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

Good morning, everyone. Welcome to this morning session. I'm Vivek Arya, head of U.S. semiconductor and semicap equipment research. I'm absolutely delighted to have the team from Marvell. Joining us, Matt Murphy, President and CEO; and Willem Meintjes, CFO of the company. I'll go through my prepared questions, but if you have anything you would like to bring up, please feel free to raise your hands, and I'll be sure to get you in. It looks like we should have had a much bigger room, Matt.

Well, it's because of you. I mean I've never seen standing room only before, Vivek, because of you.

I wish. I wish. Thank you. So welcome, Matt and Willem. Really appreciate you guys being here.

So maybe, Matt, let's just start with just a high-level state of the union. A lot of kind of mixed currents in the semiconductor industry, consumer enterprise not doing well, other areas starting to improve, right? So give us a sense of how you see demand now versus what you thought it would be at the start of the year?

Sure. Yes. Maybe if I can frame it a little higher level for a second. I think this cycle we've just gone through has been very unusual, right, different than any one I can remember over the last 30 years in that it seems very staged in terms of when companies, based on their end-market exposure, were coming out of the cycle, if you will, right? So last summer or last year, some folks started showing demand weakness. They were resetting. Everyone's kind of got their own path. So over the last couple of quarters, we -- our peak quarter was in our third quarter. And basically, in that sort of 6-month period, we did see changes in the demand, right? We saw data center slowdown. A lot of us got surprised last year on cloud, right, kind of how fast that turned. And yes, with the overall macro, interest rate environment, inflation, et cetera, things slowed down. So reason I say all that is we took action, okay? And if you look at our first quarter, while our owners at the time were disappointed with the guide for Q1, that's it. We actually overachieved a little bit, right? We had a beat on our Q1 revenue and earnings. We guided up Q2. And we basically said, look, storage is coming off the bottom, huge positive because that will be a tailwind for us. Cloud is growing again double digits from Q1 to Q2. 5G, which was up 25% in Q1 sequentially, that's still hanging in there and going to do that at least through the third quarter. And even in the enterprise area, Vivek, which we proactively took some measures with our customers and channel to get the inventory down because just an anticipation of that slowing, that business on a run rate basis is still so much higher than we sort of ever projected if you go back to like our Analyst Day from a couple of years ago, that it's very healthy. So while there's a lot of mixed messages on the macro and a lot of end-market differences going on for Marvell and where we're at in our cycle, we've sort of cleaned it up. And now from here on out, it's really, I think, a very positive trajectory. And of course, we've got the AI contribution, which a lot of companies don't. And we'll -- we're going to see that this year, especially in the back half and even into next year. So we've got a lot of tailwinds in the company now. I know it was tough for some investors to go through. Unfortunately, I think every company has got to go through their own other side of the cycle, and that's what we've done.

Excellent. So Matt, since you opened the AI box, let me open it further.

Can we close it real quick? Actually, I don't mind talking about it. Why not?

[indiscernible] much bigger box.

We might as well [indiscernible] box. Yes, sure.

So I think Marvell is kind of unique in that you have multiple shots at goal, right, from an AI perspective, right? You have the compute side, right? You have the networking, the electro-optics side and parts of the storage business that are also exposed to that trend. And on the call, you mentioned you had about $200 million last year, doubling this year to $400 million. So could you just give us a sense of what is the networking aspect of what you see AI today? And then what is the networking plus compute aspect as we go through it? Because we heard many numbers on the call. I just want to make sure we have a way of differentiating between the networking electro-optic side and the compute silicon side?

