All right. Welcome back to the Barclays Tech Conference. I'm Tom O'Malley, Semiconductor and Semiconductor E quipment Analyst here at Barclays. Pleased to have Jean Hu, CFO and Treasurer of AMD. Thank you for being here, and thank you for being a staple here for the last couple of years as well. We really appreciate it.
Yeah, thank you for having us.
So speaking of last year, we sat here a year ago and chatted on market share and how you have come from a place both on the PC side and the server side where it was really a second source story coming from the bottom, kind of moving to parity and in some instances to leadership. So can you talk about that journey and really from here, where are the aspirations focused on across those businesses?
Yeah, yeah, I appreciate the question. We are very pleased with our performance in both the server side and also on the PC side. As you know, in the server CPU market, starting from a very low single digit market share a few years ago to now, last Q3, we reported the earnings. We got to 34% market share. So it has been tremendous. I think when you look at that success, first thing is AMD is always focused on product, innovative product generation by generation. So when you look at the server side from first generation Naples until each generation now at the Genoa and the Bergamo fourth generation family, we got tremendous TCO performance, not only performance per watt, but performance per dollar. And now we launched Gen 5, which is the Turin family product, will continue to drive the leadership performance.
So technology-wise, it's about leadership, not only architecture, process technology, chiplet, and also packaging technology. But also, if you look at the team's history, it's always about consistent execution. So each generation of roadmap, we delivered on time with leading performance. Third thing I would say is our team is really good to work with the customer closely. So not only we got feedback from prior generations, but also from customers. So our product is not just about leadership performance, but also it's what exactly customer needs. When you look at that, is that really help us on the server side, we absolutely are powering the most critical workload across both enterprise and cloud, right, from enterprise scale up and cloud scale out everywhere, the major applications across all different markets. So it's very exciting. On the PC side, it's also exciting if we actually made tremendous progress.
When we look at currently our product portfolio lineup, we actually have the strongest product portfolio in the PC market in both the desktop side and also in the mobile side. In the desktop side, in Q3, our market share got to 27%. On the client mobile side, we got to 19% market share. Again, it is about technology leadership, how we can provide customers performance efficiency across the board. So exciting journey, but the momentum will continue and will continue to drive the leadership technology and the product.
Super helpful, and all of these good things are happening, and of course, I'm going to zone in on one area where you're a little underindexed, which is on the enterprise side in terms of server, so I guess the question there is, why hasn't the success translated as quickly? I know enterprises are more slow moving, and what kind of steps can you take to further penetrate that market?
Yeah, we absolutely in the cloud market, we're fairly represented. And in enterprise, we're still very much underrepresented. When you look at the technology-wise, the TCO performance from AMD is both in enterprise and in cloud. So we absolutely can not only address cloud-native applications, but in enterprise, there are more diverse applications from virtualization to database to all different kind of ERP. We actually can provide the best performance. However, enterprise side is a little bit different on the go-to-market approach. As you can imagine, in cloud market, it's all about the TCO. And once they switch, its volume production adoption is very quick. But on the enterprise side, you really need to convince each CIO with 1,000 enterprise customers that they have the TCO performance.
So, you do need to have a lot of feet on the street to address each CIO to give them proof of concept so they can see the TCO performance. What we have seen is over the last couple of years, we made a tremendous effort to increase our FAE to support each enterprise customers. Now we're seeing the benefit of that. I think the last five quarters, each quarter, our enterprise business has grown double-digit. And that accumulation of effort, you're going to see more momentum going forward. So the combination of focusing on the go-to-market, at the same time, continue to provide best TCOs, we are also seeing enterprise customers start to upgrade because in today's data center, we all know power, space, those are very limited.
So, modernize data center, upgrade data center. AMD's solution can actually provide the best TCO from a power perspective and from performance per dollar perspective. So, it is why we feel pretty good about continued progress, especially with the Turin launching in Q4 right now. We're going to see momentum into next year.
A question that I think I've heard a lot this week, and I had the co-CEOs of Intel here this morning, is, obviously, you don't wish ill upon your customers. But when there are changes, customers are going to change their preferences and profiles and conversations. You mentioned the enterprise game is really a ground game. You go CIO, CIO, CIO, and it's those conversations that win you market share over time. Have you seen any change in the very recent history, as in two weeks, in those conversations? How would you think about your business there given the change?
I think what AMD has always been doing is assuming our competition is going to do very well. So our job is to make sure that's the assumption and that we stay competitive and drive the leadership performance and also make sure our customer gets the best economics. I think to a certain degree, when you look at the enterprise market, the porting process actually is not that difficult because x86 for both of the companies, it's the same instruction set, and it's really about convincing people we can provide a better TCO, which we do.
