Marvell Technology, Inc. (MRVL) Arete Tech Conference (Transcript)

Marvell Technology, Inc. (NASDAQ:MRVL) Arete Tech Conference Call December 6, 2022 12:45 PM ET

Company Participants

Raghib Hussain – Chief Operating Officer

Ashish Saran – Senior Vice President of Investor Relations

Conference Call Participants

Brett Simpson – Arete Research

Brett Simpson

Okay, thanks everyone for dialing in. It’s Brett Simpson from Arete again. Now our next session is our last of our tech conference but the good news is we’ve left the best to last. So I’d like to welcome Raghib Hussain, COO of Marvell Semiconductor. We also have Ashish Saran, who as you all know, is SVP of IR at Marvell. So Raghib, Ashish, thanks for coming on today. Really appreciate it.

Raghib Hussain

Yes. Thanks for looking forward to this.

Brett Simpson

Great. And as most of you know, Raghib was Co-Founder of Cavium and has a long history in the networking space. And then we’re definitely going to cover off some of those areas in our discussion over the next 45 minutes. And it’s also quite timely that we’re having this session right after Marvell reported late last week. And I think we’ll also look to Ashish for a few questions around the sort of what — how to think about the business post quarter. So maybe by way of starting out, Raghib, maybe you can start with a quick intro and spend a brief period talking a bit about the strategy at Marvell post all the acquisitions you’ve made, how the portfolio sets up, particularly in comps, in data center and autos more broadly.

And then we can maybe talk a little bit with Ashish about what are the sort of points to highlight post results and what’s the that for you, Ashish.

Ashish Saran

Sounds good, probably, maybe you can lead off with our data infrastructure strategy and I’ll take it from there.

Raghib Hussain

Yes. Okay, thank you. Thank you very much. My name is Raghib Hussain. As you introduced, I was a co-founder of Cavium. And then, of course, since we made Marvell, I think Marvell is driving product strategy as well as business selection and so on.

I’m Vice President of Products and Technologies at Marvell. In Marvell, we a few years ago, about 4 to 5 years ago, we did actually put together a strategy to focus on data infrastructure. It’s a very well thought through strategy because we did realize as data volume is growing, it has to drive the kind of build-out of the data infrastructure. Because if you really want to drive the value out of this data, the economy, data economy will be driven by how good of a scale do you have in the data infrastructure. So obviously, it had to be done at the right efficiency levels and so on. So which means you need to optimize essential silicon products to enable this data infrastructure, right?

And that is the vision that we had and that is what we have been driving for the last 4, 5. And of course, to enable that, you have — we recognize that you have to be able to handle all aspects of data which means to be able to store data, move data, process data, secure data which means that the components which are needed are really the networking high-bandwidth interconnect, the storage, how do you store data, data-centric compute as well as security. So those are the areas we focus on and we made it a goal that we not only want to be a technology leader in each one of those areas but we also want to have our market-leading products in each one of these areas.

So over the last 5 years, our — all the things that we have done internally, or organically or inorganically are all focused with that mission to achieve that goal. So that’s why all of our M&A had a focus. It’s not like a random okay, whatever is available, go get it a good deal. We never looked at it. We came on with the plan that say these are the areas that we are good at. We are going to invest organically. And these are the areas that we need — maybe we need to augment through M&A and that’s exactly what we did. So today, the reason I’m so excited and the reason I’m kind of still driving here and I believe Marvell can be very fast-growing company in driving data infrastructure because data growth is happening, it’s still happening and is happening.

Actually, the rate is increasing, right? So if you look at the overall continued long-term model, data infrastructure will keep growing at a much faster rate compared to any other segment and especially the markets that we focus on which is data center, carrier, automotive and enterprise. So out of those 4 markets, 3 markets are growing at a much faster rate which is data center carrier and automotive. And in enterprise, we are sharing — gaining shares — so as a result of that, our growth is good over there. So another key item of indices is that a lot of fundamental building blocks needed for each one of these markets, products needed. We can easily share the investment across various markets, right?

So that is why I believe the choice of our end market that we chose and the target market based on the size of the market and the growth of the market. It was very well thought through strategy. And we have been successful in executing that strategy through our organic and inorganic investments. And that is why I personally believe I’m excited about it that I believe Marvell is the most, well-positioned company to provide the essential semiconductor needed to enable this scale in the data percent.

