Why Marvell Wants Lumentum: Vertical Integration in AI Optics With Cignal AI

0:00 All right, hello everyone. Welcome to another semi-doped episode. We are bringing this one news takes to you live, kind of hot off the press. There's big news about Marvell potentially acquiring Momentum. And so Dick and I thought, what better than to bring in our friends from Signal AI to talk about this?

0:20 So Signal AI is a market research firm. They've existed for a little over 10 years and they focused on networking components and hardware. They have clients up and down the supply chain from hyperscalers to laser manufacturers. So I wanted to introduce Andrew Schmidt and Scott Wilkinson. So, guys, thanks for coming on.

0:44 Andrew, do you want to tell us a little bit more about Signal AI?

0:47 Yeah, so we our primary focus is providing quantitative and qualitative market research to companies that are in the industry. Our primary focus is not the investment community, it's providing the data and the you know what is unseen to the companies that are operating in the industry. And I've been doing this for 10 years.

1:09 Scott's been doing this for six years now, Scott? I lost track. Yeah. A while. And um before that, we were both in the industry. You know, Scott will introduce himself, but uh my background, I did uh market research for a different firm called Infinetics Research. And before that, I worked for a semiconductor company for a long time called Vitesse Semiconductor, focused on optical communications.

1:38 So Scott, you want to uh yeah, so I'm the lead analyst here for um for optical components. That covers everything from inside the data center to the pluggable coherent and other coherent devices that are used on long haul. Um after grad school, I worked primarily as product management and a little bit of marketing for some large companies and some startup companies um sort of around the around the world.

1:59 And I've been here I think six years, but uh it's the best job I ever had. It's a great job. I could probably I probably should know that, but it's it's been a while.

2:07 And one thing I gotta add is Scott and I we talk a lot, we don't always agree, and that's okay, but we haven't even talked about this news that came out last night. So it's we're gonna really be kind of off the cuff here on terms of opinions.

2:21 Perfect, perfect. Then it will be even more real, and I think our audience will love it. Um, so yes, thanks for joining us last minute. Uh, we literally, I I texted Vic this morning because uh I think it was like last night, I texted Vic and you know, last night, uh my time was like, dude, we have to talk about this news.

2:37 And then this morning I thought we should bring in some experts. And so Vic said, I'll reach out to them. So thank you guys for just jumping in last minute. I love that you haven't even talked yet. So, Vic, why don't you tee up our audience? What is the news that we're about to dive into?

2:48 Yeah, uh, this is uh quite uh news for the optical industry and all his networking stuff. Uh, these are the like two of the biggest names in the in the area that is always in the news. And it's like watching like worlds collide when I read this news from Bloomberg. It said, Marvel Technology is an advanced uh talks to buy the you know data center optics maker Lumentum.

3:13 Um and it puts it in the cusp of a potential mega deal amid the race to scale up uh in the artificial intelligence sector, I should say superintelligence perhaps. Anyway, uh Bloomberg didn't catch on to that. Um, the announced the transaction may be announced as soon as next week. And uh Lumentum has a market value of like 83 billion and Marvel is like 221 billion.

3:41 And if Marvel acquires Lumentum, it would likely rank among the 10 biggest tech and telecom mergers ever, according to Bloomberg. So it's anyway, it's like a really big deal because as far as I understand it, and we'll hear from our experts, uh, Lumentum is a a prime supplier of lasers to the data center industry, and we are at the cusp of transitioning into the optical uh interconnect world for data centers in a way that we have never seen before in history.

4:14 And Lumentum, in what my perspective is that is is the king, you know, it's the top of the pile here, the the at its peak and prime. Uh, but why would they sell the company to Marvell? And why would Marvell pay so much for it, right? So that's the whole question here. What exactly is going on?

4:35 This episode of Semi-Doped is presented by Crusoe. They build the power, the data centers, and the cloud behind AI. Did you know Crusoe can help you tune an open source model on your own data for just a few bucks? Frontier models are very good, but they don't know your company's terminology, your part numbers, or how you like your reports formatted.