So, I guess, broadly speaking, about the market versus the competitive dynamic, there's this view that dollars are being sucked out of the traditional server market and into the AI ecosystem of which you've had benefit. We'll move to AI after. But just on the traditional server market, when you look out at next year, is this a year where you see some recovery or is it similar to what we've seen over the past couple of years with maintained muted spend?
Yeah, yeah, great question. Maybe let's take a step back. Our view has been we are in a super compute investment cycle, and of course, Gen AI has been driving quite significant investment and also adoption of Gen AI. You can see a significant increase of accelerator market, very fast expansion. At the same time, when you really think about general compute, what CPU is really powering is actually foundational critical workload, right? From enterprise virtualization, your database, your ERP system, even including your storage, it's powered by general CPUs, and then in the cloud, the same thing. When you look at the cloud native workload from Shopee, Amazon, social media, Facebook, WhatsApp to video streaming, Netflix or Zoom, all those things are powered by general compute. That's CPU.
So we do see like, OK, AI is going to grow much faster, but the demand for fundamental applications when everybody increasing engagement in their platform is going to continue to grow. And of course, the innovation we have been pushing out is we can have more core count. We can provide more performance. So we have been supporting the continued demand increase. But overall, it is a very large market. We continue to see strong demand in cloud. We see modernization. We see the limitation on space and the power. So customers actually need to upgrade. They also see their platform engagement is increasing. They need more CPUs. In enterprise, we also start to see the early signs of refresh cycle. It's the same logic. You need more compute to support your applications, but you have a database, data center, and the power limitation.
So you want to get the best TCO from your suppliers. That's why we do think this CPU market not only is going to grow, but also we're going to continue to be able to gain share because of the performance.
OK. Regardless of how that traditional market continues to grow or at the rate at which it grows, you have seen some really impressive growth in the MI series this year. I think if we sat here a year ago and you had confirmed to the crowd that you'd do more than $5 billion, people would have been shocked. It's a really, really strong ramp, so can we talk about how that is going in terms of the total ramp today? You're going to see a transition in product for really the first time in that series, so when is that transition occurring? And then just any kind of change in your outlook versus the last time we kind of spoke?
Yeah, yeah, thank you. Last year when I was here, our MI300X revenue was zero. And so it's amazing during this 2024, what we have done as a company is from zero to going above $5 billion for this 2024. That's a great success. I think when you look ahead to 2025, first at a high level, the backdrop is we continue to see the continued investment in AI infrastructure built out. That has been ongoing with all our customers. And you guys can see the third-party data also. Secondly, use cases have been increasing dramatically, right? Every week we see some new use cases in AI, which definitely when you do the inferencing, when you have those use cases, you drive return on investment. So the backdrop of the market continues to be very strong. And for us, the team has been executing extremely well. MI300 ramping successfully.
And now MI325, we launched literally this quarter. We'll start to see revenue in Q1 next year. And then MI350, second half next year. So when you look into 2025, we actually have a much stronger product portfolio versus 2024. And at the same time, the market backdrop continues to be really good. More importantly, when you look at 2025, it's just multi-year journey we are on. The way we think about each product we're addressing is always multi-generational, not only deliver the product execution by each generation, but also engage customers. In AI case, right, we also continue to invest in software. We're acquiring ZT Systems to also build our system expertise time to market. So overall, 2025, we feel really good about the opportunities ahead of us. And more importantly, we'll be able to continue to drive the trajectory and the momentum of our business.
So first, I want to talk about the cadence of product transitions and then about customer diversity. But in terms of upgrade cycles with different chip families, generally there's digestion, and then there's a ramp-up period of time. And you guys have done a really good job of kind of masking that even with your first transition last at the end of last year in December. Can you talk about how you manage that, how you manage through that? Is that a conversation with customers? Do you naturally see a wind down quicker than you see a wind up? Anything that you can give just to talk about the cadence of those ramps?
So what we're seeing in the AI market is you are seeing many different models and different diversified needs. So it's not like, OK, the most advanced model, they are large clusters, training, inferencing. You need the most advanced model, the technology and GPUs. But at the same time, increasingly, we see so many models across different inferencing. So different customers actually have different needs. And our engagement with our customers, existing customers, have always been multi-generational. So you don't just sell them one product. The engagement tends to be really deep. It's not only MI300 we're selling, but MI325, MI350, even MI400. And then we have been broadening our customer base. We talk about over 100 customer engagement. Then different customers really have a different need. So we do think you will see MI300, MI325, MI350.
