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

Atif Malik, I cover semiconductors and equipment stocks here at Citi. It's my pleasure to welcome Raghib Hussain, Chief Strategy Officer, Executive Vice President Networking & Processors, also known as the brains at Marvell; Ashish Saran, VP of Investor Relations, and as one investor pointed out to me recently wherever Ashish goes, the stock starts going up; and we have Chris Koopmans, Head of Marketing and you heard Matt talk about him on the last earnings call, how important he is. I'll kick it off with a list of my questions and then we'll open it up to the audience to ask your questions. If you have a question, please press the push button in front of the mic, in front of you, and ask your questions.

Raghib, I am going to start with some big picture strategy questions first. You had a busy year last year. You acquired a couple of companies. You divested the Wi-Fi connectivity business. If you can just talk about the rationale behind those actions.

Okay. So good morning, everybody. We -- a few years ago, like about 3 years ago, we started overall strategy, how should we really pivot our company and what should we focus and what should we defocus and so on. As a part of that strategy, we decided to focus on infrastructure and really move away from consumer. So several things that we have been doing over a year -- over several years is the part of those strategy that how do we serve our infrastructure customer better to serve their needs and the demand that they have and how do we focus our resources and our R&D to deliver value for them. So if you look at 3 transactions that we did, one is Wi-Fi. Overall, Wi-Fi, the majority of the -- though a small piece was focused on infrastructure, but majority of the business was really consumer. So we had to decide whether we add more consumer stuff, so which means we did not have various other components, which is needed for IoT-type of market or the other option is that we divested and really take that money and invest in where we are going. So that was part of it. In addition, our customers have always been asking for the ASIC solution. So Marvell traditionally and combining Cavium as well, we have been doing merchant silicon as well as we have been doing what we call CSSP, customer-specific silicon product, right? So these are the products which are based on our core IPs, but every customer -- in some major markets, they want to add their own -- sometimes their own critical IPs or differentiation to really differentiate their end product with their competitors. So we have been working in various markets, especially the storage and the carrier market, wireless infrastructure, those kind of solutions where we work with the customer, combine their IP with our core products and then come up with customer-specific product. But then there is always in -- especially in the hyperscale as well as in carrier and higher networking, there's a desire to control certain product themselves. So this is the area where customer wants to do the overall design themselves, but then they need a partner to do the overall manufacturing back-end design and so on, which is called ASIC business. So we had that as a part of our overall strategy and when we looked at an option that, "Hey, do we do it organically in-house? Or do we look at options?" So in reality, if you do organically in-house, it is hard to hire the right team. It takes time. It takes burn overhead and so on before it becomes breakeven and so on. When we look at the market, this team, which was -- which is originally IBM ASIC design team, which was part of Global and then they spin out at Avera. They were one of the best team in ASIC industry. They have very reputed track record of delivering quality high-end products over time for infrastructure, and then they were a little bit handicapped as a part of Global, but then now combining with Marvell, they have access back to all the foundries and IPs. So it made perfect sense for us to acquire that and to become part of the solution because now we can offer our customer complete business model all the way from the ASIC side, all the way to merchant. So that was also part of our overall strategy, and it is working out really well. We -- initial meeting that we have had with customers that it is received very well. And then, of course, we were looking at the overall -- our networking side, we had, especially, when it comes to automotive and Multi-Gig, we had good product lines in 100 megabit to a gigabit. And Aquantia had really done a good job in Multi-Gig and the -- all the way to 25 gig. So we looked at it, it made all the sense to combine these -- combine forces out there and make really the winning solution. So this combination actually puts us to be a leader in the Multi-Gig solution as well as in automotive Ethernet connectivity. So that kind of gives you a background of all of our philosophy. Now of course, as you know, that our strategy is really using our processing platform that we have, which we had as a part of Cavium, which is really a multi-core processing platform called OCTEON and combining with these other networking as well as combining with the ASIC, now we are able to offer customer solutions, which are kind of derivatives of all these platforms.

Raghib, the last point you made about the platform and I feel like investors have not fully latched on to this concept that the OCTEON process that you have is being leveraged over multiple end markets. It's kind of similar to what I think NVIDIA did a few years ago, where they used GPU across multiple end markets: gaming, data center and auto. If you could just talk about a little bit more, maybe Ashish you can jump in on how important this platform appeal is?

