Everyone, thank you for joining us. Thanks for joining AOI's session today at OFC. I'm Monica Gould, Investor Relations for AOI, and with me is Dr. Stefan Murry. He's Chief Financial Officer and Chief Strategy Officer for the company. We're gonna have about an hour session here. We'll have about 15 minutes for Q & A, following Stefan's presentation. If you can hold your questions until then, that would be great.
We're gonna walk around the microphones. This session is being webcast, and there'll be an archive available on the website later this afternoon. As a reminder, we will make some forward-looking statements, and actual results may differ. Please take a look at our Safe Harbor statement on the website. With that, I'd like to introduce Dr. Stefan Murry.
Thank you, Monica. It's a pleasure to be here. Nice to see a lot of well-known faces and some that are new to the story. I'm going to go through a number of topics today. We're gonna focus on, because there's a variety of people here that, you know, some know the story, some don't. We're gonna do a little bit of a little retrospective on, you know, how we've gotten to this point where we're at in our cycle in terms of the business.
We're gonna talk about our markets and customers. I'm gonna spend a fair amount of time today talking about our proprietary manufacturing technology. I think this is, this show, the OFC Conference, this is my 28th OFC that I've been to, also AOI's 28th. We've been here, I've been with the company since it was founded.
This is a technology show, and we wanna focus a little bit on the technology. For us, the technology that we're most interested in is really the technology for manufacturing, how we manufacture these transceivers. Our markets are very dynamic. Our customers are very demanding, and we need to have the best manufacturing technology out there. We're gonna focus on that for a little bit, and then we'll talk about, you know, some of our financial numbers upfront at the end and a little bit along the way as well.
Diving right in, just to kinda bring everybody up to speed on our most latest financial numbers, on Q4 and our outlook for Q1 of 2025. Our Q1 revenue was $100.3 million, which was in line with our guidance range. Notably, data center rose about 8% sequentially, largely driven by increased demand for 400G products.
I'll talk about our product mix between 400G and 800G as it evolves here in a little bit. What's driving that is not only organic growth in the 400G business among certain of our large data center customers, but also new customer additions that we have and more that we're expecting to bring on in the next few quarters. The other portion of our business, which is not something.
I'm gonna talk a lot about at this show because this is more data center focused, is that we do have a large business in cable television, specifically in amplifier products that are used to enable high-speed broadband over cable networks. That business more than doubled sequentially, largely driven by shipments of 1.8 GHz DOCSIS 4.0 capable smart amplifiers for a number of network upgrades that are going on in North America.
Again, I won't talk much about that at this show, but that is a significant portion of our business. On the gross margin side, we had gross margin about 29%, versus 36% in Q4 of 2023. A large part of that decline has to do with certain non-recurring engineering projects that we had the prior year that come in at a very high gross margin. Some of that, you know, apparent downfall was really just related to the difference in margin profile between those NRE projects and our traditional manufacturing model.
Revenue for 400G products, as I mentioned a moment ago, was a highlight, an increase of 40% year over year, 17% sequentially. That's in line with the long-term trend of growth in 400G as customers grow out their latest generation of infrastructure for their large data centers.
At the same time, you know, we've seen a gradual, I would say very gradual actually, decline in 100G, and that's pretty much what we'll expect to continue to see is flat to perhaps slightly decreased demand for 100G, coupled with growth in 400G and 800G. I will talk about that, you know, the trends that we expect to see in 800G shortly. We have made a lot of progress on our 800G, as I noted in our earnings call a few weeks ago. You know, a number of customers are beginning to give us their, I would say, very clear demand signals, specific forecasts, you know, specific part numbers, specific product types, and forecasts that are associated with that.
All those forecasts are aligned with, you know, our production plans, which I'll show you in a few minutes, in terms of, you know, how we're planning to ramp that production. Basically, at this point, our ability to deliver 800G is kind of what's gonna set the tone for our 800G revenue. Right now, we have very limited production capacity for 800G. As we bring on that additional manufacturing capacity, you know, we expect that our 800G revenue will continue to grow throughout the year. I'll give you some snapshots on that later.
Again, on the cable TV side, and I think this will be the last word I'm gonna say on cable in this presentation, really, is that we did receive our first orders for our new DOCSIS 4.0 next generation 1.8 GHz smart amplifier products from a large North American MSO. That's the beginning phase of their upgrade program that will eventually culminate in upgrading a substantial portion of their network to DOCSIS 4.0 capable networks.
For the first quarter of 2025, our guidance, sort of bracketing that $100 million range, so $94 to 104 million, we're guiding for gross margin between 29%-30.5%, which should give us a non-GAAP net loss in the range of $3.6 million to basically break even, and a loss per share in the range of $0.07 to essentially break even, using about 49.6 million shares outstanding.
We did note in our earnings call, just to, you know, bring everybody up to speed, that we do expect to make sizable investments this year, especially over the next several quarters as we expand production, largely for 800G and 1.6T, although we will also increase our 400G capacity somewhat. We expect to spend between $120 million and $150 million in total CapEx this year. You will see why in a few slides.
Again, you know, to kinda take a step back, that was our most recent quarter performance. Why is this happening? You know, what is fundamentally driving our demand? It is really growth in data centers, growth in internet connectivity, and to a lesser extent, growth in telecom and FTTH interconnectivity as well. We do have products in those areas, mostly legacy products.
