Good afternoon, good morning, depending on where you are. Thanks for joining us here at the Rosenblatt Age of AI Tech Summit. I am Mike Genovese, the Cloud and Communications Equipment Analyst, and super pleased today to be joined by the executive team from Fabrinet. We have Seamus, Csaba, and Garo, the CEO, the CFO, and the Vice President of Investor Relations, here with us today. And this is going to be a fireside chat. I'm going to ask questions, but I'd also like to get questions from the audience. So you should all be familiar by now, if you participated earlier today or yesterday, there's a control panel where you can ask questions, and they'll come directly to me. And I always make a point to make sure to ask them in our 45-minute session here with the team. So first of all, just thanks for joining us.
Nice to see you.
Thank you, Mike. Thanks for having us. We appreciate it.
Great. Well, Seamus, to start, I mean, you know, I think that this is the Age of AI Investor Conference. There's probably, I'm sure, investors here who are focused on the AI opportunity. But I also want to give some sort of background on Fabrinet, so some history and current business. But maybe we can particularly emphasize how you fit into the datacom transceiver value chain and how we should think about you as being similar or different to other companies that make datacom transceivers like Coherent, Lumentum, and InnoLight. So very broad questions, just sort of background on Fabrinet, but more specifically then to beginning to touch on the AI transceiver opportunity.
Yeah, so thanks, Mike. I guess the first thing is to point out we're a contract manufacturer. Fabrinet's a contract manufacturer. We don't have our own products. We make other people's products. Specific to, let's say, the NVIDIA relationship, we've been working with Mellanox for about 10 years and have been their chosen contract manufacturing partner for optical networking since then. Mellanox was acquired, of course, by NVIDIA in 2020, and we continue to be the sole source contract manufacturer for their optical design. So we make NVIDIA's product, the product that they've designed themselves. So separately, they've approved optical networking products from Coherent and InnoLight to work in parts of their system, parts of their network. But those products are designed by Coherent and InnoLight, not by NVIDIA.
So that'll be the key difference is their OEMs who have their own products, and we're a contract manufacturer who manufactures NVIDIA's designed product. But like I say, our relationship with NVIDIA, it's probably about a 10-year relationship if you take the Mellanox part of that relationship into account. So while it might look like an overnight success, it's been the better part of 10 years in the making.
Is that other business that's not the 800G transceiver business, this new one that started a little more than a year ago in revenue? Is it all lower speed transceivers for the most part? Is that what 400G, 100G transceivers? Is that what it is?
400 and 800. 100 gig, we don't produce too much of anymore. 100 gig has become quite commoditized. So we produce typically 400 and 800.
Okay. Are you aware of, do you know if you're sole source at 400 as well?
Yes.
Okay. So you've maintained that. Okay. Right. Yeah. No, I mean, there's a lot more we'll come back on and touch on that. I mean, I did want to start a little bit broad, but I guess you dove right into what everyone is interested in, so I applaud you for that. But so if we broaden out just a little bit, I mean, if you go back over the last 10 years, I mean, the top line CAGR, I think, has been north of 15%, thereabouts, which I don't think that your competitors, and to be clear, we're talking about the contract manufacturers, larger EMS providers as your competitors, they're not growing nearly that rate. But I also don't think until maybe now with some of these AI opportunities, your datacom customers, your telecom customers weren't growing that fast either, right?
So how do you, I mean, how does Fabrinet consistently outgrow the market here? I mean, you've done it so long. There's got to be some kind of secret sauce to it, and you seem to be the person to ask.
Yeah. Well, I suppose a couple of things. The first thing is the secret sauce is that there is no secret sauce. I'll maybe explain what I mean by that. Just to point out the competition, yeah, the competition in the EMS space, the rate of growth has been, I would say, single digits in the same time period where we've managed to grow 15% compound for the last 10 years or so. That 15% in our case includes virtually no growth through acquisitions. We did one small acquisition in the U.K. back in 2016, which was a very small revenue. We subsequently shut that operation down. So there's effectively no growth by acquisition. It's been all organic growth in our case. Whereas many of our competitors, they have done fairly sizable acquisitions, and their growth, albeit maybe more modest, has come through acquisition, whereas ours has been organic.
