Hi. Good afternoon and good morning, everyone. Thank you for joining us today for Asia Vital Components First Quarter 2026 Results Webcast. AVC is the leading solution provider for total solution of the thermal and the mechanical solution that offers liquid cooling solutions such as cold plates, manifolds, and the CDUs, as well as thermal modules like cooling fans, heat pipes, and heat sinks, et cetera. My name is Shannon Shi, and I am the coverage analyst here at Morgan Stanley. We are very honored to have Matthew Shen, the Spokesperson of AVC, and Bill Chen, the IR Manager here with us today. We look forward to their insights and the comments on the company and the overall industry. The management team will first walk us through first quarter results and also provide some forward-looking commentary.
After that, we will open it up for questions. At any time during the webcast, you can send in your questions in the text box on your screen. Now, I would like to turn it over to Bill for opening remarks.
Thank you, Shannon. Hello to all investors, analysts, and media friends. Thank you for taking the time to join Asia Vital Components Q1 2026 Earnings Call. I'm today's Spokesperson, Bill. Before we begin, as usual, I want to remind everyone the forward-looking statement in today's presentation are based on the company's current view and the forecast of the future business environment. Actual business condition may be affected by various uncontrollable factors such as the global economy, industry competition, and raw material prices, and the supply chain. Investors are advised to make decisions based on their own judgment. First, let's look back at the first quarter of 2026. For AVC, this was a very memorable quarter.
Benefiting from the explosive demand for AI servers and our integration advantage in the total thermal solutions, both revenue and profit this quarter set a new single quarter historical records. The single quarter revenue reached TWD 49 billion and a 110% growth compared to the same period last year. At the same time, the single quarter gross margin reached 29.8%, and the single quarter EPS hit 20.2 for the first time. Even though there were fewer working days in the first quarter and is a traditional slow season for electronics, the addition of AI server product made our slow season not slow at all. We have also witnessed the trend of transition from the traditional fans and the 3D VC cooling technology to liquid cooling systems.
This not only increase our ASP, but also significantly improve our product mix. We believe this trend will continue. I will quickly go over the CAGR and the financial data. If any investors want to look closely, we will upload this the detail to the public information platform after the meeting. Basically, the company's revenue, EPS, and the gross margin are getting better every year. We will continue to work harder to keep deliver great results in both revenue and profit. Next, let's report on the revenue by product. Our thermal product account for 63.6% in the first quarter, showing a significant growth compared to both the same period last year and the previous quarter. At the same time, system integration saw a noticeable drop.
This is mainly because the system integration product were pulled ahead of the schedule in the fourth quarter, so there were fewer orders for complete computer related product in the first quarter. This adjustment in proportion will also clearly reflect in our gross margin. Now, looking at the revenue by application. The share of the server and the communication applications has increased to 66.4% of our total revenue. Our complete thermal solution is the main reason server client choose us. Also benefiting from the booming development of the AI, server related shipment will continue to expand in the future, bringing significant change to the company's profit in both quality and quantity. Next, looking at the non-operating income and expense. Basically, there hasn't been much change.
The larger amount are still from customer samples and the tooling revenue, which we classify under others item. We will quickly skip over the balance sheet. Now, looking at the financial ratios for the first quarter. There are no major changes. Also, because our company are pulling goods faster than last year. Our inventory turnover days are short-term significant compared to the same period last year. This is a positive impact for the company as a whole, reducing our inventory pressure. Finally, looking at the cash flow statement. Our capital expansion in the first quarter was over TWD 3.6 billion, once again showing the company's determination to actively expand product production for client orders. At the same time, because operating cash inflow increased, our free cash flow are also grew while expanding production.
This perfectly fit the company's attitude to being steady yet proactive. In conclusion, AVC is at the forefront of the AI wave. We have complete vertical integration capabilities. From cold plates, manifold, fans, and heat sink to chassis, we can provide complete system-level service. At the same time, the company hold a optimistic view on the operational outlook for the full year of 2026, and is committed to creating more long-term value for our shareholders. Let's conclude the financial data report. We will move into Q&A session. Thank you.