Ashish Saran

Yes. Perfect. Great introduction. Maybe I’ll just add 1 kind of a couple of minor things there. From an M&A perspective, not only do we acquire kind of the right companies, these are basically all kind of the number one market share leaders in their space, right? So it wasn’t just about we need an asset. We actually got essentially the best assets out there. And if you this thing from a building block, as Raghib mentioned, if you are a data center, what do you need in compute, you need acceleration, you need security, you need storage, you need high-speed networking, right?

All those building blocks are now one of the same roof and with the acquisition of Avera which does custom silicon, I think this is where we really have a unique differentiator, right? It’s our ability to mix and match our IP along with our customers and give them something which is very unique but much faster time to market relatively lower cost and much higher probability of success. I think those are some of the things that Raghib will touch upon later when we talk about our cloud optimized program. In terms of near term, I mean, yes, we just our Q3 results, we guided Q4. Our first actual results were very strong. We grew revenue 27% year-on-year.

Now clearly, we’re heading into an environment where clearly macro pressures are in play and semi industry is coming off of very high growth here, including us. So we didn’t guide down our Q4. But if you stand back and even at the midpoint of our Q4 guidance, only our revenue growth is still going to be a very impressive, north of 30% year-on-year, well above our long-term target. And when we look at really what we are seeing in the near term, it’s — most of what we are seeing in Q4 is really an inventory correction primarily on the storage side, right? Within data center, it’s hard to strive and its SSDs. This is an area where we supply to our OEMs and then they supply the end customer. So visibility is probably not as good and we were — I would certainly say a little bit surprised by the amount we are seeing — correct itself.

The good news is, instead of kind of kicking the can down the road, let’s just kind of do this as quickly as we can. So expect that to kind of bottom out in the next couple of quarters. And as we look into next year, what’s really nice for us is 1 is we see a few quarters of inventory correction but all our key growth markets are still very much intact. And these are fundamentally growth markets. These are not markets which grew outside through the pandemic and now basically go ex co is quite the opposite, right? This has fundamentally secular growth in front of them. On top of that, we have a number of, call it, somewhat market independent growth drivers. We have a number of cloud optimized silicon designs which are now already going into production and will ramp to an aggregate of $400 million of revenue next year.

The 5G business is doing very well. We expect that to be a strong growth driver next year. Our automotive business just hit another milestone, we crossed $200 million annualized in revenues in Q3. So that’s going to keep growing. So overall, I would say we’re set up extremely well. As we get into next year, like I said, we’ve got a few quarters of, call it, an inventory correction to manage through. But then beyond that, I think we’re very much back on the firm beyond the growth at.

Brett Simpson

And before we get into the sort of nuts and bolts of the strategy, maybe just 1 follow-up, Ashish, is China. Obviously, there’s been a decoupling and some concerns building about the appetite for China semis demand, particularly for U.S. companies. But can you maybe just frame China as I think you talked about it more in terms of the size of the business and it’s clearly weakening near term. I think you mentioned 30% for Q4. But help frame the sort of China situation for Marvell and how you see it playing out?

Ashish Saran

Absolutely. I mean China is clearly a very large customer for the old semi industry. I think we have the consumption numbers. We’ve seen somewhere here, 1/4 of semis actually did come. We have a much lower exposure not because we don’t want but just happens to be without advanced technology platform, we’ve seen much more of an attached to it, I would say, to more kind of big U.S. hyperscalers, so big multinational networking companies. So our average exposure across the business to direct China OEMs in Q4 is going to be below 10% revenue, right? And even when taking a higher number, call it, a few quarters back, it was still probably in the low teens, right? So first point is China is important, it’s call it a little more than 10% revenue in general but it’s now a big part of our business.

We did see softening starting in Q3 and accelerating into Q4 from China, clearly, I think in response to some of the macroeconomic conditions they are managing through. And most of the impact we saw is in our Enterprise Networking business, right? So I would say at this point, I would say you should assume China is kind of, to some extent, bottomed out. I think it’s slightly [indiscernible] at this point. And I think in general, you should assume China as a percent of revenue is going to be somewhere in this, call it, 10% plus or minus range. This is a great question because I think a lot of people look at our Ship 2 number which obviously is a very high number but we realized most of that just gets into and stem back around and comes like fairly low.

Brett Simpson

Okay, that’s clear. That’s clear. And I mean, maybe for Raghib, I guess there’s been some concerns building amongst investors around the growth study in Marvell and I appreciate the near-term inventory correction. And a lot of this is centered around storage. But is there a risk in your mind as we go through this downturn that this could spread into other parts of the business? Do you — when you look into the setup for 2023 or even 2024. Are you thinking differently about some of the opportunity sets in front of you? And specifically, when it comes to like hyperscaler visibility, how do you feel sitting here today in December, looking out into ’23 and beyond?

