Okay. Thank you, Cathy. Good afternoon. Welcome to Compal's Q2 2026 results conference call. I am Randy Abrams, Head of Taiwan Research at UBS, and I am joined here by Mr. Tony Bonadero, CEO, Mr. Jack Wang, CFO and spokesperson, and Ms. Tina Chang, Senior Director and Head of IR.
Just to note, too, as the operator mentioned, 2:30, there will be a Taiwan drill, that the cellular speed may come down. So for better connection, you can join Wi-Fi at that time, from 2:30 on. With that, I will turn it over to Tina. She will start with a financial review, followed by Tony for the business update and guidance, and then we will be happy to open up to questions. With that, over to Tina.
Thank you, Randy. Thank you, everyone, for joining today's call. This is Tina from Compal IR. Before we move to the details, please help us turn to page two for the safe harbor notice. Please take a moment to review the statement. Okay. Now let us move to page five on the product mix. Compal's second quarter revenue grew substantially to TWD 238.3 billion.
Non-PC contribution continued to stay around at a 35% level, up about 5 percentage points year-over-year, and on path to their 40% goal. The non-PC growth was primarily driven by the ramp-up of our AI server business. Next page for the income statement. Second quarter gross margin was 4.6%, impacted by the rising component cost, particularly in memory prices. While cost passed through supporting ASP expansion, the expanded revenue basis also diluted the gross margin ratio.
However, through our continued product mix optimization and enhanced operational efficiency, the second quarter gross profit dollar increased by 4% quarter-over-quarter as well as a 4% year-over-year. Operating profit at TWD 2.8 billion, grew 8% quarter-over-quarter and a 9% year-over-year, outpaced the gross profit growth, reflecting the benefit from our expanded operational leverage.
Now, on the non-operating line, we booked a small amount of interest and income in second quarter. However, the major contribution came from about like TWD 1.8 billion in the mark-to-market valuation gains on the financial assets. All in all, the second quarter net profit was TWD 3.1 billion, increasing 59% quarter-over-quarter and substantially year-over-year, and the EPS was TWD 0.73. Next page for the first half's P&L.
For the first half, the overall trend was similar with the continued growth of the server business, ongoing product mix optimization, and a greater realization of the benefit from our transformation. Compal's operating scale and the profitability both improved compared to the same period of last year. The net profit for the first half was TWD 5.1 billion, grew 91% year-over-year, with the EPS of TWD 1.18.
Let us turn to next page on the balance sheet. For the balance sheet, end of the second quarter, our cash position was around TWD 83 billion. We have invested in working capital investment with cash conversion cycle days lifting slightly to 52 days. Given our business expansion, the liability ratio slightly up to 71%, but still maintained at a relatively healthy level.
Lastly, book value per share in the second quarter was TWD 31.3, both improved quarter-over-quarter as well as year-over-year. This concludes the financial review session. Now I would like to hand over to Tony for the business update. Tony, please.
Yes. Thank you, Tina. We have talked a lot about, in the past, about the transformation that Compal is going through, so I thought I would give you an update on the Compal AI transformation, as we call it, AI inside and out. If you look at the pie chart, everything on the right side is external. PCs, it is basically the businesses and customers that we serve. Everything on the left side is internal AI transformation we are doing.
So, A, AI devices. We are positioned very well to capture AI PC refresh super cycle. While there is still some Windows 10 end of life, we see a lot of focus on future proofing these replacement purchases with NPU silicon, NPU capable silicon and AI PC offerings. We see a premium AI PC mix lifting revenue and margin. First half unit margin improved if we strip away ASP effect and volume impact.
B, on AI server, Q1 2026 server revenue already equaled all of FY 2025, with AI servers over half of our server mix. Additionally, we are dedicated to delivering an end-to-end solution and are now building back-end service capabilities coupled with regional production advantages to stand out in a crowded market. We are starting to see the lift of the AI server business with these results today.
C are the five businesses, the five plus one we have been talking about. Focused growth engines, dedicated portfolio of early growth engines converting from bets to scale. 5G UE, IPC, grew about 106% year-over-year MedTech CDMO, physical AI hospital robotics with strategic alliance with NVIDIA, and currently four Taiwan leading hospitals with U.S. hospitals, planned and ready to follow. Then D on the left side is our margin inflection from AI transformation. Unit margin improved year-over-year.
Non-PC mix was up about +5 percentage points year-over-year, as the next transformation boosted efficiency, diversified revenue mix, and restructured the low margin portfolio. COMPUTEX 2026 was an exciting event for us. It was really one of our first large kind of coming out, showing what the brand is all about, and it got incredible reaction. We basically were able to show what Compal's future is.
