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

Welcome, everyone to this session. We're really delighted to have the team from Marvell, Willem Meintjes, the CFO; and Ashish Saran, Head of Investor Relations. I'm Vivek Arya from BofA Securities semiconductor team, and like in most sessions, I'll go through a list of my prepared questions. But if you have anything you'd like to bring up, please feel free to raise your hand, and a mic will be given and you're more than welcome to ask your question. But with that, very warm welcome to you, Willem and Ashish. Really appreciate you could join us at our conference.

So maybe let's get things started at the top. Marvell recently reported earnings, right? Maybe, Willem, just give us kind of a state of the union. We have seen this sort of, as people describe it, this tale of 2 cities: AI is awesome; legacy, not as much. So how are these crosscurrents playing out for Marvell now? And then how do you see them kind of trend through the rest of the year?

Sure. Maybe just baselining on the actual quarter. We guided $1.15 billion and reported just slightly above $1.16 billion. And that beat was really driven by data center and more particularly, optics. When we look at some of those other markets in enterprise and carrier, they were actually slightly below what we had contemplated in the guide and more than offset the strength in data center and optics. So we continue to see strength in data center and optics. And then looking ahead to the guide, we guided slightly above where expectations were, and again, that was really driven by the ramp in our custom program that's starting, continued healthy demand for optics, and then a snapback in consumer where we had worked with our big customer there to really get through the inventory digestion. It's sort of well understood in the gaming environment. But the bottom line is that more broadly, we indicated for the year, we spoke about $1.5 billion of AI revenue for the full year. And that was only a couple of months ago, and we've seen progress and actually expect that to be higher both on custom and on optics for the full year.

Yes. Maybe just to add a couple more things, just kind of to your comment on forward looking. I mean when we look at the growth we guided for Q2, it was, call it 8% revenue growth at the midpoint. We see that accelerating significantly in the back half of the year, both in Q3 and Q4 on a sequential basis. So we see a lot more growth, again, primarily data center, but we do expect to see some recovery start in some of the businesses, enterprise and carrier as well. And as well as our automotive business, right? And as we go through that work process, you will see us drive very strong operating leverage. You'll see some pretty nice expansion, op margins as well as on EPS.

Excellent. So I'm sure we will come to everyone's favorite topic of AI soon enough. But let's maybe get the other segments discussed first. So enterprise plus carrier plus auto industrial, right? I think you kind of guided them flattish going into Q2. What are the puts and takes in these different segments? And where do you see the recovery sort of happening first and then others may be staying a little muted until next year?

Yes. I'll start and then Ashish, you can jump in. But maybe let's just start with carrier. I think the end market clearly is going through a very significant correction on top of an inventory digestion period. So counterintuitively, what we're seeing there is that we've actually got a new design that we've actually started seeing orders for that we expect this ramp really in Q4. And so that gives us more confidence in that end market in terms of the first off being the bottom and having visibility into the second half. The second one is enterprise. And when we look at the commentary in the broader ecosystem, I think very encouraged by commentary where the end customer demand has started to return and that there's inventory being worked on. Our customers are still sitting on quite a bit of inventory. And so we're being very conservative in terms of calling that recovery. And so clearly have seen the bottom being in the first half, and that sort of informed our guide, and then see a modest recovery. On auto industrial, really driven by automotive. I think our design wins initially was really towards the EV side, and we've seen that -- a very, very healthy ramp there. But I think it's well understood that EV went through a little bit of a pause. But as we look into the back half, we see that growth reaccelerating and the more broader design wins that we have even in the ICE vehicles driving a nice setup for us to start growing again in the second half.

Yes. I think it just comes down to the inventory, right? I think the demand signals are positive now, and that's what we've been waiting for. Because the last few quarters, the demand has been coming down. So that's stable and starting to grow, not just a question of inventory. I mean, a perfect example is the consumer market, where it was consigned to one customer, you clean it up and the demand, the revenue snapped back, yes. So I think we just have to wait and see how that happens in these other markets. But it will. I mean, at some point, you're going to see switches and routers and base stations start to ship again at meaningful levels, and we will see that recovery in our revenue as well.