Raghib Hussain

So as I said earlier, the — if you — if you look at the multiyear view, their model itself, the trend is upward. The overall investment that these hyperscalers are going to do and planning to do will keep continuing. There may be there’s an impact of inventory digestion or some adjustment of the year-over-year CapEx and so on. But if you look at the overall trend line, that has to be — that is upward. And the reason for that is if you look at really, these cloud guys, their model actually is to — still there’s a huge amount of business that we need to grab, like especially the hosting. So there are 3 big hosting guys right there.

I should say three big guys focusing on hosting, let me put it this way, right? And the goal of each one of them to land as many customers in their cloud as possible. And one of the key driver of the value proposition of driving this is to keep driving the efficiency, right? How much — because we all know the performance of the cloud is limited by the capacity and the three aspects of capacity the bandwidth. — which is the number one by the way, because the performance of the system is dependent on the bandwidth and then the compute and a lot of time which is not that much paid attention to is the storage. Most of the cloud guys actually measure their customer in terms of storage because when there’s of sales, okay, I’m bringing this new customer, the question asked to the sales guys is, okay, how much storage are you bringing? Because that is their sticking the way to have a stickiness on the customer by having more of their data in their cloud.

So to provide the value proposition to their customers that why they should move, they have to provide the lower latency storage, they have to provide the high bandwidth capability and so on. And as a result of that, they cannot just take the decision — hey, just because something is happening in the industry. I will not switch to, let’s say 800 PAM4 or 1.6T. Because whoever makes that decision actually probably going to save a few hundred million dollars but probably going to lose hundreds of billions of dollar business in the next 2, 3 years because once you are benign, you are behind. So — and in terms of how much money is needed for this investment versus how much money these big CSPs have does not make any sense to let somebody else take the lead, right?

So that is — if you look at that — how the technology is transforming and how these needs are actually implementing efficiency the going forward trend is positive, right? So that is why I say if I look at the — take a long-term view, I am very optimistic and very bullish actually about it. I do not think it is — these things are going to slow it down.

Ashish Saran

In fact, just to add to that, what we’ve seen and to some extent, right, there’s a focus on maybe some ROI in the short term from some of these customers, actually, the push for optimization is uneven higher, right? So our design win opportunity funnel has actually gotten bigger, quite frankly. I think their impetus to say we actually want to move to more — I mean you’ve heard one of our — one of our — 1 of the large cloud customers had a big conference last week. And if you noticed right, a lot of what they talked about was more optimization, more ability to really improve the total TCO. But overall investment, the message with the overall investment will keep increasing because that’s what drives the revenue engine in the future, right? So even start as to your specific question, while yes, it’s going to go through an inventory correction. Once we are through it, yes, our anticipation is it’s going to go back to growth within data center because there are alternatives side, the total storage capacity, nearline is still going to be critical — for acceleration will remain critical. And then we can talk later, we actually have some very innovative new architectures around CXL to help them even more optimize. So I think we’re in the driver’s seat and helping these customers get to their long-term goals.

Brett Simpson

We’ll definitely circle back to CXL a little later. But I guess, as we start to look at coming out of this period and we start thinking about recovery. Where do — where does Marvell have undisputable kind of leadership positions today? And where do you see the share gains are going to be most pronounced over the next couple of years? And what are the product areas do you think that Marvell has a real opportunity to scale up in where you can take that individually?

Ashish Saran

Yes. I think first and foremost, as you know we’ve got market leadership by far within the optical connectivity space, right? When it comes to electro-optics in the data center through the Inphi acquisition and we essentially added more resources right after we acquired them. So very much expect that — as Raghib mentioned, 800-gig PAM4 [ph] is already being used in volume in AI applications. And as you go into the next generation of switch architecture, we’ll be driving the 1.6T terabyte road map for the industry essentially. So this is an area where I expect to continue to lead.

I think we’ve already made major strides right in the 1 area where we did not have a great socket was in data center switching. But with the acquisition of Innovium, right and our ability to now get more substrates going forward, you should expect this is going to be a big growth area for us, right? And obviously, we got hard at work at essentially coming out with the next generation of high-speed optics in the companies with silicon. So I think you should expect us to make some pretty big leaps, I think, on the switching side. I think when it comes to cloud optimized silicon, this idea of basically customizing our IP to our customers we saw this, I should really get further to Raghib and Matt, who saw this right after the Cavium acquisition.

