So, I'm delighted. We've got our conference here in New York, Fireside Chat, with Lumentum and Michael Hurlston, company CEO. So, some Q&A. We'll keep an eye out if folks in the audience have some questions. But maybe just to kick it off, if there's possibly anybody who's not familiar with Lumentum, how do you like to introduce the company to a new investor?
Yeah. Look, first of all, thanks for having us. I mean, I can't thank you enough. I think when we first met, again, as I came back into the optics world, it was in the OFC conference, and you were very generous because I didn't know a thing. So, you steered all the questions away from me. Now you're going to hit me with everything.
Oh, yeah.
Okay.
Honeymoon's over.
No, the honeymoon's over. That's it. That's it. Honeymoon's over. Okay. You know, I think fundamentally we are an infrastructure provider to the data center. We've now got most of our business focused on data center. And within that, we sort of think of about two pieces. On one side is semiconductors, where we're providing compound semiconductors and components to the hyperscalers and to the data center opportunity. And then on the other side, as you know well, we've started talking about systems, and we're pulling together whole systems, whether that's an optical circuit switch or a transceiver that we sell into the hyperscalers. But fundamentally, we are kind of the optical backbone that goes into the data center, and that's created a whole universe of new opportunity for us.
So, I like to try to structure these conversations with a little for people who are new and something for the folks in the weeds. Why don't we sort of start out at a little bit of a higher level? Because the term laser covers a lot of stuff. So, maybe double-click and give people a little bit of education as to what are the lasers you're making, what don't you make, what's important in terms of lasers. So, sort of double-click as to, yeah, where does it matter for Lumentum?
Yeah, you're right. Lasers mean a lot of things to a lot of people. And we'll keep aside industrial lasers and medical lasers and military lasers, all of which exist at different power levels. Inside the data center, fundamentally, you have two types that go into these transceivers, and then a third type that's emerging that's serving what we call the scale-up or scale-out opportunity. So, the primary vehicle today for lasers is incorporated in these transceivers. And for those of you who don't know, transceivers, it looks like a little USB drive that plugs into the back of a server or switch and converts electrical signaling to light. And the thing that's actually doing the transmission of light is a laser. And you have two dominant types today. One is a CW laser, continuous wave, and that is primarily what we call silicon photonics.
You're using a shutter or an external modulator to send light along a fiber. And then there's another type that we call an EML, an externally modulated laser. It means the laser itself is actually modulated, right? So, there's an advantage to EMLs. There's an advantage in certain cases to CW lasers. We predominantly make EMLs. They're the more difficult type of laser to manufacture. There's more masking steps in an EML, over two times more masking steps than there is in a CW laser. They are typically harder to get to yield. They're harder to center in the process. And by all accounts, we have somewhere between 50% and 60% of the world's EML volume. So, that's where we play. We do make CW lasers.
And then the third kind of emerging type that I refer to inside the data center is an ultra-high power laser, which we see being applied both to scale-up optics and scale-out. First, it will be scale-out, which means now connecting from a rack to a switch bank. And then scale-up is an interesting opportunity where you can actually use it inside the rack itself in various degrees. So, we like where we're positioned on lasers. We think it's really our dominant currency. We think of ourselves, back to your first question, perhaps as a laser supplier first and then a transceiver supplier a distant second, where other of our competitors would say that they're a transceiver supplier first, and they do much better in that market than perhaps we do.
Now, I think there's another way to sort of further subdivide in that when people talk about the EMLs used for data centers, they sometimes attach it a speed, 100 gig EML, a 200 gig EML. So, what's the point there?
Yeah. So, again, a super question, Simon. You understand this business better than I do. 100 gig, you know, primarily you have eight of these lasers per transceiver. So, when we say 100 gig per lane, it's eight times 100, and you get an 800 gig transceiver. So, today, the state of the art in these pluggable optics, which are taking those electrical signals and converting to light, the transceivers are 800 gig. The industry is moving to 1.6T, 1.6 terabit per second, and that shift is occurring now. And primarily there, you'd start shipping 200 gig EMLs per lane, 200 times 8, 1.6T. There are instances we see where you get 4 by 200 to make an 800 gig transceiver, as you know. Several customers are taking 200 gig per lane and 800 gig.
But the primary form factor for me, EMLs, is eight of them per transceiver, and you multiply eight by the speed, and that gets the speed of the transceiver itself.
So, your answers have been very focused on inside the data center. Now, a number of years ago, Lumentum acquired a company called NeoPhotonics, which gave you some narrow linewidth lasers. How does that sort of fit into the narrative? What's different about that market?