Our aim is to grow, first of all, what do we target? We target to grow at a rate of about 2x the rate of growth of the industries we serve and 3x the rate of growth of the EMS industry. And as you rightly point out, our compound annual growth rate for the past 10 years has been around 15%. The basis for our growth and how we accomplish that is to focus on high-growth industries and the fastest-growing companies and technologies within those industries. If you capture the fastest-growing companies with the most difficult products and you capture those products very early in the product lifecycle or before they're even designed, that's really how we do so well. Our attention, again, is mainly on what we're currently producing for our customers. We're simultaneously working very hard on the next-generation products.
New products are a critical part of our success, and it's something we pay a huge amount of attention to. For example, we would always look to make sure we're on the roadmap for, again, not just the product that everyone is aware of today, but the next one and the next maybe two after that. We're usually two or three generations out in what we're working on with the customer. It's never something we can talk about publicly because customers are very secretive, and that's part of the service we provide is that kind of confidentiality. Not all new products are successes. Sometimes we work with the customer for several years, and at the last minute, they decide they're not going to launch the product. That's okay too. We don't get too excited when that happens.
But fundamentally, to have the kind of success we've had, it fundamentally comes down to execution and reputation. To win the next generation products, you have to have been doing an excellent job on the current generation products, and you have to be able to tick every box in terms of cost, quality, delivery, responsiveness, flexibility, relationship, all of the above. You have to be obsessed with customer service and obsessed with giving the customers what they need. Our industry, it's an interesting industry. It's called contract manufacturing, which the name would suggest that we operate, and we have a contract, don't get me wrong, with each of our customers. But generally, the way we operate is whatever the customer needs, we're here to service their needs.
The answer is typically yes whenever the customer asks us to do something to provide a crazy ramp or sometimes to push something out. We're very responsive, very flexible. We think we're the, well, we believe we're the technology leader in our industry because we're producing the most complex products. We also have benchmarked ourselves against our competitors. We also believe we're the lowest cost because we have a very compact footprint. We don't have redundant factories around the world that are a drag on our earnings. We keep things very compact and very tight. Really, all of our operations and all of our people have to be contributing. We keep our costs very low and our execution very, very consistent. It's really down to reputation and execution. That's what I mean when I say there is no secret sauce.
I mean, we don't have IP. We don't patent anything that we do. We just make sure we do it. We like to try to do it better than everybody else. And sometimes we succeed. We don't always succeed, but in general, I think our batting average is pretty good.
Thanks. Starting to get questions coming in from the audience. A lot of my questions that I have here are really going to dig in on the datacom opportunity specifically to NVIDIA, but also beyond. I'm getting some tougher questions from the audience on sort of the telecom side. I'm going to throw a couple of these at you now, so in terms of weakness in telecom, I guess the question is, is that a function of one or two customers, or is it really broad-based across the portfolio in terms of what you've been seeing in telecom?
Yeah, it seems to be broad-based. Of course, we don't have every customer. We don't produce every product for every customer. We do, I suppose, have a fairly unique insight in the sense that we're either producing for most of the companies in the space, or certainly the components and the products we make for our other customers is shipping into those customers. So I suppose we do have a kind of a unique perspective in that sense. The telecom weakness, that's inventory digestion, I suppose. It's been going on for over a year. It's pretty broad-based. If you look at our telecom revenue, if you go back five or six quarters ago, the high point for us in terms of our telecom revenue was $405 million in a quarter. And then last quarter, we shipped $286 million.
So we're down 24%-25% year-on-year in our telecom business. Now, we've still managed to grow at the same time because of our explosive growth in datacom in that same time period. So our growth in datacom came along at a very good time. But nevertheless, telecom is and continues to be quite soft. And I suppose the positive in a sense is I think all of the telecom headwinds are kind of behind us now. I think we're at a low point. When we get back to growth, I'm not sure. I hear talk about green shoots and things like that. We haven't really seen evidence of that in terms of actual demand yet, but I'm sure it will come, but it doesn't seem to be imminent. So yeah, it's pretty broad-based. It's across the industry. It's several companies.
One of the things we track is the inventory levels at the telecom equipment companies. We track that very closely when they report their numbers at the end of each quarter. For many of those companies, their inventory levels, as measured in days of supply, it's the best way to kind of normalize the metric rather than looking at the dollars. But when you convert to days of supply, inventory levels are remaining stubbornly high. Until the inventory starts to come down, we don't really think we're going to see much growth in the telecom business. That said, it does present an opportunity for us to try to win additional telecom business and maybe take some share away from our competitors. It's a good time to be knocking on the door to try to convince companies to bring on another source or a new source.
