Good morning. Good afternoon, I should say. Thanks for bearing with us, and welcome to Coherent's 2026 OFC Technology Innovation Briefing. I wanna thank all of you here in the room with us, as well as all of you that's online. We greatly appreciate the interest. Before we begin, I've gotta refer you to the customary forward-looking disclosure statements on slide 2 in the presentation deck. It will be posted to our website, in the investor relations section of our website later today. The full presentation, all of the disclosures today are covered by those disclosures. Any forward-looking statements are covered by the disclosures. They apply to any and all forward-looking statements that may be made. Now for the agenda. CEO Jim Anderson will provide an overview of our data center communications segment. Jim will be followed by Julie Eng, our CTO.
Julie's gonna discuss scale-out and scale-up. She'll be followed by Dr. Beck Mason, EVP of Semiconductor Devices. He'll discuss lasers and indium phosphide capacity, and he'll be followed by Dr. Sanjai Parthasarathi, our Chief Marketing Officer, who will discuss scale-across. That'll take about 60 minutes. We're gonna follow that up with a Q&A session that'll be about 30 minutes long. Finally, Jim's gonna provide some closing remarks. Last but not least, I'd welcome any feedback from any of you following the event. Regardless, if I could be of any help at any point in the future, feedback or otherwise, it would be my pleasure. Without further ado, I want to welcome Jim to the stage.
All right. Thank you, everybody. Thanks for being with us here today. Really appreciate you being here, especially here in person. I know there's a lot of things going on at OFC, a lot of demands on your time, so thanks for spending time with me and the rest of the team. Thanks for everybody that's joining us online as well. We're here in Los Angeles at OFC. I was walking the floor this morning, and this has gotta be a record OFC in terms of size and number of people. I mean, each OFC keeps getting bigger and bigger, and I think it's really a testament to how important photonics, optics continues to become across so many different applications. Coherent always has a big presence at OFC.
We've got a lot of products and technology that we're demonstrating here at OFC, and happy to give you an overview of that today. By the way, the beginning of OFC, the first keynote kickoff this morning was by our own CTO, our illustrious CTO, Julie Sheridan Eng. She's famous now, so she's nice enough to carve a little bit of time out for us for our event here. Anyway, thanks again for being with us here today. I think there's probably never been a better time to be in the photonics industry than right now. The photonics industry, optics, the innovation that we're bringing is touching so many different applications across so many different markets, data center communications, industrial applications.
Today we're gonna focus on data center, but it's incredible to see the reach that photonics is having across so many different industries. In data center, which again, that's the focus of today's discussion, if you look at data center today across the scale-out and scale-across networks, already 100% optical, 100% optical networking and growing at an incredibly rapid rate, as we continue to add compute capacity across the network. But what we're really excited about is kinda the next frontier of bringing optics to scale-up part of the network and to help drive, you know, future data center innovation with photonics. Now, the reason that we're now starting to see photonics migrate into the scale-up network is the same reason that photonics took over scale-out and scale-across.
The reason is that, you take any length of distance in the data center, whether it's three meters or three kilometers, and if you crank up the data rate, and if you crank up the total amount of bandwidth, eventually you hit a point where photonics is the most power efficient, fastest way to transmit that data across that length of distance. Now what we're starting to see is to continue to drive data center architecture, now adoption of photonics in scale up. That's great for the photonics, optics industry. That's a massive TAM expansion for the industry, but really good for, Coherent as well. When we look at Coherent, again, I don't think Coherent has ever been in a stronger, better position than we are today.
We're gonna talk about today, the opportunity in front of us. Kinda think about it in two parts. Part one, we'll talk about all of the existing engines of growth that are driving our growth today. If you look out through the coming years, that's about a $50 billion market opportunity, probably a little bit more than that. That covers, just for a few examples, that covers pluggable transceivers, so we'll talk about our ramp of 1.6T. We'll also talk about the roadmap to 3.2, 6.4 and beyond. We'll talk about DCI transceivers, transport and transmission, and then all the different optical components that go into those products. Certainly talk about our existing growth engines.
What I'm really excited to talk about today, and what you'll hear from the team as well, is the new growth engines. What's driving, we believe an inflection in our growth rate moving forward and acceleration of that growth moving forward, and there's four in particular that we wanna talk to you about today. If you look in aggregate across these four new growth engines, collectively, those add over $20 billion of incremental market opportunity over the coming years. These are not things that are way out in time. These are things that are starting to ramp over the next few coming quarters, and that's why we're really excited about that. Number one, we'll certainly talk about optical circuit switch, OCS, really excited about that. The market is much larger than we thought a year ago at OFC.
The number of usage models, the number of usage cases has grown considerably, and we are on a very fast ramp of ramping our manufacturing capacity to meet the demand that we're seeing in front of us. That's already ramping today. That's ramping each quarter throughout this year and into next year. The second big growth area for us is CPO and NPO. I've asked Julie to spend a good amount of her time talking about CPO and NPO today for both scale-up and scale-out. That is a huge incremental TAM for the optical industry and especially for Coherent, and this is near term. We'll start to see first revenue from CPO in the second half of this calendar year.
The ramp will start in scale-out, but towards the end of 2027, we expect to start to see the ramp of scale-up as well. The next area we want to talk to you about is Multi-Rail. This is a new product category. We've got tremendous excitement from customers in this product category. Again, this is near-term revenue. This will start to ramp in the first half of 2027. The fourth thing that we want to talk to you about today in terms of new growth engines is thermal solutions. Now, this is something we haven't talked a lot about to date, but given that it's starting to ramp next year, we want to start to talk to you more about this.
This is us using proprietary materials technology that Coherent has developed for industrial applications and repurposing that for data center applications. Not just thermal cooling, but also harvesting heat from the data center and recycling that into electricity. We'll talk about that as well. We're really excited about the new growth areas. We think that as these come on and they layer on top of each other, that provides us with an inflection and acceleration of our growth rate over the coming quarters. In aggregate, if you look at all these are all a tailwind to our gross margin and our business model. That's what we'll talk about over the course of today.
I do want to take a few minutes here at the beginning to talk about the partnership with NVIDIA that we announced, a couple weeks ago. We've had a long history with NVIDIA. Over 20 years, NVIDIA has been a customer of ours, but this is a step function expansion in that partnership, and we're really excited about it. I was at GTC yesterday at Jensen's keynote, where he talked about CPO in both scale-out and scale-up, and so great to hear them already talking about that. This partnership is entirely incremental to what we're doing with NVIDIA, today, and it's all around CPO for both scale-out and scale-up. The supply portion of this agreement, the development and supply portion, is a $ multi-billion-dollar, multiple-product opportunity, and it spans out through the end of this decade
The different products that go into this partnership with NVIDIA, not just the laser, but the other products as well, we'll talk about through the course of today. We're happy to answer any questions about this partnership when we get to the Q&A as well, but really exciting for us. You hear from me all the time. You hear from me at least once a quarter, probably more often than that. Since today is more geared around products and technology, I want you to hear from the experts in the products and technology. We got three sections that we'll go through. Number one, our CTO, Julie Eng, will talk us through scale-out and scale-up, both the existing growth engines as well as the new growth engines. She'll talk about pluggable transceivers and the roadmap there.