Got you. No, great question. And just to frame it from the very beginning, this is not a new thing for us that we suddenly just said, oh, wow, let's go do something in AI, okay? I mean I remember when we were doing diligence on Inphi in the summer, fall of 2020, and they basically laid out their road map and all their opportunities. And I very vividly remember that team and Loi and Ford talking to me and saying, "Look, basically, here's a block diagram of all these AI clusters being built." And here's the optical interconnect and it's -- and we think that basically at 800-gig, this is really where it can inflect. Now that was assuming AI was a small part of the total cloud spend, right? But that was something where that team and then following on with Marvell, we worked very closely with our customers. And we're already on the next-generation road map to really line up the networking side or the optical connectivity. So that's been a multiyear investment, right? This isn't just something that's brand new. And that basically generated around $200 million of revenue last year. That's a number we gave you on the call. And that was almost all of our AI exposure in Marvell. In parallel, well, and just to finish that, we said that our AI revenue from Marvell was going to at least double this year, okay? So $200 million last year, so at least doubling this year. Hopefully, it can be more. We'll see how that goes. And the bulk of that is driven by the 800-gig ramp, right, which continues. And because of the truly explosion in network traffic, some of our DCI products that have -- which is the 400ZR stuff, that's also levered to that trend as well. And we've got some way that we can allocate and break that out. So that's a little bit of it, but the bulk of it is still the networking at 800-gig PAM4. Then you asked a question about, well, then beyond that, when you have sort of networking and compute, then how does that play out? And that's also been a multiyear journey. We acquired, some of you remember, an asset called Avera from GLOBALFOUNDRIES in 2019, which was really the intact IBM Microelectronics' ASIC design team that have been around for 25 years together. I mean they've done 1,000 ASICs as a team over their history, so very experienced team. We acquired it at the end of 2019. We've pivoted from the GF road map, which was no longer sort of on the advanced node. We got them on our TSMC platform and got engaged in a number of custom silicon areas, one of them being in hyperscale and some of those being for AI as it turns out. And that design win funnel we had kind of built up during 2020, when we bought Inphi, the reception was so positive by our customers about that acquisition and how we were going to integrate it and fund it that, that led to us actually closing and winning a number of cloud-optimized custom silicon designs. So when we look at those -- and we articulated those in the fall of 2021 that we had won these designs and kind of long term, the revenue would be like $800 million in total for those. Now what we've said, and it's really going to hit next year, is the lifetime revenue of those same designs, that same basket of designs has increased, right, from when we won them several years ago, which is good. And the second is that the percentage kind of allocation between AI and non-AI has gone up dramatically. Like we just sort of ballparked it at 20% of the total back when we announced them. It's the majority. It's well over half. So some of those are kicking in next year as well. So we've got another at least double from this year to next year. The networking and optics is going to continue to grow, but also then we start layering in some of the custom silicon opportunities as well. And then there are other things that we haven't quite put in the whole model because we don't want -- some of them are sort of futuristic. But we have a very, very strong switching platform at 51.2T that came from a lot of the IP from Innovium, which is a company we acquired shots on goal at AECs, right, which is a technology that I think as you get to 100 gig per lane with the PAM4 technology from Inphi, that plays well to us, and then even longer term CXL. So I think there's a rich set of things we can do both in the custom compute side and the networking, optics and beyond. And then finally, on storage, because of our leading position, both in flash and HDD controllers, over time, that will just -- we're not putting any outsized growth numbers on that. That will just continue to kind of grow because exabytes are only going to sort of expand from here, not shrink, given the amount of data being generated and the value of that data probably increasing a lot more with AI being a key driver in the industry. So I think that's a little bit of a unique thing about Marvell, it's not well understood. We're actually pretty broad in kind of what we sell in there. And I think over the long term, that should serve us well.

So Matt, just to clarify some of that. So if I look at the electro-optics side, sort of $200 million last year. The bulk of the $400 million this year. Is some of that $400 million part of the custom silicon that you used to suggest was a certain number, $200 million this year?

Yes. No, no.

So none of that? So that custom silicon is very small this year?

Correct.

Overall?

Correct. Yes, yes. And apologies to everybody for -- in the spirit of transparency, trying to give numbers, then it turns out -- everything ends up looking like an optical road map, 200 gig, $200 million; 400 gig, $400 million, AI, custom. I mean, I know it's confusing. We tried, we tried. But it will get simpler over time assuming we deliver and the revenue comes in, then everybody will be happy, not worry about the decoder.

So $400 million this year, mostly electro-optics, a little custom silicon. Next year, as it potentially doubles, then the optics parts continues to grow, and then you layer on the custom silicon?