And I think we put our foot down and I think we’ve got a massive lead on everybody else. I mean you can talk a little bit about it. I think it’s very important to understand that customization is not something just wake up 1 day and say, “Well, I can build a standard product. I think a lot of this credit goes back to the original design of OCTEON. So let me pause and I’ll let Raghib talk about that flexibility and then we can talk a little bit more. But I think it’s a very important point to convey.

Raghib, if you can talk a little bit about the flexibility of that platform?

Raghib Hussain

Yes. So I think a lot of people — even a lot of customer went that route and learn hard way and came back. So it sounds like, oh, we can get ARM core from [indiscernible] and we can just put them together and we can put together some and put a kind of data centric processor. It’s easy to say and very complicated to get it done. Because how do you really implement the interconnect, how do you do the balancing of resources. How do you allow the sharing of the — a lot of common resources in which requires a lot of tuning and a lot of expertise that we’ve developed over the last 20 years.

And that is what is the kind of unique aspect of the platform that we have which we call — call it, OCTEON platform, DPU platform whichever platform you want to call it because there are different people, people are calling it differently. But in the reality, it is a platform which allows you to provide what we call the data-centric compute. And what is data-centric compute? It is a compute which allows you to manage, handle a lot a high volume of data to be able to take in a high volume of data, process in real time that high volume of data based on the application needs of processing and then manage it means move around and store properly and so on. So this whole platform needed to process the high-volume of data requires a very good system which has a very well balance of a general-purpose compute, like our own type of compute plus the kind of some more specialized type of process like DSP and so on.

And then, some more specialized blocks had — more of a hardware abstraction blocks, right? Things like car compression, crypto or AI processing, a bunch of other type of video processing and so on. So the combination of these blocks, how do you design those resources? And how do you allocate the share — because they all need to share the common memory, for example. They all have to share various IOs and so on. How do you implement the system that it is a very balanced system and gives you the most efficiency out of the full SoC. This is where the unique expertise and IP lines, okay? And this is where you can bring a lot of differentiation.

Now based on the target application, you can tailor this combination more focused to, let’s say, video processing or more focused to, let’s say, AI processing or more focused, let’s say, implementing disaggregated historic solution, right? or maybe more focus towards security processing, right? So these are various kind of data-centric processing and application SoCs that you can develop. But there is 1 unique commonalities in all these things how flexible are your interconnect of sharing these resources, right? And how well can it scale across because in a typical SoC, there are hundreds of engines running there, right? There are general purpose core, a special purpose core, micro engines, hardware engines, all that.

They all need to share these reporters. How do you scale as you keep adding these capabilities, how do you scale to the higher and higher bandwidth capability to process higher bandwidth. And that is what is unique about this capability that we have that we developed over the last 20 years.

Ashish Saran

Perfect. Thank you for that very much. So Brett, maybe just finishing off a few of the other growth drivers and then we can go deeper, if you’d like, in any of these. I think I can just say going through our end markets. I think on data center, obviously, you have opportunities in CXL which I still think were leading the market and developing new solutions, we actually announced a very big design win on the call last week I think in 5G, you know the story really well. I think most investors do. So you know we’ve been leading in that space. We’ve driven massive revenue growth. We just announced an ongoing extension of our collaboration with Nokia right. This is the same OCTEON 10 which Raghib just talked about our DPU platform. I think in automotive, our Ethernet business continues to win design wins. We crossed an important milestone well on our way to be kind of more than half the market.

So I think got a number of IP growth engine and a number of these that will drive growth even next year, even despite being a challenging macro, in my mind, 5G, cloud optimized, automotive. These are all going to drive year-on-year but despite what is probably going to be a tough economic environment.

Brett Simpson

Yes. And do you still see Marvell growing the business in 2023, just given all the puts and takes that you see at this point going into 2023?

Raghib Hussain

I think we’ll see what actually happens. I mean, the reality is it’s a tale of 2 halves, right, the first half probably the industry is 1 where there’s obviously going to be some inventory renormalization and that’s going to have a pull-down effect. On the other hand, we know we’ve got these multi-hundred million of incremental revenue to our drivers. So as we get through the end of the year, obviously, the year-on-year growth will start to become meaningful towards the end of next year. It just depends on whether they completely neutralize these other, we end up growing a little at this point to be completely honest.

I think as Matt said on the earnings call, our expectation is we’ll be able to hopefully drive some growth. We just don’t know quite frankly at this point far out. But I think the setup for growth coming back and accelerating as we get later into next year, I think that would look quite good.

