Thank you everyone. My name is Jeff Koche, and I'm an Associate Analyst here at Raymond James covering data infrastructure and semiconductor stocks. With me today is Applied Optoelectronics CFO Stefan Murry. We have a fireside chat format today, and we'll leave it, some time for questions at the end. Thank you for joining us, Stefan.
Thank you for having me this morning. I appreciate it.
To begin with, for new investors, can you just take a second and introduce the company, including what markets you serve, your main products, and what problems you're helping customers solve?
Sure. AOI is a 29-year-old company, founded and headquartered currently in the Houston, Texas area. We're actually located in Sugar Land, just outside of Houston. We serve mainly 2 major end markets with a couple of related markets as well that are smaller for us. Currently, the biggest market that we serve is the data center market. No surprise to many of you in the room, but there's been quite a lot of activity in the data center market. I'm sure we'll spend some time talking about, you know, more about that throughout the morning here. Data center is our biggest market, as I mentioned. Our customers there are typically the large hyperscale data center operators, and we manufacture optical equipment that's used to interconnect between switches and servers within their data center infrastructure.
You know, importantly, I guess, to contrast that with some other companies, we don't make the switches, we don't make the servers, we don't make the racks, we don't make the cables, but we make little fiber optic adapters that help connect those pieces of equipment together, glue, if you will, that glues together those network elements within the data centers. That, that accounts currently for about two-thirds of our business. We have a smaller business in cable TV outside plant equipment, mainly manufacturing amplifiers and nodes for large cable operators.
Charter being our largest customer there, as well as, you know, a number of smaller regional multiple system operators, both in the U.S., Canada, and other areas around the world. Those two make up, like I said, almost all of our business. We have some small business in fiber to the home and telecommunications. Most of that is sort of legacy business at this point. The two growth areas really are data center and cable.
Great. Yeah. Let's hit on data center. I think that's what a lot of people wanna talk about. AI appears to be changing this business pretty rapidly. You know, how has your business evolved, and where do you sit in the industry landscape? Specifically, how are customers deploying your product? Is it in the back end? Is it front end? General purpose?
Sure. First of all, it's worth pointing out that AOI has been in the data center business really since optics were first used in the data center. A little historical context there. If you go back, you know, prior to about 2012, 2011, 2012, the data centers were physically smaller in size than they are today, and the speeds at which those network elements within the data centers needed to be connected was also slower than it is today. The implication of that is for many years, they were able to connect the network elements within their data centers, the switches and servers, just using copper cable, like, you know, an Ethernet cable like you might see at your home or business.
Around 2011, 2012, the very largest data centers, Amazon being the first, recognized that as the data centers physically grew in size and the speed at which the switches and servers were operating began to, you know, or continued to climb, they recognized that those copper cables weren't going to be an effective solution for, you know, utilizing transport mechanisms over the distances that they needed to span. They started to look at optical solutions. There were optics, transceivers that existed before that data center advent in 2012, but they were designed mainly for telecommunications applications, meaning they had, they had to be utilized in outdoor environments and other things. We were really the first company to develop purpose-built data center transceivers in the 2012 timeframe.
That was in conjunction with Amazon. We've been in optics and data centers since really optics and data center was a thing. Today, obviously the growth that we're seeing is really everywhere within these large data centers, but certainly AI is what is driving a lot of that growth. I'd like to say, you know, look, AI in a vacuum doesn't really do much, right? You need to connect AI to something else outside. Those general purpose networks, the things that operate within the data centers to format and route data outside of the data center, those are really important parts of the network for the data center operators.
Certainly the growth of these large language models, the growth of, you know, generative AI type capabilities and the really, really large compute clusters that are necessary to, particularly for training in those models, but also more and more for inference, as those needs grow, and they've obviously grown dramatically, and we expect them to continue to grow dramatically, that places a greater, you know, a greater degree of emphasis on the interconnection between those network elements, in particular, at higher speeds. 800 gigabits per second and above is typically a speed that you would see used in the back end, the actual compute infrastructure. The front end, where you're doing more of the post-processing and things like that, tends to be 400 gigabits per second and below. Although, you know, even those networks are evolving to use faster and faster speeds these days.