Yeah. Yeah. Sorry. Let me just ask the other question from the audience that we have here, which I guess is kind of a bigger picture question because you have, I think in the industry, right, you have companies. I'm not sure whether you have sort of periods of time where more companies are insourcing and more companies are outsourcing, or whether it sort of just goes company by company. Some are insourcing, some are outsourcing, and it becomes a wash. I mean, I think getting specific questions about Lumentum, who seems to be insourcing more. But what do you think about these issues in terms of your TAM and then whether customers are insourcing or outsourcing?
Yeah. I mean, in general, overwhelmingly, customers outsource. They generally don't have their own manufacturing these days. There's really a couple of exceptions to that. Coherent being one, Coherent have had a strategy historically of being vertically integrated, which of course we support. That's a good strategy for an OEM, but also of having their own in-house manufacturing. We think they could really reduce their cost and improve their working capital by outsourcing much more. So we'll be hoping to meet with their newly appointed CEO to impress that upon them as soon as possible. Lumentum, they have their own factories in Thailand. They have two factories in Thailand. And they have a kind of a hybrid strategy, I suppose. They insource certain products and they outsource certain products. And we've seen that strategy unfold over the last while. And that's an evolving strategy.
They've been very open about that with us. Our relationship with Lumentum is very good. We don't necessarily have to agree with all their strategy in order to support them in the furtherance of their strategy. So that's the approach we're taking. Whatever they need from us, we're happy to support them. Whatever they want to do, we're happy to help them in any way we can, even if it means that in some cases they're moving business from our operation into their own operation. That's their choice. They're their products. Again, our job is to execute on Fabrinet's strategy, not to opine on the strategy of our customers. So we're here to support them. We're here to help them in any way we can.
Yeah. Okay. Great. Well, I mean, if we go back to this datacom growth that you've been seeing, I know you're in the market of servicing your customers. So I guess from your perspective, I'm sure what you can answer very well is what are you doing to make sure that you remain the sole source of the customer's branded products? How do you sort of ensure that? But also any insight you can give us into just help people understand why demand for this product is so strong. Your understanding of what the customer is doing, I'm sure at a high level, just to help frame this market opportunity and sort of what's driving it for you. So two questions there.
Yeah. I think, yeah, we're the sole source, as I said, for the NVIDIA design product. We've been working with them for a long time. And again, what looks like an overnight success has been years in the making. NVIDIA, they started to really ramp the 800 gig transceiver products in March and June of last year, which is when we saw that impact on our datacom revenue growth. Mellanox and NVIDIA, historically, they've been comfortable having sole source manufacturers as long as the manufacturer is doing a good job in terms of and can produce at the required quality levels, volumes, deliveries, etc. NVIDIA is a very demanding customer, and we make sure that we're up for the challenge and that we meet their needs.
So our job is to make sure we're ultra competitive, very supportive in meeting all their needs, and that we work very hard to ensure that they have no reason to look elsewhere. The products themselves and where they're used in the network, I think the big difference and what's driving a lot of the volume is historically, the main use case, let's say, in a traditional data center for these transceivers would have been, let's say, rack to rack, connecting the racks to each other. Whereas in the case of these AI data centers, optical interconnect is used pervasively all through the whole system within, let's say, within the shelf, within the box, shelf to shelf, rack to rack.
So there's a kind of an order of magnitude increase in the sheer number of interconnects in any given data center, which is driving this kind of explosive growth in 800 gig. So that's really the background there, Mike, in terms of how we're the sole source and how we plan to really work hard, work really hard. We take nothing for granted. We don't have any rights to the business. We have to earn every piece of business we have, and we have to earn every piece of business and the right to keep the business. So we take nothing for granted. We work very, very hard to make sure we satisfy all of our customers' needs, not just NVIDIA.
Yeah. So for investors following the AI market, obviously, quite a few investors paying attention. There is this debate between what is the potential for a slowdown, an inventory correction, a digestion of lots of things that have already been shipped versus the idea that this is all going to be supply constrained for several more years to come. Whether we're talking about GPUs or we're talking about transceivers, that it's kind of up into the right sequentially into the future without any major downward lumps. So I mean, it's kind of a, I mean, there's people on both sides of that. And anyway, I don't want to say what side I'm on because I want to ask you the question of what you think about that debate.
