Welcome back, everyone. We're very happy to have Marvell here, Matt Murphy, Chairman and CEO, and Ashish Saran, Head of IR. So first off, thank you for being here. Really appreciate it. You guys just presented earlier last week. I think a good way to kick off, for those who didn't listen to the call, can you just give us the biggest takeaways from the call and just the moving pieces in how this next quarter, you see playing out?
Sure, yeah. I think, maybe just start at the top-
Yeah.
And then we can go into the quarter. It's been a really interesting year. If you look at kind of where we were at the beginning of the year, we were faced with, you know, a lot of churn in the industry, a lot of transition that was happening. We decided to take control of our own at that time and kind of adjust to the new reality. So as things happened throughout the year, first is, as we sort of reset and kind of framed our, our outlook for the whole year, we actually ended up delivering pretty well to what we thought. If you went back all the way, like, you know, call it in the March quarter, but we got there in a very different way.
We're going to get into that in a minute, in terms of where the revenue came from.
Yeah
And what, what sort of the dynamics were. But we definitely had a recognition that, that things had changed. So I'm very proud of our team. I mean, operationally, I think we took care of what we could control ourselves, but we took measures to basically refocus our R&D efforts, and really pivot, pivot our focus to accelerated computing. We'll get into that a little bit later.
Mm-hmm.
We had gross margins, you know, in the low 60s%. We talked about getting it back to the low end of our range by the end of the year. And we talked about also sort of in the context of our, the accelerated computing opportunities, also focusing our OpEx down and actually lowering. Those things we executed on. As we went from basically $460 million a quarter in OpEx, we're going to exit at $430 million. We've got the margin back up to 64% at the midpoint. And net net, we're about where we thought we would be on top line for the year, back to where we were back in March.
Now, what happened is, one, we had just seen the beginning signs of generative AI, and we were besides the impact of that, which basically we thought would be about $400 million this year and $800 million next year. At the end of this year, you have sort of the core markets recovering, you know, inventory correction, so on and so forth, and then we'd be off to the races. And I think what's happened is, now going through the quarter, we had excellent results in our data center business. In the third quarter, we overachieved what we thought.
We actually grew 21% naturally from Q2 to Q2 to Q3, and then we guided it up in the fourth quarter to grow well at mid-30% range, you know, on from a top-line perspective, sequentially. So data center is roaring. Biggest portion of that is AI, which now is well over $200 million per quarter exiting the year, which is really a positive. On the flip side, like, the core business, which we thought would sort of recover, is still going through a cyclical correction, and that's going to take a couple of quarters. But those markets are very, very important, and revenue and profit contributing markets from them, so forth, will normalize back.
But in the meantime, we've got our business firing on all cylinders, and it's providing some buffer against weakness in other parts of the business.
Super helpful. So why don't we dive into the area of the business that's doing quite well, data center. So I've lost track over the years. I think it's helpful to kind of reset. So the way I've kind of thought about that business is you've got this optical portion that's clearly growing very quickly. You've got this storage portion, which has, you know, been bouncing off the bottom, and then you've got this host of, you know, switching and legacy Cavium business. Can you just try to shape for us the relative sizes of the businesses? Because I think with the optics portion growing so quickly, people have lost track of the relevant size.
Sure. Yeah, maybe I'll give you a couple of data points and a way to think about that business. And maybe I'll start. I'll start with storage, and I'll kind of work my way, way back up. So our storage business historically was actually quite a stable and more predictable portion of our data center business. We had pivoted storage business over the last few years from a consumer kind of footprint to data, to data center. That, and then to call it at run rate, that was about $200 million per quarter. It's a way to think of the data center business for storage. It dipped, you know, well below $100 million per quarter at some point, so very much below kind of what we thought consumption is. It's now trended back up, right?
So it's back up at $100 million-
Yeah, it's $100 million.
-per quarter. And over time, at some point, I mean, all this stuff is going to kind of work its way through the system. It's just gotta, and it'll trend back up towards that run rate. So that's, that's sort of one piece to think about. The other one that you, that I think was mentioned more in the basket of other things we do, we actually call it out switching business. And so we, with our own R&D plus the acquisition of Innovium, now have a you know, a product focus and quite a big opportunity in data center switching. And so when we acquired Innovium, we talked about that being kind of a $150 million run rate business.