If you will, COMPUTEX 2026 was the trailer of what Compal plans to achieve over the next 18 months or so. Creativity, the future of PC, this is our bedrock, and we continue to bring innovative concepts like Agentic AI PC solutions and others. Acceleration AI infrastructure, this is server, and new use cases to demonstrate, such as quantum AI applications and others. Future care and connectivity are growth engines across MedTech, 5G connectivity and audio.
As we go into the revenue highlights, revenue again was TWD 238.3 billion. 34% of that revenue was non-PC, basically flat, one percentage point down quarter-over-quarter, basically flat. Year-over-year, we saw growth of 32% total, 52% in non-PC revenue growth. We are not just a PC company anymore. I said that at COMPUTEX, we are really not. We are starting to expand beyond that. Non-PC revenue share, again, up five percentage points year-over-year and has been staying there, well on our path to 40%.
We are growing across both PC and non-PC, but obviously non-PC growth has been really significant for us. The notebook PC segment, notebook revenue was TWD 156 billion, 20% growth quarter-over-quarter, 23% growth year-over-year. Notebook shipments were up 5% sequentially on seasonality, while minus 13% year-over-year due to component supply constraints and price increases.
Q2 is a kind of prime seasonality for some Chromebook, and there was a lot of Chromebook shortages with the lack of small core availability. A shift toward premium and AI PCs lifted average selling prices supporting a higher value product mix. FY 2026 PC shipments are expected to decline at a mid-teens rate broadly in line with the market.
Surprisingly, we saw the first half was almost flat. 133 million last year, 132 this year, almost flat. We think the second half is a very different story, and we think the second half is probably 15%-17% down, as we continue to see higher component costs weighing on consumer demand. Overall unit profitability is constant despite volume drop, and underlying unit profitability is improving if we strip away the effect of buy-sell inflation and volume impact.
The improvement is due to a more focused portfolio, higher margin product mix, and transformation efficiency. The non-PC segment, TWD 82 billion revenue, quarter-over-quarter growth of +15%, year-over-year of +52%. Vast majority of this growth was driven by our AI server business or our server business. Non-PC revenue, again, +52% year-over-year, led by servers with contribution from smart devices and intelligence systems.
AI servers ramped to 70%-80% of server revenue in 2Q 2026. New neocloud wins expanded the growth pipeline. Our FY 2026 non-PC mix targets to approach 40%, and yes, 34%, we are well on our way. We think by the end of the year, we will be there. If you look at the server and AI infrastructure, some of these, everything on the left there we have announced.
We are gaining a lot of momentum with AI cloud wins across three different regions, Europe, Asia, and U.S. Focus on Compal's value proposition of delivering end-to-end solutions, their global manufacturing expansion, our addition of a lifecycle management component to this, and bringing our efficiencies to supply chain and operations.
Aside from entering more parts of the value chain, we are now building backend service coupled with regional production service advantages to stand out in a crowded market. Server and AI infrastructure, we have two new sites coming online, Taiwan and the United States, and expanded capacity across four countries. Ramping by the end of 2026 are all of these facilities. One capability standard, close to demand, resilient. Up to full rack scale L11 integration.
In Taiwan, we have a facility which will be an NPI facility, AI server and cooling manufacturing, full rack level integration, and an in-house R&D hub. In the U.S., we will have onshore base for U.S. customers. We will have board assembly through full rack L6, L10, L11, and full stack U.S. center, including R&D. China and Kunshan will still play an important part of this, serve China and Asia demand. Group NPI knowledge base as well.
In Vietnam, we continue to ramp up mostly PCBA production, PCBA scale hub. It is really one of our highest volume board sites that we have. One of the biggest highlights of our COMPUTEX showing was in the physical AI space as we rolled out PolyMedics, which is a concept of physical AI for healthcare, spanning across arrival, treatment, surgery preparation, tracking, and training.
This is obviously playing to a huge need that we have today. You have heard a lot about robotics and healthcare. Most of that is a humanoid robot walking into your hotel room and sticking a thermometer in your mouth. We think that is probably some time before that actually happens. This, we believe now this technology is available immediately, right? It is available technology that can roll out and scale today.
Our whole goal around this is to develop an operating system and a deployment system for hospitals to make it easy to own, control, operate physical AI within their facilities. Lastly, transformation gains to sustain and fund external growth. The transformation has been pretty amazing. We have been doing it for about 18 months. We have been planning it and doing it for about a little over two years. We have identified 70+ use cases for digital.
We have implemented those across the company. There are core use cases already deployed in R&D, manufacturing, procurement, and supply chain that are actually providing significant savings, significant operating savings to us today, and also helping us fund the external world, the future growth, scaling growth from AI devices into server and intelligence systems.
The foundation remains PC and smart devices. Scale remains server and infrastructure first and foremost, and then 5G connectivity, and the businesses we are developing on frontier MedTech continue automotive, physical AI, and hybrid. Thank you.