Yeah. I mean, I suppose we don't really get involved in those type of debates because it calls for a lot of speculation and opinions and things like that. And we tend to, we prefer to talk about what we've actually done rather than speculate on. We're not industry experts. We're contract manufacturers. We're not industry experts. That said, it seems to us that the demand for these high-speed, short-reach, low-latency 800 gig transceivers is pretty insatiable right now. We would normally have 13 weeks visibility with our customers. We would have longer visibility currently with the products we're working on purely so that we can position capacity and components and everything else. And while we can't really talk about demand for specific products, we're confident that the demand looks to be very robust and looks to remain strong for the foreseeable future. There may be an inventory correction.
I don't know. That's not my sense of things. It doesn't feel like anything we're shipping is going into inventory. It feels like the cakes are being eaten hot out of the oven.
Okay. That's fair. Is it safe to say, I don't know if you would comment on this, but that this would still be the kind of fastest growing user of new manufacturing space, that this datacom and this application is consistently for the next and should be for the upcoming quarters, at least one or two?
Yes. I think that's fair to say. And I think if you look back on our success in the last while, if we didn't have the manufacturing space available in one place, it's one thing having space, but if it's dotted across three or four different geographies and everything else, you just can't support the customer. I think one of the things that was very fortunate for us and for NVIDIA is that we were able to ramp these products in very short order in one location with one team and really able to provide the customer the support they needed in one. So yeah, having capacity available is critically important. And we certainly have that.
Okay. Great. Well, look, I mean, I think that if we go back, as you said, sort of March, June last year, you were the first real company to show a ramp in 800G, which to me makes sense because it's those GPUs that were designed to be connected at 800G that needed it. And then there seemed to be a lot of hullabaloo over the next six to nine months from this conversation about this qualification of NVIDIA's added other vendors. They've added Coherent. They've added InnoLight. What does that mean? Does that mean that Fabrinet is losing share? I know you're not a market analyst, so however you're comfortable commenting on this. But really, why do you think the customer needed to do that?
Any insight into how sort of that business model differs from the business model that you have and sort of how the product gets to the end user and how the end user pays for it?
Yeah. So how the product gets to the end user part, whether, let's say, first of all, everything we produce, we ship to NVIDIA and only NVIDIA. We don't ship anything direct to NVIDIA's customers. We ship only to NVIDIA. Coherent and InnoLight, we actually don't know. Do they ship direct to NVIDIA only? Do they ship some to NVIDIA? Some to NVIDIA's customers? We don't have any particular insight into that. I guess the key thing to remember is NVIDIA, they're not an optical networking company. They're a GPU and a high-performance compute company. And for NVIDIA, optical interconnect is something that they need in order to be able to get the GPUs to talk to each other. But fundamentally, they're not an optical company. And their goal is to sell systems and secure multiple sources for all the components if they could, I suppose.
But if they had stuck to only having NVIDIA optical interconnect in the whole network, they probably would have limited their ability to ship GPUs. So I think it was a smart move on NVIDIA's part to bring on Coherent and InnoLight. And again, we still have a 100% share of the NVIDIA design product. But then Coherent and InnoLight, they compete for the merchant transceiver business. We don't really talk to NVIDIA about what's going on with their other suppliers or their strategy about when do they use a Coherent transceiver versus when they use one of their own. So we don't have any particular insight into that. Our focus is just on doing the best we can for the customer and getting them what they need.
I think we hit on this earlier, but it did come in from the audience. So I'm going to ask it because I asked you earlier if all of the other NVIDIA/Mellanox revenue was optical components. But I guess, I mean, more directly, I'll ask, is there anything else that you make for them besides optical transceivers for that customer?
We make. That's what we make for them. We're their optical, their prime optical contract manufacturer.
Yeah. Yeah. Okay. Great. I just wanted to make sure that I'm asking the questions that people ask in the audience. So I guess I get asked this question all the time, and I know how I answer it, but I guess I should ask you for people who have it. But when we talk about sort of the things that you're connecting or that your transceivers are being used to connect, which I think are servers and switches, basically, right? Servers and switches, and server to server, switch to switch, server to switch. When we talk about whether the switching protocol in use for what's going on there is kind of above the layer that you're at, is NVLink or Ethernet or InfiniBand? Do you care at all about that?