Exactly. Exactly And we don't have that quite -- that -- we don't want to and we don't know them. And quite frankly, it's early, Vivek. I mean things are very dynamic. If you even look at the electro-optics stuff, if you went back to last quarter, even on the call and afterwards -- the quarter before, why aren't you guys seeing more in AI? And then it sort of hit between last quarter and this quarter. So we continue to see the backlog building and demand increasing in this electro-optics area. So that's why it's -- I think it's premature to sort of try to call a mix for next year, but just to say we feel very good about the overall aggregate business at least doubling and where it all lands, I think we're still early, quite frankly.

Understood. Would it be fair to describe, Matt, that you are now more optimistic about the custom silicon business for next year? There were some media reports earlier today about Marvell winning something with a well-known hyperscaler. So any -- I know you don't want to be customer-specific, but what is giving you that incremental confidence about custom silicon for next year? Is it better customer commitments? Or what's changed?

Well, I think a couple of things. One is, as it's become clear that the mix has shifted between traditional cloud infrastructure and AI and everybody in the room understanding the criticality of these AI projects. You should assume that the importance of those has just cranked up in the last 6 months, right? So -- and the other thing that gives us confidence is we started these like 2, 3 years ago, right? It's not like you want it a quarter ago and now you're -- you can't get the product out that fast given the level of complexity for these customized silicon things. So I think a combination of time elapsing, the importance of AI and probably our customers also going through their process over the last 6 to 9 months of figuring out their priority, right? What's their road map? What do they really care about? Where are they going to spend the money? So that's firmed up quite a bit and just given us a lot more confidence. And by the way, the products are also farther along, right? Like we gave 2 examples, there are more. But one where the product is already back and working great, right? The other one is taking up this quarter. So time helps with this and also the shift in sentiment. And firming of our customers' plans, I think, gives us confidence in the numbers for next year.

Got it. How broad is this overall ASIC opportunity? Because there's one very large incumbent on the ASIC side. So what's your value proposition when you go and talk with the hyperscalers, like you should engage with us because of these reasons? And do you see these projects extending over multiple years? Or is this a stop cap because there are shortages of GPUs and whatnot? So just what's your kind of go-to-market value prop? And then what is kind of the sustainability of some of these opportunities?

Yes. Maybe I'll start with the sustainability first. These are -- because these projects are literally multiyear, right, to go execute, you can't really blow with the wind, if you will. So I think that's what's been a really helpful thing is as our customers have gone through their road map alignment, we've got a pretty clear view and they have a clearer view of what they want, okay? And they also -- and we also need to think about when you engage, it has to be kind of multigenerational is the mindset because -- otherwise, you're recreating the wheel. You're not getting the advanced learnings. You need to sort of, on our end, have the commitment to the next node, to the next package technology, to the next interconnect, right? So all those need to be there. So I don't think this is -- there's plans adjusting because now there's a shortage of GPUs. They would have had to have started a couple of years ago on that. And these are very strategic projects and commitments that, in our view, complement and augment what the top merchant supplier provides, which is like incredible technology, right? So this is not a zero-sum game where if Marvell wins an accelerator or our large competitor wins an accelerator, then somehow NVIDIA goes down. Like the whole pie is growing for this, right? And we view this as every company is going to have their own strategic priorities within that. So I think that's the first step. The second -- first part of your question was the value prop. And I think what's happened in the industry is everything is consolidated and there's -- and quite frankly, the cost and the complexity to really be able to engage in the type of technology we're talking about at 5- and 3-nanometer reticle-busting designs, multi-chip module, custom interconnect, I mean you're talking about thermals and power dissipation on some of these products probably being kilowatt. I mean these are big, big complex chips and they're for critical, critical infrastructure applications, right? So if you sort of say, well, who can do that? And by the way, where I don't have to worry about geopolitical risk or -- because that's another factor, right? Where am I going to source the -- source my future from relative to the design capability. I think it's really us and a large incumbent. And I think both companies have very strong technology and both -- we both compete vigorously and aggressively for these sockets. But I do think in the end, the pie is growing so much. I think there's clearly room for 2 strong companies plus a strong merchant company to really make the -- make our cloud customers' ambitions happen. And we could get into speeds and feeds and what they have and what we have and how -- and there's a lot of, I think, very legitimate kind of technical and business arguments we make and they make. And there's other things too, relationship and how you treat and what else do you sell the company? So there's a lot at play. But I'd just say at a high level, I think both companies -- I think the whole pie is going to grow. So I think the large incumbent's probably just end up doing a lot together.

