Q4 2025 Earnings Call Transcript

Good afternoon and welcome to Marvell Technology, Inc.'s Fourth Quarter and Fiscal Year 2025 Earnings Conference Call. All participants will be in a listen-only mode. [Operator Instructions] After today's presentation, there will be an opportunity to ask questions. Please note, this event is being recorded. I would now like to turn the conference over to Mr.

Ashish Saran, Senior Vice President of Investor Relations. Please go ahead.

Thank you, and good afternoon, everyone. Welcome to Marvell's fourth quarter and fiscal year 2025 earnings call. Joining me today are Matt Murphy, Marvell's Chairman and CEO; and Willem Meintjes, our CFO. Let me remind everyone that certain comments made today include forward-looking statements, which are subject to significant risks and uncertainties that could cause our actual results to differ materially from management's current expectations.

Please review the cautionary statements and risk factors contained in our earnings press release, which we filed with the SEC today and posted on our website, as well as our most recent 10-K and 10-Q filings. We do not intend to update our forward-looking statements. During our call today, we will refer to certain non-GAAP financial measures. A reconciliation between our GAAP and non-GAAP financial measures is also available in our earnings press release.

Let me now turn the call over to Matt for his comments on the quarter. Matt?

Thanks, Ashish, and good afternoon, everyone. For the fourth quarter of fiscal 2025, Marvell delivered record revenue of $1.817 billion above the midpoint of guidance, growing 20% sequentially and 27% year-over-year. Our data center end market was the primary growth driver, fueled by strong AI demand and execution. In addition, we saw continued demand recovery across our multi-market businesses, including carrier, enterprise, networking and automotive and industrial.

I am pleased to report that we achieved GAAP profitability in the fourth quarter and expect this to continue in fiscal 2026. Our record non-GAAP earnings per share of $0.60 exceeded the midpoint of guidance growing 40% sequentially. This earnings growth rate double our top line growth rate underscores the substantial operating leverage in our business model. For the full fiscal year 2025, we delivered $5.77 billion in aggregate revenue with our data center revenue growing 88% year-over-year. We ended the year with our AI revenue substantially above our $1.5 billion target from April 2024's AI Day and we also expect to very significantly exceed our $2.5 billion target in fiscal 2026. Our overall revenue growth accelerated in the second half of the year, driven by our custom silicon program ramps along with continued strong growth in Electro-Optics.

In fiscal 2025, we also drove a record $1.68 billion in operating cash flow and significantly increased capital returns to our stockholders through stock repurchases and dividends totaling $933 million in aggregate. I'm extremely pleased with our fiscal 2025 results and even more excited about our outlook for robust year-over-year revenue growth in fiscal 2026. We are poised for a strong start to the New Year, forecasting revenue growth of over 60% year-over-year in the first quarter at the midpoint of guidance. Let me now discuss our results and expectations for each of our end markets. In our data center end market for the fourth quarter, we achieved record revenue of $1.37 billion growing 78% year-over-year and 24% sequentially.

These strong results were driven by our custom AI silicon programs ramping to high volume production. Additionally, we benefited from strong shipments of our Electro-Optics products and Teralynx Ethernet switches with revenue from both product lines growing double-digits sequentially on a percentage basis. Within our Electro-Optics franchises we continued to see strong demand for our market leading 800-gig PAM products and our 400ZR DCI products. We also began shipments of the industry's first 1.6T PAM DSP and 5-nanometer process technology.

To further optimize AI interconnect performance, we accelerated the cadence of next generation products by introducing the industry's first 3-nanometer 1.6T DSP featuring 200-gig per lane electrical and optical interfaces. This new Marvell DSP enables customers to reduce 1.6T optical module power consumption by more than 20% compared to its predecessor. And we expect to go into production in the second half of this year as 1.6T adoption accelerates. While tremendous focus remains on GPUs and XPUs, the distributed nature of computing AI makes connectivity just as critical as the individual processors.

As a result, the design of the next generation of accelerated infrastructure is inexorably tied to how efficiently data can be moved on and off the accelerators and throughout the cluster. This increase in the role of high speed networking and AI data centers is perfectly aligned to our strength as an industry leader. And just like compute, hyperscalers are also customizing networking, and we are seeing similar momentum for flash-based storage, HBM and CXL-based DRAM pooling. We'd announced the design win at Meta for a custom NIC at OCP last October, and we are seeing hyperscalers more broadly adopting similar strategies.

This is evident in our recent design wins, which now include multiple custom NICs as well as a follow on custom CXL memory solution. We are also enabling new interconnect technologies such as CPO and LPO for scale up fabrics and coherent-lite DSPs for emerging campus-wide large scale AI data centers. Earlier this year, we announced Marvell's Breakthrough Co-Packaged Optics architecture for custom XPUs, enabling customers to integrate optics into future custom accelerators. First demonstrated at OFC 2024, the heart of Marvell's CPO platform is our 6.4T 3D silicon photonics engine.

It integrates hundreds of active and passive components in a single unified device and builds on multiple generations of silicon photonics innovations that we have been shipping in high volume in our DCI modules for several years. Co-Packaged Optics can enable an increase in the size and scale of AI servers, which currently rely on passive copper interconnects. We expect this transition from copper to optical interconnects will significantly expand Marvell's interconnect revenue and market opportunities. We are engaged with customers to evaluate this advanced technology and we anticipate a multi-year period of trial system development ahead of wide-scale industry adoption for CPO.

