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

Okay. Why don't we go ahead and get started? All right. Good morning. My name is Harlan Sur, semiconductor Capital Equipment Analyst here for JPMorgan. I'm very pleased to have Raghib Hussain, Chief Strategy Officer of Marvell here with us. We also have Ashish Saran, Vice President of Investor Relations. Thanks for joining us this morning.

So I'll kick off with the first few questions. Raghib, as Chief Strategy Officer, you're driving the future technology and product direction for the company. And despite a tough demand environment in both your storage and networking businesses in calendar year '19, it was a solid year for the team in terms of design win traction across your networking and storage businesses, adding new capabilities and revenue opportunities with the Aquantia and Avera acquisitions. Help us understand how the design win pipeline and the acquisitions will drive revenues over the next few years and also diversification of the business?

Thank you, Harlan. Good afternoon, everybody. So if you look at the overall Marvell business now, combined with the latest Avera and Aquantia combination, we have pretty strong position in various segments. So starting with, let's say, 5G. And we have positioned, from the processor point of view, transport baseband processor and now with the combination of Avera, we have DFE as well. So we have multiple customers and multiple design wins all across the various combination, not only including processors, but also the switches and Fis and so on. So that is -- that puts us in a very strong position for 5G. And as the 5G segment of course grows, these design wins are going to drive in the coming quarters and years. If you go from there in automotive, we had pretty strong position in automotive in the networking side, because as you know, this car architecture is moving towards more zonal architectures, where things are connected using Ethernet. We had a strong position in the gigabit and 100 megabit connectivity, both switches and Fis, which is auto-grade secure with a combination of Aquantia, now we have a leading position all the way to multi-gig, all the way to 25 gigs. So ranging from very -- 100 megabit, very basic connectivity to the main cores of the car main network, which is much higher bandwidth. So those designs, as they go in production, they are going to start driving revenue as well. And then in our traditional segments, in our enterprise, it's still going through the transition of going from gigabit to multi-gigabit. We are winning our designs in switches and Fis, not only in the campus, but we are also winning in other segments now. So those are going to drive segment -- drive growth in the coming years. And coming to the storage side, we have designed -- and with the flash side -- the custom flash, as well as we have designed in near line storage. So those are going to start driving revenue in the coming years.

So I appreciate that. So back at the 2018 Analyst Day, the team put out a revenue growth target of 6% to 8%., with your networking business growing low double digits year-over-year type CAGR, your storage business is growing at a low single-digit type CAGR. Given the design win pipeline in networking and storage, the acquisition, some of the near-term dynamics in the markets, has your view on the longer-term revenue growth profile changed, maybe networking growing a little bit faster? Maybe storage looking flattish? I'd love to get your views on that.

Yes, let me start with that. So I think, if anything, the design win success has actually made those goals even more achievable. Clearly, in networking, we already had a strong position when we did the Cavium acquisition, but more recently, we've added more design wins in the base station space as well as in the automotive market, right? So I think that double-digit CAGR in networking is, if anything, potentially even stronger. I think on the storage side, while we've certainly gone through a transition, where the HDD business, probably smaller today than what we anticipated. But we knew that business is going to transition towards enterprise data center, more flash and that certainly is happening, right? So once we've achieved stability, which I would say at this point, we have, we grew the business appreciably in the third quarter, right? And it's, again, flat to up in Q4. So I think storage has kind of crossed its hurdle. I think going forward, we do still expect very much that this is a low single-digit kind of GDP-like business, but it's obviously very attractive to us because while it has low growth rates, the investment profile is fairly lean and significant free cash flow generation.

Near term, the team and the industry is facing a broad slowdown. Your storage business is going to the notebook transition away from HDD, CPU shortages, which is impacting both your client HDD and SSD businesses. In networking, you've got the enterprise and campus upgrade cycle, which is taking a bit of a pause because of the macro uncertainty. Your core business before acquisitions and wireless divestiture, we estimate, was down about 20% year-over-year in the October quarter. It looks flattish sequentially going into the January quarter. Obviously, a good sign that fundamentals are potentially bottoming, but help us understand how you see some of these headwinds playing out in terms of length and magnitude of the weakness?

