Good morning, everyone. I'm Samik Chatterjee, and I cover hardware and networking companies at JP Morgan. For the session here, I have with me Fabrinet Management. I have the pleasure of hosting Seamus Grady, CEO, and Csaba Sverha, CFO. Thank you both for coming to the conference. Thank you to the audience as well. Seamus, I'll start you off with a macro question. I'm fully cognizant there's lower visibility in some cases, given where Fabrinet sits in the supply chain. Given the customer behavior that you've seen, maybe if you can share any insights in terms of what you're seeing in terms of order patterns or even visibility from your customers, anything to be concerned about from a macro perspective, given sort of the customer behavior you're seeing?
Yeah, thank you, Samik. Thanks for hosting us, and thanks for the question. We have not seen any change, really, from our customers in terms of their ordering patterns or their behaviors. We haven't seen any slowdown, and we don't expect to, based on the visibility we have with our customers. We think they still need everything we're doing for them. We haven't seen any change. Even with the tariff situation, it didn't cause any change in behavior. We think, fortunately, probably primarily driven by the fact that a lot of the products we're making, or the vast majority of the products we make for our customers, are for infrastructure-type applications. They're not particularly exposed to consumer sentiment. We have not seen any slowdown in the macro environment.
Okay, thank you. I'll get into specific capabilities for Fabrinet. I wanted to ask you about the breadth of capabilities that the company has, because historically, we've been used to thinking about Fabrinet as a leader in optical manufacturing. Does the recent engagement with AWS, or even sort of a similar group of customers that are looking for non-China-based capacity, does that lead Fabrinet to expand its breadth beyond optical manufacturing?
Yeah, I think while 75%+ of our revenue comes from optical, that's split between telecom and data com. A lot of the other markets that we serve are similar markets. For example, industrial laser is an important market for us. The technology is not that dissimilar in terms of how the products go together. While our focus is on optical communications, it's not exclusively. We're happy to veer outside of that, where we can help the customer solve a problem. We're happy to support the customers in other areas. Again, optical communications, automotive is an important category for us, industrial laser, and then one or two others. I think we're not adamant about staying as a pure-play optical. We're adamant about remaining as a contract manufacturer. We won't be getting into the product business.
In the contract manufacturing space, we're happy to provide as many services as our customers need. One of the areas we have been expanding is kind of deeper down into the stack, doing a lot of advanced packaging for our customers. It is not something you necessarily see on our revenue line, because we do that advanced packaging for our own use, where we make the components for our own use in the next level up in the product. That is an area we have expanded deeper down the stack. Also, over the last few years, we have veered into more system assembly, again, where we have a lot of content that makes sense for us to do that. We are very open to expanding, but we are also quite selective.
We want to make sure we do it in a thoughtful way and in a way which adds value for the customer and allows us to preserve our kind of financial performance. We do not want to dilute our margins or anything like that. Yeah, lots of opportunities there.
Okay, just following up, I mean, is the primary criteria then is you're already helping a customer on the optical side, and then you're sort of looking at other opportunities that allow you to have a bigger sort of share of wallet for the customer that allows you better margins. Is that really the focus, or would you go independently? Do you see necessarily a differentiation in doing that standalone basis without any optical business for a customer as well?
If it is the right customer, yes. I mean, for the right customer, sometimes we will do something for a customer that is a kind of a proof we will use as a proof point, just to showcase our capabilities and show them what we can do in terms of delivery, flexibility, technology, responsiveness, quality, all those things. Also cost. Our cost, our footprint is very compact. Our cost base is quite low. In contract manufacturing, cost control is very important, and being able to produce at the lowest possible cost is very important. Yeah, we will happily, for the right customer, we will go in at, if you like, an insertion point and help the customer solve a problem with the view then to expanding out from there into other areas.
Okay, got it. Moving to data com, can you maybe just talk about the length of the engagement with your primary customer in data com, how that positions Fabrinet to benefit materially in the 800 Gb generation in data com, and sort of how that positions you for 1.6 Tb, which is what everyone wants to sort of hear more and more about?