Brett Simpson

Yes I wanted to talk a little bit about content growth drivers within business, particularly data center. And I think you were talking earlier, Raghib, about bandwidth and data speeds and feeds are going up significantly. We’re starting to see servers embrace PCIe 5. We’re going to soon see PCIe 6 coming and Ashish mentioned, obviously, CXL and the standardization that’s going to come over the next 2 or 3 years. When you sort of look at these generational jumps and probably do the same with PAM4 DSP versus NRZ and we could go into various different levels of speeds and feeds going through generational jumps. Can you just talk a little bit about what that does to Marvell’s business?

The content sets within these generational jumps. And whether there’s a consolidation play that’s also going to arise from this — maybe we can start with the PCIe landscape because this shift to PCIe 5 and then PCIe 6 feels like it’s going to be a good growth driver for Marvell.

Ashish Saran

So definitely. So first of all, as you go from the lower bandwidth to the higher bandwidth, the number of players that can bring the solution will keep getting reduced, right? So that is 1 angle that puts Marvell in a unique position because we — at this point in the industry, we have best technology, whether it’s optical and copper in high bandwidth connectivity, right? Secondly, all the customer to drive the right bandwidth, they need to transition to a newer solution, right? So in other words, let’s say, if they had their own standard product, they need to switch the product which support PCIe GEN6 or they have the custom, many of the DCI speed actually have custom solution. So they need to switch to the newer custom solution with these capabilities, right? And this is where it is a great opportunity for us because we have actually invested in these areas. And combined with our other custom capabilities, it puts us in the best position to gain more design and more market share over there. So that is on the custom side. So there’s a huge aspect of how many people are capable of that and how many people can provide that custom solution with stability.

And then coming to the standard product side, if you look at the — on the connectivity side, when it comes to high-speed copper as well as the optical both connectivity, right? So we have the market-leading position, right? So as these — especially, let’s say, PCIe Gen6 or even the Ethernet connectivity closer to server and it goes to the higher and higher bandwidth. The need of these specialized product to be able to handle the proper signal strength and all that will keep growing, right? So whether it’s the PCIe Gen6 or whether it’s a higher bandwidth like 100-gig type of Ethernet you need a different type of, for example, re timer product or other type of connector we’re going from an RD to, let’s say, electric optical or are on the AC type of connectivity and so on. So all those things drive for actually going in the direction which is our core strength, right? And that is where our opportunity is to gain more market share and that is the growth — the share growth is going to be driven by a lot of those things in the data center connectivity, if you say.

Raghib Hussain

At abstract is to a high level, very simply put, when you have more bandwidth, you need more speed but you need more speed. The link speeds go up pretty dramatically. At certain point, the link can be a simple analogy. You need to manage the data. It’s the whole idea of data processing, like the data, moment you need to do that, that’s where our strength comes in. And you can look at these examples in multiple end markets, right? If you look at why is automotive market adopting Ethernet, it’s because those speeds are going to the point where point-to-point analog simply doesn’t work. So you need more advanced technology. If you look at the data center, you look at kind of the simplest numbers link which has been out there for a long time is the switch, right, top of rack switch to your silver connect.

At some point, you are already starting to see where you need that asset copper cable is not going to work anymore, right? You need [indiscernible]; that means DSP technology and guess who’s the leader of PAM4 DSPs, right? And there’s other examples. But I think very simply, as you think about bandwidth growing, link speed growing, you need to manage data, if you need to manage data, this is a line for you. So this is exactly what I’ll give is on to talk about.

Brett Simpson

Yes. Maybe we can change a little bit directions to talk about switching. And maybe starting off with enterprise switching and it’s been a great year actually for Marvell. I think you’ve taken quite a bit of market share and it’s been a good market conditions, if you like. Can you talk a little bit about what’s happened in the last 1 or 2 years in enterprise switching and had tremendous growth this last couple of quarters. How much of that was the market versus share gains? Where do you think your share is today? And I know we’re going — maybe we’ve overshipped into that market to some extent, the growth maybe isn’t going to be as dynamic going forward — but can you talk about some of the drivers around the upgrade to 10 gig switching or multi-gig switching in enterprise?

Raghib Hussain

Yes. I mean the story really — I mean most of the revenue you see, Brett, gets shown up in what we call our enterprise networking end market. And the story here is not just switching. It’s really switches as well as the PHYs, both are very important elements. And this really slowly really started 5, 6 years back. In Raghib’s opening remarks, he talked about while we did a lot of M&A, we had a lot of self-help as well, right?