Yes, I think the majority of them -- the vast majority were really macro-related, right, where I think the situation -- the global trade situation had the biggest impact, especially enterprise spending, which obviously affects a good part of our storage and our networking business. To the extent some of these headwinds start to get behind us and some near-term peer results we've heard as early as yesterday certainly seem to indicate that there is signs of stability, right? And I think what's really interesting for us is most of the growth you're expecting this year, the year we're in right now, is really on R over R idiosyncratic growth drivers, right, which is our own product cycle. To the extent that the headwinds we've been facing actually fade away and actually become tailwinds. I think you should see even more growth, right? So I think we feel very good about how we're set up going into this calendar year. And to the extent some of these headwinds, which certainly there are signs, right? You've got the Phase 1 of the deal in place. To the extent those things go through, businesses feel more confident about restarting IT spending, we'll be significant beneficiaries of that.

So Raghib, back to you. Given your role as a driver of innovation, technology and product leadership for Marvell. I think the key elements to a successful networking and storage semiconductor business, whether it's standard product, ASIC, semi-custom, is big, complex chip design expertise, having the right IP building blocks, digital, analog, accelerators and just as important, the software firmware and systems architecture. Help us understand, relative to some of your other peers out there, competitors, Broadcom, Intel, others that do these big chip designs. And even some of your customers' internal silicon design teams, how is the -- how does the Marvell team stack up on these metrics? And how are some of the acquisitions going to augment your capabilities here?

That's a very interesting question. So if you look at the way Marvell is positioned today, it's a very unique position there. It's not only the high-end chip design capability, but it's also the combination of IP and the system know-how that we have. So if you look at various target markets and the IP that's needed, so first of all, it's starting with the processor solution that we have. It's not only the CPU ARM core, but it's a combination of processor with the SoC, with the combination, how do you augment various type of processing into DSPs or into hardware acceleration, that is pretty unique to Marvell. In reality, I do not see anybody even coming close to that. And that actually gives us a very good platform, this whole OCTEON platform, which can actually go into various markets. So as you may have seen that we started with the enterprise security market, and we have been leader in that market for years and years. We expanded from there into broader enterprise market. And then we expanded into really 5G wireless market, so baseband processing and now the needs of the processing is increasing in, let's say, Intellian radio head, so we are expanding into -- further into there. So that gives a unique platform -- for example, if you take example of 5G, the same software architecture is in -- can be in transport, baseband as well as in the remote intelligent radio head. So a similar architecture, having this flexibility, to having a programmability and ease of transition from various kind of application, it gives a very unique capability that Marvell has. Now similarly, when you go beyond 5G, let's say, if you take example of automotive, we not only have Ethernet capability, in fact, we are leading in -- when it comes to Ethernet connectivity in automotive. We also have other component, which is AI inference and security as well as processing. So as these architectures evolve, we are very well positioned to be the leading player in that market as well. Security, we already talked about it. When it comes to virtualized cloud security as well. Our Liquid security product is very well positioned in addition to all the firewall security appliances out there. And then when it comes to switches and Fis, we already have a pretty established position in the campus and enterprise and then we are expanding from there into carrier and other markets, especially, the edge market. So as -- the way I see it, this combination of the IPs that Marvell has, which is a combination of networking, security, storage and compute, this is unique. I'm not sure if any company has that -- even come close to that. And that -- those are the IPs that are needed for infrastructure -- future infrastructure solution, whether it be wireless infrastructure, wired infrastructure, automotive or even the edge compute. So I feel pretty good about this disposition. And in fact, we are leveraging all this capability to gain market share in these markets.

Harlan, just adding a little bit is -- I think, inhibition to the IP pool, which Raghib just talked about, I think the addition of Avera also is pretty unique, where we have a solution set where the delivery model is very flexible, right? So it's no longer the traditional either you buy a full standard product, something we built for everybody or you do a full custom ASIC where you, as a customer, do all the IP development, right? Now we have a full spectrum of capabilities. And I think that is going to be a very unique differentiator, in addition, to all the IP elements we talked about.

