Marvell Technology, Inc. (MRVL) Earnings Call Transcript & Summary

Thanks, everyone, for being here. I'm Tom O'Malley, semi and semi-cap analyst here at Barclays. We have Willem Meintjes and Ashish Saran with Marvell. Thank you very much for joining us. I appreciate it.

Yes. You're welcome.

You're welcome.

Why don't we kick off with kind of the topic du jour, the custom silicon business? Can you just describe the distinct advantage that you clearly have over some of your competitors, both on the back end side in Taiwan? And then also the differentiation that you have versus some of your customers here in the U.S.?

Yes, sure. Yes. Thanks for that. Yes. So maybe the place to start is just if we look back, right, we really started with this business back in 2021, right? If you remember that, yesterday, really, the first time we announced that we had these design wins in 5-nanometer. And if you fast forward to today, we're actually shipping in production with all 4 of the hyperscalers at this point, right? And so really, I think a good way to look at it is just sort of tops down for a second, and we can talk about the technology. But if you look at the premier sort of sockets across the landscape around custom, we're shipping the accelerator, right, with what we call Customer A. We have a custom CPU, ARM CPU that we're shipping in production with Customer B. And then we announced Customer C, which we've said a couple of times now that we expect to be our largest opportunity. And really, I think that is a reflection of the fact that our technology is as good as anything out there, right? And then on top of that, there's now these very clear proof points that we've gone [indiscernible] silicon to production, which is extremely important when the technology cycle is compressed like this. Maybe you can talk a little bit about the...

Yes, I think the -- I mean the other one, as you've probably seen publicly now is iPad. I think we've mentioned back at our AI Day, not that far back, that we look at all of these as multigenerational because it's really a continuous technology. I think our networking capabilities are almost as important as our ability to drive these very high complexity silicon. And with one of our largest customers now you've seen basically a pressure, which is very unusual, by the way, which really talks about having this agreement across the next 5 years, almost to the end of the decade, which clearly covers Custom AI Silicon. In fact, that's the majority of the agreement revenue, is really tied to custom AI silicon. So it's not even about the next generation. It's really about multiple generation of technology. I think you saw we had an Industry Analyst Day yesterday, where we talked about additional innovation on the Custom Silicon platform. Even interfaces like memory, which for years have generally been standard, we're optimizing those interfaces as well. And we've been working with the entire ecosystem, including 3 of the large memory-makers, and there was a press release out yesterday. So I think it's the full platform. I think it's the big differentiation, quite frankly.

Yes. And this sort of trend towards technology moving faster, we think that really plays to our advantage really on 2 things, right? One is we're investing in sort of not just the next generation, but even the generation after that technology being sort of N3 going to N2. But then also, I think the opportunities in the landscape is multiplying because of this faster cadence that you're seeing, and the fact that ultimately, I think, design resources are constrained and having a partner that really amplifies the amount of silicon that you can go address, we see as really playing to our strengths.

Helpful. So if I look at the customer announcement, obviously, Amazon is public now, do you have a wide range of technologies that you're servicing them with? You mentioned custom silicon is the largest, but you have AEC, there's some PCA products as well. So can you talk about why was that included there? Was that a direct effort from them to wave the flag and be, okay, Marvell is able to offer all of these products? I've never really seen a larger agreement with those other items listed in it. So it stood out to me. Why do you think it was important for them to put that in there? And is that just them trying to say, hey, we have a second source? Or is it them really putting their foot down and being like this is something that we don't leave it?

Yes. I think the first thing to say is this is -- it is very significant, right? I think you really haven't seen something like that. And I think your take there is exactly correct, right? Now when you look at it, I think what it acknowledges is that when you're looking at accelerated compute, the problem really isn't just a compute problem. It's a system-level problem, and there's very few companies that can actually enable from a design standpoint, end-to-end, having a perspective around not just compute but how do you network all of that together on the front end, on the back end, scale up, scale out. And that portfolio of technologies is directly linked to that, right, where it's like, hey, we're going to compete long term versus the merchant offerings out there. You have to have a full stack view on how everything is connecting. And I think that agreement is really a reflection of that.

Yes. I think the new areas you mentioned, AEC, CXL, PCIe, all of these are moving into an era where there were niche technologies in the past, where they were basically used on lower speeds. Typically, the older modulation scheme called NRZ, all of these are moving into PAM4. And PAM4 basically is the basis of our kind of market-leading optical DSP business. So it's not a surprise that we would be in that market. Our customers have relied on our PAM4-based DSPs, right, for building out their large hyperscale data centers. And now the very same technology is entering essentially these AI clusters as well. So you should expect that. Obviously, it's great to see it's in a public press release with this particular customer. But the reality is this is what's happening with all our customers. It's actually very, very broad.