I think one thing that's just pretty important to state is that we haven't seen any change in our market share or anything like that. I think we have good visibility and there hasn't been any significant structural changes. It's really an inventory digestion period.

In fact, if anything, in our wireless business, our market share is going to go up with the 5G products. As you know, it's not just replacing with 5-nanometer our current generation of product, which obviously is good, but we have a very large socket replacing an alternative solution, which starts ramping this year and becomes a lot more meaningful next year. So market share is actually going to be higher going forward.

And just one last thing on enterprise. So not a surprise, right, that some of your large customers still have a lot of inventory on their balance sheet. Are there any Marvell specific drivers? Any specific product catalysts that you think can help you get out of this? Because where I'm coming at this is I think Matt had suggested that a normal carrier, normal enterprise looks like a $1 billion-type segment. So when should investors start to kind of broadly -- if I just put them together, like $2 billion, when should we look forward to that sort of zip code?

Yes. No, I think we outlined the design win on carrier. If you look at enterprise over the last couple of years, we believe we've taken quite a good amount of market share. But looking forward, I think maybe slightly grows faster than the market, but fairly in line with that. I mean I think that your specific question around when do we see that come back to the $2 billion run rate, we're being a little bit conservative, right? And so we're saying probably at some point next year, you can start approaching that. But for the full year next year, certainly, we don't see that right now.

Yes. I think on a run rate basis, possibly. I think it's really come down to inventory reduction. I think once you start to see that in -- and by the way, the inventories are never going to go back to pre-pandemic. That was a different world, just in time. No one is going to do that anymore. I think you just need to see -- and by the way, the inventories have come down. I think we had another large enterprise customer report overnight yesterday, right? And I think they will -- again, similar to the other large customers, they were very positive. I know what they're seeing out here. So I think the moment you see the inventory start to move down meaningfully, and that's obviously tough for us to guide whether it's one quarter, but it's probably not 6 or 7 quarters. It's probably in the next few quarters. And I think you'll start to see, as you've probably seen, Vivek, in your past cycles, when things go down, your instinct is to moderate when they'll come back. But what actually happens is at some point, the inventory is normal and then I think this snaps back. It's just tough to model that. But that's probably what's going to happen. We just -- it's tough to call it sitting kind of where we are.

Got it. Absolutely. Well, the good or bad thing is that those 2 segments keep on becoming a smaller part of a sale, so maybe they have less of an effect than they used to.

Not by design, but yes.

So let's talk about the data center business. Maybe just double click on that and help us walk through what is the AI portion, what is the non-AI portion? And then what are the trends in each? And then we can go through the specific AI dynamics.

Yes. Maybe I'll start and Ashish, you can jump in. So our AI motion is really -- there's the connectivity piece, and that's mostly optics today, both DCI and PAM, and driven by the 800 gig PAM today. And then we've got custom that we're really in the very early stages of ramping. And so what we outlined at AI day is in data center, about $1.5 billion this year relating to AI. And what we said is both -- we said about 2/3 was optics, so call it $1 billion and $500 million on custom. And we actually see both of those being stronger today, even just a couple of months. And we were pretty clear about that being the floor, the $1.5 billion, and that we actually expected that to continue to go up, and that's what we've seen. On the non-AI side, it's a broad range of products, includes switch on data storage, our fiber channel product, fully broad range of product. And so for the data center infrastructure portion, we've really seen that bottom as well and then we've seen growth continue. On-prem has actually been fairly challenged. And even in Q1, I believe it was actually...

