Marvell at Cantor Fitzgerald: All In On AI Strategy

On Tuesday, 11 March 2025, Marvell Technology Inc (NASDAQ: MRVL) presented at the Cantor Fitzgerald Global Technology Conference, revealing a bold strategic pivot toward artificial intelligence (AI). The company highlighted its focus on data centers and custom silicon, alongside challenges in the competitive AI semiconductor market. While Marvell expressed optimism about future growth, it also acknowledged current market hurdles.

Key Takeaways

• Marvell is shifting its focus to AI, with over 90% exposure in data infrastructure.
• The company projects significant growth in data center revenue, aiming for $26 billion.
• Marvell’s strategic relationship with AWS is expanding, incentivized by revenue milestones.
• The acquisition of Inphi has strengthened Marvell’s networking and connectivity business.
• Marvell is leveraging R&D spending to drive innovation and growth.

Financial Results

• R&D Spending:

- Growth is being driven with a minimal increase of $10 million, reflecting efficient spending.

• FY25 Performance:

- Revenue reached $5.8 billion, surpassing the $5.3 billion consensus, with AI as a key driver.

• FY26 Forecast:

- Expectations have risen from $6.7 billion to $8.02 billion, with a $500 million upside.

• Data Center Revenue:

- Incremental revenue of $1.8 billion to $1.9 billion was added last year, with a target of $26 billion.

Operational Updates

• R&D:

- Significant investments in data center technologies, with 80% of investments focused here.

• Organizational Changes:

- A dedicated data center business group is led by Raghav Hussain.

- Non-data center products are managed by COO Chris Koopens.

• AWS Strategic Relationship:

- Expanded design network usage in the cloud, with AWS incentivized by revenue milestones.

• Custom Silicon:

- Full volume ramping for a major customer, with unique IP investments and packaging partnerships.

Future Outlook

• AI Market:

- Marvell aims to address the need for lower power and higher compute density in AI technologies.

• Competitive Positioning:

- Custom silicon viewed as a primary alternative to NVIDIA, focusing on advanced nodes and scale-up networks.

• Networking and Connectivity:

- Emphasis on silicon photonics, broadband circuits, and system modeling for optimal IO blends.

• Co-packaged Optics:

- Seen as an opportunity for incremental SAM, targeting inside the rack and accelerator clusters.

Q&A Highlights

• Investor Concerns:

- Addressed revenue growth concerns, highlighting strong performance and future outlook.

• Custom Silicon Competition:

- Full-service providers are expected to dominate, with Marvell focusing on proven designs.

• Connectivity and Custom Silicon Synergies:

- Optimization of all IO, including HBM interfaces, to enhance compute on accelerators.

For more detailed insights, readers are encouraged to refer to the full conference call transcript.

Full transcript - Cantor Fitzgerald Global Technology Conference:

C. J. Muse, Semiconductor and Semiconductor Equipment Analyst, Canner: Well, good morning, everyone. My name is C. J. Muse, semiconductor and semiconductor equipment analyst at Canner. And it is truly my pleasure to be hosting Fireside Chat with Marvell.

We have Matt Murphy, CEO and Willem Mancos, CFO. Welcome, gentlemen. Yes. Thanks, CJ, for having us.

Matt Murphy, CEO, Marvell: Good to see you. Very good to

C. J. Muse, Semiconductor and Semiconductor Equipment Analyst, Canner: see you. So we were going to play the Dropkick Murphys on your entrance in, but New York union rules state that you can’t have vans from I see.

Matt Murphy, CEO, Marvell: I’m shipping up to Boston tomorrow.

C. J. Muse, Semiconductor and Semiconductor Equipment Analyst, Canner: So you’ve only got me. That’s right. I hope that’s okay. So it’s a great to have you here. We have a very, I think, broad audience.

So the thought was maybe start at a high level and then kind of take it down. So I guess just to start and to level set the audience, if you could start perhaps with your recent decision to pivot Marvell all in on AI. What was the thought process? How have kind of logistics investments within the company changed with that pivot?