4:56 So you keep stuffing instructions and context into the prompt of an expensive frontier model to get the job done. But for simple, repeatable work like reports, an open source model trained on your data can actually match a much bigger model. Think of a new grad with three months of on-the-job training.

5:13 You don't always need the PhD. And the open model is cheaper and faster to run. Fine tuning used to mean renting GPUs, working on orchestration and Kubernetes, writing training code, and babysitting the entire setup process. But Crusoe's serverless fine tuning removes all of that. Upload your data sets, pick an open model like GLM, Llama, or JeepSeek, and they handle the training.

5:38 You pay per million token used, not per GPU hour. And on the smaller models like Quen38B, $2 can take you a long way with millions of training tokens available. Once the training is done, you can then deploy them on Crusoe Cloud in just one click or download the weights and run them anywhere. Your data only fine-tunes your model.

5:58 And it doesn't have to be work data either. You can try it on how you write emails. Pick a task, compare the tuned model to your current prompt, and see how it does. Crusoe's giving all new accounts $5 in credit to get started today. That's enough to start tinkering with fine-tuning right now. Links in the description.

6:17 Yeah, so let's hand it over to our signal AI folks. And feel free to zoom us all the way out and just remind listeners like, what does Lumentum do? What does Marvell do? Why would Marvell acquire them? You guys take it away.

6:30 So the the thing to re-if you were to rewind the clock 10, 15 years ago, there were tons of small optical component companies. Um at least seven or eight that I could think of off the top of my head. And over the last 10 years, there was a you know a consolidation of all these companies into just a few.

6:52 And in essence, right now, the the two biggest suppliers of optical components, not counting people who do the assembly, like um some of the folks in China, are coherent and momentum. And so we we transitioned from a period of time where there were tons and tons of companies all competing with each other to really just having two major suppliers of lasers for inside the data center, lasers to power coherent optics, um, and components that are used in line systems.

7:26 And now what we're seeing is that these components are obviously a lot more valuable today than they were five years ago. And people, I think, are beginning to recognize the importance of vertical integration. If you want to be strong in the market, particularly in a period of time where things are moving fast, you need to control the development and you need to control the supply chain.

7:46 Well, in that case, you need to have both the chips and the optics in order to have a solid footing.

7:52 But to be clear, this will be the first company that owns DSPs, all the DSPs, lasers, and modules, right? I can't think of any other.

8:01 Well, Acacia has maybe some of that. Um, that would be the closest uh analog, I think. But you're right. I mean not on the scale.

8:10unknown Yeah.

8:10 What does uh Broadcom play a role in this? Uh do you think uh the Marvell acqui acquiring Lumentum puts them the closest because Broadcom actually has internal lasers too, and now Marvell will have internal lasers too?

8:24 Yeah, I think the difference is that Marvell is is heavily in the module business and Broadcom is not. And so you you can you know Broadcom, their optics business, other than some of the work that they've done at CPO, operates at an arm's length from the rest of the rest of what they do. But at Marvell, they're very closely coupled on the coherent side in terms of building DSPs and shipping those modules.

8:49 Uh so it's it's reasonable to think that they would they would do the same thing for uh Pam 4-based data center optics uh if they were to bring something like this in.

9:01 But what's interesting is that with Marvell, they have the coherent DSPs and they've done very, very well, especially 400 gig. The idea was that they would sell them to anybody, but they would also build their own. And that's been relatively successful. There's a lot of people out there using them, but if you look in terms of volume, it's very hard to compete against somebody who's supplying you and competing against you.

9:21 And that's what some of Marvell's customers have sort of sort of complained under the covers about uh that it's it's hard to compete against Marvell when they're building the same thing that you're building. But on the Datacom side, that's never been an issue because they've sold Datacom DSPs to everybody and they've never built a module before.

9:38 Um well, it'll be interesting to see how that plays here. Certainly on the Datacom side, any module anybody can build, someone's gonna buy it. Um there's just there's not enough capacity right now in the world, but that's not the same on the on the coherent side. But it'll be it'll be very interesting to see for the first time we have somebody who who is trying to be a commercial provider of the DSPs and competing against those customers with their own completely vertically integrated internals.

