Advanced Micro Devices, Inc. — 2024 Q1
Transcript
Each turn shows the speaker, their inferred role, the section, and that turn's net sentiment (×1000).
Greetings, and welcome to the AMD First Quarter 2024 Conference Call. [Operator Instructions] As a reminder, this conference is being recorded.
Thank you, and welcome to AMD's First Quarter 2024 Financial Results Conference Call. By now, you should have had the opportunity to review a copy of our earnings press release and the accompanying slides. If you have not had the chance to review these materials, they can be found on the Investor Relations page of amd.com. We will refer primarily to non-GAAP financial measures during today's call, and the full non-GAAP to GAAP reconciliations are available in today's press release and the slides posted on our website.
Thanks, Mitch, and good afternoon to all those listening today. This is an incredibly exciting time for the industry as the widespread deployment of AI is driving demand for significantly more compute across a broad range of markets. Under this backdrop, we are executing very well as we ramp our data center business and enable AI capabilities across our product portfolio.
accelerate our Data Center growth by ramping Instinct GPU production and gaining share with our EPYC processors; launch our next-generation Zen 5 PC and server processors that extend our leadership performance; and expand our adaptive computing portfolio with differentiated solutions.
Thank you, Lisa, and good afternoon, everyone. I'll start with a review of our financial results and then provide our current outlook for the second quarter of fiscal 2024. We delivered strong year-over-year revenue growth in our Data Center and Client segments in the fourth quarter and grew 230 basis points of gross margin expansion. For the first quarter of 2024, revenue was $5.5 billion, up 2% year-over-year as revenue growth in the Data Center and the Client segment was partially offset by lower revenue in our Gaming and Embedded segment.
Thank you, Jean. Paul, we're happy to poll the audience for questions.
[Operator Instructions] Our first question is from Toshiya Hari with Goldman Sachs.
Lisa, my first question is on the MI300. You're taking up the full year outlook from $3.5 billion to $4 billion. I'm curious what's driving that incremental $500 million in revenue? Is it new customers? Is it additional bookings from existing customers? Is it more cloud? Is it more enterprise? If you can sort of provide color there, that would be helpful.
Great. Thank you, Toshiya, for the question. Look, the MI300 ramp is going really well. If we look at just what's happened over the last 90 days, we've been working very closely with our customers to qualify MI300 in their production data centers, both from a hardware standpoint, software standpoint. So far, things are going quite well.
And then as my follow-up, I was hoping you could speak to your Data Center GPU road map beyond the MI300. The other concern that we hear is your nearest competitor has been pretty transparent with their road map, and that extends into '25 and oftentimes '26. So -- and maybe this isn't the right venue for you to give too much, but beyond the MI300, how should we think about your road map and your ability to compete in Data Center?
Yes, sure. So look, Toshiya, when we start with the road map, I mean, we always think about it as a multiyear, multigenerational road map. So we have the follow-ons to MI300 as well as the next, next generations well in development. I think what is true is we're getting much closer to our top AI customers. They're actually giving us significant feedback on the road map and what we need to meet their needs.
Our next question is from Ross Seymore with Deutsche Bank.
The non-AI side of the Data Center business, it sounds like the enterprise side has some good traction even though the sequential drop happened seasonally, Lisa. But I was just wondering what's implied in your second quarter guidance for the Data Center CPU side of things? And generally speaking, how are you seeing that whole kind of GPU versus CPU crowding out dynamic playing out for the rest of 2024?
Yes, sure, Ross, thanks for the question. I think the -- our EPYC business has actually performed pretty well. The market is a bit mixed. I think some of the cloud guys are still working through sort of their optimizations. I think it's different by customer. We did see here in the first quarter, actually, some very nice early signs in the enterprise space, sort of large customers starting refresh programs. The value proposition of Genoa is very, very strong, and we're seeing that pull through across the enterprise.
And I guess as my follow-up, just switching over to the Client side. I noted you guided it up sequentially. Any sort of magnitude around that for the second quarter? And perhaps, more importantly, when you talk about the whole AI PC side of things, do you believe that's more of a units driver for you, an SAP driver, or will it be both?
Yes. So I think, again, I think we're pretty excited about the AI PC, both opportunity in, let's call it, the near term and even more so in the medium term. I think the client business is performing well, both on the channel and on the MNC side. We expect clients to be up sequentially in the second quarter.