Matt Murphy, CEO, Marvell: Yes, great question. And great to see everybody. Thanks for the opportunity. So I think the harder pivot on the all end AI obviously built over many years. I became CEO in 2016 and very early on we made the decision back then to pivot to what we called the data infrastructure opportunity, which our view and my view at that time was that ultimately was going to drive the biggest TAM growth in semiconductors on the go forward.

Even though Marvell was consumer levered, we sort of knew we needed to make that change. And a lot of you guys are investors, you followed that story and sort of methodically year by year organic investment, M and A divestitures, all the different moves we made. And that TAM has sort of exploded, obviously. And obviously. And in particular, the data center side of it has continued to increase at a tremendous rate.

And so as we found ourselves now in the last few years with like 90% plus exposure in the data infrastructure and I think down to like 5% consumer consumer. What’s happened is that the composition of that market has skewed big time to data center. I mean it’s like 75% of the TAM now for chips that we do is in that market. And that’s been driven primarily by the increase in AI. So it may sound like, hey, yes, let’s go after AI, it’s like this new thing.

But for what we do, it’s the biggest end consumer of the types of things that we’re good at. On top of that, I would say that our pivot at Marvell eight or nine years ago to data infrastructure, the pivot to All In On AI is an extension of that, but it’s as consequential. And we can talk about what that means for our other businesses, which we actually feel really good about. But in order to really execute and succeed and win in this big market transition and dislocation, which you only see three or four times in a long career, our view is the best way to succeed and to change our fundamental trajectory is just to completely laser focus and take on constantly a five plus year type of time horizon on our investments and we’re making that big bet. And so far, we’re really happy with where it’s landed.

Makes sense. Anything worth kind of highlighting in terms of R and D focus, evolution, hiring, engineering, you announced the engineering, you announced the program with Amazon, would love to hear kind of strategically how things may have evolved? Yes. I think from an R and D standpoint, very pleased with not only how our team is performing, but how we’ve grown it. And even with all the ins and outs and the market cycles up and down and all the challenges that semi companies face, I think we were looking at this data.

We’ve grown our R and D spending very consistently to take people to take people who are working on say one set of technologies and move them and we’ve got that ability now inside of Marvell to do that. It’s actually really hard to do in companies. You have business units and you have people with teams and nobody wants to get of their team. And anyway, so we’ve actually established a culture where that’s part of the operating system of the company. So in the last couple of years, we’ve really pivoted the R and D resources ahead of the market share growth we saw and the revenue growth into the data center area.

I mean, we were running probably 80% of our investments plus in data center the last few years back when the revenue was 50%, sixty % of our revenue was in data center. Now it’s 75%, so you start to see those things catch up. But on top of that, we’ve really focused the effort around a dedicated data center business group and that’s led by Raghav Hussain, who’s been my longtime partner and and President of Marvell. He was a Co Founder of Cavium. He and I have worked kind of jointly and hand in hand to really create a vision for Marvell back in 2017 when we were doing the Cavium deal.

And so this is like a sort of dream job for him. This is really what he’s great at. And then we’ve taken the non data center end market customers and products and opportunities. We moved those under Chris Coopens, our COO. And part of that was just to get the focus, right?

Not just the focus on the data center, but actually also having a senior executive on my team that was going to really manage for with its own P and L approach and its own profitability profile and investment profile for the multi and investment profile for the multi market business as well. So just because we decoupled them doesn’t mean one got sort of deemphasized. In fact, we think we can recover and grow that multi market business back up. But that’s another big change we made. Again, those are hard to do when you go from product technology organization to end market focus.

It usually creates a lot of noise and a lot of pain and everybody is upset. This was actually a very smooth transition inside dedicated data center, engineering group, business group, all in focus, no distraction, and I think that’s going to serve us really well.

