I guess we can go ahead and get started. My name is Kathryn Ta. I'm the Vice President of Investor Relations here at Lumentum. Really, really glad to see you all here. See a lot of familiar and friendly faces and, as expected, a packed room with everyone sitting around the periphery. Don't see anyone standing just yet, but I think that will happen as people leave some of the other events. Today we're gonna be talking with you about illuminating the networks of tomorrow. I'm super excited to be doing this event right after yesterday's events at NVIDIA. I'm sure you all have questions about what you heard at the GTC keynote. I think we'll be able to clarify a lot of those questions in today's presentation, especially in Wupen Yuen's presentation.
A few brief remarks. I won't read this whole thing, but I need to read part of it. Today's presentation will include forward-looking statements that are being made under the safe harbor of the Private Securities Litigation Reform Act of 1995. These statements are subject to risks and uncertainties that are set forth in our filings with the SEC and are based on our reasonable beliefs and expectations as of today, and we undertake no obligation to update or revise these statements. As I mentioned, we will have Wupen Yuen as one of our speakers. First, we will have Michael Hurlston take the stage and introduce our topics for today and give us a state of the union on the industry. We'll have Wupen talk about all things CPO and OCS.
We'll have Wajid at the end, so make sure to stay until the very end. The last slide should be exciting as well. After our prepared presentation, we will have a Q&A session. At that time, it'll be about an hour from now. I'll invite people to line up behind this center microphone here, and then you can state your name and your firm and ask your question. We'll all in all take about an hour and a half of your time. With that, I would like to invite Michael to the stage.
Thank you very much. How's everybody doing?
Great.
Everybody's good. Everyone's gonna have to sit through Wupen Yuen and I to get to the last slide. That's what Kathy basically said. We're filler. Wupen Yuen, did you know that?
I know that.
We're just waiting for Wajid. Is that true? Okay. Well, Wajid, it's all up to you, buddy. The model. That's anybody ever cares about. All right. Look, we wanna just give you some sense of where we are as a company. You know, I think that there are sort of four building blocks that often don't get recognized. Of course, we have a deep networking knowledge. What we have done with our products, if you think about it, is we've evolved. We've taken technology that started in telecom from our telecom heritage, and we've now evolved those into data center and hyperscaler applications.
We've really upped the ante and taken existing technology and evolved it, and that's why I think we have a really big lead because we've spent years and years perfecting these technologies, and now here we are in a position to ship them to hyperscalers in different ways, shapes, and forms, and that's given us, I think, a pretty big defensible moat. Of course, lasers, right? We're gonna spend a lot of time, and Wupen Yuen is gonna talk a lot about our laser technology. This is something very fundamental to us. Most of our business is component-driven, semiconductor-driven, and as such, we expect to continue to evolve our margins towards semiconductor-like margins. One area that's a big example of this evolution of our technology is the optical circuit switch.
Wupen Yuen is gonna describe this in detail in his remarks, but if you think about the OCS, it's really based fundamentally on a WSS block that we've perfected for years in our telecom applications, right? That is something very, very much that we've done, and we've taken that and reapplied that to the OCS. Then finally, we've got indium phosphide capacity. We probably have more capacity than any company on the planet. As we think about these various ramps that we're gonna talk about, EMLs, CW lasers, ultra-high-powered lasers for scale up and scale out, it's coming on the backs of a capacity that we've built and we've continued to build upon during the 10 years and 15 years that Lumentum has been in existence. That's kind of our, the basis of our company.
Speaking of the indium phosphide capacity, Kathryn went back and looked at this for us, and the first data point that she gave was sometime in 2023, where we were still a little bit in the doldrums of this telecom crisis. We were sort of bottoming out, probably coming down in terms of revenue as a company, and we gave our first data point on EML capacity, okay? Since that time, since FY 2023, we've increased our EML output by 8x . We've been very, very successful and continue to be successful in terms of ramping our indium phosphide capability.
If you look at this metric right, which is a new data point, we are saying that we are going to increase, further increase our indium phosphide output by 50% as measured from the last quarter of the calendar 2025 until the last quarter of calendar 2026, we will be up by another 50%, which is a remarkable achievement. We've talked about 40% over the last three quarters. We're now upping the ante to say we are going to increase this indium phosphide capacity by an additional 50% on top of what's already the industry's largest base. Okay? Despite that and despite our ability to continue to evolve our indium phosphide capacity, we are still under shipping the market and frankly, that gap is growing as this chart shows.
We see the lanes increasing by an 85% compound on a compounded basis. Our output is increasing something less than that. Even as we talk about additional adds to our indium phosphide capacity, we are falling further behind. The demand is significantly outstripping our ability to supply. We've given the figure of merit that we are under shipping the market today by about 25%-30%, and as I say, we see that gap growing considerably over the next couple of years, primarily driven by the demand on our ultra-high-powered lasers for optical scale out and optical scale up, as we'll talk about here in a minute. We're increasing our capacity. We're doing everything we can from a leadership position, the leader in indium phosphide capacity, but we are still under shipping the market significantly. Okay, we've talked about this repeatedly.
Wajid, Wupen, and I have been on the road with Kathy, and we've talked about our four growth drivers that exist in Lumentum that are sort of separate from what's happening in the hyperscaler market in general. These four drivers are cloud transceivers, our optical circuit switch, our scale out CPO for scale out, and CPO for scale up. These are the four major drivers, and I'm gonna try to update all four. On our cloud transceiver business, we do feel like we've turned the corner. If you look out on the floor, we are one of the very few companies that's showing a 400 G per lane demo, an optical transceiver that is capable of doing 3.2 T, eight by 400. We are in the lead pack now of transceiver companies delivering 1.6 T.
We feel like we're ahead, and we're now executing very well from an engineering standpoint. Our profit margin is improving. This has been a real struggle for us. We've actually not done a great job with this business, as I've told you many times, but we're now starting to really see cash flow coming out of the business. Our 1.6 T shipments start this summer, so we'll start seeing our first 1.6 T shipments, which we expect to see improved gross margin on. And then our vertical integration, bringing our own lasers to help with the margin will start happening in the summertime. We're starting to see a lot of good signs. From an engineering perspective, we're executing a lot better than we have, and from a manufacturing side, we're starting to turn the corner.
We're really starting to execute and see better margins, better deliveries, and improved revenue generally coming from our cloud transceiver business. The second growth driver, this is an important update. We've talked to you guys about $400 million of revenue shipping in the back half of the calendar year. We just closed yesterday a new multi-year, multi-billion-dollar agreement with a large OCS consumer. We are really excited about this. We believe that this will lead to multiple years of revenue growth and revenue enhancement on our OCS line. Our OCS line has been something that we think is performing remarkably well. We're well ahead of competition, as we've said, particularly on the 300 by 300, the high radix switch. This is another proof point that we lead here.
Our revenues are coming up, and we're seeing now engagements and LTAs we'll be able to fashion by virtue of the position that we have in our OCS market. We are very much on track to be shipping this $400 million of backlog in the H2 of this calendar year. We like where we are from a reliability standpoint. Our MEMS-based solution that Wupen Yuen will go into some detail we think is field hardened and has given us a competitive advantage, a moat, that is gonna be very difficult for competition to overcome. We think our design, right, if you open our box as opposed to other boxes that might be shipping, we feel that we have by far the most elegant manufacturable design that leads to better margins, that ultimately will lead to better throughput on the manufacturing line.
One thing that Kathryn put down is it's important to recognize our solution is future-proof because we can ship. We're agnostic to any wavelength. It simply mirrors. All we're doing is reflecting light. Any wavelength that comes in can go right back out. That is important as people think about different bands that they may want to use the OCS product in. Okay, our capacity, right? We're going to talk a little bit more about this, particularly on UHP. There's been tons of discussion. There was discussion yesterday from NVIDIA around optical scale-up. We'll go into that in some detail. I guess Alexis was talking a little bit about it today. Rafiq was just showing me a slide.
It does seem like there's a lot of conversation now on scale up, but in order to meet that opportunity, we really have to build capacity. What you're gonna see from us in the next couple of slides is what's the plan? How are we going to actually get there? The first thing that we're doing is we're ramping our UHP lasers in our San Jose fab, and that actually is going well. We're already shipping, as I've reported many times. We're actually shipping our UHP lasers for optical scale-out applications. In 2027, we expect another step up in capacity for our UHP lasers as we start shipping out of our United Kingdom fab in Caswell. We'll continue to build that, as it says, and I'll describe that in the next slide.