Our next question is from Matt Ramsay with TD Cowen.
Lisa, I have sort of a longer-term question and then a shorter-term follow-up one. I guess the -- one of the questions that I've been getting from folks a lot is, obviously, your primary competitor has announced, I guess, a multiyear road map. And we continue to hear more and more from other folks about internal ASIC programs at some of your primary customers, whether they be for inference or training or both.
Yes. Sure, Matt. Thanks for the question. So look, I think one of the things that we see and we've said is that the TAM for AI compute is growing extremely quickly. And we see that continuing to be the case in all conversations. We had highlighted a TAM of let's call it, $400 billion in 2027. I think some people thought that was aggressive at the time. But the overall AI compute needs, as we talk to customers is very, very strong. And you've seen that in some of the announcements even recently with some of the largest cloud guys.
Lisa, as my follow-up, a little bit shorter term. And I guess, having followed the company super closely for a long time, I think there's been -- there's always been noise in the system from whether the stock price is $2 a share or $200. There's been kind of always consistent noise one way or the other, but the last 1.5 months has been extreme in that sense.
Sure, Matt. Look, I think I might have said it earlier, but maybe I'll repeat it again. I think the demand side is actually really strong. And what we see with our customers and what we are tracking very closely is customers moving from, let's call it, initial POCs to pilots to full-scale production to deployment across multiple workloads. And we're moving through that sequence very well. I feel very good about the deployments and ramps that we have ongoing right now.
Our next question is from Aaron Rakers with Wells Fargo.
I apologize if I missed this earlier, but I know last quarter, you talked about having a -- securing enough capacity to support significant upside to the ramp of the MI300. I know that you upped your guide now to $4 billion. I'm curious how you would characterize the supply relative to that context offered last quarter as we think about that new kind of target for? Would you characterize it as still having supply capacity upside potential?
Yes, Aaron. So we've said before that our goal is to ensure that we have supply that exceeds the current guidance, and that is true. So as we've upped our guidance from $3.5 billion to $4 billion, we still -- we have supply visibility significantly beyond that.
Yes. Okay. And then as a quick follow-up, going back to an earlier question on server demand, more traditional server. As you see the ramp of maybe share opportunities in more traditional enterprise, I'm curious how you would characterize the growth that you expect to see a more traditional server CPU market as we move through '24 or even longer term, how you'd characterize that growth trend?
Yes. I think, Aaron, what I would say is there are -- the need for refresh of, let's call it, older equipment is certainly there. So we see a refresh cycle coming. We also see AI head nodes as another place where we see growth in, let's call it, the more traditional SSD market. Our sweet spot is really in the highest performance, sort of high core count, energy efficiency space, and that is playing out well.
Our next question is from Vivek Arya with Bank of America Securities.
Lisa, I just wanted to go back to the supply question and the $4 billion outlook for this year. I think at some point, there was a suggestion that the $4 billion number, right, that there are still supply constraints. But I think at a different point, you said that you have supply visibility significantly beyond that.
Yes. Vivek, let me try to make sure that we answered this question clearly. From a full year standpoint, our $4 billion number is not supply capped -- I'm sorry, yes, it's not supply capped. It is -- we do have supply capability above that. It is more back half weighted. So if you're looking at sort of the near term, I would say, for example, in the second quarter, we do have more demand than we have supply right now, and we're continuing to work on pulling in some of that supply.
Maybe one, not on MI, but maybe on the Embedded business. I think you sound a bit more measured about Q2 and the second half rebound, which is similar to what we have heard from a lot of the auto industrial peers. But where are you in the inventory clearing cycle? And if Embedded has a somewhat more measured rebound in the back half, what implication does that have on gross margin expansion? Can we continue to expect, I don't know, 100 basis points a quarter in terms of gross margin expansion because of the Data Center mix? Or just any puts and takes of Embedded and then what it means for gross margins in the back half?
Vivek, thank you for the question. I think the Embedded business declined a little bit more than expected, really due to the weaker demand in some of the markets, very specifically, communication has been weak. And some pockets of industrial and automotive, as you mentioned, it's actually quite consistent with the peers.
Our next question is from Timothy Arcuri with UBS.
I also wanted to ask about your data center GPU road map. The customers that we talk to say that they're engaged, not just because of MI300, but really because of what's coming. And it seems like there's a big demand shift to rack scale systems that try to optimize performance per square foot given some of the data center and power constraints. So can you just talk about how important systems are going to be in your road map? And do you have all the pieces you need as the market shifts to rack scale systems?