Helpful. And you're not sizing the Amazon opportunity, but there is contra revenue associated with it. You have other players in this market that actually have contra revenue, you can map out how much that customer is over time. Any comment on the size of that deal? And if not, the real question is, when you talk about the third customer and the opportunity there being the biggest of the 3, in terms of size of hyperscaler, that guy is kind of the biggest already. So what gives you the confidence that, that opportunity is going to be bigger than the one that is currently ramping today?

Yes. I would just say -- I would just take you to what the full opportunity looks like in the market, right, which we outlined for all of you guys, and it's very, very large out in time. So we identified a total data center TAM, which Marvell can address, of about $75 billion out in calendar '28, $40 billion of that essentially is in Custom Silicon, right? And our revenue, if you remember, at AI Day, we had indicated this year, the year -- the fiscal year we're in, we thought we'd get to something $0.5 billion of custom silicon revenue in AI. Now the reality is we're running well ahead of that. But that just gives you an idea of we're, call it, $0.5 billion-plus of customer AI revenue this year, and the TAM is $40 billion by calendar '28, right? And our goal was to get to about 20% share of that number. So from, call it, something sub-billion dollars today to about $8 billion in the space of, call it, the next 4-plus years, I think, that's the goal we are looking for. And these customers are very much a big part of that. The opportunity is very large with each one of them.

Helpful. And I think that after earnings call, Matt mentioned, we're tracking a couple of hundred million ahead this year. He said, think about it flowing through to a couple of million ahead next year. So everyone loves to run, do the math, $1.5 million to $2.5 million. If you're running a couple of million ahead, why don't you update the AI number? Was that just a decision you don't want to get caught up in the AI number? Or is that just since things are good we're going to let it go?

Yes. The market is great. It's still very dynamic. I mean, look, we talked to you back in April, which is 1 fab cycle away from where we are today, just to put in perspective, right? And there are several hundred million dollars ahead of that target within essentially 1 fab cycle. Trying to see -- we know we'll be ahead next year. How much ahead? Well, I don't think we want to cap the upside opportunity. I'll leave it at that.

Super helpful. When you think about the next generation of this custom silicon moving to 3-nanometer in kind of the out year, there's often debate about, hey, will you see a split roadmap between leading ASIC guy and potentially like a back-end ASIC player? You answered this question a lot, but I think it's useful to kind of describe, why don't you see some of your customers using a split road map where you would go back to a customer that you used historically. What advantages, I asked kind of in the first question? What differentiates you? But like why wouldn't you see a customer in that instance move to multiple sources?

Yes. I think we try to up level that, quite frankly, and solve that for a lot of you by very publicly saying that this is a 5-year multigenerational because look, you know the cadence of these technologies now. So it's one every 2-plus years essentially, right? If you have a 5-year window, you basically -- it's not just about the next generation, it's really about what do you do for the next 5-plus years. Given how fast the technology has evolved and given that getting information on and off, the die is almost as important as the compute die itself. You can see why customers want to work on a long-term basis, right? It's who do I work with for the entire signal chain, all the way from the compute die all the way to the front end of my network. And that's where you see we've got a very large proven merchant business, which we can bring to the table and help them optimize every part of that data chain. I think that's the way you should look at it. Now, are there going to be opportunities for other pieces of silicon? Of course. I mean, I can't talk for my customers, but I can tell you what we see, which is we see -- and you've seen that in our numbers and our guidance and the forecast we gave you, back to the TAM, we see this as one of the largest opportunities in front of us. And we have already got agreements and design wins in place to get us to that target.

Yes. Maybe just to add, one angle is if you look at our internal investment, right, on our merchant portfolio, we're investing in these IPs to enable a whole net booking portfolio, right? And so that's the core reason why we have this IP, and we're able to go and drive that through that portfolio and really productize and get it to be very optimized and high yielding. And so that is the core of the IP that we've been able to go and use to go address on the custom side, right? And so we're not developing this IP sort of vice versa, right? Whereas if you look at the broader market, and if you say, hey, we're going to go buy this IP from a vendor who's just, for the first time, brought this IP out, assuming there's those models out there, but it's not productized, right? And it's not hardened in multiple chips and multiple SKUs, where we're running super high volume on. And that's really what you saw a lot in sort of the previous generation. And really, there's not a lot of results where you can point to and say, hey, this was a huge success using that model, right? And so I think when you look at some of the most successful programs out there, it was where you're using a vendor that has that internal IP, right? And I think the ecosystem is recognizing that. And clearly, the design wins that we've highlighted is very aligned with that.