I know. That's what everybody wants to talk about. Yeah, our customer, NVIDIA, they were our biggest customer last year. The relationship started actually with Mellanox, an Israeli company called Mellanox, who have been a customer of Fabrinet for several years. A few years ago, we started to ramp up these transceivers that we ended up shipping in significant volumes to support the Hopper ramp. We started to produce those initially in our Israel facility. We have a small operation in Israel, a small but important operation, where we did the new product introduction and really used it as an on-ramp to move volume production to Bangkok. We did that for the customer. Of course, the Hopper product launched in the early stages of that ramp. The only transceivers that were approved in the network were the Mellanox-designed or NVIDIA-designed transceivers that we were producing.
I mentioned NVIDIA acquired Mellanox in 2019, so that's where the relationship started. In the early days of that ramp, we were producing 800 Gb and 400 Gb transceivers, single mode and multi-mode, that were the only ones that were approved in the network. We ramped with the customers. They ramped significant volumes of their GPUs. Of course, later, they brought in other, we call them merchant transceiver manufacturers, such as Coherent and InnoLight and others. Now the business is we have a share, and then I guess the balance of the share is split between the merchant transceiver manufacturers. We think that pattern will probably repeat itself somewhat. When 1.6 Tb launches, 1.6 Tb will be used for the Blackwell Ultra product. We're ready to go.
We've been really gearing up and converting capacity from 800 Gb, converting over to 1.6 Tb, because really our role with the customer is on the new products. When the new product comes out, we're there to support them to get it ramped up very quickly with their own design initially. I think at a later date, then they'll bring on merchant transceiver manufacturers like they did before. That seems to be our role, for the new products to really be a solid, reliable source for the new products to allow them to ramp quickly.
Okay, and just I know you talked about this on the earnings call as well, but just for investors here, maybe just flesh out the timing on what's your latest thinking on 1.6 Tb when it ramps to become more material for you?
Yeah, it's not really for us to say, because 1.6 Tb, as I said, will be used in the Blackwell Ultra product. I know the customer has talked about Blackwell Ultra launching in the second half of the year. We're ready to go. We've been working hard on qualifying the transceiver and all the various flavors of components that go into the transceiver to make sure that when the ramp comes, we're ready to go. It is really up to the customer to announce the product launch. I don't want to inadvertently announce the product launch on behalf of the customer by announcing when the transceiver is ramping. We've shipped some 1.6 Tb volumes. We'll ship more this quarter.
Our data com business will be up this quarter, primarily driven by 1.6 Tb, but we'll really wait to see what the customer says in terms of when the product is ready to launch.
Okay, so what you're shipping now is the customer building some inventory before the launch, but really when you see the ramp happening is when the customer launches the product?
Yes. Yeah, we think we'd probably ramp, I guess, a quarter or so ahead of the full launch of the Blackwell Ultra product.
Okay, great. More a question on sort of relevance of EMS companies like yourself, just the engagement with the primary data com customer, followed by your engagement now with AWS. Is that changing your view on sort of how relevant manufacturers like yourselves are going to be optical manufacturers relative to the more vertically integrated transceiver IP companies like a Coherent, for example? Where is the market sort of overall moving towards?
I think there's enough business to go around, first of all, for everybody. I think this market is growing so fast. If you look at how the market will break out, there's still a significant portion of the market that will be serviced by, let's say, NVIDIA's own design. They talked on their last earnings call about, I think they said, roughly 50% of their networking revenue came from hyperscalers, and the other 50% came from enterprise and merchant customers. The enterprise and merchant customers, I think, I'm not 100% sure, but I think the majority of them still use the NVIDIA full stack solution, including the networking. Then of the hyperscalers, many of them want to move and have moved to sourcing merchant transceivers.
In the end, I think the real saving for those hyperscalers will be if they're able to design their own transceiver and have someone like Fabrinet manufacture it. We're, again, as I mentioned, we're a pure-play contract manufacturer. We don't have our own products, and we're very clear about that. There's all kinds of reasons why. The main one being, if we have our own products as a contract manufacturer, it can upset some of our other customers who might think that at some point in time we might start to compete with them. We have taken the decision that we won't get into the product business, but we'll help our customers, including if a hyperscaler wants to develop their own design and have their own design transceiver, we'll happily make it for them so they can make big savings in doing that.
I think there'll be, if you like, three buckets. There'll be the NVIDIA design transceiver. There'll be the merchant transceivers, and then there'll be the hyperscalers who design their own transceivers and have Fabrinet make it, or someone like Fabrinet.