Got it. You mentioned one important aspect, right, which is that process road map, right? And it seems like that as we go down the complexity curve, it's possible that semi companies with this IP will actually be more at the forefront of doing this as opposed to the internal teams, right, within the hyperscale. Because this is -- this looks like it's going to get more difficult over time, right? It's not getting easier. So what is the role of process in this? Like are all these designs on 5 and then you have a road map to 3 as well? Or how is that relationship?

Yes. For the revenue that we sort of called out a couple of years ago that we have won, that was all in 5-nanometer, right? And we see our 5-nanometer technology platform extending for quite a long time. I think it's going to have a lot of legs just because I think each of these steps now as you've gone kind of sub-10, to really make the lift to do the next one, it's hundreds of millions of dollars of investment, right? Maybe it's not $1 billion, but it's significant, right, to go make that next step. And the cost curve isn't helping anybody, right? The kind of cost per transistor area really isn't -- you're not getting that Moore's Law benefit on cost anymore. You're getting it on power and performance. So I think applications tend to get more selective. But if you don't have the next road map, you're not going to win the current one. You have to invest. And we do see a strong pipeline of 3-nanometer opportunities. But it's also not just the node, right? I mean we do a lot of work. We're much more quiet, I think, than the other companies in terms of our sort of chiplet and die-to-die interconnect capability and how we package these things up. But that's a critical part of the equation as well because I think, especially even if you go down to 3-nanometer, you're probably going to have a hybrid type of approach. Core silicon and 3, the CoreLogic and then you're going to have a lot of I/O probably at 5. And so I think that's an area where you go back to how can you help your customer and how can we help co-architect. And to your point about internal -- going like fully internal, I think it's just -- the economics just don't ever pencil out. I don't know how if you're a CFO at one of these very large companies you sit there and say, yes, I know you need to hire 2,000 people and build your own semiconductor company inside our company. But do you really need to do that? Or can you just have like a really good front-end design team and then find the key partner who can get you the right cost, yield, supply? I mean think about the supply, too. People have not forgotten what we just went through. And for us to be able to ramp these products, we did multiple long-term agreements for substrate capacity as an example. That was like a year or 2 ago. And that's for like next year and the year after, right? So if you just sort of wake up 1 day and say, I want to go build my own 3-nanometer kilowatt power dissipation, monster ASIC, how are you going to do that? So I think the role of key ASIC partners like us is going to be more and more important. And I think we'll be there to, again, help these large companies do the chips they want but do it in probably for them a more economical fashion and kind of carve out of place. So they can also control their own destiny, too, right, because they know their workloads and their own internal leads better than anybody.

Now, Matt, one thing on the economics of becoming more exposed to ASICs, right? Typically, these are lower gross margin products. But how does that relationship work? Because I know that they also fund you on the NRE side, which from -- if I'm right, you reflect more on the OpEx side. So ultimately, you get the EBIT dollars. It's just that it reflects in a different mix of kind of gross margin, right, versus EBIT. So talk to us about -- right, you mentioned that exiting the year, you plan to get towards that lower end of that 64% to 66%. So as these ASIC opportunities get bigger, what happens to gross margins and EBIT margins?

