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

All right, everyone. Welcome back to the Barclays Global Tech Conference. I'm Tom O'Malley, semi and semi-cap analyst here. I'm very happy to have Wupen Yuen with Lumentum. Thank you for being here.

Thank you.

So it's been some crazy times of late. You've seen an unprecedented demand in AI kind of come into the market. You guys have talked about certain areas in your portfolio in which things have gotten tight. So maybe start with like a mark-to-market. Where are you seeing tightness? Is this across the portfolio? And then maybe how long do you think that dynamic lasts?

Yes. Actually, to be honest, I think the short summary is that we're seeing tightness across the board. Every single product line we have today is -- we are basically capacity constrained. Demand is exceeding supply. And of course, certain different areas have different levels of tightness, indium phosphide, lasers is really, really tight. And then -- but even the scale across optics, also very tight. So overall, we are seeing the demand to be -- especially on the backlog and commitment customer making is really through 2027. So that's what we're seeing, right? For example, our EML lasers is basically sold out to 2026 and largely booked through 2027, right? So whether capacity will increasing continuously, it's actually spoken for fairly quickly, right? So we see that trend lasting for the next 2 years.

And in this type of dynamic, how do conversations change with your customers? Because to get that much assurity on supply, obviously, there's an economic ask that you'll have for them as well. So how is this impacting pricing broadly? And then maybe talk about the time line, LTAs, contractual agreements. Are you seeing those that exist for 6 months, 1 year, 1.5 years? How are you seeing those?

Yes. I think the -- let's take laser, for example, right, that we are -- for the largest customers, it's typically a 2-year commitment. So we definitely -- the price is not going to be the same as before, right? So because what we're seeing here really is that the longer it is, you're not supposed to get a discount, given this kind of a sellers market. So we actually then raised the price, of course, relatively flat price over those 2 years, but definitely at an elevated level compared with before, right? But the larger customers typically are the ones that are willing to give the longest contract because they also see the demand, right, for the next 2 years.

So how does that work with the ability to only supply some customers? Is it the guys that are willing to sign up for a longer period of time, they get supply first? Is it the guys that are more strategic? Because when we were on the road a couple of months ago, you guys talked about like you guys are the ones that are going to allow larger customers to be able to have product or not, right? So how do you go about choosing who's getting supply?

Yes. I think one big thing here, we want to make sure that, one, we're not a king maker around our end customers, right? So we're trying to make sure that, hey, are we -- do we have a good visibility about who's selling to who, right, the end customers. So we want to make sure that we don't favor Amazon over Google, for example. We want to make sure that we can supply them well across the board, right? So that's one very important element. Of course, pricing matters. Of course, geopolitics matters, all this playing a role. But end of the day, we want to make sure that every single cloud customers are -- we actually support them visibly, right, that they can see that we're supporting them.

So why don't we go into some of the product verticals. So on EMLs, in particular, you've talked about a 40% capacity increase over the next couple of quarters. How is that progressing? You have a facility in Thailand. How is that coming along? Should we expect that to be pulled forward given how quickly the market is kind of growing?

Yes, I think on the laser front, right, so definitely, the 40%, we're actually well on track of that. And then it's not like one step jump, right? It's a quarter-by-quarter, we're increasing. And the reason for that is actually, we don't have like one piece of equipment, oh, this is a bottleneck. You move it, it's going to jump up. No. We have like incremental additions over time. So we're doing well in the ramp. Yield is holding up pretty well, which is anything from 3 to 4-inch, right? In the beginning, it was a little bit rocky, but now we have gone through it. So yield is holding up well. So overall, I think we're actually on track on the chip side, right? On the assembly side, we're definitely pulling in, right? The demand on the OCS, for example, the demand on the module, for example, all these are kind of accelerating. So we're definitely pulling in the capacity there, too, right, kind of accelerating beyond what we were thinking about before.

So it sounds like it's not a particular tool that's got you short. It's just -- it's multiple tools.

Multiple tools at different capacity level, right? Imagine you have like 20 steps, 20 machines and everyone has a different capacity level. As capacity requirement goes up, okay, then some is going to fall short, right? It's not like one over here and one over here is really short, it's actually really across the board, right? So kind of incrementally, you're tracking all the equipment and to ensure that we get the capacity we need.

Okay. In the same vein, you've kind of talked about allocating for either EML or SiPho modules. I think that, that's been an important kind of distinction as well. Maybe talk about why you would choose to put capacity towards one versus the other. And then what's your current view on your directing of capacity?