Okay. All right. Thank you, Tony, for the remarks. I will kick off, but I will turn it to the operator just if you want to give instructions, and then maybe I will start with a first question. We can turn it over to Max just to get the instructions for the Q&A.
Thank you, Randy. All attendees, if you want to raise a question, please just raise hand feature, and then we will allow you to unmute yourself. Thank you.
Okay. Thank you. I will start with a first question, I think just to start with an outlook, like maybe an initial view coming off second quarter. We could start with the notebook business, how you saw kind of your order trend, if there was any pull-in activity just ahead of rising memory price.
A follow-up to the second half where you are expecting a kind of a weak market overall, how much it is limited by supply, a shortage of certain constraints versus demand impact? I will start just with notebook, and then we can broaden to other parts of the business.
Okay. I think we are facing kind of a fundamental margin reset in the PC industry, right? Component costs are rising faster than the market will tolerate price increases. I mean, that is just a fact.
It is turning what used to be a volume and efficiency business into a battle to protect every basis point of gross margin. It is kind of what we are seeing. We think that the growth in the second half, most of the replacement cycle probably underway, probably has happened already.
We think that on the commercial side, those customers are thinking very carefully about where they are spending their CapEx, and a lot of it is going to AI-type resources. We do believe in the future, and even today, that AI PCs can represent one of the lowest costs of AI computing out there, right? It is one way to look at it. But we believe that that market is down kind of mid-teens. We think corresponding consumer markets are similar.
As I mentioned earlier, just we continue to monitor as component costs continue to weigh on consumer demand, but we are seeing wearables, tablets, PCs, other things, just constantly impacted. Overall, we think for Compal, the third quarter growth continues, driven by AI servers, with server revenue expected to increase by high double digits quarter-over-quarter.
PC units are expected to remain flat for us, basically, while continued ASP increases should provide additional support to revenue growth. Smart device revenue is expected to decline, like we just talked about, with basically watches and tablets and things like that all under pressure. Consumers all going, "Yeah, I'll wait till next year. It's just too expensive right now." That may not be a great strategy because next year might be worse than this year.
Yeah.
This is kind of what we think. 2027 may actually have more margin, or more memory pressure, more memory inflation, and in a lot of cases, you just can't get what you need. Then we think 2028 returns back for to 2026. They're looking ahead in fourth quarter AI server momentum, again, to remain healthy. PCs and other consumer products, we'll continue to monitor that impact.
Okay. That's great. Let's turn it to the operator just to see if you're showing any questions. Otherwise, we can ask a few follow-ups from the floor here.
We don't have any raised hand yet.
Okay. I will ask a question just to the. Because you mentioned 2027 could be a continued headwind from all this memory. When you are doing the initial bidding for 2027 projects in Notebook, how is the competitive environment? You talked about fighting for every dollar. Just how does it look from an environment, and what is your strategy, is it in terms of trying to maintain market share, unit share, dollars, or focus on certain mix? Yeah, kind of more how the industry looks in your strategy.
Yeah. I think from it is not every dollar.
Yeah.
It is every penny actually. I think, as we look ahead to what we think is going to happen in the rest of this year and in 2027, there is a problem just with the memory inflation, right? That is going to continue, for us, I think. What our strategy is to continue to compete for the higher margin type products, to look for the AI PCs, where we are very well-positioned there already. Our mix is, I think, ahead of.
Of our, like.
Industry mix.
Around 50% of our shipments is AI PC related at this half. In terms of the revenue contribution, even higher because higher ASP. Dollar-wise, it is about around 60% first half on AI PC side.
With everyone focused on AI, it comes to the competition side. There are two parts. There are your traditional competitors, how they are approaching the market. A year ago, it felt like there was a bit more coming from China. How do you see those two, your traditional competitors and the China competitors now?
There was a lot of noise last year around the Chinese OEMs. Still formidable competition. We took a step back and kind of analyzed what we do versus what they do. Some of the perceptions that maybe people had about why they are able to achieve such aggressive pricing wasn't necessarily true.
Maybe there are subsidies, maybe there are other things that we do not know about. We believe that, especially with the big three tier one PC OEMs, there is a very specific AVL stack that you have to deal with. When you look beyond that stack and you say, "Well, if I do not build it your way, but I build it my way, then I price you, okay, here is your quote on what you want, but I can do this if you let me.
Right.
If you allow me to go and to bring more vendors, different vendors, some variety into the AVL. Voila, we were competitive.
That is great
And have won deals based on that. We believe we have kind of figured that out. We could not have started the transformation at a better time than we did because we have, last year, in 2025, across all of our manufacturing facilities in nine countries, the average productivity improvement, UPPH improvement, was 28%. That is unheard of in our industry right now.
It is.