Do you have visibility into hyperscalers becoming a bit more active in trying to design their own transceivers? Is there any more sort of activity on that front that you've seen?
I would say the motivation is there, largely driven by the volume of transceivers that they're purchasing and the potential savings they can make by having their own transceivers. I think there's a degree of inevitability. You can have an OEM transceiver. OEM companies, 35%-40% gross margins are not unusual. ODM, probably early to mid-20s. But then if you have your own ODM, they still own the design. If you go all the way and have your own design and have Fabrinet make it, there's significant savings to be made. My sense is they're all considering this, but maybe they're at different stages of the process.
Got it. The reason I also ask that question is we've seen a lot more EMS companies like Jabil, for example, which we cover, have become a lot more vocal about their intent to become a more full-fledged EMS supplier in the optical space. Are you seeing the competitive landscape become a bit tougher for Fabrinet? I mean, for a long time, you were really the only company in the space. I mean, every customer would go to Fabrinet. Is it starting to look a bit tougher now on the competitive?
I think that's a bit of a misnomer. I think we've had a lot of success in winning business, certainly in the space, but we compete every time. There's no such thing as a customer comes to us and doesn't go to anybody else. I mean, every piece of business we win, we compete with everybody else. We find we do well, though, when we compete, because we have this combination of deep capabilities and low costs. It's a fairly compelling proposition for our customers, we find. Yeah, I know Jabil have talked and others have talked. We come up against several competitors all the time. It's not unusual for us. We like the competition. We enjoy it because we find we do well. It keeps us sharp. It makes us stronger, we think. We also think there's plenty of business for everyone.
The business is growing at such a fast pace. We couldn't possibly produce everything for everyone. We're going to stay very focused on, as I say, being a contract manufacturer. We're not getting into the product business. I know Jabil have maybe some of their own products now. That's a decision they've taken, and we wish them well in that. I think Jabil is a fine company, by the way. I think all of the contract manufacturers, all of the top contract manufacturers, Jabil, Sanmina, Flex, Celestica, and others, they're fine companies. They do a very, very good job. I just think because of our degree of specialization, because we're very focused on this, this is really all we do. We develop deep expertise. We can do more for the customers, and we can do it at a very compelling cost.
We plan to keep it that way, and we welcome the competition.
Okay, got it. Since cost is such a big focus, global footprint is what competitors to Fabrinet are now looking to offer to some of the customers, just given the amount of focus on tariffs by country at this point. I know most things are on pause for now, but how would Fabrinet then counter what more of your competitors, which have more global capacity, look to offer to customers on that front?
Yeah, I think if what we were doing for our customers was kind of simple PCBA production and box build, yeah, we'd be rightfully worried if that's what we were doing for our customers. That's not what we're doing for our customers. What we do for our customers is much more complex and difficult than that. We do everything from literally taking the wafer, singulating the wafer, doing the advanced packaging, all the way up through component assembly, through subsystem assembly, all the way through complete product assembly. We do a lot more for our customers than a traditional contract manufacturer. That capability that we've built up, it has taken us over two decades to build up that capability in Thailand. We have deep, deep expertise that doesn't lend itself to moving easily.
The other thing, I think, if you look at the tariffs and where they might end up, we started off with the reciprocal tariff on Thailand, which was then reduced to, it's now 10% like everybody. I guess there's some new 90-day window now, which ends in July, I think. Wherever the tariff ends up being, probably if I had to guess, I'd say 10%-20%. I don't know. Nobody knows. I'm just guessing. Let's pick a number. Let's say it's 20%. I think the way our customers will look at it is, on a landed cost basis for them, it's still more cost-effective to have Fabrinet manufacture what we do in Thailand than to try to start to do what we do in the U.S. on a landed cost basis. It would still be more cost-effective. Our capabilities are not easy to replicate.
Certainly, we have no plans to look at that. We're very happy with Thailand as a location. We think we have a lot of, again, deep expertise that should stand us in good stead.
Okay, got it. Moving back to the technology side, and I know we touched upon 800 Gb and 1.6 Tb earlier, but what do technology transitions like 800 Gb to 1.6 Tb mean for Fabrinet? Because on the transceiver side, the sale of the transceiver, obviously, the transceiver becomes more expensive, typically one and a half times ASP increase for a 2X sort of bandwidth increase. What does it imply for Fabrinet content as we go through these technology transitions from 800 Gb to 1.6 Tb to 3.2 Tb?
