Lumentum Holdings Inc. (LITE) September 9, 2026 Conference Transcript

Thank you, everyone, for joining us today. And welcome, again, to day two of IT Tech Conference. My name is Papa Silla. I'm the US Optical Networking Analyst here at Citi. And I'm very pleased to welcome Wu Pan Yuan, who is the president of Global Business Units, and prior to that, president of Cloud and Networking. And of course, we have Kati, who you know very well, the VP of IR of Lumentum as well. First of all, I'm very glad to still have you now. I wouldn't be surprised if, Wupen, you cancel last minute to secure more capacity at this point. Well, this is important. This is really important. Thank you for joining us. And maybe to get started, and I will ask questions, and then we'll leave the last maybe 10 minutes for your questions as well, if you have any. To get started, Wupen and Katia, I guess, given your role into AI, I'm curious to first have your take around, But also optical and networking is playing a more and more important role into this AI infrastructure. So just if you can touch on that as well before we touch on the company in general.

Yeah, let me jump in here. I think the AI really is where at the very beginning of AI, right? If you look at the whole industry here, yesterday I was selling this article talking about, you know, Anthropik in a year signed up $517 billion worth of compute, right? So today, the demand of AI way exceeds that of supply. And we're not seeing it changing. It's very different from the past cycle of the optical, which is really for infrastructure, for communications. This is really for the intelligence, the generation, and also serving the intelligence. So the demand is just absolutely crazy. And then we don't see that changing for the next at least couple years, if not longer. And as a result, right, As AI making progress into faster throughput, bigger models, more sophistication, the agentic AIs, the connectivity requirements are also increasing. Every generation of GPUs or TPUs is getting more and more bandwidth. And then with copper running into the limitations of physics, then more of the optical has to be used for interconnecting these XPUs. And therefore, we're seeing the demand for optical connectivity is also increasing as generations goes on. Therefore, we feel both the demand side and also the technical requirement continue to escalate. And all this actually is good for the optical world we're in today.

Papa, if I could just add on to that. Thank you again for hosting us at the conference. It's been just a fantastic conference for us. We really appreciate that. I'll just add on to what Wupen just said. If we think of data center interconnect or the telecom market being sort of a 1x multiple, what we're seeing now for scale across is probably 10x of what we saw for DCI. What we're seeing now for scale out is probably 10x of what we're seeing for DCI. What we're seeing for scale up starting in the second half of 27 for the first instance of CPO going to scale up is probably a three to four X above the 10 X that we're seeing for scale out. And then when all of scale up goes optical, it's probably a full 10x above the 10x of scale out. So there's many multiples of optical bandwidth that we are needing to deploy in data centers. And we're not yet even talking about scale in. I think you may have a question for us later on scale in, there's another 10x multiple there. So it's mind-boggling in terms of what we're enabling, in terms of enabling the compute.

Absolutely. It's been quite a fascinating journey. And I'm curious on, I guess it's been not too long since last earnings, but just curious on the state of supply-demand gap at this point. Where are we? Is it extending? Any general kind of update on that front?

Yeah, I think it's still extending, right? So today, we're still short everywhere, right? The laser, for example, we're short. EML is short. Seattle laser is short. And then we don't see the shortage going away in 27, right? I think 27 is going to be another major year. People forecast just the scale-out modules going to be doubling the volume on 26 to 27. Laser supply is not really catching up to it, right? And in 28, the scale-up is going to happen. that's going to drive yet another 3 to 4x of increase of demand. So we really don't see the laser supply demand gap actually close within the next couple of years. And then we're doing the very best we can. To this day, we're still allocating capacities, in some cases, based on pricing, expertise, and things like that, in order to make sure that we serve to the most profitable opportunities. That's the case today. The pricing remains very healthy. All the deals we're making today are equal or higher pricing than the escalator pricing that we already had before. So overall, we've seen the pricing environment continue to be very favorable, and that is kind of the signal that the supply-demand gap still remains.