I won't spend a lot of time talking about those markets here because we're really focused right now mainly on the data center at this show and in cable TV and data center overall for the company. Speaking of the driving factors for those markets, you know, in cable TV, I mentioned a moment ago, we do see some strong demand coming from network upgrades in the cable space, again, to enable mostly higher bandwidth in the upstream or return path direction. There's also a number of new opportunities that we see in cable TV nodes and optical devices for cable TV as these network upgrade projects roll out.
On the data center side of things, I think this should be probably no stranger to those of you who are at the show and probably most of you who are listening to this on the webcast. There are some significant growth factors associated with data centers, largely surrounding generative AI and, you know, the training and inference that occurs with those AI networks.
It has major implications. The architectures to most efficiently optimize, specifically the training of these neural networks, has some major implications on high-speed connectivity between the network elements that make up these training networks. What I mean by that is, most of our customers are intending to roll out very massively interconnected clusters of GPUs that need to have very high-speed connections between them. That's the most efficient way to optimize the utilization of those very expensive GPUs.
and it's the most efficient way to architect their networks from a power perspective as well, which is, of course, critical in these emerging AI networks. We're going from a model where we have sort of smallish clusters of compute, interconnected at, you know, relatively lower speeds, 400 gigabits or, or even below, to a world where we have sort of massively interconnected thousands, tens of thousands, even hundreds of thousands of interconnected GPUs, all of which, in order to facilitate the interconnection of those GPUs, have to be connected at very high speeds, 800 gigabits, moving to 1.6 terabits per second in the not-too-distant future. That's the trend that we're seeing. You know, our customers are, are largely the companies that you all think of when you think of, you know, modern AI infrastructure.
The large hyperscale data center operators, they are the ones that tend to build out these large, massively interconnected networks. They are the ones that are our primary customers for our current generation of products and future generations of products as well. Because of the speed at which this technology moves, there is a fairly continuous need to expand, replenish, and refresh the infrastructure that allows these devices to work. We do not think that this is a trend that is sort of one and done, right? I think those of us who use AI in our daily life know how instrumental that can be to facilitating, you know, efficient operations of our daily lives.
I think that's only gonna grow in the years ahead as the capabilities of AI continue to expand and as the ability of the data center operators to deliver the infrastructure that facilitates greater and greater AI applications really proliferates as well. Interconnectivity is very much a key part to making that happen.
Without the high-speed interconnectivity, you simply can't interconnect, you cannot massively interconnect the number of GPUs that are needed to run some of the models that they're gonna be needing to run in the future to enable the type of services that we all would like to have. You know, we think that our optics is not just a sideline, but actually a pretty critical part of, you know, what it takes to build these cost-effective and very capable AI data centers.
I mentioned a moment ago that we have, sort of legacy business in telecommunications and fiber to the home. I won't talk a lot about that. Those are areas that we've spent, you know, significant time developing technology, over the years. There are some, trends that we expect to see, in those areas that would contribute to growth down the line.
Realistically, over the next, you know, year or two, most of our growth is really gonna come from the data center and cable TV markets. Specifically on the data center market, this is some figures that came out of Omdia, a few months ago. It was in November of last year, showing that the total high-speed market, so 100 gigabit per second or higher, data center products, data center optical products are expected to exceed $25 billion in 2026.
That gives you a CAGR of about 22% overall. If you focus on the 800G, in particular, because that really is the specific part of the model that's being driven by AI compute, especially, that 800G CAGR is about 52% from 2023 to 2029. Very strong growth trends in this market.
Again, trends that I think are well-supported by, you know, our own intuition based on, you know, what we're hearing in the popular press and probably what most of us are using in our daily lives in terms of AI applications. That's, you know, sort of our market and customer information. We'll talk a little bit more about, you know, some of the customer trends that we're seeing in a few slides. I did want to give everybody a snapshot as we sort of transition to talk about AOI's role in this.
It's nice to be a technology company. It's nice to have, you know, 28 years of developing increasingly more advanced optical devices. Technology for technology's sake isn't really all that helpful for us. Our customers are demanding of products, good products that operate reliably and manufacturing infrastructure that can be scaled, reliably. AOI has spent a long time, you know, developing our manufacturing capabilities across the board.
I think we're well-positioned now to, not only, benefit from the trends that we've talked about in terms of increasing utilization of advanced optics in AI, but also to be able to transition to a US-based manufacturing model over time, which we think is a significant advantage, with some of our US customers who are looking for, you know, solutions, given the uncertain geopolitical environment that we're in today.
We do have manufacturing locations in three sites: Houston, just outside of Houston in Sugar Land, Texas, which is where our headquarters is. That is where we were founded, back in 1997. Ningbo, China, which is our largest manufacturing operation currently. We have about 2,000 employees there, and a little bit over 1 million sq ft of manufacturing space. Taipei, Taiwan, closing in on 1,000 employees.
This number here is 870. We have actually grown a little bit since the time this slide was prepared. We have about 350,000 sq ft of manufacturing space there. As we grow our footprint, the advanced manufacturing, the automated manufacturing that we are talking about, has been and is continuing to be installed in Taipei and will be added in Houston over the course of the next few quarters. I will talk about some of those specific plans here shortly.
We also have two other centers of presence geographically, one here in San Jose. That's a small center, just three employees basically doing sales and sales support for some of the customers here in the Bay Area. We have a sizable presence in the Atlanta area up in Duluth on the northeast side of Atlanta. That's where we do most of our cable TV product development, and some of our sales and support functions for the cable TV market that we're in.
As I mentioned at the outset, you know, while this is a technology show and you're gonna hear from us and from a lot of other companies about new solutions, new technical solutions that companies are offering to enable higher-speed interconnection, to enable bits essentially to be transferred from one network element to another within a data center, let's say, at higher speed, lower cost, et cetera.
