US semiconductors, semiconductor equipment, and networking equipment stocks here at Citi. It's my pleasure to welcome Matt Murphy, Chairman, CEO, Marvell Technology, as well as Ashish Saran, friendly neighborhood IR, to our fireside chat. I'll go with my questions first, and then open it up to your questions, and if you have a question, please raise your hand and the mic will come to you. Welcome, Matt.
Yeah, thank you. Great to be here.
I always have to remind Matt that this is where the Marvell transformation started, because, you know, seven years ago, or eight years ago?
Eight years, yeah, 2016. This was the first conference I attended on this exact day.
Yeah. Matt was here with Rick Wallace on-
Rick Hill.
Rick, sorry, Rick, not Rick, Rick Hill, on Marvell, talking about Marvell 2.0, at that time. But what a transformation, honestly. I think we haven't seen as many examples in terms of, you know, a company trying to grow into these large markets like you did in 5G and then now data center AI, with a set of assets like Avera and Cavium, and now you're there. Your custom ASIC is ramping and, so there's nothing like, you know, seeing it. But Matt, maybe you can start off with the kind of state of the union, where we are, generative AI, how do you view the Cloud CapEx spending trends within that custom ASIC opportunity, and what has surprised you over the last couple of years?
Sure. Yeah, great. Matt, thanks for the lead off. Yeah, I think maybe just to bridge it, you know, in terms of where we were and where we are, you know, we had, you know, basically a company back then that was 65-70% consumer, kinda end market exposure. Really, the rest of it was enterprise, and that was kinda it, right? There wasn't any 5G, there wasn't any cloud, there wasn't any automotive, those types of things. And so, you know, you fast-forward with all the work that we did, which was a lot of organic effort, some M&A, you know, a bunch of very actually critical divestitures, right, that we did to enable us to fund our future. And so here we are now, company's now 70% of its revenue is in data center.
We've got a very aggressive R&D positioning to take advantage of the Gen AI, and just in general, the trend towards accelerating computing systems and architectures, probably one of the best-positioned companies. So our R&D positioning is even higher than the 70% sort of revenue exposure. We're very committed to this area, and the trends that you mentioned. I mean, this all started, you know, really in earnest for us in our May 2023 earnings call, where we basically had taken our time to really kind of formulate our outlook. We called out at that time, "Hey, we saw $400 million in revenue for calendar 2023 and $800 million in revenue in AI for 2024." I think we were one of the earlier companies to try to size that for investors, you know, try to be helpful.
It's kind of astonishing if you think about how far things have come along. Here we are in September of 2024. We just came off our earnings last week, and we're talking about the $1.5 billion that we had sized for 2024, with $2.5 billion for 2025, now being substantially higher. So you know, you can kinda look at it like a year-and-a-half period, and you've almost effectively doubled what we thought was, and I think the market thought was a pretty, pretty aggressive plan. So we see this TAM growth continuing in a very aggressive manner. You know, the cloud CapEx trends we're seeing are certainly encouraging for our business and the level of investment that I think very smart, sophisticated, large companies are putting in. And, design win activity, you know, and design win opportunities really off the charts.
We just got done with our strategic roadmap planning meeting. We do that once a year. It's really our capital allocation review, and I'm just blown away by the technology in this company and actually how we're positioning that R&D now to be not just for the next generation, but the generation after, and really going from being a fast follower, quite frankly, to a leader in this area. So all that's very exciting from a state of the union standpoint. We think we're in the right end market. The TAM is like super sized versus what we had ever anticipated this company could sort of look at in the past. And we like where we stand right now.
Great. On the custom, is there one thing that seems to be working well for you is your ability to offer your client a platform for them to pick the level of IP they need from you, going from packaging to SerDes to memory to processor cores. I guess the first question we have is, if you were to rank those IPs, which ones tend to be the key selling point to your customers?