Let me now turn to our current custom silicon programs. Marvell has successfully ramped highly complex 100 billion plus transistor XPUs and CPUs from initial samples to high volume production on first pass silicon. Our custom business continues to gain momentum as customers increasingly rely on Marvell to help them achieve their custom silicon ambitions. As I mentioned, our two leading AI custom programs are in high volume production, and we expect growth to continue.

One of these is a custom ARM CPU, which we expect we'll see expanding adoption in our customers' data centers. The second program is for a custom AI XPU, which is also performing extremely well with significant volume production ahead. In parallel, we are fully engaged with this customer on the follow-on generation of this XPU and planning for a production ramp once it completes its sampling and qualification cycles. As a result, we expect our revenue from custom XPUs for this customer to not only grow this year, fiscal 2026, but continue to grow next year fiscal 2027 and beyond.

Additionally, we are making tremendous progress with the new design win announced at our AI Day in April 2024 for a custom AI XPU with an additional US hyperscaler. Marvell's engineering team in close partnership with the customer has successfully completed a number of key technical milestones during the joint development process. As a result, we believe we are well on track to meet our customers' desired schedule to start production in calendar 2026. Engagement is also multigenerational and we expect it to result in a very significant amount of incremental revenue for Marvell over the next several years.

I'm very pleased with our custom revenue achievement for fiscal 2025, driven by multiple program reps. This success coupled with strong progress on upcoming custom programs gives us even greater confidence in our ability to achieve our long-term market share targets for custom revenue. We are continuing to invest in all aspects of our technology platform, including advanced process nodes, electrical and optical SerDes, high speed die-to-die interconnects, embedded memory, custom HBM, 2.5D and 3D packaging and silicon photonics. This week we announced the demonstration of the industry's first 2-nanometer silicon IP for next generation AI and cloud infrastructure.

Produced on TSMC's 2-nanometer process, this working silicon is a critical part of the Marvell platform for developing custom XPUs, CPUs, switches and other technology critical for next generation accelerated workloads. Now let me turn to our outlook for our data center end market for the first quarter of fiscal 2026. We forecast that the cloud and AI portion of this end market will continue to drive sequential double-digit revenue growth. For the on-premise portion of our data center end market, we expect the seasonal sequential decline in revenue to partially offset growth from cloud and AI. As a result, we expect our overall data center revenue to grow sequentially in the mid-single-digits on a percentage basis.

Now let me turn to Marvell's enterprise networking and carrier infrastructure end markets. In the fourth quarter, enterprise networking revenue was $171 million and carrier infrastructure revenue totaled $106 million. In the fourth quarter, we saw continued recovery in both of these end markets with revenue collectively growing 18% sequentially. Looking ahead to the first quarter of fiscal 2026, we expect aggregate revenue from enterprise networking and carrier infrastructure to grow sequentially by approximately 10%.

We are pleased with the continued recovery in these two end markets, although this forecast still anticipates Marvell products shipping below end market consumption. In the consumer end market, revenue in the fourth quarter was $89 million declining 8% sequentially. For the first quarter of fiscal 2026, consistent with our prior comments, we expect seasonality in gaming demand to drive a sequential decline in revenue from our consumer end market of approximately 35%. Over the next several years, we continue to anticipate our revenue from the consumer end market to be approximately $300 million on an annual basis.

Turning to our automotive and industrial end market, fourth quarter revenue was $86 million growing 3% sequentially as we continue to see a modest recovery in this end market. Looking ahead to the first quarter of fiscal 2026, we anticipate continued sequential growth in the automotive end market. However, we expect this to be more than offset by a decline in revenue from our industrial end market, where order patterns can be lumpy in any given quarter. As a result, we project our overall revenue from the auto and industrial end market to decline sequentially in the high-single-digits on a percentage basis.

In summary, we drove tremendous revenue growth throughout fiscal 2025, significantly scaling the company from an annualized revenue run rate of $4.6 million in the first quarter to over $7.2 billion by the fourth quarter. Building on this expanded base, we anticipate strong year-over-year revenue growth in fiscal 2026. Our AI-driven data center end market is expected to remain a key contributor further supported by the ongoing recovery in our multi-market businesses. Marvell's data center end market accounted for 75% of consolidated revenue in the fourth quarter. Reflecting this rapid transformation during fiscal 2025, we purposely redirected our investments towards data center relative to our other end markets to fully capitalize on the massive opportunity created by AI.

We recently evolved our organizational structure to fully enable this strategic transformation. All products focused on hyperscale customers are now managed by a single cloud data center group led by Raghib Hussain, our President of Products and Technologies. We have merged the rest of our end markets into a single multi-market business group led by Chris Koopmans, our Chief Operating Officer. Marvell has solidified its position as a leading provider of data infrastructure semiconductors with the unique business model spanning full custom to full merchant solutions.

