Good afternoon, and welcome to the Marvell Technology Incorporated Second Quarter of Fiscal Year 2024 Earnings Conference Call. All participants will be in 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 this 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 second fiscal quarter 2024 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 available in the Investor Relations section of our website. Before I turn the call over to Matt for his comments on our performance, let me highlight several new product announcements, starting with our electro-optics portfolio. First, we announced the release of Orion, the industry's first 800-gig coherent optical DSP for pluggable modules. Orion is our fifth generation of coherent DSPs and is produced in 5-nanometer technology incorporating our 112-gig SerDes, enabling 800-gigabits per second of throughput within the tight power and space constraints of small form factor pluggable optical modules. Orion enables high-performance probabilistic traffic shaping in addition to supporting standards-compliant transmission modes. We expect that Orion is well-positioned to continue to drive Marvell's leadership in coherent technology in both the carrier optical transport market as well as data center interconnects, or DCI, in the cloud market. In parallel with Orion, we launched the industry's first 800-gig DCI ZR module, our COLORZ 800 platform. We will be demonstrating our 800ZR module at the ECOC Conference in October. Marvell pioneered DCI…

Thanks, Ashish, and good afternoon, everyone. For the second quarter of fiscal 2024, the Marvell team continued to execute, delivering revenue of $1.34 billion. These results were above the midpoint of our guidance, primarily driven by demand from AI applications growing faster than our prior forecast. Our non-GAAP operating expenses were better than guidance due to an acceleration of the cost reduction plan we outlined last quarter. As a result, our non-GAAP earnings per share was $0.33, $0.01 above the midpoint of our guidance. We are pleased with our performance for the quarter in a challenging macro environment. Let me now move on to reviewing our results and expectations by end market, starting with data center. In our data center end market, revenue for the second quarter was $460 million, growing 6% sequentially, well above our guidance for a flat outlook. We were able to outperform our guidance in this end market because of accelerating demand for optical products to meet the continuing expansion of cloud AI deployments. Our overall revenue from cloud grew over 20% sequentially. Notably, revenue from both cloud AI and standard cloud infrastructure grew sequentially, with AI growing faster. As expected, revenue from the enterprise on-premise portion of our data center end market declined significantly on a sequential basis in the second quarter, reflecting a weakening enterprise market. As you heard in detail last quarter, AI infrastructure requires a staggering amount of high-bandwidth connectivity, best provided by an optically connected infrastructure operating at the highest available speeds. Marvell is enabling AI with a broad range of solutions, which include: PAM4-based optical DSPs and AECs for connecting accelerator clusters inside AI data centers; DCI products for connectivity between regional data centers; low-latency high-capacity Ethernet switches for fabric connectivity inside data centers; and custom silicon for compute acceleration.…

Thanks, Matt, and good afternoon, everyone. Let me start with a summary of our financial results for the second quarter of fiscal 2024. Revenue in the second quarter was $1.341 billion, exceeding the midpoint of our guidance, declining 12% year-over-year and growing 1% sequentially. Data center was our largest end market, driving 34% of total revenue. Enterprise networking was the next largest end market with 24% of total revenue, followed by carrier infrastructure at 21%, consumer at 13%, and auto/industrial at 8%. GAAP gross margin was 38.9%. Non-GAAP gross margin was 60.3%, growing 30 basis points sequentially, driven by cost improvements, partially offset by a weaker revenue mix. Looking ahead, we expect gross margins to continue to improve in the third quarter and then to increase significantly in the fourth quarter. As Matt told you, the recovery in storage continues to push out, which is negatively impacting our product mix. However, in the fourth quarter, we project a significant improvement in our overall product mix to lead to stronger gross margin. We expect this improvement will be driven by our continuing growth in data center, while wireless carrier and consumer revenue declines on a relative basis. In addition, the Marvell team continues to execute well on our efforts to reduce costs. As a result, we continue to target non-GAAP gross margin returning to the bottom end of our long-term model of 64% to 66% in the fourth quarter. Moving on to operating expenses. GAAP operating expenses were $727 million, including share-based compensation, amortization of acquired intangible assets, restructuring costs, and acquired -- and acquisition-related costs. Non-GAAP operating expenses were $448 million, $7 million below guidance. We are pleased to report that we accelerated our cost reduction plan we outlined last quarter. We remain on track to execute the remainder of…

We will now begin the question-and-answer session. [Operator Instructions] Our first question today is from Tore Svanberg with Stifel. Please go ahead.