Yeah. I think, you know, one of the key things, particularly in the accelerator area, is, you know, leadership in I/O and leadership in connectivity, and that's always been a strength of Marvell classic. We added to that with the addition of Avera. They had very good high speed folks. You know, obviously, then we added Inphi and really turbocharged the whole effort. And now from an IP standpoint, you know, really have one of the most, if not the best-in-class roadmap in that area. That's extremely critical. I think the other one is. You know, and you can go through the list of sort of IPs. The other one would be bringing to the table the custom kinda compute capability that we got from the Cavium acquisition.
You know, that team, you know, very experienced, having done very high-end processors, going back to the lineage with Zack, actually, where most of that original engineering team came from. And increasingly, we're seeing that being requested of us actually as Marvell on the question of how do you go faster, and how do you go more parallel, and how do you go more concurrent with our customers so they can increase their beat rate on product releases? Adding that as a design, front-end design resource has become actually a critical differentiator.
But then on top of that, and you can go through all the IPs and the different things we have, you know, I just think the ability to have a compelling roadmap, two generations out, have the strategic partnerships that you need with folks like our foundry partner, TSMC, the IP vendors, the big OSATs, and then actually having the skill and the scale to integrate these very complex components in this whole ecosystem into a product that can actually ship in high volume with yield so that the customer can get the TCO. You know, that whole package, and then being able to do that over and over again, is very compelling.
And so if you're part of that ecosystem and you're just a piece part, it may be difficult over time, we think, to really pull off, you know, the execution that's ultimately required. And we think a company like ourselves is very unique in our ability to pull all that together. So. And then also do it in a trusted, partnership-oriented way, multi-generation. And the final point I would make is having the other technologies we have in-house in terms of our optics platform, our higher layer switching, and our ability to kinda look at all that together within the ASIC and custom roadmap.
It allows us to help become kind of a thought partner with our customers on what are they thinking about next, and how can we leverage all of these pieces together from a system point of view, to save power and cost and improve performance. Like, every watt counts, okay? Every watt counts. It's a huge deal on every one of these, every one of these hops in the system. So having, again, all those pieces, really makes us very strategically relevant in that conversation.
Great. And Matt, you guys have talked about ramping four chips across three different customers at your Investor Day, two this year and one in twenty-six. And if you were to pull the curtain a little bit in terms of these hyperscalers and how they think about their investments. You've been very clear that at the end of the day, you know, Broadcom and Marvell are the two companies that have the full suite of IPs and expertise to make the most sophisticated chips.
But the question comes up with investors around the sustainability of these sockets. If I were to look at your peer, who's been engaged with TPUs for the longest time, they've been doing it for seven, eight years, so there's a lot of stickiness to these things. And Raghib has talked about this as well. These are very complex, radically sized chips. So help us understand how your hyperscalers are looking at the two ASIC providers and/or landscape, and in terms of you know, are they looking to, like, dual source? Are they looking to partner on a much longer period of time?
Yeah. No, I think the... These programs are so strategic, you know, and mission-critical to these companies, right? And they're so involved in terms of the trust you need to have and the kind of tight coupling and integration of the two engineering teams, that almost by default, you would just have to think of these as one generation out, two generations out, three generations out, especially at the speed you're trying to go at. So there's no, and there's sort of no room, and there's no engineering capacity out there at any of these companies and anybody to do, like, dual source, right? So you're gonna pick your partner, you're gonna go all in, and then if you do a good job, then you should probably, you know, earn the right to get the next one.
And quite frankly, the reason you got the one you got is usually 'cause you had a great vision on the next one, and you had a compelling roadmap. So it's not to say that there isn't gonna be dynamic changes in the market. There's still a lot that needs to get settled, okay, in terms of... But the dance card isn't done, you know. I mean, this is one of those things where you've got a whole new market opening up, lots of opportunities, and we're just very. We just feel very honored to be at the table to go participate in many of these directly and bid on the other ones. And then we'll see how it goes. But we basically came out of nowhere in this area, right? We didn't have a custom silicon offering, right?