So moving on to more of the specific product-level dynamics. The big focus for Marvell is your 5G base station design win pipeline that started to unfold in Q3 of last year with your lead customer, Samsung. You've got $3,000 plus of content per base station. Your second customer will start to ramp in Q4 of this year. All of this is core Marvell-developed. Now we layer in -- on to that, your Avera ASIC team. They've got a strong position in digital and analog content that sits more in the sort of remote radio head inside of the tower. We believe they also -- Avera also brings a new set of 5G customers to Marvell, including Ericsson, ZTE and potentially even Huawei, if the ban comes off. Help us understand what are the strengths of the Avera team in 5G and how that complements Marvell's position in modem, networking and control plane? And how big do you think now your total silicon SAM opportunity is in 5G, combination of Marvell with Avera?

Let me start with just the SAM opportunity really quick and then Raghib will add on in terms of the capability matrix. So I think if we just do very simple math, right? So just say there's roughly 1.5 million macro base station ship worldwide. And that's kind of an average number for a number of years. It goes up and down off that base. If you look at the total addressable opportunity between core Marvell plus Cavium and now Avera, it's roughly $4,000 of content in a full 5G base station, right, once it's fully deployed. So that's about a $6 billion TAM, right? And of that, let's say, 1/3 is not addressable based on certain Chinese customers, which obviously we wouldn't be deploying into. So it's about a $4 billion opportunity, right? So that's a very, very large serviceable market. So that's how to think about the total opportunity, and then in terms of the rest of your question, I'll let Raghib add in what Avera brings to the table.

So as I said earlier, we had pretty strong position coming from the main leader, compute side of the platform. So starting from the transport, expanding into the baseband processing and then for expanding further into intelligent radio head for massive MIMO processing and so on. . Now what Avera bring is they have a lot of expertise in the DFE side. So -- and ASIC capabilities. So in all of these situations, there are parts that -- component that customer want to design themselves and they want ASIC services, and then there are component where customer want to partner which we call CSSP, which is customer-specific silicon product, which we do with bunch of other customer. And then certain customer products like Control Plane Processor are more of the off-the-shelf type of thing. So at this point, with the combination of Avera, we have all the models, we can provide all the 3 type of solution for the customer. And in fact, we have all the IPs and capability to address the complete range, ranging from the DFE, all the way to the transport processor.

So just sticking with Avera, a strong ASIC team. This is the IBM ASIC team that has a world-class reputation for large, complex ASIC designs and strong IP portfolio. We tend to focus a lot on 5G wins that Avera has, but I actually think that the bigger opportunity that Marvell has with Avera, is the big, complex ASIC market, targeting cloud and hyperscale data center guys that are starting to do their own silicon. We think that this is a $2 billion, $3 billion per year type of market opportunity for these ASIC-type products, AI and deep learning ASICs, security and storage acceleration and so on. When you combine Marvell's IP portfolio and networking expertise with Avera's ASIC team, how well positioned are you in cloud and hyperscale? And have you guys already secured some ASIC wins here?

So very well-analyzed. In fact, overall, SAM, is a little bigger. It's about $4 billion, if you combine the -- all the AI hyperscale networking as well as wireless infrastructure. As you rightly said, the Avera team is an exceptional team. They have a long experience -- and the industry trust. I mean, if you go and talk to customers out there, they really admire the capability and things that they have done in the past. The issue with them was, they were kind of restricted because of the part of Global for the last several years. And as a result of that, they were able to service only those customers who are willing to, of course, use the Global as a foundry. Now with the combination of Marvell now, they have not only access to -- of course, I mean, they can use any foundry, but also all the advanced node capability that we have developed in Marvell itself, not only the methodologies and so on, but also the IP itself, which is needed for those high-end ASICs. So with that combination, now they are very well-armed. And in addition, what -- the way we see -- look at the problem in Marvell, we look at it from the system-viewpoint of view. And that actually puts us in a unique position where now we can go to a customer and really address their problem. "Hey, this portion can go into ASIC and this portion we can augment with the standard products." And that's how we come up with a complete solution. So in fact, for example, they can have a similar software architecture going into ASIC as well by using our solution into the ASIC aspect as well. So those actually enable them very well. We are engaged in various -- I should say, high-value designs out there. In -- But of course, we recently closed. So those are in the initial stage. We expect them to win significant design in both networking as well as the hyperscale.

