Welcome, and thank you for standing by. I would like to inform all participants that this conference call is being recorded, and parts of this call may be reproduced in JP Morgan research. If you have any objections, you may disconnect at this time. I would now like to turn the call over to Harlan Sur.
Yep. Thank you, Drew. Happy New Year, everyone. Thank you for attending J.P. Morgan's Virtual Fireside Chat series. We normally host the Marvell team at CES, which is next week, but thought it'd be a good idea to get in front of what will be, I think, a pretty busy week. My name is Harlan Sur. I'm the semiconductor capital equipment analyst for the firm. Very pleased to have Matt Murphy, Chief Executive Officer of Marvell, and we also have Ashish Saran, Vice President of Investor Relations, here with me today as well. Marvell stock, as you know, is one of our top picks in the semiconductor sector for this calendar year. Leadership in cloud data center and 5G infrastructure with their networking, compute, storage, and custom ASIC solutions.
Currently seeing a very strong tailwind from AI and accelerated compute, and an emerging pipeline in automotive networking and compute products as well. So, gentlemen, Happy New Year, and thank you for joining us today.
Yeah, Happy New Year, Harlan, and great to see you-
Yes, absolutely
... on the call. Thank you. Yeah.
Mm-hmm. Yep, thank you. So, you know, your data center business has seen a strong sequential and year-over-year inflection in growth in parallel, right, with the strong adoption of GenAI. Here in the fourth quarter, for example, your data center business is growing 50% year over year, represents more than 50% of your total revenues, and by our estimates, the strong double digits year-over-year growth will continue through this calendar year. But before we address the strong demand trends here, just wanted to touch upon the other half of your business, which has seen some cyclical headwinds, namely enterprise networking, carrier, and your storage businesses. And so, if I may, starting with your enterprise networking business, you know, the team has been proactive all of 2023, right, under shipping demands over the last few quarters to flush out excess inventories.
That business is roughly down about 30% year-over-year this quarter. It's gonna be down more like 35% year-over-year in the April quarter by our, our estimates. Can you, can you guys just give us a sense of the inventory situation at your customers, and does the team believe the business will find the bottom in the April quarter?
Yeah. Hey, great, Harlan. Thanks, and let's kick it off, as you said, with Enterprise. So that business, if you just take a step back, has done extremely well for Marvell and the whole team. I mean, if you look at sort of where that was two or three years ago, it was well below $1 billion of revenue. You know, kind of peak pandemic, we got it to about a $1.4 billion run rate, which was a little frothy, and I think everyone knew it. It was just how do you sort of manage it through the pandemic crunch? To your point, we started saying in December last year, that we believed-
Mm-hmm
... Enterprise would slow down. We were gonna take proactive steps to manage inventory with our customers and through the channel, and so we, we did that actually throughout calendar 2023 in a, in a very, I think, thoughtful way. And what's happened is by the, you know, end of the year, kind of exiting and then as we look into early part of this year, the inventory that we wanted to burn down is we've done a pretty good job of that, but the demand at the same time has weakened.
Mm-hmm.
And so, you know, given high interest rates and sort of the macro and the overall setup, plus China weakness, in there, you know, that business we said was, as you pointed out, was gonna be down year-over-year in the fourth quarter and be soft, you know, in the first part-
Yes
... of this year. But the way to think about it, I think, for investors is, you know, if you ignore kind of the quarters where there was some level of bubble, you know, $275 million-$300 million a quarter is probably a good number, you know, kind of through cycle. And then from there, once inventory is normalized, and that's the run rate of the business, you know, we've always said that Enterprise for Marvell as a market is kind of a low- to mid-single-digit grower as a market, and then we can grow a little bit faster, right, through content gains and some share gains still in front of us. So it's a nice business. It's a foundational business for us.
It's performed extremely well, but despite our efforts to control inventory, the demand environment has proven to be a little bit softer than we would've thought, say, six or nine months ago when we project it out.
Perfect. And then maybe moving to the carrier networking part of your business, kind of similar question, right? But with respect to your 5G infrastructure and Metro Long Haul franchises, by our estimates, this segment's gonna be down 40% year-over-year, this quarter, down more than 50% year-over-year next quarter. Any signs that April quarter will be the bottom? And more specifically, how big is Nokia for the team as a percent of your overall revenues, right? Investors have been concerned about the recent headwinds on the AT&T O-RAN loss and some of Nokia's other sort of near-term challenges.
Yeah. Great, great questions, Harlan. So, maybe take them in order. So again, as a backdrop, similar to Enterprise, you know, the business in carrier has ramped, you know, significantly for Marvell, if you just go back a few years-
Mm
... both on the back of the 5G ramp, which our team executed extremely well on, and, you know, we probably grew that business for, I don't know, eight or 10 quarters in a row, maybe more. I mean, it just, it's performed extremely well and in line with what we, we signaled actually when we, when we acquired Cavium and sized the opportunity. So that we're pretty pleased with. We picked up a nice franchise in long-haul optics through Inphi, and that's also performed very well through this cycle. And, and I would note that while some of the year-over-year numbers are, are down quite a bit in Q4 and, and even signaling out to the first quarter, Q3 was a record.