We acquired an asset called Avera, which had been in GlobalFoundries, but had its roots in IBM. It was the IBM original custom silicon team, which at one point did have market leadership, and for a number of reasons, that asset, you know, hadn't really performed kind of where they thought they could be. When we acquired them in 2019, we got everybody on our new five-nanometer technology platform, combined all the SerDes effort, et cetera, and we've put together a very compelling offering, which now, by the way, we're not, like, trying to cobble together a new process node and bring in a new team. Like, we're like four years into this now, and so it's actually really exciting what we can bring to the table now versus even when we won those initial designs four years ago, right?
That was kind of on more on a trust, and here's your PowerPoint. Now it's proof of concept. It's actual silicon. It's our execution. So we think we're just extremely well-positioned there, and if you do that, then you should ultimately build customer trust, and you should be able to get multi-generations of these programs.
Great. And maybe switching to Ashish on the optics side.
Mm-hmm.
You guys saw a lot of strength late last year into early this year, and you've been more kind of conservative, guiding to kind of flat growth for the back of this year, but on the last week, you talked about now sequential growth in that business for the fiscal year. What changed, and you know, there have been concerns around maybe a delay in 1.6T . Is that an issue for you guys?
Sure. Yeah, I think if you remember, second half last year, we saw very strong growth in our optics business as people are anticipating this very strong AI ramp starting. We're the primary supplier of 800 gig optics, PAM-based solutions by data centers, as well as 400 gig, which is connecting data centers together. And then with that very strong ramp, you know, like in anything else, when you see this very steep ramp, you worry at some point, is there gonna be an absorption phase? So we've been anticipating a level of flatness. Now, the reality is, we thought that's gonna happen in Q1. It didn't. It was up very strong. Same thing in Q2, and at this point, when we look at our bookings and backlog for the back half, it's very clear that there really is no slowdown.
I think the demand for our products, both inside as well as between data centers, remains very strong, and in terms of the next generation technology, which is 1.6 terabits per second, we were first to market last year at OFC, if you remember, is when we actually started sampling that product, well ahead of everyone else. We're basically getting through qualifications, and we are starting to ship for production in Q3, which is now, so I wouldn't say there's really been any delay in 1.6 . I think this was always the plan. There'll be module partners shipping their product out probably towards the end of this year, and we think 1.6T starts ramping more meaningfully next year, at multiple customers. In the meantime, eight hundred gig obviously has a lot of long legs, right?
It's not just gonna ship in AI, but next year, standard cloud infrastructure, which has so far remained on the prior generation, 200, 400 gig, also starts transitioning, so 800's mainstream workhorse this year, next year. 1.6, think of that layering on, on the 1.6T side, so we very strong optics roadmap in front of us.
Great. And then the question comes around on DSP competition, and you guys have been very clear the last couple of years that you have a very high market share and, you know, there are little competition. And can you just talk about, you know, the impact of maybe vertical integration for some of the GPU makers to be doing their own DSPs or, and how do you see the competitive landscape?
Yeah, maybe I'll take the front end of this, and then-
Yeah.
You can add on the competition side. I think for us, you know, when we acquired Inphi, which is where we got the product lines and so forth, they've done a great job, executed well, had very high market share. And I sort of agreed with their philosophy at the time, which was: look, it may not stay all the way up there, but we should maintain a leadership position, right? And we've actually continued to exceed the market share goals that we thought possible, even when we due diligence the asset, you know. And since that time, we've really turbocharged that team. I mean, we had an outstanding DSP and mixed signal team at Marvell. We've combined those together.