10:06 Um that'll be unusual.

10:09 Totally. So I'm hearing a couple different things. One, I'm I'm hearing talk about uh for Marvell that this is a vertical integration play, so they can own things, you know, from the fab to the lasers to the DSPs, potentially all the way up to the transceiver, kind of the whole nine yards. Um then on the other hand, I've heard people also talking about um maybe this is really just about laser supply and capacity, uh, just essentially like they're trying to buy capacity to make sure that they can do what they were already planning on doing.

10:46 Are those two different ideas? Are they one and the same? Like how how do you feel about the second idea is wrong.

10:52 I mean, they Marvell doesn't need laser capacity. The only thing they buy are tunable coherent lasers for their ZR modules, and you know, they have very strong relationships with you know several suppliers. So I don't think that's it.

11:07 No, this will be a new business. I mean, they they have sort of a startup company that they're working with to do CPO with their stuff, but that doesn't need the kind of supply we're talking about with Lumenta.

11:19 That's interesting. So um you also mentioned, Scott, that uh there is a certain competition that's coming up because um Marvel is competing against their own customers now. Uh just so that I have that clear, is that coming uh from the DSPs that Marvel is doing? Um and like on on the other side, like uh who is the competitor that you are referring to and how does the dynamic work really?

11:52 So Marvell right now, for example, they their coherent DSPs are used by a lot of people um who build 400 gig uh modules. Essentially, if you're not one of like Acacia Sienna Nokia Nokia or one of the Chinese guys like Huawei, um you're buying it from Marvell almost always. There's some others. Okay. Siena sells some, Acacia sells some, but um a lot of folks out there are buying them from Marvell and building those modules.

12:22 And we have at least at least one supplier who was doing that told us that they are essentially going to have to exit that market because they can't compete against someone who's vertically integrated with their own DSP. Now, if if you're working in a specific market where Marvell doesn't play, like China, or you're building a specific device that Marvell doesn't build, then you're not competing against them.

12:42 But in Datacom, those differences kind of disappear. Um so it'll be interesting to see if if anybody cares. Um again, the volumes right now are so big that nobody else can build as many as as need to be built. So at least for a short while, I don't know that it will matter, but it will be the first time that that's happened where someone is building the DSP, using it internally, and selling it externally.

13:05 Now Lumenta's been doing that with lasers for a while. So it's not entirely unprecedented. But the DSPs are a little harder.

13:14 So one follow-up question. So if you said Lumentum is doing that internally, uh what what are they they don't build DSPs, do they? So what uh how are they using uh lasers internally? So they they Lumentum sells some modules.

13:30 Those modules primarily use their own internal lasers, except when they've talked about sometimes it's cheaper to buy them externally and internally build higher margin things. Um and that has not been a problem, um it doesn't appear to be at least, uh, that they've been able to do that. Uh Coherent, their competitor essentially doesn't sell the lasers externally, they use them all internally.

13:53 Um but that's the kind of model to be looking at on. We've just never seen that model on a DSP where somebody's building the DSP internally, using it on their own internal modules and selling it externally.

14:03 So I mean a good example would be your Lumentum, they bought a company called CloudLight. They make silicon photonics-based data comp transceivers. So in order to build that, Lumentum takes their own internal CWEM laser. They take and they use that as part of the optical front end that they also manufacture in silicon photonics.

14:24 Then they purchase a PAM4-based DSP, you know, from Broadcom, Marvel, or something like that. And those two things, the optics, PAM 4 chip, are almost all of the value that are inside the transceiver. The remaining value is putting it together. So the challenge becomes there's this one part that's very valuable, this Pam 4 trend, this Pam 4 chip.

14:47 So if Lumentum is building these modules and they have to sort, or let's just say another company is building these modules and they have to buy the Pam4 chips from Marvell, well it's in you're in a situation where Marvell isn't necessarily going to sell those Pam 4 chips at such a price that they could be undercut by their competition.