We are seeing strong investment and accelerated infrastructure from both established hyperscalers and the number of well-funded new market entrants as they race to build million XPU training clusters. These customers are highly incentivized to increasingly use custom infrastructure to augment their merchant solutions. Recent developments in the AI market such as the advent of reasoning models are also expected to continue to driving strong demand for compute, networking and storage semiconductors. As a result, we remain very optimistic about our both our short and long-term growth prospects and our role in enabling accelerated infrastructure.

We look forward to updating investors on our business model and the significant opportunities ahead of us at our Investor Day on June 10th in New York. With that, I'll turn the call over to Willem for more detail on our recent results and outlook.

Thanks, Matt, and good afternoon, everyone. Let me start by summarizing our full fiscal year results. In fiscal 2025, Marvell delivered $5.767 billion in revenue with strong 37% growth in second half revenue compared to the first half. This growth was primarily driven by our AI programs in the data center end market as well as continuing recovery in our other end markets.

For the full year on a GAAP basis, our gross margin was 41.3%. Operating margin was negative 12.5% and loss per diluted share was $1.02. On a non-GAAP basis, our gross margin was 61%, operating margin was 28.9% and earnings per diluted share was $1.57. We delivered over 1,000 basis point improvement in our non-GAAP operating margin through fiscal 2025 from 23.3% in the first quarter to 33.7% in the fourth quarter. We also drove a record $1.68 billion in operating cash flow and significantly increased capital returns to our stockholders, returning $933 million through dividends and buybacks. Moving on to our financial results for the fourth quarter of fiscal 2025. Revenue in the fourth quarter was $1.817 billion exceeding the midpoint of our guidance, growing 27% year-over-year and 20% sequentially.

Data center was our largest end market contributing 75% of total revenue. GAAP gross margin was 50.5%. Non-GAAP gross margin was 60.1%. Moving on to operating expenses.

GAAP operating expenses were $682 million including stock-based compensation, amortization of acquired intangible assets, restructuring costs and acquisition related costs. Non-GAAP operating expenses came in at $479 million in line with our guidance. Our GAAP operating margin was 12.9%, while non-GAAP operating margin was 33.7%. For the fourth quarter, GAAP earnings per diluted share was $0.23. Non-GAAP income per diluted share was $0.60 reflecting sequential growth of 40%, which is double the pace of revenue growth demonstrating the significant operating leverage in our model.

Now turning to our cash flow and balance sheet. Cash flow from operations in the fourth quarter was $514 million. Our inventory at the end of the fourth quarter was $1.03 billion an increase of $170 million from the prior quarter to support the strong growth we are experiencing in our business. Our DSO was 51 days, decreasing by nine days from the prior quarter.

We returned $52 million to shareholders through cash dividends. In addition, we repurchased $200 million of our stock during the fourth quarter. Our total debt was $4.06 billion with a gross debt to EBITDA ratio of 2.06 times and a net debt to EBITDA ratio of 1.58 times. We have seen continuous improvement in our debt ratios as we have driven an increase in our EBITDA through fiscal 2025. We were pleased to receive an upgrade to our investment grade credit rating from Fitch in January, citing their positive outlook on Marvell's strong operating momentum from robust data center demand, structurally improved leverage metrics, strong market position and strengthened cash flow profile.

As of the end of the fourth fiscal quarter, our cash and cash equivalents were $948 million increasing by $80 million from the prior quarter. Turning to our guidance for the first quarter of fiscal 2026. We are forecasting revenue to be in the range of $1.875 billion plus or minus 5%. We expect our GAAP gross margin to be approximately 50.5%. We expect our non-GAAP gross margin to be approximately 60%.

Looking forward, we anticipate that the overall level of revenue and product mix will remain key determinants of our gross margin in any given quarter. For the first quarter, we project our GAAP operating expenses to be approximately $712 million. We anticipate our non-GAAP operating expenses to be approximately $490 million a modest increase of approximately 2% from the prior quarter. It is worth noting that this step up in OpEx is lower than what we would normally expect due to the typical seasonality in payroll taxes and employee salary merit increases in the first quarter.

This is because we expect more leverage from our custom model and expect higher NRE in the first quarter on a sequential basis. As a reminder, NRE is treated as contra OpEx. For the first quarter, we expect other income and expense, including interest on our debt, to be approximately $43 million. We expect a non-GAAP tax rate of 10% for the first quarter.

We expect our basic weighted average shares outstanding to be $867 million and our diluted weighted average shares outstanding to be $818 million. We anticipate GAAP earnings per diluted share in the range of $0.14 to $0.24. We expect non-GAAP earnings per diluted share in the range of $0.56 to $0.66. I am very pleased with Marvell's execution on all our key financial metrics throughout fiscal 2025. As we enter fiscal 2026, we intend to continue driving strong operating leverage and expect to make significant progress towards our long-term non-GAAP operating margin target of 38% to 40%. We will also remain focused on generating strong cash flow and returning capital to our stockholders.

I'm very excited about our future prospects and look forward to strong year-over-year revenue growth in fiscal 2026. With that, we are ready to start our Q&A session. Operator, please open the line and announce Q&A instructions. Thank you.

We will now begin the question-and-answer session. [Operator Instructions] Your first question comes from Ross Seymore with Deutsche Bank. Your line is now open.