I've asked her to spend an extra amount of time on CPO and NPO, given the huge growth there. She'll talk about OCS and thermal. The second section will be led by Beck Mason. Beck leads our semiconductor products group, and I've asked him to do a bit of a deep dive into our laser portfolio and our indium phosphide capacity. This is an area that we get a lot of questions from investors, a lot of interest in this area. He'll talk about all of the different laser technology that we have, the EML, the CW, and the VCSEL lasers, how those are used in both pluggables as well as CPO, NPO. I've also asked him to spend some time on our indium phosphide capacity because we're ramping that very fast.
We're doubling that this year, and he's going to share with you our expectations of capacity expansion next year as well. Then finally, Sanjai will wrap up with scale across. He'll talk about DCI transceivers, transport transmission, but also that new Multi-Rail technology system that starts to ramp early next calendar year. If anybody needs a photonics doctor today, we've got you covered. We've got three doctors of photonics in the house today. Before I hand it off to the three doctors, I just want to reiterate, thanks for being with us here today. We're at an inflection point in an acceleration of our growth. It's really that combination of the expanded use of photonics in the data center, the broad range of technology that we're bringing to our customers. You'll definitely see that today.
The manufacturing expertise and the scale of manufacturing as well. With that, I'll hand it off to doctor number one, our CTO, Julie Eng.
Thanks, Jim.
Sure.
Thanks, Jim. Hi, good afternoon, everybody. Nice to see a lot of familiar faces, but for those of you, if I haven't met, I'm Julie Eng, I'm CTO of the company. I've been with the company over 20 years. I've been in optics my entire career. I want to echo what Jim said is, it's a very exciting time to be in optics. I mean, if you think about it, the AI data center is basically a massively distributed supercomputer, and when you do parallel processing, the performance of the system depends a lot more on interconnects, and we are the interconnect. Optics is now becoming a key architectural element in the architecture of the AI data center, and I think things like our partnership with NVIDIA show how that our customers see that as well.
I feel very happy and confident, not just about the market. The market and the opportunity are great, but I feel very confident about our roadmap and our depth and breadth of technology. Yeah, let's talk about the data center, starting first, defining our terms, the domains inside of the data center. The scale-up domain is the networking together of multiple XPUs or processors, accelerators to act like a single compute node. The scale-out domain is the networking together of those nodes to form that distributed supercomputer that is the AI data center. Talking first about the scale-up or scale-out, that's a heterogeneous environment. What I mean by that is you might have equipment from different people that have to plug and play.
The distances can be very different, 10 meters to the rollover or 10 kilometers up through the ceiling and across a warehouse data center. Flexibility is very important in this portion of the network, and this portion of the network is optical today, and it's all served by pluggable transceivers. The scale up is very different. This is a closed system, and what I mean by this is it's owned by the person who networks inside the rack or a small cluster of racks, and they don't have to operate with anything else, interoperate. It's copper today. The distances are less than 10 meters. Because it's in the rack, density, cost, and power consumption are very important. As we see this domain transition toward optical, it's a great fit for co-packaged optics.
Of course for us, since it's 100% copper today, this is a SAM expansion. In the flexible portion of the network where flexibility is prioritized to scale out, you see pluggable transceivers because they offer flexibility. There's a robust roadmap that we as an industry agree on of increasing bandwidth density, and I'll talk a little more about that. There's a standardized multi-vendor ecosystem which gives security of supply. They're easily serviceable because they plug in through the front faceplate of the switch, and they defer the architectural commitment.
What I mean by that is once you deploy the switch on the floor, because the optics plug into the switch, the data center operators can decide after the fact whether they want that switch to go 30 meters and use VCSEL-based transceivers or 500 meters and use silicon photonics-based transceivers or 10 kilometers and use silicon—I mean, indium phosphide-based transceivers. They can make that decision literally after the rack is on the floor, and they can change their mind after the fact if they want to. That's a lot of flexibility that's very valued by our customers. On the other hand, co-packaged optics for us is a SAM expansion. What is co-packaged optics? Co-packaged optics is really just like an LPO transceiver that's made smaller, and it sits right around the switch or XPU like that picture on the bottom there.
Why would you do that? The main reason you do that is because you don't have to drive the electrical traces to the front of the box, and that reduces power consumption. Since you're not pushing all the optics through the front of the box, that helps you increase faceplate density. But you give up something for that. Now everything is inside the box, and you lost that great flexibility that I just talked about. It's a trade-off where you need bandwidth, higher bandwidth density and power and better power consumption compared to if it's more important to you than flexibility, then you would do that. That's a really good fit for that scale-up portion of the network. Let's just talk about what is CPO really?
Really, CPO, I call it an architectural repartitioning of the optical transceiver. There's a picture of one of our transceivers up there. What you can see is that all the elements, first you see that column under pluggable transceivers with the green checks. That green check is everything you need, the significant optical components that you need inside the pluggable transceiver. Now, those green check marks also happen to be our capability. What that means is every critical optical component that's in a pluggable transceiver we design and manufacture. Now, the CPO I'm showing at the bottom, that's actually a rack unit that would plug into a rack, and where those white, little pluggable things are, that's the front of the rack.
We're looking at it from the top down, and what you see is there's the switcher XPU, and then, like say, the modulators, detectors, laser drivers, transimpedance amplifiers go in those little orange boxes all around the switcher XPU. In some architectures, you plug the laser in through the front. You might be familiar with the ELS term, and those are those white elements there. When you look at what's in CPO, you look all the way over to the right, and all those checkboxes are what's in CPO, and all those checkboxes are actually things we have internally. I'll show you a little bit more about that. We have a strong roadmap for pluggable transceivers. We're continuing to ramp 800G. We have a broad portfolio of 1.6T.
We have a full suite of transceivers, including all of the types of lasers, as well as TRO, FRO, and LRO. Here at the show floor, if you haven't been there yet, we're showing three different versions of 1.6T around three different DSP vendors. I'll talk to you a little bit more about 3.2 and 6.4 on the next slide. In order to increase bandwidth density of pluggable transceivers, we do two things. The first is we increase the data rate of every line, and the second is we try to fit more lanes inside the box. On the left-hand side, you'll see what we're doing for the industry to go up to the next speed node to drive the bandwidth density of pluggable transceivers.
We're showing on the show floor a 400G full-length demo with an indium phosphide EML. That's our own EML designed and manufactured in-house. We're also showing it with our own 400G photodiode, also designed and manufactured in-house. On the bottom, what you see is a 400G with silicon photonics. This would have our own CW laser, our designed silicon photonics fabbed in a silicon fab. That's actually the first time that anyone's shown a 400G with silicon photonics. We got what's called a post-deadline paper, which sounds like we just missed the deadline, but it's actually the late breaking news of the conference. That'll show on Thursday afternoon. That's like a very proud moment to have a post-deadline paper, and that 400G silicon photonics is one.
On the right-hand side is what we do to increase density. You might have heard of this new form factor. It's called XPO. Today, our OSFPs are 8 lanes of 100 gig or 200 gig. The XPO actually jumps to 64 lanes and partly does that by integrating water cooling into the module. This was announced by Arista and ourselves and a bunch of other partners supporting this form factor. It's literally just the next generation of pluggable form factor. The reason you would do this is it enables what was shown at the conference, essentially collapsing something that used to take 4 rack units. You can now collapse down to 1 rack unit. It's just higher bandwidth density.