Yeah.
The quarter we just completed. It was, you know, $317 million. It grew, grew 17% year-over-year. It did extremely well. So if you normalize it, and again, this has always been a lumpy business for semiconductor companies, the carrier-
Mm-hmm
telco market, but if you kind of smooth it out, you know, in 2023 or fiscal 2024, you know, we'll do over $1 billion in that business based on Q4 Street, and it was about $1 billion in our prior fiscal year, in fiscal 2023. So think of it as a billion-dollar-ish type of business, clearly down in the fourth quarter and in the first part of next year. But we do expect, based on the design win position we have and the quality of the carrier business that we have, that that should come back again up to that run rate once we clear inventory and the spending environment improves and actually goes back to normal. And then from there, we have a nice product cycle in front of us on the wired side.
On the Coherent DSP portfolio, we have our new 800 gig product out. It's higher ASP, it's higher content per port, so that should help us. And then in 5G, we still have, we still have content gains in front of us that we've already won several years ago, but those platforms haven't gone into production. With respect to Nokia, there was a lot of questions about, about them, who's one of our key strategic partners-
Mm-hmm
... when there was some news about their situation with respect to some of their operators in the United States or to, with AT&T. And what we said at that time is what our customer had said publicly is that it was, you know, it was a mid-single-digit type of impact, mid- to high-single-digit impact for them, and that that was gonna take two or three years for that business to sort of transition down. And then what we've said, when we look at it, again, at kind of normal run rate, it's a mid-single-digit % type customer of ours. So if you take a mid-single type times a mid-single, just take five and five as an example, you know, you're talking about 25 basis points of-
Yeah
... of impact. So it's very de minimis, and on top of that, we do have content at the other supplier. So, you know, it, it's not—it's, we can manage it. I think in the scheme of the lumpiness of that business, Harlan, I think this is not something that we're concerned about.
Yeah, no, and I think to your point, I think the team has a pretty strong portfolio of design wins with all of the remaining 5G infrastructure players, right? Samsung, Ericsson, ZTE in Asia as well. So, very, very strong position overall in that space. So it ends up-
Yeah, strong-
Probably, probably being a zero-sum game.
Yeah. Yeah, strong and diversified, and we've... You know, because we didn't have a lot of 4G legacy, we really bucked the cycle on kind of telecom spending for several years, right, you know, relative to some of the other semiconductor companies that had a lot more kind of 4G and 5G together. So-
Yeah
... that drove some of our outperformance. But, you know, it's a business we're managing cyclically right now, and we're just focused on the long term and the design win position and driving the roadmap.
Perfect. And then just the last segment, which is still seeing some, I think, cyclical headwinds, right? Your data center, HDD, SSD storage controller business, you know, shipments have inflected, right? They grew slightly sequentially last quarter. I think they're gonna grow slightly sequentially this quarter. But it's still trending about 40% below your normalized historical kind of revenue run rate. Now, flash fundamentals, right, we cover Micron, we cover Western Digital. Flash fundamentals are starting to improve, and more importantly, it looks like your HDD customers are getting more confidence on nearline cloud HDD sequential growth going forward. Are you starting to see signs of a more pronounced and sustained recovery within storage?
Yeah. We are very pleased to see the end market data points that started flowing through at the end of last year. We're hopeful that those continue in terms of the end customers of ours bottoming out and then resuming growth in exabyte shipments and unit shipments and so forth. So that's encouraging. I think we don't have any other update for us at this time. What we had said was, you know, we had thought when we went back, you know... Again, we tried to call the ball earlier this year in March-
Mm-hmm
... but for the full year, we sort of said, "Look, we think probably by Q4, that business kind of comes back to some, maybe not the full way, but, but pretty much, you know, clears inventory by then, and then we've got growth." That didn't happen. We said it, it slid right, and we've continued to... You know, and we've not attempted to call the ball on that one again. We're gonna just let that play out, Harlan. As the end market recovers, we'll eventually, all- whatever inventory was out there of our components and their supply chain will clear, and eventually, this will, this will, this will resume. We'll get into it when we talk about data center, but, you know, that was within our data center segment.
Mm-hmm.
Total storage was about an $800 million a year business, and at some point, it was well below $100 million a quarter. It's come up off of that, but to your point, it's still, still down-
Yeah
... as we talked about Q4 and even into the first quarter.
... But, but Harlan, to your point, we, we have also seen on a sequential basis, we have also seen our revenue from storage and data center tick up, right? So for us, Q1 was the bottom, which was well under $100. It has ticked up in the next few quarters, so we are seeing the same thing. But I think to the point we've all made, collectively, it hasn't come back anywhere close to where the normal run rate was, and I think that's a combination of excess inventory still being consumed at the large cloud players, as well as I think we've all heard some level of, you know, try to recover old storage capacity. You know, do, do some level of recovery, as they need to manage their CapEx, but that's again, a one-time temporarily.