Right now, one of the things we're doing, we obviously had this engagement with NVIDIA. You saw the output of working to pre-sell this capacity. We had a $2 billion investment from NVIDIA. We are going out, Wupen Yuen, Wajid Ali, I don't know if John Bagatelos is here, our sales guy, Sri, who runs this business, are now working with other customers to sign up long-term agreements and capacity arrangements whereby we would sell the rest of our capacity. I mean, we are already close to sold out on all of our UHP in spite of the fact that NVIDIA basically took almost everything that we can make. We still have a little bit of room, and we're trying to figure out how to go and sell that. To help the project, we're announcing today we finally closed our fifth indium phosphide fab.
This indium phosphide fab is in Greensboro, North Carolina. It's an old Qorvo fab that we just closed literally this morning. The agreement was closed this morning. That will up our capacity very, very significantly and allow us, we think, to be able to meet the demand that we see from Nvidia and from our other customers on optical scale-up and optical scale-out. We expect to be shipping out of this Greensboro fab by 2028, right? That's how long it's going to take. It's a nice fab in the fact that it's a brownfield. It's already staffed, it's already operational. In fact, the trick we're going to have is move out some of the Qorvo products that they're shipping out of that fab and bring in our own reactors, our own EPI reactors, and bring it online for what we need for indium phosphide.
The great news is there's already a clean room there, already fully staffed, already fully operational, which gives us a running start into being able to produce in 2028. This is a really big step for us, a commitment that we're gonna continue to invest in our indium phosphide capacity. Okay, optical scale-out, right? As I say, we are already shipping optical scale-out product. Jensen talked a lot about it yesterday at GTC. We're doing extremely well. We're sole-sourced at the moment. You know, we expect competition to come in, but the reality is this market is huge. This market is very, very significant.
We have set expectation that we'll have our first $100 million revenue quarter in the end of 2026, calendar 2026, and then deliver on this multi-hundred million dollar commitment that we talked about in our last earnings call in early 2027. We really see some good things. It's been a surprise to us how well the optical scale-out product from NVIDIA has done and how well customers are receiving it. It's been a pleasant surprise. Surprise enough to us that we got additional backlog, and we believe that we'll be able to execute to that and deliver incremental revenue in the first half of 2027. Similar to what you'll see here in a minute on optical scale-up, one of the important vehicles we have is this external light source. Being able to pull together a turnkey package to enable customers beyond our lead customer.
That's a critical tool because there's just not the optical engineering that you see anywhere else. NVIDIA, Google, these two companies have incredible optical engineering departments. The rest of the market, not so much. In order to engage them, we need something a little more turnkey, and this is what this ELS does, right? That will give us about 2x the top line that we see today. For every ELS that we ship, it's about 2x the revenue. Margins come down a bit, but it's still very, very profitable, above our gross margin, above the gross margin target that Wajid is going to share with you in a minute. We really like it because strategically, it's an enabler. It really enables us to engage the rest of the customer base. Okay. Optical scale up, right?
This is the big story, I think, and there's a lot of debate around optical scale up. It's definitely happening. Does it mean that copper goes away? No, of course not. It doesn't mean that. I think Jensen said that yesterday. You're gonna see a hybrid environment starting the H2 of 2027. We will be shipping our optical scale up solution, and it won't be entirely at the expense of copper or retime copper. You're still gonna have hybrid environments. It's across the longer runs, as I think Kathryn and I have described to you, three-meter cables and up will use optics inside a cluster, and then for shorter runs, you're still gonna use copper. It's not a winner-take-all. I think Matt Murphy was talking about that yesterday.
You're gonna have a blended environment, but what it represents for us in the optics industry is pure incremental upside. We have never been in this position before. We have never shipped optics inside a cluster. We believe that as you go on this copper wall, which Wupen Yuen is gonna describe in a minute in his slides, as you start moving down the copper wall, there's gonna be increasing opportunity to ship optics. We will have optics inside the rack as well as across racks inside a cluster. This is gonna be nothing but upside for the optics industry, and we can see every customer, not just NVIDIA, every customer moving along this vector, with the possible exception of Google, who has their own architecture by virtue of the OCS, which does almost the same thing. It enables optical scale up.
Every other customer is interested in some way, shape, or form in what's happening with our co-packaged optics solution and our ultra-high-power lasers. Again, similar to what we talked about, we will enable this ELS, this turnkey module, such that we can engage customers that don't have the optical engineering expertise that our primary customers do. This is a real vehicle for us to engage the broader market. Okay, so just to kind of finish up and then turn it over to Wupen Yuen, we're still just the warm-up act. You realize that? We're just the warm-up act.
Warm-up act.
Okay. All right. We're just the warm-up act. You know, cloud transceivers, we talked about this sort of $1 billion cap that we were gonna have. Obviously, we're blowing through that. Our business is improving dramatically. Still a gross margin headwind. We have work to do to improve the margins in this business, but we are improving it from an engineering execution standpoint, getting into the lead pack, which is really good. OCS, right? We have this new multi-billion-dollar order or commitment that we received. This is super good. We expect it to be ramping beyond $1 billion in calendar 2027. Building on the $400 million that we've described in the back 1/2 of 2026 and accelerating. Some people have asked us, "Is that just a bubble, a sort of an order seed?" The answer is no.
We would expect to see significantly more business across calendar 2027 on the OCS. On optical scale out, right, our first instantiation that was talked about last year at GTC, we're shipping. We're already in business on that one. We've got this multi-hundred million-dollar order that we need to fulfill in the first half of 2027, so things are going extremely well. The big opportunity clearly is scale up. The first instantiation of scale up, which is again, sort of moving across links inside a rack, we estimate to be 3x-4x larger than the initial scale out CPO, so big, big bump. As we get inside the rack, which inevitably will happen, it's 10x larger in terms of the number of lanes that we see.
We're now unlocking a new indium phosphide fab to meet all this capacity and still, as I said many times, we're under shipping this market. We're still vastly under shipping where we think we can be given the demand we see. Forget about the rest of the industry. The demand we see is so very, very high, we cannot build these things fast enough. We're sold out, as I said many times, through calendar 2027, even as we bring on all this additional capacity. Okay, Dr. Yuen , you're gonna wow the audience with the technical aspect. Is that true?
My goal is to ramp it up.
Ramp it up.
Get ready for budget.
Okay. This is perfect.
That's my goal.
Okay. All right. Thank you, guys.
Good morning. That's a picture before the AI tornado. I think my hair is getting grayer and less. Let me start with the GTC yesterday. I don't know how many people went to GTC, but the GTC, there's one very, very clear message, is that the agentic AI is going to change the world, and the amount of token generated is gonna skyrocket, right? We see that. We actually feel it. All these numbers that Michael just talked about, the laser numbers, I mean, just give you one example, some customers are asking for 1 billion lasers a year. Starting with a B. Used to be in the millions, now it's in the billions, right? The amount of scale is just unimaginable.
You think about, you know, this, what's driving all this really is agentic AI. We are still creating the intelligence. When you are serving intelligence at scale, and we're actually allowing the agents to do a lot of work for you automatically, it's a big deal, right? Just last week, we had a dinner with hyperscalers. They're telling us and say, "Hey, you know, we're hiring new people now. Their job is to work with five agents or six agents every day." Their job is composed of working with agents. All these agents are AI agents, right? Just imagine when consumers and moreover enterprises start to take on this kind of approach to work. I sound like Jensen Huang a little bit now. It's crazy, right? Actually this time it is very different.
It used to be the case where, you know, in telecom, so-called bubble, fundamentally it was for communications. Compute fundamentally is determined on how many people you have in the world and how much time you can spend on YouTube. That's fundamental, right? This time, we're actually creating intelligence and serving intelligence at scale. I don't think anybody can say we can go back to pre-AI anymore. I think that era is over, right? Going forward, and this has a lot of implications on optical networking, right? Because it used to be optics is a way for communicating. Now optics is part of compute. The AI does not scale, period, full stop, without optics, right? So there's a few very important things. The first of all, now all the traffic generated in data centers by machines, not by humans.
That actually drives a lot of so-called East-West traffic in a very, very big way. The features are, number one, you need to have all these networking elements become non-blocking. Meaning you don't want your AI to be stuck, right, at a node, at a switch or something like that. Therefore, low latency, non-blocking is a key feature. Why is that important? Because it used to be the cloud networking, there's now so-called aggregation, right? The more aggregation you do, the less bandwidth there is. But in the AI networking, there is no aggregation per se. You have to be making it non-blocking to reduce time, right? That's a very, very big change from what it has to be. Then also to do achieve that, then you need to have the highest bandwidth connectivity possible.