So the actual engineering workforce now we have on this particular product area is, like, second to none, and we're driving a best-in-class roadmap, both on nanometers, on architecture, on a proliferation of different SKUs to really cover the whole market. So from a competition standpoint, the way we're addressing it is, and by the way, the market's moving so fast, we still believe execution, first to market, product leadership, that's ultimately gonna drive it. There are other factors, right? There are other competitors. There's some talk of vertical integration. You know, our view is just over time, we can, we'll still maintain that very high market share if we continue to execute on new products, and that's really what we're doing. I don't know if you wanna address the other part of it.
No, I think, I mean, I think the other point is, you know, we offer a full set of solutions. It's not just when you hear there's a DSP, it's actually a whole range of DSPs. We offer all different configurations. We have supporting analog components, TIAs, drivers. We just talked about silicon photonics very recently. So I think that's the other part of it, is it's not just a point solution. It's the fact that we actually offer the customer every type of solution they're really looking for.
Okay. And as part of your AI portfolio expansion, an area that is often less talked about is your AI switch chip, the Teralynx 51.2. Can you give us the state of the union on your early efforts, traction, and the ramp?
Yeah, we're pleased so far. We acquired a company to get this capability called Innovium. Marvell already had a very capable switching business and team, but that was more focused on enterprise and carrier. So when we acquired Innovium, we actually combined that as one much larger business unit. We hired an outstanding general manager to run that business, who had a strong lineage in this area. You guys saw him at the AI Day. This is Nick, and he and the team have done a great job of actually combining the two teams, pivoting that resource and roadmap to the cloud switching and AI switching applications.
So we got, you know, way more people working on it than we did at Innovium standalone. We released the first product this year, our 51.2 T switch, Teralynx 10. It's on the Marvell five-nanometer technology platform, so it's our series. It's not third party. It's our IP. It's kind of the best of breed between the Innovium architecture and Marvell's proven silicon platform. It's gotten great reception with customers. We're sampling it. We're actually shipping it now, the first units.
Yep.
And we'll see how it goes. You know, it's a competitive market, but we're also driving, we think, a very competitive roadmap in that area as well, and we're very committed to that opportunity. It's a little longer term for us, but it's a TAM that we showed at the Investor Day, being very substantial, right? And kind of a key part of the equation. And having the leadership on the optics side, coupled with the switching opportunity, we just think it provides a nice solution to our customers in the market, especially as AI hits, and there's more of a diversity of applications now and use cases, we think that opens up some ability to go gain market share, you know, over time. But we've got to earn it through our execution and our customer partnership.
Yeah, and in the meantime, I mean, the business itself, when we acquired it in Innovium, they already had a 12.8T product, which is shipping in high volume, and that business has grown very consistently, right? So we're very happy with the scale-up. It's already a fairly substantial business for us within our data center. And the 51.2 just simply adds on top of that.
All right. Then, it was good to hear last week that you're expecting carrier infrastructure, enterprise networking markets to kind of rebound. And I think the million-dollar question is like, when do you expect to reach the one billion of normal run rate? Any thoughts on those markets, and where they stand?
Yeah, I think both of them, you know, clearly are shipping below consumption. We did have several quarters, especially sort of during the supply chain crisis, i.e., post-pandemic kind of boom. So that overshoot ultimately, you know, manifested itself in an undershoot that we've been dealing with. That found its bottom in the first half, which we had hoped was gonna happen, and so we were very encouraged internally to see the third quarter. Now, collectively, we're guiding those up like mid-single digits growth, so it's coming off the bottom. Q4, we try to be helpful to investors because there's a big concern, like, you know, when is this gonna come back? What's it look like?
So we gave some color on Q4 based on the backlog we've already laid in, that should grow even faster in Q4 off of Q3. So that's a real positive sign, and we don't have an exact timeframe. I think it's hard to call the ball exactly the quarter in which those return to roughly, let's call it two billion combined, would be kind of a, like a good baseline number. But we do anticipate, even from the fourth quarter and through next year, that recovery will continue. You know, I think there'll be a point where at different times, they probably inflect up a little bit more just because I do think we've been under shipping so much that that'll happen. Plus, in the case of carrier, we do have some additional content that's rolling in that we didn't have before.