Hi, guys. Thanks for letting me ask a question. Matt, I want to go back to the comments you had in your script about the confidence in growing with your lead XPU customer, not only this year, but next year, there's -- as I'm sure you appreciate, a ton of debate about a competitor taking sockets, et cetera, market share. How would you suggest we reconcile your confidence and their confidence simultaneously.

It doesn't seem like both sides could be direct.

Yes, Ross, thanks for the question. And I very much understand the debate that's out there, and I appreciate a lot the significant interest from the analyst community and our investors on this topic. So let me try to be as helpful as I can given the constraints we've got around customer confidentiality and their concerns about their own programs and their confidentiality. Okay.

So we're very pleased with the ramp of our current lead XPU program. We delivered this with first-pass silicon success. We see significant volume production on the current generation in front of us and you can already see the ramp in our numbers from starting in Q4. In parallel, we have been deeply engaged on the next generation of this AI XPU with this customer. I said this in my prepared remarks, but I'll just go over it again.

And we're planning for a production ramp once it completes its sampling and qualification cycles. So to be crystal clear, we do expect revenue from these custom XPUs with this customer, not only to grow in fiscal '26, which is the year we're in, but to grow in fiscal '27 and beyond. And I just want to also note that I'm not including revenue here that would be from other products such as networking or connectivity that would be incremental to custom. And certainly, we've got a great setup, I think, overall with this account relative to the opportunity set, but it's really on both.

And then with respect to the question that our customer may be working with someone other than Marvell on a next generation XPU, it's just something we can't comment on. What I can comment on is visibility for the products that we will be building for the customer. And as I said, this extends to the next generation of our current AI XPU program.

Thank you.

Your next question comes from Vivek Arya with Bank of America. Your line is now open.

Thank you for taking my question. Matt, just one clarification. If you could help quantify the mix between AI and non-AI in your data center for Q4 and Q1? Is it like two-thirds, one-third.

So any quantification there would be helpful. And then a little more generic question, which is that when we look at Q4 and Q1 data center and AI, they are very strong on a year-on-year basis, but the extent of the beats is somewhat modest when we contrast that with a very strong spending at your largest customer. So any commentary there would be very helpful? Is it a supply issue?

Is it just early days of the ramp. So any commentary there would be helpful. Thank you.

Yes. Thanks, Vivek. Maybe I'll answer the second part first, just on the overall setup. If you look at it, we've seen very strong growth the last few quarters.

I think really, really actually above overall market. I think we were up data center revenues, 25% sequentially Q2 to Q3, 25% Q3 to Q4. As I said in the prepared remarks, our AI and cloud revenue, as that portion of the data center business is up double-digits from Q4 to Q1. So these are all sequential numbers by the way. And of course, on the year-on-year, if you look at Q4, we're like up 77% or something in data center and I think Q1 were pretty similar. So we're very pleased with the trajectory of this business in terms of just the sequential and year-over-year growth rates.

I think it's -- that's tracking really well. On the AI side and how we think about it is and I'll just give you kind of maybe a flash point just as we sort of exit the year, but a good overall proxy. So today AI is more than is more than half the revenue. So it's now the majority.

It's crossed over. And I think as we go forward, that's going to just continue to increase. And so because overall Marvell is now 75% of our revenue kind of plus going forward in data center, that's going to end up being a really good proxy. Underneath that, about half of that is if you just look at overall data center now, just to provide a little more color.

About half of that is in Electro-Optics and custom now has grown to about 25%. And about a quarter, a little bit more of the data center revenue and then the balance is made up of everything else. So again, it's a nice trajectory here. We've got AI that's sort of outgrown the total.

It's the majority driving the business. Data center is now a significant total portion of overall Marvell. And then the mix continues to grow nicely with custom really ramping. So hopefully some of those data points are helpful, but we're very pleased overall with the trajectory of the business.

Thanks, Matt.

Can we get the next question please?

Operator?

I'm checking. Folks, please hold while we get the operator back online.

Your next question comes from the line of Timothy Arcuri from UBS. Your line is now open.

Thanks a lot. Matt, I wanted to clarify the answer you just gave on the breakdown of AI revenue. You said more than half is AI, which I get of data center. And then you said half a data center is optics. Did you mean half of AI's optics so that the AI breakdown is basically roughly half is optics and then half is custom ASIC. Is that what you meant?

Yes, sorry. Yes, Tim, thanks for the clarification to be precise. It's overall data center of those numbers I gave you.

Okay. And then you had guided AI to $2.5 billion this year. I think you said you used the word very significantly. You're going to exceed that in fiscal '26. What does that mean?

I guess most of the sell side is thinking like $3.5 billion. So that certainly would qualify as very significantly. But is that what you mean when you say very significantly by like $1 billion? Is that the type of thing that would qualify as that.

Thanks.

Yes, Tim. Thanks for the question. Yes, we're definitely leaving this year open-ended in terms of what we can go do. Last year, we had talked about $1.5 billion.

We blew through that -- this year, again, we anticipate being substantially above that. I'm not putting a number on it just yet. I think that there's a lot to go here in terms of the momentum in the business and the opportunity set in front of us. And so we're, right now, we're kind of keying off the last update we did, which was in the AI Day from last year and then we'll find the right appropriate time in the future.