Our contention is that technology for technology's sake is not that helpful. If you are a technology provider, it's equally important or perhaps even more important to have a manufacturing model that can support rapid scale-up and can support, you know, time to market that the largest hyperscale customers demand. These are customers that are used to getting products that they need very quickly and having product development cycles that can run very, very quickly.
and manufacturing models to support them need to be quick and flexible, and ideally, you know, able to scale without long periods of ramp-up in that scaling process. Because if there's a long period of ramp, you know, that reduces your time to market or, you know, decreases your time to market and makes it much more difficult to satisfy the demands of these large data center operators as they evolve. you know, I had a meeting just before this one with one of our large data center operators who was, you know, kind of lamenting the fact that really their forecast visibility, it goes out a few months.
It goes out beyond that at a baseline, but they're not even sure internally exactly how much demand they're gonna have, you know, going beyond that because they know they're gonna need at least this much, but there's so much extra potential out there that they see that they're just not really sure they're gonna be able to get all of the, you know, data center elements that are needed in place at time to build all those networks. There is a lot of work being done, not just in the optics business, but in all the other parts of these large data center networks that need to get built to be able to make sure that all the pieces come together to be able to build these advanced data centers.
You know, that's really, I think, the challenge for this industry is how to scale up, quickly and flexibly enough to meet the demands of these large data center customers. For us, our viewpoint is that manufacturing needs to be largely location agnostic, okay? A couple of years ago, this was not a factor. Nobody really cared where your stuff was manufactured as long as it could be done reliably, as long as it could be scaled up.
You know, how it got done, you know, where it got done was not that critical. Given all the changes in the geopolitical environment that we've seen, the location of manufacturing now becomes important, not only as a result of, you know, potential tariffs or other trade barriers that drive up costs, but also just because of supply chain integrity, right?
We saw during the pandemic, or many of our customers saw during the pandemic, that different geographies, different manufacturing locations, were able to deal with the perturbations that were caused by the pandemic more effectively than others. In some cases, you know, some areas really did not deal very effectively at all with those supply chain disruptions.
The lesson there for our customers was that they need to be paying a little more attention, I think, to where their products get manufactured and by whom, and what process, you know, the manufacturing goes through, because those all have implications in an uncertain environment where nobody really knows what the next supply chain disruption could be. You know, working with companies that have geographic diversity and the ability to build products using a differentiated supply chain is very important to them.
We also think manufacturing, obviously, you know, table stakes for any kind of advanced manufacturing process, it needs to have quality and reproducibility. As you scale in particular, you know, building one product is one thing, building 100 is another thing. Building 100,000 or a million is a whole different ballgame altogether. Doing that with quality and reproducibility is essential for our customers. We think that our process, you know, needs to be able to meet those demands. It needs to be highly scalable, as I mentioned. Demand profiles from customers can change dramatically.
and even though we're expanding our customer base by adding multiple different hyperscale customers so that on, you know, on an average basis, we're gonna hopefully see less perturbations or less, you know, cyclicality in the market, it's still very important for us to be able to adapt to those changes because our customers themselves need a supply chain that can be, very flexible and adaptable.
On the flexibility standpoint, you know, as we talk about automated manufacturing, which is what I'm gonna talk about over the next few slides, flexibility has to be designed in from the beginning. If you're, if you're building a, a robotic assembly line, it's not easy to change that process as you go from 400G to 800G to 1.6T to whatever the next generation, 3.2T or whatever it is.
You have to design your products on a platform such that over generation and next generation and next generation, you can utilize essentially the same manufacturing techniques. If you have to reinvent the wheel, you know, every time data rates change, again, it's gonna affect your time to market. It's gonna affect your ability to scale. And that's not the manufacturing type of process that our customers are demanding.
And of course, all this rolls up into rapid response to customer and market demand. That's ultimately what we and all of our competitors, need to be able to provide to these large hyperscale customers. Specifically talking about, you know, some of our automation capabilities, I'm gonna, I'm gonna show you some of those capabilities here over the next couple of slides. This is all on the web, by the way.
I see everybody taking pictures, but it's all gonna be on the web. Feel free to take pictures or not, but you'll be able to get it either way. I mentioned earlier that, you know, for us, the automation starts with our ability to have a platform that can scale flexibly from one generation to another. The ability to design a multi-optical channel transceiver that can operate at 400 gigabits per second, 800 gigabits per second, 1.6 terabits per second, et cetera, et cetera, as the technology evolves.
Doing that on a platform basis so that we can meet the demands of those successive technology generations is really, really important. We spend a lot of time working on this, you know, platform technology. The advanced manufacturing process itself has to be highly integrated, and has to have, you know, very high precision elements.
A lot of these technologies, a lot of these manufacturing steps that we have, require extremely tight tolerances, and they require you to hit those tolerances reproducibly, you know, time after time over millions of parts that are produced. Developing a process and a set of equipment that can do that is certainly non-trivial. AOI has our own in-house manufacturing automation development team, some 200 engineers almost that are focused on developing the equipment that's used in these.
I'll show you, you know, some of those equipment, and some of the stuff that we've developed over the next few slides. It's really important that when we're developing these new automated technologies that we maintain the precision and reliability of that manufacturing process that we've developed. Critical, of course, also is the inspection process.
No matter how good your manufacturing elements are, you know, mistakes do happen and variances occur. Being able to inspect those products rapidly and on an automated basis is also something we spend a lot of time doing. We have, you know, significant manufacturing experience in this type of platform technology over multiple years.