You know, you can only do that one time, right? So I think as that plays out, sometime next year, sometime, you know, I guess now it's this year, you're gonna start to see a snapback, right? And I think that's what we're all looking for. And I think in the meantime, the other good news is the inventories and supply chain over time have come down, right? Just inevitably.
Yes.
We've been undershipping, our customers have been undershipping. So I think once that inventory overhang is really starting to get a lot smaller, and as the demand starts to come back, I think that's also what happens very quickly once it starts to happen.
Yeah, and the good thing is, based on our research efforts, I know you guys have talked about it, but the design win pipeline on storage is quite strong, right? As the industry moves on the flash side to PCIe Gen 5 controllers, on the HDD side, moving to 20 TB plus type storage platforms, I think you guys have a very strong design win portfolio. So when it does come back, I think it's gonna come back quite strong, as you mentioned, and, like I said, the team is extremely well positioned there.
Yeah, storage is a franchise for Marvell. We have leadership technology there, leadership market share, and it's a fundamental endemic part of our data infrastructure strategy, you know, in terms of having storage as a key leg to the stool.
We've always said, you know, that we don't, you know, we don't like the volatility of that business. This goes back to even the first day I became CEO, and we were just way over indexed in this area.
Mm-hmm.
You know, a lot of the M&A work we even did and the diversification of Marvell wasn't to say, "Let's de-emphasize storage," it was, "Let's get the whole company bigger so that that portion of it becomes a smaller part of the total," if you will. But if you talk to the team that's driving the new products there and the design wins, to your point, there's a very robust design win funnel, and new products and technology are gonna be required at storage, for sure. It's just not growing at the same rate as others, and it's got more volatility to it, but it's an important business for us, for sure.
I appreciate the insights there. Okay, so let's move on to the strong growth vector, right? That is your data center business. Again, more than half your revenues, strong sequential and year-over-year inflection over the past two quarters on AI and cloud spending growth, expected to continue through this calendar year by our estimate. So Matt, maybe you can just help us with a rough sizing of the different products you sell in the data center market, how that's evolving with the larger role of AI, and then can you just also update us on, you know, what you're seeing more broadly for your AI and your custom silicon revenue outlook for this year?
Yeah, yeah. Great set of questions. Maybe a couple points up front. So we're very pleased to see this business achieve the revenue concentration that it has. We've always targeted this business, even if you go back to our 2021 investor day, right? When we laid out kind of the design win opportunity, the cloud-optimized leg to it, Inphi coming in, you know, and just the market growth rate and size of the TAM, driving this to half the company revenue was always the goal, and in fact, at the end of calendar 2021, before the 2022 kind of reset started to happen, that business was like 42%-43% of revenue or something, and then-
Yeah
... we went through some transition. So that's good. That's actually by design, if you will. That's not. I mean, sometimes you get there the way you don't want to, you know, like have the businesses go down and, you know, your mix shifts, but the 50s is actually a good number. And we'll talk about the AI portion of that, which is super exciting. But if you just kind of normalize and then just in the short term, as you said, you know, the business is on fire.
Mm-hmm.
I mean, in Q3, it was up over 20% sequentially. We guided the fourth quarter up, kind of mid-30s, the total data center segment, sequentially, right?
Yep.
So, so huge, huge, huge sort of strength heading into this next calendar 2024, which is great. So if you take Street and you look at this past year, let's call it our fiscal 2024, total data center is about $2.2 billion. The way to think about that is that the largest portion of that is optics, okay? And that's-
Mm-hmm
... you know, even a little bit more than half of that, and that's, most of that's, you know, a lot of that's driven by AI, but also the traditional PAM stuff. So again, thinking about buckets, you've got the optics at over half the business. Then we talked about storage, and while that, you know, was an $800 million a year run rate, it went down. We had that whole discussion, so that's the other-
Yeah
... that's the other you know, kind of foundational layer of our data center business.
Mm-hmm.
But the third one to think about is, think about this as kind of connectivity that's not optical. So these would be like management switches that sit on NICs or PHYs, retimers, gearboxes, NICs and Smart NICs, and then also the Innovium switch products in that bucket.
Yes.
So think of it as this connectivity that enables data movement, you know, in the data centers, and that, and within that bucket, the biggest one is the switch portion of it.
Mm-hmm.
Okay? And that's done extremely well for us. We're very pleased with that acquisition. We're driving a very aggressive roadmap in that area. We had said, you know, prior to the break, that business was gonna see strong growth this year we're in now, right, from last year. And our five nanometer product's doing really well. So that's actually a pretty—it's the category people don't talk about as much, and I think there's gonna be more to come this year in that area, Harlan.
Yeah.
