about one of my favorite market segments for the AMD team, and that’s the embedded markets, right? And good to see the sort of strong near-term growth dynamics in the team’s embedded business. Very diverse end markets, industrial, auto, comms infrastructure, test, and measurement. And given their strong market position here, the Xilinx team is in a good position to catalyze EPYC, rising CPU attach rates to their FPGA, and adaptive solutions, right?
I think that embedded x86 CPU is about a $6 billion to $8 billion per year market opportunity. And the AMD team has fairly small CPU share here. So, given your year with Xilinx in the portfolio, can you just give us an update on the synergy unlock? And what is the AMD team doing to aggressively drive higher compute attached to Xilinx’s products?
Jean Hu
I’ll start, and Ruth can add since she actually led the integration of Xilinx and AMD. I think, first, Xilinx business has been performing extremely well. As you said, it’s industrial, it’s aerospace, defense and a lot of the drive come from the digitization and automation, especially in the industrial or aerospace vision. Those kinds of applications, we do see not only Xilinx FPGA gaining share, but they also incorporated the SoC, which have content gain. So, it’s a tremendous business right now.
And what we are seeing is when we combine with AMD, the embedded processing market is opening up to the whole company. And we never paid much attention in the embedded market, but you are absolutely right. It’s a large opportunity with more than 60,000 customers from Xilinx side. We actually see tremendous leverage.
We are getting design wins, some of them we announced. For instance, in security firewall, in networking side, we continue to really getting more and more market this year. It’s actually a very large opportunity for long term because the product life cycle in those markets tend to be very long. But we are really pleased with the initial design win opportunities. There are more, right?
Ruth Cotter
Yes. And I think we’re both engineering-led first companies, which meant that the teams came together very quickly and identified very commonality as it relates to how can we bring the IP of both companies together as we look to advance the long-term road map. More to come as we continue to work through that. And then I think rolling AI and all the cross-company AI efforts under Victor, Victor Peng, and making sure that he has oversights not only across all the different pieces of IP but also the 6,500 customers that Xilinx brought into our company as we look to fan that AI strategy from supercomputers to cloud to endpoints to PCs as we think about those multiple opportunities, it’s also quite exciting.
But the teams have come together well. The cross-sell opportunities are compelling, to Jean’s point, as we think about us offering what was Xilinx’s greater scale, the ability to bid for larger bids and opportunities. And likewise, the opportunity to sell AMD products into their — what was their former customer base is very attractive to us as we look to grow that CPU share, to your question.
Harlan Sur
Yes, absolutely. On the competitive front, we continue to see ARM-based servers continuing to take share from x86 at a very gradual pace, right? And most recent estimates, I think, show that ARM-based server CPUs make up about 4%, 5% of the total server CPU market. You have one start-up out there. Obviously, Amazon has been deploying ARM-based servers for several years now. We hear about some of the other cloud titans having some programs there.
Are these ARM programs mostly internal specialized programs or are they being targeted for broader data center footprint compute workloads? And how do you see the broader potential proliferation of ARM-based servers in the data center?
Jean Hu
I think in data center, the demand for compute overall just increasing significantly every year. From the angle we see, ARM largely is being deployed either in customer ASIC or semi-custom programs because they have a specific workload they really want to optimize. It’s all about the TCO in the end. So, from our side, we definitely — Bergamo was one for the programs. We really try to address the same TCO concern in cloud-native applications to really offer customers the best TCO so they don’t need to do their own customer programs.
At the same time, we have the ARM capability, both from Xilinx and from Pensando acquisition, and we have a really great semi-custom team. So, if really there’s true economics of doing some kind of customer chip or even ASIC for a particular workload, if it makes sense for us, definitely, we are open for that.
Harlan Sur
On the financial side, it looks like gross margin is bottoming here in the first half of the year at 50%. If the full-year growth profile plays out, you’ll see revenues growing in the second half quite strongly. You’ll be seeing your gross margins expanding as well. Understand the better mix dynamics, right, i.e., like stronger data center revenues in the second half. But Jean, you particularly mentioned PC improvements in the second half as a significant contributor to the gross margin expansion.