C. J. Muse, Semiconductor and Semiconductor Equipment Analyst, Canner: Yes. And CJ, just maybe one additional little bit more tactical comment on the R and D, but we’re really pleased on the leverage we’re driving. If you look at the progress we’ve made over the last year and then this guide into Q1, where typically we’ll see sort of a mid single step up into Q1. And this time around, when you look at that guide, we guided up only $10,000,000 And it’s really a reflection of the overall model and the NRE that we’re able to recognize as contra OpEx and really encouraged on the trajectory and focus on driving continued leverage right through next year as well. Perfect.

Maybe sticking with the high level thought, perhaps you could give us kind of state of the union. What are you most excited about? And what are the challenges do you see for the AI semiconductor market in ’twenty five, ’twenty six? Yes. I think

Matt Murphy, CEO, Marvell: the thing I’m excited about is also the challenge. So it’s all wrapped into one, which is that I’ve been in the industry for thirty one years, okay, and I’ve had the chance especially in my prior company to get exposed to kind of all these end markets and all these different end product cycles like laptops and Internet and cloud and all I’ve seen all these just like you would, you’ve been doing this on the other side of the house. But what happens when you hit one of those inflections is, I’ve just seen this a few times, you end up going, having to go just flat out every day on your engineering execution, your innovation that’s needed and you always feel like you’re behind. That’s the challenge. And the reason is, and with AI, it’s very pressing is this technology needs to be fundamentally lower power.

It needs to fundamentally have more and more compute density. It needs to have a better TCO. There’s all this massive economic productivity unlock in the globe that can come from this. But today, if you just and look, it started to hit a major inflection just with the combination and really led by my good friend, Jensen, at NVIDIA with the fundamental leveraging Moore’s Law plus system engineering approach plus software plus everything. Finally, it’s created an opportunity for AI to really take off, but we’re still at the beginning stages.

So it’s an exciting time to be in semiconductors because where we sit now at Marvell and I think there’s only a couple of companies that are really at the bleeding, bleeding edge. The whole industry is now being driven by HPC. That’s a big shift. I mean, this was a mobile PC driven world on technology for the last thirty years. In the last two or three years and even TSMC has said it, right?

I think their shift to their business is more HPC than mobile now and that’s where they’ve oriented their roadmap. So the challenge and opportunity is there’s a lot to do. We’re at the early stages. I mean, we’re going to need multiple successive generations of these technologies, I think, to really get it to where it needs to be. And the challenge is it’s really hard stuff.

And that’s a good thing for us because I think there’s scarce few companies that have the ability to design these large scale 100,000,000,000 transistor chips, get them done first pass silicon, which we’ve been able to do. All of our five nanometer products actually for data center have been first pass successes. That’s been a great track record approach. But also the broad suite of intellectual property you need, packaging technology, the thermals, the system modeling, and then the ability to actually have the manufacturing year, multi generation. And you got to have scale to do that and you also have to have the right capable team.

So those are all reasons we feel good, but yet every day I tell you we feel like we’re just it’s just an all out, all out assault to keep up with what the market’s requiring. And I think that’s a great place to be. Once it slows down and you go to five year cycles or seven year cycles and nobody needs a refresh and the next nanometer is it’s a different world. And we are the farthest thing from that at

C. J. Muse, Semiconductor and Semiconductor Equipment Analyst, Canner: the moment. Yes, no, I agree. I think McKinsey said $1,000,000,000,000 semi revenues by 02/1930 and we’re modeling 800,000,000,000 this year led by AI. Right, yes. We’re almost Yes,

Matt Murphy, CEO, Marvell: the trillion dollars always look like how we’re going to get there and actually, especially when the down cycle is when those reports were coming out and it definitely looks feasible. But it’s 100%. The only thing that gets it there is AI. It’s not coming back because of consumer tech and smartwatches and phones and things like that. That’s not going to get us there, right?

What’s going to get us there is the data infrastructure market led by AI as the primary use case in private.

C. J. Muse, Semiconductor and Semiconductor Equipment Analyst, Canner: So that’s a good segue to custom silicon. So perhaps you could kind of speak to how you think about your competitive positioning there. And you talked in the most recent conference call in terms of ramping at full volume for one large customer, another soon. Could you perhaps start there?