Harlan, there's a question from the audience, I think.

Yes, why don't we bring the microphone over here?

Just a quick one following up on that. There's been this long-standing debate between FPGAs and ASICs, and I'm not a chip engineer, but when you think about the expense and mass cost and complexity of leading edge, you would think that lends itself to this idea that FPGAs serve a purpose there because you have more critical mass required to justify the cost to spin ASICs. So Harlan put a number out there, you said, maybe I think it's a little bigger. How big does a particular design win need to be to justify the cost of spinning an ASIC for one of these cloud scale vendors who has a particular application or set of applications that they want to cater to with a dedicated chip design?

So it's a multiple. It's not only the size of the design. There are multiple reasons to switch from FPGAs and ASICs. So traditionally, if you look at where FPGAs fit well. So any time you want to put -- try something new, some new standard or something which you are not sure that how would it evolve or change you, you generally go FPGA route, but as the standards become solidified, then you look at 2 reasons: Either you want to take a look -- am I able to service the needs of the market through my FPGA or not? In other words, am I hitting the performance and power targets that is needed for that end product, right? As you know, in many of the products, when you go in -- out in the field, there are specific box that you can have limit of number -- amount of power you can have or amount of watt and the performance required is established by the industry. So if you take example, let's say, it's a 45-watt, whatever best you can get in FPGA, but if you are unable to compete with somebody else who is -- who has ASIC solution, at that point, volume doesn't matter. You -- if you want to win in the market, you have to go ASIC route. So that's one. The second thing it is, anytime something goes into volume, any volume product, right? So then you need to see cost-benefit analysis because you need to see that, if I keep doing an FPGA, FPGA has much higher costly versus the ASIC. Now ASIC, you need to argument people, they say, "Okay, you will have to spend design capabilities -- the designing efforts to design it." But guess what? FPGA designing requires similar type of efforts because it is not like easy to just program FPGA using C programming and all those sort of a thing. FPGA has its own complex programming, which is similar to ASIC design. So it's RTL -- it's called the RTL programming, right? So in other words, the complexity of design ASIC is similar to complexity of designing FPGA for that purpose, right? So in the reality, you need to see for my target market, what do I have to do in terms of performance power to win. And if -- can I address it with FPGAs or not? Secondly, what kind of changes evolving protocols are happening? And if it is -- if we can address it through the flexibility of program -- really like in our case, using the DSP processor or using the programmable engines or using the general ARM cores, then that is much more easier and cost-effective to maintain, and that is another reason to move to ASIC. So it's not only the volume.

Yes, let me add a couple of things because I think people sometimes -- and Raghib mentioned this, but I think there's this view that an ASIC is a completely fixed function device, right? So either do an ASIC and once it's done, it's done. That's a thought fallacy, by the way, right? So in a lot of our solutions, where we have our processor cores -- which, by the way, by definition, are programmable, right? So you make a choice in what you put in a fixed function, but there's a lot of flexibility in the DSP cores as well as the ARM cores, which again, like I said, by definition, you can rewrite the software stacks as the standard evolves, right? The second thing is with -- and this is pretty unique I think to us, which is with our flexible model, where we can start with the OCTEON product, which is a standard product but then create custom variants, the effort involved is not like a full multiyear ASIC effort, right? It's a much, much faster. Case in point, the second customer we added in wireless base station is going to get that chip in roughly a year from the first customer, which is not possible in a full ASIC. The reason is, we start with an existing product family, make a change and within basically a year, we can spin a new chip out, right? So the total investment is a lot lower. So I think the traditional benchmarks of full ASIC don't necessarily apply. We've actually lowered that standard. So it's much easier for customers to get -- call it, a semi-custom product.

Sure. Any other questions? Let's talk a little bit about automotive. With the Aquantia acquisition, the team acquired some key capabilities in Multi-gig Ethernet and also added to your design win content opportunities in Automotive Ethernet. . The Marvell team already was developing several product families of auto Ethernet products and getting solid design win traction. Sort of help us understand with Aquantia now, help us understand like, how big the Auto Ethernet SAM opportunity is for Marvell over the next 3 to 5 years? And when are we going to start to see meaningful auto revenues from the team?