A lot of great work went. That provides a lot of operational efficiency that allows us to be more lean in our RFQ and other responses to customers.
Yeah. Okay. I will ask one more then we will go back to the line. Actually, to clarify the notebook, I think where you said double digit. You are flat year to date, it sounds like, for the industry. Is that double-digit decline more just maybe over second half, or second half falling off even more to get to down teens for the year?
Market-wise, second half 2026 over second half 2025, we think it's about 15% down.
Okay. Your expectation is to be I actually might have mentioned it, but is it to be track in line broadly or to mix up and try to outperform the market? It sounds like you're stable into third quarter.
Yeah. Pretty stable.
Okay. It's more to you have resilience, I think, with being able to take back some share.
That is right.
Okay, good. Let us go back to the line to see if there is some questions.
Thanks, Randy. We do not have any raised hand yet.
Okay. You can also I will have my email if you want to send. If you do not want to ask, send it to my email, randy.abrams@ubs.com, and then I can take some from the floor. So I will look out for them here. Okay. Yeah, to Oh, go ahead.
Yep. Sorry, Randy, we have one question. We allow it? Okay, so. The question is from Irene. Irene, please.
Okay, go ahead, Irene.
Hello, can you hear me?
Yes.
Yes.
Hi, Tony, Tina, and also Jack. Maybe two questions from me. First is, I see that you show some neocloud customers in your presentation slides. Just wondering, within your AI server business, what is their contribution today, and what is the expected mix heading to 2027 from the neocloud customers? That is the first question. The second question is on your CapEx. Perhaps, appreciate if you could share your CapEx guidance for 2026 with us, and maybe also some preliminary colors on the CapEx in 2027. Yeah.
Tony, you start with the server side. So server, maybe I give you some numbers, maybe Tony later can have more details. So server actually accountable, even for first quarter is around 5%, right? We elaborate that before, right? And second quarter server is roughly high single digit for our total revenue.
As you can see on the slide, we are saying among our server revenue, it is roughly 70% now is all driven by the AI server. Right? If you recall, at Compal, we ever say that we target about 10% of server contribution, for 2026. Right? So we are on track of our path to the result for this year on our server side. So this is some data for you. Tony, maybe you can
Yeah, I think the neocloud, some of them we publicly announced. Some of them, they asked for that kind of press material to go out. The neoclouds, I would say, provide a lot of the growth that we saw in Q2. We expect that to continue for the balance of 2026 while we ramp up enterprise customers and we move into more diversified business when our facilities in Texas and here in Taiwan are ready.
I think next year we will see a pretty healthy mix of neocloud, of tier one cloud service provider hyperscalers, and of at least one large enterprise business. So we will see a hopefully very well-diversified customer portfolio in our server business.
Yes. All right.
Cool. There is the 2:30.
Okay.
Yes.
There is our 2:30.
Yeah.
Yeah. If everyone is still online. Should we do the CapEx question?
Sure.
Irene wants for the CapEx question now.
Okay. Right. For the CapEx, I think we ever give a guidance for this year, the CapEx is TWD 18 billion. TWD 18 billion, right? Is the guidance for this year. We are roughly tracking in line. If you have a look at the financial report, details for the first half, we spent around TWD 9 billion. We spent roughly half of that, right? We are on track toward to the TWD 18 billion for full year.
Some additional color here is that among the TWD 9 billion in the first half, we are spending around TWD 6 billion- TWD 7 billion, which is the server related, right? As Tony mentioned that right now we have expansion in Taiwan, in Daxi, right, and Vinh Phuc for the SMT line, as well as in U.S., Texas. Right? From the SMT to the L10, L11.
Overall our spending on the TWD 6 billion- TWD 7 billion is to the server, right? This is the color on the CapEx side. As for the question, is there any color for 2027 CapEx? We haven't yet have the number, but overall, as you can see that if the server continue to be the very high-growth potential for Compal as a big-growth pillar, right, for the next few years, and we continue need to have the capacity support. We think like this year and the next year definitely will be the expansion year, capacity year for Compal. Yeah.
Okay. Is that TWD 6 billion -TWD 7 billion out of the TWD 9 billion or out of the total full year CapEx for server?
TWD 6 billion - TWD 7 billion is just for the first half.
Just first half.
Right. But second half, we continue spending for Taiwan, for Vinh Phuc, and for-
Okay
Yeah, for U.S. Right.
Okay. Irene, do you have a follow-up question? Okay. Okay, Max, are you showing other questions at this-
Yes.
Yeah, good. Oh, go ahead, Irene.
Sorry. Yes. Maybe also a housekeeping question for me. I see that OpEx ratio was down the second quarter. You mentioned it is primarily from the operating leverage. I am wondering, how should we model for the OpEx, heading to second half or even to 2027?