Right. So definitely one most important factor there is really kind of gross margin per wafer, right? So if you look at that, the EML is better than CW laser for silicon photonics. And then 200G EML is better than 100G EML, right? So that kind of in large is our kind of our -- how we're thinking about allocating the capacity. However, right, even though we are sold out for EMLs, we still want to make sure that we are seeding the CW laser opportunities. Because silicon photonics is important. It is going to take more share on the longer term. And we have a really, really good CW laser, right? So I want to make sure that we are in the game as well. So we basically say, hey, majority of the capacity is out there for EMLs. And then -- but we're also still providing some capacity to CW lasers to the customers so we're in. And also, we're providing some to ourselves for our own modules starting from middle of next year, right, to ensure that we can bridge the gap of our own laser supply to for our own modules.

How fungible is that capacity? Let's just say customer demand profile changes or you internally decide that you want to split that dynamic a little bit differently. If you wanted to take capacity going towards EML and move that to SiPho, what's the time frame that you could get that done in?

It's actually just the lead time, right? Today, both the CW laser and EML lasers are made in our Japan fab, right? So really, it's just the lead time of changing the wafer start. 200G, 100G stable lasers is actually fungible capacity in our Japan fab.

So let's just say, if we move towards 1.6T, I'd love to hear your opinion on. Do you think that the world moves more towards silicon photonics versus EML at 1.6T? And do you feel like because your existing footprint is so fungible, like you could move your capacity very quickly to address that? So yes, just general thoughts on technology for 1.6T and...

That's a good question, right? I certainly, number one, the fungibility is there, right? That's why we want to make sure we're designing to every customers, right, with silicon photonics laser too. So if their demand changes, we can always change, right? That's number one. Number two is actually then we do see photonics actually -- they are kind of got designed into the transceivers slightly later than EMLs, but not by too much, right? So we definitely see SiPho will take -- the market share increase will be faster than the 100G generation, right? But on the other hand today, though, given the supply constraint situation, it's not about technology choice anymore. It's about which one you can get lasers for, right? Can you get EMLs or can you get CW lasers? So our largest customers all have 2 solutions. And depending on the laser availability, they will go with that solution, right? And keep in mind today, the silicon photonics supply chain is also constrained, right? So now you have constrained silicon photonics, you have constrained on lasers, constrained on EMLs. So every single customer of ours, you're trying to figure out is how can I supply to the demand based on the constrained supply chain that I have.

In prior generations, you've also seen VCSELs take share as time has gone along, generally lower cost option. Is that a reality at 800G? Is the performance good enough to get there? Do you think people move in that direction? And then same question for 1.6T. You kind of talked about SiPho in that way, right, where it takes share over time. Is that the same way you think that VCSELs will be at 800G?

Yes, 800G definitely -- VCSEL has always been there, right? When the hopper was ramping in the first -- in the beginning, it was a lot of VCSEL-based modules, right? And then it got shifted over to silicon photonics and to EMLs because they didn't have the capacity. Now capacity is back. So we see 800G kind of coming back, the multimode kind of coming back a little bit, not to the previous level, right? Now for 1.6T, despite some of the vendors pushing for 200G VCSELs, we're not seeing it being adopted broadly, right? People have already made the mind up and say we're going to shift over to single-mode based architectures, actually infrastructures. And then we have not seen the effect of coming back. So we're seeing the constraint of the lasers and silicon photonics and things like that. But we're not seeing people say, hey, we're going to use 200G VCSELs as a replacement. We're not seeing that.

Got it. less of a factor.

Less of a factor. Because it's very limited, right? When we do like 30 meters, it's just too limited, right, for the -- especially nowadays, all the racks, right, are really high power. And then the rack spacing becomes wider and wider, you really need about a couple of hundred meters to cover all the scale of networking. So 30 meters is too short to cover the use cases.

Got it. Maybe switching gears to another vertical, and I do want to come back to kind of RO scale architecture and solutions to that. But why don't we start with just OCS. So I think on the last earnings call, you guys talked about OCS going from kind of single-digit millions to something around $100 million per quarter by December of '26. If you listen throughout the quarter, you guys have kind of been incrementally more positive there. Could you talk to why you're more positive? Are you seeing more customer engagement? Is it a broadening of customers or kind of a deepening of your existing customers? And maybe just from a technical perspective, I get a question a lot of times, like is this replacing packet switching? Is this existing alongside packet switching? Maybe spend a little time explaining where this fits in.

Yes. I think, first of all, I would say that we're getting deeper with the customers. And then today is a primary use case, there are actually 2 primary use cases. One is the scale up, right? It's a scale-up architecture that really drives the volume, right? The second one really is, I would call it, a multi building campus scale across, right? So you have like, say, 3 or 4 different campus buildings, big data centers, 100,000 GPUs each. You want to connect them at the campus level. These are 2 major applications, right? We also see the scale up, which might be a replacement of the spine switch, right? We also opportunity for scale up. That will also be a replacement for the kind of call the NVLink switches, but those are kind of further out, right? So today, the dominant one really is a kind of scale across level application and then the scale-up portion, right? So these are the big ones. And we see this scale-up portion is actually strengthening. Our position is strengthening, the demand is strengthening. And this customer also starts to sell their solutions outside, right? That's also an upside. So all these add together, we're seeing -- that's what give us increasingly more confidence, right? It's because we're seeing our progress. We're seeing backlog visibility, and we're seeing our customers' demand because of their own success, right? That's all factored together and we say, this is looking better and better.