Maybe before diving deep into the different kind of segments, Cathy, maybe one question Recently, you updated your 2028 target to $40. I think it probably came earlier than most expected because you typically do that, I think, at OFC. Just curious on the timing.

Yeah, sure. Again, thanks for the setup there. So since our last earnings call, we learned that we have more opportunity in OCS than we previously had understood. That visibility in OCS enables us to be very confident about, for the first time, I think ever, putting out a fiscal year 28 guidance on EPS of $40. So the upside that we're seeing from OCS enables us to get there. And that OCS upside is coming from our largest customer. And it's a function of more units, more units that we need to ship to them. And it probably is in large part due to their success in the market and in our success with them in enabling their success in the market as they begin to sell hardware. Yeah, I will probably circle back to OCS, but just a quick follow-up, I guess, is that more back loaded, kind of more 27, 28, or is that kind of every single quarter you are seeing more demand for us i think it does layer in as we move through the the fiscal quarters of 28 so this is a 40 eps in fiscal 28 and it kind of pertains to all of the quarters of fiscal 28. you should probably think of it as layering in linearly and increasing as we move through calendar or fiscal 28.

Got it. Very helpful. But maybe perhaps I can start with the CPO, NPO part, and then we'll get back to OCS. I think during last earnings, the discussion was around CPO, particularly for your main customer. Demand is coming even faster and stronger than expected. I was hoping you can pull the curtain a little bit and discuss it in more unit than expected. Is it kind of the winds are coming faster or is the customer pulling forward this timeline? Just any color on what's driving that sentiment.

Yeah, I think that's a great question, right? There's a lot of noise about the CPO, NPO story. First of all, on the CPO piece, right? What we're seeing here is the schedule is the same. Schedule is not changing, but demand is higher, right? It appears that there is more demand for the CPO. First, it's going to be in a switch, right? The scale-out switch. later in the second half, 27, as we talk about, it will be for the optical scale up, right? We're seeing both actually going up in volume. No change in timing, but intensity is actually increasing, which is a very good sign that CPO is doing well, right? Getting good tractions in the market, and technology is continuing to ramp well.

Yeah, no, that's very helpful. And it seems like recently NPO is even more popular than CPO, which is very hard to do, right? But I'm curious, I guess, if you take 27, 28, 29, is there a year in which you can see NPO potentially be bigger than CPO?

Okay, let me take the minute to describe the story. So let me sit straight and describe this, right? So if you look at the, first of all, why is NPO even there in the first place, right? You know, at the OFC time, when we talk about the CPO, we thought the CPO, okay, was going to be a one-customer-only story, right? And that's going to be one-customer scale-up story. Ever since then, what I found out here really is that the optical scale-up is now industry-wide story. And to support the level of volume and demand for optical scale-up in the time frame in 2028, the industry is needing to use both CPO and MPO approaches to serve optical scale-up, okay? So if you look at the overall market, we think it looks like the following. CPO will be half of the market. MPO is the other half of the market, right? So I would say just from the volume point of view, it's kind of 50-50. Now, look at what we serve. We serve in the external light source, high-power lasers, right? We see the CPO world being an exclusive external light source market. And then for the MPO market, half of that will also go with external light source, okay? And the remaining half, the customer are discussing between IOS or integrated light source versus external light source, but there's more and more of movement toward external light source. The reason is very simple. Because we have such a high optical density at, you know, with optical engine being placed so close to the heat source, which is the accelerators, there's a great concern what happens to the laser when it's so close to the heat source, right? And therefore, to ensure reliability being strong, there's a strong desire to move the laser outside and so that, you know, you don't have to worry about the laser being a reliability-limiting factor, right? So we're seeing that, and we discussed also in the earnings call, this is the reason why we believe that the NPO is a net addition to our opportunity, because when we're thinking about the CPO before, We didn't think about the NPO being a mechanism for the industry to jump on the bandwagon of optical scale-up. I hope that explains the situation.