That gives us a very large data set, which we've used to train AI models to be able to do these visual inspections in line in a very, you know, highly automated and reproducible way. You know, companies that do not have the experience with the same type of manufacturing and the same type of platform, you know, a lot of those, the data set to train these neural networks just does not exist.
We think that that data set itself is a significant advantage, as we move to, you know, more automated inspection models. In process material handling, again, both for cost reduction purposes, that is, you know, labor element reduction, as well as for reproducibility and scalability, we need the ability to move from one process step to another efficiently, and in a way that does not involve a lot of human contact. It may, you know, I do not know how many of you have seen a modern optical manufacturing facility, but a lot of times it is characterized by discrete process steps. You have got your wire bonding, you have got your die bonding, you have got, you know, various different process steps that go on.
is kind of like there is a machine to do this, and then there is somebody that carries it from this machine over to the next one and so on and so forth. Our process is a very highly integrated process from beginning to end, and it utilizes some carriers and some magazines and things like that that we have developed that allow you to transport product automatically from one station to another.
Again, we are avoiding human labor content, and we are avoiding the possibility of having human error enter that process by, you know, just simply the transportation from one step to another. Finally, testing. When it comes to capital expenditures, the test equipment, as technology advances, speeds go up. Test equipment tends to go up very, very rapidly, with newer data rates.
In particular, at the very early stages of technology development, that test equipment can be very, very prohibitively expensive. We have developed a proprietary technology that increased our test station throughput by about 20 times. It really dramatically reduces the per unit cost of testing and allows us to scale that testing as well, in a cost-efficient way.
This slide goes into a little more detail on some of what I have talked about. There are 17 different process steps, you know, kind of major grouping of process steps involved in the manufacture of our transceivers, ranging from, you know, eutectic or soldering type of processes all the way through various different sealing processes, coupling, optical coupling, and so on and so forth. You could see those 17 different processes listed here.
The ones that are marked in purple, that have the little purple rectangle next to them, those are processes that were the equipment that we used to do that was in-house developed, okay? Basically, 14 out of the 17 different types of processes here were developed in-house. We've spent eight years, and I mentioned earlier, you know, almost 200 engineers that we have that are currently working on this.
I'd like to say, by the way, that, you know, we were very prescient and that we knew that at some point, you know, global trade barriers were gonna become a thing and we were gonna have to look at where to optimize our manufacturing. Reality is, we've done manufacturing in China since 2006. As time went on, we realized that the labor cost there was growing dramatically. There was a lot of labor inflation.
and so we started developing automated manufacturing back, really in the 2016 timeframe, specifically to address the then, you know, rapidly inflating cost of labor, not so much because we had an intention of, you know, eventually moving to the United States or some other geography, but just simply to control the rising cost of labor. what we realized over time, particularly during the pandemic, was that automation had other advantages besides just, you know, kind of allowing us to control the labor inflation.
It really had advantages in how we do manufacturing. I've had customers tell us that, you know, on the back of the pandemic, the reason why they continued to do business with us, one big reason was because we were really the only company that never decommitted on their optics purchasing during the entire pandemic period.
We were always able to deliver what they needed. A large reason for that is because we had, you know, the highly automated process. It was not as automated as what you are gonna see here, but it was still pretty advanced, you know, for its time. Since that pandemic period, we have only increased the amount of automation that we have had. We have applied it to, you know, almost all of the process steps that are involved.
Again, all this equipment is proprietary to AOI. It is not something you can go buy, you know, off the shelf. It is something that we have had to develop over, you know, an eight or nine-year period. We think this is a key element that now enables us to move manufacturing back to the United States and to do it cost-effectively and scalably for our customers. I am gonna show you a video here.
The audio was not unfortunately working, but you are not missing too much because it was really just a little bit of, you know, side music. We will put a little background music just so you can see it. It is going to run you through just a very basic snapshot of what this manufacturing process looks like in our factory.
Before I play it, I just want to emphasize everything you are seeing here is actual, I mean, this is video that was taken recently in our manufacturing plant in Taiwan, you know, in actual production of transceivers. It is not a cartoon. It is not a, you know, AI-generated video. It is real video of, you know, our manufacturing process and operation. I will go ahead and play the video now. For those of you online, you should see it pop. There you go.
Again, it's probably a little bit hard to describe. Some of you in this room and some of you online may have had the opportunity to visit an optical manufacturing plant. I'm sure most of you probably haven't. You may be sitting here thinking, "Well, isn't that the same thing that everybody has?" I can assure you, not based on my own observation, 'cause I'm not lucky enough to get to visit all of our competitors' manufacturing operations, as you can imagine.
Our customers have visited those operations multiple times and ours as well. The feedback that we've gotten from across the board from all the customers that have visited has said, "You guys have the most automated, most advanced manufacturing process that we've seen of anybody for these types of optical modules." Okay?
We're very excited, you know, to be able to show a little bit about that. I know it's probably less meaningful to you maybe, but hopefully you get a flavor for the extent to which, you know, these processes are really highly automated. I don't know if you caught that right there at the end of the video, but even things like putting the dust cap in, putting the label on, putting it in the pick, the packing box is automated.
What that's allowed us to do is, reduce the manufacturing cycle time by more than 35%. And perhaps, you know, more important even is that we've decreased our labor hours, by close to 90%, for the manufacture of these. You know, 90% about of the labor is, is gone from these.