Yes, thank you. Matt, could you just elaborate a little bit more on the storage business. I am just trying to understand all the dynamics there. So, it sounds like there is a delay in the recovery, obviously, because there's still some inventories out there. But I'm also trying to understand the growth in storage associated with AI. I know these are different architectures, right? But if you could give us any more color on the digestion of the inventory on the compute side, coupled with how storage could benefit more on the AI side?

Yeah, sure. Hey, thanks, Tore, for the question. Yeah, on storage, I would say the best guide post is to look at the end customer commentary and both what the hard drive companies are saying as well as the flash industry, and just relative to the sort of extended time it's taking for inventory to work down at their level. That being said, eventually, it's going to come back. It's a little bit longer than we had modeled, but at this point, we're following the market and following our customers, and hopefully that's sometime at the beginning of next year, in the first half, but we'll see how that goes. But eventually, it will come back. On the AI question, it's kind of interesting. We continue to make attempts to model that. We don't have a great model for storage at this point. We've got I think an excellent view now of where we sit in the block diagrams relative to our optical products, our custom silicon, networking, but the storage impact, in our view, at a high level is net positive, but we don't have a particularly helpful model to how to think about that. Certainly, AI should be an overall tailwind for storage. But I think there is -- that's all caught up in the inventory that's being worked out at the CSPs.

Sounds good. Thank you.

Yeah.

The next question is from Timothy Arcuri with UBS. Please go ahead.

Thanks a lot. Matt, so AI revenue was supposed to be I think $400 million this year, it's actually going to exit at $200 million a quarter. So, it's going to -- it's, obviously, going to be higher than $400 million for the year. Can you talk about how much of whatever the number will be for the year? Probably $500 million or $550 million something like that I would think. How much of that is going to be custom ASIC? And then next year, you said that AI would be greater than $800 million. Obviously, it's going to be quite a bit greater than $800 million. Can you maybe update that number and give us an idea of how much of that's going to be custom ASIC? I'm guessing $150 million, something like that, but I'm wondering if you can give us some more milepost there. Thanks.

Yeah. Hey, thanks, Tim. Yes. So I think, very encouraged by the demand trajectory upwards on the AI segment for us. Exiting the year at $200 million run rate, that is still mostly driven by the electro-optics platform, the DSPs, TIAs, and drivers at 800-gig. There is probably going to be a little contribution even in the fourth quarter from custom silicon. But that really ramps up more meaningfully next year. But we look at this as just a very positive sign that a quarter ago when we did our -- made our best estimates as to the opportunity in front of us, the fact that exiting the year will be at next year's run rate is very positive. So think of it as this year mostly electro-optics, that's going grow again next year, obviously. And then you need to layer on top a more meaningful ramp in custom silicon. But at this point, demand is so far up into the right, it's actually hard for us to put an exact number or give you a refined number on next year, other than it's going to be obviously bigger than we said the last time around.

Right. I guess, just -- but is the -- so there's obviously this big surge happening during the back half of the year. But if you took the AI piece, does it continue at that sort of linear rates through the year, or does it sort of flatten off as you kind of go out through next year? Just trying to shape that. Thanks.

Yeah. I think it's -- well, I think the way you should think about it is, in Q4, we're already at $200 million. It's going to be at a higher run rate throughout fiscal '25 each quarter. So, it's going to keep growing on the optics side, and then you're going to layer in custom silicon on top. So, we're not -- we're certainly not saying it's going to flatline at that level. We're just saying it's coming a lot faster, demand has been a lot better, supply chain team in Marvell is doing a great job to source the components and the capability that we need. So, I think it's a great setup for next year relative to the AI revenue.