So there's enough going on there that we feel very comfortable getting back to kind of a solid baseline and then growing from there, kind of at market or market plus. And of course, look, if things come back stronger, if there is a bigger broadband build-out that happens as a result of AI someday, if some of these enterprise things we had, you know, enterprise picks back up more than we thought, you know, there's that's great, we'll take it. But I think for investors, just getting to that solid base back to that $2 billion is really what we're driving right now.
Great. And just finishing it off on the auto industrial guiding mid-single-digit growth, you know, amidst kind of negative data points on the auto demand. Just curious what you guys are seeing there.
Yeah, I don't think anything unique. You know, I've taken a look at, you know, even before Marvell, I was involved in the automotive business in my prior job, and so I've very much watched kind of industry-wide, the commentary around automotive inventory and digestion and some of the dynamics. And I'd say what we're seeing is consistent with that. And we're not the, we're not the bellwether here, you know, but it is an important business for us. It's just smaller in size relative to data center. The trends are exciting in terms of the connectivity attached that's happening in all the new vehicles that are coming, including combustion vehicles. But in the short term, there's probably not a lot of market insight we could give, just because I think what I've read from other people is kind of what we're seeing.
The only thing I'd add is that, I mean, in the automotive market, you know, we don't need unit growth at the end market level for our revenues to grow, right? For us, this is a new market opportunity. It's all about Ethernet replacing essentially legacy point-to-point networks. So the overall automotive unit assumptions can be very basically pretty flat in our view, and you would still see very significant growth as Ethernet just becomes a bigger part of cars going forward. And the nice part is, when it started off, it was mostly in EVs and hybrids, but now it's also going into ICE, traditional internal combustion cars. So the volume of opportunity is also increasing quite nicely. I think the near term is just the typical inventory correction, which you mentioned, but I think as I look forward, you know, we certainly see growth coming back to this market.
Great. I'll pause here and see if there are any questions in the audience.
Hi. Hey, thanks for the opportunity. So there's been, obviously, all the hyperscalers are building data centers, and there's been some discussion that they need to build the quote-unquote, "telecom infrastructure," because these data centers are getting very distributed. I just wonder from your product portfolio, in terms of coherent DCI perspective, are you hearing early discussions of that type of demand coming out, or is that something more aspirational, maybe for 2026 or 2027?
Yeah, I mean, why don't you tee it up, and I'll, I'll add to it.
Yeah, in fact, I think we certainly saw this, in fact, even, even last year. You know, we've been one of the key... You know, we, we basically founded this market on DCI, pluggable data center connects, you know, starting a number of years back with a key hyperscaler partner, and that was at a hundred gigs per lane. Essentially, we introduced our four hundred gig part last year, which is now shipping in volume. Normally, we would introduce the next generation, eight hundred, another few years from today. We actually pulled that in pretty significantly, primarily because of exactly what you suggested. As hyperscalers look at this massive AI spending, the size of these data centers, the number of data centers, they see a need for a lot more bandwidth between them.
That's exactly the reason we pulled in our roadmap and actually had our 800 gig product sampling as of last year. In the as we look at the demand for this year, you know, we had a question on optics from Atif earlier. You know, while we've certainly seen upsides on the PAM side of the business, which is inside data centers, we've also seen very strong demand for DCI products, driven very much by what you described, which is a lot of activity by the hyperscalers to kind of control that connectivity of that, even that 80 or 100-km length.
Some of the hyperscalers asked us for even longer reach products. So earlier this year, we introduced a product which can actually reach across 1000 km, right? Just because this market's becoming a lot bigger than what we originally thought this market is. It's almost probably two X bigger than what we thought it would be a while back. I think we can-
No, I think that was good. Yeah.