If you can do that, if you can reduce the labor content by 90%, then you can feasibly manufacture products like this in the U.S. I say that not only because, you know, labor is more expensive here than it is in other places, but frankly, the type of labor that was needed to manufacture these modules some years ago just is not present in the United States in the quantity and concentration that would be necessary to scale up a factory that can produce hundreds of thousands of optical modules per month. It is just not that type. We just do not have that type of, you know, manufacturing labor available to us here like we do in other parts of the world.
It is really, really important to us to have a process that can be scaled without having to add a lot of people. And we have been able to accomplish that as well. While we have done the reduction cycle time and the decreased labor time, we have also managed to dramatically reduce defects during this process. DPPM, defective parts per million. I will not give you the exact number, but it is significantly below, you know, 50 defective parts per million.
Fifty DPPM is kind of an industry standard for sort of, you know, good quality control during your manufacturing process. We are well, well below that with this automated manufacturing process. John asked a good question a moment ago during the video. You know, was that Taiwan or was that Houston?
I wanted to talk about, you know, our kind of plans throughout the rest of this year in terms of capacity additions. I'd love to give all of you guys who are trying to make your models the specific number of units that we're producing. For obvious reasons, I'm not gonna do that. I did want to give you a flavor for how that capacity is gonna evolve over time.
What I did here is I took a baseline number, whatever we're producing in March 2025 for the Taiwan numbers, and, you know, whatever our beginning volume in September 2025 is for our US production. I'm showing how that evolves over time. What you're gonna see is that, you know, cumulatively, by the end of the year, we'll be producing about eight and a half times what we're currently producing in Taiwan.
In Texas right now, we have negligible manufacturing capacity specifically for 800G. Once we start in the third or fourth quarter of this year, we'll pretty rapidly expand that up to about four times the capacity that we have. That's our plan for this year. We obviously have plans to grow that capacity more as we go forward. I wanted to give everybody kind of an idea on how this evolves over time and where it's evolving to, to answer John's question about geography. We have, as I mentioned a moment ago, a small manufacturing capacity for 800G in the U.S. right now. We do have a reasonably sizable manufacturing capacity for 400G active optical cables in the U.S. right now.
It is not like we have no experience doing this manufacturing here, but this will significantly expand the capacity of 800G in particular that we can produce in the U.S. That is the plan for this year. As I mentioned at the outset, I wanted to spend most of the time talking about our manufacturing technology, but I will just mention a little bit on our technology roadmap as we look towards the next couple of years.
I have looked at this in a couple of different ways. One is sort of the optical data throughput per lane, so moving from 100G per lane to 200G per lane, eventually to 400G per lane. I have also, on the bottom, tried to highlight the kind of technologies from an optical standpoint that are relevant here.
You know, vertical cavity surface emitting laser technology, and silicon nitride type photonic, silicon photonics modules. I did not mention here the EMLs and some of the edge emitting lasers that are also used in this, but certainly that's a technology that's being used today as well. Predominantly wire bonding type of modules, but also some flip chip.
As we move to, you know, 200 gigs and particularly by 400 gigs per lane, we expect our viewpoint is that more of this is gonna go to silicon photonics. It's gonna go to probably newer material systems that are higher speed and enable the 400 gig per lane silicon photonics technology. Certainly when you go to those kind of speeds, our viewpoint is that it's gonna be largely flip chip assembled.
You can imagine, as we move, you saw in the video that, you know, a lot of our products currently are wire bonded as we move into a flip chip realm. You can imagine a lot of those automation engineers are busy at work today, developing the automated manufacturing for the flip chip type, models that will evolve over the next couple of years.
The other nice thing about this automation capability that I'll just mention before I move to sort of, a few more financial numbers. I'm gonna hit those pretty quickly because I wanna leave enough time for Q & A here. The same type or similar types of automation capabilities that we've applied to our data center products can also be applied to our telecom products, our FTTH products, and even our cable TV products.
Even though those are somewhat different, the knowledge and experience and capabilities that we have in designing this automated equipment can be utilized in other areas as well. We are looking at ways we can leverage that experience and those, you know, the automation capabilities that we have and use them across our product lines as we move forward. It is just an area to be aware of as we move into the future.
Finances, I'm gonna go through these pretty quick 'cause again, I know you have a lot of questions and all of you can access our financial numbers both online in this presentation as well as in our various public filings. We did end last year with about $249 million in total revenue. That was about 15% growth, compared to the prior year.
It's the highest year in revenue that we've had, really since 2020. That was an outlier. Really, 2019 was below that. It's been a while since we've had a year this good. You know, we believe based on the trends that we're seeing that this year will be significantly better, in terms of growth. I talked earlier, you know, Q4, Q1 are gonna be relatively flat with each other.
The reason why we're not seeing bigger growth in Q1 really just comes down to our ability to manufacture. We didn't have enough manufacturing capacity in Q1. We're bringing more capacity online. As we ramp that additional capacity, we expect that number will continue to grow. I went through the Q4 numbers again. I'm not gonna really highlight that too much.
Balance sheet wise, the only thing I'm gonna say on here, you know, we did talk about at the outset, a CapEx figure between $120 million and $150 million. You're looking at our balance sheet saying, "Look, you know, we've got $80 million in cash on the balance sheet." Clearly, you know, we're gonna work on methods to finance those expenditures.
We are looking at a variety of different options, you know, some bank facilities and other things that we'd like to do, as well as, you know, the ATM program that we announced earlier this year. Really with that, I think that brings us to just about 20 minutes before we're done. I want to leave five minutes or so for everybody to get onto their next meeting. That will give us about 15 minutes for questions.