Matt Murphy, CEO, Marvell: Sure. Yes. We feel very good about our competitiveness. And in fact, the contrast of where we are today from where we were when we won those initial five nanometer designs back in ’twenty one, ’twenty two, ’twenty three, call it, now the last couple of years comparing, I mean, literally we won a lot of those initial designs off of a five nanometer test chip and great PowerPoint presentations from our ASIC team, right, which we had acquired from GlobalFoundries, which was an asset called Avera, which had been inside of IBM for a long time. It was a very successful chip design team, had never designed on TSMC before, had always used IBM GlobalFoundries processes.

And with Marvell, by the way, when we got into this ASIC business, our prior most advanced node was 12 nanometer. We had no products on seven nanometer. We made the decision to skip it. And the reason I bring all this up is we went from that and actually capturing those designs, getting the trust of our customers to go take that risk and do it. We executed really well, as I pointed out, in terms of the success rate.

I’m super proud of our team. But we’re like it’s like a 180 from where we were then to now. I mean, we are on we have the silicon, we have the proof points. We actually have invested ahead of the curve now in very unique IP. We’re at the bleeding edge on the TSMC shuttles, the packaging partnerships.

It’s just like we are ready to execute and we’re pushing the limits on the next gen of everything and even two generations out, where before we were kind of wondering how are we going to catch up. But we absolutely caught up. And this sort of full, you know, Marvell technology platform approach from process package, IP, manufacturing is, I think is what’s setting us apart. And I think to your point, there’s very few companies that that can actually do what we do. And we do think that over the long term, despite the fact that there are different business models and different approaches these customers are gonna take to get their silicon ambitions fulfilled, we still believe the large portion of the TAM in the future that ends up shipping is gonna come from full service, fully equipped, custom silicon companies like Marvell.

There will be other exceptions. It’s not a binary thing, but I think the bulk of what ends up getting produced gets produced by people like us because in the end, you can try to do it yourself and you can try to augment it and you can try to license things and bootstrap it together and maybe you can, in theory, save some money. But in the end, are you going to get your chip delivered? Is it going to yield? Can it work first pass?

Can you underwrite the success of that and derisk the program? And that’s proven to be a very valuable thing that we can bring to the table. We can show we can get these to production. And as I said earlier, time is of the essence because if you’re late and you have to spend the thing two or three times, you might as well throw it away, right? It becomes irrelevant as long as the market leader keeps pushing the state of the art.

C. J. Muse, Semiconductor and Semiconductor Equipment Analyst, Canner: So a few months ago, you talked about expansion of your strategic relationship with AWS multi generational five year agreement, both custom silicon and connectivity. Can you speak to, I guess, how has your relationship, if at all, deepened since that announcement? How we should think about that structure and what it means to kind of your multi generational, multi year kind of commitment to one another?

Matt Murphy, CEO, Marvell: Yes, the announcement was a big deal. It was a huge deal for us and for them, and I think industry wide it was quite unique, right, to see something like this. It was born out of though really years of working very closely together on two directions. The first is, you know, we’ve we’ve become a very strategic supplier to them, of custom silicon, of networking products, connectivity products, and and I think that is is sort of very much strengthened over the last few years to where that partnership is very deep. At the same time, we were one of the early semiconductor companies to enable our EDA in the cloud and our design engineering network portions of it to be done in the cloud.

And we chose early on, we chose and we went all in with AWS for that. And that was that was a risk in an investment just like they took a risk in an investment in us. And working together, we’ve made that a very efficient tool and platform for Marvell. And so part of this strategic agreement we have is that we’re going to expand our usage of their cloud services and our design network, which I think really puts us on the forefront of major design companies leveraging that capability. So so and and for them, that’s a big deal because they have a very strong teaching customer that’s in there every day, helping to improve their their performance, and that helps them overall in terms of their business they’re trying to drive.