Great, Matt. Thank you so much.

Thanks, Tim.

Your next question is from Blayne Curtis with Barclays. Please go ahead.

Hey, guys, thanks for letting me ask the question. I wanted to ask, it's either Matt or Willem, I guess, on the gross margin. Can you just remind us what the headwinds are that you expect to resolve in Q4? Because I guess storage is taking longer, but I think the gross margin in electro-optics are quite good, so you would think that would be a tailwind. So, could you just walk us through the puts and takes on the gross margin guide, that would be helpful.

Yeah, thanks, Blayne. Yes, certainly the optics is a tailwind. And then in addition to that, we've signaled that wireless carrier is really stepping down in the fourth quarter and consumer is also really stepping down. So, the combination of those things. And then in addition, our cost structure for our products is really improved. And the overhead on manufacturing and we see that benefit, it's really starting to flow through more significantly in the fourth quarter. So, storage certainly is a headwind for us as that recovery is pushed out.

Okay. Thank you.

The next question is from Matt Ramsay with TD Cowen. Please go ahead.

Yeah, thank you very much, guys. Good afternoon. Matt, I think you addressed the AI revenue commentary. I wanted to ask about the sort of goalpost for custom silicon. We had been in -- we've had that conversation around $400 million and then $800 million and then it got pushed out a little bit. And there has been I mean -- I don't think any of us have seen anything like the CapEx shifts that have happened around Generative AI and spending patterns at a lot of your large customers. What we've observed is, a lot of custom compute ASIC programs in flight, some of them are for sort of CPU offload and whether it's called DPU or SmartNIC or whatever you want to call it, some of them are for smaller model inference custom silicon programs at hyperscale. Maybe you could talk a little bit about where your engagements are? What you're seeing in terms of timing and magnitude? Because it's been such a tumultuous CapEx environment with AI versus traditional compute. And if you have any updates on some of those numbers around the custom compute programs and timing, that would be really helpful. Thanks, guys.

Yeah, great. Thanks for the question. I'd say a couple of things, Matt. The first is on your question about the custom silicon opportunity, which really if you sort of summarize the different pieces of it as you mentioned, it's really of all to provide accelerated computing, right, in a custom format for these various companies. I think the overall CapEx shift towards accelerated computing is clearly going to benefit those that are participating in that segment, including us. And we do see -- and some of that overlapped, and that's why last quarter we tried to decouple them somewhat. But some of that revenue obviously overlaps with the AI numbers I gave earlier. But it's certainly a positive setup. We talked about having two different products last quarter that were kind of the lead ones that were tied to AI. One of which was in sample stage. That product is looking very good and most likely go into production first pass. The second product taped out as we expected in the second quarter, and that's also planned to ramp up next year. The size and the timing of those are still to be determined. We're working with our customers on those. But the overall shift in CapEx spending is clearly a tailwind on that business. So I'd sort of say to calibrate it today kind of tracking towards what we updated you guys on the last time, and then next year, it really depends on how much it moves over, Matt, and how much it can actually -- how much business can be driven. So, we're not calling that, other than to say it's -- the custom stuff for us is mostly going to be driven by AI next year in terms of the...

Got it. No, thank you, Matt. Appreciate it.

Yeah.

The next question is from Harsh Kumar with Piper Sandler. Please go ahead.

Yeah, hey. I think when you talked about gross margin drivers for fourth quarter, Matt, you mentioned that you might be expecting a significant recovery in revenues as well. Did I, a, understand that correctly? And if so, could you talk about what might be happening in the fourth quarter to drive that recovery?

Yeah. You're saying overall, Harsh?

Yes.