Yeah, I think the price uplift will probably not be as big as you would traditionally expect. There's a few reasons for that. One is because we've been kind of gearing up for a while now to get going with 1.6 Tb. We've been able to look at how we take cost out, how we reduce the material cost, how we improve the yields, things like that, but also to make sure that we have a little bit more Fabrinet content. Sometimes when we're able to convert some critical components from purchased items to made in-house, we can drive a lot of cost savings for the customer with that. If we're taking a component that, for example, a supplier makes and makes 35%-40% gross margin on, we can take out 25%-30% of the cost by producing it in-house.
That will reduce the ASP, but it really improves the stickiness and also the margin. We've had lots of time to work on all of those things. The transition from 800 Gb to 1.6 Tb, it's 200 Gb per lane. It's a new design, so that comes with its own challenges, but that's what we like. That's what we do. We're ready to go. Like I say, whenever the customer is ready, we're ready to ramp.
Okay. Just to confirm, when you say price is going to increase less than historically, you mean the price for the transceiver customer that you're saying?
No, I mean, let's say our price to produce. Typically, general rule of thumb, when you went in the old, let's say if you went from 100 Gb to 400 Gb, the price would double, more or less.
Okay. Your content would double?
Our price, yes. When you go from, when you double the speed, you would typically increase the price by about 1.5. This time around, we think that increase will be less because of various factors, including more Fabrinet content.
Got it. Got it. On the same topic, as we look at 800 Gb transition to 1.6 Tb, I think one of the investor concerns has been, as you said, you're sort of helping the customer transition to 1.6 Tb, so your 800 Gb revenues are coming down. You're getting ready for 1.6 Tb. One of the concerns investors have expressed is that you might continue to sort of lose share in 800 Gb, and that offsets the increase of the ramp that you get in 1.6 Tb such that overall data com revenues for you really are more sort of starting to plateau. How would you address that concern?
Yeah, we think ultimately we'll follow the customer's lead. We have been converting 800 Gb capacity over, and I think as we start to ramp 1.6 Tb, the current version of 800 Gb will probably begin to taper off. With the new version, with the 200 gig per lane EML design, there will be a 1.6 Tb, but there will also be an 800 Gb version of that. So 800 Gb won't go away. It will just change from VCSEL-based to EML-based. The question, the imponderable, is for how long will we be the only supplier of 1.6 Tb? Probably a shorter time span than we had with 800 Gb because I think the merchant manufacturers will be ready earlier.
Okay. Got it. Moving to another technology, CPO. How prepared is the company in terms of catering to next-generation optical communication products like CPO, and where would Fabrinet sit in that supply chain?
Right now, we kind of sit in the middle of it, I would say, in between the kind of foundry and the system assembler system manufacturers. The first one that's been talked about publicly was the switch product that we're in the supplier ecosystem for, again, our customer NVIDIA. They rolled out that product at GTC, and they showed all the customer names. We're part of that ecosystem, and it's a very impressive group of companies that are all working together to get the customer's product out on time. Our role is, if you like, the way we see it is there's an insertion point for us where we'll be working on silicon photonics-based technology and providing some of the advanced packaging and optical manufacturing process technologies that we've developed over the years.
We see it as an opportunity to start with that and then hopefully work out from there and do more for the customer. We have two other co-packaged optics projects underway with two other customers that they have not named publicly, so we have not, or they have not discussed publicly, so we have not named them, but we have three CPO projects underway at the moment. I think CPO as a technology is very interesting. I am an engineer myself, so I also looking at new products. It is very interesting to see the technology and the different approaches of the various companies, but it is still quite early days. I think we are a while away from CPO ramping to significant volumes or anything like that. There is a lot to be worked out yet.
Okay. A couple of follow-ups to that. The two additional customers, is that engagement with hyperscalers, or is it more engagement with some of the chip companies that are then going to support the hyperscalers?
There are two existing customers of ours. I do not really want to get into the nature of who they are, but yeah, we have two other projects under the way. For us, it is interesting to see the different approaches of the different companies. They are not all approaching CPO the same way, which is very interesting for us. Our role really is to help them develop the manufacturing processes that will allow them to take the CPO products to scale and volume.
