Here at Barclays. Happy to have Chris Coldren, SVP Strategy and Corp Dev here at Lumentum. Chris, great to have you here.
Great, thanks. Thanks for having us, Tom. A lot of good meetings today.
All right. It's been a good day so far. We'll have one more tomorrow, but let's just start off with the recent acquisition you did. It's about a year ago now.
Sure.
You did Cloud Light. You kinda talked about, you know, intersecting a market that was really taking off in the data center interconnect space. Can you talk about what's happened since that acquisition and maybe kind of the trajectory of that business going forward? 'Cause you obviously had a lot of positive things to say at the last earnings.
Sure, sure. I mean, I think what's happened in the acquisition as well as the overall broader data center market over the past year has been really quite amazing, right? The demand is very intense, and we supply a lot of optical components that go into other people's transceivers. And that was our thesis that, okay, how can we get more exposure to this overall market, get a seat at the table with cloud operators directly as kinda being one layer removed? Since then, that's proven out.
And what I mean by that is not only, you know, do we believe we have a nice growth trajectory for the transceiver business, but also by having that seat at the table, a lot more insight into what we should be doing on the component business and a lot more opportunities to drive the technology roadmap of our component business, which then obviously ultimately will feed into our transceiver business as well. So, very, very good and exciting. And, you know, as we've highlighted, announced a couple new customers for the transceiver business and, you know, expect over time we'll announce more.
So before I get into the different customer details, let's talk about the larger market. So you've heard a lot of different forecasts from individuals in the market.
Sure.
Some as high as kind of like the 20 million forecast number into 2025. Do you guys have a rough approximation of kind of where things settle out this year and where things are going?
Yeah. I mean, maybe simple mental math here is that we think of guys that are building large AI clusters are buying about four transceivers for every GPU they deploy. So I think 20 sounds low to me in that I expect there'll be more eight to 10 million GPUs next calendar year. So that'll drive more like 40 million transceiver units. So, you know, a lot of growth, a lot of units, and then frankly, a lot of chips that go into those units. So, all good.
Okay. So I think like the debate when you guys acquired Cloud Light was that it was primarily a Google provider, right? And they had silicon photonics technology, and you guys were like, "Look, we can take this technology, we can take these transceivers, and we can expand them to other customer bases." Let's start with that Google customer first. So when you did the acquisition, you had talked about a product transition that occurred.
Mm-hmm.
Pretty soon after you did the deal. Was that a speed transition, technology transition? Can you give me your best, help me understand what happened there?
Yeah. I would say most simplistically it was a, you know, you're further down the road and know how to build better product after having a year experience, a better design. I can't opine on what the customer's necessarily doing in their data centers, but I'm imagining they've similarly learned ways to build a better data center. And so that drove a different set of transceivers with better performance, better capabilities that now we're ramping up again. And that's not too surprising. I mean, we also look to other customer engagements we have and, "Hey, can we intersect, you know, in this generation of the architecture, or is it in that sort of the 1.5 version of something before they go to the next generation?" The same thing is true.
Gotcha. And I think that the trajectory from a revenue perspective that you kinda talked about was like a $80 million-$90 million run rate, kinda bottoming at that $50 million run rate in the back half of this fiscal year, really stepping back up. If you look about, you know, 12 months down the road, obviously all these hyperscalers are expanding. We just talked about how you may see a $40 million per account year. Like, what is the right way to think about the trajectory for that specific customer? Can that be double the size of where you acquired the business?
I guess the way I think about it is, you know, the December quarter and March quarter where we're clawing back up to.
Yeah.
To where we last left off, and certainly there's the opportunity to do a lot more, both in terms of just share, but as well on, they're growing, so I definitely think we can do more. I'd hate to speculate if it's 2X, but I certainly think that's not in the realm of impossibility, but we'll see.
Yeah. Okay. And then to the customer expansion side, you had mentioned two new customers. One customer, you said that you had gotten, you had an award qualification early next year, shipments middle of calendar year 2025. And then on this most recent call, you talked about a second customer, which had been in the qualification process, which you actually shipped to in the December quarter. So maybe talk about your ability to get those design wins. Is that something you were approached on? Anything on speed or architecture type that you can share?
Yeah. And maybe at the start, kinda clean up, it's probably a little confusing. It's just two customers wanted to sort of engage commercially versus technically in a slightly different order. But at the end, you know, we're now at a spot where, yeah, I think maybe the way to think about our business is we're focused on the highest speeds over the last 12 months. Probably 80% of our revenue has been 800 gig.