But right now, we're trending extremely well. And you can really start to see it in the numbers. As I mentioned earlier, at some point, this data center revenue is now 75% of Marvell where I gave you the AI percent. So you could start using that as, I think, a really good proxy for where things are going.

Okay, Matt. Thanks.

Your next question comes from the line of Harsh Kumar from Piper Sandler. Your line is now open.

Yes, hi. Thanks for letting me ask a question. Matt, I think you mentioned that you have three kind of custom ASIC customers for you. I wanted to talk about, I wanted some color on the stickiness of these customers.

Kind of what are some of the triggers that might cause some of these customers to go look for another design partner? Is it price or is it just 100% performance if you can't produce and start looking and what is the second part of the question is, what is the possibility of you being able to attach your networking products to some of these custom ASIC players or if you do could you just help us understand what's being attached today?

Yes. Thanks for the question. I think the first is that we actually have custom engagements with all four of the major hyperscalers and then and then two of them in compute with a third coming just to clarify. And these, look, these are very complex chips.

And our view is consistent with sort of what we've been saying, which is over -- when you look at the TAM that's out there, which is, we sized back in April of last year for Marvell like a $75 billion data center TAM, which is probably, if anything, just gotten stronger and a big portion of that coming from custom. The reality is to service that the customer base is really looking for partners with a couple of key attributes. The first is technology leadership and technology investment and IP. And we just announced, as an example, just a couple of days ago, our working 2-nanometer platform, we've got a very competitive offering on custom HBM, die-to-die, advanced packaging.

So that's part of it. The second is to have the manufacturing scale, the ability to go drive the right design methodology in terms of first-pass silicon success and be able to actually ramp these highly complex upwards of $100 billion transistor type of chips into production. And they're also looking for partners that have the relationships with their suppliers like at Marvell as an example, we've invested heavily in the supply chain. And then finally, partners that are flexible in their business model that can look at different approaches to different chips all the way from traditional ASIC services to full-blown build-to-spec type of solutions and we provide all those.

So in the end, those are the types of different things. And when you sort of add all that up, there's really our view still is that it's ourselves in one other very large competitor of ours that can service that market. So these sockets are very sticky. At the same time, each generation, you've got to go bid and you've got to go compete.

But we've been very successful since we started winning these big custom silicon programs back in 2021, getting awarded those and now taking those into production and then looking at the next generation we're very comfortable with where we stand. And so there is a high degree of stickiness. There's a high barrier to entry. But at the same time, you got to win them each time.

And then the final thing I'd say is just from a design opportunity perspective, that pipeline just continues to expand. And within that, we continue to close very significant designs as recently even as our fourth quarter. So those are the dynamics we still see and it's competitive, and everybody is in a race, but we feel like we're in the best position we've ever been in, quite frankly, relative to the technology, the platform and our ability to go really drive in a very, very focused manner the execution around what I think is the biggest opportunity I've seen in a very long time in the semiconductor industry, which is the AI super cycle that's still in front of us.

Thanks, Matt. And then for my follow-up, I was curious if you could expand on your comment around, you said, I think, growth in data center, cloud, plus AI will be double-digit on-prem will be down. Can you help us think about what you mean by that? What is double-digits? Is it teens or higher than that? Just because investors are super sensitive to growth rate in your cloud and AI business?

Yes. I mean, we're getting a little more precise. And certainly, this is the guide, but double-digits is double-digits and it's not 20% or something. I mean, it's above 10%, and that's the way you should think about it.

And clearly we're -- we'll see where the quarter lands, but that's what we're seeing at the moment. You also asked me about attaching networking. So I just want to cover that real quick and we'll go to the next question. That has proven to be extremely sticky, obviously, and that is probably the other leg to the -- that goes hand-in-hand with the custom designs is that, I said some of this in my prepared remarks, but the value of the solution now is really moved just beyond who can do the best sort of individual compute piece of silicon.

I mean that's critical, and you've got to have the right sort of performance, and you've got to take advantage of Moore's Law and all those things. But the way to get ultimately outsized performance in the solution is you've got to couple that with the networking and the connectivity. And I think the IO from scale up to scale out to all the various different interconnect type of solutions we have is proving to be a very compelling reason for customers to use us because we can co-architect those together and we actually get with the compute custom silicon opportunities. And when we do that, by the way, and especially when we engage on something like an accelerator of CPU, we get very early information and perspective on the customer architecture.

Again not just the piece of compute, but then all the IO that's required and then we drive our solutions around that. So this is all by design in terms of how we formulated our technology platform approach. But it is a much more compelling value prop for our customers versus maybe competitors of ours that only have one of those pieces.

Very helpful, Matt. Thank you.

Thanks.

Your next question comes from the line of Aaron Rakers from Wells Fargo. Your line is now open.

Yes. Thanks for taking the question. I'm going to stick with the AI question as well. Matt, You had mentioned that you had four customers in the hyperscalers that you engaged and you mentioned two in compute and the third one coming.

The fourth, I just want to be clear, is that also a compute opportunity or is that referring to the NIC custom NIC opportunity that you're currently engaged in? And then also on that same narrative, how do you think about the custom logic opportunity around HBM 4 and the timing of when that might be impactful for Marvell? Thank you.