Both so-called scale out, which is connecting the GPU or TPU clusters, and also scale up, which connecting all the XPUs at its full line rate, right? You just have to optimize and maximize the bandwidth as much as possible, which is a really big change. Finally, it's also one thing that matters as well is that now there's the traffic flow is not really in terms of small packets or so-called mice flows. It's called elephant flows. The flows are actually huge, right? What does that mean? Actually, that now it opens the door for a circuit switching rather than a packet switch. This is where optical circuit switching, the OCS, comes into play, right? These features of how the traffic pattern is changing, how the traffic amount is different from the past, is actually creating this new opportunity for optics fundamentally.
I have to tell you, famous last words, this time is different. For that reason, you see this huge ramp, right, of the CapEx. Frankly, this number, when it showed up, right, a couple of months ago, it wasn't really a big surprise. Because you hear, you know, in the news that Anthropic, OpenAI, all these guys saying, "Hey, you know, if I had more compute, I would have more revenue." Period, right? You see all the agentic work, you see all the Claude Code. Now, what's it called? OpenAI yesterday talked a lot about, you know, now it has become a thing, a phenomenon now, at least in Asia. People talk about feeding my lobster, right? That's a term now in Asia, right? All these things it's driving.
Therefore this huge ramp of the CapEx from this top four hyperscalers really is not a surprise. If you look at all these hyperscalers, they're investing in securing capacity. You know, most famously on HBM, you know, committing all the way through 2028. You see it in the news everywhere. You look at the TSMC talk earlier this year in their earnings call. They say, "Hey, we are seeing our AI semiconductor growth rate, CAGR, to go from mid-40s to mid-to-high 50s over the next five years." I'm frankly not surprised at all, right? Because again, the paradigm shift, using the cliché, the paradigm shift of serving, generating and serving the intelligence and increasing the scale is driving all this.
I have a lot of confidence in terms of not just 2026, 2027, and 2028 and beyond. This is why Michael talked about we're investing so much into the fab capacity and still not nearly enough to serve all demand that we're seeing. Okay, this is a complicated slide, so let me slow down a little bit. There's been a lot of discussion on optical scale up, scale out. You know, what does that all mean? Let me start with something very fundamental, right? The lowest line there is most fundamental. Really is what's driving all this, right? Fundamentally, right, to scale AI, to scale the compute, you want to package more and more of compute and memory together, right? That's the first unit you talk about, right?
Therefore, you see more of the CoWoS, you know, from two reticles, one single die reticle to two reticles, to four reticles, to panel scale. People are talking about how can I pack more compute and more memory on a single so-called chip. It's not a chip anymore, a single package, right? Secondly, you want to do, you want to actually form a large so-called coherent domain of these XPUs and then connect them at a full native rate, whether it's NVLink five, six, seven, eight or whatever, things like that. That number is going up too, because the compute is increasing and the bandwidth needed to bring out the data is also increasing at the same time, right? Therefore, there's a very, very strong need to go to say, okay, can I increase my, you know, bandwidth per lane? Okay?
Today we're at 200 G per lane. At 200 G per lane, you can still do inside the rack with copper, right? Yesterday, Jensen Huang showed this huge cartridge of cables. That's all copper. At 200 G per lane, copper works. Going to the middle plane, right? It probably doesn't go the full two meters, but going from, you know, the servers on the top to the middle of the rack, it still works. Right? That's what we called the phase I . Today the phase zero is that we keep all that copper, right? But the scale out portion network, we start to use CPO. We call it CPO phase zero. Okay? In this network we also use OCS.
The OCS is used at a higher aggregation level, whether it's so-called spine or super spine. This is part of our business today we're shipping to. Actually another part of the business we're shipping to, which is a big one, is optical scaled up. That's for a very specific customer. That's actually today. That brings us from today, 2026 into 2027. Okay. Now the next phase, you stay at 200 per lane of copper. Now you want to again, keep in mind largest coherent domain possible. Now you want to scale your XPUs across multiple racks. How do you do that? Copper doesn't go that far anymore, right? You want to use optics in between these racks. Carry the full traffic of the entire, the XPU traffic.
That's when we see a major bump up of the optical demand. We estimate about 3x-4x, depending on how much traffic is, you know, diverted, right, over optical. We also see that's opportunity for OCS. We call it low dimension OCS, roughly 60, 70, 80, you know, by 80. 60- 80 by 60- 80. That size of OCS, that will be a good complement into this kind of phase one scale up world. Okay. Finally is phase II. Phase II is when the copper even inside the rack become very challenged. Either is very expensive or is very power hungry, or it doesn't go very far or all of the above. That's when optics will go all in and to start replace more and more of the copper.
Then fiber runs all over the, not just, you know, across the rack, but now inside the rack and across the data center. Right? At that point you need to use again large radix of OCS to switch and to route the traffic. Right? Like Michael said, which is really, really important, today optics has zero share. Zero share in scale up. Any uptick, any share increase of optics in that huge volume space is a huge deal for optics, right? Therefore we look at it as incremental gain for us and for our peers rather than a loss of any sort. Right? This is how we're looking at this phase zero today through 2027, probably the next phase, 2028, 2029, 2030-ish, maybe late 2029, 2030, go to the next phase.
For the next five years, you're gonna see a continued increase of optical content in scale up, paired with increasing use of optical switch to connect all these fibers and route traffic. If you then, you know, pull out a little bit, right, now we'll look at the whole evolution of optics. We identified four key elements that's not Lumentum specific, but Lumentum happens to be well-positioned in all four of them, right? First one is CPO. CPO is the ultimate interconnect. Nothing beats CPO, right? As data rate goes high, you just have to use CPO to get you scalable, low cost, low power consumption, connectivity. Second thing is optical transceivers, right? Transceiver is going to be complementary to CPO. CPO is not a completely overnight replace, transceivers. There's no way. Nothing changed that fast, right?
The transceivers has an advantage of being able to provide a more granular units of bandwidth. You can plug on the servers. You can plug on some other different places, give you a instantaneous actually upgrade of speed because it is actually easier or less challenging to implement the latest optical technology into a transceiver. You know, you know, as one unit, right? Then you can use that to scale the network much faster. Sometimes you also want to maximize the bandwidth per fiber. The transceiver also give you the opportunity there by providing you design flexibility, right? So transceiver will be there for many years to come. Third piece, which we didn't talk about yet so far, is high-speed lasers. Today, Lumentum is the largest EML laser supplier in the world. 100G, 200G.
We're working on 200 G right now. In this OFC, we're gonna demo, as Michael talked about, a 400 G differential EML, right? Again, the high-speed lasers is again instantaneous, the fastest path to increase the bandwidth per connectivity unit, right? We're gonna demonstrate this module for the first time, you know, using a differential EML. We believe that the high-speed lasers will continue to be really, really important as a co-evolution of the CPO technologies. Finally, is OCS. We believe that OCS is fundamental to the evolution of AI networking. It will be there. It will be there to route traffic, to be able to provide the connectivities at a very high scale, from scale up to scale out, to spine, super spine, to protection.
All these applications will find, you know, use O-OCS being a very useful tool to, for, implementation. This is a chart that, you know, this has gotten a lot of debate, right? I want to say here really is that there is roughly a physical line between the optics and the copper. That line depends on whether you use so-called direct attached copper, pure copper cable with nothing on it, which is the lowest cost, lowest power. Unfortunately, it doesn't go very far anymore, right? It runs to a limit about one meter or so at 200 G per lane, and at 400 G per lane, probably doesn't go very far. Right? You have this called the ACC, which has a retimer inside.
You have an AEC, which has a DSP inside. The more compensation and capability you add to the link, the more power it consumes, the more latency it generates, and also the more power and cost will go up as well. Right? There's a blurry line over there, but we believe that at 400 G per lane, copper will run into a big deal, big challenge. Despite the best technologies TSMC and analog world can produce, it's gonna be very challenging to really have a highly reliable 400 G per lane copper. People are definitely trying, right? Last week, Broadcom talked about they are trying to make this as a PAM4, PAM6. I think we'd recognize that. But still, to broadly apply the copper at 400 G per lane, it's gonna be challenging.
Therefore, we see in that world, there will be a mix of the copper and optics, you know, inside the rack, depending on the quality level required and also frankly, depending on how much volume the CPO world can actually supply. Right? Overall, what we see this is that at 1.6 T is intra-rack copper, inter-rack optical. At 3.2 T or 400 G per lane, you're gonna see intra-rack copper probably mixed with optics, and certainly in between the racks will be optics. Again, it's all net up for optical components. Let me spend a few words a little bit here on the high-speed lasers. We talk about a little bit, right? Why is EML laser, as investors know, so attractive? Because the laser includes a light source and a modulator.