Yeah, I mean, this is now scale across. So, what's happening, you know, we talked about, and you and I probably for the last three or four months have talked about data center interconnect, where you're really trying to link together numerous data centers to create sort of a larger instance. But scale across, as my partner in crime, Kathy, would say, is sort of a smaller circle in the Venn diagram. And here, you're actually running an inference model. The data center itself, one single data center, can't contain that inferencing model. And as such, now I need multiple properties to run the inferencing model. And scale across now has become much more of a thing. You've written about it. Several other of your colleagues have written about it. And it's a big opportunity.
It feeds into these ZR and ZR Plus modules, which are coherent light, as you know. One of the key components in coherent light is a narrow linewidth lasers that sits at the front of that module. But then they also involve pump lasers, like amplification for traveling, getting signals to travel over larger distances, coherent components themselves, all of which we make. You know, I think caught both of us by surprise, quite frankly, into our Q2 guide, our December guide, was buoyed very significantly by the scale-across opportunity. Our customers here are primarily the Cienas, the Ciscos, the Nokias, and they're doing phenomenally well, right? You cover Ciena, and they're kicking butt on the scale-across opportunity, and they're dragging us along primarily with these narrow linewidth lasers.
So, it leads nicely to sort of the next question. We often hear from the OEMs, like the Cienas of the world, as well as other prospective partners, like an NVIDIA, as an example, about these customers thinking about doing more vertical integration. So, their messaging is often, we're going to get more vertical integration. How should we think about the risk to Lumentum that a laser becomes something they vertically integrate?
Yeah, I think, you know, all of these guys are purportedly working on that. And meanwhile, we've actually made a decision to step back a little bit from the entire integrated module. So, we compete with Acacia, Cisco, we compete with Ciena, we compete with Marvell on ZR, ZR Plus modules, and they have the DSP that sits on the back end. And we've made a decision as a result of that to sort of back away from that market. That's tough for us to compete on because that DSP is such a key piece of the bill of materials. So, we've said, look, this is probably not something that we want to compete in long term. Where we do want to compete is on the components. And those narrow linewidth lasers are something that we sell to everybody.
Reportedly, all the guys are working on their own internal solution. We don't think that anybody is particularly close. And even if they were, there's so much demand out there, Simon, we can't fulfill it all. I mean, one of the biggest things that we're getting into right now with the customers is we simply can't fill all the demand. So, it wouldn't surprise me in the asymptote, let's say, you know, sometime in 2027, 2028, I don't know, whenever these guys can get the narrow linewidth lasers to work, that we're splitting the business in some way. And it doesn't worry me all that much because the demand signal is so very high.
Now, that's sort of one of the other things that I wanted to kind of delve into a little bit because it's not just the customers, but we hear from Coherent ramping production, Broadcom ramping production, the Japanese players like Sumitomo. So, we're frequently asked, when does the market come in balance? So, everybody has the same messaging, right? Demand exceeds supply. When do you think the market gets into better balance?
Yeah, it's a hard one for me to answer because as we've ramped, and you're right, I think the other folks are ramping as well. The Sumitomos, the Broadcoms, all of these guys have talked about ramping supply. I still think we're getting further behind. So, what we've said to you and to others is, as you entered sort of the end of our fiscal year, we were roughly 20% behind demand. If we look at the forecast we're seeing for the middle of the year and the end of 2026, we expect to fall further behind. We're saying we probably think we're 30% behind. And that, in our own book, we're adding 40% additional capacity on a big number, right? Because we're by far the largest supplier of EMLs today.
So, when we say we're adding 40%, it's probably bigger than a Broadcom or a Sumitomo can do when they're saying double and triple and all these other numbers. I mean, we are the king of this market, and we're falling further behind. When do we hit supply? I just can't tell you. I mean, it's a great question. Based on the LTAs that we have, based on the requests that we have, we don't think that we hit equilibrium in 2027, at least. And then beyond that, I simply don't have great visibility.
And I think you've talked about adding 40% by June of this year. That's relative to what date? What's the?
Yeah, so it's as compared to our September quarter close. So, into our guide is factored a piece of that in the December guide. And then the next two subsequent quarters, we'd expect to add pieces of additional capacity to get us to a total of 40. You know, we've been asked today by several analysts how we're doing. You know, we're tracking to that. I feel very comfortable we'll be able to hit that commitment. And beyond that, what we've said to you is we're going to adopt this really interesting strategy where we virtualize three of our indium phosphide fabs. We're going to take three different properties. We have two of which are rather underutilized today. Surprisingly enough, with all these constraints, we have two properties that are relatively underutilized. And we'll try to run different steps of the process in each of those three properties, right?