Your question is, what's sort of the view? Yeah, obviously, we're guiding one quarter at a time, but to give you some perspective because we gave you a lots of puts and takes including commentary about carrier and wireless, which we've been actually saying for several quarters now relative to that, 5G run taking a pause. Overall, we expect revenues to be up again in the fourth quarter. We were sort of consistent with what we outlined in last quarter's call, which is that Q3 and Q4 growth would accelerate. So kind of think of that as -- growing again in the fourth quarter sort of like we did in Q3. But really driven by cloud and AI, and AI being the big driver. But also as we said in our prepared remarks, standard cloud infrastructure revenues are also growing very nicely through the year, not only from Q2 -- Q1 to Q2 and Q2 to Q3, but also into the fourth quarter and beyond. And that means if you kind of do the back of the envelope, data center becomes a bigger part of our revenue and AI becomes a bigger part of our revenue by the fourth quarter. And then overall Marvell, Inc. is also up. So, I just kind of frame it at a high level, we did set out a long time ago to have a diversified strategy within data infrastructure by serving multiple end markets with a suite of products. So, we've got enterprise, 5G, automotive, data center. And as you've seen even in this dynamic environment, certain quarters some are performing better than others, and then we're sort of blessed to have markets that then have kicked in at the right time. So, even in a tough macro right now, Marvell is continuing to grow throughout the year starting from our first quarter, and we anticipate that growth obviously going through Q3 and then through Q4. But the mix shifting by end market, if that makes sense. I know you will create your model and -- we give enough commentary I think to put the pieces together.

No, we appreciate it, Matt. Thank you so much.

Yeah.

The next question is from Christopher Rolland with Susquehanna. Please go ahead.

Hey, thanks for the question. And this one is for Matt. Just kind of an amazing revision here on the electro-optics portion. Since you've seen this inflection, you probably done some more research here, how are you thinking about the attach rate for these products per GPU, call it, is it one for one? Are you thinking it could be two for one? I've seen some research that suggests depending on how many layers there are, could even be three per one. And is the revenue that we're talking about here, all 800-gig PAM4 DSPs? Is there anything else related in that as well? Thanks.

Yeah. Thanks, Chris. Yeah, I think it's fairly similar to our view last quarter. We were pretty clear about the direct attach, which was the one-to-one you mentioned. Understanding that as you get to the upper layers of the network, there is more, and there is a range and you probably size the range. We're still I think refining our exact models there, so I don't know if I have an exact number to give you. But I think you're thinking about it the right way, starting at the direct attach and then building higher. And the second part of your question, it's all 800-gig for AI, at this point. We have strong traction on our next-generation products at 1.6T, which would be starting sometime next year. The products that are at frequencies lower than 800 gigabits are typically served for the traditional cloud infrastructure, which is also seeing a recovery as well. So, hopefully that's helpful.

Fantastic. Thanks, Matt.

The next question is from Ross Seymore with Deutsche Bank. Please go ahead.

Hi, guys. I don't know if this one is for Matt or Willem. But I just wanted to talk about the puts and takes to next year's gross margin, and I know you're not guiding that far out with any specifics. But it seems like you're going to have a number of custom products that will be going up in the data center side, the storage side should recover at some point in time. But the general question is if custom products tend to be lower gross margin and storage and some other areas like enterprise eventually come back cyclically, how do we think about the puts and takes in your gross margin versus that 64% to 66% historical target range?

Yeah, Ross, maybe I can start and Matt can add. So, when you look back, certainly over the last couple of years, we've grown both our carrier and ASIC business sort of faster than the rest of the business. And we were able to maintain our gross margin within our target range. We're targeting to get back to that 64% exiting this year and then to maintain that through next year. But clearly, it's sort of early to decide exactly how big the ASIC ramp is next year. Now if we do show outsized growth there, that would negatively impact our gross margin, but certainly our view is that, that would be very accretive to operating income and to EPS. But it's too early right now to know exactly the extent of that. Hopefully, that's helpful, Ross.