Right. This is Tina. Let me give you some color on the OpEx. OpEx actually is TWD 8.1 billion, TWD 8.2 billion, right? So absolute dollar in the second quarter, right? Actually sequentially it is growth. Sequentially it is growth. The ratio wise definitely is this decrease is because we join operating leverage, right?
We have expansion on the top line of the revenue side, right? The company, actually, we give the guidance for the OpEx is on the absolute dollar, right? We ever say that, the company will continue to will be managed well on overall the spending to have a more efficiently to use our overall resources. We are targeting about roughly the single digit growth, right, year-over-year on the OpEx, the dollar wise. This is continue to be our target and our goal for this year.
If you look at the first half, the number, the year-over-year growth is around 3% year-over-year growth on OpEx. It is roughly in line with our revenue and overall the company target.
Okay. Great. Yeah. Thank you. Okay, Max. If Irene's finished, Max, is there another question on the prompt?
We do not have any questions yet.
Okay. A few follow-ups on the server. Actually, one to start with the traditional server business. There has been a lot of this strip from the Agentic AI. Do you see that. So it looks like a lot of the investment is AI server, but is there potential you see in terms of more volume to go after in the traditional server market?
Yeah, I think so. We are not seeing that translate into demand just yet, but yes.
Yeah.
The agentic stuff is very interesting. We were talking before the call that the article I just read said that the world's built less than 1% of the compute power that we need to. Especially when you're talking agents and a software programmer goes from an agent to managing 50- 100 always-on and live active agents. The silicon just isn't there. Yeah, so we see all kinds of interesting opportunities for inferencing, for agent, for agentic compute, if you will, and other data center infrastructure.
Okay. Turning to the AI business, talk about the transition. Is most of the business still L6? When you start to bring up the Taiwan and the Texas site, how does that transition over to L10 and expand a bit? It sounds like it's a customer base and also a mix-up where you can get more content per server.
Yes. The L6 business. The L10, L11 business is over becoming our L6 business. That is why you see the significant revenue growth. Do you have any other comment on that?
Actually, if you look at the first half, the number we have, the major contribution on the AI server actually is already L10.
Yeah.
It is already
It is about 80%, 70%.
Yeah, it is about 70%. About 70%.
About 70%, yeah.
The rest of that is the general purposes. That will be the L6. That was the L6.
Most of that revenue is from B300, and most of that being L10 manufactured here in Taipei today.
Yeah.
When the other facilities get ready, of course, we'll have much more capacity, and we'll be able to do more regional customer service.
Okay. How are you thinking on product transition? We're getting a little bit later in the Blackwell family, starting to get into Rubin. How do you see just the continued B300 just continue to go into your end early next year. Is there any transition pause? When do you see us kind of moving up to the Rubin generation?
Rubin CPU.
Yeah.
Yeah. I think there's these product generations move so quickly, by the time you get a platform and get it debugged and get it producing tokens, you're already onto the next generation, yeah. We don't see that slowing down. We don't see much tail either, much overlap between those products. We think B300 runs strong. The demand is very strong through this year, and as the Vera CPU comes in, we'll start to see that take over. That transition will happen, I think, pretty quickly.
Okay. As you transition, it looks like from market, all this high component costs, the ASPs, the pricing of these systems should be much higher. How's the thought in terms of margin percent? Is kind of strategy keep when you're negotiating dollar margin. I had a notebook.
One of the ODMs for market was a dollar margin, but try to move toward a percent margin. Should we think of it as dollar, so components inflate, it's a good dollar profit, but it's a lower gross margin? Or just if there's a way to think, because that's a huge ramp-up in the price of the systems, Blackwell or Rubin, or just in general with these AI servers.
Yeah. We're not percentage-based.
Yeah.
Yeah. Not percentage-based. So, like we're seeing with notebooks, you're seeing the margin compression actually happen as the ASP increasing. It would be great if it were percentage-based.
If it were percentage, yeah. I understand. Yeah, what we thought was a little optimistic on that. In terms of components, it sounds like it threw a demand pressure in PC, but how much is supply, whether it is CPU supply or memory supply, actually limiting shipments? Is it an issue for some of the server ramps, getting components?
Yeah
Is there enough prioritization to those that there is less issue with ramping up the servers?
Yeah. Depending on what customer we are talking to or dealing with, the buy-sell components may be different.
Yeah.
They manage a lot of that relation, some of that relationship, but it is everything. If you do not do the supply chain dance very carefully with PCB and even just wiring and connectors and everything, if you do not have that plan, have enough DSI, have enough POs to secure supply, you are going to be short of something, for sure. We have seen, for the most part, like in Q2, for example, we had a dip in May only because we could not get the parts, but then we had a very strong June because we got the parts.
Okay.
I think that dance continues forever, it seems like.