But, but, but I think, you know, that's, that's one that we're very bullish on. And then the 4th group, which has a lot of kind of current investor interest, and certainly we're, we're, we're completely, you know, laser-focused on, is the cloud-optimized bucket, the custom silicon piece. And maybe just to contextualize a little bit of that, so, so on that one, the big driver there is obviously AI, that we've talked about, for, for the cloud and custom silicon portion. And if you go back, you know, to kind of where we sized that opportunity a few years ago, we said that's about $800 million a year run rate, you know, sometime between fiscal 2025 and 2026. That was our 2021-
Mm-hmm
...investor day sizing of the custom opportunity. We expect that with the current wins we have and the production plans from our customers and our forecast, that we will significantly exceed that number, okay? So that's super positive in terms of where that business is headed from where we thought just a few years ago. Now, what we didn't anticipate a few years ago was the AI portion would be so big, so that's kind of the overshoot, if you will. And, you know, we don't even know how to cap it at this point, but it looks very encouraging to us.
And I think at some point, you know, if you kind of roll out a few years from now, I think the custom silicon versus the optics contribution, you know, probably is similar, or certainly I think the custom group is gonna challenge our optics group, would be my point.
Yeah
... in the data center area. So that's all looking really well. Now, maybe just a couple more points, and then we can kind of go back and forth on a little more Q&A. On the AI piece, just to kind of, again, recap where we were and where we are, in the March investor earnings call, we talked about, you know, kind of baselining the AI revenue now for Marvell. So let's talk about the biggest piece of our data center, which is AI. At about $400 million this year, and they-
Yeah
... Or sorry, in fiscal 2024, and then-
Mm-hmm
... fiscal 2025, $800 million. And that was mostly, the $400 million was mostly driven by optics, okay?
Yeah.
That was what we said then, just to kind of baseline. And then what we said is, hey, the next fiscal year, when it goes to 800, you know, the optics was also gonna have good growth. And so just to kind of sift through it all, investors, and what we signaled was, you should assume there was probably $a few hundred million of custom in there. That's kind of what- that's the inference, right? If you said $400 million, it's mostly optics, that grows, and then the rest of it is custom. So I'm just kind of now baselining on what we said and where we were at that time.
So since then, okay, both in the subsequent two earnings calls and, and as time's elapsed, you know, that opportunity for AI overall has substantially surpassed our prior projections, okay, both for this year and for next year, and in both the optics area and in custom silicon. Optics this year and next year, and then custom silicon, obviously, for, for next year. As far as the kind of what's underneath it, so yeah, like on the optics side for this year now, we're, we're basically blowing through that $400 million, right? We're at- we said we're $200 million plus next quarter, or the- I'm sorry, in Q4, with that being pretty much all AI optics in the 800 gig platform that we have. So that's doing very well, and we expect overall optics to grow again next year, AI optics.
On the cloud-optimized side, we have two key chips that are really gonna drive the revenue in fiscal 2025. Both have taped out, huge customer pull to get these products into production, okay? We've had teams working kind of nonstop, depending on where the chip is at, but basically, both of them are out of the fab. They're at different stages of qualification, which we can get into, but one is a little bit earlier, and one's gonna be a little bit later through the second half of this year. But both of those are on track. They look really good. We have strong forecasts from our customers for those. We have build plans that we're executing to. We're working the whole supply chain to make sure we have what we need.
And so when you kind of add it all up, and you look at it from that perspective, the overall $800 million that we were sizing for AI just a few quarters back for fiscal 2025, we're just gonna completely blow through that. Now, how big can it be, and where is it gonna land? You know, we're still finalizing that, but it looks very, very positive, both on customer forecast, pull from sort of their end customers, and by the way, also, and even over the break, you know, really continued positive sort of milestone achievement on the NPI to ramp these to production. So we're very bullish, and I guess the final point I'd make is, that's all great for fiscal 2025, but depending on when they exactly go into production, we're working on that.
But fiscal 2026, you actually get almost a full year then of full, full revenue, which is actually gonna be even higher, right, than it would be for 2025. And those programs only continue right in the out years, plus there's next generation design activity we're working on. And so it's just a very, very exciting time right now for Marvell. Overall AI, cloud silicon, our optics, and then all the other pieces, too, underneath. It's, I think we're very, very well positioned in this area.
Perfect. Yeah, I apologize, I dropped off for a bit there. So if I ask some of the follow-up questions that are repetitive, just let me know. But as we enter calendar 2024, and just summing up everything that you've said, right? Talking about the growth tailwinds in data center, some of the cyclical headwinds in the business, overall design win pipeline remains strong. I think if I look at the consensus estimates from Marvell, right, they're modeling the business to decline sequentially 3%-5% sequentially in the April quarter, but drive sequential growth thereafter, with calendar 2024 revenues expected to grow sort of high single digits year-over-year. I'm not asking you to endorse the consensus numbers, but directionally, is this how you see the business trending this year?
What are the programs where you have the highest level of confidence on growth in calendar 2024?
Yeah, gotcha. Yeah, you, you missed some, you missed some great stuff. That's too bad you dropped, but I think everyone else heard it, so don't worry about it. Yeah, I think the way to think about it is, yeah, we see data center year over year driven by AI, but also the just the traditional cloud infrastructure stuff growing very strongly-
Yes
over the year. Very strongly. In particular, in the back half because of some of these cloud-optimized programs kicking in. And that sort of then, you know, kind of correlates to the overall Marvell picture, right? Just because if, if you took overall Marvell for a second, right, you've got this first half with some of the other markets we talked about the beginning being down, but, kind of overall, so that'll... Whatever. That's gonna do whatever that's gonna do. But from a full year perspective, yeah, we see data center up huge year-over-year and, driven by AI.
