Good morning, everyone. I'll get the next session started here as you take your seats. I have pleasure of hosting Fabrinet Management, including CEO, Seamus Grady; and CFO, Csaba Sverha. [Operator Instructions]
Seamus, Csaba, thank you for taking the time to attend the conference. I'll start you off with a bit more of intro question. Just for the new investors in the room, can you just start with a bit of background on the company as well as history for the benefit of new investors, and then we can dive into more specific questions. Thank you.
Thank you. Thank you, Samik. Good morning, everyone. I'm Seamus Grady, CEO of Fabrinet. Fabrinet is a contract manufacturer. We are the leading contract manufacturer servicing the optical communications industry. We have probably 50% market share of the outsourced manufacturing activity in that industry.
The company was founded in 2000 by Tom Mitchell, one of the co-founders of Seagate and went public in 2017. Our manufacturing footprint is primarily located in Thailand. We have about 3.5 million square feet in Thailand and about 12,500 people there. We also have an operation in Santa Clara, an on-ramp to Bangkok or an NPI facility. We have a small custom optics operation in New Jersey. We have another NPI operation in Israel. And we have a passive optics operation in China, but the vast majority of our contract manufacturing footprint is in Thailand.
We've been growing -- we target to grow at about 2x the rate of growth of the industries we serve, 3x the rate of growth of the contract manufacturing industry. With that, we've been growing. If I look over the last 7 years, we've grown our revenue at 15% compound annual growth rate, and our EPS has grown at 20% compound annual growth rate in that same period. Yes, we specialize, as I say, in the optical communications industry. Historically, that business was split approximately 7 -- primarily telecommunications being the main part, about 70%, with datacom being the lesser part.
In the last few quarters, that mix has shifted. We're now shipping more datacom than telecom through a combination of significant increases in the datacom business, while telecom has been soft due to inventory digestion this past while. The last quarter, we reported our datacom revenue was up 150% year-on-year and up 6% quarter-on-quarter. So we're very fortunate to be participating in the 800-gig ramp that's going on right now to support AI, datacom, our data center expansion around the world. So we're very optimistic about the future and about the markets we serve.
Great. For the question that I'm trying to get a view from every company we host over the next couple of days is how you're thinking about your end markets 12 months from today? When you look at telecom, datacom and the industrial exposure that you have, and I understand it's probably more of a collection of what you're hearing from your customers, but how do you think about where those end markets would be 12 months from today?
Yes. We're optimistic, I think, about all or certainly most of the end markets we serve. If I take them, each in turn as you mentioned in telecom, telecom is down today as the industry goes through some significant inventory digestion. But at some point in time, we think that will begin to improve. And we're -- I think we're optimistic for a couple of reasons.
One is, at this point, I think there's really just upside in telecom. It's at a low point right now. It's only really up from here. And secondly, it gives us an opportunity to try to win additional telecom business from our customers that we would either win new programs or take business away from our competitors. So we're using the time with this inventory digestion that's going on. There's a little bit of a lull in the industry. So it's a good time to push for new business. So we're trying to win additional business right now.
Datacom, I think the -- obviously, the explosion in bandwidth that's required to support these AI data centers has been very positive for our business. We've shipped a lot of 800-gig interconnects over the last while. And I think we will continue to do so for some time. And then that industry will transition to 1.6 at the GPU level and 800 gig at the switching level. So there's a lot of growth left still at 800 gig, but also then 1.6 and further products beyond that.
In addition, the rest of the datacom world had really just finished the transition from 100 gig to 400 gig. So there's another transition to take place from 400 gig to 800 gig. So I think there's an awful lot of growth drivers in the datacom world and it certainly feels like with this explosion in AI data centers that the need for bandwidth is exponential growth rather than linear growth. Maybe the growth in telecom is more linear, but the growth in datacom looks to be exponential.
Our automotive business, it's about 10% of our revenue, maybe a little bit less. That's made up of some traditional automotive business and also LiDAR and EV charging. The traditional automotive business is very stable. Doesn't go up much. It doesn't go down much. The LiDAR business, we've won most of the major players there because in terms of the technology, LiDAR products are very similar to optical products in terms of the way the products are put together. So we've been quite successful there.
And again, a little bit like telecom, we think there's really just upside there. A lot of the LiDAR volume hasn't really begun yet. So as and when those LiDAR companies begin to ramp, we think we're well positioned to capture that upside. And in the EV charging business, our big customer there was going through an inventory digestive, which seems to be behind them now. That's increasing again. So I think our automotive business should begin to increase in line with the industry growth from here on in.