Yes. Thanks, Aaron, and great two questions. So, yes, on the first one, I think you got it right. You actually nailed it.

So that and that opportunity sets bigger than just that product as an example. But, yes, the fourth customer, we had a -- we're in production and we publicly announced with custom NIC in that area. And then on custom HBM, yes, that continues to be actually a critical IP for us and I think a critical part of what we see the customer roadmaps aligning around. And so we're very engaged right now both with the key memory partners of ours who we work with jointly with our hyperscale customers.

And the benefits are quite significant that we're able to propose in terms of the each front area reduction on the logic die, the ability for our customers to get significantly more compute packed into that die and then also improving the throughput between the HBM and the compute and then actually being able to have a higher density memory solution wrapped around our customers' compute die. So all those are still very valid and I think that's going to be a significant sort of industry trend that we're going to see in the next generation of these high end accelerator products. Thanks.

Thanks.

Your next question comes from the line of Tom O'Malley from Barclays. Your line is now open.

Hey, guys. Thanks for taking the question. Appreciate it. So I kind of wanted to pivot a little bit to the technology side.

So Matt, in the script, you talked about increasingly optical connections, taking over areas where electrical connections were today. So I kind of wanted to get your opinion like when you look at the world of just optics in general and particularly Co-Packaged Optics, do you see that at first happening kind of inside of the rack or do you see that kind of happening in the aggregation layer. And then what does that mean for your DSP business? Like obviously your share of the optical world is a little bit different than your share of the switching world.

I just would love your commentary on where you see kind of Co-Packaged optics coming first? And then kind of as it relates to your product set, where do you feel like you have some strength.

Yes. Great. Thanks, Tom. So, yes, maybe some perspective.

So I think the first is in the history of CPO, I think, this is really prior to this [indiscernible] change in the last few years of the growth in accelerated computing. The holy grail that's been out there for a decade plus, maybe close to 20 years is that the aggregation layer or the scale-out was ultimately could go to CPO. And there's been a lot of POCs in this area. There's been a lot of different efforts in the industry.

I mean I was at OFC last year. And that's kind of where a lot of the air really did come out of the balloon. I think not just from us, but our big cloud customers sort of have said that's a very long-term type of opportunity. And that's been our view, too, by the way, that it's something that we're paying attention to.

We've got some investment there. But what's really happened in the -- and by the way the same thing sort of happened with LPO, which was linear optics. 2023, there was a ton of noise on this topic that, that would be able to replace DSP-based pluggables. That has not happened. But what's emerged, as I mentioned in my prepared remarks, is that in the scale-up and the connectivity in the rack in between accelerators and inside the cluster.

That's where there's a lot of opportunity to go active and accelerated in terms of the connectivity and to go optical from passive. And that could be accomplished a couple of different ways and Co-Packaged Optics is a very interesting and compelling way to do that. We demoed our 6.4T product a year ago at OFC. We're definitely investing in this area and we have a lot of traction here relative to us presenting that as a solution as part of our ASIC platform.

And then also with LPO same thing. I think that may find a home in some of these connection points, again, within the rack. But these will be important applications, especially in closed systems. There's a lot of technology that needs to get developed and a lot of work to do.

But we're very much investing here and we will be in this business to compete with best-in-class technology. The question is, when does it want to get adopted? What is the qualification process, what are all the manufacturing and yield issues we're going to have to deal with. So it's complex, Tom.

But this is not something that we're backing off of. In fact, we think it's a critical part of our long-term technology investment to not only have best-in-class DSPs and pluggables, best-in-class AECs, as an example retimers, a whole suite of connectivity solutions, but also investing in the next generation where we can enable significant breakthroughs in power and performance. We just have to balance those off with the reality of this is very complex new and not traditional sort of basic CMOS technologies. These are more complex.

So you'll see more from us at OFC on this. And certainly this is an area to watch. But we think it's going to take some time for it to finally work its way through industry-wide into real high-volume production. Thanks.

Super helpful. And then just on the near-term, obviously, on the optical side, it sounds like things are better there. But there's been noise as there is about many things these days in relation to AI about potentially some headwinds in terms of inventory at 800-gig or the pricing dynamic as you move to 1.6T. Could you maybe comment on the health of both of those nodes and kind of what you're seeing in terms of the growth trajectory for the 800-gig world this year and just the health of pricing, et cetera, just because there's been a lot of noise there.

Thank you, Matt.

Yes, yes. No, we feel really good. Optics has been very, very strong for us. We saw a big uptick in the second half of last year in terms of orders.

We've been fulfilling those. We see demand very strong this year, particularly in 800-gig very healthy demand. We do see a 1.6T transition happening, but quite frankly, 800-gig continues to be the workhorse. And we'll be ready to support both our 5-nanometer 1.6T device, but we're very excited about our 3-nanometer product.

It's getting a lot of great traction because of the power savings. I mean it's 20% per link. So it really adds up when you talk about the massive scale and the number of interconnect points in these massive clusters that are being built, so all that looks very positive, Tom. And of course there's always noise in the system.

This is a very hot area. I'm saying AI overall. But from a Marvell perspective, sharing what we see, we see very healthy strong demand continuing.

Your next question comes from the line of Harlan Sur from JPMorgan. Your line is now open.