The insides of the transceiver are roughly the same as they were before. These can support because it's got 64 lanes of 200 gig, it can support up to 12.8 terabits per second. We're also driving an innovation at the OIF, which is a standards body, called the high-density form factor, which does the same thing but ups the data rate to 400 G, which is 25.6 terabits per second. What you can see is there's a long, robust roadmap of pluggable transceivers, and, you know, we're a leader in that area. We're also showing the XPO live in our booth as well. I often get the question. Well, first, actually, I'm sure your eyes are going to go to the SAM, so let me talk about the SAM.
We are upping our SAM for CPO. Our new SAM for CPO, our estimate is $15 billion by 2030. I often get the question, you know, "What can you make for co-packaged optics?" I've tried to make it really explicit here. I'm showing you that same rack unit that would plug into a big rack. This is the top-down view, and in the front are those ELS, external laser sources. We could sell a silicon photonics PIC. We design silicon photonics PICs. We use them in our transceiver. You can use them in CPO. We package the silicon photonics PIC and/or VCSEL, actually, you could package into something called that little CPO or NPO module, or sometimes people call it an optical engine. That would be all the orange things around. That's something we sell into CPO and NPO.
The external laser source, which I'm sure you've all heard of, that has our CW laser in it, which Beck will talk about. We also, I think you know, make, design, and manufacture a significant fraction of the world's isolator. The isolator protects the laser from back reflection, which would otherwise hamper its performance. We also design and manufacture thermoelectric coolers, which keep the laser at a constant temperature. Either we could sell at any of those levels, the external laser source, CW laser, in isolator, thermoelectric cooler. Then on the other side, you need to get the light from that external laser source to the silicon photonics PIC. That takes a lot of passive optics, and that's something maybe we haven't talked about quite as much externally. They're usually inside of our transceivers.
In this case, what you see is you can use this, small passive. It's a very precisely manufactured and designed passive, actually, called a Prism Micro-Lens Array. It takes the light from the fiber, and it shoots down into the silicon photonics. It also requires something called polarization maintaining fiber, which we also design and manufacture internally. When you put those together in a fiber, something called a fiber attach unit, and we design and manufacture those internally as well. You can see the broad array of products that we can make for CPO. I want to show you our, what we're showing at OFC. This is what we're showing at OFC this year related to CPO and NPO. One comment I should make is NPO just means it's not on the package.
It's still inside the box, very close to the switcher XPU, and that's what NPO is. What we're showing, I think we have the broadest and deepest portfolio of CPO technologies, and I'll show you how here. On the floor, we're showing a 6.4 terabit per second silicon photonics-based socketed CPO. Another moving forward on CPO is looking at sockets. The advantage of a socket is that it makes it more easily serviceable. It means you could have different types of CPO that would all fit in the same, you know, location. There's a downside. It takes space, and there's some maybe signal integrity trade-offs there.
There's a lot of interest in socketed CPO, and so we, together with others, founded a socketed CPO MSA. It's called Open CPX. We just announced it last week. This is a silicon photonics-based socketed CPO demo. It's 6.4 Tbps, and it fits about on the size of the top of my thumb. If you think about that for a second, you've probably seen an OSFP, and it's the equivalent of 4 OSFPs, but now I'm putting it on my thumb with the exception of the external laser source. Super proud of that, very excited about that. With that, we are showing also the demo is with our own fiber attach unit, which has our own micro lens array and our own polarization maintaining fiber.
We're showing that with our ELS, which has our indium phosphide CW laser, our isolator, and our tech. It's end-to-end coherent solution. We're also showing VCSEL-based CPO and NPO, and the VCSEL actually is an interesting potential for silicon photonics because the power is very, very low. The energy efficiency is good. When you look at a VCSEL-based solution, it has a path to what we call one picojoule per bit, and maybe that doesn't mean that much to you. It's basically between four and five times lower power than the silicon photonics solution. But it doesn't go as far. It's a shorter reach, but there's a lot of short-reach distances inside the data center. In the VCSEL-based solution, you usually have the laser in it, and so there's some pluses and minuses of that.
I do think that these will coexist in CPO, NPO, just as they have in pluggable transceivers. I'm showing there in the picture our 2D Datacom array of VCSELs. VCSELs are usually made in a 1-dimensional array, but now Beck's team is making them in a 2-dimensional array. Next to that is a little NPO, near packaged optics, that's VCSEL-based that you can also see on the show floor. You can see how small it is next to that penny. Finally, we're showing, I believe we're the first to ever show, a CPO based on indium phosphide. This is 4 lanes of 400G indium phosphide Mach-Zehnder modulators integrated with semiconductor optical amplifiers. That's important because semiconductor optical amplifiers are very helpful in closing the 400G link budget.
Down there you see our 400G AI. Again, I think the depth and breadth of our CPO and NPO is unmatched actually in the industry. I'm very proud of that. I think that for reasons like this is why you see such a strong partnership with NVIDIA and people who are working, you know, to be the leaders in CPO and NPO deployment. Okay, changing gears now, staying inside the data center, but talking about optical circuit switch. We are also increasing our SAM for the optical circuit switch from $2 billion to $4 billion.
Jim alluded to. We feel like maybe we under-called it last year at this time, and I'll tell you why I think we under-called it, is that the use cases are actually broader than what, let's call it, I can say, speak for myself, than what I thought last year. The customer base is broader. I think the customer adoption is faster than we thought. Let me try to walk through those pieces. What really is the use case for optical circuit switch? When you think about it, I think the primary use case, there might be like a lot of sub-uses of it, but the primary use case is our customers can send a software command and then reconnect how all the fibers are connected inside anything that's touching the OCS, how all those fibers are reconnected.
One example that might matter to our customers is suppose they're doing like a huge training model one month, right? They want all the fibers connected in a certain way because they're gonna use almost all their GPUs for that one training model. Two months later, let's say, they wanna do a whole bunch of small training models. They might want the fibers connected all differently. What they can do is send a software command, and the fibers get connected differently. That's very important because XPUs are so expensive too. Being able to almost on a job level optimize the topology for the workflow, for the maximum utilization of those expensive GPUs is very, very valuable.
I think as our customers started to realize this and think of all the different ways they can use it, that's how we're seeing the use cases expand. One example, some of our customers have talked about having a hot swap rack, and if something happens in one rack, they literally use the OCS to fail over to the other rack to keep their uptime and not lose where they are in their training. We are actually seeing customer interest in scale up, scale out, the spine, and scale across. All different kind of locations and use cases. As Jim mentioned on our last earnings call, we've shipped to over 10 customers, and we are shipping into production deployments. There on the left, you see the 320 by 320 OCS system.
Just as a reminder, we do all of that software ourselves in-house, and we have a very skilled software design team that our customers are often complimenting us on how good our software is actually. Okay, just a reminder that our technology, we believe, is differentiated. We're using the liquid crystal technology. MEMS is kind of like a moving mirror. It can either move up and down or tilt to move the beams around. Anything that moves, usually you don't want something that moves inside the data center if you can help it. In addition, the liquid crystal takes less than 10 volts. In the liquid crystal, we're just turning molecules, so that's a lot easier. To turn those molecules is only 10 volts, whereas a MEMS system can sometimes be 100 volts, 200 volts.