When we take it, fully yielded, that laser, you know it works. That's super attractive, and compare that with silicon photonics today. silicon photonics today, you have to get a so-called CW laser and launch into an array of, silicon photonics chip. The trick is that you have to make sure all the lanes yield at the same time. Whereas EMLs, you put one at a time. You know every single one works, right? That's why people always use EML when they go to a new speed node because they know that thing works. That thing doesn't require to yield the whole solution at the same time, right? Another thing also EML does really, really well is unless you want to have many bandwidths, many wavelengths to put a lot of bandwidths into a single fiber.
With silicon photonics, it will be challenging. Because now you have to put all these lasers, all different color of lasers, align them together, hopefully they all work together. Whereas you use EMLs, you can have all the wavelengths put into the module, and then you know it's gonna work. Right? These are the reason why the EMLs are attractive and always be the first to market, first to volume, for a new speed node. Frankly, we don't see that changing. Now, at 400 G per lane can be even more interesting, right? Again, we're gonna demonstrate a four by 400 G module with differential EML. This can be a bit more interesting because silicon photonics has a challenge at 400 G per lane, right? The physical property of silicon doesn't easily lend itself to a very high speed.
Therefore, a lot of people trying to work on different tricks, different solutions to address Silicon Photonics. So far, we've not seen a very compelling manufacturable solutions, and therefore, people are talking about different materials, barium titanate or, lithium niobate, right? But these things do not have a proven supply chain, do not have a proven deployment, reliable track record, right? It will take time for them to develop into a solution that people can really count on, right? Therefore, we believe that the EML remain to be a very attractive solution at 400 G per lane. Again, I'm gonna do a little bit advertisement here. Please come into our bio booth and see it. This is the world's first 400 G EML demonstration in a pluggable module.
You know, Lumentum's pedigree in this regard has been lasting the last 25 years. Again, you don't achieve such a milestone, such a feat, overnight, right? The design and the process manufacturability all tied together is what enables us to be not only able to design something, but also be able to instantaneously ramp the product, right? We're confident that we'll be ready when 3.2 G era, you know, is arriving upon us. We'll be able to scale our 400 G EMLs to the scale that's needed to support our customers. For that reason, we continue to see the penetration of indium phosphide into the data centers increasing its market share, right? It's now very, very clear.
I think last time when we met here at OFC, there was still a lot of discussion about, oh, is multimode gonna survive, gonna last for a long time? Now what we see here is that our customers and customers' customers have put a major decision that they will holistically use single-mode fiber because single-mode fiber is much more scalable in terms of how much bandwidth we can put into single-mode fiber, and they want to have infrastructure that is scalable, sustainable going forward, right? For that reason, we see indium phosphide's market share continue to increase in those high-speed areas and also in the CW laser areas or UHP for CPO, applications like that indium phosphide will be the horse to bet on. Again, that's why Lumentum is investing so much into the indium phosphide world.
On the CPO transceivers, I want to add some colors to what Michael just talked about, right? Some of the customers are really driving to increase the bandwidth per fiber, right? We see people talk about a lot of high-radix, you know, connectivities, but some band customers are trying to drive up the bandwidth per fiber to scale their AI clusters. You know, our modules are designed very well to fit that application. Then definitely 1.6T with Tomahawk 6 switch now, you know, coming into the market, and we're now shifting our production over to 1.6T, which is challenging to do, which we're actually now early to market, and we are gonna ramp our production up, and that will carry a higher gross margin and better economics for us.
Finally, what we really want to talk about is that not only are we going to integrate our lasers into our 1.6T modules, but also silicon photonics that we use in some of those modules are also for internal. This is the reason why we understand the trade-off between silicon photonics and indium phosphide EML lasers. We understand trade-off very, very well. We're integrating our componentry also into our modules for two reasons. One is, of course, it give us a better gross margin. Also secondary, really important is it secures our supply. Today, people are coming, knocking on the door trying to buy modules also because you have your own lasers, and that makes a difference to them. OCS, this is super exciting.
You know, the Lumentum history of OCS starts in year 2000. Our first generation of so-called wavelength active switch used the MEMS technology that's fundamental to our OCS today. Through all these years, there have been a lot of products deployed in the field. Lumentum has always been a leader in WSS until today. There are lots and lots of these MEMS devices deployed in the field. Every single problem that anybody has ever seen on MEMS devices, we've seen it. We've solved it. We've overcome them, right? That 25 years of experiences, know-how, the processes controlled, manufacturability, and equally important is that our WSS scale has already exceeded the OCS scale our customers want us to ramp to. What does that mean? That means we know what it looks like.
This is a very, very important thing because before you know what it looks like, you don't know what to expect, what to anticipate, where the traps can be, where the yield shortfall can be, where the process, you know, problems can be. We have seen all that stuff already, right? Now for us it's really, okay, let's go scale the supply chain. Let's go scale the machines. Let's go scale the training. For us, we know what it looks like to scale the OCSs, and we have field data to prove how reliable the MEMS mirrors are in the field. There is a lot of very attractive features of OCS. You know, to summarize, I would say the MEMS-based OCS is as transparent as you can get from an optical box point of view, right?
What do you want out of an optical switch? Really is you want to switch the traffic without paying any bit of optical penalty. That's what you want, right? I think MEMS mirror comes closest to that. Our insertion loss is very, very low. For a large 300 by 300 OCS, it's less than 1.5 dB. Our return loss, where the light doesn't come back, it goes one direction, it doesn't come back, very important for certain applications, is extremely low. It's scalable, right? It's transparent to wavelengths, data rate. Nothing bothers it. It's the same switch. Another thing, latency. Unlike a packet switch, which depending on the route it takes, congestion or not, this switch has a fixed and low latency, right? In the tens of nanoseconds.
Basically, how much time the light takes to travel, you know, between those two mirrors and in and out. Power consumption. This is passive, well, kind of passive. There is no transceivers on the switch. Light goes in, got switched, light comes out, right? Power consumption of this switch, number one, it's fixed. It doesn't scale with data rate. Number two, it's less than one-tenth of that of packet switch. As packet switch goes higher and higher speed, that difference is gonna become bigger and bigger. It's proven reliability. We know how to make it. Furthermore, you can address wide range of radix from very small to very large, right? The MEMS design and optic design is all scalable, right?
Again, we see a lot of opportunities for OCS, and we believe that our technology strengths built on 20 years of evolution, experiences, and manufacturing scale will enable us to be able to capture this market pretty uniquely in this market space. Then we think that in a few very important applications, right. Today, we all know one of the biggest application is particular approach to optical scale-up, right. In that case, what we do see is a 1.5- 1 attach rate, right. Meaning one of these XPU will require 1.5x of the optical port for that application, right. That is really to enable certain topology, certain connectivity, to scale the AI. Second thing here, we call it optical scale-up.
Now, this depends a lot on how the cluster is designed, how much traffic is coming out of this. There's a wide range. Our estimation is that the port to, you know, XPU attach rate is roughly 2-to-1 to 10-to-1. Actually, some numbers are going even higher. If you go with pure DR optics, this can be 20-to-1, right? It depends on how much bandwidth you put per fiber, and that will change the OCS demand drastically, right? So again, as we see more optical scale up, we're gonna get a better visibility and better certainty on this port attach number, but it's high. It's a lot higher than today. Finally, at a spine, super spine scale across level, it's roughly between 0.2-to-1 and 1-to-1.
That we see today, right, in other applications. Overall, again, we see a wide range of OCS use scenarios, use cases, and we see a wide range of the attach rate and overall added together. Again, it goes from a very small base or a zero external market base to something very meaningful. It's all upside for Lumentum. For that reason, we talk about OCS ramp. We just talk about this multi-billion-dollar deal that we just made with a customer, and we now have visibility into the for the next couple years. This is all pre-XPU scale up. Then when we add all this together, I think this line, you know, has a multi-year runway.
We think this will complement the packet switches and is uniquely suitable for AI clusters and AI networking. With all this story all tied together, we're really bullish on the market, right? This year, roughly we have $18 billion TAM across all the products that we can cover. In five years, we think we're gonna go to $90 billion+. You're gonna see a very large growth of scale across huge growth of scale out from zero to nothing in scale up and huge ramp on OCS, right? If you hear our story about everything is from zero to something, right? The OCS is almost from zero to something. Scale up from zero to something.