So that’s very strategic. And then for us, and then and then on top of that, we entered entered into this agreement where they there’s a warrant structure where they’re incentivized if certain revenue milestones are achieved that they get an opportunity to participate in the upside of Marvell. And the scale of that is significant, and it covers both custom silicon products as well as it covers networking and connectivity products. So it’s a very all encompassing partnership, and I think it’s just a I think it’s a symbol of the deepness of the relationship, but certainly codifying it in this way only makes it stronger. So we’re really pleased, and they’ve been a great partner to us.

Perfect.

C. J. Muse, Semiconductor and Semiconductor Equipment Analyst, Canner: If I could ask maybe a more narrow question, I think from an investor perspective, the breadth and scale of growth of revenues for this business disappointed near term. And I guess, is that a reflection of perhaps pull in of business earlier, perhaps conservatism, perhaps second half weighted? How should investors kind of think about that?

Matt Murphy, CEO, Marvell: Yes. Well, let me give it from two angles. I think the first is, and and, you know, as a as a public company CEO, you’ve got to kind of manage these two dynamics. Right? One is is your internal plan, your internal view, how’s the team doing, how are we tracking, and your commitments, by the way.

Like, what have we said? What are we doing? And then there’s the external world always. Right? Equity markets, competitive rumors, landscape, etcetera.

And and sometimes those things diverge. Okay? So from our standpoint, set aside the whisper, the expectation, we’re extremely pleased with the performance of the company. You know, we had a huge, beaten raise from q three to q four. There was a big step up in the revenue As we ramp some of these key programs, that was great.

We performed well against that ramp. We executed. We guided ahead of consensus for Q1. It wasn’t as high as people wanted, but as far as what we’ve said we were gonna do, we’re actually way ahead. I mean, let me just take you back a year real quick just to we were sitting here.

Consensus for Marvell for for fiscal twenty five at that time was 5,300,000,000.0. We ended at 5.8. So we upsized 500,000,000 in like nine months, mostly driven by AI. That was pretty cool. The same time for fiscal twenty six, the year we’re in, consensus a year ago was like six seven, and now consensus has us at $8.02.

So that’s 500,000,000. Again, this is actuals plus sort of street forecast. So in like basically a year, we’ve got $500,000,000 last year or more and we’re projected to have on top of what was already looking like pretty strong growth. And when we look also year over year from fiscal ’twenty five to ’twenty six, there’s a very few companies I think that if we can deliver this type of growth for the street forecast, it’d be like astonishing, especially at our scale. And then when you step back too and you look at just the net add per year, we’re adding an incremental data center revenue, like year over year, I think last year we added like $1,800,000,000 or $1,900,000,000 and the Shear Street said $1,800,000,000 or $1,900,000,000 and $26,000,000 we said we’re going to grow to in terms of so again, we feel really good about how we’re doing.

Now there’s we can talk more about the external environment. But as far as where we’re tracking, I mean, we guided the best quarter in the history of the company, we guided the best EPS, we’ve GAAP profitable. We’re buying back shares. SBC is under control. Like we got great TAM outlook.

So from our standpoint, actually nothing’s changed. It’s only gotten better over the last year since the AI Day. And and I would say even since the December results that we announced in December, if you just look at sort of with the administration change and all the kind of news about the the overall AI market, big capital projects being announced, sovereign data centers, rumors of huge spend by the hyperscalers in the future, it looks really positive. All those data points look more incremental over the long term. So if I didn’t know what was going on in the stock market, I’d we’d be having a glass of champagne and say, good job, Matt.

You know, you took a trip to this company. It was kind of a mess, and you sort of turned it around, and you bought these assets, and you integrated them, and you figured out how to get into this AI market, and you got the thing up to, like, 8,000,000,000 in revenue this year, and cool. So, you know, that’s contrasted with the reality, which I understand and I appreciate and I, understand the frustration by investors with certain externalities that are happening. And I’m trying to be as helpful as I can, but you guys also would understand these partnerships I’ve built with these customers. They’re so deep.