Yes, sure, Timothy. Thanks for the question. For sure, look, our customers are engaged in the multigenerational conversation. So we're definitely going out over the next couple of years. And as it relates to the overall system integration, it is quite important. It is something that we're working very closely with our customers and partners on. That's a significant investment in networking, working with a number of networking partners as well to make sure that the scale-out capability is there.
And then just as a quick follow-up. I know this year it looks like it's going to be pretty back-half loaded in your server CPU business, just like it was last year. I know you kind of held our hands at about this time last year sort of on what the full year could look like and how back-end loaded it could be.
Yes. I mean, I think, Tim, I think the best way to say it is our Data Center segment is on a very, very strong ramp as we go through the back half of the year. Server CPUs, certainly, Data Center GPUs, for sure. So I don't know that we're going to get into specifics, but I could say, in general, you should expect overall at the segment level to be very strong double digits.
Our next question is from Joe Moore with Morgan Stanley.
I wonder if you could address the profitability of MI300. I know you said a couple of quarters ago that it would eventually be above corporate average, but it would take you a few quarters to get there. Can you talk about where you are in that?
Yes. Thank you, Joe. Our team has done an incredible job to ramp MI300. As you probably know, it's a very complex product, and we are still at the first year of the ramp, both from yield, the testing time and the process improvement, those things are still ongoing. We do think over time, the gross margin should be accretive to corporate average.
Great. And then as a separate follow-up. On the Turin transition on server, I know when you had transitioned in generally, you said it could take a little while, that there were significant platform shifts and things like that. Turin seems to be much more kind of ecosystem compatible. How quickly do you think you might see that product ramp within our server portfolio?
Yes. Joe, I think from what we see, look, think Turin is the same platform so that does make it an easier ramp. I do think that Genoa and Turin will coexist for some amount of time because customers are deciding when they're going to bring out their new platforms. We expect Turin to give us access to a broader set of workloads. So our SAM actually expands with Turin, both in enterprise and cloud. And from our experience, I think you'll see a faster transition than, for example, when we went from Milan to Genoa.
Our next question is from Stacy Rasgon with Bernstein Research.
For my first one, I wanted to address the MI300 ramp into Q2. So you said you've done $1 billion, give or take, in cumulative sales, which puts it at maybe, I don't know, maybe $600 million in Q1. You're guiding total revenues up about $225 million into Q2, but you've got Client up, you've got traditional Data Center up, you've got Embedded flat. Gaming is going to be down, but I'd hazard a guess that the client and traditional Data Center offset it, if not more. Does the MI300 ramp into Q2? Is it more or less than the total corporate ramp that you've got built into guidance right now that you're expecting?
Stacy, thanks for the question. You always ask a math question. So I think, in general, it is more. The Data Center GPU ramp, it will be more than the overall company's $200-some million ramp.
Okay. So that means Gaming must be down like a lot, right, if client [indiscernible]
Yes, yes, you're right. Gaming is down similar zip code like Q1.
Got it. Got it. That's helpful.
So maybe -- yes, maybe let me give you some color about the Gaming business, right? If you look at the Gaming, the demand has been quite weak, that's quite very well-known and also their inventory level. So based on the visibility we have, the first half, both Q1, Q2, we guided down sequentially more than 30%. We actually think the second half will be lower than first half.
Got it. That's helpful. For my second question, I wanted to look at the near-term Data Center profitability. So operating profit was down 19% sequentially on 2% revenue growth. Is that just the margins of the GPUs filtering in relative to the CPUs? And I know you said GPUs would eventually be above corporate average. Are they below the CPU average? I mean they clearly are, I guess, in the near term, but are they going to stay that way?
Yes. I think you're right. It's -- the GPU gross margin right now is below the Data Center gross margin level, I think there are 2 reasons. Actually, the major reason is we actually increased the investment quite significantly to, as Lisa mentioned, to expand and accelerating our road map in the AI side. That's one of the major drivers for the operating income coming down slightly.
Our next question is from Harlan Sur with JPMorgan.
On your Data Center GPU segment and the faster time to production shipments, given you just upped your full year GPU outlook, how much of it is faster bring up of your customers' frameworks driven by your latest ROCm software platform and maybe stronger collaboration with your customers' engineers just to get them to call faster? And how much of it is just a more aggressive build-out plan by customers versus their prior expectations given what appears to be a pretty strong urgency for them to move forward with their important AI initiatives?