And Ross, maybe -- it's Matt, I'll just add. I think the -- clearly, the custom business carries a lower gross margin, we've been very open about that. But I think looking out to next year, it's a little early to call it, right? We don't know the rate of recovery for storage, that's a big part of the equation. We also have what's the -- even within like AI and cloud, what's the optics revenue going to be versus the custom stuff. You've got automotive continuing to perform well and growing, which is a higher-than-average gross margin category for us. So, we've got a lot of irons in the fire relative to various businesses that may or may not pick up at different times. As Willem said, we had a pretty consistent ability to manage gross margins in the range that we were targeting, understanding there is mix issues all the time. We've been going through a period here Q1, Q2, and Q3, where we've had it for a longer period, I'd say, unfavorable gross margin relative to our traditional mix. But it's such a dynamic environment, it's really hard to call the ball exactly, Ross, on when all those markets sort of burn through inventory, kickback back in, what is the recovery look like. But when we look at it a very high level, we continue to believe we have a very nice balanced portfolio of different products, technologies with different business models behind them that generate different gross and operating margin profiles. And as an example, like Willem said just because some of these custom programs are at lower than corporate average gross margin, especially if they ramp significantly, they are extremely accretive to operating margin and operating income at the bottom line level, which ultimately over time is what we're laser-focused on is driving earnings per share and driving operating profits for the company. So, high-level answer, but maybe it just frames it up, we need a little more time in front of us to really figure out what that looks like for next year.

Fair enough. Thank you.

The next question is from Vivek Arya with Bank of America. Please go ahead.

Thank you for taking my question. Matt, I was hoping you could help us size how big storage is. Currently, in the quarter you reported both kind of data center and outside of data center. And I forgot whether you mentioned whether that's going to be up/down/flat in Q3 and Q4, or can it kind of hold at these low levels? And what are you looking for to inform you as to when it starts to grow sequentially, like how much excess inventory is out there, or do you think that it can actually hold at these low levels? So, just help us kind of set what the right baseline view is of storage as you get into next year.

Sure, thanks, Vivek. It's a great question, okay, I think it's in some ways a million-dollar question. Let me tell you where we're at so far. So, obviously, Q1 was very, very low in terms of our storage revenues, and in particular, we're really talking about data center storage, if you think about it. The consumer piece is kind of hung in there because of some the specific applications we have. But that's -- what's moving the needle is the data center side. In Q2, it had recovered, which was nice to see. It was coming off a bottom, right around $100 million, let's call it, something like that. And then, from there, we said in our prepared remarks, Q3 we'd see a modest recovery again. So that would be up. It's still not sloping towards I think where we thought it was going to be a quarter or two ago relative to the year-end exit rate, and I think most of the end customer commentary suggests this is more of a first half of '24 type of recovery. But I guess the good news is, it's come off the bottom, good growth in Q2, it's going to grow again in Q3, and it will grow again in Q4. But it's just not quite at that slope and that rate. When we've dug in with the customer base kind of all the way to the end companies that consume the storage, it's hard to get an exact clear picture of where the inventory is and the timing, as I mentioned. However, what we do understand is that there's a lot of activity on qualifying the next-generation drives in technologies. There's a big TCO benefit. We haven't been able to detect anything structurally that's changed per se, although we continue to study that. So, in our mind, it looks like this has to come back to where it was at some point in terms of the number of exabytes that are being shipped or consumed, and it should grow from there for all the reasons that's been growing for the last 20 years. And AI, there was a question earlier about that, it's probably a tailwind on that. But I think with how far back we are on the supply chain, I'll just conclude my remarks by saying it's just very difficult to get a full read on where it all is, what's going on at the end customer level, and when it comes back. So, all we can do is sort of look at what our customers are saying and talking to us about and follow their lead. Hopefully, that's helpful.

Yes. Thank you, Matt.

Yeah.

The next question is from Srini Pajjuri with Raymond James. Please go ahead.

