Lumentum Holdings Inc. (LITE) Earnings Call Transcript & Summary

Yes. Okay. Let me describe it again, a super question because there's just a lot of range to it. So the large customer, NVIDIA, right, really is adopting CPO. What is CPO? CPO simply means that the optical engine is going on the same substrate as their main compute, whether that's a switch or whether that's their GPU, okay? NPO, what does that mean? NPO means that the optical engine is off the substrate. It's somewhat decoupled from the main compute substrate, the main switch substrate, if that's where you're trying to put the optical engine. And that opportunity for us is we view as, again, very much additive but actually larger than the CPO opportunity, at least in the near term. Why? Because what we see from customers is a higher adoption rate of optical lanes, at least to start than we would see from NVIDIA, very high adoption of optical lanes. That's, again, great for us. So NPO, NPO has a number of different forms. One form that you talked to is putting the laser in the same package as the PIC as the laser driver, making it a complete optical engine in a single package. For us, again, we see ourselves as largely a component supplier. We want to supply components. We also see ourselves as a great packaging house. If you look at everything that we do, you have lasers that ultimately get into a package. And that packaging capability is an underappreciated part of Lumentum. We can apply that packaging capability to the whole optical engine, integrating the laser in with all of these other components to make a complete optical engine in the case that you have the laser inside that package. That's super interesting and very complex. A number of the NPO and certainly, the early CPO architectures are different. They pull the entire laser content outside that optical engine to a separate block. We call it an ELS, an external light source, and that actually moves off of this whole substrate issue and out onto the face plate of the tray. And in that configuration, again, we can compete effectively for the ELS. NVIDIA, look, they have the best optical engineers in the world, right? They can compete with us any day of the week. Still, they decided to adopt in their early architectures this ELS, and we think we can compete for that ELS. It's not so much a high degree of packaging capability there, right? It's more an issue of we control a lot of the build of materials, and we think we can get a decent margin while competing for a bigger piece of the revenue pie. For other customers for NPO, they've interestingly enough adopted a very similar architecture where they've decided to pull the lasers out of that optical engine, creating again an ELS. And there, because the optical capability might be less than in NVIDIA, they're actually requiring us to ship the entire ELS to participate in their opportunity. So great revenue for us because we control these high-powered lasers, which are extremely difficult to make, we are able to participate at great margins and in a pretty meaningful way.

It sounds like you have a great opportunity with NVIDIA on ELS.

Yes, and other customers. I mean I think that, as the NPO customers, Gianni, think about deploying, several of them are saying, look, we want to take this NPO architecture in the same direction that NVIDIA did and deployed this ELS.

Yes. So the health of the transition, I guess, the CPO. Okay, great. So perhaps we'll shift on the fab side of the business. Let's touch a little bit on to the supply chain. You previously mentioned Greensboro, we'll see first revenue in 2028, and I believe full production in '29. This is a fab size for billions of annual revenue. You currently have multiple LTAs and capacity, right, signed. So 2 things here really. Firstly, what did you see that gave you the conviction to commit first? And secondly, now that there are customers queuing up to underwrite the capacity, how do you decide how much of the future build stays on your risk versus theirs?

Yes. Look, this is a complicated problem, and you're not the first one to ask about it. And that is we have to bet so far ahead of when we would see orders. Our build cycle, we talked to the investors about acquiring that Greensboro fab in early calendar 2026. And it's not operational until early calendar 2028. So it's a 2-year cycle to get that darn thing into production. And that's pretty fast. I mean, I'd say our team that is doing it is probably the best in the planet. The team that we inherited from Qorvo that drives that fab is a really good team. And so I think 2 years is a pretty aggressive cycle. So we had to invest in 2 years ahead. Now we obviously had these discussions with NVIDIA that have emerged with, one, a very substantial long-term agreement that we've characterized as a multibillion dollar deal. And then, of course, the investment where they helped us considerably buy that fab and equip it and outfit it. So it's been a great partnership with them. We are looking to get similar types of deals with NPO customers. The NPO customers, as I said, we see, at least in the near term, a larger opportunity with the NPO customers than we do with the CPO customer. We believe the CPO customer, if we -- if they want it, they could take all of that capacity and then we'd be in the business of looking for additional capacity. At the moment, that hasn't happened more. I think, timing issues. I think they certainly could just given the demand signal we're seeing. And so what we're trying to do is lock up additional business now with these NPO opportunities to fill that fab up and more, right? And it may end up being something where we have to go out and get additional capacity down the road. But as I say, this NPO opportunity has surprised us with both the timing and the magnitude vector. And in order for me to invest, I would definitely need to see a similar type of commitment as we ultimately got from NVIDIA. They've been a great partner. I think they're a great partner for the entire industry, but certainly for us in terms of laying some of my fears about expending all of this money and deploying it into a fab that has a multiyear cycle before you see it pay off.

If that additional capacity would indeed come, it would be like a 2028 story, right, onwards? It will be -- because that's when you see inflection of the CPO, right?