They’re so trust oriented, and they’re underwritten with NDAs and contracts and the whole nine yards, but at the end of the day, it’s a trust business. And I have gotten the honor to to know the CEOs of these big companies, and they’re trusting me. They’re trusting me. Not just with, like, executing to to to meet what they need, but with with their fundamental, like, we know their road map. We know all the ins and outs in some of these things.

I’m not gonna I can’t talk about I would never do that. And I think every investor here would understand, but I me and my team will try to be as helpful as we can so we frame the long term opportunity, but some of these short term things, it’s just hard to comment. And we’re no stranger to rumors at Marvell. I mean every once in a while, there’s OFC 2023. You guys remember that one?

Linear optics was gonna wipe

C. J. Muse, Semiconductor and Semiconductor Equipment Analyst, Canner: out our DSP business. That up.

Matt Murphy, CEO, Marvell: You know, stock plummeted, what’s that?

C. J. Muse, Semiconductor and Semiconductor Equipment Analyst, Canner: I wasn’t going to bring that up.

Matt Murphy, CEO, Marvell: I remember, I mean, it’s been all to DSP. So we’re just going to, you know, we’re a very heads down execution oriented focused company, drive the stability, drive the long term growth, take care of the team.

C. J. Muse, Semiconductor and Semiconductor Equipment Analyst, Canner: Makes sense. So maybe a higher level question on custom silicon. I think in the last twelve months, the market has decidedly made the call that if there is a second source or primary source vis a vis NVIDIA, it’s custom silicon. And I guess within that construct, there’s two large players, you and Broadcom with the IP, the advanced packaging, the SERDES, whereas there’s others, I would highlight kind of in Taiwan and Mainland China design houses with less kind of IP. So curious, as you look at the landscape, as you look at the push to three nanometer and below, how does the differentiation increase and how I don’t want to call it market share, but how kind of the prioritization of just two players occur over the next one, two, three years?

Matt Murphy, CEO, Marvell: Yes. So I think a couple of things on the market dynamic. This view that there was going to be one big merchant guy, NVIDIA and then custom silicon, and then probably not a not a lot of room left, that was that was me and Raghav’s high level thesis in 2018. Okay? Why

C. J. Muse, Semiconductor and Semiconductor Equipment Analyst, Canner: don’t you

Matt Murphy, CEO, Marvell: call me? Well, no. We we did. We bought we bought Avera in 2019. We announced it in, like, May.

And, you know, it was born out of all these meetings we were having with these hyperscalers. And we had an inference chip, by the way, from Cavium, the M one k. I don’t know if you guys remember this. Syed may have talked about it, maybe he didn’t, but we had an inference chip. We had a great team.

It was awesome. It was the best performance of any seven nanometer chip in the history of mankind. There’s no customers for it. None. Everyone wanted their own.

And our so we shut it down. We redirected those resources in, at the end of twenty eighteen, early ’20 ’19. We acquired Avera, and we settled on that journey. And I think that thesis clearly has proven to be the correct one. I’m glad we didn’t try to do an m one k and an m two k and an m three k, and we’d be still telling you some story with no revenue and said we’ve got revenue in this area.

And I think that’s so I think that’s going to continue. And then within that, as you point out, there’s sort of the full service model, right, which is the end to end suppliers that can do everything. As I said earlier, I think that’s still going to dominate in the future, the large chunk of the revenue that ships through. But clearly with the with the amount of spending going on, there’s going to be room for experimentation, there’s going to be room for companies to try to, to do it a different way, which is more like do it yourself, but augment with some with some design resources, license a lot of IP, use a whole bunch of partners and and try to pull it together. And and so far, that’s that approach historically has been kind of a mixed bag.

What’s really shipping today primarily has been the full service model if you just look at the percent of the revenue. So I think you’ll see that, and I think there’s a valid reason for companies to to try to do that. But to to answer your question, it’s only gonna get harder. I mean, we’re already out looking at a 16 and a 14 at TSMC, the next two nodes beyond two. We’re looking at what that complexity is gonna bring.