Hey, guys. Thanks so much. Just a little two-part question here. I guess you see a recent IPO forecast, you know, people are looking towards close to $1 billion for some of their, you know, their parts, which are more, you know, peripheral parts. I guess you guys are entering the retimer space now. Do you guys see yourself having anywhere close to that in terms of of an incremental, I guess, TAM? I just had a second one. I'd love to get your thoughts on kind of the co-packaged optics and how that plays into your, yeah, your markets as well.
So the first part of the question, just to clarify, is the PCIe Gen... PCIe retimers?
Yeah. How much-
The first part.
Yeah, like, how much incremental do you guys think that? Yeah. Why don't I start with that one? You can do the second one.
Sure.
Yeah, I think we-- I don't think we sized that market yet. I mean, there's been kind of a, you know, we're obviously on Gen Six, so there was five before, and that's been around for a while, the retimer opportunity. So we haven't sized it exactly, but I think we-- it was a good time for us to enter, you know, when this, this transition from Gen 5 to Gen 6 is when you go from NRZ to PAM. So just kind of like in the optical area when Inphi came in at that time when there was a technology transition. It's a really good fit for us. You know, we can leverage all that goodness we have, both on, you know, with our engineering teams there. It's an incremental effort for us versus some ground up new thing. And, you know, we can also leverage all the benefits of Marvell, right? In terms of our scale and cost and ability to execute. So we've announced products now. We're not sizing the TAM. I don't know-
Early days, yeah.
It's early days, but we'll, we'll be in there, and there is interest from customers, and certainly we're, they're very familiar with us because we're a big supplier to all those companies on all kinds of different components, but in particular, in connectivity, we're, we're well trusted there, so we'll see how that one goes.
Yeah, I think in terms of CPO, I mean, just taking a step back, I think if you should imagine, we're engaged in funding pretty much every type of optical roadmap where the market lead in this particular space. I mean, clearly, if you talk to our large customers, they look at kind of pluggable retimed DSPs, pluggable optics as being the primary path for a very long time. But then we do see certain use cases where near package optics or potentially co-package optics have a play. And I think one of the key elements to be successful there, you do need a very advanced silicon photonics platform. This is exactly what we've been shipping for the last seven-plus years in our DCI products. Earlier this year, for the first time, we made it available as a standalone product offering, which is our 3D SiPho package.
We call it a light engine. It can actually scale all the way to six point four terabits, essentially in a single component. So we certainly see that potential, in particular, as we have this very large and growing custom business, as we look at N plus two, kind of, you know, time frames where there is certainly a potential. So you should imagine we're working with multiple customers. We already have the key piece of technology, which is the 3D SiPho Engine. So we'll talk more about it, but it's certainly one of the tools we have in our toolset as time goes on, and you really need to get the optics a lot closer to the silicon over time.
You have a question?
Thank you. I thought Atif was just ignoring me, like usual, so the question I want to ask you, just more thinking broadly about Marvell, because clearly everyone's excited about your AI opportunity, the ASIC opportunity. That gets all the airtime.
Mm-hmm.
But at the end of the day, if the other businesses, the legacy businesses, aren't performing, other than maybe some inventory restocking from under-
Mm-hmm.
It gets harder to build the real bull case, particularly around margins. So just... Can we just take a step back and talk a little about some of the other legacy businesses? What's going on, away from like the overship, undership, and-
Mm-hmm.
Like, what is the underlying trend in storage, in networking when you think about the next three to five years? Thanks.
Sure. Do you want to start with this?
Yeah, I can kick it off maybe. Yeah, so I think if you kind of stand back and I mean, those are good markets. I think the growth rate in those markets inherently tend to be, call it in the mid-single digit, plus or minus range, but 100% agree, these are very long-running markets which have very good margin potential. And I think what we've done internally is said: Hey, look, we know they're gonna get back to this roughly $2 billion plus, right? It'll take some time. And then the growth rate from that is still gonna be, you know, $2 billion plus growing at mid-single is still an appreciable amount of extra revenue every year, which you can drive at pretty decent margins.