Outside of the customer who kind of evangelize this technology, is there a lot of greenfield expansion where you're seeing new data centers get deployed and customers are saying, I'm going to deploy optical switching along with packet switching in tandem. Are you seeing OCS kind of eradicate some spend that you would need on traditional packet switching? I guess that's the question that people ask a lot times, like should I be worried about players in the traditional packet switching environment because OCS is coming on? Like how would you...

How do I think about it? I think right now, I would say the following, right? So right now, it is really an architectural decision. Okay. I'll answer the first one first. We definitely are seeing much more interest in OCS. People, especially from the AI model companies, they definitely are interested in exploring and even in some cases demanding the use of optical switches, okay? But it's actually right now, I see it's building on top of the packet switches. They're using it because architecturally, it is a better choice, right? There's all the benefit about cost and then power consumption, all that stuff like that. That will play out over time. In the near term, I don't see it kind of cannibalizing the packet switches. It's going to be supplementing the packet switches. Over time, when it goes to more into the spine or even into scale-up architectures, I think it will kind of eat into the share of the packet switches, right? But the market is going to grow really, really fast. So you're going to see packet switches still grow, but then an increasing amount of the share is going to be allocated over to OCS. That's how I see it.

And so will that business as well, OCS be limited by supply like your other business? You kind of started saying, we see constraints across our entire portfolio. Like you are starting from a very low number, getting to a moderately sized number there. That is in that same camp, right, where you are limited by constraints?

Yes. Right now, because the OCS supply chain is still pretty new, right? So the whole needs to be primed to support the volume, right? So right now, it's not so much of a constraint because of the capacity -- fundamental capacity per se, is just it's new, right? So it takes time for the supplier to adjust to it, right? We're not seeing like supply constraint fundamentally like we see on the laser side for the next couple of years, but depending on how fast the OCS will grow. And we already started to think about how we're going to plan for that to enable the further ramp up of OCS.

Got it. Switching gears to CPO. It sounds like late '26 is kind of the time frame where you'll see scale out technology at first, which is revenue towards you guys there. When you look at like volume of switches that are moving to CPO, do you expect a significant volume transition in late '26, early '27? Or is that something closer towards the end of the decade?

That's a fantastic question. I think our customers don't know, right? So they're asking us really to say, hey, do you have fungibility between the CPO laser and the regular lasers, right? The answer is yes, with some fungibility. But we do see that I think -- let me just give you some rough ideas. I think by '27, by '28-ish, we should see, let's say, 40%, 50% of the switches of the first user of this CPO to be a CPO based, right? They have the incentive to really drive the adoption of CPO, right, and then selling the whole solution. We'll see at that vendor in 2028 time frame, [indiscernible], you will be like 50-50 between a pluggable based versus a kind of CPO based. That's our best view, right? And then themselves actually have a large uncertainty. They are trying to gauge from their customers, too, right? But meanwhile, I think it's fair to say we see the CPO activities picking up at every single hyperscalers, every single one of them. And not just for scale out, but also for scale up, right? So we definitely think that 2026 ramp is happening and '27 will continue to ramp. And we're going to see the share of CPO increasing from '27 onwards to '28, probably reaching 40%, 50% in '28. And then around that time frame, probably scale up will start to show up as well. right? That's how we're picturing the CPO demand.

Yes, 40% to 50% is a pretty large number going into that '28 time frame. I guess if you're already at that adoption rate with that large first customer, what is the barrier to moving to scale-up architectures? Because I think the idea that we've heard as an industry is scale-up is where most of the benefit comes. It's just it will take a little bit longer. What's gating you from getting into scale-up architectures and systems?

That's a fantastic question. I think from a system point of view, there's still a power consumption, cost considerations from the application perspective, right? But also from -- if you look at it from our biased angle, capacity, right? Because, again, the scale up is going to be several x of the demand, right? So we're going to scale -- we're already ramping very aggressively. Now we're going to actually tilt that slope by another several x, right? So we're already planning for that and thinking about it, how we're going to actually address -- actually enable, right, what to enable the scale-up opportunities, right? Again, there's multiple conversations, not just one customer, right? So how we're going to do that as an industry. So that's what we're actually already starting to work on.