Okay.
A little bit of 400 gig because that's the market. There's still 400 gig shipping. And so these new customers also are 800 gig and some 400 gig, is what we'll be ramping up. And it's something we highlighted in, you know, it seems like an eternity ago, but at OFC earlier this calendar year, we highlighted that there's initial adopters at 800 gig. We know who those folks are. Now there's a wave of other adopters, a sort of second adopter, 800 gig wave. Same thing will happen at 1.6T. There'll be a limited set of folks and then a second wave of 1.6T folks adopting.
In terms of, you know, there were existing engagements pre-acquisition, but I would also say we really changed the nature of the dialogue with customers by, you know, effectively taking a private company public, U.S. headquartered, presented a manufacturing strategy, which was really leveraging some of our vertical integration on components, leveraging our large operation in Thailand. It's a big focus on end customers over time to lessen, if not eliminate, dependence on China manufacturing. So I would characterize it as we have the right stuff, meaning when suppliers or, sorry, customers would say, "What is it I want in an ideal vendor in this new geopolitical world?" But also AI has changed things, the product cycles have moved more quickly. It used to be the bulk of the market was kind of in the, you know, the slightly slower speeds.
Now the bulk of the market's at the highest, you know, bleeding edge technologies. So now, being a vertically integrated player, being a leader in the component technologies, at least from an optical standpoint, what would enable jumping to the next speed? We're an innovator there. And then the U.S. company, non-China manufacturing, that aligns with where hyperscale cloud operators ultimately wanna see suppliers look like. There's not many of us that look like that. And so I think we're benefiting from that at this point in time. Our strategy is to take that, you know, the door's open, the seat at the table, execute on what we have taken on.
What's really, I think, great about our story is what gets me excited about coming to work every day is that, you know, what's being shipped today and maybe, you know, in the near future is largely technology that had already been in the works, you know, pre this.
Yeah.
Big AI surge. So we're kinda just supersizing older technology. It doesn't scale as well as you would like it to scale as you look to the future, particularly around power consumption, but to an extent cost. But we've got a lot of tools in our tool chest to go solve those problems, then bring it to bear. Previously, it was like, "Oh, okay. The meat of the market's not at, you know, once something becomes commoditized, it became high volume." Now high volume is at the bleeding edge. So it's much more attractive for us. And customers view what we can bring to the table very differently. They're not looking for a commodity supplier. They're looking to say, "How do I solve this problem that nobody else can seem to solve?" And it's gonna be solved by high performance electronics, not us alone, and high performance optics.
Yeah. So U.S. domestic-based vertical integration, leading edge versus potentially N minus one, N minus two. The one area I wanted to dig in a little bit more on is the technology side. So.
Sure.
Almost, I mean, I don't think it's coincidence that SiPh was what the predominant technology was at Google, right? And if you're looking at the market today, you're seeing more SiPh deployments because of a constraint in the EML market, right? How much of your ability to, you know, transfer that technology is the driving factor versus the, you're a domestic supplier? Like, can you rank order like where people are coming to you for?
Yeah. Well, maybe I kinda turn around and say that yes, Cloud Light, their transceiver modules were silicon photonics-based, are silicon photonics-based. We continue to develop silicon photonics-based. So, you know, we kinda have all the tools in the tool chest.
Yeah.
VCSELs, EMLs, high power lasers, silicon photonics. Kinda the right tool for the job is kinda the way we look at it, and you know, just pragmatically, they already had existing designs. You can't change something overnight. It takes a couple years to transition to a new product design, and at the same time, we don't have enough manufacturing capacity. We're sold out on our EMLs, so it makes sense for us to continue to drive the silicon photonic products and roadmaps, even though there might be some "margin benefit" if you think about it.
Yeah.
But then we're robbing from selling something externally, which is a margin loss. So right now, the best thing for us and shareholders is to pursue this path. But the real strategy, if you look out over a longer time period, is we're gonna be adding a lot of manufacturing capacity for EMLs and lasers. Too, we believe that the technology roadmap for the indium phosphide-based lasers yields more benefit in terms of higher speeds, lower power consumption over time, and so that you'll see a blending of all the technologies, but maybe a little bit more of a growth in the EML side of things over time.
For us, the logical time to kinda bring EMLs into our transceivers is both, one, we have more capacity, which will happen, you know, next calendar year, but at the same time, when we have an EML technology inside of the transceiver that truly provides a differentiation at the transceiver level, so maybe disproportionate share or price or some other benefit that we wouldn't get by any other approach, then that's a real win.