15:09 Um, the only thing that's that's different about this market is there are other suppliers of Pam4 chips. So even if they become vertically integrated, somebody like uh Inalite can still go to Broadcom, they can, you know, maybe even go to NVIDIA, they can go to Credo and source these things. So there is a market for these DSPs, but as Scott said, this is the first time we would see somebody who's completely vertically integrated on the Datacom Transceiver side.

15:37 I have a follow-up question there. So if Marvell has uh a ton of shares selling these DSPs, and now they're buying the laser so that they can do the optical engine and the DSPs. And but but with that, they're essentially competing with their customers. That would suggest that over time their customers are going to start sourcing their DSPs from someone else, and that uh Marvell's DSP share could be eroded a little bit, but at the same time, they have this new business of selling the end-to-end component, the transceiver, the module, to these big hyperscalers.

16:16 Um, is this something where maybe they're okay with losing some DSP business to gain this other business? And by the way, if we're moving to NPO and CPO where it requires less and less of a DSP, is this the right kind of long-term trajectory anyway?

16:33 I think that's a that's a that's a great way of looking at it. Is in essence, you're making the trade that you know by vertically integrating, I can gain more value than by selling the piece part, even if it means even if it means I might sell fewer of those parts. But I mean, it really goes beyond all this that we've been focusing on because you know, Scott's done a ton of work on OCS.

16:53 Um, that's a that's an emerging area where there's a ton of you know startups, but really only one supplier that's shipping in major volume. Um and that's going to become a critical piece of infrastructure in the next five to ten years inside data centers. And Lumentum also has, yeah, as you pointed out, a lot of work that's been done in high power lasers for CPO and NPO.

17:16 And um, so that that dovetails well with the direction that Marvell's going in terms of trying to move to greater photonic integration.

17:25 Yeah, it'll be a strategic decision internally. I mean, assuming this goes through, but um I think it might not even happen. Right. Yeah, it may not even happen. But I think if my business card said uh DSP salesman, I'd be a little worried.

17:38 Looking at it from Lumentum side, uh you they have pretty good technology across many domains 200 gig EMLs, ultra high power lasers for CPO, um OCS, and when you mentioned, Andrew, that there's only one company shipping this at any kind of volume, even though there are a ton of startups working on all kinds of variations of OCS, uh you're referring to Lumentum, right?

18:05 So this is one company that has all this portfolio of products. And why would anybody consider uh getting acquired by Marvel? Like, wouldn't the board of directors turn it down and say, like, no, our growth prospects are like far more superior because we have all the best technologies. So why why would they even entertain this acquisition deal?

18:27 Uh when you get somebody's offering you a deal that's probably you know beyond 10x your current revenue, it's kind of hard to turn that down. I mean, things can go wrong. So I I don't um I mean it's a question of valuation. Yeah, locking the value down. Yeah. So I I um that's not our that's not our jam, right?

18:50 We don't we don't look at stocks, we don't come up with valuations, we don't try to figure that out. But my my guess is that's that's the calculus that they would be going through. And if if I recall the CEO Hurlston of Lumentum, you know, his his last posting was at Finissar, and he was he was there for only I can't remember exactly how long, but it wasn't long, maybe a year or so.

19:13 And yeah, he organized uh the the deal to sell the company to Coherent. So it's consistent with some of the work that he's done in the past. Um so we'll see. Uh we'll see what happens. My my personal feeling is that that lamentum is probably just as valuable or more valuable in the hands of you know someone else who wants to vertically integrate.

19:36 Um, you're talking about you know, even companies like let's say NVIDIA. I mean NVIDIA bought Mellanox years ago as a way to vertically integrate the networking components with the GPUs, and it's logical to think that if if they want to continue to that down that path of vertical integration on the component side, they do need an optical component and they also need an OCS component.

20:02 And remember, NVIDIA put two billion dollars into the momentum.

20:05 Yeah, also into Marvell, though, and coherent. Yeah. So what happens in this uh this crazy dynamic of NVIDIA being involved in this? So what happens to uh I have like several things can happen. So what happens to NVIDIA's two billion dollars in Lomentum uh when it lands in the hands of Marvell, who also got two billion?

