Asia Vital Components Co., Ltd. (3017) Earnings Call Transcript & Summary

Okay. So in general, both GPU, I mean, only -- I'm only referring to [ co-pay ] itself, in general, both GPU and ASIC product that decide complicatedly of copays getting more and more complicated. So in [indiscernible], it may be obviously one piece of [ co-play ]. It means achieve and it's really a simple piece of co-play. Now when we say [ one ] [indiscernible] of co-play, it may be contained like several pieces co-play and this may contain -- it has lots of different types. Maybe [ 1 2 3 1 2 4 ] maybe upside patent side, both sides are co-play [indiscernible]. So I mean, in general, the complexity of co-play itself become more and more completed. So that's why we are able to provide our [ g ] the value to start to work with client in a very early stage to contribute to our capability in the go through strength through the [indiscernible] passes and in the end, success for these mostly comes production and during the [ NPI ] phase. Of course, when the product is getting more and more complicated, more and more complicated, the [indiscernible], for sure, will increase. So this is a general trend we see in both GPU and ASIC margin for co-play itself.

Yes. Thanks, Eric. I'm Matthew. And I would just like to add to that answer. I think Eric already gave a very complete answer, but what I'd like to add is, when we look at spec in terms of thermal, we really are looking at a multitude of factors. So GDP would be one, voice would be another, form factor is another, system complexity, as Eric mentioned earlier, is another. And all of these have a direct impact on the design complexity of what we deliver to our customers. Now we look past the '26 into '27, I suppose, already on very many exhibitions such as [ GTC ], such as [ Computex ], you, our investors and including ourselves would already gained some insight into the immediate next generation or next 2 generation. Past that point, our thermal is really heavily dictated by what the customer system-level solution looks like. And what that means is before our customers have decided what they want, specifically in what they want -- they want their system, they want their rack, they want their trade to look like, if it comes very difficult for us to share in a nonspeculative manner of what we think the future trajectory is. Now that being said, I think history is always a good feature. And if we look at the previous generations leading up to this moment in time, what we will be -- what we can clearly see is that system complexity has continued to increase, from NVIDIA's HGX platform to their MGX platform, going from [ A ] cards in an HTS tray into 72 GPUs in the MGS system. So that massive step-up in complexity also was always driving a lot of our growth in the past few quarters. If we look on the ASIC side, the same story is developing, where you need more complicated system-level solutions where customers are trying to put more components in the [ tray ], trying to put harder components in a tray, that's been driving a significant architectural change in the tray level design, which then gives us a boost not only in content but also on margins. We should continue to expect seeing this going forward. Now I understand that was maybe a bit of a nonanswer. If we look at who we're working on right now and what we are trying and exploring right now with our customers, with our future -- on our future projects, the things that we're investigating right now include changing the [ TIM ] material, changing the material of the cold plate, controlling for weight, controlling for hotspot on the cold plate, continuing to push for higher GDP, continuing to push for higher flow speeds and also looking into 2-phase liquid cooling. So all of these directions are potential directions that customers will continue to want further improvements are on. And because every system is so different, we are having to look at all of these different aspects, look at these different development trajectories to really make sure that we can continue to stay on top of what our customers' future -- well, current and then future system architectures will look like. So just to add some colors there.

Yes. Now shifting gears to maybe more of a short-term question. Would you say you are more optimistic in terms of, let's say order demand on some certain projects versus say, a quarter ago or a couple of months ago?

I will answer this first and if, Matthew, you have other input, please go ahead later, yes. Like what I mentioned previously that AI demand is real. The demand is still very strong and will continue to spend based on our clients real order, real focused that we even received some down payments first before our expenditure. So we are still after we speak for the overall AI -- general AI product and we have high confidence that we will be part of the journey growth with our client [ LCD ].

Yes. I would just like to add to that, if we're looking at the last few quarters, it always is the case where when we are further away from the delivery, the mass production date, we are more cautious about forecasts that our customers give us. And I think Eric mentioned a key point earlier, which is that we have received prepayment on our expansion or received prepayment on the capacity that we are delivering to our customers. Ultimately, the forecasts have -- well, on select projects, have continued to be revised upwards. What really gives us the confidence that this forecast is real, that we can continue to expect strength in AI is customer willingness to pay us beforehand. And at this current moment, many of our customers are willing to do this for us, to prepay us for expansion, to prepay us for capacity, to prepay us for building buffer stock and inventory for them. And so as we approach the end of Q3, these are the signals that we are paying more attention to rather than just looking at the forecasted figures, which are historically always prone to change.

[Operator Instructions] [indiscernible]

I'm like so whether you can shed some light on the art and the science of setting ASP on products that are going through a significant capacity expansion. So is there any rule of thumb and say, for example, your ASP will go down x percent if capacity go up by y percent because you can then share lots of that value with your customers, and that will help you getting further orders because you are very competitive on capacity delivery and the price. So like how do you -- what's your philosophy offsetting that ASP right? Is gross margin a target or goal in doing so? Or is that just an outcome?