Do you make some transceivers that have a different shape for InfiniBand and Ethernet, or it's all the same, the same actual interface that gets plugged into the switch, if that makes sense?
From our perspective, it's the same. We're completely kind of agnostic on whether the, let's say, the protocol being used is InfiniBand or Ethernet. We really don't care. Spectrum-X and Spectrum-X Ethernet could well be a strategy that helps drive NVIDIA's continued success. But we really are agnostic about the underlying technology. We really don't care. We make all flavors of devices, whether they're InfiniBand, Ethernet, multi-mode, single-mode, AOC, or transceivers. We're just ready to support the customer no matter what the market is demanding of them. So we're completely agnostic about, let's call it, the protocol that's used in the transceivers.
So I assume, I mean, because you do do this, right? I mean, you make silicon photonics transceivers, I mean, high-end telecom things. I mean, if a datacom customer came to you, I mean, whether it was a GPU maker or hyperscaler or somebody like that, and they said, "We have a design. We have the intellectual property, but we want you to make us an EML-based transceiver or a silicon photonics transceiver." I mean, there's nothing about that you're limited to just making VCSEL transceivers.
Oh, no. No, not at all. We're very comfortable. Again, we're agnostic. We're very comfortable making all manner of transceivers irrespective of the technology. And we're kind of the leader in putting those products together and knowing how they go together. So yeah, we'd be very comfortable doing that.
Great. So let's talk about more. What is the opportunity there for having more than one customer in datacom? Should we be patient? How patient should investors be waiting for that opportunity?
Yeah. I think it's still, if you like, a relatively new, if you like, category, this optical interconnect for AI applications. So for us, obviously, we have the customer that if you were in my shoes in one of the other contract manufacturers, you'd want to have NVIDIA as your customer. So we have NVIDIA. That's great. And that's a great relationship. We're very happy with that. But of course, we're working to expand in that space. And there's really three, beyond NVIDIA, there's really three additional, if you like, AI growth paths for Fabrinet. And these are in no particular order. One is hyperscalers or anybody else could be hyperscaler, could be an enterprise company who decide to build versus buy. So if they decide to design their own optical interconnect and they need Fabrinet to produce it, we'd be happy to do that.
Again, provided that it's their own IP, we will not be a product company. We've been very clear about that. Some other contract manufacturers are in the kind of product space. We're very clear that we will not be in the product space because we understand how deeply it upsets the customers when their contract manufacturer starts to be a product company. So we don't dabble in that business. So like I say, we're happy to produce for a hyperscaler as long as they own the design or somebody else other than Fabrinet owns the design. The other opportunity is, let's say, other merchant transceiver suppliers who would be looking to compete with Coherent and InnoLight. There's a few companies out there who would be looking to compete with Coherent and InnoLight within the NVIDIA ecosystem.
And we think because of our relationship with NVIDIA and because of our experience, we think we should be well positioned to help those companies as they try to break into that space. And then third is any competitors to NVIDIA. Although we believe NVIDIA has a huge head start, I'm sure others are working hard to catch up. So they will need similar solutions to the NVIDIA-type architecture. And again, we'd be happy to support them with that. And there's absolutely nothing precluding us from doing that. Again, our only caveat is we don't want to own products. We're not a product company. We won't own IP, but we will happily produce other companies' products all day long.
Yeah. I mean, I mean, in this particular case, it makes a lot of sense that NVIDIA wants an NVIDIA-branded cable. And I'm sure probably has a preference to sell that and probably has ways to make more margin on it as well, I would imagine, selling it. But I guess more broadly speaking, rather than looking for these one-offs, I mean, is there a way to think about, for instance, a hyperscaler, what would put them over? I mean, because they're not, as you said earlier, they're not optical companies. And well, they're not optical production companies. But what would put them? How close are they to being optical design companies and to sort of taking that next leap? Because I already see them getting very heavily involved in the customization when they work with Coherent and Lumentum and AAOI. But yet, it's not their intellectual property.
There's a large amount of design work being done by those guys. Is there just a way for us to think about help investors understand? Is it just going to be these one-offs that appear out of nowhere, or is there a framework to think about why customers would do that?