Yeah. I wanna dig into that a little bit. You're always one question ahead of me. It's. I have to catch up. So EML versus silicon photonics, you've chosen right now, and you talked about this on the last call about selling the EML product externally. There are gross margin benefits to that. The market's in a shortage. In terms of the customer diversification there, are you selling that across a broad range of customers, or are you really selling to a certain few that are kind of at the leading bleeding edge, just given the product type?
At the EML.
EMLs. Oh my God.
Yeah. I would say we're supplying to just about everybody. You know, we there's not that many EML providers in the world.
That's what I'm saying. Yeah.
Certainly there's even fewer that are at the highest speed. So we provide. Maybe there's a few accounts we aren't in, but I would say just about every major transceiver vendor, including those that have some level of their own vertical integration, given both our technical differentiation, but also, I mean, the world doesn't wanna be dependent only on one person. And so including if that's your own internal supply.
Yeah.
So, you know, for us, maintaining silicon photonics also provides a diversity of, you know, supply internally for supply stability in addition to technology benefits. So yes, we're supplying very broadly and, you know, benefiting from everybody else's success, if you will, as we partner with them on high performance EMLs.
'Cause I was gonna say, I mean, like, if you look at people who have actually announced 200 gig per lane 1.6T solutions, it's actually not that many.
No.
And so you have a defined customer set there, 'cause you're not using—well, I guess some guys could use 200 gig for 800 gig solutions, but I would assume you're probably targeting those of higher value. So I guess competitive dynamic is the only other question there. If the market's really good and there are capacity constraints, you guys had a lead at 100 gig EML. You have a lead at 200 gig EML. How far ahead of you are you of the competition? When do you see, you know, increased competition market? We can talk on the supply side in a little bit, but.
Yeah. I would say that there, there's a very small group, two, maybe three folks that can make, us included, 200 gig EMLs. Then you talk about the manufacturing capacity. None of us have enough collectively to supply. I would say that if you look at history, the same thing would happen at other speeds, but transceiver product cycles were four years. So then there would be some stragglers that eventually would.
Yeah.
Figure out how to catch up, you know, maybe two to three years into the product cycle, and that would bring prices down and, you know, competition. I think now with AI driving a two-year product cycle, being two years behind means you're just gonna be left behind, ultimately. That has yet to happen, but.
Yep.
But that I think is a reasonable thesis. So I do think for at least the next few years, presuming, you know, the industry remains on, I mean, GPU and the AI industry remains on the kind of growth trajectory that supply's gonna be a big problem in addition to needing new technology every two years to keep up with the GPU roadmap.
And then at 100 gig per lane, you saw VCSELs really get a lion's share of the market, or at least a generous size of the market against. It's very difficult to kinda break down the actual share by laser. But at 200 gig, clearly VCSELs are behind. So you made the comment, you know, two years behind could mean you're left behind. If there is a technology solution, and you guys have been wonderful VCSEL makers in the past as well, if there is a solution that gets to market, is that something that you think people will use, or is there a real technology barrier to using it at that speed?
Yeah. I think that at 100 gig per lane and what we've all ramped up over the last years kinda what existed as opposed to what's new. I do think that, you know, 200 gig per lane VCSELs will take a while. Eventually, people will figure it out. And then the question will be, do they actually offer any power consumption advantage? And I think there's ways in advanced EML that we can reduce power to the level that they're.
Mm-hmm.
We may have an advantage over VCSEL-based solutions.
Mm-hmm.
So, you know, we're working it in both ways, in terms of we have all the technologies in-house. And frankly, VCSELs will also play other roles in data center over the long run. There's still a lot of electrical interconnect links that are probably gonna be pushed towards optical because the electrical world is gonna struggle to scale to much higher speed and density. That's a place where even large VCSEL arrays conceptually, as you know, we've trailblazed in the consumer world.
Yeah.
But you know, all of a sudden you say, "Hey, I want a high density, you know, multi-thousand or, or even couple hundred interconnect." That's, that's what people are basically asking for to connect literally the pin-outs of switch or, or, or logic, ASICs. So, you know, we're pursuing all paths, but we do believe there's a good couple years here where 200 gig per lane will be, EML based. And then right around the corner, there's gonna be 400 gig, per lane, not right around the corner, but two years out.
Mm-hmm.
You know, if you're two years late, you're gonna be left behind.