No, that's a very, very helpful color. And kind of just to double-click on NPO specifically, I think you discussed before, there are two versions. One in which the laser is inside, more of a 100 milliwatt kind of laser is needed or so. And then outside, more of a 400. I'm curious kind of, and then you mentioned it seems like most will be outside, but is that like 75%? And the reason I'm asking is I think there's a lot of discussion around China competition, them being able to do the 100 kind of version of it. Just curious if you can kind of paint.

That's a great question, right? So let's go to the 25% overall market of the ILS piece, right? Even for the ILS, you really don't want to use low power lasers. Why? Because imagine, you know, if you look at a NPO engine, it's a size of like this big, one inch by half an inch, right? It will pack 6.4T inside. Compared with a 1.6T module today, it's half of your iPhone size, right? So optical density is a lot higher inside this little package, and therefore you really don't want to use too many lasers inside. You want to use as few lasers as possible so it doesn't fail, right? So therefore, there's a very strong push to also use high-power lasers. It's not going to be 100 milliwatt laser. It's going to be 150 milliwatt lasers or 200 milliwatt lasers, right? And so therefore, I would categorically say that for the entire NPO-CPO applications, it's going to be 120, 150 milliwatt and above of high-power lasers, which is not the traditional 70 milliwatt laser which is used in today's pluggable market, right? So a lot of the chatter about the CW laser supply from China in particular, is really targeted today to serve the pluggable market, which we talked about earlier But for the NPO-CPO, what we see really is a strong push toward less lasers, more power, and then to feed both the EOS or iOS applications.

And even in the US, I assume practically for the 400 ones, you don't see many competition at all.

Yeah, we see one competitor, right? you know, who's coherent that NVIDIA also signed a LTA with, right? And we're really not seeing anybody else in this high-power laser area.

Yeah, no, that's very helpful. Now that we, I guess, for the CPU and NPO, you set a very clear picture on the demand front. I guess now the question is supply. And one important kind of aspect of that is the Greensboro, which is supposed to come in late 2028. I guess any potential kind of drivers for that Greensboro fab to come in faster? Any color on the timeline? What would pull it forward? That's a great question.

I think it may give you a holistic picture on how fast, how we're thinking about ramping our ultra-high-power laser for CPL and MPO, right? So first of all, we are ramping today in our San Jose fab, right? We have a line of sight to ramp from today all the way through 2027, so that's the first ramp. Second ramp is we're ramping this ultra-high power list also in our UK fab. That will come in in second quarter, 2027. We'll continue to ramp through 2028. That's the second ramp. The third ramp is our Greensboro fab. It will start to ramp in second quarter, 2028, or mid-2028. That's the third ramp, right? So in composite, you can imagine there is a three-wave of a ramp that continue to support the UHP laser for both the MPO and CPO applications, right? What can change that? Really, I think it's so far we're on schedule for all these different ramps. And, you know, what can really change the schedule is the qualification timing, right? We have to submit a PCM for customer to qualify the product. If we actually accelerate that qualification of the product, we can actually put it in and vice versa. So that would be the biggest variable, but we're definitely working really hard to ensure We'll get all the quality, all the data ready, so that the PCN approval can be a pretty straightforward process.

And I'm curious on particularly the fab that exists already. I'm sure you are using it to some capacity or so for, let's say, telecom and so on. So I guess it must be a very tough decision because demand is strong everywhere. So I'm curious on how you think about allocation. Are you kind of giving priority where margin is strongest? are you also looking into the future on let's say CPU or NPO could be a bigger time than the other?

That's a great question. So we are, in principle, we're adding capacity. We're not taking away, right? For example, you know, when we're ramping our UHPV laser in our St. Jose file, we're adding capacity. So we can actually keep our current product, which is also ramping. And also in the UK, we're adding a clean room. We're putting the new equipment, the latest and greatest state-of-the-art equipment into that clean room and ramp the USB-A laser that way so we don't sacrifice our current products. So that's how we're thinking about it, really, is to ensure, because all these products are ramping at the same time, the telecom scale across is also ramping, as, you know, Cassie just talked about, and therefore we don't want to sacrifice those, right? We're adding new capacity, you know, six-inch capable capacity to ensure we have the highest efficiency possible when we start to ramp these products, right? So by and large, we don't sacrifice, we add.