Okay. It sounds like you guided. I should go back to big server, high double-digit growth is the outlook. When you think the high double, are we thinking, is there a way to think it is because I have heard different things for high double. It can be 30% - 50%, but it could actually be closer to triple digit.
The type of acceleration and how you see that there will be a higher and higher base. It feels like you start to get to 10% by, if your year-end target was 10%, it could pull in. Does it look like it kind of continues into year-end, or there is some data on the supply side?
Yes. Could be. It depends. It just depends on how Q4. A lot of it is truly material base. Right? Again, it is customer you are dealing with, can they get the parts that they are bringing to you from a buy-sell perspective? Do we get enough early warning on the demand that we can go secure the parts that we are responsible for?
Yeah. Do you have it? Then I will take it back to the line in a second. Maybe just one thought. Do you have it, like strategy midterm to go after it? It would take a different amount of working capital to do the bigger racks like this NVL72. That is one side. I am curious too, if you see opportunity. I think at COMPUTEX you showed MI, like AMD boards.
Yeah.
They are also moving up to rack scale.
Yeah.
Do you see on both camps kind of that opportunity to go after?
Yeah
Bigger rack scale?
Yeah. We believe the larger rack manufacturing will start in the Q1 timeframe of 2027.
Okay.
Yeah.
Okay, good. Actually, I'll go back to Max if questions on the line.
Yeah. We have next question from Anthony. Anthony, please.
Oh, thank you. Thank you for taking my question. I have a quick question regarding the server business, especially on the margin side. How do you see this server profitability, given that you already are really on track on the 10% of total sales scale this year? How do you see this server margin trend, going forward, given the dynamics between switching server and AI server, and also the mix between L10 plus product versus L6 board level? How do you see this margin trajectory, and could you share some color for this?
Right. Maybe I give us a number first, then Tony Bonadero can follow on more color, right? Right now, if you look at the server margin, right? If you compare to few quarters ago, we are doing the L6, right? L6 is a more moderate level, so the margin definitely is a bit higher. Right? When we move into the L10 system, right, so we have a higher ASP.
But if you look at the Q1, Q2, well, currently we have in the first half our servers are. Servers still enjoy a very healthy growth margin. So server growth margin at the first half is still higher than our corporate average. Right? Afterwards, but of course, we are going to have a different customer to come in, different project to come in. That would be definitely depends on the different projects and the customers on the margin profile. Tony , if you want to add.
No, I think that is right. I think the margin profile of the customer will vary. I think neocloud and what we are shipping today, B300 and other things, can have more profit than others. If you look at the large rack systems we were just talking about, given that the GPU silicon providers take home such a huge portion of the BOM, it is what it is. It is what it is.
So we believe that our strategy is to make sure that we execute well in the factory, that we have high yield rates. We make sure we are not spending money or letting margin leak, if you will, where we do not need to. But also making sure our offering includes lifecycle management and all the back-end stuff that we believe has some significant profitability to it as well.
Yeah.
Okay. Anthony, do you have a follow-up question?
Yes, I do have. Thank you. So, follow-up on this margin. When you talk about the healthy margin and higher than corporate average, does that refer to both gross margin level and OpEx level? A quick follow-up will be, if we look at just 70% of AI server in the first half, does that also apply to your comment on the higher than corporate margin?
Yes, I think you are right. Yes.
Got it. Thank you. Back to you now. Thank you.
Thank you.
Thank you.
Okay. Actually, a couple follow-ups on the server margin. There has been some talk about industry consignment models, where certain things like memory you can, whether have it off the balance sheet or not showing up. I should not say off the balance sheet, but not showing up in your buy-sell.
Right.
Is there any negotiation going on or any potential some of these cost shift to consignment, or it still looks like a lot of the AI server, what you're doing, but mostly going to stay buy-sell model?
I would say most people are comfortable with what they know, so buy-sell seems to be the most thing. However, I will also say that we're in an age of very creative finance tools and other things. So people are having lots of discussions about how you can do that. We've had a lot of those discussions, but right now. It will vary customer by customer, of course. But right now, we don't have a consignment model. We don't have an operational customer with a consignment.
Yeah. Okay. To blend the margin where it's above corporate, is it a thought you go for a little bit of a transition period as you bring up U.S. and Taiwan site? So maybe it goes toward corporate and then the mid long term, relative to notebook business, accretive to margin. Is there kind of a thought, maybe it's a flow where it could come down a bit or come toward corporate or below corporate and then matures?
But really, I think for the margin thing, I think what's important for Compal is that still look at the absolute TWD dollar wise, right? So the margin ratio could be different, depends on the customer's business model as well as, you asking about buy-sell consignment, right? There's a lot of a different arrangement and even on the L10, L11. What's all important is that which is EPS accretive to the company, right?