They may coexist for a while because just different customers. You're meeting different customers' needs.
And then there's the aspect of customer diversification as well. Obviously, re:Invent was very recently. You have one large customer that's well known today. But in terms of customer diversification, could you talk to where you're seeing green shoots maybe with other large customers and your ability to kind of expand? Is that a function of just seeing more inferencing in the market, or is that just really customers taking more time to come to your platform?
Yeah, of course, we talk about our large customers like Microsoft, Meta, Oracle, and not only MI300 and our software, ROCm, are powering the most demanding workload and the model with our key customers. But also, we are engaging with other customers. AWS, we have been engaging with AWS. And of course, for third-party workload, you have to work with enterprise customers also. So that engagement continues. We are also engaging with other hyperscale customers. So you should expect us not only with existing customers. We're going to broaden application, both inferencing. In Meta's case, we're also doing training with Meta. The model applications are going to broadening. But we're also engaging with new customers, addressing new applications. That is how we think about building a business for the long term.
Customer diversification is going well, it seems. That's checkmark number one. Two is, I think that increasingly you've seen the diversification and the differentiation in the market between NVIDIA and other players is their ability to scale to system-based solutions, right? They have their own protocol. They're scaling up and scaling out. They're clustering. I think what's been more difficult for non-NVIDIA players just broadly has been the fact that you need a bunch of different players to come together in order to get that scaling architecture to work. So you obviously acquired ZT. That's part one. But how is it going in terms of your ability to replicate that sort of systems-based architecture? Are you running into kind of any walls in terms of that process? And how are you thinking about scaling that architecture over time?
Yeah, yeah, that's a great question. You are absolutely right. When we think about Gen AI market, when we look at the opportunities, $500 billion, we talk about it, we strongly believe the majority of the market is going to be addressed by general-purpose GPU, including system-level software solution, the rack, the cluster level. You really need to have a system expertise, have a strong software base to support that future market because that's what provides the best TCOs. If you look at our journey, not only do we have the annual cadence on the GPU side, we are increasingly investing in networking, in buying ZT Systems. That will give us system expertise to build rack-level and cluster-level solutions. And also our team continues to push forward about software investment. That is what we think the majority of the market will be.
There are some ASIC opportunities, ASIC market also, but for us to build that overall solution to be a major player in this market, ZT Systems is one of the examples. We expect to close the transaction early next year, and our MI350 will get some benefit, but MI400, which is 2026, will get a full benefit of ZT Systems, so it's an exciting time.
OK, and then I think that you've been very prescriptive, which has been helpful on the gross margin ramp of the MI series. And then there's also to be the layer on of a more systems-based architecture, so you said over time moving more towards corporate average and then perhaps surpassing that. Could you give us an update of how that margin structure is progressing, when you see that crossover point? Anything that varies in terms of your initial expectations?
Yeah, I think first thing is the gross margin is a very important metric to AMD. With all the R&D investment we are making, the gross margin really is a reflection of your IP and your engineering excellence. So when you look at the 2024 versus 2023, I think 2023, the fiscal year, our gross margin was 50%. And in 2024, we really have been improving gross margin. We are expecting 2024 is at 53%, quite a significant expansion. Going forward, that continues to be our objective to expand the gross margin. On the data center GPU side, we did talk about it's below corporate average. I think given the large opportunities in front of us and especially how fast this market is expanding, of course, our priority number one is to meet customers' need to address the large market opportunity.
As a CFO, you always think about the gross margin dollars, right? Percentage is super important. But when a market is expanding so quickly, you can get more gross margin dollars. That is our priority. But over time, when we look at how complicated the technology for data center GPU, it's absolutely the segment that you should expect us to continue to improve gross margin over time. It will be accretive to corporate average.
Helpful. Using gross margin as a bridge here, I want to go to the PC side and talk about AI PC. So we've heard a whole bunch of different things about what the AI PC market is. It's updated pretty frequently. What is your view of the AI PC market moving into 2025? And we've heard it's a gross margin headwind. It's a gross margin tailwind. It's been all over the place. What is it for AMD? Is this something that should help ASPs?
Yeah, our view has not changed. We always said that even at the very beginning, we think AI PC adoption in 2024 is going to be moderate. It's really in 2025 because you do need all the AI PC applications for customers who want to buy AI PC. And when you look into 2025, we do think there are going to be more applications. We do think with Windows 10 end of life, you will see a refreshing cycle. When people upgrade their PCs, if you have an AI PC, if you have applications, you would expect people will upgrade to new AI PCs. And also, you are going to see more offering of AI PCs. So we do think 2025, of course, the overall PC market in our view is going to grow, maybe low- to mid-single-digit.