And then beyond laser as well, now you are also doing ELS, or you plan to do more ELS as well. I guess, how large do you see that opportunity relative to the laser longer term, and any color on, is it one customer mainly currently, or is it pretty diverse? Do you want to take that question?

Yeah, I mean, we announced our first ELS win on our last earnings call, and that was with our existing CPO customer, we know that that's just a relatively small portion of their ELS total demand. We expect the timing of those first ELS shipments to take place in the second half of calendar 27. And our estimate is maybe 15 to 20% of their fleet is about the right amount of their share that we think that we will win. Looking forward to the NPO opportunity, I think we can be even more excited about our ELS capacity to take share because it looks like, as Wupen just described, the majority of the NPO opportunity seems to be on the ELS side of the equation rather than the ILS part of the equation. And many of those ELS customers would also like us to be an ELS supplier, not just a laser chip supplier. So I think we have both of those product opportunities to look forward to. We haven't yet announced any NPO wins, But I think that we can expect to see something in the next handful of months, at least in terms of a first customer that we may be looking at.

That's very helpful. Maybe jumping to the transceivers kind of slash EML, EW part. Just curious there, it seems like there as well demand is pretty strong. I guess, are you seeing it more for your 100 gig, is it more 200 gig EML, or is it kind of CW, just any color on what you're seeing on that front?

On the EML side?

On the, yeah, I guess the lasers for the transfer. The lasers for the transfer.

Yeah, I think today, definitely 100G EML remains to be the workhorse. 1.6 T or 200 G per lane is ramping, right? So I think we talked about we exceed 25% of the mix, right? We think it will continue to go up. So that's going to be predominantly our output. We talked about before we also are allocating strategically a portion of our capacity, minority of our capacity for CW lasers, right? Really for three purposes. One is that when our end customers are short and asking for some help on CW lasers, We try to help them to ensure that they can kind of bridge over with the deployment. Two is that we want to make sure that we have CBLAs also seated at every opportunity that we can to ensure that when we have more capacity in the future, we have a design wing, design slot we can run to. The third piece really is that we're also supporting internal modules because our modules are basically photonics, and the customers really need us to make sure we deliver our modules. We're also allocating a little bit of our laser to our own module need, right? But overall, I would say predominantly our allocation today is still EML. Workhorse is 100G, but increasing amount of 200G. And then our own several lasers use the three different purposes and the minority allocation of share.

And then for both, I guess, 100 gig and 200 gig, similar question to earlier, the lasers on the CPO, NPO. So what are you seeing in terms of competition, both U.S. and out of China as well? Are you seeing much competition on 100 gig? Because, again, there are a lot of noise.

So, frankly, we are not seeing merchant competition on EMLs, other than who we already know, right? The Japanese and some of the U.S. company. But we're not seeing new entrants of EML coming from China. There is one vendor who's producing EML internally for their internal module use, but their noises were heard in terms of putting EML into the marketplace, but we're not really seeing it on their radar screen in terms of being a competitor that we're fighting share with. To this day, our EML pricing remains very favorable. And, you know, if anything, we're really not seeing the competition.

Got it. That's very helpful. Maybe last question around the transceiver, such kind of lasers going into the transceiver. It seems like for 1.60, the idea is silicon photonics might have a bigger percentage of the mix. But I think there's a lot of debate on by the time we get to 3.2 T, EML will come back. Can you just discuss, is that, I guess, your point of view, and two, can you kind of discuss why would that be there?

Yeah, we got the question a lot today on this 400G per lane kind of technology, right? So first of all, silica photonics, while it's really good for integration, it has a material property limitation on the speed. So 400G is going to be really a stretch for silica photonics. And can you do it? Yeah, probably can, but you have to pay a lot of price on power and power consumption. our output and power consumption so um and india fastify uh for 400g is we have already demonstrated it we know it's rampable uh so it's a short bet for four energy uh per lane so the industry is still trying to figure out one technology people are thinking about really is not silicon but uh simply listen nowbe right but as a new technology uh you know unclear whether the supply side will will be able to be there when the volume goes to tens of millions. So I would say, if you look at today, for 3.2 t, India phosphide is a short bet. Sympatholim will be probably pretty challenged. And Tiflin or Sympatholim is promising, but the supply is a little bit challenge or a little bit questionable at this time.