Great. Thank you, Bill. Now we will move into the Q&A session. Again, as a reminder, please send in your questions in the text box on your screen, and we will get to them shortly. While we are waiting for the questions to come in, maybe I can kick off the first question. Could we get a sense about the latest outlook for 2nd quarter? Whether we will continue to see business momentum keeps growing, and how does that imply to the margin trend?
Yeah. Hi, everyone. This is Matthew from AVC, and I'll be taking Q&A. Thanks very much, first of all, for Bill giving us a really good overview of our finances. Now, to answer the question on how we view our trajectory for the second half of this year, very much our new products and our new projects with our leading CSP customers are coming out in the second half of this year. That's when we'll start to see things change more significantly. As for Q2, we are still working on the same projects extending from the second half of last year. Naturally, in preparation for what's coming next, our customers are rebalancing their resources.
Despite this, we still expect to see sequential growth throughout the entirety of this year, and we are very, very excited, particularly for the second half of this year.
This implies to the margin trend, right?
Margin, we will not be able to comment quantitatively at this moment, but, you know, it's always been the case in the last few quarters that our margins have improved as a result of product mix. As server and networking continues to be the fastest growing segment for our business, this should have a creative effect on our margins going forwards.
I see. You also mentioned about, you know, second half business probably will be more exciting given much more new projects coming in and also new customer addition. Will you be able to help us understand a little bit about our business momentum split between first half and the second half this year?
Yeah. Once again, I'm not able to share quantitative figures, but we are excited about the second half of this year for our new projects. We are working with all of the Hyperscalers. We're working across America and China. We have the broadest breadth, the largest breadth amongst our peers, and we are excited to roll out our thermal solutions to power the next generation of AI.
I see. At GTC this March, we see that NVIDIA again introduced quite a few different chips and also different rack design variety from Vera Rubin, from Vera CPU Rack to NVIDIA Groq 3 LPX, even we are gonna to see CPO switch rack, et cetera. Are these gonna to all design with liquid cooling? If it's all designed with liquid cooling, we gonna to be again the major cooling components vendor for all the solution?
Yeah. Good question. For any future generation, not just for Vera Rubin, but for NVIDIA, for ASIC, for U.S., for China, any time there's a new piece of tech, our sales teams are always very upfront in trying to win these projects, and this has always been how AVC has grown in the past. We partner ourselves with the cutting edge of tech, providing our best services, our best R&D capabilities to these leaders, continuing to deepen our partnership with them. you know, zoom into right now, at GTC, NVIDIA announced that there was gonna be six types of trays to form their entire ecosystem.
This is very interesting because first of all, what this tells us is to deliver real AI impact and value. You know, you need to have more than just one chip architecture. You need to have some for training, some for inference, some for memory, some for agents. This branching in the architecture is actually presents, you know, what NVIDIA calls an ecosystem, but for us is a great opportunity to further expand our know-how and our expertise in thermal. We are already working with all different companies for all six of these trays. They currently seem to be all fully liquid cooled, and we are very excited to walk at the cutting edge of tech with our leading customers in this way.
Although a lot of these projects, we don't have a set timeline laid out in front of us at this current moment, we are there for when our customers need it. We will work to our customer's timeline and make sure that they can always have the best thermal solutions for whatever project they're building.
I see. That sounds pretty promising, in terms of the overall offerings going into next couple quarters. Also just related to, you know, recent thermal design change, into the new GPU that's going to coming to the market soon. Does that mean the thermal component design upgrade will be paused a bit? Does that mean once all the production yield issue, et cetera, being redesigned, then it's going to be potentially able to be adopted in next generation, maybe ultra generation next year?
This is regarding to, like gold plating.
Yeah, gold plating.
Okay.
Yes.
Yeah. The gold plating change on our end has been largely discussed in, you know, with us and with our investor community as of late. Now I need to mention here that the change in gold plating has not been confirmed. It impacts our other important companies in our supply chain more than it does us. In any case, AVC's mission is to provide the best solutions for our customers in consideration of not just thermal resistance and thermal capabilities, but also in reliability, in serviceability, in availability, manufacturability and capacity, right?