Speaking of which, like, we had two announcements yesterday from competitors, Coherent and Lumentum, on partnerships and equity investment from NVIDIA. You're in long-term agreements associated with those. You have a long-term agreement with Amazon. Maybe you can talk about that first.
Sure, yeah. Well, I think to a large extent, what we're seeing is a recognition on the part of the large hyperscale operators that the amount of industry capacity to manufacture optical transceivers and the optical components that go within those transceivers, things like laser diodes and photodiodes and perhaps some other elements that are used within those transceivers, those are not. You know, they're beginning to recognize that as we move forward with these large compute clusters, there's not likely to be sufficient capacity for everybody to get all of the modules that they need. I think many of the large hyperscale operators have recognized that, and they're sort of staking their claim to try to, you know, obtain as much capacity as they think they're gonna need, by doing these long-term deals. You mentioned Amazon for us.
We actually have three out of the top five large data center operators as customers. I would say, you know, just within the last few months, we've had more and more interest about securing long-term commitments for the capacity that we have today and capacity that we're bringing online. That's particularly acute in the sense that companies that have laser manufacturing capacity, which we do, and Lumentum and Coherent do as well, and I think that the announcements yesterday mainly revolved around securing capacity for the production of laser diodes. So that's indium phosphide fabrication plant capacity to make laser diodes, mostly laser diodes for these data center applications.
Again, when you look at the universe of companies that have optical technologies that can make transceivers, there's a certain subset of companies that can do that. When you narrow that down to companies that actually make their own optical devices in-house and so have the capacity to expand and grow their ability to make those indium phosphide devices themselves, that's an even smaller subset of those companies, and I think that's really where the bottleneck right now is in the industry is on that laser capacity, and certainly an area where we're expanding along with the transceiver manufacturing. Both of those areas need to go hand in hand for us.
Yeah. Yesterday's press releases definitely seemed like they focused on U.S. production too. Maybe you can talk about, you know, how much of your, of your product is internally sourced versus externally, how do you get to... And where are you at in getting to, like, a 200G per lane EML, and getting a 1.6T product?
Sure. A number of different questions embedded in there. I'll kinda walk through that, I guess. First of all, when it comes to the technology evolution that we're seeing today, 200G per lane is certainly a part of our plan for 1.6T specifically. We have those products today, but it's not based on EMLs or VCSEL technology. It's based on silicon photonics. That's really the direction that most of our products within the data center are going in the future is a silicon photonics type solution.
In the silicon photonics world, what's needed is a narrow, a high power, narrow linewidth laser that can then be modulated with the silicon photonics photonic integrated circuit, or PIC, to be able to encode data on that light. The lasers are needed. The silicon capacity is also needed, although typically, silicon has much larger industry capacity available than the III-V components like lasers. I think it's important to realize too, and one of the, one of the reasons for some of the announcements recently, is that as the industry evolves from sort of EMLs and VCSELs towards more high power, narrow linewidth lasers for silicon photonics and eventually for Co-Packaged Optics, those optical devices are physically larger in size, significantly larger in size than the earlier generations of devices.
If you think about, you know, the amount of wafer real estate that's available and the processing capacity for those wafers, as the die sizes get multiples in terms of size, well, that has implications on the amount of capacity that needs to be brought online, and I think that's a, you know, very important reason why NVIDIA and others are trying to secure this capacity early, so that when it really becomes crunch time, you know, in a year or 2, that capacity will be available to them.
Yeah. I definitely think our take was that these agreements were centered around, you know, scale up.
Mm-hmm.
Scale up network, maybe going Co-Packaged Optics. Are you having discussions with these, your larger customers on? Where would you sit in that, would that be another long-term agreement, or is that captured-?