Thank you. Hi, Matt. My question is on the custom silicon side, I think you called out two programs that -- in AI that you expect to ramp next year. Can you talk about the design win pipeline? And also I want -- if you could touch on, I know it's difficult to predict what kind of revenue opportunity -- forecast the revenue opportunity this early, but could you give us at least some sense, because some of these programs -- we do hear that some of these programs do get canceled. So, just want to make sure that you have that visibility that these programs are for real, what you have for next year. And then, if you can talk about other opportunities beyond those two programs that you talked about, that will be helpful. Thank you.

Yeah, great question. Let me start with the, are they real, because I think that was the big question. Probably two quarters back, Srini, and even at the end of last year when I think the reality of the macroeconomy really hit the cloud companies pretty significantly and that's when you saw those companies at a very high level take actions on headcount, take actions on expenses, look at program priorities down at the infrastructure level. And part of our call in Q1 and Q2 was sort of advising investors how that had changed. What emerged from that was a shift from traditional cloud infrastructure to AI, now we've seen that play out through the first half of the year. So, those programs not only are getting funded there, there is extreme urgency around them. And so, now we're at a point given where these products are at, actually talking about when do we start wafers, how do we make sure the capacity is lined up, how do we have three-way calls to make sure we're all dialed-in on what's needed. So, that gives us quite a bit of comfort, and that -- and we were further along. In a quarter from now, two quarters from now, we'll be even farther along relative to being able to provide more visibility on what that ramp looks like. But we feel good about that. And certainly going through that process, we -- for all the programs in the pipeline, we have both one an open a good view of where our customers want to go. On the opportunity size, just to give you a sense, if you look at the overall Marvell's total design opportunity pipeline, the biggest segment is data center, it wouldn't surprise you, but it's a huge portion of…

Thanks, Matt.

The next question is from Harlan Sur with JPMorgan. Please go ahead.

Hi, good afternoon. Thank you for taking my question. I know there's been a lot of focus on your electro-optical business tied to AI. But I think the bigger portion of your electro-optical business is focused on the persistent upgrade cycles across the entire cloud data center footprint, right? And I think these tailwinds are still in front of the team, right? I think you still have -- there still is one more cloud titan that is yet to upgrade to 400-gig across its data center. You still have the 800-gig 1.6T optical upgrade across the remaining three titans. And then on DCI, you're still ramping 400ZR DCI, but you still I think have two more cloud customers that are in the pipeline. So, are these broad upgrade cycles starting to line up to fire next year, maybe even some starting in the second half of this year, maybe it ties into some of your commentary on strength in broader cloud. But just trying to get a sense of the ramp profile on these upgrade cycles.

I think you did a great job summarizing it. The AI piece has obviously become very meaningful and we gave a sense of what that level of revenue is going to be in the fourth quarter. So it's pretty amazing how fast that's grown obviously. But to your point, the standard cloud infrastructure side of connectivity is having a nice recovery off of some inventory that was built last year, that's now becoming a tailwind as that burns down. You pointed out that there's still a major cloud company to go through their upgrades. So that's also in front of us. As we transition each of the technologies, in terms of speed, there is always an ASP bump and it makes sense, because you're delivering, 2x bandwidth for not 2x the price, right? So, we're on a nice sort of share the benefits with our customer trajectory there. And even at 1.6T for next year while AI will lead it, right behind it, it's going to be a key part of the overall infrastructure build probably the year after. And then the regional data center stuff, which is really the ZR products, also just doing extremely well this year. I mean large upsides due to Generative AI. And then, we said in the prepared remarks and we just put out the press release, I think it was today or it was yesterday about 800ZR, and there'll be more to come on that. But there is another frequency kicker coming in probably faster than we would have thought if you went back six or nine months ago because that between data center bandwidth is actually now becoming a bottleneck. So just tremendous across-the-board opportunities that are not just AI, but it's really driving our cloud growth, and that's -- a lot of that we're seeing in the second half and we're going to see it through next year. So, we have both of those irons in the fire for Marvell overall growth.

Perfect. Thank you, Matt.

Yes.

The next question is from Ambrish Srivastava with BMO. Please go ahead.