Yes. Harlan, thank you for the question. What it really is, is both us and our customers feeling confident in broadening the ramp? Because if you think about it, first of all, the ROCm stock has done really well. And the work that we're doing is hand in hand with our customers to optimize their key models. And it was important to get sort of verification and validation that everything would run well, and we've now passed some important milestones in that area. .
Yes, absolutely. Just going back, just kind of rewinding back to the March quarter. So similar to the PC Client business, right, which declined at the low end of the seasonal range, if I make certain assumptions around your Data Center GPU business, x that out of Data Center, it looks like your Server CPU business was also down at the lower end of the seasonal range. By my math, it was down like 5%, 6% sequentially. Is that right?
Yes. Harlan, this is Jean. I think the Server business has been performing really well. Year-over-year, it actually increased a very strong double digit. I think, sequentially, it is more seasonal, but we feel pretty good about continue gaining share there.
Yes. And if I'd just add, Harlan, to your question, we did see strength in enterprise in the first quarter. And I think that has -- that offset perhaps some of the normal seasonality.
Our next question is from Tom O'Malley with Barclays.
I just wanted to ask on the competitive environment. Obviously, on the CPU side, you had a competitor talk about launching a high core count product in the coming quarter, kind of ramping now and more so into Q3. You've seen really good pricing tailwinds as a function of the higher core capital. Can you talk about what you're seeing in that market? Do you think that there's any risk for more aggressive pricing, which would impact your ASP ramp for the rest of the year?
Yes. When we look at our server CPU sort of ASPs, they're actually very stable. I think we -- again, we tend to be indexed towards the higher core counts. Overall, I would say, the pricing environment is stable. This is about sort of TCO for sort of the customer environment and sort of our performance and our performance per watt, our leadership. And that usually translates into TCO advantage for our customers.
Helpful. And then just a broader question to follow up here. So I think you got asked earlier about the importance of systems. But on your end, how important is the Open Ethernet Consortium to you being able to move forward to systems? I know that, today, you obviously have some internal assets and then you can partner with others. But is there a way that you could be competitive before there is an industry standard on the Ethernet side? And can you talk about when you think the timing of that kind of consortium comes to market and enables you to maybe accelerate that road map?
Yes. I think it's very important to say we are very supportive of the open ecosystem. We're very supportive of the Ultra Ethernet Consortium. But I don't believe that, that is a limiter to our ability to build large-scale systems. I think Ethernet is something that many in the industry feel will be the long-term answer for networking in these systems, and we have a lot of work that we're doing with internally as well as with our customers and partners to enable that.
Our last question is from Harsh Kumar with Piper Sandler.
Lisa, I had two. One is for you and one perhaps for Jean. So we recently hosted a very large custom GPU company for a call. And they talked about kind of mega data centers coming up in the near to midterm, talking about nodes potentially in the 100,000-plus range and maybe up to 1 million. So as we look out at these kinds of data centers, from an architectural standpoint, it's not a situation where winner takes all, where if somebody gets in, they kind of get all the sockets?
Yes. So I'll talk maybe a little bit more at the strategic level. I think as we look at sort of how AI shapes up over the next few years, there are customers who would be looking at very large training environments and perhaps that's what you're talking about. I think our view of that is, number one, we view that as a very attractive area for AMD. It's an area where we believe we have the technology to be very competitive there.
That's wonderful. And then maybe a quick one for Jean. So Jean, I put everything into the model that you talked about for June, I get about more or less a $400 million rise in the June quarter over March. You mentioned that both MI300 and EPYC will grow. Curious if you could help us think about the relative sizing of those 2 segments within the growth? I'm getting -- the point I'm trying to make is I'm getting roughly about a $900 million number for MI300 for June. Is that -- am I in the ballpark? Or am I way off here?
Harsh, we're not going to guide a specific segment below the segment revenue. I think the most important thing is that we did say Data Center is going to grow double digit sequentially. I will leave it over there. Subsegment, there are a lot of details.
There are no further questions at this time. I'd like to hand the floor back over to management for any closing comments.
Great. That concludes today's call. Thanks to all of you for joining us today.
Thanks.
This concludes today's conference. You may disconnect your lines at this time. Thank you for your participation.