I think it's a function of kind of volume and spend. If they're spending enough on optical interconnect, I assume at some point someone sits down and does the calculation and says, "Hey, we're spending X amount. We could develop our own product for 10% of X, whatever X is, and here's the ROI." I think it's just economics. I don't think there's any particular technical reason why they couldn't other than these products are not straightforward. They're very difficult. We know this. We don't design transceivers, but we work with the companies who do. And these products are difficult to get right, and it doesn't happen easily. It's not like traditional electronics or solid-state products. These optical products, you're dealing with data that's moving around at the speed of light. So it's pretty special to be able to design these. So it's not easy.
Historically, hyperscalers have tended to get into, we call it, kind of white-box type businesses. But no disrespect to storage companies, but storage racks are a little bit more straightforward than an optical transceiver. So it is technically a lot more challenging, but it's not beyond their capabilities if they put their minds to it. But I think it really is a function of volume and revenue and the economics of the situation.
Great. Good answer. Well, so Fabrinet was the first to market, I mean, I think, as far as we could tell, right? First to market with 800G. Would we think that something similar there would happen with 1.6 because it would be the NVIDIA system leading the kind of new, not even the new one, but the new, new one coming out, right? That'll be leading the charge to needing 1.6. And so since they're the first ones with that need, you would be the first ones to provide the product. Is that a fair way to think about it?
Yeah. We certainly hope so. If you look at, obviously, we can't talk about specific timing of what we're doing for the customers because we're always careful not to inadvertently announce a product launch on behalf of a customer. They get upset if we do that. But if you look at NVIDIA's roadmap that's published and publicly available, it shows that 1.6 terabit switch, 1.6 terabit interconnect will be used in the GX200 generation of products, which is slated for 2025. Now, again, if you look at their website, it doesn't say whether it's January, June, or December, but 2025. So that should mean that 1.6T is more of a, maybe more of a calendar 2025 story than a 2024 story, calendar 2024 story. But certainly, we'll be working hard to make sure we're ready to ramp when that comes along.
Yep. So yeah, I mean, could you talk more about, I guess, the pricing that we would expect there and then what's going to happen to 800G when 1.6 starts to ramp?
So pricing, we can't really comment on because we don't have real visibility into, let's say, NVIDIA's pricing, what they might charge for these products. But I can confirm we're working very hard to make sure that so if you go from 800G to 1.6, historically, for that type of a change, if you were to compare similar transitions in the past in, we call it, traditional datacom transceivers, you would expect if the speed doubles, you would expect the price to increase, or let's say our cost, our price to increase by, call it, 1.5, something like that. We don't think that would be the case here. One of the things we see with NVIDIA is they're incredibly well organized. They're extremely thorough in how they go about running their business. It's really interesting. If you look from the outside looking in, you have all this explosive growth.
So you might imagine things are very chaotic. They're actually not chaotic. They're very well organized. They're very thorough. They're very diligent. So we've been working hard with NVIDIA to make sure that the price increase, let's say, as we go from 800G to 1.6, that we minimize the impact of that because we, of course, want them to be able to ship as much 1.6 as possible when the time comes. I think because of the application, I think they'll be the leader. They'll be the first to market. They'll be the leader. And we certainly hope that they'll be using their own interconnect for as long as possible, of course, before they bring on other sources. But they will bring on other sources. I'm sure they will. But there's a lot of new, let's say, new moving parts with 1.6 as well.
Obviously, a lot of things have to go right. The GX200 is a new product. I guess with the new GPU, but the GPU should be kind of proven out at that point. And then the transceiver itself, we're looking at 200G per lane optics rather than 100G per lane. So there's a lot of moving parts, a lot of new, let's say, new technologies, but it should be good once it gets going.
Great. Yeah, Seamus, I've heard you talk a lot recently about this opportunity where the way I think about it is that Fabrinet is making sort of at the transceiver and the ROADM and other really key component levels, the most important, highest margin, anyway, most complicated, hardest yield type of products for the industry. And then you've talked about having this revenue opportunity to kind of move up to simpler systems level, which is kind of just snapping these subsystems together into a box. And instead of them having a separate contract manufacturer like a Jabil, you do all the complicated stuff, and they do all the easy stuff, and then they're paying both of you. They can pay you to do it.
And so that sounds really interesting, but I don't know how to sort of quantify if that's happening and just what the impact of that could be. Just help me think more through that opportunity.