We've achieved that in terms of what we thought it could do, and that's actually growth happening now in that business. It's going to go again next year. So you've got that layering in, which is exciting. There is a basket of other products in there. You've got some DPUs, you've got some retimers, you've got gearboxes, those types of things. We have to size that exactly, but it would probably be the smallest of those other two, so a little bit smaller. And then really the balance of it is our optics business. Okay, so then the optics, there's really two pieces. There's the sort of inside data center piece, right? Which would be the optical DSPs, TIAs, and drivers. That serves both traditional cloud infrastructure applications, that's sort of a 200-400 gig PAM products.
And then you have 800 gig for AI, and then both of those kind of have a roadmap to basically double from there. And then within that as well, we have our, our cloud optics, which is DCI or data center interconnect, that connects data center. So, that's been the growth engine, that's the biggest portion of the total data center business, and, it's performing extremely well. It's the biggest portion of our AI revenues today, and from a year-over-year perspective, we expect it to grow again next year, even though it's been red hot, this year.
Yeah. Let's dive into that biggest portion of the bucket there.
Just one more, I think, you know, the other one is, which is becoming more relevant for custom silicon. The custom silicon, in the scheme of things, what is ramping only this year, right? So last year, that really was a big contributor to our data center revenue. We talked about that getting towards, call it, close to $400 million this year, right? And this is mostly in cloud applications, not AI. And then next year, you should expect to layer on top of that as we start ramping in AI, right? So that's the other piece, which historically wasn't a big part, which Matt can talk about it, historically in data center, but it's starting to become fairly important this year. Really becomes a much bigger contributor as we get into next year.
Yeah.
Yeah.
Yeah, I'm glad you teed that up. I figured there was going to be the whole discussion was going to be about that. So lay it out there. But yeah, 90% of your question-
Yeah, that's really where it's not a lot today, but, but certainly for next year, it's going to be, you know, significant more with talking about.
So why don't we start with the optics as we go into the custom AI tools? And I think, one, you know, you're clearly a part of a broader platform that's being deployed right now with NVIDIA, and you're clearly well positioned there. Just looking at the, that—what's been deployed in the market and the numbers in your optical business. But, you know, right now, it's so captive InfiniBand, it's so captive NVIDIA in a lot of these deployments. What changes if the world moves more standard Ethernet? Does that help you? And if you look at the market, do you think that, you know, your customers will push for a more open ecosystem to what they're facing today?
Yeah. That's a great question. We look at it a little bit differently. I think the way I would frame it is, and this goes all the way back. I mean, I remember when we were doing due diligence on Inphi, okay, in 2020, basically in the summer, the big theme of that sort of discussions was, hey, there's going to be this PAM transition from NRZ to PAM transition. There's going to be an upgrade in DCI, and that's sort of for traditional cloud. And then, oh, by the way, we're designed into every AI cluster in the world, and here's what they look like, but the volumes are fairly amazing. And that's sort of a, here's the product set that goes there. We've been working with those customers.
So from a deployment standpoint, Tom, I mean, that's the one partnership you mentioned is super important, but we're kind of broadly used within those clusters today across the market. And we're agnostic. I mean, the technology is agnostic, whether it is InfiniBand or Ethernet. We also service the higher layers of the Ethernet of the AI like a cluster to cluster, for example, connection would be also optically interconnected. That would use our products. So for us, it really, that whole debate doesn't really matter. It's going to drive a kind of optical connectivity one way or the other.
In general, when you kind of blend all of the different AI architectures out there, we generally have, on average, more than a 1-to-1 ratio of the AI processor itself to the optics. We're kind of okay. Our market share is still pretty high, so we're kind of all of them, and it's a little bit of a picks and shovels type of situation. You just supply into the market, you do a good job, and we're not too worried about that mix per se.
Okay, we officially made it halfway through before I mentioned the existence of, of the win, but-
Are we going to do any questions during, or will we do it at the end?
We're going to do at the end. So yeah, if you guys have any questions, we'll save the last two minutes to just get out.
Okay, perfect.
But just on the custom ASIC side, I think there's a lot of confusion in the market. You're guiding to some revenue growth sequentially into the first quarter. Can you talk about one, you announced some bigger programs. Can you just explain the difference? I think this question comes up a lot. When you look at your efforts versus some of the smaller Asian ASIC makers, why do you guys win? Are these different solutions you're going after? I mean, people compare the two of you. What would you bring up as, you know, why you guys are superior in a given design win at a cloud guy in particular?
Okay, sure. I think I'll break it to two pieces.
Yeah.