Hey, good afternoon guys. Thanks for taking my question. Great to see that you captured the follow-on AI XPU program after your current XPU program, which is ramping now with your major cloud and hyperscale customer. Matt, just to clarify.

So is the new follow-on XPU program, a training XPU as well? Is that a calendar '26 ramp? And is that at 5-nanometer or 3-nanometer? Any more color there would be helpful.

And then for Willem, your inventories were up 20% sequentially, which in a strong demand and product cycle environment like typically implies strong future growth, but you compare that to the 3% sequential total revenue guide for April. There seems to be some disconnect. So the way to interpret this is that the 20% sequential growth in inventories is more reflective of the AI strong growth profile ahead for the team?

Yes, I'll start off, Harlan. So, yes, so couple of things. So again given the confidentiality wrapper we've got, here we go. The first is you should assume this is a -- it's a very high volume program and it's a continuation of what we're doing.

It on the -- and you should assume just in general, on every next generation type of device, you're going to see no transitions and technology advancements as you go forward. So you should assume that. And then also on a timing perspective, all I can say there is we'll be ready to ramp when it's time, and we'll manage that transition. We're very confident in our ability to manage that transition successfully with our customer, but that timing is something that we're just going to have to see when that's ready and we'll time it around that.

And I can't really comment on what my customer plans are in this kind of detail, they just don't like it, and I don't blame them. And I'll give -- I'll let Willem take the inventory question. Thanks.

Thank you.

Hey, Harlan. On Inventory, that's really in support of continuing growth in our custom programs that we see over this year as well as very strong growth on the optics side. There's really those two dynamics driving inventory. But if you look at it on a day, it's actually flat quarter-over-quarter.

Got it. Thanks for the inputs, guys.

Your next question comes from the line of Mark Lipacis from Evercore ISI. Your line is now open.

Great. Thanks for taking the question. Matt, I think there's always a concern in the market around this risk of a deceleration in spending or maybe even a digestion period. And you seem to express confidence in the visibility as you go through this year.

And I wonder if you can help us understand the [indiscernible] that you look for on your dashboard that tells you that there's this risk of a digestion period. And I think maybe as part of that, I guess, I've always thought of Marvell as like a COM IC company. Now you just have this processor business that's really ramping. And I wonder if you kind of deconstruct that risk assessment?

Do you think about those two businesses differently?

Yes. Okay. Yes, I'll break it into the two pieces. So we do a lot of triangulation, Mark as a company relative to how we forecast our data center business in particular.

And so we do take a lot of the top-down data into account in terms of CapEx trends and things like that. But ultimately we're so joined at the [indiscernible] because of these key -- the key nature of these programs and ultimately our products and especially where they're just completely sole source being a potential gate for our customers. We've been -- we plan our supply chains very carefully together. And I think you can see the evidence of that over the last two years as we ramp with sort of the advent of ChatGPT.

And it's amazing to think of where we were if I actually go back to this call two years ago and the big breakthrough was we were saying we were going to do $200 million in AI revenue in calendar '23 and $400 million in '24. And now we just finished '24 and we're talking about blowing through $1.5 billion and doing $2.5 billion plus. So just as a quick side note, it's just kind of remarkable. But we scaled, right, along with our customers and generally continue to sort of exceed expectations and be able to ramp with what they need. So when we look to the rest of the year and even, quite frankly, on a multiyear view, it's a very strong outlook.

And some of that is our own specific things because we have new programs, let's say, in the custom area, that are ramping that are incremental. We also have -- we just have additional sockets. We have customers transitioning from a technology perspective, let's say, an interconnect. So we have a lot of positive tailwinds.

Now we're monitoring and certainly, the supply chain, it needs constant review. But at the moment, again, we see a very strong setup for strong growth in our fiscal '26 and our fiscal '27 driven by both the custom programs ramping as well as continued growth and strength in connectivity and in higher layer networking and in things like storage coming back. So all these look like a positive setup at the moment.

Great. Thank you. Very helpful.

Your next question comes from the line of Ben Reitzes from Melius Research. Your line is now open.

Hey, guys. Thanks for the question and thanks for the intermission there in the middle. That was a nice break. The question that I have is with regard to sequential growth, Matt, and then a long-term question.

Did you clarify that AI revenue could grow sequentially through the year? And did -- when Tim asked his question earlier? And if so what's your confidence? And then if you could just kind of step back also, Matt, a lot of noise due to the speculation around that customer and content there.

But when you step back, you have a goal of $15 billion for the data center long-term by calendar year '28 are you seeing the progress in your custom business still to hit that number because nobody on the street is even close to that. Thanks.

Yes, Ben, thanks for the questions. So, yes, on the assumptions through the year, yes, I didn't, I think what I said was in Q1 our data center business ex the on-prem stuff was going to be up double-digits, coming off of overall data center like 25% a quarter sequentially. So I didn't comment throughout the whole year, but you should just you should just assume. It's not a bad assumption, right, or would be a fair assumption to think that it's obviously going to continue given the strength in the business and the momentum that we're seeing and just from what we're looking at from a year-over-year perspective.