We’re looking at things like backside power. We’re looking at things like side power. We’re looking at things like CPO and scale up networks and how to integrate all this stuff and two generations out on package technology and the cost structure that’s going to be required to go drive all this. And I’m just looking at all this in my engineering reviews and thinking to myself, how is somebody going to do all this themselves? Maybe somebody can and they can try to cobble it together.

And that may work sometimes, and and I’m not saying it’s going to be 0% of the market. I’ve never said that that this is binary. That will exist. It has existed. But I just still believe based on how complex it’s going to keep getting, the large customers, I think, are gonna rely on the full service companies to really underwrite the success and just, like, don’t take any risk and get the best engineering team, best technology, and someone that’s gonna get in there and be able to fix it, yield it, ship it, and and you don’t have to worry.

And that’s worth a lot, especially in a world where manufacturing cycle times now are astonishingly long on these new EUV based technologies. I mean, it’s twenty something weeks in the fab. So you do an all layer change and you spin the chip and, oh, it didn’t work. Sorry. But we went cheap on the PD and we didn’t have a close timing.

We didn’t have to have all the advanced verification techniques like that a Marvell would have, and then it didn’t work. And then you spin it, and okay, I’ll see you in a year. It’s really risky. So these are the dynamics that we see.

C. J. Muse, Semiconductor and Semiconductor Equipment Analyst, Canner: Makes sense. And maybe pivoting to networking and connectivity. Inphi, obviously, tremendous acquisition, figured growing your optics revenue by more than 150% versus calendar ’twenty two, really pivoted to high value data center interconnect and switching solutions. With OFC just around the corner, what are you most excited about in terms of the near term trends?

Matt Murphy, CEO, Marvell: Yeah. I think there’s a lot of good stuff happening, and OFC is always a blast. Always something fun to look forward to. But no, I think we have we have well, you’ll if you guys are there, come see our booth. I mean, we’re going to have a great lineup in terms of speakers, in terms of demos.

And, yeah, I don’t want to steal the team’s thunder, but, you know, I think the the suite of technologies that we have, the really the full end to end in connectivity is very unique. You know, the combination of having silicon photonics in house, broadband analog circuits, advanced DSPs, new use cases, things like AECs, retimers. There’s a lot of companies there’s companies that have one of these or two of these or some of these. We’ve got the broadest range and all at the sort of bleeding edge and we can stitch them all together. We also I mean, not only that, we take all those technologies and we we form this group in the reorg we did called our cloud platform group where now we’ve implemented very sophisticated system modeling capabilities, so we actually can help our customers decide what’s the right blend of IO that they need, what’s the best trade off in terms of how they want to architect their interconnect.

And we can do that because we can provide everything from SIFO to linear optics to full full DSPs in the most advanced node to electrical solutions instead of optical and everything in between. It’s very compelling because at some point, this stuff’s so complicated and it’s such a mix now. It’s it’s totally, heterogeneous that going, like, supplier by supplier may be an approach, but there’s also a value in pulling the pieces together. So that’s how we think about our our business there. And we couldn’t be more thrilled with the Inphi team.

They’ve done an outstanding job. It’s been a great integration. We’ve just, like, completely came together. We staffed the heck out of it, took a lot of talented Marvell folks, put them in there. One giant connectivity engineering organization.

You don’t even know who’s who anymore. It’s, and it’s and it’s, yeah, it’s way outperformed the deal model, obviously. So thank God because that was really expensive when we bought it.

C. J. Muse, Semiconductor and Semiconductor Equipment Analyst, Canner: So not getting in front of OOC, but perhaps getting ahead of GTC, where I think co package optics is going to be a critical area of focus. So perhaps you could give us kind of how Marvell sees the adoption curve there where I think you’re of the view that that’s a handful of years out as opposed to near term.