Sure
in your current one?
You know, for us, we're certainly having discussions about both scale up and scale out applications for our technologies. Both are part of the equation. I would say for us, we're kind of agnostic in the sense that, you know, we're selling optical modules that connect things together, whether they're used in scale up or scale out applications. To us, it has implications on the volumes and things like that. At the end of the day, we're gonna need both of those technologies. You know, the devices that we make can be utilized in either of those applications. We're not particularly beholden on one versus the other.
Great. maybe just talking about last quarter, we seem to have a firmware issue that impacted your 800G results. sounds like that might be the resolution, but maybe you could just recap that for investors real quick.
Yeah, I mean, I wanna kind of touch on part of that. It really didn't impact our results dramatically. We didn't have a lot of 800G in the forecast anyway. We were supposed to ship $2 million of 800G in the quarter that didn't get shipped because of this firmware issue. As we're called upon. You know, initially, when we started working on 800G, we had a certain subset of the network that our customers intended to use us in. Those, that network application space has now broadened to the point where we're being asked to interoperate with a number of different platforms that weren't part of the original scope of the project.
As those new platforms are brought into the mix, that means we have to sometimes do some interoperability work to ensure that they're gonna work smoothly across, you know, all the variety of different switches and other network elements that are out there. That was, you know, resulted in that shift, if you will, from, you know, Q4 shipments into, you know, anticipating Q1 shipments for 800G. The firmware issue, just to, you know, make sure everybody's, you know, we're on the same page. The firmware issue itself, I wouldn't even call it an issue. It was really more of a, again, it was a new platform that was brought on. To the extent that that was a problem, it's been resolved at this point.
We've shipped the new firmware, and we don't expect it to be, you know, an issue going forward. What I think is most critical for that is that for the last several quarters, as we've talked about our trajectory in 800G, we've consistently said that it was a Q2 2026 ramp, and that's still the case. Regardless of whether we ship in Q4 or Q1, the ramp starting in Q2 2026 is still on schedule, and that's really probably the most important point for investors to understand.
You still have, you know, you're still forecasting really good sequential growth for 1Q. I think that's 400G driven. Maybe, like, what is driving that? Is that the front end?
That's front-end application primarily, most of the growth that we're seeing is from 1 customer, you know, that we started to sort of re-engage with, last year. The trajectory there is, you know, it takes this customer a while to sort of qualify the solutions again, doing that interoperability testing that we talked about in the 800G realm a moment ago. We had to go through that on the 400G as well. That was completed last year, we're starting to see that ramp now. I think, you know, 800G, as I've said most of last year as well, 800G will follow a couple quarters after 400G. We're starting to see that initial ramp in 800G. A couple quarters from now, that's where we're gonna start to see that ramp in 800 gig.
Yeah. I think you've really characterized the business as being supply constrained. I mean, your success being dependent upon ramping that, and you've laid out a progression from getting from 100,000 capacity for 800G, 1.6T.
Mm-hmm.
to by mid-2026 being at 200,000 and 500,000 by mid-2027. Maybe you can walk investors through this?
Sure.
You know, talk about some of the challenges.
Sure. One of the things that is somewhat unique about AOI is that we've spent a lot of time innovating in the method of manufacturing the transceiver. I mentioned earlier that we have our own laser fab. I think that's a key differentiator. The other key differentiator for us is the automated technology that we've developed for manufacturing these transceivers. As we are called upon by the industry to really ramp production, I think that automation is. The advantages to that automation are starting to become apparent for us because we can replicate production lines. Because they're almost entirely automated, we can replicate those lines pretty easily.
I mean, it takes some time to order the equipment and get it installed, but it's a lot more predictable ramp process than it might be if we had to, you know, hire a new factory, you know, build new factories, hire a lot of people, train them, go through the learning curve and whatever. That automated production I think is very attractive. The other nice thing about the automation is that it allows us to geographically locate the production in whatever place is most advantageous to us. Right now we're seeing, and you touched on this earlier with the announcements yesterday, we're seeing a lot of interest in U.S. based production.