Scale out already today, but with compute scaling like this, you need a huge amount of scale out still, right? Don't forget the blue bar there, which is scale across. There's one important rule to keep in mind. However much compute you put in data center, you need to have a proportional amount of networking to connect data centers. While data center is growing like crazy, the networking is also gonna go like crazy, right? Lumentum has a bunch of products that's well-suited in that market. I would say you look at this whole pie chart, whole bar chart, Lumentum's really well-positioned in all these different areas, scale across, scale out, scale up, and also in OCS.
We do see that increasingly in five years, we're gonna see about more than about 1/2 of the traffic will start to go into the scale up domain that optics can cover. A little bit advertisement. This thing actually matters. Want to spend some time on the demonstration we're gonna do at the booth. Please swing by to take a look. First of all, there are two demos that we're working with NVIDIA on. One is to use our UHP laser for so-called CPO by using an optical module. Okay. Each such laser carries four 200 G lanes, so each 1.6 module will take two UHP lasers. Okay. There's a module from NVIDIA in our booth using our laser inside. There's one demo.
Second demo, also from NVIDIA, is our 200G EML put inside eight of them for 1.6T modules, right? These are two demos that we'll work with NVIDIA on. The next one is what we just talked about. It's a 400 per lane, the world's first 400 per lane differential EML module, right? This is gonna be the next standard, we believe, for the 3.2T module generation. The next two things are for scaled up. There is, you know, VCSEL, potentially VCSEL-based scaled up applications for certain more of a, you know, interchip connectivity kind of application that we're demoing.
We're looking at a higher power of the lasers that can actually cover even more optical lanes or a WDM-based laser that can now enable the next generation of optical scale-up technologies, up to 800 G, 1.60 per fiber or more. Finally, on the last column, we also have two scale across, multi-rail scale across demonstrations. One is a coherent optical channel monitoring. Now we want to monitor all the rails at the same time. Second thing is called a digital gain equalizer. Again, it's multi-rail ready, right? Again, we're gonna show you technologies covers scale out, scale up, and scale across. These are new technologies that we will continue to drive our revenue going forward, and then enable us to capture increasing amount of TAM, the big bar that we just showed you on the previous slide.
Thank you very much. Hopefully, I've built up enough for Wajid to take over.
My gosh, the pressure. Oh my God.
No pressure.
No pressure. Okay. Good morning, everyone. Thanks again for joining today. It's great to see all of you. You know, many familiar faces from the analyst days that we've had over the last couple of years. You know, there's been a common theme among the analyst days and that is it's the company's view that there is a shortage of indium phosphide capacity. It was even before all of these orders that Michael and Wupen spoke about that we had a thesis around how we should be investing our capital, you know, in spite of the fact that we were actually, you know, hovering at around $300 million-$350 million a quarter of revenue and barely generating any type of operating profit.
We were going through that transformation, and we had a lot of conviction. Certainly Wupen Yuen convinced us a lot around you know what the CapEx needs for the company would be. You know last year when we got up we talked about you know $500 million a quarter and then eventually a $750 million quarter. You know one of the reasons we've been able to to beat those numbers is because we really leaned in on a lot of the CapEx decisions back in fiscal 2023 and fiscal 2024 and in fiscal 2025. You know that was without a lot of the orders that we have right now. You'll see in a slide or two our backlog is quite robust.
I think that would probably be the measured way to talk about it. Again, what you're seeing from us is we're thinking out to 2028, 2029 and 2030. The announcement that Michael made about, you know, the new facility in Greensboro and the investments that we're gonna make in that in order to enable another $5 billion of annual revenue. Those are chip margins. I think all of you are familiar with how chip margins look. That will really, again, provide another pivot for us, another leverage for us to improve the business model of the company. You know, we're not gonna get off that track.
We're gonna continue to invest on our front-end capacity to expand our indium phosphide capacity, both at our Rose Orchard fab, which is our San Jose fab. At our fab in Caswell, we're gonna be expanding to, you know, put in more clean room space, and then at our new Greensboro fab as well. As well as in Sagamihara, as we move around the mix of our lasers, especially with 200 G and like Wupen Yuen spoke about 400 G really ramping up, we're gonna have to make investments there as well. We're gonna stick with that strategy. That strategy has been working for us, and we've been able to execute on the financial numbers that we spoke about last year, being able to beat expectations.
I think the other thing that's worked very well for us is very early on when Michael joined, he said, "Hey guys, you're not gonna be able to hit these numbers unless we start investing in our CM partners." What you'll see is that we are investing across our CM partners really to provide us with scale. You know, we had an internal management debate, "Hey, do we give up a little bit of margin?" You know, the decision was made that no, we're gonna get more operational efficiency if we have a number of CM partners. We'll be able to scale with market demand. We have a number of people rowing in the boat with us.
You'll see we've made that decision, and that's worked very, very well for us, and, especially with the multi-billion dollar deal that both Wupen and Michael spoke about, that's gonna help us a lot. That scaled strategy on CM is gonna help us a lot as we look to execute to significantly bigger numbers. On our operating margins. Does anybody remember what we showed last year in terms of our operating margins? Anybody? Okay. We had talked about eventually getting to 20% operating margins. For fiscal Q3, that'll end in March, you know, we've guided to north of 30% operating margins.
A lot of that has happened because we were able to execute on DCI demand and EML demand and, you know, with the product mix really working in our favor on both of those product segments, we've been able to improve the gross margins of the company, and I think many of you are now modeling us into the low- to mid-40s% for gross margins, and we'll give you guys an update on that as well. Capital structure. Up until a couple of weeks ago, our net debt was quite high. You know, thanks very much to the generosity of NVIDIA and their management team, we've had an investment of $2 billion.
I remember when that wire transfer came through. It was a really interesting moment. I got an email from our chief accounting officer, and it had a picture, and I was like, "Wow, that's great." That was fantastic. Now we've given up a little bit of dilution. You know, we have 2.9 million more shares out there. As you're updating your models, you should certainly account for that. You know, that really gives us the ability, and we could have done it on our own. We had taken a look at this internally. Should we do it on our own or should we invest with someone?
Obviously, with the commercial agreement that we have with NVIDIA, that really kind of leaned over to the decision that, hey, let's invest alongside with a leading AI infrastructure partner and really strengthen our balance sheet at the same time. Probably about 1/2 of the $2 billion will be investing in strategic CapEx. Even though we've got a great CM strategy for back end, in the front end, we will probably be upping our CapEx investments over the next couple of years. Probably about 1/2 of that will go into strategic CapEx investments, but then the rest will really allow us to have firepower as we think about how we should vertically integrate from an M&A standpoint.
Obviously, we're gonna need some more working capital to achieve the numbers that I'm gonna show you in a couple of slides. Okay? The difference between last year and this year, this slide says it all. Last year. By the way, this is all gonna be on our website. Unless you're taking pictures of me, you should. I don't think you are. This will all be on the website, but you're welcome to take pictures. The difference between last year and this year is last year we had a forecast. We had Wupen Yuen coming into my office and Michael's office saying, "Hey, this is what customer X told me. This is what customer Y told me." Now, you know, now it's real backlog.
It's real long-term orders. That really gives us the visibility and the confidence to make the CapEx decisions that we're making, right? Generally, we like to have our CapEx pay back within a year of it being operational. You know, being able to have this type of backlog from our customers certainly gives us a lot of visibility, allows us to communicate better with our supply chain, and I'll talk about our supply chain in a minute. And really invest what we need to from a business standpoint to make sure that we don't miss any opportunities. Because I think outside of the financials, you know, Michael's allowed me to sit in on more customer meetings than in the past.
Our customers are really counting on us, and I've seen that firsthand. We really can't disappoint them because there are even bigger customers that are counting on them that then count on us. This type of visibility really helps, and it allows us to, you know, make the right type of decisions from an inventory standpoint, from a headcount standpoint, and all of those other things that come along with making decisions under certainty rather than making decisions under uncertainty. I talked about the CM manufacturers. Actually, Michael and I had dinner with one of our leading CM manufacturers last night, and we were talking about how we could, you know, continue to expand our business with them.
You know, at the pace that we're growing, you know, we need to have multiple sources of supply from a CM standpoint and really be able to take those steps within the manufacturing process that don't have sensitive IP, right? The parts of the manufacturing process that have sensitive IP, our new leader in Thailand, KW, has got a great pedigree from Jabil, and he's able to really work with the CM manufacturers and say, "Okay, well, this part of the production process I'll take, and then this part of the production process you should take." And that has really helped us from a scale standpoint, in order to be able to virtually manage the type of output that we need to. Because, you know, last year at this time, we were delivering $400 million of revenue. We've been able to double in a 12-month period.