The reason why we highlight that, even though if you look at second quarter, we have the same margin ratio dilution, right? Because we have the different product mix, right, so the customer mix. If you look at the absolute TWD 1, it actually compares turning back to the growth, right? This is what the company focus on, right? I think that is so for our focus.
Yeah.
Yeah.
Let's see if we have a question from the line.
We don't have any questions.
Okay. I will just ask a couple final ones, then we will go back to the line one more time just to see if any questions. Actually two, since you brought the financing, not to say creative financing, but when you start to go toward rack scale business, just how you see in terms of different working capital requirements, if you think kind of debt or additional financing, or you feel like there is pretty good headroom to grow the business with the resource you have.
Yes.
Yeah.
I think we have looked at the cash we need to run that business, and you do not need it all on day one, right? It ramps over time, but we have facilities to take care of that.
Okay. Makes sense. Then a question on other growth pillars, like a lot on server, but just between auto, medical, 5G, smart device. It seems like the consumer side probably headwinds because of the kind of inflation. But other sides are kind of more growth areas.
Yeah. We're really interested in this PolyMedics thing that we did, the physical AI. We've been talking about robotics, and we've been talking about MedTech and healthcare for a long time, something our chairman is very passionate about, the healthcare space. We started this project about a year ago, and I've embedded many members of our innovation team, have spent months in hospital wards with clinicians, understanding their pain points and their problems, and then turning them into use cases and solutions for this physical AI.
We're really excited about that. We think it's something that, again, can be deployed now. Technology's available today. We're not waiting for it. We're kind of leaning into that business, so we like that one a lot. Automotive, we've kind of pivoted away from not very attractive at all EMS business looked at developing point solutions directly for OEMs.
We have an infrared solution for cars to meet a requirement in 2029 about pedestrian detection and avoidance that we think is a winner, and we're getting that same feedback from a lot of automotive companies.
We like the segment, don't like where we were playing, so kind of moved higher up into the higher margin businesses where we're providing value and real solutions, real point solutions for global automotive OEMs. We'll continue to invest in the consumer stuff. It's just not a great year to have a consumer. We need COVID. That's the one time we need a COVID. But not a great year to get new components.
The last point, final note, then we'll see on the line. Initial take, looking toward year-end, fourth quarter, I know we're not necessarily at guidance time, but it's a big step up on server if you kind of see that we're still in that phase where each quarter it should be stepping up, and then it sounds like we should expect outside of server nothing exciting into year-end, like for the notebook and the consumer business. But server, should we get another step up where it actually could bring up the whole company outlook looking beyond the third quarter?
Oh, yeah. In 4Q, we see the AI server momentum to remain healthy.
Yeah.
Yeah. It should start building.
Okay.
One of the toughest obstacles for us to overcome was that we love you guys on the devices side, on the consumer side, commercial PC side, et cetera. But you haven't done this. Now we're doing it, and we're doing it at scale, and we're about to start doing a whole lot more.
Yeah.
I think that gives a lot of confidence, and we expect that to continue into Q4.
Yeah. Sorry to ask one more, but is it a very competitive bidding environment for servers? It feels like it's strong environment. Everyone's trying to get these data centers up. Notebooks are I feel they're somewhat always competitive. How does it feel, like competitive intensity as you're coming into the market? Everyone kind of fighting for share, or everyone grabbing a lot of opportunity?
Yeah. I think it varies, right? I think some of the cloud providers have very aggressive deployment plans and need capacity, right? They need capacity. It's not like you can really go shop these things around and then say, "Oh, this guy's TWD 10 cheaper. I'll let them build it." Because there's so much enablement that needs to happen for an ODM to serve a cloud service provider.
First of all, they have to come in, look at your factory, make sure you're capable, et cetera, do all that type of stuff. Then they have a whole set of requirements that you have to adhere to. Then you have to handshake. You have to have an IT connection. You have to have all That's not insignificant.
Yeah.
Once you're in, competition among the established ODMs can be tough.
Sure. Yeah.
They will all stay within a certain margin range. To get in is very hard, and we found ourselves at a very fortunate time in the industry where people are looking for more capacity and therefore more trusted partners.
Got it.
They like not only the facilities we are bringing online, but the quality of the facilities, the quality of what we are putting inside those buildings, the investment and automation that we are making, and the team that we built, the humans that are running this. We built a really, really good team from all over the industry that does have that experience. It continues.
Okay, that is good. Actually, Max, I will see if any final questions from the line.
Yeah. We have a question from Anthony. Anthony, please.
Thank you. Thank you for taking my final question. I have a couple of follow-ups. Sorry, still about server side. Just want to double-click on your server customer. If I heard you right, you mentioned you may have one new tier one neocloud and also a large U.S. enterprise as customer. Is it into next year, or they will start to contribute in the second half this year? Is that an L10 plus kind of products or like a MI300 kind of product?