We're doing it today. We're running our EPI process in our United Kingdom fab and then shipping that to Japan for finishing for all the E-beam and lithography and things like that, and then we'll adopt that more globally across our second Japanese fab, our UK fab, and the Sagamahara Japan fab, which is the one that's so very constrained at the moment.
Then I wanted to talk a little bit about this 200 gig cycle in that I think you've given us a metric. You expect it to be 10% in the March quarter. Now, is it 10% of revenue or volumes? How should we think about that value?
Volumes.
Volumes and so, I've assumed the ASPs cost more than 100 gig per lane.
Yeah, it's, you know, 100 to 200. It might be the same rough ratio.
So, how should we think about the trajectory of that mix and how that affects the business? And what I'm really getting at is, so you're an analyst, you're trying to build a financial model, and I've now got another variable. It's not just capacity increase, but ASP increase. How should we think about that math?
No, no. I mean, that's a super question. What we've said from a trajectory standpoint is we would expect that the 200 gig per lane lasers, as you said, 10% in Q1, calendar Q1, going to 25% in calendar Q4. So, it'll be an increasing part of our overall mix as our mix is going up, as the number of units go up through that time period, which, as you correctly call out, leads to more revenue. It's an ASP bump of roughly two times and then a margin bump, right? Because the gross margin on the 200 gig per lane is definitely better than the 100 gig. So, you've got a number of different tailwinds in our EML business. Again, capacity going up, mix shifting from 100 to 200, all of which are super good for us.
And there's been this, I think, long-running debate about which laser wins, EML versus VCSELs versus CW. And I feel like one of the messages that came out of the ECOC show back in September was that CW and Silicon Photonics was taking over the market. Now, my theory is you say that if you can't buy EMLs. So, I'm leading the witness, of course, and I'm sure you're a little biased, but how do you think about the competitive forces among different architectures and what's sort of the right answer for the industry?
Yeah, look, we've been honest. I mean, I think that there's no doubt that you'll see a mix shift towards CW lasers at the 1.6T node. There will be more, as a percentage, there will be more Silicon Photonics-based transceivers at 1.6 than there were at 800. That doesn't mean that the number of EMLs go down. We'd expect that the number of EMLs go up because the number of transceivers is going way up, even in the phase, and we haven't touched this yet on co-packaged optics. Co-packaged optics certainly eats into transceivers a bit, but the market is growing so very significantly that we would expect, in the face of some share loss, that we would still expect the EMLs to go up in a rather pronounced fashion. That changes again. I mean, you understand it as well as I do.
At 3.2T, the next node, our folks think, hey, Silicon Photonics doesn't work. It runs out of juice, particularly at reasonable throws, reasonable distances. And as such, the market will ebb back to EMLs. I would also tell you, and this is very true, all the initial 1.6T designs are EML-based, right? There's very few, if any, Silicon Photonics-based designs that are coming. They're easier to make. EML, you have one throat to choke, right? Because we're doing the modulator and the laser. But they're just generally more robust designs. So, people are going to lead with those, and then the CW lasers, the Silicon Photonics-based will come in on the back end.
So, I want to pivot topics to the optical circuit switches or OCS. Interestingly, I don't know if you caught this, the market researchers at Cignal AI, they increased their forecast.
Oh, they did.
No surprise.
Okay.
They're now.
What's the new number?
2.5 billion by 2029, which maybe proves a little bit conservative, I think, but we'll see. Maybe, first of all, help people understand what this market is, what you do here, and how you're seeing that market developing.
Yeah, no, I think that number is very conservative. I'm sure you have a model in your mind, Simon. It seems wildly conservative in terms of the TAM. You have three different dominant use cases for OCS. You have a spine switch replacement, and what we're seeing is a lot of greenfield deployments that are coming out. New data centers are adopting the optical circuit switch in the spine layer, and that is probably the first volume that certainly we're seeing with a slower ramp, right? And then there's a use case where you have TPU interconnect, and this is really a scale-up. I mean, for all respects and purposes, you're using OCS inside a rack to connect multiple TPUs together, that is a use case today that exists, right? There's one hyperscaler that uses that model, and, you know, from our perspective, obviously a big opportunity.
We don't know yet that that particular customer will be looking for other alternative suppliers. The good news for us is they use a MEMS-based solution today. We would expect that that familiarity would give us an inside track because obviously we're a MEMS-based, as you know, as we look at that one as an opportunity. And the third one is kind of an interesting opportunity where you're now to have GPUs inside a cluster. And as bad a rap as optics gets, and you've been around the industry for a long time, it turns out that GPUs actually fail more often than optics. And when you have a model, an inferencing model that's costing you tens and, you know, fifties of millions of dollars to run, you cannot have a GPU that fails.