I think the first one is important to address and kind of clarify where we're at on this. So, we've articulated for some time, even if you look back to our 2021 Investor Day, that we had won a number of very important cloud-optimized silicon custom programs. That's all the way back, a couple of years back. As we, as we've moved forward since then, those designs have now, many of them, new product development, they've taped out, and kind of pre-production. So that's across a wide range of things. A few of them were targeted for AI applications, and there's two in particular for next year that are going to drive very strong revenue growth for Marvell.
And we, and tomorrow, I just claim that for everybody, we had sized back in the 2021 Investor Day, somewhere between FY 2025 and 2026, that being kind of the $800 million a year revenue contribution in that time frame. I'm not sure the exact order. We were a little off on this year in terms of when the timing happened, but it still looks good in that time frame. So that's all on track. What's happened within that is, from a couple of years back, more of it's gonna come from AI than we thought, and the numbers should certainly could be larger.
I mean, quite frankly, we just, you know, with the with how fast the Gen AI contribution to our revenue happened, I mean, we've been upping that number every quarter if you just look at our optics business. So there was a little confusion after the call. I just wanted to clarify it. Our ramp is very much on track for next year in this area. We're very excited about these programs. There's a hot minute of activity, and I'll get to your second question, in this area because of just the sheer amount of CapEx and kind of, quite frankly, disruptive opportunity this has got AI globally, right? Just to improve productivity and improve businesses. So there's a need and a desire for our customers to have a blend of merchant products that they're going to go buy.
In certain use cases, there are certain customers, and they're also going to have and probably need more of differentiated custom designs that they do themselves. So then you say, well, who's going to benefit from that, right? Who would win and who would lose, and who would do well? I would characterize us as one of a couple of companies that really operates at the very highest end of capability. And so what differentiates us versus our peers is, one, our best-in-class I/O, particularly on SerDes. This is something that Marvell was not really known for five or six years ago. Our introduction in five nanometer with our full five nanometer process technology platform and I/O offering was best in class. It is best in class, and that's gonna continue in three nanometer.
It's gonna continue beyond. So that's been very well received by our customers. The second is your ability to actually tape these things out first, because I'm right, could be upwards of 100 billion transistors. You know to pull them off. It's one thing to say, you can sort of do something. It's another to actually tape it out and have a path, either A zero or just a minor spin. And then you have all the other aspects of it, you know, sourcing the advanced packaging, the substrates, the glass, being able to yield it, and being able to deliver with the model. So you add all that up, it's quite a significant barrier to entry.
I think there's gonna be, to your question about the other peers at the other end of the spectrum, I think there's gonna be different business models, right? That ultimately are gonna play themselves out. I think some of that's gonna be desire for a certain business model, and does it, does it work or not? We're gonna have to see. But I just think, from our lens, you know, what it really takes to design one of these chips, the sheer human power required, the know-how, and also the capital and the capacity to go invest in this area. I look out, especially at three nanometer and beyond, I think the world is gonna need companies like Marvell and the select few group of people who can really afford to invest at this scale and actually deliver what's required.
So we'll see how it plays out, but we don't play at the low end. We don't do things that are in the commodity area, and if someone else can do it for cheaper or better, we can't add value, then we're probably not gonna pursue it. So some of those companies really aren't our competitor per se. That's not to say anything, that we're better and they're not. It's really just we're in a different segment.
Yeah.
So those, some of those folks you mentioned don't really show up when we're bidding per se on something.
Helpful. Very helpful.
Yeah.
Let's just switch sides away from this really good data center business that's become more of the focus.
Yeah.
Your other businesses are clearly still going through a cyclical downturn, and at some point, you're going to see that recovery. Could you just update us on when do you think that recovery may come, to the best of your abilities right now? Then I also wanted to give you the opportunity on the carrier side. There was obviously news out this week on AT&T switching from one provider to another. I know historically there have been some contractions out there. Any comments you guys have there?
Okay. Yeah, maybe I'll break them both out. So let's start with carrier. So that one, look, it's interesting to call the exact plan on the recovery. And one reason is we just had our best quarter ever in the history of the company. $317 million in revenue was up 17% year-over-year. Over the last probably 3+ years, you know, our wireless business is completely cranked, and we basically got it to the level or even more of where we indicated back when we bought Cavium, and we talked about the Marvell opportunities. That actually is a great success story for the company.
It's played out, and we started signaling. I think anyone that paid attention in the last two quarters, we were pretty, pretty, specific in including in our earnings call, directly saying, you know, Q4 we believed would be down significantly from Q3, just because we had visibility to when some of those NPIs were going to finish ramping. And also starting to see where that market probably was gonna take some level of pause. So that's played out, and that's gonna take a few quarters. It's carrier, it's lumpy, you know, it's, it's a little bit of anybody's guess, but high-level numbers, if you look, in calendar fiscal 2024, this current year, if you take our Q4 guidance, it's about $1.1 billion we'll do in carrier. It was about $1 billion or so the year before.