Yeah. So I guess the other end of the question is, like you've heard in the transceiver market, either from suppliers of system-based AI architectures and/or from other earnings calls, constraints in the interconnect market that are prohibiting deployments. EMLs is certainly an area that's come up. I'm assuming that there are several areas, but we know that EMLs are a bottleneck. When you talk about your ability to scale capacity, I feel like every two years, and this is not your fault, the market's going one way or the other, and you're scaling or you're bringing down, or you're scaling or you're bringing down. What's the right way to think about your capacity from a quarterly basis or a yearly basis or contextualize your ability to scale capacity from a dollar's perspective?
You know, how much are you willing to scale that with the knowledge that, like, "Hey, maybe this is the direction the market actually is going. We can take a risk here and scale a bit more"?
Yeah. I mean, certainly, as we highlighted in our last earnings call and, and subsequent discussions, that we're adding significant capacity in calendar 2025 and a couple tranches. Now, the reality is those capacity decisions, those decisions were made.
Years ago.
Made it ways back. Today, we're making decisions to invest because it takes quite a while, you know, from building buildings in some cases or acquiring wafer fab equipment, which may have multiple quarter lead times. So, you know, I do expect that as we, you know, exit calendar 2025, that we will have, let's say, 2X the capacity of what we are exiting calendar 2024 at, maybe more than that, and from a revenue standpoint, you know, as it depends on the ratio of 100 gig to 200 gig per lane EMLs because the 200 gig per lane has a higher dollar per square millimeter of wafer size, if you wanna think about it that way, they're higher ASP chips.
Yes.
At a given size. So that also is a revenue uplift aspect of the capacity additions, but we're not done. There'll be more capacity. Our goal is certainly to add capacity as fast, if not faster, than the market is growing, and I think, Tom, there's another important thing to keep in mind that, you know, transceivers are gonna grow as fast as GPUs, maybe slightly faster because, you know, and as the networks get bigger, they become more transceiver intensive, but another key point is that the number of optical lanes within transceivers is also growing, right? Started at four, now there's eight and four, kinda blended between 800G and 400G, and it's gonna go to all eight lane. And then maybe eventually there's 16 lane. And if it doesn't go to 16 lane, then you're gonna need bigger chips.
And so the Indium Phosphide content is actually growing faster than the transceivers are growing.
Yeah.
which puts more pressure on capacity. And something you brought up that I think is also important for investors to understand that even in a silicon photonics-based transceiver, there's indium phosphide lasers. And it kinda turns out the amount of indium phosphide's about the same because yes, you may use one laser to.
Multiple lanes.
Against multiple lanes, but the laser might be twice as big to go to two lanes. So, you know, in our analysis suggests that you're using about the same indium phosphide, whether it's SiPh-based or EML-based. And so because of that, we are very focused on adding a lot of manufacturing capacity for wafers.
I wasn't gonna go here, but I thought it was interesting that you mentioned it. So silicon photonics, obviously handling multiple lanes potentially with a single light source. There is worry that when you move to 400 gig, that you have such a density of light in a given transceiver, it would cause thermal issues. Is that something that you hear? I mean, there's gonna be issues with every type of light source, right? There's cost, yield, VCSELs are two years, one and a half years behind. How do you think about that?
I mean, it's certainly, I think that's more manageable, but I do think it will maybe turn it around a different way. Our roadmaps are to focus on reducing the sort of the power consumption to drive these photonic devices at very high speeds 'cause the power consumption goes up as the speed goes up. So really innovating at that level is a way that then we can pack more devices in a given power envelope. So that's certainly in the works. There's not a wall there that's upcoming that I'm aware of, but certainly we do need to improve to enable the scalability. Again, I go back to that's what makes this exciting because now we're differentiating on performance, availability, not price, right, per se.
Price is always important because there's obviously the less they can spend, the more they can add. That's really the logic here. But I think ultimately being able to differentiate, that's where our key expertise is on the ability to drive performance and scalability, is what's gonna separate winners and losers, I think, over the long run than just who's willing to drop price and go to the bottom.
Okay. Switching gears over to the telecom side really briefly.
Sure.
You used to define the market as transmission and transport.
Okay.
With the transmission side being more the transceiver business and the transport side being more ROADMs. When you look at, like, what you're seeing from a telecom recovery perspective, I think you used to make the comment, like, you know, ROADM deployments are a good indicator for broader transmission health, but it seems like in the market today, we're hearing more about these transceiver sales. Could you maybe describe what you're seeing in your business? Where is that return to strength kinda coming from?
Yeah. I think there's a couple factors that maybe kinda muddy it a little, but I think, you know, you can't ignore the fact that we're kinda coming off of a very large inventory correction.