I would submit to you that if we were still manufacturing our transceivers in a more manual process the way many of our competitors are, you know, it would be very difficult to move production back to the U.S., not only because of the economics of moving large manual production processes back to the U.S., but also just finding, you know, the amount of skilled labor that would be required to do that in the U.S. Again, that automation allows us not only to scale up faster, but also to be able to locate that in a place like the United States where our customers perceive, the supply chain to be more secure if it's made in the U.S., and I think that's a big advantage.
We can do that with only a modest increase in cost. It costs us maybe 10% or 15% more to produce our products in the U.S. than it does in a similar production plant in Asia. You know, that little cost adder I think is perceived by our customers to be well worth it in terms of guaranteeing the security of the supply chain. You know, if you look at where our customers or how our customers are looking at their supply chains these days, and I'm sure this is no surprise, there's undoubtedly dozens of companies here that are involved in various aspects of the data center infrastructure build.
I'm sure they'll all tell you that our customers, the hyperscale customers, are really looking at this as an existential opportunity or crisis, depending on how you look at it. They need to build these data centers to secure their position as a leader in AI. If they can secure that position early, then they think it's gonna deliver, you know, dividends for many, many years or decades to come. There's a lot of urgency around building these data centers. There's a large number of different types of components that are needed to build a data center. You know, you hear about energy, you hear about GPUs, you hear all these different things, you know, fans and power and cooling and all sorts of things.
Any one of which can derail your efforts to build a data center. I think there's a great deal of emphasis on supply chain integrity with our hyperscale customers. A few years ago when, you know, maybe the market was a little bit less robust, there were a number of suppliers out there, and I think the reasoning went, "Well, look, if we have a supply chain disruption here or there, you know, there's other suppliers we can backfill with. It's not gonna be a, you know, it's not gonna be a problem slowing down our actual growth." Today, that perception has changed dramatically.
I think they're very, very keenly focused on the integrity of their supply chain, and the U.S. is perceived by them, I think rightfully so, to be, you know, the most secure place to be able to produce products. If you can build stuff here without incurring additional cost or too much additional cost for them, that's a huge advantage, and we think that our automated production allows us to do that.
Great, great. Given the supply shortages or tightness, given the shift to higher cost regions and the long-term agreements, can you talk about pricing, price per gig? I mean, it's typically come down, you know, 15%-20% per year until AI hit. You know, how much is it right now for 400G, 800G, and where do you think it's gonna go?
That's a great question. First of all, I would say, I think the historical trends in terms of price reductions are probably not really on the table at this point. As you mentioned, you know, in a, in a supply constraint situation, I think, you know, offering, you know, double-digit price decreases every year is probably not something we're gonna have to contend with for a while. As far as the price per gigabit, we tend to think. I mean, most of these technologies are kind of in the vicinity of $0.50 per gigabit per second. An 800G transceiver is in the $400-ish range, you know, the 1.6 terabit is perhaps a little bit less than that on a gigabit per second basis, but somewhere between $700-$800 for a, you know, just 1.6 terabit device.
This will be my last. I do wanna hit cable TV here.
Sure.
just on the CPO-
Sure. Everybody is here very excited to hear about cable TV.
On CPO, you know, It sounded a lot like those announcements yesterday were CPO-focused, high-powered lasers.
Yeah.
They do carry high ASPs. You know, how is AOI positioned to participate in that market specifically?
As I mentioned earlier, I kind of alluded to this, but one of the key technologies that's needed for CPO, Co-Packaged Optics, is a very high power, very narrow linewidth laser, and that's a relatively unique technology. AOI is one of the companies that has that technology. We've had it for some time. We actually developed it, believe it or not, for LiDAR applications some time ago. The LiDAR application, you know, back in, back in the days when LiDAR was gonna be ubiquitous in cars and everybody was gonna be sitting in the back reading newspapers instead of driving. Hopefully, that'll still happen one day, but as a commuter, I look forward to that day. It isn't here yet.