Okay. The second part, when you talked about the transition from 800 Gb to 1.6 Tb and the content there, as much as you get involved in advanced packaging and other capabilities on the CPO, how do you think about then the step from 1.6 Tb content to if you were to go to sort of CPO around the 3.2 Tb generation?
Yeah, I think, first of all, I think pluggables will be around for a long time. I do not think CPO is going to take over from pluggables in the short time frame that maybe some people are expecting because serviceability is critical to the hyperscalers. The pluggable transceivers are a big advantage in the hyperscale world that the serviceability is paramount, and they have to maintain that. Whereas when you go to CPO, that is much more difficult.
On the last earnings call, you highlighted modest headwinds you're facing in relation to new product ramps. Can you just talk about the materiality of why? I mean, you do new product ramps all the time. Why is the materiality now higher than what we've seen in the past, and how quickly do those start to moderate?
I think generally we do new product introductions all the time, but they're usually spread throughout the course of the year. There's no kind of cadence where they all hit at the same time. Now we have a few of them that are hitting at the same time. That's why I think Csaba called out.
Yeah, we have a modest headwind in our Q4 forecast baked in. It's going to last a very short period of time, so it's going to be a temporary headwind because we have such great opportunities ahead of us. We have to put capacity and then startup cost in place upfront of the ramp where we don't have the revenue. Once these programs start to ramp in our first part of fiscal 2026, we anticipate this to go away. It's not going to be a headwind anymore. Besides it, it's probably around 20 basis points in Q4 guidance.
Any more color in terms of all these ramps are hitting you at the same time, but which segment is it?
It's multiple. We have, of course, the 1.6 Tb ramp that we talked about. We have the Ciena business that's in qualification phase at the moment, and that will ramp, I think, as we go throughout the year. I mean, it's a new product, so we have to wait and see how the product is accepted. Will there be any design hiccups along the way? Who knows? It's a completely new product, but that will ramp as we go through the year, through FY 2026. The Amazon business that we've won, that's in, again, in qualification right now, that will also ramp. That will ramp probably earlier than the Ciena business because that's an existing product. Probably over Q1 and Q2, that business should ramp.
We still have some telecom wins and share gain where we're taking business away from competitors that would be ramping as well. We just have a lot of, it's a good problem to have, but we just have a lot of these startup costs all hitting around the same time, which should position us well for FY 2026.
Got it. Got it. Let me just open it up and see if anyone in the audience has a question. Okay. Let me continue here. You mentioned on the call the opportunity to execute plans in regards to Building 10 at Chonburi. What's driving the confidence? Are you getting longer visibility from your customer in terms of their pipelines to sort of drive that confidence?
I think, so right now, just to be clear, we don't have any plans to accelerate. The Building 10, just to kind of let our current footprint without Building 10, we have two campuses in Thailand. Between the two campuses, we have 3.5 million sq ft of manufacturing space. Building 10 will be an additional 2 million sq ft. That would take us up to 5.5 million sq ft. Building 10, we started it in January. It's about an 18-month lead time to get the factory built. We're about 15 months away from completing Building 10. The comment about potentially pulling that in if we needed to was really more of looking at our internal kind of scenario planning.
Everything we've kind of talked about, the products that we just mentioned there, so the 1.6 Tb ramp, the Ciena ramp, the Amazon ramp, and also the continued growth in 400ZR and DCI generally, coupled with the back-to-growth in the underlying telecom business, that's all, if you like, already factored in. I think we'll be fine with our current footprint with all of those. It's more if we're successful in winning a few others that we're working on but haven't talked about publicly that we may need to pull in. If we needed to, we could pull in some parts of the building. We wouldn't need to pull in the whole 2 million sq ft, but if we had to pull in a few hundred thousand sq ft, we could.
Okay. Got it. Just focusing on the AWS or the Amazon business, when you decided to issue warrants to the customer, what does that sort of, one, what was behind that decision? Because you haven't done such transactions in the past. Secondly, what does that give? How should investors interpret that in terms of visibility into revenue from that customer long-term?