Got you. All right. Broadly into 2026, you're getting, I would say, more substantial 1.6T volumes. 800G, I think you said before, is still supposed to be the largest node in the market for next year. Could you give us some view on what you think total port count looks like next year or just optical ports in general for 2026?

Yes. I think optical ports, our estimation is probably about 60 million ports, 60 million, 70 million ports in that range. Actually, the number kind of doubling, almost nearly doubling year-over-year, right? This year, probably 40 million, 50 million, next year, not doubling, maybe 50% increase. Next year, probably 70 million, 75 million ports. And we're getting -- if you look at it from the GPU count, you can do some attach rate calculation and you can look at the TPU multiplier, right? And then what we'll hear from the hyperscaler directly is probably around 75 million ports, around which I would say probably 55-ish or 60-ish is like 100G -- or 800G and 15 to 20 is 1.6T.

So that's largely the hyperscalers.

Hyperscalers including whether it's NVDL, whoever it doesn't matter overall the market, right? That's what we're looking at.

Got it. And then your customer exposure within that vertical, like Cloud Light was well understood to be kind of a Google provider very early on. In terms of like the diversification and your scale, like do you still feel like you have a very good position at that customer versus where you were...

Yes, we are. So I think we originally -- our strategy a year ago was, let's go proliferate and win more share. We're like, well, actually, we don't need to do that. Because every single customer is so demanding, lead customer is particularly demanding. So we want to make sure that we focus on serving them well. And then just -- we have 3 customers. just the 3 customers can keep us really busy, right? So we want to make sure we're serving them really well. And then, yes, while position is still very strong, right, in those lead customers. And we have road map visibilities multiyear into the future, and that's what we're focused on.

When you look at those 3 big customers, longer term, looking to get to kind of $1 billion business or so, what are the biggest challenges in getting from where you are today? Is it manufacturing? Is it yields? Like where is the hang up? Because I would assume that today, if those customers could get more modules from you, they would happily do so. Where is the hang up?

It's really just capacity. It's really just capacity. I think every single customer, frankly, today, if I were to win more customers, I just go over capacity. You'll get it, you win it, right? Especially if they -- here you have your own laser supplies, "Oh my God, please come in," right? So that's the dynamics, right? So really for us, it's like, okay, let's be careful with this, right? Let's make sure we can produce and deliver the capacity, right? That's really the only gating item.

Because historically, I remember at the 400G transition and then early 800G a hyperscaler would qualify 2, 3, 4 modules, and then there would really be very little requalification process, right? What was set, was set from that point forward, you got [indiscernible]. Now if you have supply and you go rebid that project, are they like welcome in? Has that dynamic changed?

Yes. I think a little bit, right? So basically, now the customers are still favoring the existing supply chain because they have a lot of trust. They know these guys can ramp, these guys have good reliability, quality, all that trust takes time. Especially now opportunity cost now for them becomes really high, right? Everything has to ramp instantaneously. They have no time to qualify new vendors. However, we have been getting a lot of invitations. They say, "Hey, are you interested in bidding for this opportunity?" We're like, why? Because we're short. We heard you have your lasers. right? So that's become a common theme, right? But we're like, okay, let's be careful, right? Let's make sure we hold on to our current customers really well because the capacity is short in the industry, right? Even the DSPs could be short because TSMC is moving capacity from 5 nanometers to 3 nanometers, and they're struggling to supply the world's GPU, TPU, XPU demand, right? So everything is going to be short. And therefore, even if you want to invest in the capacity, you may not be able to get the parts, even semiconductors, right? So that's the challenge I think industry is facing. So -- but opportunity is there. If we want to grab it, it's there.

Helpful. So in the last quarter, you look at all your large new opportunities in the data center, you have switching, you've got your module business, you've got your laser business. It was surprising where you saw a lot of your upside was actually in the telco business, where you saw like more broad telco strength. Could you maybe point out where you're seeing that strength? Is this just the ZR, ZR+ market actually starting to come to fruition? Or is it a certain subsector of component that's going in there?

It's actually broad-based, right? So the term scale across was invented, right? That really was the case, right? So you look at our portfolio, all of our other products, whether it's wafers management or WSS or pump lasers or tunable lasers or modulating products across the board, every single product line is increasing, right? So basically, what we're seeing here really is that you have a supercharged growth of data center optics, and we're talking about already here. And then all that -- some portion of that is leaking out means we're leaking out to the wider area networks, right? So really across the board, everything is actually growing, right? So if you look at our kind of the forward-looking view, there's kind of this baseline not at 2x, 3x per year growth, but at 50% year-over-year growth, that kind of thing. And then on top of that, then you have this supercharged data center growth, right? That's how we're looking at it. So still, we're seeing that trend continuing.

Well, things sound fantastic across the board. Thank you very much for joining me.

Well, thank you. Thank you for the opportunity. That's great. Thank you so much.

Thank you.