Yes. I think enterprise on-premise. I mean that's -- by the way, it's the smallest part of the data center now. It think cloud, which includes AI, would be the vast majority. I'm talking something probably north of 80% of revenue. Enterprise on-prem is smaller and has the same trends we talked about enterprise networking. For the reasons we all understand, enterprise is still on the suppressed side. That has also bottomed in Q1 as far as we can tell, and you'll start to see a recovery there as well. But to Willem's point, I mean, AI clearly is, and it should be, and it's growing fantastically for us. I think this is a reference. Last year, it was 10% of company revenue, like something north of $0.5 billion. And with the $1.5 billion floor, probably do better. If we just look at Street estimates, that assumes AI is probably somewhere around 30% of the company, right? So you're basically tripling over this 1-year period. But what's also been very nice is that the standard cloud infrastructure, which clearly had gone through an inventory correction in the first half of last year, you're seeing renewed investment there as well. And we've seen that in our optics business outside of AI. We've seen this in our switching business. Storage has been coming back, and it sounds even more positive, and that's feeding into our second half kind of growth assumptions as well. So I think it's been very nice to see standard cloud investment also start to grow again, and we have certainly seen the benefits from that.

Got it. I find Marvell's optics business fascinating just because there is such a strong growth rate and correlation with just the build-out of accelerators. And we just see that these cluster sizes grow exponentially. What used to be 20,000, 30,000, now NVIDIA's saying 100,000. I think you have friends at [indiscernible] at even 1 million at some point. I imagine that has very positive implications. So talk to us about how you're seeing the attach rate of optical transceivers, right, today? And where do you see it going as the size of these clusters [indiscernible] to grow?

Yes. So I think where we see -- I mean, this impacts both our DCI business. I'll talk about that a little bit separately, but the biggest impact is on the interconnect inside the data center. And there's 2 levels where you will see this impact. The first and foremost, of course, is you've got a connection going from the GPU cluster or the accelerator cluster to the first level of switches. That is not -- doesn't change with the size of the cluster itself, right? What changes is as you go to these much bigger, overall cluster sizes, right, we go from 30,000 to 100,000 or 1 million, the number of switch layers expands dramatically, right, because you have to connect these multiple clusters together. And each switch layer means you add another entire new set of connections, right? So instead of 2x to 3x ratio between a DSP versus an accelerator, as you go to these higher cluster sizes, that ratio can go significantly higher. And we are seeing that, by the way, right? So we're seeing that impact. The other impact you're also seeing again in a positive area is that the connections within these clusters inside them historically have all been passive copper essentially, right? The market leader is still staying on passive copper for at least one more generation, but a number of the other hyperscalers with their custom solutions, and some of them we are part of. So we are very intimately aware of the architecture. Those are switching from passive copper to an active connection, either in AEC, which we are also participating in now, and that becomes a much bigger market next year. And some of them are also looking at going optical because of the size of even the connections between the accelerator inside the cluster. So we're certainly seeing that, and I think that's what gives us confidence that we'll probably keep outgrowing the market in terms of overall optical attach. The other consequence is as you build these very large data centers and you need to connect them to each other, we're seeing a lot of growth even in our DCI business, data center interconnect, where historically, there was one large customer who kind of created the business with us. But today, it's spreading to pretty much every hyperscaler. So we've actually won designs now with pretty much most of that community, and we've been shipping our 400-gig product in high volume. We actually pulled in our 800-gig product because of higher demand. And very recently, a number of customers were looking for this data center interconnect to stretch beyond the historical 50-mile, 80-kilometer connections. So we introduced a product which can go 1,000 kilometers, which really helps some of the other broader architectures, especially for AI connectivity as you go from taking your training models and now you need to inference across multiple data centers, the DCI product is actually a big part of it. So yes. 100% agree. Optics is super exciting and kind of we see a lot of growth there.

Yes. And one piece that I would add is just that we're very excited that we're actually ready with 1.6T. We're [indiscernible] to sort of enable that next generation of accelerators going into next year. And we sort of start seeing the beginning of that at the end of this year and then much more significantly next year. But yes, the team has continued to execute really well on our technology road map and really being the leader there.