Also, our customers have spoken publicly about how high voltage components fail more often than low voltage components. We believe we have differentiated reliability, which I think our customers would agree with us on. We're seeing strong sequential growth. We're ramping capacity to meet our growing demand. As Jim has mentioned on the earnings call, we're shipping 64 by 64 and 320 by 320 into multiple customers. We have a 512 by 512 in development, and I think you will be able to see that on the show floor. Finally, just wanted to end up with our thermal management. These thermal materials, they're in our industrial segment. Because in this case, we're using them toward the data center, we decided to present this to you today.
This is the first time we're putting a TAM on this, and we're putting $2 billion by 2030. Let me tell you what it is. These are very specialized thermal materials that today we sell into semicon equipment, for example, and other applications. They're finding a new home in the data center. Why is that? The reason that is these accelerators, XPUs, GPUs, they run at like a kilowatt, and getting the heat out of them is very important to their reliability and their performance. A couple of the materials that we have are very valuable for that. We have ThermaDite, which is our patented, our own name, and it's our own proprietary material. It's actually a diamond silicon carbide ceramic.
Since it's a ceramic, you can like fire it, you can machine it. We literally use our own industrial lasers to machine it. That helps wick the heat away, you know, efficiency of heat transfer, double that of copper. In addition, we're seeing interest in silicon carbide substrates because that can get the heat away about twice as fast as silicon, actually. Then finally, as Jim mentioned, we have some ways in which thermoelectric coolers that we use to keep a constant temperature in the laser can actually be used in reverse to take the waste temperature delta of your waste heat and turn it into energy, actually, with for energy reclamation in the data center.
These very, very interesting thermal materials, we have a lot of customer engagement, because it's just very valuable to our customers to be able to improve the reliability and the performance of the XPUs. With that's the data center for today. Thanks for your time, and I hope you see the excitement that I feel and the breadth and depth of our strong roadmap. Let me hand it off to my colleague, Beck, who will talk to you about lasers and indium phosphide capacity.
Thank you, Julie. Good afternoon. My name is Beck Mason, and I lead the semiconductor device business group at Coherent. I'll give you a little overview today on some of our laser technology, the breadth of our technology in semiconductor photonics, and also talk about what we're doing to scale up indium phosphide capacity. I've been in this semiconductor photonics business for a little over 27 years. I started my career as a device researcher at Bell Labs, and I can honestly say that there is no more exciting time in this field than right now. The pace of volume growth is unprecedented, and the pace of technology evolution is faster than it's ever been before. It's both exciting and challenging for us. Coherent is a very broad and deep technology leader in semiconductor photonics.
We have arguably one of the broadest portfolios of devices of any company in the world. We make very high-speed indium phosphide-based EML and DML lasers, and we have solutions that go up to 400 Gb/s. We've been shown to be a leader with the first company in the world to demonstrate differential EML technology, and I'll talk a little bit about that later. It's important 'cause it enhances overall link integrity for data center transceivers. We also have very high volume indium phosphide CW lasers. These are the devices that are used to power silicon photonic transceivers today that we sell across our data center customers. We have even higher power versions of those that we're developing and engaging with customers on which deliver 400 mW per laser.
These are the critical components that are enabling the external laser sources for CPO growth. Finally, we're also a VCSEL manufacturer, and we're one of the leading VCSEL manufacturers in the world. We have extensive gallium arsenide wafer fab capability, and we've developed VCSEL solutions that scale from hundreds of gigabits per second to 200 gigabits per second, and then in aggregate capacities in array-based solutions into the terabits per second. I'll show a little bit more on that technology today as well. To date, we've shipped over 1 billion VCSEL devices to the field in communications and over 500 million indium phosphide devices. If we count the non-com semiconductors we make, it's well over 3 billion units. Indium phosphide, I think most people have heard how important indium phosphide is to the scaling of data centers, and specifically AI data centers.
Currently, we are running 4 separate indium phosphide wafer fabs. Our first fab in Fremont, California, is where we've developed our high-power CW laser and high-speed photodetector technology. That's a 3-inch fab, and it's a fab where we've got more than 20 years of experience delivering devices into the data center market. Our next fab in Sherman, Texas, is the first fab where we've deployed 6-inch indium phosphide production. It is the most advanced indium phosphide fab in the world. It's a very large facility with capability to scale to meet the huge growth and demand we're seeing. It's the first place where 6-inch indium phosphide has been put into production, and we are making multiple different product platforms in that fab. It's also a very high-volume gallium arsenide fab for us, where we make our 3D sensing products.
The next fab is in Järfälla, Sweden, which is just north of Stockholm. This is also a fab we've had for more than 20 years. It's the home of our advanced EML technology. It's also where we make our most complicated photonic integrated circuits that are used in the scale-across applications. It's the second fab where we've installed our six-inch production capability and are ramping our capacity significantly. Finally, because of the unprecedented growth we've seen, we are adding a fourth fab to our indium phosphide group, and that's our wafer fab in Zurich, Switzerland. This is also a fab with a very long history. It's a very high-volume six-inch gallium arsenide fab today, and we're adding a substantial amount of six-inch indium phosphide capacity into that location as well to scale it up.
Overall, as I said, we're seeing unprecedented demand for indium phosphide devices, and we're seeing it across the full range of our products from many of our customers. We're in the process of locking up long-term strategic agreements with these customers to guarantee that they will have the capacity they need as they scale, and for us to guarantee that we will have secure supply to deliver to. We are focused really in Coherent on having the most advanced technology possible in our fabs, and we've done that by going to the largest wafer size. Going to 6-inch indium phosphide doesn't only help us with scale, it lets us deploy the most advanced semiconductor process equipment in the world into those lines. That gives us higher yield, better throughput, lower labor costs, a whole bunch of benefits.
In fact, we're currently running three main categories of devices on six-inch indium phosphide: EMLs, high-power CW lasers, and high-speed photodetectors. In all three of those categories, we're seeing higher yield and better throughput efficiency on our six-inch lines than we've been able to achieve even on our very mature three-inch production lines. As I said, we're ramping capacity in Texas and in Sweden and now adding the capacity in Switzerland, and we fully expect that we will double our total capacity by the end of this year, and we're planning to double again or more by the end of next year, and we're going to keep increasing after that. The scale of growth for us is very rapid and very fast, and we wouldn't be able to do it if we didn't have such an advanced platform to work on.
I'm going to switch gears now a little bit and talk about innovation because innovation is a core platform of what we're doing. We are today a technology leader in EML technologies, and EMLs are really important devices for data center transceivers. If you want to know what an EML is, it's a very, very tiny laser that we fabricate a very, very fast modulator in front of. That modulator is used to turn on and off that light and encode data at up to 400 billion bits of information a second. Then we aggregate multiple of these together in transceivers, and we can deliver aggregate data rates up to 6.4 terabits per second or beyond. We have been in high-volume production on 102 gig EMLs for some time.
We have last year demonstrated our 400G EML technology. We're demonstrating even more capability there today, and we continue to ramp and grow those platforms. We were also the first to introduce a new type of EML called a differential EML. The reason this is important is it provides much better signal integrity and much better modulation efficiency. That matters because as you go up in data rate per per wavelength or per device, you need to get more and more link budget to meet the requirements in the data center. To do that on those higher capacity links, everybody wants to move to these differential EML technologies. Being a leader in that space has really driven a huge interest and a huge growth in the demand for our semiconductor devices in transceiver modules.