20:26 That's one question. Secondly, now that nobody else other than Marvell has access to Lumentum's uh lasers, will everybody flock to Coherent?

20:37 Well, first of all, I think NVIDIA got equity, so they already own a small portion of uh portion of Lumentum and Marvell. So that's that's really I don't think that that merger is really much of an issue. They would just end up owning some of those companies. I I don't want to speculate what would happen in terms of people flocking to you know from Lumentum to Coherent.

20:58 Um that's that's kind of far down the road, and a lot of that will be dependent on how Marvell manages the merger. They they they may come up with arrangements that that give their current DSP customers or Lumentum's current optics customers the the peace of mind that they need to continue to to use those products.

21:19 So it's too early to jump to conclusions about something like that, I think.

21:22 Yeah, Lumentum has expanded their indium phosphide capability uh what four times in the last couple of years. I don't remember exactly the numbers, but they they had a two time and then they had another doubling. So they are definitely scaled to sell to a lot of people, spill lasers to a lot of people.

21:37 I don't think they could use them all internally.

21:40 Right, right. So Lumentum is not taking Marvell will not take Lumentum capacity away from customers. They'll continue to sell to Lumentum's customers, right?

21:50 It is that what we're saying. Remember, Marvell's not making uh datacom transceivers, so it's not like they're sourcing those internally anyway. They would take Essentially the source and the sink both internally. So it's not like there's some new demand that's going to materialize because of that. Yeah.

22:06 Um unless they up their capacity for transceivers, which they they could do, but it it's not like they're if they're already scaled to sell lasers externally and use them internally within Lumentum, that's not going to change if they become part of Marvel. It's just that other component of it, the DSP, becomes less expensive.

22:23 That makes sense.

22:24 And in terms of uh Indian phosphide capacity, uh, when it goes to the hands of Marvell, do you think Marvell is in a better position to help expand indium phosphide capacity by further building out more fabs? Or is that like nothing to do with it?

22:43 I don't know. I think Lumentum is perfectly capable of doing that without Marvell.

22:48 Okay.

22:49 I don't think there's any there's any real link there. I and I um okay, okay.

22:57 So it has nothing to do really with uh with the fabs uh and the expansion capacity whatsoever. So in your mind, I would say, like, what do you think is the grand plan behind this entire uh what what's your if you have to tie it all together and come up with like a conspiracy theory of why Marvel would get momentum, what's your theory?

23:19 Oh it's it's what we started off the conversation with. It's about vertical integration. It's always vertical integration, okay. Yeah, we've gone through a period of consolidation in the component industry where you know these products used to sell for 25% gross margins, they're now selling for almost double that.

23:34 And the value of those components is recognized to be much higher. So if you can take those and integrate them into a larger system, essentially wrap sheet metal around them, you can make more gross margin dollars as a result. And that's that's the opportunity that's that's being seen here. There's there's also the the co-development aspect of you know, if you're controlling the optics, you're controlling the chips, you're controlling the system potentially, depending on where these things go.

24:00 Um you you can you can more tightly couple and build a superior product.

24:06 And if there's any time to do it, this is the time to do it. Everything is up and to the right. Um all your projections are gonna be positive. You can see upside to just about every decision you make. Um strike while the iron is hot and do it right now. I can understand that the timing makes sense. It was a little surprising, but it does make sense.

24:24 Yeah, I mean, I honestly the the the combination of semiconductors and the chips with with the lasers is not a new idea. Um I've been talking about it for more than 10 years, especially when you had all of these optics companies that are were selling for three or four times the gross margin dollars. But the resistance then, because I had conversations with companies in the industry as well as some of the folks on the investment side, is you know, if you're a chip company and you have to you have a 50% gross margin business, why do you want to go buy a business with 25 or 30% gross margins?