And I think it's a perfect segue. I think you've talked historically about the investment that you share across your different technology portfolios and how, obviously, there is a difference between the gross margin profile of the custom silicon and then the flow-through to the operating margins. So you kind of gave some color about the gross margins going forward. Do you think that -- obviously, you're leaving the door open to upside on the custom silicon side, how are you flexing that in terms of your gross margin? Is that kind of already accounted for in that low 60s gross margin number, the potential upsiding of custom silicon? Or is that -- would you think additionally maybe some gross margin pressure?

Yes. Maybe I'll sort of talk about it a little and then kind of get back to your question directly. I think if you look at how we scope these programs, right, we sort of outlined this gross margin for a custom program, call it, 50 plus/minus, right? But setting that aside, all of our programs are designed to hit this 40% -- 38% to 40% operating margin target that we have as a company. So whether we're shipping a $1 of custom or a $1 of merchant, it really doesn't matter, ultimately in terms of the flow-through to EPS at the bottom line. And I think investors have been recognizing that, right? And so when you look at the custom model, when you double click, well, why is it different? Well, first of all, the customer is investing alongside of us, and we recognize that, as we call it, nonrecurring engineering, or NRE, where it gets recognized as a contra OpEx. And so our actual outlay and the fact that we didn't have a captured customer there, means that the risk profile is obviously a lot lower, right, versus if you're just doing a merchant, you're taking all the risk where, in this case, the customer is obviously taking a ton of risk because they're making that investment, right? And so from a gross margin standpoint that it's just never going to look the same when you have that certainty, right? And so we've been very consistent on that model and believe that, that remains the case. So the faster we ramp customer, the faster we get to our long-term operating model, right? And so to your point, I think how we scope next year really is how we see it today. If it upsides significantly, yes, certainly, it can go -- the gross margin can go lower, but that would be a great thing, right? That's a good problem to have.

Super.

Yes.

How about transitioning to the optical side, which has really been the growth vector leading to this custom silicon uptick? So you announced a new product right with earnings, the 1.6T [indiscernible]. And you've seen continued leadership at every speed transition for Marvell. Can you talk about when you go into '25 there clearly is more of a battle now outside of what you would historically define as traditional leading-edge optics? There are LRO, LPO to a certain extent, there's a lot of noise. How can Marvell remain kind of that leader in this optical market with the different flavors in profiles kind of coming to market?

Sure. Yes. So I think first, as we saw -- this has been a high-growth business for several years. We saw tremendous growth really starting when AI kicked off, really, and that's over -- been well over a year at this point in time. The business continues to grow very strongly. Just even in Q3, we said it grew double digits sequentially off a very, very strong base. It's going to grow even faster in Q4, just to give you some context. And we see very strong growth for next year. In fact, as you can think from a lead time perspective, a number of -- the first half of next year is already kind of coming in from very strong bookings. So I think to your question on 2025, I think at this point, we're already seeing some very strong trends for next year. And I think our optical business, we're going to basically be talking more about interconnect, which is broader because to the question earlier, even AECs are starting to come into that market, right? So I think our interconnect franchise remains very strong. We've been ahead. The reason we've had such a strong position is because we've had a leading product, right? That's the reason we've had this market position. We introduced our 5-nanometer 1.6T part over a year back. It's now in production. As far as I'm aware, the only ones in production. And as we're a market leader, this technology cadence keeps going up. Even though we just started shipping our 5, we've already introduced our 3-nanometer product. And I think that's the basis of why we expect we will remain basically a market leader in this space. And this is really talking about the PAM DSP inside data centers, and I'll get to your question on alternative technology in a second. But let's not forget, we also have a very, very strong market position within DCI, which connects data centers together. The demand on that continues to grow, especially as you think about larger training clusters, you can't get enough power in a single location, you will see more data centers come up. So we see a very strong trend in our DCI business as well. AEC is, of course, a brand new market for us, multiple design wins. Our go-to market is shipping through our cable partners. We started shipping AECs in Q1 of this year. It's ramping very nicely. I think it really starts to hit strong volumes next year at multiple hyperscalers. In terms of -- it's actually very exciting. In some ways, when you hear about more competition, this is to me, capitalism 101. It means the market is getting a lot bigger than what people anticipate. And that's exactly what's happening because of AI. And yes, there are multiple paths people are looking at to solve some of the next-generation problems, right? So I think it's about traditional networks, which tend to be heterogeneous, meaning there's different equipment from different eras. I think our customers have stood up at OFC and said, look, pluggable DSPs are probably the primary path. Having said that, if you look at some of the other places, think particularly about what's called the compute fabric, some people call it essentially the scale-up network. That's all copper today, right? It's copper on a motherboard or it's copper on what's called a [indiscernible] cable. It's completely passive today. But again, as bandwidth keeps going up, you can see why that would be a natural place where you would want to go down, either active electrical as an intermediate step, or at some point, you may have to go optics all the way. Now the density of this connection is actually higher, significantly higher than what you see in the traditional networks today. And that's why people are looking at, look, can we find an alternate way which is lower power, the business tend to be shorter? And that's where I think you will see some technologies like LPO as an example, which by the way, we also announced the products for yesterday. So your question timing is very nice. I appreciate it. We're also seeing this is where in the longer run, you may want to start putting optics a lot closer, right? So there are things like near-package optics, co-package optics, need a silicon photonics platform to do that, which happens to be the basis of our DCI business. We've been shipping this thing in high volume for a very long time frame. We bought our SiPho package that we call it our light engine at OFC earlier this year in April. That is going to be the basis of what we do in this particular market. So I look at actually these new alternative technologies as actually very exciting. It's opening up a new TAM. As long as that compute fabric stayed copper, that was not a place Marvell would historically play in. But if that starts to move into the space of either active electrical or, at some point, optical, it's a massive opportunity, I think, for a number of us, Marvell especially included.