And then finally, industrial lasers, a smaller part of our business. That's been like 5%, 6% of revenue. It's been -- the growth has been disappointing. We've been essentially flat to down a little bit in that business. We haven't lost share to any competitors, but that industry has been down, and our job is to convince that industry to outsource more. That industry is still largely in-sourced. And we would like to work with the companies in that space to outsource more to people like Fabrinet.
So again, even though it's starting from a small base, we see a lot of upside potential and opportunity there. So overall, I would say we're very optimistic about the future.
I want to discuss 800 gig, but before we do that, just how you're thinking about AI being adopted internally in the organization. Coherent just talked about manufacturing being an area where they potentially could leverage AI. You're obviously a manufacturing company as well. Do you see opportunities? Do you see tangible improvements in what you do from adoption of AI?
We do. We're a consumer of AI. We manufacture the equipment that connects all the equipment together, but we're also a consumer of AI. So it really has great potential, great opportunities for us in some of the more backroom activities. So a lot of logistics work, purchasing and planning, a lot of the tasks that can be automated. We've said about automating a lot of those.
So it's really about eliminating waste, eliminating non-value-add activities. And then as we do that, those savings we're able to generate, we like to keep some and pass them along to our customers. So it's really more about finding ways to make ourselves more efficient, reduce our cost so that we can pass those savings on. And we've been quite successful in a few areas. Nothing that we would ever sell commercially with some really good tools we developed and that we now use internally that are very effective.
Let's move to 800 gig. How should we think about the drivers for the 800-gig business? How they've changed over time? It ended up putting Fabrinet in a leading position, leading market share position in the 800-gig market but also you're now starting to see some moderation in the sequential growth. So just walk us through what was that sort of industry dynamic that put you in the leading position and why you're maybe seeing a bit more moderation on that growth on a sequential basis?
Yes. So we started off our relationship with -- we supply NVIDIA with their 800-gig transceivers and a range of other products, 400 gig as well. Our relationship with NVIDIA, it might look like an overnight success, but it's been 10 years in the making. We started off as a key supplier to Mellanox probably a decade ago, I think originally. Then of course, Mellanox was acquired by NVIDIA in 2019. And we've had a long-standing relationship with Mellanox, a great relationship with Mellanox. About 3 years ago, we established our own on-ramp facility in Israel in Yokneam, across the carpark from Mellanox. And the purpose really of doing that was to start to introduce these products. So we started off in Israel. We did the initial builds in Israel, the initial kind of debugging of the process, then we transferred manufacturing to, again, our volume manufacturing facilities in Thailand.
And then we set about ramping the volume fairly significantly. Of course, we -- there was nothing to talk about publicly until NVIDIA became a 10% customer, which is at the end of our fiscal year, our fiscal year ends in June. So I think in our earnings call in August, was when we first announced that NVIDIA was a customer. And at that point, they were already a significant customer and greater than 10% customer.
So we've been building and ramping 800 gig. I think, in the beginning, NVIDIA just use their own 800-gig interconnects in these products. Then as time went along for various reasons to do with making sure they didn't become bottlenecked and that the transceivers didn't become a limiting factor in selling GPUs, and also to meet the needs of their customers, they brought along 2 other approved sources. Not for the products that we make, the products we make, we're a contract manufacturer, so we make NVIDIA's products.
But they brought it on 2 other companies, Coherent, the previous speaker, Chuck, CEO of Coherent and also InnoLight. So they have been ramping and shipping significant volumes, but we've also been doing quite well again. Our datacom business last quarter was up 150% year-on-year and 6% quarter-on-quarter. So the volume is quite strong. The demand is quite strong. And I think we -- our objective is to continue to be the leader and to continue to be the preferred contract manufacturing partner for NVIDIA.
I suspect NVIDIA will bring on other approved sources. There's lots of a few other companies out there who have talked about getting approved in that regard that merchant transceiver space together with Coherent and InnoLight. And I'd say if we were here a year from now, you'll probably have 4 or maybe 5 companies improved in that space.
One of the questions we get asked a lot is how does NVIDIA make the decision as to whether to go with the Fabrinet manufacturer transceiver versus let's say, Coherent or InnoLight transceiver, we have no idea how they make that decision. And we wouldn't ask them. That's their business.