Yeah, to be conservative, I would say that we will prepare for it in the second half of 2026, and we will ramp it in very early Q1 2027, and we will see the business expand from kind of what we're doing today with B300 and other things into get into full rack solution. But we'll see that volume, the OEM volume, really in 2027.
Okay. Got it. Yeah, that leads you, I think, some news in your Daxi factory announcement opening. Chairman mentioned some 30%-40% AR revenue contribution next year. Should we confirm it as a guidance, or that is not your official comments?
No, that is the guidance, right? We believe that 2027 that we will see 30%-40% of Compal's revenue generated from the server business, and a vast majority of that generated from AI server.
Got it. In that case, the PC mix should dip below 50% next year, right? Because you still have other business in the non-PC business.
Yeah. Certainly, if we hit those numbers, that is true, because we have other smart device businesses, phone business, other businesses that way. So our percentage of revenue PCs will dip below 50%, yes.
Got it. Got it.
But we still like that business. We still like that business. We are still investing in that business. We want to grow that business, but we want it to continue to become a smaller portion of revenue only because the others are obviously growing.
More balanced than before.
Yeah, yeah.
Understood. Makes sense. My final question would be to support this very strong AI server revenue growth next year. In terms of your balance sheet now, I think it is almost going to, say, a little bit into net that area soon. How do you see this, your capital plan? Will you, say, announce any convertible bonds, and how do you look at your capital structure? Will you leverage more on debt or, say, equity increase?
Yeah. I think, Tony, I give the answer first, right? Overall, Anthony, I think you are right. If you look at it cash flow wise, definitely we are expanding, right? The spending on the cash flow definitely support the business growth. But in terms of, as we said, like a bank facility, we still quite have a capacity there.
If you look at the reliability ratio of 71% compared to right now, the industry, the average growth, we are still relatively the household level. Bank facility-wise, we still have some good support over there. But of course, we were going to prepare, right, for the future, if the business continue to have a very high growth potential. We are also at the same time to look at the different, the funding vehicles, right? We are not excluded the possibility, right?
For the other vehicles, like everyone talking about the ECB or the other equity financing, but we are all paying attention on that, but it will be really, really depends on how is the further progress of the business. Yeah.
Okay. All right. Good. I will just squeeze one final, and then I will turn it to Tony for closing remarks. I am actually curious because you brought up the 30%- 40% revenue. Operating, you know, you touched OpEx is kind of growing single digit this year. Next year, do you get leverage or to support that kind of growth, we should maybe accelerate the OpEx growth? It is kind of early, but do you need to start growing more double-digit OpEx just to handle that kind of big server expansion?
Yeah. Tony, I think rather than to give a specific, the numbers guidance of that, I think the rationale behind that, of course, for the growing business where we give the actual choices, right? We are spending OpEx on that, right? But the other part, we are also doing the reallocation of the resources, how to spend in our OpEx more effectively efficiency. We are doing that kind of a balancing. That is the reason why you see the result, right? The company, we can still manage overall the OpEx growth on still the reasonable level.
Yeah. Okay, great. With that, I will turn it over to Tony to give some closing remarks. Thank you.
Yeah. Again, perfect timing. Okay, guys. Thanks for joining the call. Hopefully, you see the effects of our transformation and what that has done for Compal so far. Again, we have been at it for about two years. It took about six to seven months to do the diagnostics and to get the mobilization ready, and then we have been at it for about 18 months doing the transformation. Now we are really starting to see it pay off.
We have made the bets. We have made the investments. We have the facilities coming online in the second half, hired the teams and the right people to manage them. Across acceleration, future care, connectivity, creativity, everything we showed at COMPUTEX, kind of what I termed our trailer, and what I say is we are not just a PC company anymore. It is true. We are now executing. We are not just talking about those businesses.
We are making bets, executing on those bets, writing checks, building buildings, hiring people, and we are very encouraged by what we think is ahead of schedule results for the server business, which we, again, continue to see through the remainder of 2026. Then we think 2027 could be a really bang-up year for us.
The transformation continues. Schedule right now is through 2027. We aim to be the most digital-enabled ODM on the planet. We have seen firsthand the impacts of GenAI and now in the last quarter or so, agentic AI, as we start rolling that out in certain areas, and how that is driving operational efficiencies. We are excited. We are excited about what the future brings. Thank you very much.
Okay. All right. I want to thank everyone for joining. Look out for the replay link. Yeah, at this time, I think everyone can disconnect. Thanks, everyone, for joining. Thank you to Tony, Tina. Yeah, we'll join you again next quarter, I think. Yeah, thank you.
Thanks, everyone.
Take care, everyone. Thank you. Bye.