I'll kind of wrap it up there and just, you know, sort of the floor open for questions. There are a couple of microphones in the back. If you have a question, just go ahead and raise your hand. I'll try to call on you and we'll get one of the folks with the microphone to come out. Yeah, Simon here. All right.
Thank you very much. Simon Leopold from Raymond James. I'd like to ask two, one of which is you had a recent announcement of a warrant agreement with Amazon. Maybe give us a little bit of sort of the backstory of the motivation and why that's happening. The other topic, which I think we're required to talk about in every meeting, is CPO. I'd like to kind of get your take on what the implications of that technology is on your business. Thank you.
Sure. You know, on the AWS warrant agreement, you know, for obvious reasons, I can't talk about too many details that aren't, you know, already in the public filings, which I'm sure most of you have already read. To give you a little, as you mentioned, sort of a backstory or a little flavor for what's going on here, you know, from what they've told us.
I think the folks at AWS are very interested in forming a deeper relationship with certain companies, certain technology providers, certain vendors that they view as having expertise in things that they're gonna need in the future. Some companies just look at that as a procurement opportunity to go out there and just, you know, find whatever's available and they're happy with that.
Other companies like AWS have taken a more active role in saying, "Look, you know, we not only want to be a customer for these products, but we wanna have a deeper relationship with you, both financially as well as," and, you know, this is the part that isn't totally apparent, but on an engineering and supply chain level, the same level of intimacy, if you will, that is implied on the financial model by being shareholders in AOI is also occurring on those other levels as well.
That is, we're, you know, I would say, increasingly, you know, more closely tied with AWS, and they're, you know, on an engineering level, on a procurement, on a supply chain level, really talking about, you know, their needs, their plans, our capabilities, and how we can evolve together to deliver the products that they're gonna need, not just now, but for years into the future. If you, you know, read that warrant agreement, you know, it's a 10-year timeframe. It specifies up to $4 billion of revenue over that 10-year period. You know, again, it's in the public filings. The warrants that are implied or, you know, offered by us as a part of that agreement are completely contingent upon revenue generation.
Other than a first initial tranche that was already granted at the signing, the rest of those warrants will only, you know, will only be granted as revenue rolls in. By the way, they're at a strike price that's, you know, significantly higher than the current market price. You know, it, it, there's no economic benefit to AWS now until revenue starts to roll in and presumably, you know, the stock price would, would go up, reflecting, you know, that additional revenue. I, I, yeah, I view it as a very positive sign, you know, that they're interested in, in, taking this relationship to a deeper level.
and I can tell you, you know, as we've begun or continued to work with them more closely during the warrant negotiation process and, you know, the parallel technology work that was going on, they've acted very much like they're interested in, in being, you know, a long-term partner for us. I think we're very excited about that opportunity.
The other question you had concerns CPO. Since we have, you know, whatever, 16 minutes, I could probably spend 16 hours talking about CPO and what that entails. I'll just say a couple of words on it. Number one, I think CPO is a very exciting technology. It's something that, you know, we and others have been talking about, at this show at OFC for a long time already.
It does appear that there's movement, and more interest about developing CPO technologies. I would say that, you know, our large data center customers are interested in that technology. I think it has promise. I can tell you, just based on my experience with these operators, they are very, very careful, rightfully so, in deploying any new technologies in their network. You know, these companies have massive footprints. They're trying to scale up new AI applications that absolutely depend on very, very high reliability in their networks and very high performance in those networks.
You know, while there is a lot of interest in CPO, my gut feeling is, based on my experience, you know, again, 28 years in the optics business and, you know, 12 years working with these large hyperscalers, that, you know, no matter how interested they are in a new technology, they do not just dive into the shiny new thing. They move very slowly and deliberately. I think it is a technology that is out there. I think it will be a part of networks in years to come. I do not think it is a major factor really over the next few years probably. Thank you.
Hi. Thank you, Chief Yongpo from BNP Paribas. I'm new to the story. Just, first a quick clarification. I saw your Houston facility makes laser chips. Can you tell us, like, what type of laser chips you make and to what extent that you are vertically integrated?
Sure. The question is, you know, in Houston, do we make laser chips and to what extent are we vertically integrated? Yes, I didn't really touch on the manufacturing of laser chips in the presentation here. That has certainly been a core element of our manufacturing in Houston. We do have a laser fab. It's an indium phosphide fab with some gallium arsenide capability, currently operating on three-inch indium phosphide wafers.
We're moving to a four-inch wafer. That gives us the ability right now to make about 1.4 million lasers a month. By the end of the year, we'll probably be scaling that production up to about 2 million lasers a month. Those lasers can be either CW lasers for silicon photonics for pumping silicon photonics platforms, or they can be electroabsorption modulators or any other combination of those.
VCSELs right now, we currently have very limited production. That's a gallium arsenide-based technology. And we have currently rather limited production of, you know, VCSELs in Houston. We do have, you know, plans to expand our production capacity for VCSELs in Houston, but it really won't be until later in the year or early next year before any of that starts to come to fruition.
Your question was about how vertically integrated are we, okay? We make our, you know, our own lasers and our own fab there. We do not make a silicon photonics platform ourselves. We do have a team that works on designing it. And we have some companies that we also work with on designing. Obviously, you know, fabbing of those silicon products, we don't do in-house. You know, we source that out.
The video that you saw talks about all the assembly work that we do. You know, we're involved in, you know, procuring things like mechanical housings and fiber. We don't draw our own optical fibers or anything like that. Relative to the key optical elements, the lasers, the photodiodes, and, you know, a number of the mechanical parts and things like that that go into making these transceivers, we make most of those in-house.