So really, depending on 3.2 t timing, right but we because of this we think that india fasa has a good chance of regaining uh the share that you had at one point at the 100g per lane generation maybe switching to to ocs and we'll open it for question if any um and then kathy i know i bother you all the time on this question but still be curious on any way i think for the transverse most of investors and sell side are now able to do some type of bottoms up and so on. But for OCS, I think everybody's still trying to get there. Any sense of, I guess, how should we think about kind of content per accelerator or kind of content per TPU? Just any comment.

On a first order basis, I think you can still think of like a 1.5 port to one XPU attach rate for the most part for the major applications that most people are thinking about um there are nuances to the attach rate as we think about the new topologies coming down the pipeline um there are some puts and takes but in general i would say wupen that it still ends up roughly 1.5 to 1 but by the time you get through all of the the puts and takes on back end of network fewer hops and some parts of the network it nets out to be about the same attach rate.

Got it that's very helpful maybe I can pause here for now and open it up for question if you have a question just raise your hand.

I'm trying to get a better understanding like is the edge of your product mode hop free window or is it peak power and then is that kind of driving the better yields? So if you could just comment on that.

Loaded question. I think from the performance point of view, okay, let's start with what customers care about. The customer cares about power, obviously, that's a big one, but they care about the noise as well, right? The noise of the laser, we call the ring noise. It has to be really low, right? And the third piece is line width, right? Because the customer uses a ring modulator. Any kind of line width excursion will cause it to be translated into intensity noise, which will then screw up the link, right? The fourth one is what we talk about here is mode hop, right? Because we look at today's signal photonics, the biggest problem with signal photonics is the laser will mode hop. The mode will jump. Once you jump, you have a link fader, right? So all these four things have to be achieved simultaneously in the laser, at volume, with good yield, and it doesn't change as you go through the packaging processes. That's a pretty tall order. I think the jury is still out, right? We have already started ramping these lasers now. We're seeing real data. We're doing pretty well. But I think for anybody who's new to this market that has such a high power, such a different laser designs, to meet all these criterias at the same time, it's not going to be easy, right? I think Coherent is also working really hard on this. I think they will overcome it at some point. But, you know, in terms of overall competition, it's not an easy laser to make. In addition, it's pumping multiple lanes, right? So you don't want to have the laser fail at all, in any way at all. So, yeah, so I think that's a differentiated product, very challenging to make.

What does that timeline kind of look like for when they catch up? Because as you said, their laser isn't necessarily stable versus yours at set power. So is that like 12 months? Is it 18 months?

We frankly don't know, right? I think from a demand point of view, I think the customer will need them to be ramping up sometime in 2027, right? We don't know if they're going to be there or not. I'm sure they're working hard on it, but I think to keep this market going, we need a second source as an industry, from a customer point of view, to have this capacity ready in 2027. Probably second half, I would guess. But I think that will be needed. The reason why I'm saying that, because we need to support the scale up in 2028, right? And there's going to be a big need of lasers, right? So I think that's a time frame that if I was a customer, I would push Coherent to have the product ready for that ramp.

Yes, I'll get us kind of go from here. So I guess on the OCS, I think often we think about, I guess, the time. I think there was an $8 billion number. I guess, in any way, probably upsiding that number, but just curious on how should we think about the time and what do you think Lumentum could potentially capture of that time?