Another place where that's really important, Julie talked about OCS. OCS deployment is growing in data centers. The one downside of OCS is it does introduce additional link loss, and so it requires higher link margin in the transceivers. That's another place that's driving the use of differential EML technology. In addition to very fast EMLs, we make very high-power CW lasers. Today, we do these in very high volume to support silicon photonics-based transceivers. The really exciting payoff of this is the same technology platforms we use to develop the 100-milliwatt lasers that we use in our silicon photonics transceivers are now being used for our 400-milliwatt lasers for CPO applications. CPO is growing very, very rapidly.
Again, we're in full production on our high-power CW lasers, both on our 3-inch platforms and our 6-inch platform in Sherman, Texas. We've got some very unique and differentiated IP in this space. It's very defensible. We have, using more advanced semiconductor processing tools, we've been able to develop a laser with very, very high yield across the wafer, very low phase noise, which gives you better link integrity, and very high power conversion efficiency, even at those higher output powers that are required. That's driving our customer engagement. We have extremely strong customer engagements across multiple customers for this technology, and we've already secured very large multi-year orders that are gonna guarantee our demand for several years to come in this space. It's definitely a place where we've got tremendous customer pull and engagement.
Finally, I wanna switch to the third class of devices for data center, and these are our VCSELs. We've been long a world leader in VCSEL technology. To date, we currently deliver 400 million VCSEL devices a year for data center and sensing applications. Our arrays for sensing applications actually contain hundreds of VCSELs per chip. We know how to make very large arrays as well. Our VCSEL technology now, we're one of the first in the world to get to 200 gigabit per second capability. That's the highest data rates people are currently deploying. We have unique technology that allows us to make single mode VCSELs, which help us to extend the reach of those in fiber networks much beyond what's traditionally been able to be achieved.
The beauty of VCSELs is they lend themselves enormously well to large arrays. First of all, VCSELs are very low power and very efficient ways to transmit data, but they're also very convenient for making large arrays. We make today two-dimensional VCSEL arrays, both with conventional top side emitting and new back side emitting VCSEL designs. Back side emitting are very convenient 'cause they enable you to flip chip the device for easier integration. Also we can integrate micro lenses on the back of those chips, and those integrated micro lenses greatly enhance their ability to package and integrate those solutions at very low cost. Our VCSEL technology is in high demand today.
We obviously are in very high volume production of VCSELs across data center transceiver applications, and we have a number of engagements with hyperscale customers working on VCSELs for CPO applications because of their very compact size and really good power efficiency in those applications. Lastly, all of these great technologies to transmit light over the fiber wouldn't be much good if we didn't have something on the other end to receive it. We have a broad array of internally developed photodetector technologies from 50 gig to 100 gig to 200 gig to 400 gig per lane capability in both indium phosphide and gallium arsenide. We make large arrays of photodiodes for more efficient packaging in data center transceivers. If you've got a multi-lane transceiver, you can use our quad array photodiodes in that application.
We make our photodiodes again in flip-chip configurations with integrated backside lenses. Those integrated backside lenses, again, are to facilitate coupling of light into the detectors with very high efficiency, very robust high tolerance for those coupling efficiency, and that drives low cost. The key to this is Coherent has a full end-to-end capability of semiconductor photonics, from the transmit to the receive side, from the highest performance EML technology through CW technology for silicon photonics down through VCSEL and Shortwave. That deep technology stack gives us a huge advantage in time to market, in innovation when we integrate those into our, CPO and, transceiver products, in better cost and better supply chain resiliency. I think that's a huge differentiator for us in the market. Thank you very much for your time.
I'm now going to pass it off to my colleague, Sanjai, who will speak about our scale across technology.
Thanks, Bob. Wow. It's great to be here, and it's so nice to see so many familiar faces in the audience here. Sorry. Put it on up here. Okay. For those of you who don't know me, I've been with the company for over two decades, and prior to my appointment as the first CMO of the company in 2019, for 15 years, I ran what is now our communications business. I ran product marketing and product management, or our scale across business. You can imagine, I'm especially thrilled to talk about our innovation and scale across. Data centers are power constrained. This means that large AI workloads now have to be distributed among multiple data centers, and that drives a need for high bandwidth, low latency connections between the data centers or the scale-out network.
In parallel, over the last, I would say 4 or 5 years, the optical networks are becoming increasingly disaggregated. From large monolithic systems, they're giving way to smaller, easy to deploy functional elements. Two such critical enablers of scale across are data center interconnect transceivers and transport equipment. Now, data center interconnect transceivers or DCI transceivers, they sit at the edge of a data center. They convert electrical signals to optical signals and then get them ready for transport. Transport is, as the name implies, are systems that take the signal and transport them, carry them from one data center to another. The distances between the data centers can be anywhere from tens of kilometers to thousands of kilometers. Let me first start with our DCI transceiver.
We have a deep vertical technology stack. Every photonic component in a DCI transceiver we make. This ranges from indium phosphide tunable lasers, the same indium phosphide platform that Beck talked about to modulators to silicon photonics-based planar integrated circuits, indium phosphide-based planar integrated circuits, detectors, passive optics, thermoelectric coolers. We even make a few ICs. We also take these components and package them into tiny subassemblies, such as the Nano-ITLA, which is a integrated tunable laser assembly. We also take the tunable laser and combine it with the receiver optics, and we make a product known as an IC-TOSA. I'm super excited to announce that our new high-power narrow line width tunable laser is now ramping in production for both in-feed into our own DCI transceivers as well as packaged into a Nano-ITLA for the merchant market.
I'm also thrilled to also announce that our IC-TOSA, an extended C-band IC-TOSA, won an award yesterday at this very OFC. By extending beyond the C-band, that enables our data center customers and hyperscale customers to drive more efficiencies from their fiber infrastructure. We have an industry-leading roadmap in DCI. Let me start with 100G ZR. You know, in the 100 ZR, we took all the complexity and all the smarts of coherent transmission and detection and compacted it into a tiny pluggable QSFP28 form factor. The reception from the customers has been fantastic. Today, we have over 40 revenue-generating customers for this product. We continue to innovate in the platform. We, in this OFC, we announced the ramp-up of a BiDi product. Okay, what is BiDi? It's bidirectional transmission.
It is a transceiver that allows you to transmit and receive on the same fiber. That doubles the efficiency of the fiber infrastructure. We also won an award, and this was yesterday, for a low latency version of the same product. You can imagine how important latency is for DCI scale across networks. Moving on to 800G, we were the first to introduce an 800G pluggable transceiver into the market for coherent transmission. We quickly followed it with an L-band version. Again, going to the L-band enables our data center customers to double the efficiency of the fiber plant. We've got 1.6T development. We are also working on 3.2T. Okay, now moving to the transport equipment.
Transport equipment serves many important functions, including multiplexing, demultiplexing, essentially taking multiple wavelengths of light and putting them on the same fiber, amplifying the signals, switching routing, and then monitoring the health of the optical signal, as well as monitoring the health of the underlying fiber infrastructure. For the past 20 years, improving efficiency has been the biggest mantra in transport equipment. We are a pioneer and an innovation leader in transport. We have a long list of unique and industry-first innovations that we have brought to the market, starting with the merchant EDFA in the late 1990s. There's too many on this list, but let me spend a couple of minutes on a couple of these products. The dual-chip pump completely changed amplification.