Your expansion plan to expand 4X capacity, how much of that capacity expansion is based on, like, more com high conviction orders, opportunities, and how much of that is, is, like, more remote opportunities?
I guess the question is, you know, kind of what's our degree of confidence in, you know, the utilization of that capacity? I would say it's very high for two reasons. You know, number one, I have customers that have literally told me, you know, "If you can build it in the U.S., we'll buy it, all of it," right? Even if that does not come to fruition, we have enough customers that are interested in seeing at least an increment of manufacturing done in the U.S. that I'm not, I'm not at all worried about being able to utilize, you know, the amount of capacity that we're trying to build here.
I know there were no absolute numbers in that, but let me assure you, you know, those numbers are small enough that they still represent, you know, a very small fraction of the overall data center market. We're not adding enough capacity to address, you know, even a high single-digit % of the market.
What we're doing is we're building an amount of capacity that would allow us to concretely demonstrate to all the customers out there that these can be made economically in the United States. Our plan for 2026 and beyond would be to further scale that up. Again, first we've got to show, just like our customers move very methodically and slowly into new technologies, we're doing the same with our manufacturing.
We're gonna add an increment of manufacturing, scale that up a little bit, show customers that we can do that, wait for the next, you know, customer commitment before we add, you know, the next increment of capacity. It's gonna be a slow and deliberate process for us. Another question up here.
Thanks for taking my question. Just two minor clarifications on the AWS warrant agreement. Is there anything that's contingent to the revenue streams or the opportunity there, is that contingent to us being able to make everything internally or there's no, it's not binding that everything must be done internally? How does the margin profile, how that plays out over the longer term?
The other part of that is, let's say that if there's a component that's under shortage, let's say like DSPs are under shortage, do we get, do we get any extra help from, you know, the relationship or will Amazon go out and say, "Hey, look, we'll help you get the, you know, your Broadcom or Marvell DSPs if you need it," if that's the one they want?
Sure. So, you know, two questions in there. You know, number one, is there a particular, you know, bill of materials or sourcing strategy that needs to be employed for these? The answer to that is, not specifically. There is certainly nothing in the warrant agreement that talks about that.
There are a number of commercial agreements that are, you know, not public that are part of that whole package of agreements that concern specific individual products. And some of those do have, you know, requirements that certain components either be made externally or internally or whatever because they have, they are trying to make sure that they have a diversity of suppliers across, you know, all the different companies that they have supplying different parts. So some of those do have a little bit of, of that.
In general, you know, we have a lot of flexibility in choosing, you know, vendors and partners and manufacturing models for these products. I mean, the way that our customers have explained that to us before is, you know, "You guys are the experts at this. We don't wanna have to dictate every element of the design to you."
They do, you know, most customers now are, you know, more concerned than they ever have been about who we're sourcing stuff from because they wanna make sure that, you know, whatever supply chain we use is gonna be viable for them moving forward. There is a lot of back and forth and a lot of, you know, common discussion between our engineering team and our sourcing team and theirs to make sure that we get this right.
Ultimately, we have a lot of influence on that. Your second question, which I forgot, I'm sorry. Could you repeat that?
Do they help us, you know, in any part of the sourcing? I guess there's, like, a market share situation.
Again, I can't really talk about the specifics of the agreements in terms of how they're gonna source. Let's just say that, whether there's, you know, direct, formal help, that is, whether they, for example, already secured a supply of critical components and they're gonna offer those to us, or whether it's more of an indirect thing. Having this type of agreement with AWS, I think, gives us an ability to, you know, let's say we just get a higher profile with a number of the key vendors that would enable us to presumably have earlier and better access to their production capacity.
Is there any way to kind of differentiate, like, for example, your warrant agreement with them and also, like, Fabrinet's warrant agreement with them who, you know, they're also saying that it's also transceiver included as well, etc., etc.?
Yeah. I mean, I can't really tell you that. I don't know anything about the Fabrinet that's not public. You know as much as I, you probably know more than I do 'cause, you know, Fabrinet's probably more forthcoming to you guys than they are to me. I don't really have any insight into that.
I have been told, you know, by people, you know, that are knowledgeable about these things that it's not AWS's intention to have, you know, sort of multiple suppliers with a similar agreement for the same type of product. I mean, that would sort of put themselves internally at conflict. I have reason to believe that whatever Fabrinet is doing is probably, you know, different from what we're doing, but.
Thank you. My last question is, just want to double-check your 200G EML schedule, if there's anything that you're pulling in forward or just in general any update there. Thank you.
No, we're still on track for 200G EMLs, you know, in the second quarter of this year, so. Yep. Thank you. I think we had a question over here.
Great. Thanks. I guess just, another one on the AWS partnership. Just curious how we think about that, the procurement, and is it dilutive or accretive to kind of the overall data center gross margin as we think about that going forward? In lieu of this partnership, is there a pipeline that you can comment in terms of additional customers like this that may occur in the future, just based on, you know, all the advantages that you had talked about earlier?
To be clear, you know, the warrant agreement with AWS doesn't specify any particular, you know, margin profile or cost structure or anything like that. You know, whatever we ultimately agree on with them for cost, and some of that's already been agreed, but, you know, obviously, I can't tell you about it.
You know, none of that is mandated by this agreement. I would say, you know, if our cost structure is such that we can't feasibly manufacture product at an attractive cost, you know, they would have to evaluate whether it's worth to pay extra money. And probably the answer to that is, you know, if it's just to obtain the benefits of the warrants, they're probably not gonna pay, you know, a significant premium on products just to get warrants.