Thank you. Overall, I would say, it is our outcome is a result. We do not -- listen, there's no methodology or secret recipe that we can see. Okay, I do this percentage of expansion, and I can gain how many percent of the ASP. We do not. What we can do in corporate level is -- the only principle is, the more complicated product, usually, we can go more greater margin. And when it's a small margin budget, we also charge higher margin when it is a super big margin. And then this is, of course, this is negotiated. But there's no clear card, no methodology that how many percentage increase on capacity equal to how many percentage of ASP increase this might be. Matthew, you want to add more color on this?

Yes. Thank you, Eric, and thank you, [ Pam ], for the question. It's truly one that is difficult to answer directly. And between it being a science and an art, it definitely is more of an art. And the reason it's more of an art is because when we do pitch an ASC or when we do work on a project with our customers, the ASP content is heavily dependent on a multitude of factors. What can be said that is common across all sorts of different thermal solutions, whether it's fans, whether it's chassis, whether it's liquid cooling module is that we, as a company, continue to value a long-term partnership with our customers, right? So what we are trying to do is to be a long-term solutions partner with our customer. What that means is the ASP price isn't something whereby we just try to intensively push up our price. We try to exploit market capacity gas. We are trying to pitch prices at a reasonable amount. That means we can cover the 100, 200 engineers that we put in on each project, but also means that we can continue to form a long-standing relationship with our customers to continue to be their main source to -- for our customers to continue to be happy for us to be their main source. And it definitely is a balance between many things. It's a balance between how many engineers we have to put in. It's a balance between what time to market that customers want, system complexity, how much time -- the time lines that we have to work to as well as some of the other aspect that was mentioned earlier, waste, GDP, et cetera. And I suppose what is good for ASP is always higher spec, higher difficulty and higher complexity products will always be accretive to our ASP and margins, which is actually what we've seen in the past few quarters, and also why AI data centers and AI server solutions will continue to drive our margin profile upwards because it is fundamentally a more complex solution than the historical products that we made in laptops, in desktops, et cetera.

Okay. That's super helpful. I have 2 more, please, if that's okay. The first one, just that if you think about your supply chain and anything that's really important to delivering the capacity growth that you have planned for, is there anything that particular of concern? Are you supply chain ready? Do you -- or are you still able to recruit and retain the engineering tenants that you'd like to have? So that's the first question. And the second question is I wonder whether you can comment a bit more on growth potential in China and AI customers or within that market in general?

Thank you for the question. It's a very good question. So I will answer the first one first. Yes, the question itself is also a good improvement to our overall capability. We will say capability is not only a [indiscernible]. It's an end-to-end, how we can make the product matters mostly use certain margin continuously. So for growth to us, co-play itself mean for itself, they are also a module, which contains lots of sub-tier components. So towards the [indiscernible] voice, first of all, I need to have the season defect capability of this module. Secondly, I want to have in-house manufacturer capabilities for those important sub-tier components, to some example, co-play [ 2Ds ] are very important self. So that's why we have our subsidiary for detail. They are -- they have a talent team working on this. And other [indiscernible], the tube, the host. Lots of competitors lots of clients, they think the whole site. When you take a shower in your home, that's what is the normal stuff, but to us not, no. It's a very important product. You need to keep it very precise. It cannot have the [indiscernible] and you need to have a stable sufficient output. So [indiscernible], you can have the whole module, human smoothly. So we also have the -- so we have those in-house capability, manufacture capability for those important components. And then the least is continues to be reviewed [indiscernible]. That's still the host itself. In a positive generation when we work with [indiscernible], almost, I would say, of the module makers, including AVC, the main capacity is [ outsourced ]. [indiscernible] it's a normal component but starting from the second generation to us, this is very important. So we [indiscernible] we urge to the management team and that will start to establish of our in-house capability. Before we already have the [indiscernible] capability, but then we adopt more manufacturer capability. So this is a very good case to help us overall to continue to improve our supply chain company to our shipment and also improve our cost structure. And those make us continue outstanding among the competition. This is our first question. And the second one?

Yes. I think the second one was a question on our opportunity in China. But just before we jump to the second question, I'd just like to add to Eric's answer, Pam, which is on talent and talent retention, we are -- there are several things that we're doing. Number one, we understand that talent -- to acquire talent isn't always just to pay the big bucks. We continue to offer the best engineers in the field, an opportunity to be working on the most critical projects in all of technology right now. This is an edge that only AVC has. We are participating in absolutely every project that every CSP has, that NVIDIA has, and we're continuing to work on next-generation solutions even for robotics, space or autonomous vehicles. So I suppose in talent retention, not only do we offer the first opportunity to work on, well, the most exciting projects. We are also setting up new offices, new locations to cater to different potential hires as well as, I think probably -- if we come back on the monetary part, we have about TWD 10 billion of options outstanding at this moment to continue to retain our best talent. And this is what enables us to have an engineering team that's more than 1,800 people to continue to have them build experience with us and continue to contribute their expertise into the most exciting project. Now the next question on China. I'll let Eric answer first.