Yes, I think from a SAM perspective, I'll outline that and then Raghib can add from a capabilities. But again, pretty conservatively, when we look at the next several years, we can come up with about $0.5 billion of addressable market between -- we do switches and Fis, so it's not just Fis. In fact, we've announced 16 design wins, this is about 3 or 4 quarters back with a number of leading OEMs worldwide, right? So it's Asia, it's U.S., it's Japan, it's Korea, a few of them started ramping last year, but the majority of them will ramp essentially for model year '21, which basically means you start shipping, call it, early fall of this year.

So second half 2020?

Second half of this year when they start ramping. Again, automotive, as you know, is a long play, right? Once -- it takes you 3 years to get in there. But once you're in there, these are like 7-year kind of model life cycles, right? And the way OEMs work is, they start with the higher-end model and then proliferate that across kind of the more bread-and-butter, kind of mainstream models. So this is going to be a very significant growth driver for us, and we are 1 of potentially 2 players. And we strongly believe we're in the lead in this particular market.

So just to add on that, of course, Marvell had a pretty good position within Ethernet -- Automotive Ethernet. Just to give you an idea, this is not like a Ethernet in the same -- exactly same as an enterprise. There's an improved, optimized version of Ethernet called T1, which is in automotive. So we had a pretty good position with 100 megabit and gigabit, both switches and Fis. And then Aquantia actually was very well positioned in multigig to 10-gig solution. So with the combination now, we have complete product portfolio of switches and Fis, ranging from 100 megabit all the way to multigig and 10-gig, all the way 25-gig. So in other words, if you look at the car architecture -- zonal architecture, the backbone are the higher bandwidth. And as you go through the various devices and node, those are the lower-bandwidth connectivity. So we have complete networking architecture for the car. We are in leading position there.

Let's turn to storage. The team has had a big focus on enterprise and cloud within the storage business, both on the HDD side and on the SSD side, we'll get to the HDD side in a moment. But in SSDs, you focus more of your R&D dollars in the data center side of the market. Standard product controller supporting NVMe, NVMe fabric connectivity. You've got some custom silicon or ASIC engagements given your strong IP portfolio. And then recently, you've started to talk about this DIY model that you're working on with some systems OEMs and some hyperscale cloud customers. So help us understand this new DIY model? And do you already have design win engagements with OEMs and hyperscalers?

Sure. Yes, I think this is a pretty -- I mean this is a pretty, pretty interesting shift in the industry, and it's not very surprising. If you think about what's an SSD, it's a bunch of silicon, right? Basically, it's a large amount of NAND, right? And it's one or more SSD controllers. . Historically, you couldn't disaggregate the solution where it's a hard disk drive, it's an electromechanical device. You had no choice, but you had to buy it in a standard form factor. I think as flash proliferates, especially, in the data center, right? The big thing you have to remember is flash in the data center is used for acceleration, it's not used for permanent storage. So here, if you can tailor the controller to your particular application, you can get much higher performance. And that's exactly what these system-level OEMs -- basically, somebody who's consuming a huge amount of flash and installing it directly in their system is coming to us and saying, "Hey, can you customize the controller for us, right?" It's not just doing the I/O., it's doing a lot more than just the I/O transfers. We started working on this -- it's been 2-plus years at this point in time. We won some of these design wins in that time frame. A fairly important design win is now, at this point, the development cycle is essentially complete, and we will be launching in volume production in the second half of this year. So it's fairly advanced at this point in time. And you should expect that as time goes on, right? And by the way, this is not unusual. I mean if you go back and look at a particular handset OEM, they did this about 7, 8 years back. They bought their own controller company, and they just buy NAND directly in the open market. It's a very similar business model, which is being employed at multiple customers. And we are pretty unique because, obviously, we've got our first PCIe Gen4 product out there, which is a standard product. But that's also the architecture for this custom controller, and you should expect more of this as time goes on.