However, we have been told, not necessarily directly by Amazon, I'm not gonna attribute this only to them, but by multiple customers, that US-based manufacturing is very attractive to them and they are willing to pay a reasonable additional cost for product that's manufactured in the US because it has a differentiated supply chain. They are willing to pay more for and presumably would give us, you know, a somewhat better margin profile. Again, none of that is specified in the Amazon agreement per se.
I guess, quickly just enter your pipeline. The next one is just on capacity. Of the $45 million, $44 and change, of data center revenue last quarter, what's the mix of 400G? Have you clarified that? As we think about the 800G capacity coming online, when does it surpass, 'cause I know you can't give us the exact units, but when does it surpass kind of the 400G current installed capacity?
Yeah. The numbers and the evolution of our 400G business is covered. Every quarter, we give those numbers broken out in our earnings call. I can, you know, call your attention to that or I can guide you to those numbers, you know, offline a little bit and point you to where they're at.
They're in our earnings call every quarter as to the actual percentages. As far as when 800G passes 400G, you know, we haven't given a specific timeline on that, but it would be sometime later this year. Again, the reason why we're not producing more 800G now is just simply we don't have manufacturing capacity for much 800G right now at this moment. A couple more questions. There's one in the back over here, maybe over here.
Hi. Hey, on the, the 400G that you are shipping now in a AI data center, could you talk about where it sits? Is this, you know, just from bottom of the rack to top of the rack? Is it 30-100 meter kind of, applications or, or longer? And who would be the main competitors there? Thank you.
Yeah. It's in switch-to-switch interconnection. It's not really into a rack. It's between racks of switches. You know, generally speaking, those span distances from a few tens of meters out to a few hundreds of meters in our case. Most of those are single-mode devices, so they're longer distance. They're not vertical cavity surface emitting lasers, which would be used for shorter distances. As far as competitors out there, I mean, you can see them out at the show. There's a bunch of people that are displaying their wares there. Certainly, you know, any of the large, multifaceted optical suppliers, companies like Coherent and Lumentum would be strong competitors of ours. We got time for probably just one more question.
Okay.
Hello?
Yeah.
All right. I do wanna try to put a baseline underneath your capacity.
Uh-huh.
Discussion. You know, it seems like, depending on how you do the math, you're gonna end up with 5-10 times as much capacity. Exiting the year is beginning. It seems like, you know, obviously, you'd wanna exclude, I don't know if you'd wanna exclude 100G capacity from that, but there are a couple of ways of going at that. Either way, it seems like you should be at, from an 800G, 400G standpoint, over $100 million a quarter by year-end, which coincidentally would be your annual Amazon demand. Is there any relationship between those two numbers? How do those numbers sound to you for general reasonableness?
I think the answer to your question is, I mean, there's not a specific relationship. It's not like, you know, we have this capacity specifically for AWS and, you know, the math kind of works out that way, but that's not really the thought process behind it.
The capacity that we're building is not dedicated to any one specific customer in any sense. That's not to say that we don't have pretty strong assurances from certain customers that they're gonna buy, you know, a certain percentage of that. But it's not like a requirement that we're building it just for one customer or anything like that. We do have multiple different hyperscale customers, all of whom have said, you know, they like our manufacturing model in terms of, you know, eventually moving more in the U.S.
Even, you know, having manufacturing in Taiwan right now is seen as a more favorable place by a lot of these customers, certainly than China. Even then, places like Thailand or other areas, I think they generally like Taiwan.
Just to follow up real quick, would you be disappointed if you couldn't do, you know, to the extent that mathematically Amazon works out to $400 million a year or $100 million a quarter? Would you expect a baseline with everybody else about the same level or, or more? Is that a reasonable thing to think about?
I mean, I think it's a reasonable thing to think about. I don't wanna say that, you know, that's the guidance that I'm trying to give you specifically, but certainly, the demand in aggregate for the other hyperscalers certainly would allow a, you know, a production level that's the same as Amazon.
You know, whether we get there or not, difficult to say, but I certainly think that's the potential. You know, again, our goal here is not to be just Amazon supplier. And Amazon knows that. I mean, you know, they don't want that either. They want us to have a differentiated, you know, customer base as well. Obviously, they wanna put, you know, they want us to put their needs early on our priority list, let's say, and we're gonna do that.
It is not their intention to have us as sort of our captive, you know, single, single supplier just for them. I think, you know, that's not our plan either.
Just sneak in there with one more. This morning, Lumentum said they were gonna participate at Amazon on the laser side, not on the module side. Are you gonna be buying lasers from Lumentum, I guess, or just for the chase?
I can't comment on that. I can't comment on that. I have said before that, you know, we often source lasers, some percentage of lasers from, you know, other companies, mainly when our customer demands that we have a second source. All of our lasers that we make in-house come from one single fab in the Houston area.
You know, that is a single point of failure for the supply chain if we only have one source. Oftentimes, our customers will mandate or at least highly suggest that we have an alternative source. That generally would mean we'd have to buy, you know, you can't just have a source that you don't buy anything from. You've gotta keep them kind of in the loop.
I will not comment on Lumentum or any other supplier specifically, but it is not unusual for us to have an additional source of laser. Yep. I think our time is just about up. I am sorry. You had one more question real quick? Sorry. Yes. There is a strike price associated with them. It is 20, so they have to pay $23, it is $22.74, something like that, in order to get the warrants, which means they are currently underwater. There are lots of thought processes behind it. We are looking at lots of different ways to do it. Thank you. Thank you all for coming. And those of you who are listening online, I appreciate your attention and look forward to getting to know many of you better. Thank you.