Got it. One thing we have noticed in optics over a long period of time is that there is just rapid price compression, right? We especially see that among the optical transceiver makers whose margins are much lower, right, than the kind of profitability. So how do you look at just the pricing dynamics of your specific DSP TIA drivers? Do you think it follows kind of a similar price compression as we see in the overall optical transceiver market? Do you think you are able to kind of hold the pricing better because of the market share that you have or uniqueness?

Yes. I think our differentiation in terms of being first and then enabling it really gives us the pricing power. And so that's why we're so focused on actually enabling that next generation. And then you have these sort of follow on where we can really optimize the product, right, and sort of maintain, from our perspective, a margin profile while enabling cost reductions in the ecosystem. And so you sort of have this [ TikTok ] when you're first to market, have the best product and then you're able to optimize that product for more broader use. So certainly, we're able to sustain very healthy margins for us while enabling cost reduction in the business, too.

Yes. I mean, basically, we are able to offer huge value to customers, right? If you're doubling the bandwidth, right, then obviously, you don't double the price. That's a significant advantage, right? And I think if you -- since we acquired Inphi, as you know, your own model actually, we've massively increased the revenue from that business, and the margins have been actually fantastic. So yes, we don't -- I think as long as you are the technology leader and you're driving innovation, and we continue to intend to do that, we -- it's not just the DSP, it's the TIA, it's the driver. It's the entire optical platform. We are now also bringing out our internal silicon photonics technology, which we've historically used only internally, as something more broadly available. So I think our total value in that ecosystem is going up, and I think it really improves power yield for our customers. And yes, we get paid for that.

Do you see the competitive landscape changing in DSPs? Off and on, we hear just -- and by the way, this is not just with Marvell. We always see, whenever there's a market leader, like very strong 70%, 80% share, there's just a lot of talk of more competition, right? It's not unique to Marvell. But we do hear talk of even whether it's NVIDIA, Credo has spoken about a DSP, right? Others, right, have spoken about DSP. So how do you see the competitive landscape evolving over the next 12 to 24 months?

Yes. I think our focus is really to have the base technology, right? And so we've got this track record of executing, right? And you can go back, right, where even when we acquired Inphi, there was assumption around like that market share not staying at that level and really getting watered down over time. Certainly, when we looked at it, that was part of the discussion. But it simply not happened. And the reason for that is that these relationships that we have with the broader ecosystem, it's not just about bringing a product to market. We actually work with them over many generations and really have these tie-ins and telemetry and all the sort of special source that's enabling us to be there as they roll out the first generation of this technology. And our product is -- we've got a track record of actually being highly reliable, right? And so I think there's much broader sort of tie-in that people miss where it's like, okay, well, if you just bring out a piece of silicon, that's really -- it's hard to penetrate those relationships and markets.

Yes. I think the other reason -- I mean, the reason you're seeing a lot more competitive dynamics is the fact that the market is just massively growing, right? So it's not just in optics. Any market, which used to be a market which was somewhat niche-y, limited to hyperscalers was growing very nicely, but it was a relatively small market. That market is absolutely exploded, right? And it's spreading beyond just the U.S. hyperscalers. So all of it, we are looking at as actually it's a very positive sign. The market size is incredibly large. It's growing very quickly. It's not a zero-sum game. And it's fairly natural that market, which is going to end up many times bigger than people thought it would, is not going to be relying on one single supplier. We kind of get that. But to me, it's actually a very positive sign, is kind of why you're seeing. I think the biggest reason is the value of interconnect in AI systems is significantly higher. If you think about standard cloud infrastructure, the server CPU does most of the compute. And when you're doing compute, there's no interconnect at all. The interconnect just feeds information in and out. In AI, as you know, one single chip can't solve the problem, right? It's the connectivity in the middle is equally important, which is why I think you're seeing all of this activity in the ecosystem, is the value of interconnect is absolutely integral to how you build these AI data centers. And I think that's why you're seeing all this competitive excitement.