25:01 Like everyone's gonna hate that on the investment side, yeah. And it was to me, it was always hard to understand because who cares what the percentages if you're gonna be making more money in terms of dollars, but it's a lower percentage of the total, it doesn't matter. Uh, but there's just a lot of importance assigned to that gross margin percentage number in terms of investment calculations that nobody did that.

25:28 And so now the ironic thing is we have a company like Marvell, who could have bought Lumentum for three billion dollars six or seven years ago, is is potentially gonna pay you know thirty times that.

25:41 Yeah. And the the margin comment, it's interesting. Look at what Hurlson's done since he's been there. Almost every call he's had has been focused on improvement margin of the margins.

25:51 But I I will say that that he's done a great job in terms of extracting the value of the product in the in the current environment. And we've transited that industry has transitioned from one where you know people would tolerate lower gross margins to you know really structuring the deals and and putting things together to get a lot of value out of these components, which you know, in the case of Lumentum, they have some very unique unicorn type devices that you just can't buy anywhere else.

26:22 So you don't sell something like that for 25% gross margin. And I think they've done a good job at at transitioning out of that mindset.

26:30 And what are those devices you're referring to that are like unique unicorn devices from Lumentum? Is it the US? 200 gig EMLs.

26:37 I mean, they were really one of the first ones out with the 200 gig EMLs. You know, everyone knows about the OCS, you know, everyone knows about the high power lasers, a lot of debate about you know whether they can sustain that or not. There's a lot of people chasing them, but um, I mean, I've I've been in the component industry my entire career, and it's real easy to say that you're gonna go do something and build something similar, but and maybe you can build a hundred of them, but when you need to go build 50,000 of them or a million of them, totally different ball game.

27:09 And so that's that's the real challenge when you see you know companies that are small and large, you say they're gonna get into this business, it's never that easy.

27:19 So I'm reflecting on the the margin conversation, and it feels like nor historically, maybe as a component supplier, that part of the value chain would have these lower margins versus someone who's taking all the components and building like a system, right? Because it if you're just a component and there's other people, you know, maybe it feels a little bit more of a commodity, which you know, feel free to push back on that.

27:45 Um I I'm and then, but then we're in this like AI era where some of these components, that's kind of been like the bottleneck trade. There's like some of these components are a bottleneck, there's crazy demand from hyperscalers and maybe a little bit of limited supply. And so it feels like you can get especially rewarded if you're one of the companies that can get to the next speed the fastest or do or do something first, where sure it was a component, but now you have these 200 gig EMLs, um, or you're like kind of first to OTS, and therefore you can really uh charge for your value, let's say.

28:20 Um does it feel like, since you guys have been in the component industry for a long time, does it feel like we're in this era where component companies can make higher margins than they historically have? And would you see that ever coming back? Or has like this AI dynamic changed things? Like, for example, I think of Credo.

28:40 Like when I first heard about AECs, I was just like, oh, okay, it's a cable. Yes, they have the DSP, it just it doesn't feel that important. But then obviously it was very important, and they were the first one there, and they could totally charge for their value on that. Um, so yeah, like zooming way out.

28:56 Like, help me, how should we think about the components structural the component industry structurally, and is it changing?

29:06 I I don't know. I I don't know what you think, Scott. I don't think anything has permanently changed. And the the idea that the laws that have governed you know how this industry works have been somehow repealed or not possible. So, but we have there's a thing I like to call IEEE lottery, right? So there's a new standard that's gonna be coming out.

29:25 Let's just say, for instance, 1600 ZR, and there's gonna be five or six companies that all try and build to that standard and deliver a product. When you fast forward a few years later, the reality is it's always about two companies that ship 80 to 90 percent of the volume. And I don't think that that's gonna be any different as you start looking at these new technologies.

29:45 Is you wanna be you gotta be first to market with enough of what's necessary in order to win the initial market share. And at that point, it's a downhill fight.

29:57 I would say it's a little different this time because it's it's not just being first, it's volume. If you're first to market and you can make 10,000 of these things, doesn't matter. If you're second to market and you can make a million, you get the business. And we see that with companies like eOptolink coming in a little bit late, maybe at 800 800 gig E, and now they're I think the number two in the world.