Yeah, and I think as, I mean, that 8 billion number isn't that old. It was something that we put on a slide just six months ago, but now I think we see upsides to that. Not only is our largest customer doing extremely well in the market, which is one source of upside but there's also this sort of prey level in rack different way to think about OCS from a dynamic partitioning perspective or an XPU player they may want to not only have you know 576 XPUs in a cluster they may want to have different numbers of XPUs in a cluster to dynamically partition that. You could use OCS. You could also use it as a resiliency network application where you route around underperforming accelerators. We now see those applications kind of coming into very serious R&D, and we are working with customers on that, And I think that gives us better visibility to, at least by late 28, early 29 type time frame, we can see the advent of a lot more OCS that will be required for those types of applications. So that probably puts upside to that $8 billion number.

Yeah, that's very helpful. And just on the $40 EPS for fiscal 28, to my understanding, it was just the EPS number. Is there any way we can kind of reverse engineer to get to the EPS? What's the implied margin? Any color on how we should think about the other lines?

Yeah, I'm hesitating to say that because we did see the sell-side models out there, yours included, by the way, Papa, that had us a little bit lower on EPS for the fiscal 28 timeframe. And we felt like, okay, well, we have all the confidence that we'll be able to meet this $40 metric by that time frame, based on the goodness of the additional orders and business that we're seeing in OCS.

Got it. Very helpful. Maybe going back to you, Wupen, on CW, kind of something that was really interesting that you mentioned, I guess you mentioned during the call was, Michael, sorry, mentioned during the call was CW margins or kind of yields improving a lot. I'm just curious on, I guess, how far can we think of kind of it getting, can it get to EML level, or kind of just in any color, or should we think about that aspect?

Yeah, I think that's a good question. I think our serial laser yield has climbed up to be higher than EMLs. And the pricing of our serial laser, believe it or not, is similar to EMLs. And then so it's a little bigger than the EMLs, but as we discussed in our earnings call, we have also shrunk the size of the CW laser to be a lot closer to that EMLs. So I would say the margin profile is not going to be as good as EMLs, but pretty close to CW lasers. And therefore, our past kind of hesitance in putting more CW lasers out there is really gradually diminishing given that the progress of MAID on the maturity of the CWL technology from the yield and the die shrink perspective. So it's still not as good as EMLs, but it's coming close.

That's very helpful. And maybe to finish it off, last question for maybe both of you. Kind of obvious, Illumintum is the name a lot of investors look at, but I'm sure from your point of view, they are part of the story that you believe maybe might be underappreciated or misunderstood. So I want to give both of you the chance to maybe part of a technology or part of the story that you perhaps want to double click on.

The first thing that comes to my mind is scale across. And this is partly our fault. We don't give it enough airtime. But we also see scale across really inflecting our pump laser business. We talked about in this last earnings call and the call before that, that we see a 4x multiplier on our volume of pump lasers that we're expecting over the next five quarters or so. we have signed up all of the major network equipment manufacturers in long-term agreements with us over multi-year period and we engage them with higher prices for pump lasers to help us underwrite the capex that we need to spend to expand that capacity so we really see scale and pump lasers, as well as narrow-line-width lasers. That really has been a story of growth. I think this last quarter we had the 10th consecutive quarter in a row of growth in that business. So it's really been a very strong foundational story for us.

Yeah, for me, I would just try to maybe emphasize our position, right? Cassie talked about scale across is 1, scale out is 10 in bandwidth, scale up is 100, scale in is 1,000. You know, Lumenton's position today is going all the way from scale across to scale out, to scale up, and later to scale in. The position, the focus we're doing on whether it is the laser or, you know, pump laser, CW laser, EML lasers, OCS, right, And then later on, the VIXO, stuff like that. We're positioning ourselves to be kind of the technology of choice in all these different areas, right? So as we ramp AI, as intensity of optical demand increasing, moving from scale out to scale up to scale in, and with the whole demand going up because the XPO number is going up, I would say Luminum's best days are yet to come. because we see the positioning of our products and technologies. We see the roadmap of our customers and working hand-in-hand with them to facilitate the realization of these opportunities. So, you know, we feel confident in that, you know, we'll be well-positioned to capture the AI growth for the years to come. Very exciting stuff.

Thank you both for joining us, and thank you, everyone, for joining us.

Thanks, Papa.