Yes. When you look at this, right, and this goes back to when you first started seeing the whole accelerated compute or GPU architecture, it became clear that the networking piece is the key part of the puzzle, right? And you can go look at sort of some of the acquisitions that we've done to enable that. But when we look at our portfolio, we're investing aggressively sort of in these next-generation technologies, right? Because what you see with the Inphi success, the key there is having differentiated technology and really leading the industry. And so we're super committed to continue that because when you look at, you can see how the networking opportunity just becomes a critical enabler here where you can solve the accelerator or compute problem, and without the networking piece, it really doesn't go anywhere, so yes.

Yes, it's a full-service solution. And by the way, customers don't know for a fact, nobody really knows, right, what some of these new technologies will actually enable. So what's important is to have all of them available. So your partner can develop actually parallel paths with you because at this point in time, we don't know what's going to succeed. But it's okay. We can do both for you. And as we hit a fork in the road 3, 4, 5 years from now, we can enable whichever path makes sense.

Super helpful. All right, transitioning away from the data center side to some of the other businesses. So clearly, a recovery path, which has begun across really all of them to different extents. But the last quarter, you took a write-down and talked about some -- it was 2/3 IP, 1/3 cash assets that you were writing down. Could you talk about -- I think the question that I've gotten most often afterwards is, is that going to impair that growth profile of the businesses coming back? Can you maybe get a little more specific where that cash cut actually occur?

Sure. Yes. Yes. So first of all, the answer is we do not expect that to impact that growth profile. We still see it sort of returning to sort of a total of $2 billion between enterprise and carrier. The way I would think about it, right, is that those markets have really become more mature and the technology cycle there has just slowed down, where we invested heavily to build a portfolio to go address that. And so we just brought out our 5-nanometer products on carrier. And so when we look at those, we expect those to have a very long life cycle. And so what we've done is really look at the data center opportunities and really taken our resources and just become way more selective and targeted where we're investing on carrier and enterprise and then enable our investment across the rest of the portfolio, while we're still driving this very significant leverage in the operating model. And so part of that question really comes back to, hey, when you look out across next year, we're aggressively driving to our long-term target. So we're going to make a ton of progress next year. I mean clearly, we're not saying all the way. But I think that sort of internal discipline where we're allocating resources is really enabling that. And so yes, I still feel very good about the portfolio, and we're just going to be very targeted on where we invest.

Yes, just to kind of put some numbers around it to help investors. Yes, I think that carrier and enterprise together are going to be clearly well under $1 billion this year, right, even if it starts recovering. I think it's -- at this point, it's a pure inventory game, right? We aren't going to build a new product, get a new customer, it's a pure inventory game, where our customers have significant inventory, which has started to come down. It's going to take several quarters, but think of this as a very nice multi-quarter tailwind, you don't have to do nothing for, it's just revenue coming in. And in the meantime, of course, our data center business remains on fire. So it's not a bad outcome quite frankly.

One more really quickly because we started a teeny bit late. Matt was very detailed on the call about talking about share-based comp coming down as well as GAAP profitability. Is there a reason why he went out of his way to call that out on the markets...

I think 2 things, right? Clearly, when you look historically, we were on a growth path, and we had sized our sort of stock-based comp for that. And over the last year, 2 years, we've been on a slightly different trajectory. And so [indiscernible], we acknowledge it as a percentage of revenue, got higher than we would have wanted, right? And so I think with this Q4 guide, we've really gotten that back on track. And then you're going to continue to see that leverage next year, where we'll be right in line with kind of what the peer group looks like. I think GAAP profitability, it's a key focus for us. I think, clearly, that enables us to get included in some groups somewhere in the future. So you should expect us to be very, very focused on that over the next year, yes.

Well, thank you both for joining. Really appreciate it.

Yes. Great. Thanks, Tom. Appreciate it.

Thanks, Tom.