The LiDAR application that we had for these large, high-powered devices, it hasn't yet, you know, secured the economic benefits that we had planned. The technology behind it now is utilized in the lasers for Co-Packaged Optics, and I think that's certainly a nearer term growth driver for us for that particular type of device than the LiDAR was.
Terrific. Terrific. Okay, let's move to cable TV real quick. Can you update your outlook for DOCSIS 4.0? Charter, you know, has said 2025, which is, Charter's one of your biggest customers. They've said that 2025 is a peak for CapEx, for their CapEx spend. How can you still grow with them, and where else do you think you can have success?
Yeah. I think it's always tough when you look at the cable market, to extrapolate from an overall CapEx number to, you know, where exactly that CapEx is gonna be spent. If you just look historically at MSO CapEx, it tends to be not dramatically different from year to year, a little bit up, a little bit down. Where that spend goes within their network can vary dramatically. You know, they could be deploying head-end equipment or outside plant equipment or, you know, CPE-type equipment or backbone. You know, there's a whole variety of areas where it gets spent.
You know, regardless of Charter's commentary regarding their overall CapEx, what we're hearing from them and from other MSOs is that they're still getting going on ramping up their deployment of DOCSIS 4.0 in the outside plant. That's not a general statement about all cable companies. You know, different suppliers that supply different technologies may have different trajectories. For us, I think, you know, 2026 and 2027 are likely to be peak years, not specifically for Charter, but for us overall. Again, a lot of that is gonna be driven by Charter, we think. We have a number of other MSO customers that are kind of following along in that path that are maybe a little behind Charter in terms of when they're starting their deployments.
Those guys are either just starting to ramp or haven't even started to ramp yet. You know, we continue to see opportunities to grow that cable business into 2026 and 2027 and, you know, hopefully, beyond that, with some newer technologies that we have in the works.
Yeah. I think you've characterized the amplifier market specifically of being, like, $5 billion over, say, 5 years. I don't know if you gave an actual timeframe for that, but-
Yeah
A round $5 billion, and the node market being maybe half that. Is that still how you'd characterize it?
Yeah, I think overall that's about right. You know, you can run some of the financial models that we have, and that's sort of where you come up with. If you look at the number of amplifiers that are out there, the price per amplifier, the nodes tend to be about half that. They're actually a significantly lower number of nodes, but the ASPs are higher. Dollar-wise, it works out to be about half of the amount that the amplifiers are, so.
Where are you at on your own node product?
We have, we built a number of node platforms over the years. The requirements shifted a little bit, with some of the newer Broadcom chipsets that were announced, a year and a half or so ago. We're re-engineering those nodes to accommodate these newer, the newer thinking on the chipsets. Those will be out later this year.
Terrific. Do you have plans for a unified amplifier or full duplex amplifier?
Currently the amplifiers that we're building don't support Full Duplex DOCSIS, we don't have any plans to support that in the future.
Maybe talk about the software differentiation you mentioned.
Sure. Yeah, yeah, absolutely. You know, one of the things, as we look out there on the cable networks, one of the things that we've seen is that This generation of outside plant equipment, specifically the amplifiers, is the first generation of products. You know, we talk a lot about DOCSIS 4.0 being a driver, and it is, but there's an additional technology that's been added to those DOCSIS 4.0 amplifiers that hasn't really existed before, and that is that these are so-called smart amplifiers. They have microprocessor controls embedded in there. They have the ability to communicate with external devices through the network. Having that capability to do monitoring and control, you know, it's a game changer for the MSOs because now they can see deep into their network.