Matt Murphy, CEO, Marvell: Yes, I think historically co packaged optics definitely was was historically more of a science project. I think there’s there’s been I mean, I was going to OFC even when I was at Maxim in, like, 02/2012, ’2 thousand ’13, and there was this was the holy grail, you know, and there was, like, small SIFO companies running around, and is this sort of gonna disrupt things or not? And it’s definitely continued to shift because it’s it’s a lot of complexity. But the issue there was always technologies maybe not ready manufacturability wise and also the impact when you try to implement that, like, in a scale out net network and in the aggregation switching area, it’s just there’s too much risk because if one of these fails and there’s a lot of issues about what’s the actual reliability of these, you’re not just taking out one switch box, you’re you’re creating a blast radius where you’re taking out a whole bunch of switches. And so the sort of network uptime goes down and is it really worth the trade off versus if you want to really go hyperscale mode and scale up a massive data center and do it fast, pluggables is the way to go.

And I we think that that’s gonna continue. But in the scale up where now in inside the rack and from inside the accelerator clusters, the the the need for high speed connectivity is only getting more because that’s gonna be a limiter on the throughput. And we think that this copper passive approach is probably gonna run out of steam soon. And so that’s where I think a new market opens up, which is new SAM for us with something like co package optics or some version of that or even linear optics as an example. And and again, we’re we’re we’re fine with either and all of those because we are the full provider.

We wanna provide the most optimized solution. But in the scale up, that would be fundamentally incremental SAM for us.

C. J. Muse, Semiconductor and Semiconductor Equipment Analyst, Canner: So it’s an opportunity, and we’ll see what happens at TTC. Perfect. Maybe last question, thinking about the synergies between connectivity and custom silicon, where scale up, scale out obviously becoming more and more critical and thinking of the way Jensen does, viewing the data center as kind of one large GPU or one large in your parlance. I guess, can you speak to as you think about your investments and your conversations with your customers, how critical and or competitive advantage is it to have kind of both businesses in house?

Matt Murphy, CEO, Marvell: I think it’s fundamental. I mean, you paraphrasing Jensen is the right way to think about it. It’s a high speed networking problem. Obviously, the compute is very sophisticated and that’s at the bleeding edge. And, and they and our customers that are doing their own custom chips have done a great job on that, but it sort of doesn’t do you any good if you haven’t optimized all the IO.

And literally down to things like, I mean, we’re optimizing HBM interfaces at this point, and you’d think, well, what’s the big deal? Can’t you just make it go faster or something or why can’t there be an industry standard? But again, we can apply a lot of our novel design approaches and use our IP to fundamentally create more more room actually on the accelerator for compute. We can treat shrink the beachfront area that’s required and shrink the HBM phi physical size on the accelerators to the point where customer can actually have more compute as an example. And then at the same time, we can enable denser HBM on the on the periphery of the chip.

I’m just giving you that as one example of if you didn’t have that innovation, there’s a throughput issue there, and it’s a there’s a limiter there. And so I think starting from literally die to die interfaces all the way building up to, you know, I mean, you can go memory to accelerator, you can go accelerator to accelerator, you can go cluster to cluster, you can you can just think of this thing. And and with Marvell, we’re we’re trying to build it literally on every hop. Right, we want to participate. All the way to all the data spinning out of the data center over the DCI network and going back and forth.

We want to own that, we want to be the key provider there, and that’s that’s the capability that we’ve built up. And it really becomes a differentiator for us because we’re not just offering to our customers, hey, we can do an ASIC job for you. What do you want? We can actually provide a full solution relative to the connectivity surrounding it. And then we could start to talk about, hey, let’s partner up as you’re doing your accelerator design on on what kind of scale up, solution might you need to go couple with that.

And then so we we view, like, for example, silicon photonics, we pitch it as part of our our, our ASIC platform. It’s not a random separate science project. It’s highly integrated into into our value prop. So those are things that are long term actually competitive advantage for us because very few digital companies are gonna have a a SIFO team inside of them that’s world class that’s actually shipping volume, which we do have today.

C. J. Muse, Semiconductor and Semiconductor Equipment Analyst, Canner: Well, perfect. Well, I think with that, we’ve run out of time. So thank you both.

Matt Murphy, CEO, Marvell: Okay. Yeah. Thanks. We appreciate it. Nice to see everybody.

This article was generated with the support of AI and reviewed by an editor. For more information see our T&C.