Yes. I mean, look, again, what surprised us with NPO is, again, timing and magnitude vector. I think prior to our last earnings call, we probably thought that it was going to be a late 2028, into early 2029 event. It seems like the market has sort of caught on to the same physics issues, I mean, no surprise. But in order to compete effectively with NVIDIA, they've said, look, we need to deploy a lot more optics, and the racks, we need to do so sooner. So it seems like now it's a late 2027 into early 2028 events. And again, the magnitude is such that it's going to put a lot of pressure, a lot of pressure on that Greensboro fab. So we need to think now about our next leg of the stool, what we can we do to add more capacity. We have a lot of differentiation, I think, in this high-powered laser. And as you and I have discussed, a lot of this is the fab. Your question talked about fab. We're part TSMC, part semiconductor design house. It's really a marriage of the fabrication facility and our design capability that makes all this possible. That's not like NVIDIA, Broadcom, the great semiconductor houses where there's a strict delineation between them and TSMC. These are very much intermingled. Our process, our process control, our fabrication know-how is intermingled with the ability to design and make these lasers at reasonable yields and reasonable margins.

Maybe staying on this fab theme. You're expanding to indium phosphide fabs in Japan, qualifying CW and EML on the newest tools simultaneously, holding to more than 50% EML unit growth by December quarter. We spent the last 3 months securing additional substrates because ultra high-power demand around the supply. My question is, what is the binding constraints over the next 18 months? Substrates at capacity, tool times or people?

More than anything else, it's time, right? I think in the fab business, there's long cycles. You have long cycles to deploy equipment into the fabs. You have long cycles to qualify products on your own tools, in your own systems. And then you have long cycles to qualify these lasers at the customer. I mean one of the things that we've said, and it's good to remind people, it's a year cycle for us to get qualified at NVIDIA. They're very discriminating in terms of how they test and qualify lasers. There's a lot of fear. Some of it founded, most of it unfounded about the reliability of these lasers. And we've been able to produce really high-quality, highly reliable lasers, but there's a huge testing cycle that goes into these things. So that's the long pull in terms of deploying capacity. It's time. There are other issues. You touched on a lot of the equipment, right? There's a lot of constraint now in terms of getting reactors, getting e-beam litho tools into these fabs, there's a constraint in terms of the substrates. The substrate problem has become very significant. We said before you got involved in the company, we said, look, we feel pretty good. We did a massive deal with a Japanese supplier to secure a lot of substrates. Now given the demand signal we see not enough, we went out and did another deal with AXTI. Still, that might not be enough. So we're always in search of substrate supply. But really, I would say that the headline is just time. I mean, this is a long cycle process, which makes it one that gives us a competitive moat, right? People don't want to get into this business for a reason.

That makes sense. Okay, so maybe let's shift towards the system business. Your system business grew 90% in FY '26 and your 1.6T modules just left the factory, that's great, with you crediting the signal integrity team because of the degree of difficulty plays to your strengths. I believe a few years ago that the market saw Lumentum as a components house, it's less of a module footprint. So what changed inside Cloud Light? And how far does the module ambition go before it collides with the module makers, but also customers.

Yes. Look, I mean the history of Cloud Light is rather a difficult one. When we acquired Cloud Light, they were doing roughly $75 million a quarter, and there were a lot of quality problems coming out of Cloud Light. Our largest customer, Google was complaining a lot about the quality coming out of the production line. And so we took a lot of steps back. Our revenue from Cloud Light dropped below $50 million largely because a quarterly run rate of $40 million to $50 million, largely because of these quality problems. So we sort of went in, we retooled our engineering. We got back to sort of a $100 million quarterly run rate by virtue of improving the quality, but we were still very late to market. We were coming in behind large competitors like InnoLight, like Coherent, like Eoptolink. We simply couldn't figure out the time to market. At the 1.6T transition, suddenly, we started to get things right. We put an intense degree of focus on our engineering and our turn time, retooled the entire process, retooled the way we thought about engineering these modules. And as you said, at 1.6T, you're getting into a world where signal integrity, for all the reasons we've been talking about with optics, it becomes really hard, right? And so our engineers have figured that out. We have a good signal integrity team. And suddenly, we got to the head of the curve at 1.6T. And we are now shipping ahead of great customers, great competitors like InnoLight and Coherent. We're first to market with a lot of our SKUs, and that's given us a great increase in our revenue. I mean we are now running well over $200 million a quarter in that business. It's doing fantastically well, and we expect it to grow considerably throughout 2027. Challenge for us, as you know, has always been the margin of that business as we've now fixed the engineering side of the equation. We've yet to fix the gross margin. It's still a challenge for us. We don't perform as well as the top competitors. And we've got to structurally fix that. So we were able to fix the engineering problem. What we're looking at as we enter calendar '27 is how do we fix the manufacturing? How do we go best in class or not -- I don't think we're ever going to get as good as InnoLight or Eopto, but I think we can get a lot better in terms of the margin.