And one of the biggest self-help areas really within Marvell was taking this Ethernet franchise which, quite frankly, under the Prime management had kind of all look sideways be as polite as I can. And so the first thing we focused on is saying, look, at the end of the day, there’s still only really 2 large companies in the world which are known for doing high-quality switches and fibers, still essentially a very small market where customers want choices. And clearly, Marvell was viewed as 1 of those but we need to invest and we did that, right? Takes — when you go fix the franchise, it takes about 2 to 3 years in semi. So it’s only the last couple of years, to your point, whether the benefits of those things are to show up.

Now it only very started to show up last year because our customers have given us new wins but they have made the new platform change yet, right? Because the whole networking market has been a little bit depressed if you remember the last few years. It’s only last year, you saw, hey, all the design wins we had started to come through, right which are in play today. Our customers introduced their new platforms, right, as the world now shifted to a hybrid working environment, handling video at a much higher bandwidth speed, at much higher security requirements, right, because you have users from multiple endpoints coming in.

I think it’s a combination of those things is what drove the strong revenue growth. So I would say share gains in switches and PHYs along with on gains, better, higher ASP switches, multigigabit PHYs, the shift to WiFi 6, a shift to more video within the office itself is what’s driven a very big increase in multi-gig. And the multi-gig PHYs, I mean, this is — it’s a pretty high ASP product. I mean the ASP is not few percent higher than gigabit. It’s multiples higher, given it’s a very difficult problem to use existing cabling but get 2.5, 5 or 10x of bandwidth. It’s a very difficult signal processing problem. So we get a lot of share there. So overall, I don’t know if anyone follows the market shares anymore, so I don’t have a public data point to give you — but you can see that we are driving, call it, on average, 20% to 30% year-over-year growth for the last few years, very recently even higher than that in that which has very low single digits, right, almost by definition, we’ll be gaining share.

In terms of over shipping, I would say this is a case where I quite [indiscernible] inability to get supply probably helped us. So it’s kind of difficult when our customers are still escalating even a few weeks back, right? So we probably were one of the “Golden screw. ” So I think the — unlike HDD, where obviously, those parts were easy to get. So we’ve been able to shift to whatever they ask us for. I do think that as you look into next year, we should be conservative to assume that the end market, China has already come down, as I mentioned earlier. I think our expectations are, our customers still feel very good. But we’ll be cautious, I would say. But we still feel good that we would expect to still see some share gain next year. Multi-gig PHYs will continue to increase as a percent of the total market. It still is very small number. They actually still on in the single digits, right? So there’s a lot of headroom here.

So we will not agree year-on-year next year should be a more kind of slow year, right? But enterprise networking but I think fundamentally, our position is very strong. And, networking is 1 of those ubiquitous technology. It’s not talked about but every shopping mall, every hospital, every data center, every enterprise, guess what, you all need networking. The fundamental need for this is never going to go away. A high-growth market, it’s a very essential market which we have for us.

Brett Simpson

Yes, makes sense. We’re going to maybe take 1 more from me before we open up for Q&A. And I think my colleague, Jim is standing by. I just wanted to touch on data center switching. And you mentioned the Innovium acquisition. And I think you’d said earlier that the business should be doing about $150 million in revenue. I think it was this year. And I wanted to sort of just get a sense as to how that is playing out, particularly on the sort of 12.8 and 25T switching market. And then I wanted to also get an update on 51T with regards to Innovium. And I know Broadcom’s talked about taping out or certainly going through some sampling. Where are you in that sort of journey towards customer sampling and eventually commercialization?

Ashish Saran

Yes. Let me take the first part and then I’ll let Raghib talk about the whole 1.6T, is really what the question should be about. That’s what actually buys a road map, right? Because you can have a switch but without the optics and come to be one. So first, in terms of where we are with Innovium, very happy with the acquisition fully integrated, right? We got and shifted over to our 5-nanometer process. This is an area where everyone knows we’ve got a very large customer which came along with an OEM. We were unfortunately very supply constrained, right, all the way till almost now. So in terms of our full year revenue, data center switch of $150 million, I think we’re get pretty close to it, I would say, maybe not exactly there but pretty close to it.

And then next year, we do expect most substrate availability, right? Probably one of the only positive outcomes of industry slowing down is going to start to get more parts. So I think we’ll see a next year. And then let me transition that over to Raghib and he can talk about how we think about kind of the long-term roadmap. Raghib?