And that’s created, I think, a little bit of confusion amongst out there. So maybe — I think part of it was because knowing that your competitor may continue to be aggressive on PC client pricing. So maybe just walk us through the puts and takes on how do we think about the second half potential gross margin step-up.
Jean Hu
Yes. So, if you look at the first half of our business and the gross margin around 50%, as we mentioned, we are actually very pleased with the strong gross margin performance in both the data center and the Embedded segment. Both segments together accounted for more than 50% of revenue. So once the headwind we see is actually on the client side because we are undershipping the consumption significantly and going through the inventory digestion in the downstream supply chain. That really impacted our segment gross margin significantly.
Second half, we do believe the inventory digestion in the downstream supply chain will largely done. And then the client’s gross margin actually going to improve second half. Of course, we’re not talking about going back to last year’s level but to improve from a very lower level to second half actually is the major driver of our second half gross margin improvement. There are other puts and takes, but overall, this one is quite significant because of how low our gross margin currently is.
Harlan Sur
And there have been also some questions going back to sort of the Embedded business. But I think the team is anticipating maybe a slight step down in the Embedded business in the second half versus the first half. Auto, industrial, service provider trends are holding up still relatively well. So, what are the drivers of the slight deceleration in Embedded as we think about the second half revenue profile?
Jean Hu
Yes, you’re right. What we see is industrial, aerospace, and the defense as far as automotive have continued to be very strong and solid demand. Second half for communication, definitely, we see weakness. You do see the global CapEx in communications side is coming down and the 5G is — actually, deployment is slowing down globally. So those kind of things impacting our wireless business and the wireline communication business. So that’s one of the headwinds. And of course, consumer, we continue to expect headwinds on consumer side. But overall, I think if you look at the year-over-year, the Embedded business continue to perform extremely well.
Harlan Sur
You’ve had time to review the team’s product portfolio, right, competitive position, customer opportunities. What’s your view on the team’s longer-term financial targets that were laid out at last year’s Analyst Day, right? 20% revenue growth, gross margins, 57% or better, and mid-30s operating margins.
Jean Hu
Yes. I think if you look at our product portfolio and the strategic direction to focus on data center, CPU, GPU, adaptive compute, and also Embedded, a very diversified business. In the longer term, we do expect our gross margin continue to go up. The 57% target for Investor Day two years ago continue to be our target. We do feel pretty good about the trend going forward. And of course, with the gross margin going up and the top-line revenue growth, the operating margin, the cash flow generation of the company is tremendous.
Harlan Sur
As you, Jean, have looked at the technology, the IP portfolio, anything that you’re seeing that you feel like the AMD team needs to add a little bit more scale, a little bit more capability as you go after client and data center sort of compute trends?
Jean Hu
I do think we have a complete portfolio. Ruth, you can add your comment, is the company is very well positioned from IP portfolio technology perspective. We always look at the AI investment, software talent, those kind of small acquisitions. If they are a good fit, we definitely will invest to address the opportunities ahead of us. But overall, we feel pretty good about our portfolio.
Ruth Cotter
Yes.
Harlan Sur
With the last couple of minutes, Jean or Ruth, did you guys have anything — any closing comments that you want to add? Anything — your discussions with investors today with your meetings or over the past couple of weeks since earnings? Anything that you want to communicate to the market in terms of things that you feel are underappreciated or key focus areas for the team that we should be thinking about as moving forward here?
Ruth Cotter
I think we feel we’re very well positioned. Obviously, there’s this mixed demand environment everybody is working through, to Jean’s earlier point. But as we come out of that, we have an incredible product portfolio. We have really strong road maps that continue to advance in both hardware and software. And we continue to leverage the Xilinx IP portfolio as well as what we acquired with Pensando as we look to gain more share in a more diversified market opportunity, which is very exciting.
And then 2023 brought AI, which we’re very well positioned to capitalize on. The team is very focused on maintaining execution as we continue to scale and grow. And we’re delighted to have Jean on board with us for the next inflection point and leg of the journey.
Harlan Sur
Absolutely. Well, Jean and Ruth, thank you very much for your participation today. Looking forward to watching the great execution by the AMD team as we move through the second half of this year. Thank you.
Ruth Cotter
Thank you, Harlan.
Jean Hu
Yes, thank you.