And so there's a big use case out there where you're actually steering traffic away from utilized, but actually more importantly, failing GPUs. And so these three use cases actually make up the revenue target that you and I have discussed. Sort of $10 million of ROC, $10 million of OCS in Q1 going to $100 million of incremental dollars for us in Q4. All three of those use cases make up pieces of what we'd expect to see. And, you know, we're super excited. I think the $2.5 billion in 2029 is very, very conservative based on what we see in front of us.
For you, what are any milestones or limitations, capacity to get from 10 million a quarter to 100 million by, just to clarify, December of 2026, calendar quarter?
Calendar quarter, right?
What needs to happen? Do you have the manufacturing set? What are your hurdles?
Yeah, our biggest limiter is ourself. I mean, the demand signal we're getting from customers is far in excess of what we can build. We've got to build the manufacturing capability. We have to build out the supply chain because this is one where we're depending on key strategic suppliers, both for the analog voltages that drive our MEMS, for the MEMS themselves. You know, if we can get the supply chain to work and we can get our manufacturing to work, I think we can beat the numbers that we've put in front of everybody as targets. But that's still a big what-if. I mean, this is a tough ramp. We're running, we're going from a standing start. We're manufacturing all of this in our factory in Thailand, which is already very constrained with the transceiver activity, some of the optical scale-out.
I mean, there's a ton of stuff that's going on in that Thailand factory, and we're trying to rejigger resources every single day.
Now, do you do the MEMS yourself, or is there an outside contractor that does that fabrication?
Yeah, it's outside, so we've contracted that out. We haven't said who it is, but it's a very, very key and important supplier for us that we spend a ton of time with. They've been working with us since the days of WSS, which you covered for years. It's the same fundamental technology that goes in our wavelength selectable switch that we applied to telecom that we're now reapplying for this OCS opportunity.
There are a number of different architectures or designs. Do you feel like MEMS has a particular advantage over, say, LCDs or piezoelectric, or, you know, it's a black box, it doesn't matter what's inside it?
No, I think it does matter. I mean, the biggest thing that this is, remember, this is not a packet-based switch. It's a traffic switch. And so when you're switching traffic, you want to have perfect transparency between one port and another, meaning no loss. And so MEMS has an advantage that it's a mirror. And then also, and we know our competitors do this, they have, we're talking about different SKUs for different wavelengths. We're also wavelength agnostic. So you can apply one SKU to C-band, O-band, L-band, whatever frequency you're trying to switch, you can deploy one of our OCSs, whereas competitors who use different technologies have to be in a certain frequency band. The knock, right, is the reliability. You definitely have pieces that are moving. You're trying to move these mirrors to direct the light from one port to another.
That we believe we've solved, and we have proof points everywhere with this WSS solution. We've been deploying this for years and years. Telcos are digging earth up to deploy these WSSs. The last thing they want to do is pull them out. So we really feel very confident we've overcome the knock, and there's a knock out there on the reliability of these things. We think we've overcome that, and then the technical advantages that we have are serious. So we do think that as people deploy this, that technical advantage gives us a big running start.
I want to go to co-package optics. I think it's interesting to tell folks that sort of a little bit of the backstory in that I think if you look at Lumentum's stock chart from sort of December of 2024 till March of 2025, it just was down, down, down because co-package optics were the end of Lumentum. And then Jensen gets up on stage and shows his slide to announce co-package optics, and lo and behold, Lumentum's a partner. I think from that day on, the stock's just pretty much gone up. Market perceives co-package optics as now a good thing, not a bad thing. Maybe unpack this a bit. I assume you think it's a good thing too.
We definitely think it's a good thing. Look, there's been a knock on co-packaged optics again for a long time because of the reliability of the optical solution and the fact that, okay, now I have a big switch box or a server box, and I have these optical engines next to it, and my whole yield is now going to be impacted by lasers. I do think that the approach that NVIDIA has taken is very smart, and that is that they have an external light source, as you well know, that looks like a pluggable. So it's basically the same as a transceiver without the DSP and the associated electronics in there. So it's a very simple solution, very elegant. We believe that there's no reliability problem, similar to what we were just talking about, but to satisfy their customers, they've come up with this form factor.