Raghib Hussain

Yes. So of course, we have very well established ourselves as what is called a cloud switch solution through Innovium but of course, we are also investing in the next generation. And the way We see it is not we do not see it as a switch is the only 1 socket, sort of a thing. We look at what is needed for the whole platform to be delivered.

So if you look at the value of the next level is really you need — of course, you need a 50T switch but you also need something to connect with it, right? So the cost that — the CapEx cost of connecting all the optical is much more than the actual switch in any data center. So the way it works out together is 1.6T is where the 1.6T works with 50T and we are making sure we are investing in both of them and we deliver as a solution because that is needed by the industry to have the right solution, right performance and efficiency and the right cost point to be tell you that for us to be able to deliver the overall solution which will drive the key type of bandwidth out of the switch and of course, through the network.

We are actually implementing both of those. We are in the — bringing in the process of bringing both of those them together. And of course, we are working very closely with our lead customer to make sure that when these products are delivered. They work together, obviously. And then in the timing wise, they are kind of delivered to bring the system-level solution instead of just 1 component and waiting for the others.

Ashish Saran

Yes, we should expect something in the next — I think in now we’re trying to get this ready to go, call it, in the near future, right? So I can’t say much more than that. But I think the most important thing is we were aligned with the cloud ecosystem which we obviously have great contacts with given everything else we do for them. I think we’re all in alignment and I think you should expect movement here over — in the near future, essentially.

Brett Simpson

Okay, great. I just wanted to welcome my colleague and Jim are you there for Q&A.

Question-and-Answer Session

Q – Unidentified Analyst

We’ve got a few questions in the hopper. The first 1 is, does Marvell own the IP for CXL and what is the relationship with Intel?

Ashish Saran

I mean on CXL, we’re basically we’ve been investing internally for a number of years, especially as cloud customers came to us a couple of years back. Pushing us that this is kind of the next big thing. This is what lets them finally completely kind of virtualize their infrastructure. This is a missing piece. And they recognize that you need someone like us because if you think about CXL, if you think about standard and Raghib, can talk a little bit more on, these are very complex. This is not like a simple expander to connect once DPU and just kind of fan out, right? It’s a lot more complex given the protocol translation. And once you start getting from expanders, to pooling devices, complexity goes up massively. So Raghib, maybe you can talk a little bit about what need for CXL?

Raghib Hussain

So first of all, CXL is an industry standard. I mean, in fact, the success of CXL is tide to how standardized it is because it cannot be, oh, 1 vendor control it and it is not going to happen. One of the reasons that PCI is so successful is because of the fact that it is a very well-understood and established standard. And that is what industry, especially the consumer of CXL, they are making sure they’re driving it that way that it happened that way, right?

And of course, there are multiple levels of IP needed to enable the CXL solution. So the interface bus standard is 1 thing, right? So it is just like think of it is as complex, if not less is as PCIe buses. But that is just the interface level protocol and how the bus work and so on and so forth. But then as you go behind this, in implementation itself, you need to worry a lot about the latencies because if the latency — if the SerDes is not tuned with the application need of the CXL application, then your whole application use [indiscernible] goes away, right? If the latency goes higher, as an example, — now what makes us unique is that we control our SerDes. So we can tell our SerDes designer to it. This is targeted for this CXL application and I need — I want you to save last 4 nanosecond in this point which sounds I needed why are you saying for a second. But in the bigger picture when nobody expand, every nanosecond counts, actually. So we are really tuning the entire IP end-to-end from the CXL interface SerDes all the way to memory DDR interface SerDes because that counts, right?

The complete the chain has to be optimized for the solution. So that is 1 aspect of the complexity. The other 1 is when you are implementing, for example, pooling or adding multiple devices and so on, how do you share the resources? How do you make this coherent memory subsystem so that it really serves the need of the CXL. And this is where we have a lot of IP that we have been developing, to provide a scalable way to connect these high disaggregated high-bandwidth processors or the end devices and so on. So those are the areas that we have invested a lot and that is why we are working. We have [indiscernible] design winner. In fact, the way we work in this, we really work closely with the customers to really understand the end application needs. And that’s how we optimize our IP and that’s how we take the market leadership.

Brett Simpson

Thank you. And I think there was a question on the relationship with Intel.

Ashish Saran

Well, we work with multiple partners, as Raghib mentioned, right, whether it’s supporting on the CPU side. So whether it’s Intel, we actually — there was a press with AMD earlier. Yes, you should imagine we work with and tell you work with AMD, we could work with kind of the memory provider. So we know very well because of our great NAND business. I remember, we also know that equation really well. So it’s a broad partnership and that’s kind of Raghibs’ point. This whole thing actually works because it’s an industry standard and it’s quarter by multiple component providers would go into that data center stack essentially, right? So you should imagine we work with everyone.