They can identify network problems, you know, before they become customer-affecting. When the industry goes out and looks at, you know, reasons why consumers, you know, might change to a different provider, a lot of times what's driving that decision is really the reliability of the network. The MSOs are very focused, and Charter in particular has taken a huge lead in ensuring that their networks have very high reliability. In order to do that, they need visibility all the way down into their network and the ability to see what's going on with the amplifiers and even taps and things like that down the line, and being able to identify network issues and send crews out to fix those problems before they become customer-affecting. The software that we've developed, helps them do that.
It utilizes machine learning to look at network impairments in real time, looks at the telemetry data that's coming back from our amplifiers, identifies amplifiers that are trending in a direction that may cause problems and allows the MSOs to send in some cases, they can repair things remotely. In other cases, they'll send crews out proactively before it becomes customer-affecting to make adjustments or fix or repair whatever needs to be done so that the overall reliability that the customer sees is very, very high, which is what these networks are demanding.
Great. Great. We still have a few minutes here for questions from the audience. If there's any questions, I'll repeat it for the webcast. Anybody have a question?
On your conference call, you mentioned that sometime later, bit later in 2027, you might be able to get to a run rate of something like $375 million a month, something like that.
Mm-hmm.
Can you explain how you get there? What needs to happen between now and that point?
Sure. I'll go ahead and repeat the question. I think you were alluding to some commentary that we gave on the call around reaching a revenue level for transceivers for data center that would, that would approach $380 million or $378 million. And that was specifically composed of about $91 million of 100G and 400G products, about $217 million of 800G products, and about $71 million of 1.6T products. When you break down where those data rates are, those are the data rates that we outlined in the earnings call. In order to get there, we need to build, you know, a significant amount of additional capacity. We talked about mechanically how that gets done.
It's taking the production equipment and machinery that we have already developed and replicating that in a, you know, at scale, right? Putting more and more of these production lines in place. Moving forward, the initial phase of that is largely gonna be concentrated in our Taiwan facility because it's quicker for us to build capacity in Taiwan than it is in the U.S. As we get into, as we get through 2026 and into 2027, the bulk of the additional capacity that's coming online is gonna be in the U.S. By the time we get to the middle, you know, that timeframe that we outlined, the middle of 2027, most of the production for 800G and 1.6T products is gonna be in the United States.
we're building, you know, adding equipment, building out new facilities, mainly in the Houston area, for the purpose of scaling up that production capacity in the U.S.
Any other questions? Okay. Maybe you can talk about, you know, how many 800G customers does that sort of contemplate? How many of the how many does it require? Like, I know the demand is there, but.
Well, I mean, when we, when we look at those numbers, those forecasts that we provided for the middle of 2027 are basically comprised of 3 hyperscale customers for us. There are other customers out there, and many of them we're in various stages of discussion with other hyperscale customers, and even customers that are not themselves hyperscalers but supply into that ecosystem. Really the forecast that we provided is based on 3 customers. We expect 800G and 1.6T probably from all those customers, maybe in a different mix. Not everybody's gonna buy exactly the same ratio of 800G and 1.6T, but we do expect business from all of those.
Yeah.
Yeah.
Great. We only have a minute left here, so I guess we always end with the last question, which is, what do you think is the least underappreciated aspect of AOI?
Well, that's a good question. you know, I think it's starting to become less unappreciated or more appreciated, depending on how you look at it. I still think it's a little bit misunderstood is the extent to which indium phosphide manufacturing capacity is gonna be a critical bottleneck for this industry going forward. I think those companies that have a heritage, and AOI's, you know, founding was related to the production of lasers. We've been doing that since 1997. Frankly, we were a research group, at the university looking at how to manufacture lasers for years prior to that. So our heritage in the manufacture of lasers goes back a long time. We have our own indium phosphide fab at scale. We're scaling that up, and I think that's gonna serve us really well, and I think that's something investors should probably focus on.
Great. Thank you, Stefan.
All right.
Appreciate your time.
Thank you very much.