There's always that debate about merchant versus module suppliers, the margins are slightly better always on the merchant side, if that makes sense. Okay. Maybe I'd like to take a question on the long-term technology risk, and this might be an interesting one for you. The eventual threat to the discrete laser is heterogeneous integration, and it comes in 2 flavors. InP bonded on SiPho, which already exist commercially. And lasers grown directly on silicon, which does not. You'll likely have first dibs in seeing this given your position in the market. So where are each of them really? Give us your bird's-eye view.

Yes. I'm glad, this shows your technical acumen. And I'm glad Kathy prepared me for this question. Thank goodness for that because it is quite a deep one. Look, this is an opportunity for us, actually. I don't see it as a threat. We're already working with the best silicon manufacturers to embed indium phosphide in their processes. And if we think if we can do that and work with them, we just have a much greater adoption of indium phosphide. So we see an opportunity, again, with the best in the world silicon processing houses to embed indium phosphide as a critical layer in their process. We see an opportunity, as you said, to -- which has been done before to embed the laser right on a silicon photonics pick. We actually think that's an enabler because some of the -- as really a component supplier first and a module supplier second, we see that change as kind of enabling a component inspector, making the modules a bit easier to make by putting the laser and solving some of the inherent alignment difficulties that you get right on top of the photonics I see. So as I say, if you're a module guy, it might be more challenging because it sort of takes moves towards more of a commoditization vector. But we see ourselves as components first. We actually think that this is something we want to enable rather than fight against.

I mean, yes, my next question would be basically exactly what you alluded to, which is if it would be to say -- it would be fair to say that even in a bonded world, someone still needs to grow and supply the [ MK ] material, right? And so the integration doesn't actually threaten that much the laser makers. If anything, it just changes where the laser lives and actually who packages it. So in a way, you still need a laser to be honest. So great. Perhaps I'll sneak one in on to the geopolitics and particularly on China. Now we know that there are some second order risks from the supply of raw materials to the Japanese substrate providers. So perhaps can you give us a bird's eye view on both the upside and downside risks around China, whether that is in the form of the supply chain or export control restrictions that could potentially cut the Chinese module makers exports, which would make the Western module and merchant makers left with potentially added demand?

Yes. Look, I mean, again, a super thoughtful question. And for us, I think it plays in multiple dimensions. We are viewed probably correctly so as a net beneficiary of a move by the United States government to ban Chinese transceivers. And I would view that as potentially a net positive. I think the challenge is that the Chinese transceiver suppliers make 70% of the transceivers in the U.S. market. It would be a problem to cut that off entirely if things aren't done in a thoughtful way. So I'm thinking now more on the macro, right, in the entire hyperscaler industry, if you suddenly said, there was no Chinese supply, that would be a challenge. That would certainly be a significant issue. Again, we think we can mobilize and respond to that challenge relatively quickly. Coherent, who is a great manufacturer of transceivers, I think, has probably mobilized and respond to that, but it will take us some time. And so in that Gulf, right, what happens? So I think however the U.S. government is thinking about that needs to be done in a thoughtful manner and work with the industry at large, including the hyperscalers to institute the right set of parameters around it. And then the concern, of course, is there some tit for tat. Right now, we've been able to really work well with the Chinese subcomponent suppliers to get us the supply we need, whether that be substrates, whether that be isolators, whether that be other raw material that we need on our manufacturing process. Is there any escalation that occurs that curtails to the first part of your question the ability for us to get Chinese supply? That would also be a bit of a struggle. Again, we have a nice attribute in that we have a global footprint. We have manufacturing, as you know, in Thailand. We actually have manufacturing in China itself, right, which can be an advantage when dealing with Chinese supply. We have laser manufacturing in the United Kingdom, which is generally a safe harbor relative to substrates and things of that nature. So we have a lot of things that work to our advantage. But depending on how this all plays out, Gianni, again, we're definitely viewed as a net beneficiary, and I'd see ourselves in the asymptote, that probably can work out just fine. But it's definitely something that I think needs a lot more thought than perhaps is being given at the moment.

Absolutely. We have 1 minute left so I'll squeeze one last one. It is 2031. Lumentum rolled out an incredible super cycle and built out more capacity for an evolving landscape. So paint us the picture. Does optical become the default fabric of computing the way copper was for the last 20 years? What does the Lumentum look like? And what is the one thing in this room should watch over the next 12 to 18 months to say that we're on track towards this?

Yes. Look, I mean, there's no doubt. You started with a question, you ended with a question. There's no doubt that optics is going to become prevalent form of connectivity in the data center, no doubt, right? And I think as we look at 2031, a huge adoption of NPO, huge adoption of CPO, huge adoption of OCS, all of which has high optical content. It's going to be a good time for anybody in the optics industry, and we're generally viewed as the leader. I think the thing that is going to be significant is the scale in, and that's the thing to watch over the next 12 to 18 months. Does this thing really take hold? Are you going to see optics not only penetrate, which I think is largely a foregone conclusion, the back plane. Are you going to see it actually go in tray and serve this high-bandwidth connectivity between memory and between GPUs, and I think we'll know more in 12 to 18 months around that particular opportunity.

Thank you very much, Michael. It's been a pleasure.

Gianni, thank you, as always. Appreciate it.