But from an R&D spending perspective, those markets don't require the same level of, you know, updating the product every two years as an example, right? So I think internally, we've refocused our resources more towards what are faster developing markets and faster-growing markets. But make no mistake, we very much are investing and making sure that we are very competitive, whether it's automotive, it's enterprise, it's carrier. And storage, by the way, has already come back a lot, right? So storage in data centers, what went down a lot last year, you should assume it's almost back to its normal run rate, perhaps another quarter or two, and that's an area we're gonna keep investing in, right? So I think you should think about...
It's just, you've got this 70% of the company, which clearly has double-digit growth figures with a lot of exciting developments in front of it, which is new opportunities. Versus you've got these more mature markets, where you will still have a consistent level of investment, and which will also certainly drive pretty high margins for us. We like those businesses, just to be very clear. Yeah.
Yeah, and we're investing in them. It just, there's such a TAM differential that it's not even, well, there's hype or it's more exciting. It's just, it's just the sheer size. You can allocate more R&D appropriately and get a better return. But those other markets we do, we are putting investment in. We have teams of people working on next generation products. We're not ignoring them, but we're balancing that with this other big trend that we've got in front of us within our spending envelope.
Thank you for taking my question. I guess at the beginning of AI, we had a very large customer who had a very integrated system, and companies like Marvell offered kind of many alternative, let's say, Ethernet or any ASIC as well. But I guess when we step back and look at Marvell higher level, we noticed that you guys are also having kind of all the different pieces to maybe have a system on your side as well. Is that an opportunity? How far away are we, and are you getting engagement on that side, providing pretty much putting all the pieces together and offering it as a system?
Yeah, well, we, you know, we do from a POC standpoint and a demonstration vehicle standpoint, leverage a lot of that. You know, because as I was saying at the beginning, when we were out of an hour of traffic chatting, you know, figuring out how to really power optimize the whole solution is very critical, and also prove out the IP earlier and in advance of when someone's gonna need it. So that's kind of the way we approach it from a system point of view. But we just basically wanna bring to the table this suite of technologies that our customer can really access. We can work with them, we can optimize it so they can build the system, but we're not gonna go build the system. We're not gonna go make the accelerator ourselves. We're not gonna go... We're gonna go enable all the stuff around that, and if somebody wants us to do a custom accelerator, we'll build it for them, but we're not, you know, we're not gonna go do that on our own. If that was your question? Yeah.
Matt, is M&A still key for you guys? And I know, you Marvell's mentioned as potential suitor for part of the asset that company's trying to sell a couple of days ago on the PLD side. But is M&A still key for you, and do you feel like you have all the right areas of IPs to go after the data center infrastructure market?
Yeah. Well, you know, I think my view on this and our view has been very consistent, actually, since two thousand and twenty-one. You know, we had an investor day, and I said, "Look, we did a lot of hard work. We did what we needed to do in terms of all this M&A. Looks great now. A lot at the time, it was a little risky. Some of the things we did, you know, we bought Cavium, we paid 85% of our enterprise value. That was kind of a big bet. You know, we went and did Inphi, paid a big multiple for that.
That was a big bet." You know, there's a whole bunch of things that now look like, okay, that kind of made sense, but at the time, there was definitely risk, and there was definitely, you know, the integration side and all the hard work you have to do. At the same time, we did a bunch of divestitures, right? There was a big cleanup of the portfolio. So my comment in twenty twenty-one was: Look, we did what we needed to do to build the portfolio that we wanted, and now it's time to reap the benefits. And our thinking at that time as well was, M&A is gonna really be tough to do because of regulatory, because at that point, even back then, it was really clear that that was just gonna get squeezed, and we wanted to control our own destiny, you know?