While this hasn't been confirmed on the singular case of gold plating, whichever direction it goes doesn't impact AVC directly because we, as shared last time with our investors in our, in our yearly report, our gold plating for Vera Rubin was going to be outsourced, and therefore, the removal of this spec does not materially impact AVC's trajectory for the second half of this year going to next year. The question on whether gold plating is a technology here to stay or not is a separate discussion. Our understanding is that gold plating is used this time because the thermal interface material was going to be changed from graphene into a liquid metal TIM. The liquid metal TIM needs to bond with gold for it to be effective, and that's why we needed to gold plate our gold.
Liquid metal is actually not a particular new piece of technology. Actually, many gaming consoles already use this for this is a consumer segment to achieve better thermal performance. This remains a viable option, while we don't see it as a inevitable option in the trajectory of thermal thermal development. It is a potential option. We wouldn't say, you know, it's the end all and be all. It's not inevitable.
We're still looking very closely, and we continue to work very closely with our customers to find the best solution that works for them.
Yeah, of course. I think that maybe half year ago, six months ago, everybody has been so worried about the impact from microchannel lid. Definitely it proves out that maybe it's not a near-term threat. Any, like, indication for the next generation thermal component design for next generation GPU that we are working on, and potentially that will help us to see offering value upgrade. Et cetera?
Yeah. The two things to comment on, number one, microchannel lid, and number two, what else are we doing?
Yes.
Microchannel lid, we are still very firm and confident in our position that not only will it not be used in the immediate next generation, we don't see it be used in any generation going forwards in AI servers because of its reliability and serviceability constraints.
AI servers are machines that are meant to run for a very long time, and our customers cannot accept any downtime. The integration of microchannel lid causes very many problems on a system level. When we're looking at having a rack run for very well for a very long period of time, microchannel lid becomes a undesirable option due to its lack of maintenance capabilities. Now, to comment on what technologies we're looking at, number one, I think it's important to highlight that single-phase liquid cooling is not a singular generation of a product. It's multiple generations that we continue to work on. We continue to find ways to improve it and continue to find ways to make breakthroughs so that it can continue to meet our customers' needs going forwards in the next many years.
The analogy is with air cooling. At the very start, air cooling was 20, 30 watts. It became a few hundred watts. In our most recent air cooling solutions, we're capable of removing heat of up to 1,300 watts, right? It progresses by generation, and this is something that we'd like our investors to keep in mind when thinking about single-phase liquid cooling as well. It's not a static technology. We're continuing to make breakthroughs on it, and this is going to be a important path to go down for us and for our competitors, as well as for our customers in the next few years to come.
Apart from single-phase liquid cooling, we're also now very eagerly looking at two-phase liquid cooling, which is when you have instead of cold liquid flowing in, absorbing the heat, and then flowing out as hot liquid, you have a coolant that has quite a low boiling point. It flow in as a liquid. It'll boil on the chip, and then flow out as a gas-liquid mixture. This, we're very excited about for two reasons. Number one, the process of boiling absorbs significantly more heat than simply heating something up, and therefore, this will give us the opportunity to further improve our thermal performance. The other thing is that coolants that boil at a low temperature will also tend to vaporize at room atmosphere at room temperature and atmospheric pressure.
This actually solves a key risk for current models of liquid cooling, which is leakages and damages that it might cause. Swapping to two-phase liquid cooling will entirely evade this problem altogether. Now, we are already able, at this current moment, to deliver up to 15% better thermal performance from our two-phase liquid cooling solutions in a like for like comparison with our single-phase solutions, and we already have end customers, end Hyperscalers that are working with us to co-develop and co-own this technology. While it's not a immediate solution, we're very excited about it. Our customers are very excited about it, and hopefully in the future, this will be part of our thermal portfolio to serve our customers in yet another way.
Yeah, that sounds pretty promising. If these two-phase cooling is adapted, is some of our existing equipment could be applied, or do we need to set up the capacity from scratch to do this two-phase liquid cooling?
Two-phase liquid cooling is an interesting one because we're still in sort of the research phase of this technology. What we can say based on what we know now is the cold plates are still likely to be made out of copper. You still likely need manifolds to, you know, act as inlets and outlets for the coolant.
If we take a closer look at the cold plate itself, it's still likely to be made out of copper, and therefore, our current manufacturing capabilities of copper machining and then copper welding will likely to still be applicable.