Again, the buckets being the AI optics and custom silicon, and then also, at some point, you know, probably, you know, later this year, and certainly into fiscal 2026, you've also got switching opportunities coming in, AECs. You know, even out beyond that, you've got things like CXL, so kind of a very nice setup for 2025 and 2026, and then beyond.
Perfect. And then for calendar 2024, can you just give us a sense on the AI revenue mix? I think you're gonna bucket it as you've got your optics business, right, 800 gig, 1.6 T. You've got your data center interconnects within that are tied to AI build outs, right? And then you also have your custom AI ASIC business. So I'm assuming that a majority of your AI revenues in calendar 2024 are still gonna be optics versus ASICs, but I don't know if you can give us a rough sizing of that mix.
Yeah, well, I-
Maybe-
Yeah, what you wanna... Yeah, go ahead, Ashish. I'll, I'll wait.
Yeah, I think, Harlan, quite frankly, when we gave our original estimates of $400-$800 at that point in time, clearly the assumption was that the majority of 800, you know, would still come from optics. But as Matt said, and you probably missed a little bit of that, you know, what we anticipated for custom silicon revenue a couple of quarters back versus what we are looking at now is a very different picture. We're anticipating a lot bigger from custom silicon, right?
So while I suspect optics will still be the majority, right, of revenue in fiscal 2025 within AI, it's not as imbalanced as what we probably thought it was a few quarters back, 'cause I think our level of confidence, the orders we have from customers on custom silicon is significantly higher, right? So I think one point is from a mix perspective, while optics remains, clearly that's a high volume business and flow today, still would be the bigger part of it. I would say I wouldn't underestimate the amount of opportunity we have from driving custom silicon revenue, even within fiscal 2025, right? So that's one point. I think in terms of the mix within optics, I would say the majority will still remain 800G PAM4. We've seen very nice traction on DCI, right, which is connecting data centers together.
I think that continues, but remember, our PAM products are tied, you know, literally more than 1-to-1 with accelerator units. And as accelerator units keep driving higher volumes and into next year, we're tied, you know, very linearly into that, that model, right? So that's why PAM really will be the majority. And then later in the year, we'll also see some uplift from 1.6 T, which obviously has a nice impact on price. So that's the way I would think about our AI revenue as you think about fiscal 2025.
Perfect. And then again, I think you might have mentioned some of this, but I'm gonna ask it again. You know, you guys have been talking about two AI compute ASIC programs contributing to the growth, right? You're actually ramping wafer starts now for the first program. I think, Matt, you talked about some of that coming out of the fab, different stages of qualification. That's for the first program, I assume. When does the next AI compute ASIC program start to ramp for the team?
Yeah, the way I would kind of frame this one, Harlan, is the design win and design pipeline activity is off the charts right now.
Mm.
And I think what's happened really in the last year... I mean, not even, well, certainly with the GenAI sort of somebody, you know, lighting the fuse a year ago, but really in the last six months, I think every major company in the ecosystem, you know, the cloud guys themselves, the server companies, the, you know, people that are also making their own GPUs. I mean, there, there's just been a flurry of activity, right? In terms of-
Yes
... where we can participate. And I think a couple of themes have emerged. The first is that the old cadence of what was sort of the AI cadence, you know, let's do one every couple of years and do one versus, you know, one after the other. There's gonna be a much faster beat rate, is our view in the industry. Now, the biggest company there, you know, NVIDIA's already doing that. I mean, it's we've talked about that publicly. You know, there's just such demand for the product and the diversity of options. So you're gonna see more SKUs, is our view. There's more interest there. We're It's actually something we don't talk about as much.
We have, well, we have, you know, a clearly huge revenue driver this coming year and the year after on the current programs we have.
Mm-hmm.
We're also engaged very broadly across the ecosystem. I mean, remember, we're not. We have a custom piece, which is we can make somebody else's AI-
Right
... for them, or their own compute for them. But also, we can participate in other platforms that are being developed from a networking perspective, from an optics perspective, and I'm talking about other, other chip companies, other server manufacturers, or people that are making AI boxes, things like that. So it's much broader than just, "Hey, can you win one more, you know, one more SKU at one more cloud company?" I mean, we're clearly gonna go do that, and the, and the opportunity is very large. But I would just say, when we look at kind of our AI task force and kind of our AI opportunity, it's not just a couple of big companies in the U.S. that are, we're doing their own chip for them, if you know what I mean.
I think there's a whole AI ecosystem that's gonna get built out, and that's where kind of our connectivity portfolio really comes in. So going back to it, strong design win pipeline and opportunity set at really 3nm now for the custom side. And then on top of that, trying to couple our full solution in terms of networking products, connectivity products, and all those other things like AECs, as an example, that really help the whole system come alive. And that's what we've been talking about, is accelerated infrastructure, right? Is all the chips that go around it and the solutions to make, to kind of unleash the power of AI. So, that I think we're gonna. It's gonna be a very busy year.