And I just think mental model, you know, through cycle, get through some of the inventory stuff and correcting, that's probably not a bad place for it to return to, right? And then you'll get a little growth on... because we've got some additional content gains coming in in the wireless area, as well as in the wired area, where we have a nice optical DSP franchise and coherent DSPs for long haul and metro that we acquired from Inphi. So, you know, we keep pretty substantial market share in this area. You know, it's in decent shape, but it's gonna kinda grow at market or kinda plus once it recovers.
With respect to the announcements on Nokia, I think kinda a high-level way to think about it is, they've called it out as kinda being, you know, mid-single digit, you know, 5%, maybe 8%, so something in that range % of their business, that over the next 2-3 years was gonna trend down. And I would say for us, you could also think of it as kind of a single-digit type % customers for us over whenever they sort of ramp back up. Just think of it as a normalized run rate. So you kinda take a single digit times a single digit, you do the math, it's not a huge number, and we also have content at other companies as well, so that's a little offset, too. So we view this as very, very much not an issue at all, actually.
I mean, and we'll have to see how it works out, market share stuff, but single digit times single digit equals not much.
Can I offer something on the other side?
Yeah.
I think that some people may not think about this as well, but obviously, if you're looking at networks that are more centralized, you need more interconnect. Do you see yourself playing in that area of the world? And could you potentially offset some of what you're seeing there with more impact on the backhaul side or even within a data center, like?
Well, I think if you look at carrier overall, that scenario, it seems like on the wire side, it has to play out at some point.
Mm-hmm.
You know, I know CapEx is tight, and everybody's sort of managing around that, and that makes sense. But at some point, and I think because of all the Gen AI stuff that's happening, plus this continued movement to, you know, cloud-based computing versus on-prem, but it's just gonna have to drive some upgrade at some point for sure.
For sure.
So that's why we're not on a specific customer announcement, you just kinda times those two. It's not a big number, but that's why over the long term, I mean, carrier's actually a good market to be in. It's just we've always said it's got a high beta.
Yeah.
And part of that was why we did what we did when we really had a head of steam on us in 5G, is we were often also diversified into the cloud, which was really the end purchase, Innovium, and then plus the Avera stuff. So we've always tried to have diversification in mind, Tom. And I think there's one case where it's serving us well. 5G, by the way, you know, earlier this year, when data center was down for everybody, everybody corrected hard, right, about a year ago in data center and cloud, and we had carrier was kinda carrying the day at that point. So, I think, I think, I think it's going to be a fine market.
A quick one before we go to questions. With the mix of business that you're describing for the next couple of quarters, you've always talked about the optical business being a gross margin, accretive business for the entire company.
Right.
As you see that business ramp and some of these other businesses not recover, what's holding back the margin profile from continuing to accelerate? Not saying that it hasn't already. I think everyone did a great job, and you called them out on the call, saying how quickly you guys got to the range. Can you just talk about where you can go from here with that mix continuing to kinda go in that direction?
Sure, yeah, and I think, you know, there's a lot of moving pieces. Mix is obviously, you know, the biggest part of it. So you're seeing some of the reason why our fourth quarter is going up like 300 basis points, right? And it's a lot fourth quarter.
Mm-hmm.
But we had signaled that normally, nominally, because we believed even, you know, a few quarters back, that the mix would go in our favor. Now, it probably didn't go in our favor as much as we thought, but we also did a lot of self-help.
Yeah.
A lot of self-help. Our operations team did an excellent job, kinda taking the opportunity with the ramp in other parts of our business to really improve our cost structure, both from a supplier negotiation and partnership, but we've also reduced our MOH, and we've really streamlined our operations around the sizing of it. So we took a lot of action there to get it to the guide we gave, and I think the way to think about it over the next year is it will still be very mix dependent. And while we have a great head of steam on it, parts of our business, like the optics area, we also have custom programs coming in, and we've got other, we've got other mix dynamics.
So I think the good news is we did a lot on our own that sort of wasn't baked in, but I continue to just sort of comment that we will be... It really depends for next year, honestly, how big this custom business gets. Now, the good thing about that is you can look at it as a negative. Well, geez, the custom is bigger, it grows at a faster rate than the rest of the higher margin. That would be bad for Marvell, potentially. Well, it really wouldn't, because the fall through would be significant, right? We actually want more, we want more revenue next year because all the development effort for these chips is largely behind us, you know?