Unidentified Analyst

Just on that point, guys, when do you think that we start to see CXL in the sort of cash coherency or the low latency examples that you gave, whether it’s 3.0 or 2.0. When do you start to see this aligning with servers, so you really start to see the community come together and really scale up to CXL as an enabler.

Ashish Saran

I think the next couple of years, I think you’ll start to see — I mean, remember, these are now — these are not experimental things with these large cloud customers. These are massive revenue-generating properties. So they’re going to go through a pretty long process of — this is a big architecture change. So we can give you an example which is, hey, we’ve got a good design win. It will take us a couple of years to kind of chip out there right and then start taking it. So you’re looking at I think there’s going to be a lot of work, right which we’ll be working with. We’ve actually got an entire platform from a software and a format perspective which we are giving to customers for simulation and modeling. The next 2 years are I think about, hey, how do you do the implementation? How does it work? In all, there’s a lot of chip development going on. So I would say you’re looking at 2 years of that and then call it calendar ’25 onwards is where you start to see physical implementation in starting a volume.

Raghib Hussain

And I would like to add 1 thing on what Ashish said that this is a platform. This will keep bringing new use cases and so on. So to me, it’s by the time CXL will be as established as it is PCI and all that, like a 10-year process we are talking about but the initial application will start showing up in the next couple of years and then more and then more and then more.

Ashish Saran

And we have first-mover advantage. I think that’s the other key point to kind of keep in mind is it Raghibs’ point, this is going to go from expanders to pooling devices to switches to accelerators over a year of lifetime. I think we’re kind of driving this with the big U.S. hyperscalers and I think that’s the key point here, right? And we made these investments, quite frankly, what they asked us to, with everything else going, you want Innovium brilliant, you got the data center switch support. We’re looking at this kind of be the next big thing and we don’t want to rely just on start-ups to make this happen. So we did have an acquisition as well but that’s the message here.

Raghib Hussain

Is the huge ECU value for all large hyperscalers exactly? That’s why everybody is so much interested in this getting it sooner and later, let it put it this way.

Brett Simpson

Yes, that’s clear. We’ll take 1 last question, if we can, Jim, over to you.

Unidentified Analyst

Sure. So last question, how does Raghib think about the size of the DPUs of content from attaching compute to networking/storage?

Raghib Hussain

So DPU content attached. So the thing is — so are you talking in the context of the data center or overall DPU because to me, anything which we said is a combination of the general compute, DSP-type compute and hardware acceleration is a data-centric processor, right? So it goes beyond data center. It has data center aspect and then it has wireless, it has even enterprise and so on. So can you elaborate a little?

Unidentified Analyst

Let’s maybe keep that in the sort of data center bucket, if you like, how do you think about sizing that market? Do we see — for example, every server has a dedicated DPU engine inside. How should we think about the time line for DPU adoption and what the adoption rates look like on a sort of 5- or 10-year view. How do you think about that?

Raghib Hussain

Yes. Yes. So let me give you a first high-level view the way you should think about it. As the data volume is increasing, the need of data-centric compute will keep growing more and more and more. right? So the thought process that thought process that every server needs a DPU, actually every server will need a DPU or more actually multiple types of DPUs or maybe the relationship of server to the rest of the DPUs will keep getting in favor of data-centric compute more and more [ph]. So to me, the overall compute market will grow exponentially driven by the expansion of the overall volume of data and because of the need of the efficient compute needs. So this heterogeneity of the compute will keep increasing. And as a result of that, the size of the market, of the data centric compute will keep increasing and compared to the organic growth of the traditional what I call the application centric compute.

Brett Simpson

Fantastic. Well, I think we’re out of time but we could go on. I’ve got a lot of other questions. We’ll do this in another time. But I really appreciate your time, guys, thanks Ashish as always. And Raghib, great to see you on our tech conference. I don’t know if there’s any closing remarks you want to lay out Ashish or Raghib but feel free.

Raghib Hussain

Thanks. We really appreciate the opportunity. We’re very well positioned to navigate through some of the short term. And I think all of the key growth drivers remain very much intact. So looking forward to continue the conversation. Very excited to be here.

Brett Simpson

With that, we’ll close out the session and thanks again, guys.

Raghib Hussain

Thank you very much.

Ashish Saran

Thanks.