It's actually -- it's really important because if you think about these are not individual products, right? From a customers' perspective, from a time to market as well as a software investment, once you start down the path of building your products around a multi-core architecture, it's very easy for them to introduce new products in a much, much quicker time frame, right? So -- and we have like real life examples. For example, with our lead customer in base stations, right, they started down the path in 4G with our baseband processor. In 5G, they gave us pretty much the entire digital platform, so not only the baseband processor, but also the embedded processor. And these are all based on OCTEON essentially, right? So from their perspective, it's a very quick turnover. As they look forward in 5G and they needed processing in the radio head for massive MIMO what they did is, they actually took an existing baseband processor, which was designed for a different purpose, but because its program is very flexible, they immediately adopted it and replaced an FPGA. And in parallel, they said, "Hey, look, in the next x number of quarters, can you actually develop us a more purpose-built massive MIMO processor using the same platform?" And it's only going to take us, call it, a year or so to get it done versus the traditional ASIC development or a traditional full chip development is typically a 2- to 3-year process. But the software reuse is a very important part of it. And I think the other thing that sometimes people get confused around this purpose-built solutions that they're not flexible which is actually not reality at all. Take our baseband processor, it actually has multiple ARM cores, which by definition are programmable, and similarly, it has an entire DSP set of cores, which are also programmable. And that flexibility is actually very critical as standards evolve. You don't need to do a new chip every time a standard evolves, right? You actually can reuse the same architecture and rewrite the software. And that's really a big part of why we think we have been so successful. And now as you realize, we have now broken into our second base station customer. And the time to market for their baseband is essentially a year, right, which is amazing because internal ASIC development, traditionally for basebands is a 3-plus year process.

So just to add a bit on what Ashish said. So he gave a particular example of like a baseband processing, for example, but I'll -- just to give you an idea, I'll step back a little bit and explain. So what unique thing that we have developed, and that was part of the OCTEON architecture is, as part of Cavium is, how do we take the general purpose ARM type of processing and DSP processing and hardware accelerators? And how do we combine all of them together where we have enough acceleration for the known blocks, known algorithm, known computing, which could be accelerated in hardware. But still have full flexibility and scalability to be able to address the flexibility needs of the application and the protocols. And I think that was unique thing we did it in -- first, in a security platform that, as you know, that 9 out of top 10 security appliances out there use our platforms. And then we expanded it into networking. And then from there, we expanded into the wireless infrastructure and other application like -- things like smartNIC type of acceleration in a data center or it can be application in what is called the brain SoC in the cars. So if you look at the building block IPs that we have and how do we connect them together combining the networking, combining the compute, combining the DSP capability, combining the security and AI, all those how do you put together and how do you focus on the target application, that gives us the very flexible way to address various markets where we can provide the flexibility, but still provide the performance, which are needed for the end application.

Then just diving into the 5G opportunity within the networking market, FPGA displacements by ASICs have started, as Ashish pointed out, for the kind of the laymen generalists, what are the drivers or the reasons behind the migration from FPGA to customized chips?

So there are multiple reasons. So let me start with the most obvious one. So whenever you have something, which is you want to quickly turn around in terms of changing protocol, and that people have a tendency to do the FPGA designing. But then in parallel, as the volume goes up, you need to translate it into the ASIC. So that is one traditional reason. But the other 2 very important reason is, every application has a box limitation of how much power can you have, power envelope, and within that power envelope, what performance are you trying to get. So even in case where volume may not be high, but if you cannot reach the performance that is required for that -- an application because of the limitation of FPGA, what FPGA can do. Or if you cannot address the power requirement, even if you are willing to pay 5x more cost, you just do not have the solution, right, versus your competitor in end market who has a solution, the silicon base, and as -- is able to achieve the performance target or able to achieve the power targets and so on. So those are the 3 vectors in play in addition to cost. One is the work power requirement at this and within that power envelope, what performance you need to address. And then within -- and then also what volume of the application is there.