Thanks, Matthew. For China, what we can share right now is since day 1, even before pandemic, AVC was able to join all the major ASIC project with China-based client, and the trend is still ongoing. We continue doing those ASIC project development, not only the [ green ] product, but also there are still much and many projects. They are [indiscernible] quickly. So we participate both of land and also for -- not only co-play itself, but also other part, even [indiscernible] with those products, we are also part of that. So we believe we will continue to have healthy growth in China market. Matthew? Thank you.

Yes. Thank you, Eric. To add to that answer, we have continued to be long-standing partners with China CSPs even before AI infrastructure, before AI ASIC took the spotlight of the tech road. So we are very familiar with China customers. We have continued to form, once again, a long-standing partnership relationship with Chinese suppliers. 5 of our 6 manufacturing facilities are in China, which enables us to supply to them relatively flexibly. And the main difference between AVC and Chinese local producers is, number one, we are maybe 5x, 10x the size of our Chinese competitors. And number two, our Chinese competitors don't meet us in terms of the design capability, the architectural solution design capability that we are able to provide. Our Chinese competitors are, for the most part, really good competitors in their ability to manufacture. But the ability to design is something that needs to be built up over time and needs to be just up over working on very many projects. And as such, for Chinese CSPs and Chinese startups in AI, we continue to be #1 choice across the board. Now in terms of our opportunity in China, once again, this is heavily dependent on what our customers want to do in their own ramp. We continue to remain committed to supporting their journey, but the growth trajectory and the forecast and the growth outlook, whether it's this year, next year, '28, is heavily dependent on what our customers' own strategic plans in AI are.

Are there any questions from other investors?

It's Pam again. I can always have one more if there's nobody else asking.

Sure, go ahead, please.

Okay. Well, just because you mentioned in the -- when we talk about talent, to attract them, you measure robotic state and autonomous. Just out of curiosity, to the extent that you could disclose, what would be your involvement or focus in the space?

Okay. Okay. For these parts, I can only share very wide, very roughly. Our client in space area, we do work with a few clients right now. And there are 2 supply [indiscernible]. First of all, [indiscernible] itself. Secondarily, the so-called AI data center on the space. We do work with clients, different clients in both areas. And the sum of client, they have -- they are under one big umbrella. So we work with the -- maybe the big [ PUA ] in the past. Now we are able to work with the PUP, PUC, PUD together because you need them, it is a technology, it's a product. So we have a very good trader. We have a long future coworking with [indiscernible] in many product mini field. So when clients come from A to B to C, now we see no AVC can help them co-work with them together on the journey and the [indiscernible] product that they want and follow the regulation doesn't simple to your diversity. Some example, the requirement they have, either human itself either location, either on products, we can all spot in. So if you comfortable to work with us. Thank you. Matthew, you want to add more?

I don't have much to add, but that -- I suppose what does excite us is the space piece because, well, it's -- going to space is objectively a very interesting segment. And as Eric mentioned, we continue to support our customers. We continue to leverage our long-term relationship with the leaders in tech to enable ourselves to continue to be #1 in any type generation. Maybe 5 years down the line, there's -- we're going to Mars. Over 10 years down the line, we're going to Pluto. We are fully confident that in any tech integration, in any solution integration, we continue to be our customers' #1 choice. Well, just to clarify, like going to Mars and Pluto is not currently in our pipeline. But just as I mentioned to say, we are committed to that long-term trajectory rather than just trying to win a project by project.

Okay. That's super helpful. And I know it's a tricky question. But I'm also more thinking about like what would be the cooling method in the space, if not conduction connection. So I'm just wondering, is there anything on the sort of science breakthrough front that you might be able to share? Or I'm also happy to wait and see when you move to product.

Very high labor speaking. It's a very advanced merging, super improved merging of existing product, existing technology. It's also conduction, it's also convection, it's also kind of evaporation. Those are all in review or in discussion. But in the end, it's just a more of the best version of existing products. Thank you.

Yes. I think to add to that, in space, our conduction, our convention, our evaporation can only happen in combined components. Ultimately, to expel the heat into space, we would have to rely on radiative cooling, which is the same way that the heat from the sun gets to the earth. This is less effective and more unavailable compared to, say, air cooling on earth. But this is exactly why we need to continue to work on finding the best way to run it, not only to dispel the heat, but also to run in a way that enables reliability because we are not able -- when something is sent into the space, there's no serviceability. There's no way to fix it if something goes wrong. So we are pushing our boundaries in multiple directions, not just in how do we expel the heat, but also how do you make some things that can be in space, that can last the lifetime of the said light -- of the space data center, and these requirements are higher, significantly higher than what we would expect on earth.

If there are no further questions left hand today's call, thank you all for joining us today, and thank you, Eric, Matthew and Bill, for your time and insights. Have a good day. Bye-bye.

All right. Thanks, everyone, and thanks, Pam. Thanks, Timson, for the questions, and everyone else who submitted early. See you in the next quarter.