Yeah.
So as we grow, we may be just growing closer to end market demand, even if end market demand isn't necessarily changing. Not saying that's the case, but there's a dynamic there. And so what you may be growing just like how much inventory was existing of a particular product. But that said, it's very clear to us that data center interconnect, DCI, is growing rapidly as we supply certain components that play into that. They also play into other things, but you can see that that's where strong pull is. Little less pull, but there is improving pull for some other products.
So that's what makes it very clear that today, a lot of what's maybe burning off the inventory and then churning to true end market demand is generally cloud-driven or, let's say, next product, next generation products, where there isn't a historical inventory levels. That's the next driver. And then the final driver, in this sort of order of importance, is just reverting to more normal, run rates as we've been very depressed compared to customer ship outs. I think that third, though, will continue on, right, and maybe reach some acceleration as we get out into calendar 2025. So I feel a lot better about telecom than we have been at least several quarters that we're starting to see things improve. And hopefully history repeats as things tend to improve. Then they tend to improve faster than you expect.
Then when they get bad, they tend to.
Yes.
Get worse than you expect, and, you know, we'll let you know when that starts to happen, but it definitely feels good.
Is that in the traditional, like, I've lost track over the years here? Is that in the traditional long-haul metro markets, or have you really seen that market move more to, like, the ZR, ZR + market where it's kind of like the hybrid data center environment versus the real long distances? Like, if you were to look at your business today, is it entirely modernized because you saw inventory get all the way worked down and new demand is kind of in this new metro long-haul area?
I would say that what we're actually seeing is cloud driving much more today.
Yeah.
I think the service providers will resume and grow ultimately. Internet bandwidth is not going down. It's so they'll continue to need to invest over time. But it's really cloud-driven. But I think maybe where there's not a change right now, but will be a change over the coming year or years is that data center interconnect tended to be focused on maybe shorter distances, but a lot of what is driving and that was driven by certain operators' architecture. Every operator is dealing with the fact that at a certain level of AI cluster size, they can't get enough power. And so they will start separating data centers and saying, "If I wanna do a million GPU deployment, maybe it's in 300,000 GPU data centers," which requires massive amounts of bandwidth connecting those data centers.
They probably need to be further apart 'cause the reason why you're separating them is beyond different power sources, which suggest more than 100 kilometers or 50 kilometers.
Yeah.
Of what we're typically looking backwards for ZR kinda deployments. So we see a lot more focus on longer distance 400 gig or longer distance 800 gig. At least that's what we're seeing start to emerge. And we're also seeing pull for the transports, as you described it. I would call it more line system elements because if you're lighting, you know, putting a lot of bandwidth through a fiber, you need amplifiers and ROADMs and other things that get those optical channels onto the fiber and across the couple hundred kilometers. So kinda pulling in general, a range of our products, just obviously there's more products as you look to the broader service provider world that we're not seeing as much strong pull for, hence kinda the little bit of fuzzy color we can provide on where the growth is.
All right. Last one for me. So we talked about your existing customer, the addition of one customer, and then the second customer.
Sure.
On the transceiver side, on the datacom side. We talked about a recovering telecom. We didn't really talk about industrial and consumer, but we can leave that to the side.
Yeah.
'Cause I'm pretty sure this is not the answer to this question, but when you look at the 500 million.
Sure.
On a quarterly run rate and you build up from where you are today, what's responsible from getting you to point A to point B?
Yeah. I think most of the dollar growth is clearly gonna be in the cloud-oriented business between chips and transceivers just given, you know, very, very strong end market demand, capacity additions coming online in the chips, new customers coming online in the transceiver side of things. So even if you kinda think about that 500 and say, "Okay, even if industrial tech doesn't change much at all, you can have, you know, nothing to, you know, a very modest telecom product recovery." And I'm more bullish than that, but even if you assume modest, you know, most of the growth is gonna be coming from these cloud opportunities. But I also believe that, you know, as we've outlined the kind of the number of customers and capacity additions, I think that more than covers what we need.
It kinda provides not that there's anything going wrong, but it provides, you know, upside if all things go as expected or the ability to hit that number a little earlier than we said, but we're not at that point yet. So, you know, we'll give an update every quarter of progress towards that. But I think it's as Alan liked to say, it's on us, right?
Yeah.
The opportunity is there, and we just need to go execute, and we should be able to get there.
Exciting times. Thank you very much, Chris. It's a pleasure having you here.
Oh, thank you, Thomas.
Yeah.
Take it easy.
Well.
Bye-bye.
Bye.