By combining the functionality of two pumps in a single pump, you get significant power, cost, and space savings, and it's essentially a step function change in amplifier performance. With the embedded OTDR, we created a market. We created an entirely new market segment for monitoring the health of the underlying fiber infrastructure. My personal favorite is pluggable optical line systems, which we affectionately call POLS, where we took an entire line system and collapsed it into a pluggable form factor. Multi-Rail is our latest in this long history of transport innovations. What are Multi-Rail systems? Transport equipment is usually housed in enclosures like the one that you see in this photograph. We call them fiber huts or amplifier huts. This one looks like it's in a residential neighborhood. There is snow around.
You can see the power going into the hut. Some of these are temperature-controlled, so there is additional power that you need for temperature stabilization of the devices there. Now with the demand and scale across, our data center customers and hyperscalers are demanding multifold increase in transmission capacity. Now, that creates a major challenge for this fiber hut. For example, if you want to go four times more capacity through the existing infrastructure today, the state-of-the-art current systems, you need four times more equipment, four times more power, four times more space, and of course, four times more cost. This is where our Multi-Rail comes in. It's truly groundbreaking innovation. What Multi-Rail does is it allows you to process four times the traffic within the same space and with a sublinear scaling of power.
It's truly a groundbreaking innovation. Now, how do we do it? Okay, I'll come back to what I said earlier. 4 times 4 rails in one tiny 1 RU or one rack mount unit. How do we do it? You've got to open up the hood of the Multi-Rail, and you'll find many award-winning innovations there, starting with the array amplification. The array amplifier basically does the function of multiple amplifiers in the same space and again, with sublinear scaling of power. How do we make an array amplifier? With the dual-chip pump. In this particular case, it is a quad-chip pump, so we have 4 chips in one package, uncooled, 700 mW per pump. Really innovative product there. Then we also have a multi-port dynamic gain equalizer.
What is a dynamic gain equalizer? Well, when you transmit multiple wavelengths or DWDM signals on a fiber and you amplify them, when the signals get to the other side, some signals are amplified, some wavelengths are amplified more, and others not so much. That's just the physics of erbium-doped fiber amplification. You need a DGE to level those wavelengths, otherwise you're gonna have issues on the detection side, on the reception side. That is the DGE, and we have compacted multiple lanes of DGE into a tiny package. Then to monitor the signal and do fiber metrology, we have a multi-port OCM and a multi-port OTDR. The OTDR is an optical time-domain reflectometer. It monitors the health of the fiber infrastructure. Lots of award-winning innovations.
The platform itself has won multiple awards, and at this OFC yesterday, we picked up two more awards for our Multi-Rail platform. The multi-port DGE and the multi-port OCM won awards as well. You can imagine the customer reaction for this product has been pretty fantastic. There's been a frenzy of design and activity with hyperscalers, with NEM customers all across the board. As Jim mentioned earlier, we are going to be shipping first revenues in early next year. We have an amazing demo of the Multi-Rail system at our booth. I'd highly encourage all of you to come and see it. I'll invite all of you to those demos. There are also demos that Julie and Beck talked about.
Really, really cool demos at our booth, so please try and make it there. I think with that, I want to thank you for your attention, and I'm going to pass it back to Jim.
All right. Thank you, Sanjai.
Thank you, Jim.
All right, I am also very excited about the Multi-Rail system. We're gonna open up for questions in just a minute. I'm just gonna summarize real quick. Look, we're at an inflection point of accelerated growth ahead of us. We talked about all the different new growth engines that we're bringing online over the coming quarters. Really it's a combination of that continued progression and expansion of photonics across the data center, the tremendous breadth and depth of technology and products that we bring to our customers, and then along with that technology, the expertise, and definitely the scale of manufacturing to be able to scale and hit the quantities that they need. With that, we're gonna open up for Q&A, and I think Paul's gonna moderate our Q&A.
Thanks, Jim. While the guys are coming up on the stage, obviously this is about. Wow. Can you all hear me okay? If you could keep your questions focused on the topics at hand for today's session. For those of you sticking around, we do have food and refreshments afterwards. With that, I will open it up. Simon.
Is there mics? Or yeah.
Yeah, there's mics that are coming around.
Thank you. Simon Leopold, Bloomberg, James. A couple weeks back, you did announce the agreement with purchase agreement investment from NVIDIA.
Yep.
You talked a bunch about the things they may be doing, but could you maybe connect the dots in terms of what's related to that commitment from NVIDIA? For example, you've got the investment. Where's that CapEx maybe going? What are you investing in? What are your priorities? What have you committed to delivering to them? What kind of visibility do you get from it? Thanks.
Yeah. Thanks for the question. First of all, on the cash piece of your question, the $2 billion of investment from NVIDIA, we're gonna use a good portion of that cash towards just capacity expansion, right? You know, just given the tremendous demand and their forecast of demand through the rest of the decade, right? This partnership expands or goes through the rest of the decade. A good portion of that will be indium phosphide, no surprise, capacity expansion, but also capacity expansion for the other products that we're bringing to them as part of this agreement. The agreement covers multiple products. A lot of those products that it covers, actually all those products that it covers, Julie covered in her section today.
If you go back to that slide that talks about all the different things that we can provide in CPO applications, that's kind of a superset of the things that we'll be supporting NVIDIA on. We're really excited. It's a step function expansion in our partnership and you know, that supply agreement runs out through the rest of the decade. Great, yeah, great partnership, and we're really excited about it.
Samik. We can get a microphone over.
Oh, hi. Thanks for the session, and thank you for the questions. Maybe, Jim, starting with you, as much as yesterday's sessions were very focused on CPO and when it sort of goes inside the rack and outside the rack.
Yeah.
You sort of dodged the question, it seems like, in your panel. Any thoughts in terms of as we sort of go through this transition of Coherent to optics, when are you expecting broader adoption of scale-up outside the rack? Then when does it sort of start to filter into inside the rack? A second one maybe for Jim or Beck and Jim, you can chime in as well. The indium phosphide capacity ramp and the 6-inch transition. Maybe an update in terms of how that's going and when you're seeing doubling of capacity next 12 months, another doubling. How much of that is capacity addition versus the transition from 3-inch to 6-inch?
Okay. On the first part of the discussion on CPO, and Julie, you can jump in if you want. The way to think about it is, the CPO coverage will scale out and scale up, right? That we talked about today. The first CPO will start to ramp in scale out, right? We'll start to see the revenue from that towards the second half of this year, basically second half of calendar 2026, and then scale out will ramp into 2027 and beyond. We see scale up starting to ramp in the second half of 2027. About a year later, we'll start to see first revenue from the scale-up portion. Yeah, no secret obviously, NVIDIA is lead customer for us, but we are engaged with multiple other customers in CPO, both scale out and scale up opportunities.
We do expect adoption across the multiple customers in both domains, scale out and scale up. Then the second question on indium phosphide, remind me.
Six-inch.
Oh, six-inch. The way to think about all of that capacity expansion that Beck showed. You know, he showed we're gonna double it this year, we're gonna double it again. That is almost all six-inch capacity, right? By the end of this year, as we double the capacity, and that incremental capacity is almost all six-inch, we'll basically be at a 50/50 mix of three-inch and six-inch capacity. The expansion again, the doubling again, all that incremental will almost all be six-inch again. Six-inch will just continue to become more and more of our capacity over time. It's really capacity expansion. We're not so much converting lines as we are expanding six-inch lines. As we get a chance, we'll convert three-inch to six-inch as well, right?
that capacity expansion is really mostly six-inch driven. I don't know. Beck, would you add anything to that?