I'll say a few words, and I'll ask you to chime in after. I think the first is you're absolutely right, right? There's a different business model for custom ASICs, if you will, than merchant products. I mean that's kind of been, I think, understood since the dawn of man, okay, right? That's -- and in fact, when we bought Avera, we outlined that, right? We said here's the gross margins, here's what it looks like. And to your point, for Marvell, which is different than some of the companies, we don't recognize the NRE as revenue. We recognize it as contra OpEx, okay? So the customers are funding the programs. We're doing that for them. The -- and I'd say if I go back to you when we acquired Avera, the whole custom effort for us, the custom business has grown pretty significantly, okay? In fact, I think it was as of last quarter, we looked that the custom stuff had actually grown faster in top line than the merchant. But the gross margins for Marvell, if you just sort of step back, actually have kind of crept up a little bit. And that's because we have a lot of other shots on goal, right? We have set the volatility aside for a minute on the storage business, right? We've been able to grow the storage business. We've been able to grow our networking business. We've done things that are accretive to gross margin like acquiring Inphi. So we've managed that portfolio. And some of this got exposed and is getting exposed now because as we've had this correction we're going through, some of our ASICs are still doing well and some of our merchants lower. And so now our gross margins come down. But for the past 4 or 5 years, we've managed that without making it anybody's problem, right? So as we -- to your question about, well, what does this look like if this really ramps up? As we look through the end of the year, I'll let Willem comment in a moment, for a number of reasons, we're confident that gross margin can come back up. We're confident looking into next year that gross margins will be higher. That being said, just let's get a hypothetical, let's dream the dream. Let's dream the dream. Let's say, hey, this custom silicon thing you guys had it turns into a monster next year, okay? And there's a whole bunch of upside to these numbers I gave you. But it's got a gross margin with the 5 handle on it, okay? But remember, that engineering work's all done. The chip's done. It's shipping so that the fall-through would be enormous, in theory, because there's no incremental cost to support it. So it really works for us. So even if we got all this great upside and the blended gross margin was lower, the operating income and EBITDA would be so much higher, it would just -- nobody would care. That's my view. And maybe at some point, we need to -- depending on the size of it, we need to start giving some more view on that, how that would sort of work for our model. But we think that, net-net, that's very positive for us. At the same time, though, we're not giving up because we have businesses like automotive, like networking, like switching, like optics, right, that are all very strong gross margin businesses for us, right? So we're not changing the range, but if the worst thing that happened was we had a little bit lower gross margin and a ton more top line and a ton more earnings, I think probably every investor in the room would say, hey, thanks. I appreciate that extra EPS, not, hey, you lost 0.5 point of gross margin because you took -- but we -- and I think the key, just the last point is we really think about -- we manage those as different businesses, right? So the one thing I want to ensure investors is the non-custom part of the business, that's continued to maintain and had expanding and very strong gross margins, very strong. And so we managed the portfolio. There's a longer answer, but I think it's like pretty much -- it's kind of top of mind for folks. So do you want to add anything?

Yes. Sure. Just to maybe bridge to Q4, right? And so exactly to Matt's point, when we look ahead here, the wireless business continues to be really strong through Q3. We do expect that to step down significantly in the fourth quarter. So that's a big tailwind for us. We're talking about the storage business and how that's bottomed in Q1, expect that to sort of continue to come back and grow through the rest of the year. We don't expect it to come all the way back to sort of our pre-pandemic level. But certainly, show nice growth through the rest of this year. And in these optical products that's really being driven by AI has really nice margin. And if you go back to sort of what the Inphi margin was and as we see that going, so that combination is all very accretive to gross margin. And then in addition, we're taking all these steps internally, right? Our supply chain, our operations team, we had sized that for this year to be a much bigger revenue year, right? And so all the actions we're taking to optimize that, working with our supply chain on where we've got dual-sourced products. So all these actions really have to flow through our inventory as that turns. And so it will take a couple of quarters, but we really start seeing that benefit in the fourth quarter. So we're very confident to be back there in Q4 and then driving that into next year. And then that really summarized our view for next year, but very confident going through into the back half here.

Got it. So even -- just to summarize that, Matt, even if, let's say, you kind of have the expected trajectory on the custom silicon side, it is still EBITDA-accretive, right, for next year? And it can be far more if it's 100%?