On the long-term, I think we're tracking extremely well to our 20% market share number. When we look back to kind of calendar '23 and then '24 and '25 and you start bouncing it up against what we said at the AI day. I think it looks very favorable. We're definitely gaining share from '23 to '24 and we'll definitely gain more share from '24 to '25. And so to get to that revenue target, you got to get there both ways.

You got to grow the share, right, from kind of 10%, let's call it to 20%. And then the market's got to develop, right, obviously. You got to get to the sort of $75 billion TAM, but both of those are trending in a very positive direction. In fact, certainly in '24 and even in '25 is just kind of big round numbers, it looks like both the market and our growth, if you just bounce it up against sort of what's the compounded growth rate you needed from '23 to '28, it's actually growing above that right now.

So now when we're in ramp phase, but that's going to continue. So I think we're in very good shape in terms of where we're tracking from a market share perspective. And certainly, if the market, by the way, is bigger. I mean remember, we gave that point of view in April of last year.

And the world has changed then since then in terms of the absolute CapEx that's being deployed, certainly, even in recent months, all of the various programs from all the different key players and the potential of what's out there, sovereign programs, government programs, programs from new entrants. So we just see this -- we just see the -- quite frankly the TAM and the opportunity for Marvell, if anything, being way larger than it was when we looked at it almost a year ago. So all those make me feel very good about the market size developing and then certainly our progression on the market share.

And you're right, that would be, it'd be an absolute home run to get there. And that's what me and my team are absolutely driving in this company day in and day out is drive the market share and help create the market, help make the TAM happen and execute like crazy and do it in a very focused manner, which is also why we reorganized the company with a dedicated data center engineering and business group to drive it. So I think the setup is really good. Thanks.

Thanks, Matt.

Your last question comes from the line of Christopher Rolland from Susquehanna. Your line is now open.

Hey, thanks for squeezing me in. And some of this was addressed, but scale up, you guys have a lot of opportunities here. Some of this could be optical, some of it could be Co-Packaged, some of it could be active copper. I think you might even have some like UAL or NVLink like alternatives, DSP throughout the system.

I guess as we see changes in architecture, either GPU or XPU, where do you think your biggest opportunities are? And when might we see some of these wins. Thanks, Matt.

Yes. Excellent, Chris, and a great question to end the session. So first of all you've done your homework. And I think the key message is we are planning on participating and enabling all of the things you mentioned, we have programs underway and developments in all those different areas.

The key thing I want to stress to you and to the investors on the call is this is incremental TAM the scale-up opportunity. We flagged it a few different times is right for sort of a TAM expansion. And it looks like that's what's going to play out at some point here. So that is all incremental.

That's not, hey, if that happens and all of a sudden, Marvell's DSP revenue goes down or something like the scale up and this type of connectivity that is a revenue upside, market upside type of opportunity. So that one, I think, is -- we're very excited about. And I think you're going to hear a lot more about that from us certainly at things like OFC and our Investor Day and so forth. But, yes, big opportunity there.

No question.

Thanks, guys.

Thanks, Chris.

Let me now turn the call over to Mr. Murphy for closing comments.

Yes. Thank you, operator. And first of all I appreciate everybody joining the call. I apologize for the gap there.

I hope everybody enjoyed a quick coffee break and we'll certainly figure out what happened. But I appreciate the interest in the company. I want to thank everybody for their interest and just kind of reflect on a few things. I think the first is we had really strong progress in fiscal '25. We started off at a $4 billion-ish, $4.2 billion, $4.3 billion type of run rate in the first half.

We're now at the $1.875 kind of guide, well above $7.5 billion kind of run rate. So that's, I think, really strong trajectory for the company heading into the year and a big change. We're also getting leverage in the model. And the operating leverage, cash flow generation, operating margin expansion, and you can even see it in things like our Q4 to Q1 OpEx, which normally would go up, as Willem said, more than we're forecasting, but we're starting to see the benefit of these custom silicon programs kicking in and the NRE contribution.

And so we're seeing it both on the top line revenue growth, but also on the expense line as well, and that is a great setup. We were GAAP profitable as well in Q4, which was great. We're going to keep driving that. It wasn't just AI, our standard cloud infrastructure side of things and data center also had a very strong year.

The programs that we maybe there was some concern about going back a year ago, can you ramp? These products are going to get qualified, what's going to happen. I think our team and I want to thank the Marvell team did an excellent job of ramping these very complex devices into full volume production, supporting our customers. Tom had a question about Electro-Optics.

We had a banner year for our Electro-Optics business and it looks very strong heading into next year. Also the multi-market businesses that was a source of concern over the past year, the enterprise and carrier and things like that. You're seeing very strong sequential growth now, sequential growth out of those businesses as they recover. And they sort of trend back towards their at least enterprise and carrier their $2 billion run rate.

And look we had a beat and raise. We're guiding revenue and earnings growth to continue in the first quarter. Customer engagement is really strong. As I mentioned, very strong design wins in the last year, great execution by our sales team and business units to win those key designs that are going to drive our future growth.

And we're very optimistic about where this heads. So I appreciate everybody's interest. I look forward to seeing all of you in various investor meetings coming up and continue the discussion. So thanks everybody.

I appreciate it.

Ladies and gentlemen, this concludes today's conference call. Thank you for your participation. You may now disconnect.