The only thing I would like to say is, the beauty of going to the larger wafer size is it's so much more efficient for us to add capacity, and I can give you a simple example. We use a lot of these very specialized tools called epitaxial reactors that we buy from a company called AIXTRON. These tools grow the indium phosphide layers on the wafer, right? When we go from three-inch to six-inch, we get 2.5 times the capacity per run in the same kind of reactor configuration. Six-inch is really efficient as we scale, and that kind of scaling goes across multiple tool sets in the fab. It's really the only way that we can scale at the rate we're going to scale is by doing six-inch.
Yeah. The simple way to think about it, 'cause you're getting more than 4 times as many devices out of a 6-inch versus a 3-inch. Every new 6-inch line we put in place is like us putting four 3-inch lines, right? We're scaling at 4 times the rate as if we were doing it on 3-inch lines, right?
Nothing? Yeah.
Paul, I have the mic. Is that all right?
Yeah, absolutely. I'm sorry. It's hard to see with the lights.
From Steve Paul. I have a question for Julie. Julie, wondering if you could walk through the TAM improvement for the OCS from $2 billion-$4 billion, and just maybe walk through some of the use cases that are driving that TAM expansion. It seems to me, I could be wrong on this, but it seems like most of the OCS deployment today is spine switch replacement. You know, do you see that as being the primary use case as part of that TAM expansion in the near term, or are there other use cases as part of that? The second part of the question is, you mentioned software being an important part of OCS deployment. Google's doing a lot of work there and have been for a long time.
How are you seeing other hyperscalers, you know, kind of address, you know, the software aspect of it? Is that, you know, something that you could, you know, add value with or are you gonna leave it to the hyperscalers and other customers? Thanks.
I think the OCS use case is the original use case. I can say, as we all know, it's Google. They published a paper on it, so I can say it. It was really kind of spine switch replacement. As I mentioned, I think the use of OCS sort of coincided with the buildup of AI infrastructure. People have broadened out their thinking on OCS, which is part of what made us think, well, we actually see the demand also. That gave us a good signal to increase the scale.
You know, when you look at it's really, as I've tried to say, it's any place. You can almost think of it as a new capability in the data center rather than a flattening or a difference of a layer or a place. It's not a replacement for an electrical switch because we're not switching packets. Think of it as a new capability. That new capability is at a software notice, I can reconnect how all my GPUs are connected. If you think about that, you can use that any place. It's not just in the spine. That's why we're seeing people have interest in the scale up, the scale out, the scale across, and the spine, actually, I think.
Because in the end, it's about operating the data center efficiently to generate as much revenue off of the GPUs that you've put in place, actually. I think that's part of the expansion. To your other question about the, well, I don't know. Did you
Software. No, go ahead.
Yeah. On the software, yes. We have an excellent software team, and the only time we talk to customers about software is for them to thank us for our great software team. But we write the software that actually controls the OCS given commands from the customer's software. Our software gets integrated into their software, and then they control obviously their network and everything like that. It definitely adds value. If your software is good and it's easy to integrate, that adds value over someone, you know, something that isn't actually. Yeah.
Nina.
I guess, could you just expand on kind of gating items to greater VCSEL adoption? 'Cause I think, you know, indium phosphide it can alleviate some of the bottlenecks that we have. It all came from a session where they were kind of saying that VCSELs would be a smaller portion of the market. Just how do you see getting over some of the hurdles that have prevented greater adoption?
Yeah. Do you mean in CPO, NPO, or do you mean in pluggable or both?
I guess both.
In both?
Yeah. I mean, the CPO, I think it makes sense at scale up because of the short distance.
Yeah.
It's greater in the pluggables.
I think definitely in CPO, NPO, there'll be short distances, and I think, you know, Julie covered that really well. We'll clearly see VCSEL adoption in NPO applications, and we're already engaged with customers there. In pluggable, you know, even in the pluggable use cases, there are shorter distance links that we see even in pluggable applications. We think there will be adoption of VCSEL-based 1.6T transceivers in some subset of the market, right? It won't go across all applications, but there'll be shorter reach applications where we expect those 1.6T VCSEL-based transceivers to be adopted.
That's good.
Hi. Matthew Niknam from Deutsche Bank. You're expanding indium phosphide capacity, but the raw indium feedstock is roughly 70% sourced from Chinese zinc smelters, which are now subject to export permit requirements with multi-month processing times. I guess my question is, how much visibility do you have on indium supply for the next 12-24 months? And are you actively diversifying sourcing away from China or do you hold strategic inventory as a buffer?
We actually have a very diversified supply chain for indium phosphide substrates. We have, and I think I've shared this in the past, we have over five different substrate suppliers today, and we work with those suppliers not just on the next, you mentioned next 12 or 24 months. We don't work on just next 12- 24 months. We work on the next one through, you know, like 3- 5 years of capacity that we're gonna need. We have, you know, in some cases, very long-term agreements in place, and that includes not just the substrates, but all the key inputs that go into that.
We believe that we have very good visibility into substrate supply, and that capacity expansion that Beck showed is, you know, we have commitments from our suppliers to supply the necessary indium phosphide substrates to support that.
Next question. Sure.
Thank you. I'm Papa Sylla from Citigroup. I guess this question is more for Julie. You had a very helpful slide showing the SAM with different areas of CPO that you can provide. I guess I was hoping if you can maybe parse through within that $15 billion SAM, what is ELS, what is ultra-high power laser, what is effect? And tied to that, I know you have the capabilities of doing all of those, but if you can maybe help us understand which areas you are already number one or number two, and in which areas are you guys ranking.
Yeah, we believe. Maybe I'll kick it off and Julie can weigh in. So on the second part of your question, we believe we'll be a leading supplier in all of the things that Julie mentioned, right? We have line of sight and very clear demand from customers on each one of the things that Julie talked about. I also wanna stress again that, you know, there's no other supplier in the world that brings that entire portfolio of technology. If you're gonna go build a CPO or an NPO, a system based on CPO or NPO, you wanna be working with a supplier that brings that whole portfolio of technology 'cause otherwise you as the customer are having to do the integration, trying to put together the supply chain.
If you go to Coherent, it's a one-stop shop of that supply chain, and that's a big advantage for us. A lot of the very critical components, and this is important as well, are done in the right geography, are done in geographies that the customers want it, and that's obviously, you know, like indium phosphide, high power CW lasers done in Texas, right? Very important that we don't just have the breadth, but we've got the manufacturing in the right place, and we've got duplicated manufacturing in a lot of cases, indium phosphide being another good example. In terms of the breakdown of the $15 billion, we don't have a breakdown today, but maybe at a later time we will break that down into more granularity.
I would say the external laser source module is a significant part of that $15 billion. Within that ELS is all the things you showed, the lasers. A lot of times there's focus on just the lasers, but there's a bunch of other components that go in there, the isolators.
Thermoelectric coolers.
The thermoelectric coolers, all the things that Julie showed today. The other big portion of that $15 billion is in the fiber attach unit. That's all of the fiber that goes from the front faceplate to the switch chip or the XPU, also goes back out to the ELS, right? That whole assembly is very complicated optical assembly, difficult to do. Very few companies can do that well. Coherent, since we've been doing that for decades, right, as part of transceivers and other optical systems that we manufacture, we know how to do that. That's a very high value piece of that $15 billion as well. Again, there's kind of key components underneath there that we supply as well, like the PMLA that you talked about.