This is one of the big reasons we've been able to do that. We expect to continue to grow sequentially as the year passes on during calendar 2026 and in calendar 2027. This will be a critical enabler for us, in order for us to achieve our customers' expectations that they're setting on us. All right. Probably no surprises here, all right? Our current quarter we've set at a midpoint of $805 million. Future quarter, we think that we can grow to $2 billion. This $2 billion quarter is without the Greensboro fab that Michael spoke about earlier. This is the EML capacity that's increasing year over year by over 50%, and that's unit capacity. That's not revenue capacity, right?
The mix is going to shift more to 200 G as the quarters pass on. That certainly provides us with a nice tailwind. This includes the OCS win that we just got, the multi-billion dollar agreement with a large hyperscaler customer. This includes CPO as well, and you know what we're going to be shipping throughout calendar 2027 based on the commitments that we have from our customers, as well as an uptick in 1.6 T transceiver demand. The 1.6 T transceiver demand, like Michael spoke about earlier, the profitability of that business and that product line has improved quite a bit. As that business is growing, earlier our position was, okay, we want to limit the growth, and we've softened a little bit.
We're still being selective, and we are only taking on business that hits certain thresholds from a margin standpoint because we'd like to manage the overall business model, but that will certainly be a contributor of growth. The current quarter, those three areas, OCS, CPO, and transceivers, that's about 25% of our company revenue. Fast-forward a number of quarters from now, and we'll talk about the timing in a minute, we expect that those three products will drive 60% of the $2 billion number that we have. You're probably all wondering where EML is. Well, EML is in the balance, right? There's a little bit of natural hedge in the model, when you take a look at where we are now and where we move to at the $2 billion quarter. Hopefully no surprises here. All right.
You take a look at, okay, how much is each one of the growth drivers contributing? You can see it's pretty balanced, right? I mean, there's a little bit more OCS than others, but on balance, it's fairly even from a growth standpoint. Now, the timing will be a little bit different because you'll see that our gross margins over the next number of quarters will go up and down. You'll be like, "Hey, well, you just hit the top end of your model this quarter. Why are you going down, you know, a little bit next quarter? And then why are you overachieving?" A lot of it will depend on product mix, right? The first thing that'll come out is 1.6 T transceivers. We'll start shipping those in the June quarter. That'll start ramping up. Then Michael talked about OCS.
Well, that'll layer in next, right? That has a different margin profile than our cloud transceivers business. That'll give us a little bit of a tailwind. When we really ramp in the December quarter and then in the March quarter next year with CPO, that'll provide another tailwind again, right? It'll be a little bit lumpy, but that's really how we're thinking about it, and that's how we'd like all of you to think about it. There's the bar to the right, to the $2 billion, right? I think this is really important for all of us to understand because we are thinking about the company to the right of the $2 billion, right? How do we go and achieve those opportunities?
Many of the capital deployment decisions we're making now, like the one we announced this morning on the Greensboro fab, are around the box to the right of the $2 billion. You know, many of the R&D decisions that we're going to make are going to be focused in on that. You can see that there's a wide variety of opportunity for us to continue to grow and to really match the market data that Wupen showed in his slides, right? He showed, I can't even remember what the number was, but, you know, significant uptick in market growth. Well, how do we go ahead and capture that market growth?
Well, it'll be with the product, the diversified product lines that we have that are on the box to the right, and we'll be happy to answer more questions about that in the Q&A. But hopefully, those, none of those are any surprises. Okay, so the model. We wanted to make sure, just like last year, you know, we had talked about a $500 million quarter, and then we had talked about a $600 million quarter, and we talked about a $750 million quarter, and we had put some timelines around that. We wanted to make sure that all of you had some perspective around what our very short-term targets are, which is the middle column, and then what our midterm targets are.
You know, in the very short term, we expect that we'll be able to hit $1.25 billion a quarter with operating margins that hover around 35%. You know, we just guided to just above 30% for this March quarter, so I'm sure everybody can kind of draw a straight line and figure out how this could also happen. When we move to our $2 billion quarter with all of the capacity ramps that we've got going on, we think that we can actually run the company at a 40% non-GAAP operating margin. This is not EBITDA. This is operating margin.
Now obviously our share count will be a little bit higher like I spoke about earlier, but from an operating standpoint, outside of capital structure, this is where we think the operating margins can be. From a timeline standpoint, and we talked a lot about this internally, if you take a look at what we internally believe and what our customers are asking for and what we're ramping up our own internal capacity for, and you were to say, "Okay, how long would it take for us to achieve each one of these targets from a run rate standpoint?" We would say nine months and nine months. Nine months from now we think we can exit at $1.25 billion, and then nine months after that we think we can exit at $2 billion.
However, I'm sure that everybody's heard about all the supply chain tightness that we've got going on in the market, right now. We think it is prudent and measured to add three months in the range to each one of those targets. The first target, it'll be nine months-12 months for the $1.25 billion, and then the second target, after we achieve the $1.25 billion, another nine months-12 months, really with the delta being any type of supply chain surprises or supply chain tightness that we see across all of our product lines. Now obviously the internal team is doing everything possible to make sure that we beat even the low end of those targets.
Since we're communicating out something that is, you know, 18 months-24 months from now, we felt that it was prudent to have a bit of a range on that. Again, the capacity for Greensboro is not included in any of these numbers that we're talking about here. Yet the multi-billion-dollar agreement that we have with Nvidia requires that Greensboro fab. None of that potential revenue and capacity is in any of these numbers, and the reason for that is because we're not talking about calendar 2028 yet. Okay. I think I'm running out of time. All right. I think I've talked about this before. We're gonna keep to the same strategy. We're gonna continue to invest in the right R&D programs with the right customers and the right CapEx.
We've got a great balance sheet now to help enable and really lean in on some of those decisions. We think that we can drive the company to another 1,000 basis points operating margin improvement from the 30% that we just guided for the March quarter. We think that we can exit at $2 billion and 40% operating margins anywhere from 18 months-24 months from now. Okay. All right.
Okay, Kathy, you're the MC.
Yeah. Go.
You don't need that? No more clicker?
Yeah. I'd just like to announce that we have now published the slides on our website. Many of you have asked me for the slides, so they're on our website as of right now. So if you have a question, I would encourage you to step up to the microphone. There's one microphone in the aisle here. Then please state your name and your firm and we'll just take your questions. Thanks, Samik.
Yeah. Thank you. Maybe Wajid and Wupen, for you the first one is, you're highlighting sort of the range on the years that your targets are based on the supply chain constraints, but most of the new growth opportunities that you're also highlighting seem to be a bit more under your control going forward. I'm just wondering, when we think about the growth drivers, how much really seems to be supply chain constrained given most seem to be a bit more in your control? And secondly, for Wupen , more specifically, you mentioned Silicon Photonics doesn't scale to the 400 G lane. Are you doing something internally on Lithium Niobate already? Thank you.
Wanna get started?
No, go ahead.
Oh, okay.
Yeah, the revenue first, yeah.
You know, like I said, Samik, you know, what we have visibility to and, you know, both from a capacity standpoint and from a supply chain standpoint, enable us to say, "Okay, look, we think that the targets are nine months from now, nine months and nine months," but we do see surprises in the market, right? You know, we'll see a certain supplier come short, and the problem is, you know, if even one screw is missing, you can't ship the final product. You know, because we are setting numbers that are for-
You know, call it the end of this calendar year and effectively the end of next calendar year. You know, we're putting in that three months because there is just a lot of volatility in the market, and there are things that come up, you know, on a weekly basis that we are able to work through generally, but we want to be measured in terms of, you know, the type of range we give on our targets. That's really all it is.
I would say a couple of things, right? One, OCS definitely is a significant driver over that nine-month period. That does rely on supply chain. So cloud transceivers are gonna be in there, where that's actually growing really fast for us. That relies some on supply chain. So two of the elements do. I think on scale-up and scale-out, less so, right? We've got a lot of that under our control. Obviously, as we think about the margins and the mix, those two elements are super important. Two of the four are definitely areas where we see supply chain pressure.
Yeah. On the indium phosphide or the, you know, silicon photonics question, we definitely also, like everybody else, we're looking at different kind of material combination. We definitely think that indium phosphide, especially with all the investment the industry is going in, has a very strong potential to be the so-called material of choice at photonic plane. We think that actually complements today's silicon photonics really, really well. That's all of it.
Thanks. Karl?