On the HDD storage business, your design win pipeline in nearline or cloud HDD, which is actually the only segment in HDD, which is still growing. The design win pipeline has expanded pretty significantly here with Seagate and Toshiba. They're now starting to ramp their 16-terabyte nearline platforms using your silicon. Additionally, you're building, as you just mentioned, a strong pipeline of enterprise and cloud SSD controller wins with the DIY model as well. You have a target to drive 50% or greater mix in your storage business and data center market. Your design win pipeline appears to be moving in that direction. How close are you to getting to this 50% sort of threshold?

Yes, I think we provided this update at Analyst Day, which is about 1.25 years back at this point in time. So at that point in time, our client business is still the biggest business within our overall storage footprint, followed by Edge and other. And finally, enterprise data center was the smallest, which was somewhere in that -- if I remember correctly, like mid-20% range, roughly. I think we've made substantial progress towards that 50%. We haven't given an update, so I think just wait for the next Analyst Day. But I can say for a fact that look, our client exposure today versus a year-plus is a lot smaller, right? A lot of that have to do with the market dynamics of a shift in the remaining part of client market from ACs to SSDs. In parallel, if you remember, our SSD business, even 1.5 years back, was already 50% enterprise and data center. So I can tell you that's obviously a larger number. We started shipping in our 16-terabyte platforms only last year, right? So again, that adds to this shift towards enterprise. So I don't think we're at 50% today, but I think we're -- we made substantial progress, and we'll certainly update you during our next Analyst Day.

The team has done a very good job of really leveraging the OpEx structure and the R&D and it seems like R&D productivity is extremely high. Raghib, when you think about calendar '20 or the next 2 to 3 years, and the R&D resources at your disposal, how do you think about deploying them? Where are you going to be? Well, where is the bias going to be towards your R&D dollars? Is it going to be -- continue to be more focused on networking? And within that, is there going to be a certain bias towards 1 service provider, enterprise, cloud, even within storage? Is it going to be -- obviously, it's going to be more focused on enterprise, but is it going to be more SSD-like? Or -- help us understand that.

So on the networking side, I wouldn't say it's a bias towards any particular product asset, but it's more of a -- as I mentioned, when we look at the market we have to balance between our established business, which is growing, which is enterprise and then campus and so on. So we have substantial investment over there and we'll keep -- we are keep updating that and keep bringing that second general program. But then we also have a substantial investment in the 5G because it's a growth area. We have a various design, various customers engaged and all those. We are investing in that. So our amount -- in the near term, of course, investment on the 5G side is a little higher, but it comes from the base investment itself because the investment that we are doing in OCTEON platform, that actually goes across various segments of the market. So I really do not think that we are being biased on any particular end market as such in that segment. Now it's come to storage side, as Ashish said, that we have both HDD as well as the SSD, but then there are a lot more variance design on SSD side, especially with this DIY model. And of course, that will require more resources.

Last question. Post the sale of your wireless connectivity business to NXP, team has used the cash, going to use the cash, has used the cash to pay for the Avera, Aquantia acquisitions, pay down more of the existing debt on the balance sheet, should enable the team to restart the capital return program in a big way. Is this how you guys see it? And remind us, what is your target capital return as a percent of your free cash flow generation?

Yes. I think the target remains 50% of our free cash flow is what we target to return to shareholders as a combination of dividends and buybacks, and as you mentioned, because of the pending acquisitions, we had to stop the buyback program. You should assume that the buyback program, as Jean said in our last earnings call, is clearly a focus for us to get restarted. At this point, the sale is closed, the money is in the bank. We've already paid off the incremental loans we took on for Aquantia and Avera. The excess free cash, so you can imagine, certainly being used -- a portion of it certainly for restarting the buyback program. So I think we'll remain fairly consistent. Again, we look at buybacks not as something which is a point in time, we look at that as part of our overall capital return strategy. So you should expect us to remain fairly active as we go forward. We obviously had excess cash so we don't need this level of cash around the business. That's a much smaller number. So we have a lot of flexibility initially.

Gentlemen, thank you very much for joining us.

Thank you.

Thank you.

Best of luck in 2020.

Yes. Thank you.

Thanks, Harlan.

Raghib, thank you.