Those design decisions are all full bore right now, and everybody's trying to figure out how to accelerate their, their NPI, right, to, to really meet the needs of the market.
Yes.
Harlan, I think you had a question on the timing of the two chips we currently talked about. I think that was part of your question.
That's right. Yes.
So I think as you noted, the first chips kind of already launched into production, right? And we talked about that, you know, sampling a lot earlier, a couple of quarters back.
Mm-hmm. Right.
So imagine the second chip is following closely behind, right? It's, it's, it's actually fairly closely behind, and we feel very good about that as well. I think in aggregate, you should expect that the cumulative impact gets a lot more meaningful of both programs by the second half of the year. And then to Matt's point, by the time we get to fiscal 2026, you've got both these large programs in full flow. But really-
Yeah
... the cumulative impact really starts to get more meaningful in the second half of the year. The second chip is fairly closely behind the first chip, right? As we mentioned on our earnings call, we already have the chip back. We already went through initial testing. Things are looking very good. You should imagine that things are proceeding very nicely from a production point of view, even on the second platform.
Yeah, and then maybe just to emphasize that, 'cause we did get some interest—some questions at the end of the year, so maybe it's helpful for investors to kind of give our perspective on this broader call. I think first, this is a when, not an if, okay?
Yeah.
To be super clear. I mean, this is mission-critical type of stuff for our customers. We know that from the forecasts we're getting, we know that from the management meetings. So, there isn't any concern there. We just have to literally get these products, you know, into production. And the second question that we got from a lot of investors, some general, some specific, which was: Hey, is there a substitution risk on these? You know, is there a second source-
Yeah
... or is there another? The answer is no. I think, Harlan, you've been around the block enough to know on the custom side of the house, some of these chips we started working on, like, two, three years ago.
Yeah.
Right?
Yeah.
So, to go and sort of have a second one, it just in general in custom, it just doesn't work that way. The economics don't work out. Custom by nature is both companies just bet on each other, right, for mutual success and partnership, and you wanna be driving the next roadmap while you're doing the current one. So, to have... But that's a risk that I did get a few questions on that in December when we had a series of meetings, and I thought it would be helpful to clarify as well that the programs look rock solid, and they're... And the timing, you know, which is even on everybody's mind, we're gonna figure that out.
All the milestones look really good so far, but at some point it's gonna happen. That's kind of even why back a few years ago, we even said, "Look, it's gonna be in this fiscal 2025, 2026 timeframe, because it's, you know, you gotta figure out the exact quarter." But we've got a much better view now at the end of this year and through the break, and even kind of looking out to this year, of what that could look like, and we're adjusting our plans accordingly. Just to reiterate, if you go back to, like, the March call we had, you know, almost a year ago, these numbers are just way up versus what we had thought at that time.
Perfect. And you mentioned this, and we've talked about this before, right? Which is we're talking about the, you know, two ASIC AI, ASIC programs that are gonna be ramping this calendar year. But your leading-edge ASIC portfolio is much broader than that, right? We've always said that we estimate that the team has-
Hey, man, I think we may have lost Harlan. So I think, I think that his question was kind of broader on what we are doing in terms of our broad kind of custom silicon footprint, which stresses, you know, all the way into carrier and enterprise. Maybe you can-
Yeah
talk about that until Harlan comes back.
Yeah, great. Yeah, and I think just, I think he's frozen, so why don't we give him a compliment so he doesn't hear it? But yeah, I think, I think Harlan and the JPM team have done a great job, actually, over the years, sizing this custom market, which includes storage, it includes carrier, it includes enterprise, automotive, you know, data center, et cetera. And why I think that's a relevant question is, part of our strategy has been to develop technology platforms that address all those markets from a high speed, high performance perspective. And, you know, as a reminder, you know, we came out very strong on our 5 nanometer platform, with best-in-class SerDes performance and all the essential building block kind of IP and readiness. And that, you know, that won us a significant number of designs over the last few years.
We announced a couple of years ago we were gonna partner with TSMC again on 3 nanometer. That platform is progressing extremely well. We're driving design wins already in that area, in 3 nanometer for custom silicon. And we think this is gonna be a very sweet spot kind of cycle for us, going from 5 to 3. That's gonna last, quite frankly, a long time, because when you start looking beyond that, the technology's still a little unproven, and it's being worked on. And we're gonna be there, you know, from a technology leadership perspective.
But I think the investment that we've put in, both in five and now 3 nanometer, is set to pay significant dividends for the company in terms of leveraging all that investment over the next few years to really drive, you know, incremental revenue and, and design wins from it. And, and I think as you go farther and farther down on kind of the, the nanometer train, the higher the barrier to entry, both in terms of just cost and personnel that's required, and know-how and IP, and then the ability to actually manufacture it in volume and yield it, and, and all the things you've got to go do. And so we think that we've positioned ourselves as one of the very, very few companies in the world that can address these types of applications.