On the custom ASIC side, I think you have suggested over $0.5 billion this year going to over $1 billion, right, next year, hopefully, a lot more beyond that. How is the visibility over this next 2- or 3-year period? You think you are able to hold on to sockets that you're winning this year? Because there's, again, a lot of noise from your Taiwanese competitors about regaining, right, sockets that you might have won. So just what's your level of confidence and visibility?

Sure. Yes. We've had this question about 25 times at this point. But yes, so I think just focusing on the first one, right, in terms of visibility, clearly, we need very good visibility to be able to get the allocation on the supply chain. You can't wake up tomorrow and go get [indiscernible] assigned to you. It's just not possible. And so because these are all multiyear sort of lockups, if you look out. And so clearly, we have sort of a 3-way dialogue with our supply chain and the customer around what supply is required. And so that -- plus in the near term, we obviously have orders and backlog, right? And so the combination of that really gives us very strong visibility in what I would say the near term. And then more broadly, if you look at the longer term, we really have these relationships that is multigenerational, right? And so I understand sort of the commentary from some of the companies in Taiwan. And we don't control what they say. We can just sort of look at what we see, and we can go and execute on our product ramp. And so we're very confident on that. I'll limit my comments on Taiwan.

Yes, I think just at a high level, just think about market dynamics where the reason why some of these designs came our way in the first place is because when you went from 14 to 10 to 7 to 5 nanometers, it's more complex. You're going from 50 gig SerDes to 112, 224, 448 in the future, chiplet architectures, much higher-density compute cores. You're looking for someone who has done compute on ARM, which is us essentially in this ecosystem. So that's the reason why this stuff came in our direction. So I have to think fast forward. So what's going to happen to the next generation? Is it going to get easier? Is it going to get more difficult? Well, the answer is pretty obvious. It's going to get more difficult. So I would say, in the same time period, we've actually increased our design win success, and we've added another customer. And the revenue from that third customer, when we look at the cumulative revenue, it's actually even bigger than what we're going to get from the first 2 customers, right? So I think if you just look at the traction we are seeing, the revenue expectations we have, we have visibility to revenue growth for many, many years at this point in time.

The one other aspect -- oh, one question to just close that, and I think, Ashish, you addressed some of it. But what is the difference between what Marvell can do that some of your Taiwan competitors cannot do?

I think it comes down to IP. It comes down to having like the best SerDes in the market. It comes down to the execution we've done on very large compute chips. I mean, this is when we acquired Cavium. This was one of the core things. I mean multicore ARM is essentially the core of those products. I think that's the biggest differentiation. If you look at the team we acquired, which helped us to custom, which is Avera, this is the old IBM team, which has done thousands of high complexity designs and done them right the first time. So I think that's the biggest differentiation. If you're looking for someone who's done vertical size chips, running at the highest frequencies, where high-speed I/O is a critical part of the architecture, has done compute, the reality is there's literally not even a handful of places to go to. It's literally basically 1 or 2. That's what it comes down to, and has the ability to continue to invest in going not just to 3-nanometers, but we're already having 2-nanometer architecture discussions, right, well in advance of the technology coming out. I think when you put all of that together, I think the proof's in the pudding, like I said. I mean, we would not have won, out of the 4, essentially 3 hyperscalers, if those things weren't important, right? Otherwise, they would have kept doing what they've historically done. I think that's what the differentiation comes down to.

Yes. I mean the other thing I would add is just our supply chain relationships. I think we're very strategic about that. I think when you -- when we went into the pandemic, I think a lot of people realize that as long as there is excess supply, no problem. Everybody gets allocated. But when you go into a situation where allocation is key, having a supply chain partner with those deep relationships, multiyear relationships, very strategic relationships, it's a critical differentiator.

Makes sense. Two questions, Willem, on the financial model. So obviously, custom ASICs bring about their own, right, different kind of financial metrics in terms of gross margin. Do you think, overall, Marvell can kind of hold the 60% plus gross margin? Or do you think it can -- ASICs become so big that it will change that -- it can change that financial model?