Yes. Just adding to that, I think it's almost like every time there's a transition, people forget that this happens every single 2G to 3G, 3G to 4G and no difference 4G to 5G where you start with an FPGA application. But then very, very quickly. Because remember, these are long life cycle programs with high complexity, very stringent power requirements, which, I mean, lends itself to really having a solution, which is built for that particular application. And then if anything, as you move down these process nodes, right, it gets even more complex from a power perspective. The flexibility, and again, an FPGA is, by the way, fairly difficult program. It's on a high level programming language, right? So you actually do need to spend a significant amount of effort to program each of these chips the first time around versus again going back to the discussion on OCTEON, as once you've done the work, essentially on our multi-core architecture, it's very reusable. It's actually quicker for you, right? So I think this -- again, this narrative that somehow an FPGA is very simple to implement versus using the solutions we provide take longer. But in reality, it's the exact opposite. And given the requirements around 5G with more -- why are you in 5G? Because you need more capability, you need more packet densities, you need lower latency. All of those things actually align extremely well with essentially using the type of solutions we provide today.

Just to emphasize on what Ashish said, and type of expertise R&D resources that you need to design FPGA is same as designing ASIC, right? So it's not like you can hire a general-purpose software program or something and they start writing FPGA. So you really need chip designer to really design FPGA solution as well, who understand how the hardware design works and the languages of that, which is called RTL. It's not a standard C programming and so on. So it is also much easier to find DSP programmers or the ARM core programmers versus the FPGA programmer. And the amount of time it takes to design is much shorter in a programmable ASIC versus the FPGAs.

Raghib, if I go back to 2018 Investor Day, the team talked about like $2,700 content per base -- macro base station. You guys approaching low $3,000 content, so you have over-delivered in terms of your design wins and the things that you control, which is the dollar content and your design wins and the things that your carrier customers control, which are the units of the macro base station. So if -- can you walk us through where have the design wins come, which areas? Whether it's embedded or in radio heads, as you've kind of grown your content?

Yes, let me kind of just do a quick kind of background or so we're all on the same page of where we started off in 4G and the journey we had so far, and it is very multifaceted, right? So again, just to kind of baseline everyone, and this is mostly on the Cavium side, right? So Marvell historically was not much exposed on the carrier side. Cavium actually got into the base station market all the way back in 3G, right? It was mostly around some -- our transport side, right? It was a few hundred dollars of content. So it gives us pretty nice stock back then. 4G is really where the breakthrough came, which is Cavium was and still is, combined with Marvell, the only, call it, merchant baseband supplier for base stations, right? And that's where we were able to break into Samsung, supplying them their 4G baseband. And the content went up from, call it, $200, almost 3.5x, right? Call it in the high $700, $800 range, essentially. And that's where you saw, in fact, if you were tracking Cavium, you saw a pretty meaningful shift upwards in their revenue trajectory. It's been a very successful rollout in 4G. The world's most populous country, India, basically uses -- essentially the biggest network is run by Reliance. It's called the Jio network. It's all run on Samsung base stations, which is all running essentially on our silicon, right? So the world's most populous country, most of the traffic, which goes on a 4G network runs on Cavium-based silicon. From 4G to 5G is where we really saw really big step-up, right, and this where we were able to not only get a more advanced baseband processor for 5G, but also locked on all the embedded processors, which does what's called transport control plane data plane essentially along with Marvell where we had now Ethernet switches and Fis, and that's where the content, back to the Analyst Day, we articulated was about 4x higher, right? So now you're in the, call it, high $2,000, approaching $3,000 range. Since then, we've actually had a number of additional design wins, right? First is, again, leveraging the flexibility of the OCTEON platform as customers required additional capabilities, massive MIMO and radio heads. There's more traffic now in the front haul link between the radio head and the baseband unit. So now you need front haul processing. Again, that's something -- anytime we use the word processing, that's something we can do, right? So this is really falling into our categories. That's been pushing the content up towards, call it, low to mid $3,000, right, on average per macro base station. And by the way, these are average numbers, right? In reality, some base station will have a higher number, some will have a lower number based on the line card density. And then most recently, we acquired Avera. And Avera does what is called digital front end. These are typically ASIC implementations, which, again, if you add all that together, the total opportunity available to us, as a combined company, is roughly $4,000 a macro base station, right? And again, if you -- back to unit comment, if you assume -- and again, we are very simple saying, "Hey, if you look at the average number of macro base stations, it's about 1.5 million units a year." Somebody has higher, somebody has lower, but use that as an average number. It's a $6 billion TAM, right? Once it's all 5G a few years from today and off that, we say, let's strip out a 30 -- about 1/3 of the market saying, it's China, it's not addressable through some of the OEMs like Huawei, that's still a $4 billion TAM for us, right? And that's really what we go after. Today, we've articulated -- we have already line of sight to, call it, $600 million-plus of revenue as Marvell and Cavium together. And then if you add another $150 million from Avera, we've got roughly just under 20% market share today. Of course, our aspirations are a lot higher than that, but it's already a very, very significant start we have in the 5G cycle.