As it turns out, as you move a huge chunk of the TAM now from traditional computing to accelerated computing, a bunch of new TAM is opening up for... That is, I think, well suited to these high-speed, high-performance advanced node technology platforms that Marvell develops. And so we're very excited about the next few years and what that shift is actually gonna be able to do in terms of opening up a much larger TAM and opportunity set for Marvell.
That's great. Can you guys hear me?
Yeah.
Yeah, we can, Harlan.
Yeah, it's-
Yeah, I apologize, I apologize for the technical issues. I wanna move over to your non-AI cloud segment, right? Because after a period of customer inventory digestion, your tier one customers, Google, Amazon, Meta, they're spending, they're back spending again, right? Primarily to support more capacity on their 200 gig and 400 gig optical networking footprints. You still have one more tier one customer remaining to hire for the 400 gig networking upgrade cycle, along with the China cloud titans, the tier two hyperscalers. Have these customers contributed to the recent sequential inflection in the cloud business? If not, you know, when do you expect these customers to start their upgrade cycles?
Yeah, no, very, very strong growth we've seen in the second half in all those product categories you mentioned, but also including switching. And-
Mm-hmm
... you're right, there was some inventory build that happened because, as you remember, at the end of the prior year, right, there was a big kind of reset in terms of how some of those companies were positioning their spend and managing their OpEx, and, and their growth rates slowed. So, so that netted some sort of inventory that we felt strongly was gonna go clear by the end of the year, and quite frankly, it really cleared by Q3, because we started to see sequential growth in Q3 in standard cloud infrastructure. We projected it again to be up in Q4, and then for full year, next year versus this year, also to be up based on all those trends, which is basically all the inventory is clear.
People are investing in their core business still, because while AI is, like, super exciting and there's, there's a massive productivity sort of bogey out there, and the whole world is trying to figure out how to leverage it to drive their businesses, the traditional cloud computing businesses still have a lot of investment that's needed, right, to make them competitive. And so there's product cycles within that. As you mentioned, 200 gig and 400 gig, at some point, that's gonna go to 800 gig. There's DCI, there's switching. So we see, we see that as a continued growth driver market for Marvell. The AI piece is obviously turbocharging the whole data center segment, but we haven't seen really any... I mean, if anything, I'd say our prospects in the standard cloud infrastructure...
have gotten better because of where we're positioned, which is more on the networking and connectivity, which kind of by design has to get invested in if you're gonna put AI capacity inside the same tenant as your traditional computing infrastructure. So you're gonna have to have that same sort of networking overlay and connectivity. So that's all positive for us. AI's gonna drive the standard business as well, kind of indirectly.
Yeah.
Yeah, Harlan, just adding to that a little bit, I think as everyone's fairly aware, a lot of the investment today on AI has really been more on training, and, you know, inferencing is gonna start to become a lot more relevant as you move forward. Once you've trained your models, you need to go monetize that investment. I think this is where you'll see a huge expansion in bandwidth across even your standard cloud infrastructure. You're gonna have to deploy those models in real time. You're gonna have training information fed back in real time. So I think there is actually a big tailwind you should expect, even on the standard cloud side, as you start, you know, monetizing your AI investment.
That's something else to kind of keep in mind as you look forward for the next couple of years.
Yeah. We're running out of time, but I did want to touch on just two more topics, if you could allow me to. So Matt, you've mentioned this several times in our conversation. So on your cloud switching, your Teralynx family, right, of products, the team has been seeing very strong orders, very strong shipments. Beyond your one main customer, can the team provide any color on other customer design wins and adoption rates? And we know you're currently sampling your next-gen Teralynx 10 switching platform, that's 51.2T. You guys drove about $150 million in Teralynx revenues in calendar 2022. How big was that business in calendar 2023, and what does that look for this calendar year?
Yeah, great, great set of questions. Maybe I can summarize it somewhat here. So, yeah, and just for the broader investor community, so we acquired a company called Innovium, which brought us this kind of high-end data center switching architecture. Marvell already had a very successful enterprise and carrier switching business unit, which we then combined with Innovium to make one higher layer, higher layer networking platform group that is addressing, you know, switching needs across all those markets. We've shifted the focus primarily to data center. And to your point, we executed on our 5nm, 51.2 T switch we call Teralynx 10.
Mm-hmm.
That was the program that was in flight from Innovium. We sort of took two steps back to take a few leaps forward to make sure we got it on our Marvell Technology platform, including our own SerDes. So that product has been sampling. It's in customer hands. The design opportunity for this is very, very large because, one, 51.2 T is going to be a big, big cycle in the industry. And with AI as part of this, again, it's a little bit like we were talking about in custom silicon. There's gonna be, I think, a diversity of applications and a diversity of solutions that are needed. And with our position with these companies, these cloud customers, and the broader ecosystem, as I mentioned, we are a very viable option at this juncture.