I think in terms of the visibility to revenue, it's a good thing because it means our, if you like, our interests and Amazon's interests are very much aligned. It's approximately 1% of the shares outstanding in return that they have the option to purchase, in return for which they have to give us a certain amount of revenue, and it's a significant amount of revenue. We're happy with the warrant. We've never done it before. I hadn't heard of it before. In the beginning, frankly, I wasn't keen on the idea, but we got to a point after several back-and-forth discussions, we got to a point where it actually makes a lot of sense for us and for Amazon to do it this way.
It really means that the customer is motivated to get us that revenue as quickly as possible so that they have the option to exercise the warrant. For us, it means it's never a guarantee of revenue, but it's more or less we know going into this, the customer is highly motivated to get us ramped up quickly, which is always great. It ended up being a real win-win. We both have the same objective. From shareholders' perspectives, we had let Csaba talk about the accounting treatment and why it will not be dilutive to shareholders. We'll make sure of that.
Basically, we had a 10% vesting upfront when we signed the warrant last quarter, so they presented a bit of headwind on the gross margins as a contra-revenue. Going forward, as these warrants will vest, this will be counted against the revenue commitment that we will be getting. Because of the size, we do not anticipate any further impact in the gross margins. In terms of dilution, obviously, we will be buying back shares in the past, as we have been doing. We will be treating it similar to a share-based compensation. There will be no dilution to shareholders going forward.
It is really indeed to the size of the warrant and the revenue commitment over the seven-year period of time, it is really a win-win situation for both sides, and we anticipate that it will have no other impact going forward basis other than this upfront 10% vesting in our Q3 results.
Okay. I will press you a bit more in terms of just on the revenue that it could mean. I mean, one of your peers who has a similar warrant with Amazon has disclosed some revenue numbers overall in terms of what they are targeting. Maybe from your perspective, is there any way to size this up for investors? Is this maybe potentially going to be a 10% customer? Is this going to be a customer as big as what your primary sort of data com customer is for you long-term? How do you see the opportunity?
Maybe bookend it for us to the extent that you can.
Yeah, we're not going to size it because our agreement with Amazon, you may have noticed in the 8K we filed, the revenue was redacted. The customer has asked us not to disclose that, so we won't, but it is a significant piece of revenue for us. It wouldn't make sense to do it if it wasn't a significant piece of revenue. Can this become a 10% customer? I think it has the potential to, yeah. They're a pretty amazing company who can really move the needle. They have very deep capabilities, and we'd be looking to, the nice thing about the arrangement we have with them or the warrant is there's no product categories that are excluded from the business we can do with Amazon. We'd be looking to do as much business as we can with them.
We see the revenue that's in the warrant as a minimum, not a target. We'd be looking to blow past that. Yeah, it has the potential to be a significant customer for us.
Okay. Great. Maybe for the last few minutes, if you can just focus on the telecom market. One, maybe just separate out the growth that you're seeing in traditional telecom equipment versus what has been clearly driving most of the growth for the industry, which is sort of ZR products.
Yeah, so our telecom business kind of breaks into three buckets. There is the baseline business that we had a few years ago. We internally called them the same store sales. That business, the high point for us of that business was in Q1 of fiscal Q1 of 2023, so almost eight quarters ago. We shipped $405 million of telecom revenue back then. Then with all of the inventory digestion that was going on, that business declined for several quarters, bottomed out a couple of quarters ago, and that business is now back to growth. It is not at the $405 million level, but it is back to growing again. That baseline business is back to growing again. The other two areas which have grown nicely for us in that intervening six or seven quarters, the DCI generally is growing, but 400ZR is growing very nicely for us.
400ZR was, I think, 10% of our optical revenue a couple of quarters ago. It's now 10% of our total revenue. It's in our telecom number. I think as we go into the new year, we'll start to break out DCI as a separate category. It will still be in our telecom revenue. Some people think that we should put it in the data com revenue because it's data com that's driving a lot of that growth, but it's in telecom. We're going to leave it in telecom, but we'll disclose the number so that people can decide if they want to put some of it in data com, some of it in telecom. That's DCI and 400ZR.
The third area are the new system wins that we've had some success in gaining business there, both the Ciena product, which is a completely new product, but also other products that we've taken share away from our competitors. All three of those subgroups, if you like, within telecom are growing for us right now.
Got it. Great. We're close to up on time now, so I'll wrap it up there, but thank you for coming to the conference. Thank you to the audience as well.
Thank you very much. Thank you, Samik.