Hi, Karl Ackerman from BNP Paribas. Wupen Yuen, you showed how VCSEL could be used for scale-up CPO. Yet VCSEL, I think, would be just 9% of laser demand, despite how
Right.
You showed later that scale-up would represent maybe one-third of that $90 billion TAM. I guess why wouldn't VCSEL address a larger part of scale-up versus other technologies, such as EML? Could you discuss your investment priorities to address VCSEL if it does become a larger TAM opportunity? Thank you.
That's a great question. I think at this point right now, our view is that the customers want to use single-mode fiber to scale for scale up, right? But you know, there is a small group of customers say, "Hey, I want to still leave the options open." Right? What if the scale up scaling of scale-up is more challenging than what we thought it would be? That really is kind of leaving that as an option, right? Second thing is actually, if you're looking at the VCSEL ecosystem, it's well established, right? Therefore, scaling of that laser and chips is not much of an issue, right? That's why people still say, "Hey, let me just keep that as an alternative solution.
Let's see how it goes," right? That's what we're still investing in. We want to make sure, frankly, we got all the scale up bases covered.
Karl, I'd say it's a timing issue too. I think certainly for probably the next three to four years, it's all gonna be indium phosphide. I think what's being discussed is beyond that for very short ranges, and that's where VCSELs work, right, are the very, very short ranges. If you believe in this thesis that we're putting out there, that there's gonna be a tremendous amount of optical scale up, you are gonna get into a world where you're gonna have short runs. In those short runs, VCSELs make some sense. I think that's where we see applicability. It's certainly not in the next four years. It's a 20, 30 and beyond type of thing, where you really start to see this high penetration of in-rack scale up. Thanks, Karl.
Mm-hmm. Thanks a lot. Simon Leopold, Raymond James. You gave us a lot of information today, but the one thing I didn't hear much about was really wide area network. I'm just wondering if you could frame your thoughts of your opportunities for DCI, particularly scale across. Number of the OEMs that have talked today about amplifiers being particularly tight. We know you do 980 pumps, but when I looked at the chart you put up, DCI looked kind of small relative to everything else. Could you maybe elaborate on this opportunity set for you?
I mean, we have the three of us sat in a meeting this morning that was exactly the content you described. We have a big share in pump lasers, as you know, and we intend to build on that. We're actually investing a lot in pump lasers. It's just on a relative basis, it's smaller. I mean, in the end of the day, the numbers that we're looking at for these other opportunities are huge. Doesn't mean we're not investing there. It's an important part of the portfolio. We've talked about ITLAs, we've talked about pump lasers, we've talked about WSS. All three, I think Wupen Yuen and the team are investing in. Again, just in a relative order of magnitude perspective, it's smaller.
It looks on the chart like it stays small, but it roughly doubles during that period.
Triple.
Yeah, I think it's.
Next three years.
Yeah. We're actually upping our output in some of these areas.
A lot.
5x. Our output on pumps are super important to us, and we're actually upping the output 5x. We're spending a ton in the capacity there. Again, just a relative order of magnitude, Simon, right? It's just smaller.
Great. Thank you.
No less important.
Yeah.
Hi, guys. Thanks for hosting the day, and congrats on that long-term model. And congrats on the Greensboro purchase as well. That'll probably be my questions. Any other details on the economics there would be great. I guess maybe more importantly, is this going to be for three-inch, four-inch or six-inch? And then you suggested ultra-high-power for ELS modules, but this will, I assume, do all of the EML additional capacity you would need, as well there.
Yeah. Chris, super good. First, I wanna thank you for rescuing this meeting this morning, right? I would have been toast without Chris. I was in the wrong hotel, apparently, like going up and down the elevator, but Chris saved me. Thank you, Chris.
Hear it out. Hear it out.
Look, I mean, I think that as we think about Greensboro, it's going to be very much focused on UHP. Consistent with what we've said, it's a four-inch, six-inch compatible line, right? We are thinking about starting on that particular line on six-inch, but that's a TBD decision. We have the ability to flex both ways, and as we start continuing our experimentation on six-inch in Caswell and in San Jose, we will look to make a decision on that. We will have the capability to swing both ways.
You know, what we've said and very consistent with is San Jose, UHP, Japan is going to be EML, and then as we communicated, I think to you last time, we're thinking about now Caswell, because the UHP numbers are so very big, shifting that which was originally intended for EML and now thinking more about it for the UHP. We're going to have three blocks now of UHP. San Jose, phase I, Caswell, phase II, Greensboro, phase iii. That's kind of how we're thinking about it.
Excellent. Online in 2028, is that?
What's that?
The Greensboro.
2028. We should be shipping in 2028. We'll be qualifying next year, right? Again, we have a nice fast run. It's already fully staffed, which is great. It's a fully operating fab, so it's just a matter of switching out the Qorvo stuff and putting in ours. That first phase will take a year just to get to the samples and then probably another year for qualification and production.
Thanks so much, guys.
Thanks, Chris.
Thank you, Chris.
Hi. This is Papa Silva from Citi. Wupen Yuen, I think you had a very kind of helpful slide around kind of putting everything together in terms of time. I was wondering, I guess, for your largest, I guess, CPO customer or in general, if you can kind of talk to the content per GPU. I guess, let's say if you take that customer, how should we think about it for Blackwell, and we move to Rubin Ultra, and Feynman? Just kind of any color on dollar content. Yeah.
Yeah, that's a complicated question. Let me think this way. If we were to implement, right? Today, let's say in the Blackwell generation, you have today's all scale out, right? Just assume that you go all CPO, all scaled out, right? You go from Blackwell over to the Rubin generation, there is a increment. I think, the scale-out bandwidth is becoming bigger, right? That's probably a, call it, a 2x increase, right? Of just the scale out based on CPO, right? Per GPU, right? Now, when you go from that to a inter-rack, scale out, that's probably scale up. That's gonna be another probably 2x-3x. That's why we put a 3x-4x, right? Think about Blackwell to Rubin, roughly 2x.
From Rubin to Rubin Ultra or something like that, the scale up happens. That'll be another 2x-3x. That's kind of how you think about from a UHP content point of view.
I don't know, Papa, that we've actually characterized how much per GPU. We get asked that all the time, right? You and others have said, "Hey, how do we measure it per GPU?" It's tough for us to measure that, right? What we're looking at is what's the demand on us, what's the backlog that's been placed on us, and then trying to give scales relative to what we see for scale out, initial phase of scale up, and then the phase II of scale up.
Got it. That makes sense. If I have a quick follow-up. On that timeline graph, you had a scale-up CPO scale-up optics timeline, and then you had the scale-out. I assume a lot of that scale-out, you have the transceivers in there. I'm curious. Let's say if you remove out the transceiver and we look at kind of optics from just kind of CPO, is scale-up by 2030 bigger than scale-out? Or at what point should we expect scale-up to start?
I don't have that number. You know, maybe I need to calculate that number for the model.
Yeah. I mean, you're right.
Yeah.
When I saw that graph, I thought the same thing you did.
Yeah.
I think that the reality is scale up is much higher. It's a timeframe. I think Kathryn put 20-30 on the chart. In that timeframe, if you just look at UHP should be significantly higher for scale up than scale out.
Right.
What you see in the mix is transceivers, right? That's what's sort of
Right. Scale out.
Shifting the number a bit. Yeah.
Yeah. Very helpful. Thank you.
Thanks, Papa.
Thanks, Papa.
Hey, George Notter, Wolfe Research. I guess my question is just qualitatively, what's embedded here on the $2 billion number in terms of, the trade-off between copper and optical in the scale-up domain, right? There's a copper distance radius associated with racks, and I guess I'm just curious how you kind of foot your numbers with that copper radius and, you know, what's embedded in how you think about your TAM and your sizing.
Yeah. I mean, George, I think what we're saying is sort of in the H2 of 2027, right, we'll start shipping scale up, and the scale up is gonna be coming sort of switch to switch. You're gonna see not in the, in the back plane itself. It's gonna be a lot of connections inside the cluster, right? You know, to Papa's question, we are sizing that from a magnitude standpoint at 3x-4x what we're seeing in scale out, but it's not going inside the back plane. What we've said is, at the end of 2028, we're gonna start to see connections inside the back plane. That's where you start to see a really significant leg up on CPO. It's not to say that first leg is small.
It's pretty significant, and we're gonna have to ramp significantly to get there. To be clear, it's more of a cross-cluster type of connection.