The seasonality is always the first half is lower and the second half is higher. That is going to be the case. But AI PC, we do think the momentum will be much more significant in 2025.
And then obviously the gross margin side of that. I know that you've done a very good job with ASPs driving some growth on the PC side. But does AI PC lend to ASP increases, or is that more of a competitive dynamic?
Yeah, I think AMD actually has, if you look at our Ryzen AI 300, we probably have the best CPU inside and the best GPU inside, and also the NPU, which is the AI accelerator. We do think when you offer more features, you should get higher ASP. That's our view because you're providing customers much more. So typically, that is how it's focused on. Higher ASP gross margin should be at least similar or better.
And then I just wanted to kind of conclude the topic on competition. You're hearing more about ARM. There was exclusivity going into the end of this year. You're going to see more engagement with the ARM ecosystem into the beginning of next year. What's your view on what the right level of penetration will ultimately be from an ARM perspective? Is that something that you factor in when you look at the growth of your business in the '2025?
Yeah, I think the first thing is I don't know about you guys. I never look at what's in my PC, like ARM or x86. I think for most customers, what they care about is battery life, performance, efficiency. So what you are seeing is AMD now has one of the strongest product portfolios in history. We are offering not only performance. We are continuing to extend battery life to really make sure we can offer customers both power, efficiency, and also performance. That's what we're striving for. And of course, ARM PC this year, the share is low. The ecosystem, especially on the commercial side, you do need to be backward compatible with all your applications. Over time, the way we look at this market is we are trying to innovate and provide a customer-best product. And AMD, we actually really view ourselves as a high-performance computing company.
Some part of our business actually works with ARM. Xilinx, they have always been partnering with ARM, so for us, it's not about x86 or ARM. It is like a high-performance compute, how we can provide customers the best performance, and of course, we have the capabilities. If a customer wants us to do ARM-based PCs, we absolutely have the capabilities.
Perfect segue into your embedded business. That's been going through a bit of a longer recovery than I think most would have expected. And you've kind of talked about some growth into the 2025 period. You've not been super prescriptive there. But two questions. One, are you starting to see some green shoots in that business? I know that one of the takeaways from the conference for us thus far has been you've seen a little bit better telco. So that's obviously not their wireless business as much, but maybe a little on the wireline side. And then two, you're hearing about a potential spinout of the main competitor to Xilinx. Do you think that changes the competitive dynamic in any way?
First, our embedded business has been doing really well from a design win perspective. If you look at it despite a very deep inventory correction cycle, our design wins continue to be quite significant year- over- year. And then when you look at the business side, it actually bottomed. In Q3 sequentially, we actually see increase. In Q4, it's more stabilized. When we look at the end market we cover, we actually cover a very broad end market. When you look at AMD's embedded business, we actually tend to be on the mid and high end of the FPGA business. So aerospace defense is actually doing reasonably OK. And then testing emulation is actually doing good. I think, consistent with what you are saying, industrial has continued to be quite challenging. That makes the demand environment and communication stabilized. But it is the headwind in 2024.
Communications definitely was one of the sectors that was challenged. And of course, automotive is small for us, but it seems like it's still struggling. So it's a mixed environment. But we do think going into 2025, overall, you should expect a gradual recovery.
I want to just ask on the broader model into 2025 really quickly. So it sounds as though the server, core server business is doing well. PC, more second half weighted. And you talked about the last earnings call, a little bit stronger end of this year, maybe a little softer first half of next year. But in terms of the total growth of the business, things sound good. In terms of the operating leverage that you can get, how do you think about spend? And if I look at you've grown quite robustly over the past couple of years. If spend is not as, if revenue is not as aggressive as those big growth years off the bottom, how do you think that you can move that OPEX lever to get more drop down to the bottom of the model?
Yeah, you are right. When we look back in 2024, our data center business and our client business performance have been tremendous. Data center is like literally with the data center GPU come to more than $5 billion. We almost doubled our data center business. The headwind, it's really on the gaming and embedded business, which they are behind us when we look ahead of 2025. On the operating model perspective, you should expect us to continue to invest in R&D aggressively because that is how we drive the multi-year generation roadmap to really continue to drive the leadership. But overall, the OPEX increase should be less than top line revenue growth. And then that's how we drive the operating model leverage. If you look at the 2024, it was similar. We want to make sure earnings expansion is much faster than top line revenue growth.
That's how we think about it. And that's how we are going to drive it.
All right. Well, I really appreciate the time. Thank you for being here. It's a pleasure as always.
Yeah, thank you for having us. Thank you, everyone.