And so there's no real update per se in this call, other than the you know, I think we're very glad we did the acquisition. We've got the... The quality of the chip looks excellent, and we're driving a very compelling roadmap, which, you know, Ashish and I have talked about as this makes progress over the year, throughout the year. We'll get that, we'll get that sort of better known as we kind of hit our own milestones and are ready to talk about that. But big investment from the company in this area. We have a dedicated general manager of this group who reports to me directly. That was a change we made last year. So we're all in on the switching area.
To your point on the run rate, yeah, when we acquired it, we said it was gonna be about $150 million of data center switching revenue.
Mm-hmm.
Executed to that, and now as we look to next year, it's gonna be quite a bit higher than that. And we said that that was gonna be driven both from standard cloud, as well as-
Yes
... a little bit of AI touching that too, but most of it is driven by standard cloud. So I think it's a real positive story there. More to come, but I think we're well positioned for the 51.2 T cycle.
Perfect. And then my last question, you know, the nice thing that I've always liked about the Marvell team is that at any given time, you know, the team is always looking forward a number of years, right? So for any given of your current platforms, you're probably already in design of the next two generations of platforms, and then you're looking to leverage and expand your core leadership in certain areas into new markets, right? And so with that being said, if you can just give us a quick update on some of your new and emerging growth initiatives, whether or not they're gonna start to contribute to revenues in calendar 2024. That's your AEC, DSP chip solutions, your CXL memory compute connectivity solutions, and finally, your 800 gig coherent DSP products for long-haul metro applications.
Yeah. Hey, Harlan, I think, Ashish, we have a VP of marketing job open, don't we? I think we... You can apply for that. I think that's basically you actually nailed it. I mean, those are the three that-
Yeah
... immediately came to mind. And just quickly on the AEC and active electrical cable area, you know, as that market's transitioning to PAM4-based modulation from NRZ this coming cycle, that's really where we're inflecting. And we actually said on the last call, it kind of got drowned out because of a lot of other topics from the call, but we do have, you know, design wins in that area that are gonna ramp to production this fiscal year that we're in right now. So that's been a great success story. That's a new emerging market, which will grow over time. The 800ZR, you know, we announced that in like October of last year-
Yes.
That was a roadmap acceleration we did. We actually inserted it in between the 400 to 1.6T DCI transition we were planning. We actually, because of our investment on the wired side, we have a coherent DSP, right? That works very, very well for DCI, and that should drive, you know, new growth opportunities for us. Because basically, if you think about it, at some point, there's been a big investment in training capacity everywhere, but as you wanna push, you know, push AI to the edge, and you need inferencing, you know, closer to where the work is being done, that's just gonna require a lot of improved connectivity between data centers. Especially as people go to regional data centers, there's also a trend where it's...
Even if you could build the data center, you can't get enough power to it, so people are contemplating building more clusters of data centers, so you need more connectivity in between. So that's sort of what the trend that's driving us to shift our roadmap. And just as a reminder, from 100ZR to 400ZR to 800ZR, we get a content-
Yeah
... per-port uplift each time because we're effectively, you know, quadrupling or doubling the capacity, so that's a positive trend. And then on CXL, yeah, we're driving some very key programs right now. That's still kind of in the future, but that has kind of not abated at all in terms of the need for disaggregating storage and memory in particular, and especially as you think about some of these AI systems and the amount of memory that's gonna be required, I think having some more flexibility in how people architect it with CXL, it becomes actually even more kind of urgent to make that happen. And so, yeah, we're pretty excited 'cause we have, you know, a lot of irons in the fire right now.
Mm-hmm.
You know, short-term growth is happening. Next year looks really good, and the year after, but to your point, there's a suite of things that we're doing. And then you should imagine, we drive a lot of innovation in the company. So there's all kinds of things we're working on that are beyond those other growth drivers that you mentioned, and that's kind of one of the shifts we've made, and maybe as we get to the close here-
Yeah
... you know, we did a lot of M&A 'cause we needed to build the portfolio that we wanted to really end up where we are today. You know, extremely well-positioned as sort of accelerated computing and accelerated infrastructure is in front of us. But we're now able, with all the kind of the key pieces, to really invest organically in some of these new areas, and we don't, we aren't as reliant on M&A as we used to be to go do that, and we're kind of just have a mindset we're gonna go control our own destiny here. And certainly, if things pop up or there's opportunities that fits that strategy, we would always look at that, and we're pretty good at M&A.
But right now, we're really focused on, you know, control our own destiny, drive our organic growth, and then with that, with the financial returns you can get from that, then drive, you know, significant shareholder returns and return to capital. So that's the spot that we're in right now, and that's why we're able to get some of these new initiatives going-
Yeah
... is because we can invest in it organically, 'cause of the business model we have.
Perfect. Well, this has been hugely insightful. Really appreciate the time spent with us and investors today. So Matt and Ashish, thanks for the participation today. I mean, we look forward to monitoring the progress of the team this year. Yeah, thank you very much. Appreciate it.
Yeah, appreciate it. Happy New Year, everybody.
Happy New Year.
Happy New Year.
Okay, bye. Take care.