Great. That's very helpful. We had Samsung here yesterday. And I'm talking to their networking guy, talking about high single-digit base station growth this year with the focus more in Japan and U.S. in second half. You guys have a base case of 10% type market share, $600 million. And is their view kind of aligned with what you're also thinking about this year?

Yes. Pretty much. I think, again, we're not -- it doesn't matter where you deploy the base station. So regionally, it doesn't matter. But again, we do look at what the rollout looks like. Obviously, Korea went first. And of course, Korea will deploy even more broadly as we get into this calendar year. But equally importantly, and this is really important for Samsung because, historically, they didn't have a significant share even in Korea, the home country, because they entered the 4G race fairly late, right? So now they've got great share in Korea. They are going to be enabling parts of Japan's network and the really big breakthrough is the U.S., right? Because traditionally, most U.S. service providers have wanted at least 3 providers, right? And it used to be in the past where you had a Motorola, you had Nortel, you had Alcatel-Lucent, these are all separate companies. Well, some of them don't -- aren't there anymore and the others got moved into what's basically Nokia-Siemens today. And obviously, the Chinese OEMs are not available, right? So Samsung has a natural position open up. But more importantly, as you're aware, in terms of having the first kind of purpose-built 5G base station, Samsung actually is one of them. So they're actually technologically also more advanced than the alternatives as well as there's a strong requirement for them to be this kind of third provider, right? So I think that's why, I think, Japan, the U.S. start kicking in. At some point, India will also do spectrum auctions, which, again, is a very strong territory, but we are very well aligned with kind of their view of how the market is going to develop. And I think you have to think about these transitions take place over multiple years, right? It took about 4 to 5 years from start to kind of peak. You should expect something very similar. So this is still very early days. But obviously, we're very, very encouraged by the fact that we already started shipping 5G silicon. So we're obviously looking for this kind of very important ramp in the second half of this year.

And then just talking about Avera, obviously, it gets you guys the digital front end you mentioned at one of the 3 carriers. What else does Avera bring to the table?

So as I said that if you look at the background of Avera, they have experience of doing really high-end ASICs. And of course, wireless infrastructure is one angle, but they also, if you go to all the way back to IBM time about 6, 7, 8 years back, they were doing the high-end switches, they were doing bunch of other kind of protocol converters and those sort of a thing. So in the reality, when they were part of Global, they were kind of handicapped because they were -- only those customers they could win, who wanted to do chips with Global. So as they've become end part of Marvell now, so they can use TSMC, they can use Samsung, they can Global. So it's giving them full freedom. And then, on top of that, it also gives them access to the Marvell IP. So the latest process node that investment foundation IP that we have done, 7-nanometer and 5-nanometer and SerDes and some of the other like PCI, DDR controller and so on. So in other words, they do not need to do all IP development themselves anymore. Before they had to do everything themselves because, of course, there's no general IP provider for GLOBALFOUNDRIES as such, right? So it gives them a lot of freedom now, a lot of flexibility to be able to go and win. And as I said, it's only been like 2 months right now, so we are in early stages, but we are going -- we have targeted the top customer in the hyperscale data center, in the high-end networking, in the carrier as well as automotive. And we are going through this road show. We have gone through 5, 6 customers so far and reception that we are getting is really great because market is really looking for some capability and provider who can really address their needs for high-end ASICs and to be a partner. So that is -- we are being welcomed very well. So I expect we will have design in each one of these segments working with -- as a Marvell ASIC offering.