Yes. I mean I think the way to think about that, Vivek, is that we're -- everybody is very focused on the gross margin. I understand that, right? But if you look at the fall-through in terms of operating margin and EPS, right, whether it's a dollar of our merchant product or a dollar of our ASIC product, both of them drives the same pull-through on operating margin and EPS, right? When we do these programs, there's -- our long-term target's, let's say, 38% to 40% in terms of operating margin, and so when we do these projects, we're targeting to be in that same zip code, right, if not somewhat accretive over time with the scale of these programs. And so I understand the discussion around gross margin. And if you go do your math, you're going to get an answer and everybody will have sort of their view. But over time, when you look at $1 of ASIC revenue, it's just as accretive to our EPS and our operating margin. And so we're really focused on growing that top line. And sort of we outlined this market opportunity in custom, right, at our AI day, where we think the market is going to grow very significantly. And we could be wrong or right in the market, but I think we have a view. And some people are even more aggressive than us. And then we don't see a reason why we can't take a very significant market share with sort of this dynamic where it's sort of us and one other player that are able to do chips in this zip code. So yes, certainly, I acknowledge that when you do the math, you're going to get an answer there in terms of gross margin. And everybody's will be a little bit different, maybe slightly above 60% or slightly below. But let's see how it plays out.

I mean, it comes down to mix size. I mean, so it's -- what we know is that the merchant products, especially in enterprise, are going to grow at some point. So you're going to get that uplift, right? But to your point, yes, we could be wildly successful in custom, which would be pretty cool. You're going to get a ton more revenue and it's going to have a very nice roll through. So it's kind of difficult, given kind of where some of the merchant markets are, it's kind of -- they're all at the bottom right now, so we have to kind of figure out, okay, do they all the way come back in Q2 next year or Q4? And what does custom look like? So we're going to need a little bit of time on the printed gross margin, but the reality is when op margin and EPS, which I suspect is the clear focus, that's going to just keep growing. We're going to drive a ton of operating leverage.

I see. And as you scale these ASIC programs, is there more OpEx leverage? Like how should we think about just the OpEx intensity? Is it -- so let's say, conceptually, if I say gross margins of these are, I don't know, 10 to 15 points below corporate average, should I assume OpEx intensity is also kind of 10 or 15 points? That's how you get that offsetting impact?

Yes. I think I would disconnect a little bit, Vivek, in terms of the OpEx intensity. I think if you about the investment that we're making on a custom program, there's, call it, x minus whatever NRE we're getting, right? And so if you look at a merchant model, right, then maybe that OpEx is like, what, 20%, right? But the scale of that is sort of finite, right? If you look at a custom model, the OpEx intensity is probably off or somewhere in that ballpark. But then you can scale that program. And depending on the market absorption, these products can be much more significant, right? And at that proof point, you get significantly more accretion. However, you have to overlay the fact that we're investing for the next generation, right? And so when you hit this run rate we're investing in at the moment, we're already looking at the 2-nanometer platform, right? And so our OpEx is really driven at this point by not these programs that's ramping, because those investments have been made, but these upcoming programs that we outlined that's out in 2026, '27, where we have to have the 2-nanometer platform in place.

But that long term, that 38% to 40% EBIT margin, do you think that's kind of line of sight in the next 12 to 18 months? Is that a reasonable...

I think from a run rate standpoint, sure. I think we can drive the business. I think the faster that custom grows, it will accelerate us getting towards to that level.

Yes, the 38% to 40%, just to be very clear, that's very much the target. And we certainly see as revenue reaccelerates, which is just doing, as we speak, is certainly very much in line of sight.

And you'll see the accretion in our model here in the second half, right? We outlined that we see growth accelerating in the second half, and you'll see that acceleration in our OEM.

Terrific. On that optimistic note, thank you, Willem. Thank you, Ashish.

Thanks, Vivek.

Thank you.