Okay. How do you think about the revenue drivers changing for Fabrinet as you transition from 800 gig to 1.6T? How do you think about the timing of that? I mean, I guess the question essentially is, should we expect an inflection as you start to ramp 1.6T and what's the timing of that look like?
I think the timing, I'm not in a position to talk about. I mean we know -- of course, we know what our customer is asking us to do, and we know what we're doing in terms of getting capacity ready and everything else. But the timing of that, I wouldn't be in a position to talk about because if I were to talk about it, I would be, by definition, announcing the product launch on behalf of the customer and they generally get upset if I were to do that.
I think what the customer has talked about is their GX200 product, which is the one that will appear to use 1.6T optics at the GPU level and 800-gig optics at the switch level, that they've said that, that launch is in 2025. That's all they've said. So we don't know if it's January, June or December. Like I say, we know what the customer has asked us to do and we'll be ramping. It would be a similar -- I suppose a similar pattern to maybe 800 gig, we'll have to build up capacity and start ramping well ahead of the product launch.
So we'll be doing that over the next several quarters, but I'm afraid can't really talk about the exact timing at this point.
Okay. Good. How should we think about engagements beyond NVIDIA as a customer for 800 gig or even 1.6T? How are you seeing the outsourcing market within the optical industry sort of evolve and how is the customer base broadening for the datacom market?
Yes. So outside of NVIDIA for, let's say, 800 gig or 1.6T and beyond. There's really 3 vectors for growth that we're pursuing. The first is other GPU companies who are going to compete with NVIDIA. They will need similar optical interconnect solutions, and we're well positioned to support them with that. So that's the first one.
Secondly, the hyperscalers, if the hyperscalers decide to go direct with their own optics, which at some point, they may decide to do that. They'll need a contract manufacturer, and we think we're well positioned to manufacture those products. Our only caveat is we will never own our own products. We're a pure contract manufacturer. We -- we'll never have a Fabrinet logo on a product that we make because we think that's -- some of the contract manufacturers do that and the best to look to them, but it's not for us because for the very simple reason that our other customers would be quite upset if we started to as they would see it compete with them. So we never have a Fabrinet-branded product.
But -- so just to recap, the GPU companies, secondly, hyperscalers if they go direct. And then third, that merchant transceiver market as others want to enter to compete with, currently Coherent and InnoLight, they'll need someone to manufacture their products. So we'll be working hard to -- we're working hard on all 3 of those. It remains to be seen, which 1 or 2 is the most likely, but they are the 3 areas for growth.
In our business, the kind of gestation period, if you like, from when we engage with the customer, who has a real need until we're actually shipping something is 18 months to 2 years typically. So this is a slow sales cycle. Like I say, in NVIDIA, we were manufacturing for NVIDIA for 8 or 9 years before they become a 10% customer. So these opportunities take time to come to fruition. You have to be very patient in our business.
Yes. Regarding the 1.6T transition, you've said it multiple times on the earnings call as well that you don't think it really cannibalizes 800 gig as it ramps. Maybe just flesh out that reasoning a bit more? Is it just because it's a different product set completely? What's driving your thinking that it doesn't really cannibalize 800 gig?
There's a couple of things that drive it. One is if you look at -- again, this is all on NVIDIA's website. There's nothing confidential in what I'm about to say. If you look at NVIDIA's website, they show the architecture of their future products and when they move to 1.6T transceivers or interconnect at the GPU level, they move to 800 gig at the switch level at the top of the rack at the same time.
So in fact, the number of 800-gig sockets actually looks to increase when 1.6T comes along. Now we're not sure if they'll populate all those sockets. But certainly, we don't see any evidence that, that 1.6 is replacing 800 gig. It seems to be additive rather than replacing. And then secondly, again, for, let's call them, regular data centers, that industry hasn't transitioned really yet from 400 gig to 800 gig. So there's -- we think there's a lot of growth left at 800 gig. A lot of runway left and a lot of room to grow the 800-gig business. And we don't see 1.6 as replacing it.
It's not like a traditional in traditional data centers when the upgrade cycle happens, let's say, the industry goes from 100 gig to 400 gig, 400 gig very quickly cannibalizes 100 gig. That looks like it's not happening in these AI data centers. They're separate applications almost.