Atif, I think there's a couple of other angles, which are pretty important as you look at ASIC. In fact, I'll let Chris make some comments there. One is the size of the market is actually quite large. The second is, it's not -- no longer this very traditional all-ASSP, all-ASIC market, but I'll let Chris add some comments there.

Yes. Sure. So in general, the market has been -- you either decide to build your own chip or you don't. And as you know, Marvell is in the infrastructure business and infrastructure OEMs tend to do both. They tend to build some of their own and they tend to buy some on the open market. As Raghib mentioned, Marvell and Cavium have typically been in the open market piece, but increasingly, they want to be able to do something in between. Increasingly, they want to be able to say, "Hey, maybe I don't want to do the entire chip myself." And so our strategy with ASIC is not just sort of black and white. Do you want to build your own? Or do you want to buy something off the shelf? Or maybe what you want to do is take our OCTEON, for example, and add your baseband into it or add some of your other secret IP into it to be able to differentiate your base station as an example or any other products. And so we're really looking at this as a full spectrum from completely off-the-shelf to completely custom. And we think that a number of the OEMs are going to want to play somewhere in the middle and evolve their business model over the next 5, 10 years as the time to develop and the cost to develop this technology continues to increase as you go from 16% to 7% to 5% to 3%. They won't be able to completely develop the whole thing themselves and there's no need to. They'll actually get a better product, if they take some of it that Marvell has already built and hardened, and selling the same ARM complex, for example, into firewalls, into routers and into base stations not without having to reinvent the wheel at every generation. So really, our strategy was not just, "Hey, now we can do ASICS." Our strategy was, "Hey, we're going to be able to develop custom products for our customers to deliver exactly what they need in a differentiated way and become a much more strategic supplier to them over the long term and letting them differentiate their products over time." And we think that's particularly important because one of our most strategic assets is our OCTEON processor complex. There's really nobody else on the market that has that. And so being able to offer that in a custom fashion versus just an off-the-shelf fashion so that they can -- it will find its way into way more applications. And we'll start to take away from the FPGA market as well as the full custom ASIC market into this sort of custom OCTEON market because of the power of the OCTEON platform.

And we do see all these high end, the hyperscalers or higher networking, they will all have their engineering team. So instead of having it as black and white, either you use my merchant or you do your own ASIC and really seeing as there's a competition with each other, what we are -- we have developed business model where we kind of see it as a partner. How can we work as extension of their team as a partnership? And having this whole business model is spectrum of shades of gray, I would say, from ASIC on one hand, pure ASIC or the pure merchant on the other hand, and we can play anywhere in between. So that actually -- that whole business model offering is being received very well.

Then moving on to the non 5G part of the networking market, the enterprise market has been weak. Your biggest customer, North America, has seen weakness. When you see this business stabilizing this year?