30:18 Just because they had the capacity to build as many as anybody wanted. Um they may not have gotten the initial build, but they're certainly getting it now. And we're seeing the same thing happening at 1.6. 1.6 gig E, right now 106 terabit Ethernet is it's starting to roll out right now. It's not a huge market, it's starting to grow.

30:35 The usual guys are sort of there at the top, but it's because it's not necessarily that they were the first ones to build a DSP or to build the module or to build a laser, they can build them in volume. So there's it's a little different this time, but only because we're capacity limited.

30:53 I think the other thing that Scott's the other thing I agree with Scott, the other thing that's different this time is the amount of value that's attached to the delivery and the enablement of these components is much higher. I mean, when you're building coherent lasers and they're going into ATT's network and it's going to light up some additional capacity, okay, and how much is ATT?

31:14 What's their revenue associated with that investment? But when you look at if I can if at my AAI cluster has access to you know 1.6 terabit interconnect, and you know, I can train two or three times as fast, that's incredibly valuable. So the the optics now are are potentially enabling a massive amount more volume, excuse me, more value than they used to.

31:39 And that is different, right? That's not how things used to be.

31:43 Ah, yes. So this is really good. So there's a lot of value that can be unlocked if you win the IEEE lottery, so to speak. Um, does it matter also that there's like four or five buyers, and when they buy, they're buying at massive volumes as compared to previously?

32:01 Yeah, I mean, the customer concentration is definitely new as well. I mean, you have you have companies who, I mean, OCS is a great example. You can build a massive business with one customer, right? That's it, it wasn't like that before. I mean, maybe like if you had one AT ⁇ T or Verizon, like that was that was a big piece of business, but but now um and we see the same thing on the hyperscale side, and and the numbers are even larger.

32:28 There there are modules being built for a single customer, like 2.4 gig coherent light. One customer in the world that wants that. There are people out there actively competing for that. Marvell's one of them, competing for that business, knowing that they will only ever sell to one customer.

32:43 Yeah, Marvell did the same thing. Well, I should say InFi did the same thing with the Color Z 100 gig at Microsoft. Yeah, that was uh essentially a single customer product from the very beginning. And a lot of other companies looked at the business, and at the time it's like, why would I want to go do one development for one customer?

33:00 Are you crazy? And uh that was the mentality you know of a decade ago, and it's not that way anymore.

33:08 Totally. So then taking it one step conspiracy theory further, maybe, do you think Marvell, by having an XPU design business and having relationships, potentially designing for these exact hyperscares we're talking about? Well, you know, think yeah, XPUs, TPUs, that kind of thing. Are they even better suited to become vertically integrated, potentially, you know, by Lumentum because they also have that relationship?

33:36 Or is it no, they're fine as just a networking company anyway, they still would have won just as a networking company.

33:42 I don't think that really comes into play until CPO is mainstream, right? Um, that's that's the one thing that that that is out there. And you have at this point essentially TSMC, which is the gatekeeper to you know having production grade CPO technology in volume. Um the the end customers don't like that, right?

34:06 They want to have multiple sources. So if you could have someone who can build the ASIC, can package it, and then can provide the optics with it, that's again vertical integration, that's the value, that's very compelling. Um, but that's down the road a little bit. Um, and I mean I don't conspiracy theory isn't really a word I like to use, but um that would essentially that would most certainly be like if this deal goes through, that that endgame would certainly be prominent in terms of where they're gonna take the company.

34:40 Yeah, it's interesting if you think about NVIDIA, whose people sort of think is their competitor for that kind of business. They have a module company, they bought Millinox, but NVIDIA is building systems and data centers. Marvell is not. They're build their their biggest business, as far as I understand, is they're building custom ASICs and others that people you know they'd put into systems.

35:00 So the link directly between that business and uh modules is pretty tenuous, at least the way it's set up right now. Not like NVIDIA. NVIDIA you can see them very tightly together. They yeah, they want to build you the box, they want to plug the their optics into it, they want to build the switches and routers that go with it, they want to sell you the software.