Relative to the technology for 800 gig moving to 1.6T as well. They're sort of 3 competing technologies that we hear of. One is the CW laser base silicon photonics, the second one being EML-based, third one XL-based. Do you necessarily see your position being stronger in any one versus the other? Or are your capabilities pretty broad and you can for your customer, do all of them?
Our capabilities are very broad. We're very comfortable in manufacturing all of the above, and we don't really have a preference for one over the other. Again, we're a contract manufacturer. We don't have our own products. So we'll manufacture whatever the customer needs us to manufacture. And we're very comfortable with any of those technologies.
Moving to telecom. Can you just give us an update on what you're hearing from your customers on inventory digestion, any sense of how many more quarters to go? And we are sort of starting to see some stabilization, but any more thoughts on when to expect a rebound?
Yes, I suppose we learned a long time ago, we don't really listen to what people say. We just watch what they do. So what we started to do a long time ago, probably a year ago was track their inventory levels, they being the telecom equipment manufacturers. And we look at that in terms of days of supply, and we compare to what they're -- what they're consuming, what they're burning.
And we try to project where we think they'll be -- they'll reach the point where they need to really start ordering again in earnest. We have been -- if you go back maybe a year ago, we had been saying we thought it could be the second half of 2024 before things start to improve. That wasn't a popular view. I got some feedback about that at the time, but here we are. So we're coming up on the second half of the year.
If there's evidence that it's improving, we haven't seen it. We hear some of the -- talk about green shoots and things like that. When we see it in the orders we're receiving, we'll be delighted. But so far, we haven't really seen that. That was the reason for the comment on the call that it could be out into calendar 2025. Again, we hope we're wrong. I'd love to be wrong on this one, but just based on what we see, inventory levels are still stubbornly very high at some of these companies.
And we also hope that it is, if you like, just an inventory correction and that there isn't some bigger maybe overarching trend going on. Again, we're a manufacturer. We're not particularly industry experts, but it does -- it just begs the question about what is the driver -- what will be the driver of growth in the telecom market? We know what the drivers of growth are in datacom, it's AI and the exponential growth in bandwidth that's required to support these AI data centers. But what bandwidth is required for AI data centers? What telecom bandwidth is required?
Right now, if you use any of the AI, ChatGPT or any of their equivalents, you send your question into the form of text or you upload a document. It uses very little bandwidth, almost no bandwidth. All the processing happens in the data center, and then you get your answer back 20, 30 seconds later. So the processing power and the bandwidth is in the data center. Out of the edge here in the telecom network, it's very little bandwidth required. So unless something changes that, unless something happens to require more processing and more bandwidth out at the edge, then it does beg the question, is there some bigger trend going on that pushes the need for the bandwidth into the data center, instead of out of the edge.
Historically, the need for bandwidth is in the telecom network because the data centers were mostly storage. Now the data center networks are mostly compute with a little bit of storage. So it's driving, again, exponential growth inside the data centers and linear growth in the telecom networks. Of course, we won't know until we get to the future and look back, but it does seem like there's something bigger than just inventory digestion going on. We hope not, but it seems like there might be.
Interesting. How do you think about then this processing in the data center impacting data center interconnect, which sits in the telecom business? Is that a driver of rebound?
It is, and it has been for us as well. If you look at our telecom revenue, we were looking at it earlier, our telecom revenue in our company. And again, we're not a proxy for the industry, but probably not a bad indicator. So our telecom revenue, if I go back to Q1 of 2023. So we're -- so 6 quarters ago, 7 quarters ago, we shipped $405 million of telecom revenue. And then the next quarter, [ $393 million, $380 million, $310 million ].
And so on. Last quarter, we shipped [ $286 million ]. So we're down more than 25% versus 6, 7 quarters ago. But within that, DCI has seen fairly significant growth actually. So DCI has been growing. DCI, for those of you who don't know, data center interconnect, they are categorized as telecom products because they connect the data centers together. But what drives their growth is actually what's going on in the datacom world.
So they're categorized as telecom products, they're in our telecom revenue. They're in everyone's telecom revenue, but they are actually datacom-driven products. And DCI has been strong for us and continues to grow for us. The biggest -- seems like the biggest growth engine in DCI is 400 ZR and also 800 ZR and ZR plus. Right now, we have 5 customer engagements in that, we call it the ZR space. So 400 and 800 ZR and ZR plus. We have 5 customer engagements there. So we're -- that's one area of telecom we're very optimistic about.