Sure. Yes, so I think the weakness is all really related to the macro impact from really the U.S.-China trade tension. I mean, that's what it boils down to. Because there's no other general weakness from a worldwide GDP perspective, right? It's all based on businesses are not confident about where should they be investing, right? Who can they sell to, who they can't sell to? So the natural reaction is you cut back on the things you control, which is your spending, right? At the end of the day, people like us get paid when companies go spend on IT, right? So that's what's been cut back. There have been no change on our design win side. In fact, we've actually added more design wins through this kind of down pay from a revenue perspective. If I stand back and look at the 3 kind of big chunks of my business, storage took -- was the first one to start to come down, which is not unusual. It almost always does. In fact, it was -- it kind of bottomed out in the first half last year. Q3 grew quite nicely. Q4, we're expecting it to be flat to up. So storage has kind of made its way through the pipe and saying, "Hey, it's actually growing again." Compute, which for us is really our fiber channel business plus some additional connectivity, took a pretty big hit back in Q1, Q2. Again, started to show signs of recovery, again, right? So I think 2 out of our 3 businesses are now showing signs of recovery. Networking was the last one, right, to show some declines and, in particular, with our kind of our only 10% customer back in Q3, which is Cisco. And I think, well aware, you heard their public call as well, where, yes, I think they certainly saw IT spending start to come down. We've anticipated similar weakness in our Q4. And we are, I would say, cautiously optimistic that we've seen the worst of it. And the reason I say that is because the root causes, again, the U.S.-China trade relationship, where they've at least been the first steps towards some form of a deal, whether it's the Phase 1 today and -- or Phase 2 at some point later. So I think that's what gives us some confidence. And in the meantime, like I said, we've actually added design wins which have not gone to production yet. In fact, the enterprise refresh cycle, which had been anticipated for a long, long time, I think, only ran for about a year-or-so before all the trade tensions kicked in and everybody kind of went back and halted most of those purchases. As an example, take Wi-Fi 6. About 2 years back, right, the view from the industry was the year we just finished would have been a very strong year for Wi-Fi 6 adoption, right, which would have driven a big upgrade cycle on the campus side with Multi-Gig in particular, which we are very, very well positioned for. We have won those key sockets. But because of macro, right, Wi-Fi 6 is seen in most cases as a -- it's an important upgrade to drive over time for good connectivity. But when you're tightening your belt, you can kind of hold off that, right? But at some point, you will need it. And new access points, typically for Wi-Fi 6, typically are specified with Multi-Gig on the Fi side. And if you do that on the Fi side, guess what, you need a Multi-Gig switch as well to take advantage of that advanced capability. So yes, we're looking forward to kind of the refresh starting again, which will add on top to 5G. We have some custom flash work we're doing, so I think we have a number of very exciting product cycles, especially in the second half of this year.

And then just moving on to the data center opportunity, it just feels like there has been increased momentum with ARM server adoption. NVIDIA made a big announcement with you guys at the Supercomputing Conference last year. Amazon announced the second-generation ARM-based CPU at AWS re:Invent in November. What has driven the kind of increased momentum in the ARM server market?

So I would say it's a constant improvement or constant growth in the momentum. It's nothing like something happened in the last few quarters. This thing is being worked for several years. And I think the primary driver for most hyperscalers are -- I mean, it was perceived as if it is an alternate of Intel which is driving, but it's not really that only. They really see that as one of the option that's having a whole ecosystem and their application is running on ARM to give them a little bit of a freedom, that if they ever want to build themselves, they can build their own processor as well as they will have more influence on doing more, what I should say, application-optimized processors, which they do not have freedom to work with the companies like Intel and so on. So -- and then the third thing which drove them also the concern of the security. You know in the last few years, there were several security issues came up. And the frustration in these people minds was that they did not have visibility to how these architecture works and so on, and that kind of made them also think that they need to have a better control on their destiny. So those are the -- some of the driving factors that have been driving companies like Amazon, Google, Microsoft and all these that they want to have ARM-based solution. So I would say, as we have always said, it's a long play sort of a thing. It has made a lot of progress. I mean, the Microsoft is deploying in Azure, which is a huge validation of the things that it is production ready, not only the hardware, but the platform -- not only the chip but the platform as well as the firmware buyers, windows software and application stack and everything. So because everything has to work and everything has to work in a standard way, the way servers are supposed to work. So yes, we have come a long way, but there is still, as we know that it takes time to ramp in the volume. So we are expecting in the second half those things will ramp. But apart from that overall hyperscale part of the industry, they are very much interested in ARM servers and it's making progress.

Let me pause here and see if any questions in the audience. Okay, I'll keep going. Not to leave storage behind. I wanted to ask about what's the long-term view on the growth rate? And when should we expect storage sales to stabilize. We're starting to hear about near line growth in HDD market, also SSD growth seems to be improving in the enterprise side. So just the puts and takes in the storage market?