There's a massive leverage in front of us, quite frankly, in that custom area because like I said, we spent the last 2, 3 years developing these products, right? There's more -- there's a pipeline, there's a big design win funnel. But as those flow through and they actually inflect up, it's at a much higher run rate than that business has ever been at. So where would you spend it? You know what I mean? We could decide to spend more if we wanted to win more projects, but that -- we haven't signaled that in our kind of OpEx trajectory. We think we got a very healthy size team. And just as a side note, we had the original Avera team, right, which came out of IBM. But recently -- and Raghib Hussain, who's our President and he was the founder of Cavium and runs that whole business for us, he actually just recently combined our kind of core processor team and the custom ASIC team. So it's actually one big monolithic organization now. So instead of -- you had a group who are doing processors and a group -- it's actually a lot of common skill sets. So our engineering capacity and our ability to go scale and go do these is we're in a really good place. And so then it's basically a lot of leverage from here. And that's the exciting thing about that business. It's got a lot of operating leverage in front of it.

Got it. And then just the last one, better margins obviously mean better free cash flow also. So as you look at the portfolio, Matt, do you think it is rightly aligned, like it has all the building blocks that you need to execute? At your Analyst Day, you laid out a 15%, 20% growth trajectory, right, which was executed on very well, but then we obviously got into the downturn. So as you look over the next 2, 3 years, is that still the kind of growth rate that Marvell is capable of? Or do you think that you have opportunity to augment it with M&A?

Sure. I'd say on the -- do we have what we need, I would actually stick to exactly what I said in October of 2021, which is we did what we needed to do to build the portfolio that was needed for this data infrastructure massive opportunity. And we got it, and now we've got to go execute like crazy. I still feel that way. M&A is hard, it's not easy, it's risky, all the things. So it's not like we love doing it, although I think it's been hugely value-creating how we've been able to take some of these assets and do it. But we're very focused on the organic trajectory of Marvell. All the benefits you get from that over time, by the way, when you do organic, like you got a bunch of free cash flow and you could just go buy back your stock or you can go -- you get tons of flexibility when you do that. So there may be other things that we do that are smaller, who knows. I mean there's -- we found really nice teams. I think in this environment, probably more pop up. But you should just assume this is us cranking forward the organic plan. And then as far as the growth rate goes, we'll -- we haven't sort of decided on when the next Analyst Day is, so it will be interesting whenever that is to benchmark. But we basically took the growth rate we sized at 15% to 20%, I believe, was based on our Q3 calendar '21, fiscal '22 run rate, which was the first quarter when we had Inphi. And that was like [ 10.76 ], I think, was the revenue that quarter. So just call it well under $4.4 billion, something like that, it was combined. So look, we overachieved last year, right? We did $5.9 billion. We'll see where we end up this year. But we've outlined a pretty compelling growth story for the company next year, and that only continues in our belief. So we'll see. I'm not taking anything off the table, right? If you kind of look, if we have a strong fiscal '25-'26 and we've got some Investor Day sometime next year and we lay it out, compounded, it may be okay, right? And then how do you look from there? We still see -- we still believe -- I'll just kind of end on this. One of our thesis has been and our beliefs in how we sort of constructed this company was the end markets that you're a part of really matter, probably matter almost the most of anything else you've got to go do, right? You need technology. So when you look at our SAM, we don't have negative SAM grower things in our opportunity set. So we think our SAM is still growing at probably 10% to 15% a year because we don't have PCs, right? We don't have smartphones. We don't have things that are ex unit. And so that's -- the infrastructure SAM is going to be the highest grower if you're a pure infrastructure. So just keep that in mind. And it's hard to have a down quarter, things get bad. Well, oh, my god, Marvell lost the plot. What's going to happen? But our view and my view even in the last couple of quarters was nothing's changed. It shifted. It's more AI. It's less traditional server. We got to do an enterprise. But the same principles are there, and we are so flexible to go hit all the shots on goal for the data infrastructure opportunity. We're the best pure-play kind of positioned company for that, that doesn't have the other things. So...

Terrific. Thanks so much, Matt. Thank you, Willem.

Okay. Thanks, Vivek. Appreciate it. Yes, thank you.

Pleasure.