Yeah, we feel like we've got a very very good competitive position on that sort of totality of CPO and MPO technology.
Jack.
Hey, thank you for the presentation. This is Karl Ackerman from BNP Paribas. I have two questions about your costs. First, I'm glad to see now you have expanded engagement on Nano-ITLA. Can you just talk about your customer engagement, your breadth of customers, with external customers, or you are using it mainly for your own program, internal?
Yeah. I believe we're using it both internally, but we also have external customers as well. Sanjai, do you want to just talk to the breadth of the external customers?
Yeah, sure. We have a lot of deep customer engagements, both with NEMs as well as hyperscalers. The tunable laser itself, we use it as a chip for our DCI transceivers, but we also package it into a Nano- ITLA into the merchant market. As you know, there is a merchant market for the Nano- ITLA.
Yeah, I was sitting in a meeting this morning where the customer was asking us to make more, right? Yeah, it's in high demand.
Nice. That's nice to hear. About your material technology, that's an important technology there. You showed the box there. What's your strategy there? Are you going to supply the entire box or components into those boxes?
Yeah, all of the above. We'll supply both the box as well as the key components into other suppliers. We'll do both.
Okay. Thank you very much.
Yep.
Jack. Is that mic on?
Is this working?
Yeah.
Hi. Jake Silverman, Bloomberg Intelligence. I wanted to ask about your material segment that you called out today. You have capabilities across silicon carbide in a press release about two weeks ago on the ThermaDite liquid cold plates, which you mentioned, I think. I'm curious how close to your engagement some of your customers across these 2027 second half of 2027 later timelines. Is this you kind of intersecting where you think power density within the racks goes and, you know, the new materials that are needed for this and, you know, should we expect to see a lot of additional announcements over the coming quarters that start to layer on top of that in 2027, and into 2028 as well?
Yeah. You should comment too.
Yep. Yep.
We're pretty excited about it. We're seeing there's engagement across multiple customers. It's a very compelling technology. Remember, there's kind of two categories. Category one is, you know, like using ThermaDite, our proprietary technology, to pull heat away from the switch chip or the XPU, et cetera, and it's much more effective than a copper cold plate at pulling heat away. Which is then obviously you can run your XPU or whatever faster, it's better reliability. The other technology that's really cool is the thermoelectric cooler kind of running backwards, where it pulls waste heat and creates electricity that's funneled back into the data center. If you think about that's pretty compelling to basically any data center owner, 'cause what's the primary constraint right now in data center? It's power delivery, right?
If I can take heat out and recycle it back into power and that's just like free energy, right? That's very compelling to every data center owner. Even if it were just, you know, small percentages of recovery of heat, that's very significant. I would say the customer engagement on both of those is really, really strong. Anything you'd add, Julie Eng?
Yeah. Just adding. Yeah, you can make a liquid cold plate, as Jim said, heat spreaders and because those materials we can form, and then also silicon carbide wafers, as a thermal wafer. In general, I think we are in a very special position that we have all these unique materials, but we have all the context in the data center. We're not just some materials company who knows nothing about data center knocking on the front door, right? We're working with the customers, and they have to learn, right? This is a new tool in their toolbox.
Kind of they learn about the material, we go back and forth, and so it takes a little while, I think, to germinate, but kind of not so a totally different product, but not so different than the OCS, that first people have to get used to it, they have to think about it, then they figure out how they can utilize it. All of a sudden, they start coming up with a whole bunch of like a cascade of options. Yeah, we were at SEMI -THERM last week. That's a big thermal conference. We had a lot of engagements. We had a talk, we had a booth. I think you will definitely see increasing announcements from us or demonstrations over time before we get to those points where we call that the revenue and the SAM yet.
Yeah.
Jim, if I could go back to my sales ideas and ask a follow-up to that. Hopefully, it was clear, but as an economic opportunity, the power challenge and the thermal challenges, were those both prominent?
Oh, yeah. For sure. I mean, for data center operators, those are two of the biggest problems, right? The data center is limited by the power delivered into it, right? You're trying to generate as many tokens as you possibly can out of the total energy that you can deliver into that data center. This effectively would help you generate more AI tokens per data center by harvesting that heat and putting that energy back into the data center. There's an immediate economic impact, right?
Yeah, that's that one. The other one is, like what we've shown with simulation and actually some measurements, is that using the thermal materials, we can reduce the junction temperature of the XPU by between 5 and 10 degrees. That can either, as Jim said, improve the reliability, but what actually I think our customers are gonna do is they're gonna run the clock speeds faster, and they're gonna get more output. The output of those is their, that's their revenue stream, actually, right? People are very powerfully motivated. It's really gone from thermal being like, oh, an after effect implementation thing to being like front and center in the architecture of the AI data center.
George
Hi. Thanks. George Notter from Wolfe Research. Just expanding on the ThermaDite discussion. I know it's been percolating inside the company for a lot of years, I think, and I guess I'm just curious what the milestones are to see, you know, traction towards customer engagement, customer revenue, products, making real business out of this. Like, what are the milestones you guys are looking for?
For me, the only milestone I ever care about is revenue. The revenue milestone is we think, based on the customer engagements that we have, and there's multiple of them, that we start to generate revenue in the second half of next year, so second half of next calendar year. To me, that's the key milestone is when does the revenue turn on? Now, there's normal engineering milestones that we go through ahead of that, you know, normal qualifications and pilots and things like that. The most important milestone is that revenue milestone.
What's reasonable in terms of opportunity for you guys? I mean, revenue run rates.
So we-
-size.
We set the TAM or SAM, I guess, at $2 billion. You know, we always try to be a little bit conservative on the SAMs that are further out in time. I think it could easily be bigger than that. If we were some portion of $2 billion, look, I think this could be certainly hundreds of millions dollar product line and could become a billion-dollar product line for the company. Yeah. Thanks.
It's hard to see in the back of the room. If I'm missing anybody, please, shout out.
Should we do this last?
Last part
Last question?
Yeah.
Yep.
It is maybe the other's question. Cori Johnson from Distribute Capital Research. In this materials business, what are the barriers to entry to that? How, you know,
Yeah
Lab-grown diamond thing, for example, that seems to pop up in other places. What are the barriers to entry? Obviously you've got some sales advantage as well.
Yeah. That's a really good question. In the materials business, the barrier to entry is very high because it's like having a special recipe, right? It's very difficult to reverse engineer that recipe. Like, if you take ThermaDite, proprietary material we've developed, you know, years and years ago, it's very difficult to reverse engineer that particular recipe. It's like somebody handing you a cake and saying, "Hey, without recipe, reverse engineer and figure out how to bake a cake," right? If you've never been given a recipe. That's what it's like. It's very high barrier to entry. We feel like this is a very well-protected area for us and something that we have unique expertise in. I mean, we were founded as a materials company, right?
This is really deep heritage here. Really what we're doing is we're taking an existing material from our industrial space and just repurposing it for a new application, right? That's what we're doing.
Thank you.
Good.
Great. With that, I wanna thank all of you. I wanna thank all of you online. Again, for those of you here in the room, if you're hungry, you want food or drink, and if you wanna speak to Jim and the rest of the team, we have some time. Thank you.
All right. Thanks everybody. Thank you.