Okay. I was going to ask you on ZR, but let me still sort of ask you, but maybe in sort of rephrasing the question, have you done any work in terms of outlining what the addressable market on ZR, ZR plus looks like? Obviously, it's been a slow market to ramp over the last few years where it's finally starting to ramp, and we can see there's sort of green light on that front. Like have you done any work on outlining what that opportunity for you could look like? And does It start to get close to what the traditional telecom business opportunity is? Or is it -- does it still remain materially below the traditional telecom opportunity in size?
Yes, I think it's -- we've done a lot of our own internal research. We think it's certainly a market that's one we're targeting very aggressively to be the leading, again, the leading contract manufacturer supporting the customers in that space. We see it as very strong. I don't think it's going to outgrow telecom. Telecom is still much bigger.
But in terms of growth, it's -- from our perspective, it's the only area within telecom, we see growing for the next while. And it looks to be a very good solution. Again, for DCI, it's very good solution in terms of the speed and the bandwidth, the power consumption and also the price, it seems to be the best solution for that -- those DCI products.
Let me open it up to the audience before I switch gears to the other segments. Any questions for the management team?
Yes. Do you see any technical challenges with manufacturability of 1.6T? Are there any major issues in terms of your ability to produce?
These products are all quite challenging to produce. They're not straightforward. They're quite difficult to produce. There's a few areas. But I think when you get to products of these speeds and these very, very precise dimensions in terms of manufacturing, it's really just a question of getting the process dialed in. It's all about the yield. If you -- if you can predict the yield, which we're quite good at. And then if you can make sure that you're yielding at or above your expected yield, life is good.
If you're below your expected yield, life is not good. So there's always challenges. And generally, the way we -- what we will do is we will work with the customer to figure out, okay, when do we launch? We let the customer decide actually. Sometimes the customer might decide to launch any customer, not any particular customer, but companies in general, might decide to launch with the product below what you might expect it to be as long as they know that everything that's shipped is good. So we'll capture the fallout in-house.
And typically, just to point out the fall out, it's -- if it was a manufacturing defect, that's our issue. But generally, it never is, it's to do with those stubborn laws of physics and things like that when you get a component mismatches, tolerant stackups, timing issues, software issues, firmware issues, it's those type of issues that just take time to bed down. So what we're very good at, I think, is okay, we build a batch. We -- it's good to get good product.
But what's more useful actually is the bad product or the product that didn't yield, why didn't they yield, what's the root cause, the real root cause, and we're really good at working with the customers', engineers' to figure out, okay, what are the trends, what needs to be worked on. And then it's kind of an iterative process with the customers, design engineers, where they go back and they change and maybe redesign or change software or change components in some cases. So that's work we kind of do in the background quietly for several quarters before we launch or before you ever hear about the revenue. But yes, these products are not easy to make. If -- that's why we like them because it's difficult for others to imitate.
Given that complexity, is it -- are your margins -- is there an upward bias to the margins as you have rising complexity with the 800 or 1.6? And then also maybe you mentioned it before, but maybe just double-click on the complexity of working that customer or the firmware and the other aspects that make it more sticky for you to keep that customer. How long are the design lead times to get in. And maybe compared to like a 100-gig transceiver that's not with 1 specific customer, just how much harder that really is and sticky.
Sure. Well, I suppose the first thing is 100 gig, now it's easy. But at the time that 100 gig was being introduced, I was replacing 25 gig, it was very difficult. So generally, things in the beginning are very difficult to produce at a consistently high yield and at a predictable unit cost, but that gets better down over time.
In that sense, 800 gig is no different. The question about the margins. For us, in general, our margins are better on newer products than they are on older products for all kinds of reasons. So we have a kind of a bias for, again, the new product pipeline. We always like to make sure we're working on obviously shipping the current products, but then getting ready for the next product. And usually, we try to be at least -- at least 3 product families ahead of what we're shipping today. So we're usually working with customers today on products that may not launch for 3 or 4 years, actually, in some cases.
And in some cases, we work for 3 or 4 years, and then we get to the point where it's ready to launch and the customer decides to not launch the product. And that's okay, too. That's their choice. But that's what we do, that's what we're good at. It's helping the customer figure out we're very good at, like I say, failure analysis and really helping the customer understand the root cause so that the customer can make changes. There's usually not that many changes that we can make to improve the yields. It's more to do with helping the customer understand like I said, design changes and software firmware changes.