So as we've kinda gone forward the last three, four years, I think that strategy served us really well. Now we're sitting here with this huge AI, you know, accelerated computing TAM that's kind of opened up to us. It's $80 billion. We talked about the AI Day. It's way bigger of an opportunity we thought we would have, and we're really well positioned. So there are things we can do to get assets, to help smaller things, to help us kinda get technology to go do that. That's pretty interesting. But we've been on a mission since really that time of like, we got to develop this stuff on our own. We can't just wait around for someone to do it. And by the way, now that's starting to show up in the P&L in terms of and just our overall financial model.
Like, we're generating very strong free cash flow now, like, consistently. We're buying back stock. The board did a $3 billion authorization. We're using it every quarter. You know, Willem talked about, "Hey, we're gonna bring share count down, you know, and start offsetting dilution." All these are really positive benefits you get when you have a relatively clean operation and you're self-funding. So the barrier for us to do, you know, big M&A is high. It's not even, is it actionable or not? That's another question, because of regulatory. So I just take everybody back to twenty twenty-one. Go watch what I said there. That's kind of the same feeling we have today.
Great. And then, kind of a villain question on gross margins. You've been very clear that the custom ASIC business is below corporate average gross margins, guiding to low 60s%, but the operating margins more than 35%. Are you guys looking at your gross margin structure, you know, structurally, that the bar has to be reset to low 60s% from 63%-64%? Or if this is gonna wait and see game in terms of when the carrier infrastructure networking markets rebound, and then you kind of revisit where the gross margin needs to be?
Yeah, great, great question. Yeah, it's definitely the latter. I think, you know, and maybe to set the stage, our view is that, you know, when we do a long-term model, then it's a long-term model. So we set the model out there. We do the best job we can based on the information we have and back then. We, you know, we had a model from the last investor day that we've been driving to, and we were in that range for a while. Now, as we went through the cyclical downturn and now coming out of it, and the mix shifting to more custom, to your point, you know, we are running gross margins below that target model.
And at the same time, and to the question earlier, like, the core businesses have not come back yet. And so we don't really know, quite frankly. I'm not ready to call it either way on where that lands. But I think as we progress through next year, and as we see a recovery, and we get a sense of the sizing of what the custom TAM and the custom revenue is gonna be for us, I think at some point we'll be in a good position, right? To say, "Hey, look, here's an update to the long-term model." We'd probably do that as part of an investor day. We'd set that as a multi-year view again, and we go try to drive that.
The change I would add, though, is I think, and we said this in the call actually, as we get a lot of revenue scale now from some of these new opportunities, including custom, even though they carry a lower gross margin, you know, the fall through is pretty significant, so the leverage is high. And so, you know, you could see a scenario where we're driving to our long-term operating margin target, right? Which is really what we're driving towards through next year. That's 38%-40%. You know, with enough revenue scale and if some of these things really hit, you could see long term, maybe an expansion on the OM a little bit, and maybe the GM actually comes down a little bit. So we need to figure out what that's all gonna look like.
I think the great setup for next year is we feel really good about what the top line is gonna look like. Street's got some estimates, we've got our own view, but it's healthy either way. We can control our spending. We're already... We've been spending at a very healthy level. We appreciate the investor support because at some point our revenue came way down, and we were under shipping demand, and we did things we needed to do to sort of keep things tight operationally. But we've kept a very healthy R&D position in this company. We haven't blinked. And so that means into next year, while we can add some incremental resource, we've got, we're well positioned, you know?
So if you can control your OpEx and you've got a big top line lift, even if the GM stays in the ballpark it is or comes down a little, the leverage on the OM could be very significant. So those are, that's kind of what's on our mind, and I think an update to that would come as part of an investor day.
That's it.
But I think the part I'm envisioning is if you envision success, that really looks like, just, you know, with scale and improved, you know, long-term OM model.
Great. We're almost out of time and on that optimistic note, Matt and Ashish, thank you for coming to the Citi Conference.
No, thanks a lot.
Yeah, you're welcome. Thank you.