We believe it's doing very well. We're already on the InfiniBand switch. We're shipping in limited productions. We would expect to see this take off on Ethernet, on the Ethernet form factor. Look, you know, if you look at Lumentum, the growth drivers on OCS, on this optical scale-out, and then we haven't talked about optical scale-up. These are huge growth drivers that I think are underappreciated. Generally, look, this is a good thing for us. We tip our business away from the transceivers, right, to the extent that we have transceiver share. We don't. We're not a big transceiver player and more toward components, which are higher margin. For us, you know, when we compare them to the lasers going into transceivers, higher dollar content.
How do you think about sizing the co-packaged optics opportunity? I know we've got sizing for OCS. We have sizing for transceivers. This is a trickier one, I feel.
Yeah, I think, you know, we feel like it's going to be for us, the sort of the data points that Kathy and I have given is we feel like it's going to be material revenue for us in the back half of calendar 2026. And, you know, we believe at least at the outset, we'll be the only supplier, at least with the opportunity that you just mentioned. We believe that several customers are looking at this, both for switches and for processors and for servers, where we can expand the customer footprint. You know, we just had an announcement from my buddy Matt Murphy now buying the Celestial AI to us. That's a nice proof point that co-packaged optics is indeed a thing, right? You're not going to pay $5.5 billion for something if you don't buy the market.
Now, are you in that project, or is that something where they're vertically integrating?
No, I mean, we could, right? So the substrate and all the PIC and all the electronics, that's all coming from Celestial, but a light source they need somewhere.
They do, okay.
You know, we have not said one way or the other where we're engaged there, but we're a likely candidate to play. Anywhere there's going to be co-packaged optics, we feel that our light sources. It just creates more opportunity for us on the light source vector.
So some of the work we've done is looking at how the sort of architectures evolve. So it seems as if when you talk about co-packaged optics, these are for scale-out, rack to rack.
Right.
So when we get to a point where optics are necessary for scale-up, my guess, 2029?
Yeah, and look, it could be sooner. I mean, we're getting indications now that the copper is running out of juice in certain applications. So you're right. If you think about the racks themselves, there's servers and switches inside those racks, and then there's a rack of higher-end switches that exist in the topology. That is where scale-out is going, where taking it from a rack and going to the switch rack, the main switch rack inside a cluster. I think what's being discussed for scale-up is something where you're going to connect in various ways and shapes and forms GPUs, because you got to maintain the bandwidth, right? Remember, when you're going scale-out, you're cutting the bandwidth. The bandwidth is no longer perfectly. It's usually a ratio of eight to one. In scale-up, you're going to have to preserve the bandwidth.
And with these higher speeds, you have to think about electric cables, copper cables running out of the juice to do that. Maybe at first, it's longer throws. There's throws inside the rack where you're connecting one GPU to another, not just along the backplane. But ultimately, I think this thing is going to go everywhere. I mean, you're going to have to have a high degree of optics even in the backplane. And this opportunity for this industry is, you know, scales of hundreds of millions, where we've been an industry scaled on thousands. That scale has changed to hundreds of thousands. Now, as we think about co-packaged optics as scales of millions, if you think about scale-up, it's tens or hundreds of millions of units. I mean, it's a huge opportunity.
But to start, back to your timeline, you know, we could see a world where this happens late 2027, earlier 2028. There are use cases we see already where copper is running out of juice.
So believe it or not, we've run out of time. I think I've got another hour's worth of questions for you.
Oh my goodness, okay. All right.
Appreciate it. We'll take it up again maybe at OFC.
I always love chatting with you, Simon. Like I said, I really appreciate it last time. You took it easy on me. You're a gentleman.
Now, let me just ask you to sort of close with, what do you think is the least appreciated or most misunderstood aspect of the Lumentum story?
Yeah, I mean, I think that everything you see now, and you're right, we've had a nice appreciation in the stock since OFC. Everything that you see is really based on the core business, meaning the EMLs, the scale across, the core sort of connectivity that we've had, transceivers contributing a little bit to that, right? But none of our numbers are the three big things that you touched on. OCS, not in the number. Scale-up, not in the number. Scale-out, not in the number. So if you look at how we're thinking about the business, 2026 will be a very good year for the company because we will start to see contributions from OCS. We will start to see contributions from scale-out optics. And we think that we'll start to see scale-up, at least we'll formulate where that strategy comes from.
So I think our best days are ahead of us.
I'll paraphrase Gary Smith's comment, which is optics are cool again.
They are cool again. Gary had it absolutely right. I texted him after our last earnings and said just that.
Great, great guy. Thank you very much, Mike.
Okay, thanks everybody.
Thanks, Simon.
Great.
Unbelievable.