35:16 Uh Marvell's not really doing that.

35:18 So it looks like they have different parts of the business. This was a question I think uh somebody asked at Computex, I think, or some some conference I was at. They're like, we look at you, you say you're a connectivity company, uh, but then if you look at it, you have you're this the CXL business that you're doing, and then you're you're doing these like uh X XPU custom silicon design business over here, but you're not really like a cohesive business unit.

35:46 You seem to have these pockets of other things that are happening. So I think that's your fundamental point is that these are isolated pockets of business within the company. The connectivity is probably the most cohesive one, but then maybe the XPU is not as related to this at all.

36:03 Although maybe I would push back and say, shouldn't that be Marvell's North Star? Of hey, now we're gonna be fully integrated on the optics side. Um, now we have on the XPU side, we have all this bespoke IP CERTIs and whatnot. Couldn't they make custom data centers at scale, maybe with like the best co-design or the fastest time to market because they are vertically integrated?

36:28 Uh yeah, that's that's outside my strike zone. Yeah, it's getting outside our wheelhouse pretty far. I mean, because then you start talking about software integration.

36:36 This is all very, you know, it's always easy to speculate on what somebody could do. It's a it's another thing of what's real, what is a realistic path that they can they can take. I mean, back to um back to vertical integration. If you know Broadcom, they got super strong over the last 20 years, and they ran a playbook across a lot of their businesses.

37:00 And the one that I saw most closely was in the Ethernet switching side, is they were one of the first ones to bring both the Ethernet switching silicon and the copper fies and all the physical layer stuff together. So if you wanted to go build a 10-gigabit Ethernet switch, you'd go to Broadcom, they'd give you the switch chip, they'd give you all the copper fies, they'd give you the software, and they would give you the reference design, and you could just go build it.

37:26 And they would price that such that if you wanted to buy just a switch, it wasn't attractive. But if you if they could kit price the entire thing, it was attractive. And they would price the kit in such a way that they could strangle some of their competitors that were just doing point products. And that's effectively what they did, and that's how they got to be the market share leader, is they were really good at building stuff and they built it on time and they built everything, and so they sucked all the oxygen out of the room.

37:54 And I think when you look at what as this going to this transition to CPO, that strategy is gonna be even more valuable. Is if you can build the chips and you have the phys and you have the optics and you can take that as a kit to an end customer, that's gonna be a really powerful solution. And we've been talking about CPO, but NPO matters as well, right?

38:13 You could have the same approach with NPO and the chips, have a tightly integrated phi. You buy my NPO, you know it's gonna work with my ASIC, I'll take care of it. I've got it covered. So, you know, whether whether that's the end strategy, I don't know. It would seem to be the logical path, and it worked very well for Broadcom.

38:33 Yeah, absolutely love it. Um, I'd not to take this too far, but do you have any other like thoughts around this whole acquisition? What should we be looking for a potential acquisition, I should say, what should we be looking for next, apart from the approval process? Uh what uh what what what would you expect coming uh is coming down the pipe based on your long experience in this industry?

38:58 I I think the good chance there's gonna be a bidding war. And it wouldn't, I mean the fact is, is this information leaked? And if it if in fact it is true, like I don't I know nothing. So, first of all, I want to be clear on that. I don't know anything about this. Um it it's clear that somebody leaked the information in an intentional way.

39:19 I mean, this is probably not an accident, and so we'll we'll see. Okay, potentially maybe they're they're trying to generate more interest for other bidders to let them know that there's activity. Um, I think there's a good chance there's gonna be other people looking to pick up this company and and run the same vertical integration playbook that we were just talking about.

39:36 And NVIDIA is you know a potential customer for that.

39:40 All right. I think we can wrap it here. This was a great conversation. Uh my takeaway is definitely that this is all about vertical integration. I think that makes a ton of sense to me, whether it's um being able to co-design, being able to run faster, win that IEEE lottery, um, and also just the experience that you can give your customers by being vertically integrated.

40:02 So thank you, Signal AI guys. This was a lot of fun, and let's do this again sometime.