The other one I had was just on the NVIDIA relationship. Obviously they're gonna become a very large customer, I think, for you guys, as I kind of play this forward. Like, how do you walk the balance between NVIDIA soaking up your indium phosphide capacity and, you know, other customers that also want that capacity? I expect you're gonna say this is not an exclusive arrangement, but again, it probably creates a lot of paranoia, I think, among the other customers. What's the view?
Yeah, I mean, it's to a certain extent been a good paranoia, right? Because I think people can sort of size it and say, "Wow, that's a big chunk of the capacity." What it's enabled Wajid and Wupen to go out and do is negotiate with the rest of the guys on very favorable terms, I'd say. Yes, it has created a little bit of a feeding frenzy. Yes, I think that's actually gonna end up being a positive for us. It is non-exclusive, and we intend to ramp this capacity pretty significantly to serve.
I forget, I think it was Wupen that talked about this sort of discussion he had with a customer looking for 1 billion, literally 1 billion CW lasers that's outside our current ecosystem that's coming in and saying, "We need as much of this capacity as we can get." It is great to be in the driver's seat. I mean, there's no question about it. We really are kind of hands on the wheel, and these two guys are out negotiating with our... There's our sales leader in the back, John, these different capacity arrangements, and we expect to be able to talk to those over the next couple of quarters.
Thanks, George.
Yeah. Again, this is Vijay from Mizuho. Thanks for hosting the day today, Michael and Wajid and Kathryn. I want to go back to Wupen. Obviously a very interesting $90 billion chart, I guess. Getting a lot of questions on that. When you look at that scale up and scale out that you showed, how much of that is the transceiver versus SiPho in that? You mentioned as you go to CPO, you have some challenges. Might be slower, might be there's more transceiver. What is the mix within scale up and scale out as you go out to 2030? On OCS also, when you look at the OCS slice that you showed, is that just one customer? Is that multiple customers? Do you see that in scale up, scale out? If you can kind of give us a little color.
Yeah. Let me answer the OCS question first.
Yeah.
as the easier one.
Got it.
The OCS really is multiple customer, multiple applications, right? That bar is big enough to be, you know, higher than anything that we know today, right? It definitely is multiple scenarios, multiple customers, that's for sure.
Mm-hmm.
Right? On the scale.
I mean, let me just add to that, right?
Yeah.
I think, Vijay, look, we're shipping to three guys today. We've been clear on that. This new incremental order is one of those three, right? To be clear, we expect to add customers. I mean, the use cases that Wupen talked about are varied. We talked about the spine switch, we talked about sort of a scale up type of approach, and then we talked about sort of a super scale up going inside the rack and getting to that high ratio of attach. The OCS opportunity for us is we think is the most exciting sort of medium term, right? As you go out, it's certainly the scale up. It's three customers we're shipping to. We expect to add more. One of those three definitely leaned in with this multi-billion dollar incremental order.
Got it.
Yeah, your first question, right? I think in that scale out bar, like Michael said, right? There's a lot, a bunch of transceivers in there, both EML-based transceivers and SiPho-based transceivers.
Yeah.
Right? Certainly 400 G per lane will be in there too. In the scale up bar primarily is more of the, you know, UHPs and ELS, right? Certainly the content, you know, here, the volume here is probably more similar, right, than the trans size because of the how much contents, module contents in there, in the scale out area.
Got it. Okay.
Yeah.
Wajid, just a quick question for you on the convert side. You have some converts out there, I guess, 28, 29. Any plans on that?
Did a bank send you?
I think I had some questions.
I mean, like I said earlier, we're very comfortable with our capital structure now, not only because of the NVIDIA investment, but the operating profits that we're generating now is really giving us a lot of flexibility.
Yeah.
It doesn't mean that we won't get to go and take a look at the maturity and what makes sense and, you know, we'll take a look at the pricing versus fair value and all that type of stuff. But there's no pressure. I guess that's the way I would position it. There's no pressure.
Awesome. Fantastic. Thanks a lot.
Thanks, Vijay.
Thanks, Vijay.
Thank you.
Yeah.
Meta Marshall from Morgan Stanley. I guess just two questions. Any details that you can give around the new OCS order? Just is it scale up? Is it spine? Is it 300 by 300? You know, just any details there. Just a second question. You know, you guys have been using Caswell as you spoke to kind of aid with EML capacity. Just what is that transition timeline to kind of move that over to UHP and then just how does that impact kind of the excess capacity you've kind of been able to deliver on the EML side?
Yeah. Thanks, Meta. You know, on OCS, we haven't said who the new customer is. It's definitely one of the three that we're currently shipping to, and they have seen our product and really believe in it to the extent that they're willing to commit quite a bit more backlog than they had previously. It is a 300 by 300, so it's a high radix where we think we have quite a bit of differentiation, and we expect to build on that. I mean, I think we're right now, if you look at our opportunities, again, sort of intermediate term, this is one that we're super excited about. We're growing our customer base.
We're really leaning into that opportunity, and we feel like it's something where we can really add to the revenue profile in late 2026 and 2027. With respect to Caswell, right, you're right. I think that we initially had targeted Caswell as EML overflow. We are looking at it now saying, "Look, this UHP opportunity is so significant, we got to bring the extra capacity online for scale out initially and then scale up as a phase II." We think that can kind of carry us through to when Greensboro comes online. What does it mean? It means that we're actually gonna rely more on Takao, the second Japanese fab, which is still very underutilized to bring up additional EML capacity. The basic strategy is now concentrate EMLs in Japan, right?
It takes away a little bit of the diversity that we wanted to have initially, but it helps us get Takao up and running and then concentrate first San Jose, then Caswell, then Greensboro on UHP.
Thanks, Meta.
Ryan Koontz with Needham. Thanks for hosting today.
This is the best for last, Ryan. Is that true?
Great.
Okay.
Super.
All right.
Maybe reflecting on your experience with the 800G transceivers, some of the delays you saw, what gives you the confidence in 1.6T, where are you in qualifications, maybe any color you can share with us that gives you confidence in that quick ramp. That seems like a fast emerging market. Thanks.
Yeah, yeah, look, I mean, at 800 G, we acquired Cloud Light sort of in the middle of 800 G. Cloud Light was never sort of first to market. They were always, you know, a captive Google house, right? That's always very clear, and we were always sort of later. It was Silicon Photonics. You know, if you look at what Wupen Yuen described quite well, most of the first designs are gonna be EML. They're easier, they're more performance-driven, and we have been pigeonholed as a Silicon Photonics provider, so that means we're typically later in the curve. When we took over Cloud Light, we actually didn't do particularly well.
We got later and later in the design cycle, our designs were well off the mark relative to timing with Google and other customers, and so we started falling behind. Now, we've reconstituted that whole Cloud Light business. We rolled it up under Wupen Yuen. He's spending a tremendous amount of time on it, and the last couple of 800 G opportunities, we were first. We were actually at the head of the curve. Even with silicon photonics, we were able to out execute the EML guys. Based on what we know on 1.6 T, again, we're first to deliver samples. Now, that's only 1/2 the battle, right? We have to get our production working as well. You have to go from samples, which we've definitely created a tremendous amount of excitement in terms of our ability to execute on the samples.
We have to then roll that into production, which is still choppy. I mean, if I'm being honest, we're not perfect in our production, and we've got to get better in terms of getting the margins up and getting our production right. It definitely looks like we're catching the early part of the curve, which is super important from a margin and revenue standpoint. Then provided we can execute on the back end, we should be in pretty good shape.
Great. Maybe a quick follow-up, if I could, on indium phosphide expansion. I know there's long lead times for those reactors. You guys feel comfortable with the, your supply there?
Yeah, I think we're in pretty good shape with the reactors. It is. You're 100% right. It's a long lead time item. It's probably not the thing that worries me in Greensboro or as we ramp up in San Jose, right? It's really around substrate supply. I mean, there is still supply tightness. I think we've got that resolved. Kathryn and I had talked to you about the seven-year agreement that we've been able to work out. We feel pretty good about where we're sitting there, but that's probably the thing that keeps me up at night most is.
Okay
... is substrates. Less so sort of reactors and getting the right tools into the fab, although it's not a zero challenge.
Thank you.
Thanks, Ryan.
Thanks so much, Ryan. With that's a wrap. I look forward to seeing all of you at our booth, which is number 1439. It's smack dab in the middle. Biggest booth on the show floor. You know, not that we're competitive at all. Yeah. Look forward to seeing you there. I have predetermined, like, office hours with our CTO, Matt Sysak, at the booth. You know if you have an appointment with us. If not, feel free to stop by at the booth, especially today, I think is the least crazy day. I appreciate all of you attending and thanks so much.
Thank you.
Thank you.