Yeah.
So now, of course, we'll work on the next generation. Hopefully, we can even expand our footprint in this area, but, you know, we'll just have to manage it. But look, more revenue is a good thing, and I think no one will be, in this room will be upset if custom does phenomenally well and grows faster than the rest of the business, say, optics. I think it'll be a win for everybody.
Understood. I want to be fair here in terms of giving the audience a chance to ask any questions. I had seen one earlier.
Yeah.
Do we have a microphone back there that we could just grab? Thank you.
Hi. I actually had two questions around your SerDes and your acquisitions of Innovium. They used to use, I believe, Cadence's SerDes, and that caused a lot of problems. How is owning them and owning your SerDes help the business, one? And then number two, is you mentioned your DSPs, and, you know, Credo's built a good, healthy $200 million a year business.
... with that, and your DSPs are arguably better. So how do you feel about, you know, having both the SerDes to increase your switching business with Innovium, and then bundling that with the DSP that you can, you know, add to optics or optical AEC cable system, et cetera, and being able to generate maybe like $20-$20 like Credo can?
Yeah, yeah, great questions. On Innovium, when we acquired it, and even during diligence, they had been working on their next generation product, which was at 25.6 terabits per production on 12A. We made the decision based on market reasons to kill that program, and we basically said: Look, we got to double down on 51.2. So if we try to do this serially, it's just gonna take forever, and we'll do that later. And we didn't believe that particular node was gonna be as big a node, and transformation was 400G. So we did that, and the way we designed that product, and by the way, you know, it's looking great. Product development came out of the fab, you know, I don't know, 4 or 5 months ago, with several customers.
It looks great. That was done with Marvell SerDes. That's on the Marvell 5-nanometer technology platform. So we effectively went back and replanned everything, our technology flow, but it was still the same Innovium architecture team as possible. And I'm so glad that we did that because we're not relying on third-party SerDes for this product. We're controlling our destiny there. And then we got a lot of know-how, too, because we had the Inphi team in place as well. So that's a great first start. We want to be even more competitive at the next generation in switching, but just so you know, the first product coming out the door is actually Marvell SerDes, and it looks fantastic. On the AEC side, we were very pleased at the earnings call. We actually had it in the script.
It may have gotten lost a little bit, but we did say we have now design wins confirmed in this area, and they're going to be ramping into production next calendar year. And we do see a big opportunity for AECs, and I think it's gonna be, you know, gonna be... It's a real market that's gonna develop over time. We like our position there because as the, as the next generation AECs are coming out, they are kind of going through the same thing the optical modules went through, which is the old ones that are kind of in production now are based on NRZ modulation. They're all moving to PAM, and so the designs we have are all PAM based that are gonna, gonna ramp up next year.
And then, of course, it's gonna go through the same technology treadmill there in terms of wanting to improve speed, capability, and throughput. We're in great shape. That market is gonna go, is going through an NRZ to PAM, with very strong position.
Yeah, maybe just two quick additional comments. One is, yeah, what we are launching now is a 100 gig per lane product, but as Matt mentioned, this market will keep accelerating, and we've already actually demoed at OCP, a 200 gig per lane electrical product, right?
Yes.
Which will end up in the AEC market. So that's a very key point in terms of our ability to lead the market. And the second is the go-to-market model on your revenue comment. Just, just to be clear, I think, you know, some of the existing players sell entire cables, right? Well, obviously, the margin would technically be lower, higher revenue. Our go-to-market is like an optical market. We actually enable resale DSPs, and we work with multiple cable companies, right? So from a revenue perspective, of course, our margin is going to be a lot higher, but the revenue, we're not going to collect at the, at the cable level. So just kind of keep that in mind as you kind of look at those revenues.
Yeah, that's a great clarification.
Yeah.
I think the two companies have taken different approaches. I think both are valid, and customers are gonna want to have flexibility on what to choose. But we've sort of
Look at what the price is.
Well, I think, yeah. I mean, we'll see. You know, I think the economics should be pretty good for the customer because we partnered with all the Tier one cable suppliers, right? Who are all very aggressive, and they're going to go pursue that business pretty hard, and they have all the scale, the background in this area. And so together, I think we're very competitive in terms of what these customers want.
I think we're over already, guys.
We are?
Thank you so much for being here.
Okay.
Have a great rest of the week.
All right. Great. Thanks, everybody.
Thank you.