Yeah. So I think, yeah, I guess the thesis and also the reality is if we're producing a significant portion of the BOM for the complete network system and shipping that today to other contract manufacturers who purchase those components, add on a markup and a margin to, in some cases, they're called pluggables for a reason. You plug them in. But if you're plugging them in and charging a 6%-7% margin just for plugging them in, there's a margin stack there that the customer is paying for. So our approach is we say to the customer, "We don't need a second margin for something we've already produced for another company. We charge you the labor to plug it in and to assemble the full system, but we're not going to margin stack." So we help the customer to eliminate the margin stack.
So that's, if you like, the economic advantage that the customer gets by moving that system assembly business to us. So we're able to do that for the customer, eliminate significant non-value-add cost, which is margin stack. Margin on top of margin is a non-value-add expense for the customer. So we eliminate the margin stack, and we're able to do it at a respectable margin for us. So it helps us grow our top line, provide great service to the customer, reduce their cost. The opportunity, we have had some success. We won the Infinera Coriant business a few years ago, transferred that to Bangkok. We also transferred the Cisco OTBU business, which we were able to win. They transferred that from one of our competitors a few years ago. But there's other opportunities we're working on.
And the last few years have been kind of tough with, if you remember, the component crisis, which now seems like a distant memory, that whole component crisis that was going on, that mitigated against our ability to move business away from a competitor. And I should point out, it is not easy to win business from a competitor. These are excellent companies that we're talking about. These are all very good companies, our competitors. They do a very good job. There's no fly-by-night companies in the top seven or eight contract manufacturers. They all do an excellent job. So it's difficult to win business away. It was even more difficult during the component crisis because there was always a, it was never a good time to build up a buffer inventory to facilitate a transfer.
Whereas now, it is more companies are more amenable to maybe transferring when they see the savings that can generate by moving to Fabrinet. And we do a very good job. Our execution is really excellent. So all that happens is they get, worst case, the quality delivery service and everything else is the same as they have today. We'd like to think we maybe improve things a little bit, but they just save money. And they save a lot of money when they move business to us. So we're working hard on that. In terms of how to think about the revenue opportunities, it won't be straight line. It'll be lumpy, and there'll be nothing for 18 months, and then all of a sudden, we'll have one.
And so in some cases, we'll get a win that we can't talk about because the customer doesn't want us to announce anything publicly because sometimes there's a competitor who's not best pleased. So we don't want to rub salt in their wounds by having a kind of a press release. So sometimes we'll win some of these opportunities, but we won't be able to kind of communicate anything. But we're still going to work hard on it.
Couple of questions from the audience, and then just a couple of minutes we have. This first one, I think you touched on it a bit when we talked about how sort of organized NVIDIA is and how they can do these very quick cycles, but in an organized manner. But specifically, the question here is, how does the acceleration of GPU new product release cycles affect Fabrinet's model versus your historical experience where the technology migrations were more gradual?
I think the technology migrations for the optical interconnect will continue to be more gradual because we're dealing with those stubborn laws of physics. So I think the technology, let's say, if you look at the speed, let's say, of the transceivers, right now, the next transition that's going to happen is switching will transition from 400 gig to 800 gig, while there'll still be a lot of 800 gig, and then 1.6 comes along. So we'll have 800 gig and 1.6. And then down the line, we'll have 3.2 and 6.4 and so on. But those transitions, I suspect, will not be on a 1-year cadence. The GPU cadence, I think, and the product cadence will be on a 1-year cadence. But I think the optical interconnect, there's only so much speed you can get. And ag ain, 200 gig optics comes along, 1.6 terabit transceivers. 800 gig is not going to go away.
It's not like 1.6 would replace 800 gig. So I think the cadence at the optical interconnect level will be over a longer. It'll be a longer.
Maybe three years between the speeds?
Maybe something like that, yeah. These products take time to develop.
Just the last question here from the audience is, can you talk about your share of the bill of materials in Tesla's Supercharger business? Anything you could say about that?
No, I can't.
Okay. Fair enough. I just, like I said, asking the questions. Well, I mean, we are at a time. I was going to ask a bit about the balance sheet and capital return and any capacity additions and get Csaba onto the call. But it was such a discussion that we went to full 45 minutes, Seamus. I appreciate it.
Thank you very much. We really appreciate it.
I hope the rest of your meetings go well and that the audience also has good follow-up meetings with you and gets their additional questions in. But it was great seeing you today. I, again, enjoyed speaking with you.
You too. Thank you very much, Mike. We appreciate it. Thanks for everyone for participating. We enjoyed it very much.
Thanks.