Hi. Thank you very much. Willem, I had a question on the cash flow side. I was trying to see if there was a discernible pattern of seasonality, doesn't seem to be. But you gave a reason for DSOs and impacting the (CFO] (ph). Could you kind of just walk us through how should we think about cash flow for the rest of the year?

Yeah, sure. Thanks, Ambrish. Yes, so this quarter certainly DSO was impacted somewhat by linearity. We do expect a nice back -- bounce-back in Q3 and some normalization. As we've mentioned, we're really focused on driving down our days of inventory and we've consistently reduced that through this year, and we expect to continue driving that down through the rest of this year. And so, yes, you should expect some good improvement in the second half here.

Okay. And my quick question for you, Matt. We're hearing a lot of companies are talking about the pushouts of CapEx, general-purpose CPU towards AI. I know you addressed it in some ways earlier. But are you seeing that manifest in your business that certain projects are being sacrificed at the AI alter?

Funny way to phrase it. Yes, I think there's two aspects. I think one is, were there project reprioritization or not within those companies. And then also, is it affecting our business right now, which is kind of the two pieces. So, our view is, on the first one, at least for Marvell, I can't comment about everybody, but for us, we really worked through that issue at the end of last year, early this year. There was a big, I'd say kind of re-prioritization that went on, first, driven by the budget squeeze and companies just trying to get more efficient. And then almost in line with that was sort of the release of ChatGPT and the realization that there was this potential massive opportunity, right, for productivity gains through AI and then what are people going to do and then sort of the race was on. So, from that standpoint, we feel very good about our WIP and what we're working on and that -- at least for us, those decisions and re-prioritizations were made. And some of those are why we were lighter on our custom silicon revenue this year. We sort of talked about that a couple of quarters ago. So those programs got delayed. So, we've I guess taken some medicine on that and we moved past it, now we've got this off to the races with the AI stuff. And then, on our current business as you saw, we had -- in Q2, we had overall data center revenues up pretty nicely and that included on-prem being down, AI being up a lot, but we also said that standard cloud infrastructure was up and it's going to be up again nicely in the third quarter and it's going to continue through the fourth quarter into next year. So, I think we've worked through that as well. And probably the reason we don't get hit quite as much there is that we're tied more to the networking and connectivity than selling CPUs as an example. And so, that overall kind of network bandwidth needs to continue to get upgraded and deal with it, because a lot of these big data centers are multi-tenant and they've got AI sitting in there, and they've got specialty built servers and storage for other things, but it all needs to work together and it all needs to ultimately have the right level of bandwidth to support the compute power. So those are some puts and takes. I think the headwind on the whole thing would -- for us really would be the storage piece, which shows up in our data center near-line and that we've just been kind of -- I answered the questions about that already. So that's the puts and takes, but overall it's actually growing very nicely for us. It will be a growth driver for us next year too.

Got it. Thank you, Matt.

Yeah.

The next question is from Chris Caso with Wolfe Research. Please go ahead.

Yes, thank you. I guess the question is digging into the enterprise on-premise a little bit more. If you give some detail on what you include in that, that's still looking to be down, does that include, for example, fiber channel as well? And do you think that business -- that this will mark a bottom to that business? Yeah, obviously, probably difficult to figure out when that starts getting better.

Hey, Chris, thanks for the question. Yes, I mean, it's -- enterprise on-premise business has just kept coming down. Now I think the reality is if you look at overall enterprise server shipments and you look at sort of the end data, that's probably been down for, I don't know, eight quarters or something. So, the end markets just not doing as well. It's mostly fiber channel that's within there. There's also some Ethernet and some NIC products as well. I don't know that we're necessarily calling a bottom, but it's got to be close just because at some point, people need to -- people actually need to buy servers and customers need to work through inventory. So, it's been a headwind. I think we're close to it, maybe needs a little more time, maybe another quarter. But at some point, that will renormalize as well. And it's obviously within overall data center, all of that sort of inventory adjustment and weakness at the end market levels being just blown away, obviously, by the AI and cloud infrastructure piece, which is still driving very healthy growth into Q3 and then another step-up obviously in Q4. And that's assuming really no material recovery in the on-prem side. But I do expect over time, if you just sort of step back and look at like fiber channel, it's been a pretty stable business for us, ever since we've owned it through the Cavium acquisition. And we don't see any reason why, at some point, that just doesn't come back to where it was. And when it does that, that will also be a positive thing for us. Hopefully, that's helpful.