So yes, design lead times are typically quite long in this industry a couple of years, 2, in some cases, 3, 4 years.
You're very clear before that you have no idea how NVIDIA makes their purchases decisions. But for investors who are attracted by your datacom exposure but may be less familiar with the competitive landscape. Could you just big picture explain how you view your competitive advantage in the industry?
Thank you. Yes. So our competitors are all the large contract manufacturers. I can name them all, but I don't want to be advertising for them, but I will. It's, I would say, Flex, Sanmina, Jabil, Celestica, benchmark on some products. Plexus, not so much. So it's really, I would say, like Flex, Sanmina, Jabil and Celestica will be our 4 biggest competitors in the kind of the global EMS landscape. There's some smaller contract manufacturers, too, but generally, the smaller contract manufacturers, they can't cope with the business once it gets to a certain scale. So they will be the biggest competitors.
We feel we're well positioned. Our focus is mostly on the optical components, right down to the packaging straight after the wafer. We don't do -- we're not a foundry. We don't make the wafer. But everything after the wafer we can do. So we can do packaging, both multichip module and also actual components. Then we integrate those components into the next level up in the product into like a sub-assembly assembly or a subsystem. So our perfect customer for us is one where we're doing everything from packaging all the way through to complete systems that we ship.
Whereas most of our competitors tend to focus more on that system-level business because that's where the revenue is. But the complexity is down at the packaging level and the precision microelectronics level. Just maybe put it in a little bit of context. Most contract manufacturers, their clean room space as a percentage of their total manufacturing space is about 10%. It's a bit lower. It's probably 6% or 7%, but let's -- we round it up to 10%. Our clean room space as a percentage of our manufacturing space is 70%, 7-0.
So in that sense, we're much more of a packaging and precision microelectronics company than a traditional contract manufacturer. So it gives us -- we think a few competitive advantage, but competition is strong out there. There's some very, very good companies who like to compete with us, but we like to -- we like to go after their business, too. So we enjoy the fight.
Let me squeeze one question that I want to get in before we run out of time. Capital allocation strategy. Can you talk about how you're thinking about capacity expansion versus share buybacks? And the primary reason for asking is often the investor feedback is Fabrinet is a great company, but what we get on earnings growth is what we get on revenue growth. How do you sort of -- is there any plan to sort of more proactively tackle that with buybacks as you go into the next fiscal year?
So on the chart, we'll talk about the share buybacks and then I'll cover the capacity expansion, if that's okay.
So we -- typically, we want to make sure that we return the surplus cash to our shareholders. So we have had a pretty good run in the last 3 quarters. We have generated over $290 million of free cash flow. So we are sitting on a very healthy balance sheet. I think this is a good time right now actually to have a very healthy balance sheet, and our cash have yielded about $0.23 per share EPS in the last quarter. So from that perspective, we are enjoying the higher interest rate environment right now.
With regards to buybacks, we do have a 2 ways of returning surplus cash is one of them is 10b5 plan. We typically run a 10b5 once a year. In the recent couple of quarters, it didn't trigger too much buybacks because of price restrictions. However, we revisit that every year in our office board meeting. So this is something that will come up again as we go into the next fiscal year, but we make strategic decisions. But for us, capital allocation is really focusing on organic growth and investing and reinvesting in the business. And with that, I think Seamus will cover the capacity expansion at large, but that's our main focus.
Capacity expansion, we have -- as Csaba said, we have about -- I think in the last quarter, we ended up with $794 million of cash in hand. We generated $8.5 million of interest income. So it's not a bad time to have cash. We have very little debt at this quarter, we'll have basically 0 debt. So a good use of the cash for us is to expand our own capacity. We -- right now, we have 3.5 million square feet of manufacturing space.
We're at a revenue run rate of about $2.93 billion. If you take last quarter, it's actual times 4, we have capacity right now to do about $4 billion. So we've lost capacity. And as we expand, we'll build additional buildings in our campus in Chonburi, we can build another at least another 2. Each additional building will add about 1 million square feet of manufacturing space, which will give us revenue capacity for an additional 1.2 billion, 1.3 billion and will cost about $55 million, $60 million, which we'll pay for with cash. And we'll take about 18 months to bring up and adds about 10 basis points. It's about a gross margin headwind of about 10 basis points. So very little downside and a lot of upside potential.
Great. I'll wrap it up there. Thank you for attending the conference. And thank you, everyone.
Thank you very much.
Thank you.