Yes. I think, first, I would say the storage market has already stabilized. I think the low point was really first half last year, right? Q3 was up mid-single digits sequentially. So we started recovering there. And then Q4, like I said, is going to be again flat to up again. So I think storage has certainly passed its downturn. As we look forward, we're still confident in what our long-term strategic outlook was, which was once the business is stable, it's still capable of driving, let's call it, low single-digit growth. And that's through basically 2 end markets: one is on the hard disk drive side, as you mentioned. 3 or 4 years back, we had almost no position in near line drives, which is these high capacity drives for cloud customers, which is really the only growing portion of the HDD market, right? And that we have now 2 out of the 3 customers very well positioned with 16-terabyte drives. We actually have not just our SoC, which is the controller plus rechannel, but we also have dual. So 2 preamps actually in the Toshiba 16-terabyte drive, right? So the content is actually very, very appreciable. And we expect that market is going to continue to grow over time. And obviously, these platforms are going to scale from 16 to 18 to 20 to 20-plus terabytes over time, very well positioned there. In addition, we have also been making a pretty big shift on our SSD strategy, right? So if you go back 5, 6 years, most of the growth in SSDs was really the conversion on PCs from HDDs to SSDs. Most of that's done at this point in time. And we saw this coming about 2, 3 years back, recognizing we've ridden that initial kind of growth curve, like the market has. But over time, the PC market for us, it is going to go x growth. So where you continue to grow in this market, which is enterprise and data center. We also recognized a very big shift, a paradigm shift in how people view flash. So HDDs are viewed as permanent storage, right? You store information, you leave it on there for a very, very long time. Flash, in most cases, especially in enterprise is an acceleration layer, right? It lets you exit an application, which means you can lose -- use fewer CPU resources. And if you take that next step forward saying, okay, if that's the case and if I can tailor my controller to that particular application I can get even better performance, right? And another side benefit for the system OEM who is consuming this is, I don't have to tie myself to one SSD controller and one SSD OEM essentially. So they're disaggregating their SSD, and they're coming to people like us for the controller, which is very optimized and they can multisource their flash from pretty much anybody they want. The first major design win, which we actually won about 2-plus years back is completing development as we speak. In fact, it's going to go into mass production in the second half of this year. So those are the 2 big reasons why we feel very confident that we can sustain low single-digit growth. And while the growth number is not that great, but the investment profile is fairly lean in storage, we can drive very significant free cash flow from this business for a very long time period.

And then auto opportunity, it's a number that's not in my model or Street model. Just to talk about the Ethernet opportunity in autos, you're designed in Tesla cars currently. What's your average content over the next 2 to 3 years? And how big is this opportunity?

Yes. So I mean, we -- if you just take a step back and look at the size of the opportunity, we can kind of conservatively say, over the next few years, we think it will come up to be about a $500 million opportunity. And we started the automotive business about 3 years ago in Marvell, built the 100 meg and gigabit Fis as well as gigabit and multi-gigabit gateway switches. And with Aquantia, we're now the leaders in Multi-Gig. So we feel like we have established a leading portfolio in this market, and we have a multiyear head start, and we feel like we can be the #1 position as this market continues to grow. Of course, I think the market will get even bigger over time. If you look at the very long future of cars and you talk about self-driving and autonomous and all these types of things, the number of Ethernet ports in cars could start to rival that of campuses and other types of markets. But in the near term, we're really sort of conservatively saying, hey, Ethernet's being used for infotainment links, for camera links, for connecting the radio to the head unit, for replacing slower old legacy technologies like MOS in LAN with 100-megabit, for example, and moving the car from legacy automotive proprietary types of network technology to modern Ethernet network technology, and then it will only take off from there. So we're really excited about the opportunity. I think you'll start to see in the second half of this year, model year '21 cars with significant Marvell Ethernet content. Beyond, obviously, the Teslas of the world and things like that, but mainstream vehicles using Marvell Ethernet technology in the second half of this year.

Yes, I think that's the exciting part because it's not. I think some people have this view that Ethernet will only take off in cars once you get fully autonomous Level 5, Level 5 plus. That's not the case at all. It's happening today, it's happening in mainstream cars, right? And that to me is the exciting part. Yes, there will be Level 5 at some point in time, and those will have $100-plus of content. But what's more exciting is if you get into the 10s and 20s, right, in mainstream cars, that's a lot more volume there, and that's starting to happen.

Awesome. We're almost out of time. Thank you, guys, for coming to the Citi Conference.

Well, thank you. Thanks, Atif.

Thank you.

Thanks, everyone.