Thank you.

The next question is from Quinn Bolton with Needham & Company. Please go ahead.

Hey, Matt, I wanted to ask, you talked about the electro-optics driving most of that $200 million of AI revenue. Can you give us some sense how much of that is InfiniBand versus Ethernet within those 800-gig modules? And do you guys see any share difference between your position in Ethernet versus InfiniBand? I know your share is very high in general in that market, but just wondering if you would say your share is higher either in the Ethernet or InfiniBand. And then I got a follow-up.

Yeah, maybe the simple answer is, first of all, as you point out, we're agnostic. We participate broadly in the connectivity independent of whether it's InfiniBand or Ethernet. And I'd say that we continue to have very high market share and penetration into both those applications. And the demand we're seeing is broad-based from both of those as well as multiple customers driving it. So, it's actually really kind of hard to break it down exactly. And I don't know that it's super helpful, because overall, the shares still very healthy, and we're agnostic. So, the same module can kind of be used in either one.

Got it. Okay. So it's pretty broad-based across both. The follow-up question is just you mentioned 51.2 terabit switch and starting to sample. Just as you look out to calendar '24, fiscal '25, do you see that starting to get deployed at the U.S. hyperscaler partners?

Yes, I think we'll have some contribution from that product area next year. The bigger ramp is a little bit further out. But very pleased, I'd say, with the open funnel on that technology. I'm very happy with the team. I mean this -- the architecture and the key pieces of that initially came from Innovium, right, which is a company we acquired in 2021. But the thing that we did upfront is we put it on the Marvell 5-nanometer design flow using our SerDes, our IP platform, our new product development process. So you kind of got the best of both worlds and teams executed extremely well on the chip. And as a result, it's giving customers confidence we can really be there for them not only on the feature set, including very, very low latency and in the right power envelope, but also being able to manufacture for high volume, which is no small task on these effectively reticle buster type switches or type products. So we'll see how it goes, Quinn, but pipeline is strong. This is going to be a big upgrade cycle for the industry. And I think it's going to be overall positive for the people that are in that ecosystem.

Thank you.

I think we'll do one -- perfect, Quinn. And I think we'll do, Gary, one more question, and then we can wrap it.

And that question will be from Gary Mobley with Wells Fargo Securities. Please go ahead.

Hey, guys. Thanks for taking my question. I don't think there's been a mention so far the influence on an extra week in the fourth quarter. Matt, I want to verify your expectation of sequential revenue growth in the fourth quarter. Is that independent of the extra week or inclusive of the extra week? And Willem, the target of OpEx for the fourth quarter, I think you said $430 million with the extra week. Is that still the target?

That's correct.

Okay. Willem, you want to go? Then, I'll go.

Yeah, sure. Yeah, just on the OpEx, that's correct, Gary. So for the 13 weeks normalized, we said $420 million. And so, for the full 14 weeks, we guided $430 million. So you've got the correct number.

Yeah. And then on the revenue side, Gary, it's been my experience doing this for a long time and having the famous 14-week quarter roll through, it's -- I think the best practice I found is, you end up really not getting the extra revenue per se, but you got to obviously count the expenses. So that's sort of the way to think about it. So, we don't -- we're not saying, hey, because Q4 is 14 weeks, we're going to have some better growth than we would have. I think it's just -- that's kind of what the market is doing. So the -- and then Willem gave you sort of the run rate on OpEx. If you look at it, netting out what 13 weeks would have been, which is about $420 million.

Got it. Thanks, again, guys.

Yeah.

This concludes our question-and-answer session. Thank you for attending today's presentation. You may now disconnect.