What we still haven't yet defined is what extra new components might be required to set up such a system. At that point, we will have more information as to what new manufacturing techniques, what we have to add into the production process for us to deliver these solutions and we'll have more to comment when that time comes.
Okay. I see. Maybe let's switch the gear a little bit for the ASIC group of the customer. You know, a lot of times, for investors, they will wondering, what's the overall liquid cooling penetration for ASIC customer. Do we have a view that, you know, roughly, what's the cooling penetration right now, and then, what would that be maybe in three years from now for the ASIC customers?
Good question. When we broadly talk about liquid penetration, what most of our investors are referring to is what solution archetype is used on the ASIC compute chip itself.
To answer this question first, the majority of solutions that we're developing and deploying from this year onwards will be liquid cooled. There will still be select air cooled, air cooling solutions on the ASIC chips.
That will, that's currently already a minority, and we expect it to continue to transition into air into the next three years. I think if you, if you asked us what we think it'll look like right now, we're probably thinking mostly continue to be mostly liquid going forwards.
I think the more important discussion is, as we've seen for NVIDIA rolling out different architectures for different chips, as well as the transition from even the compute tray for previously only the GPUs and CPUs to be liquid cooled, but in Vera Rubin for the entire tray to be liquid cooled.
I think where we should be putting our focus is actually the penetration for auxiliary components.
rather than just the GPUs and CPUs. I think there isn't much room for change for the GPUs and CPUs because predominantly they are already liquid cooled with some select projects being air cooled. I think the bigger transition and the bigger momentum and even so the bigger opportunity for us is to look at the auxiliary components that are also making this move for the next few years.
Right. Among all these Hyperscalers, for their own customized design, we understand that we have been working very closely with a Hyperscaler A, and that potentially we will start to do the business with Hyperscaler G. Could you maybe let us know a little bit more regarding what's the business development progress for the new customers, no matter it's for Hyperscaler G or the other two Hyperscaler M, whether we are gonna have potential opportunity to work together with them on the ASIC side?
While I'm not able to comment on individual customers as everything is quite confidential, what we can share is we are already working with all of the hyperscaler ASIC projects starting this year. Not only are we working with them, we are also working with practically all of the hyperscaler ASIC projects in China, which will only grow in importance going forwards in the next few years. As AVC is the known leader for thermal design, we are pleased to continue to deliver our solutions and our designs for each ASIC customer. All of the ASIC customers have been longstanding customers of us, and we are proud to continue to serve them in the best way possible and to continue to build the next generation of AI alongside them.
Right. I think at that last earnings call, I think you guys mentioned that maybe for year 2027, we might be able to see the overall ASIC customers' contribution to outgrow the GPU customers. Is that still the case?
Yeah. Yes. It very much is still the case. When we're looking at our capacity planning, our forecasts, our trajectory for both NVIDIA and ASIC, we see our ASIC customers as an aggregate sense growing in terms of their importance. They continue to be very important clients of us. NVIDIA continues to be a very important client for us as well, AMD as well. We're pleased to have the opportunity to partner with all the leaders in AI.
Great. Could you explain how different are cold plates used for NVIDIA and the various ASIC clients? Basically, the client want to understand whether it is just a minor difference in size, surface areas, or water flow pressure, et cetera, or it is a total different solution in terms of raw materials, structure, and the new technology efficiency for each cold plates for each project. Basically, just a bit more colors on that to help understand the AVC's capability over total design solution. How big is the value add on top of the hardware manufacturing? Could you please help elaborate the value of customized service that AVC could provide? That would be very helpful.
Yes. That's a large question, and thank you, Curly, for asking this question. I'll break it in several parts to explain. The difference between cold plates for NVIDIA and ASIC would be, I think a good analogy to use is suit making, right? You're making a different customized blazer for every customer. You'll still need the same stitching techniques. You still need to pick a fabric. According to every customer, you would customize it to exactly how they would want it. This is actually a really good analogy to grow, to pull with our cold plates, where for other hardware components, say power, say memory, there's one thing that you're selling to every customer. For thermal, we're starting from scratch. Every chip size is different.
Every chip heat is different. Number of chips in a tray, how much space we can work with and what the flow rates we're working with is. Based on all of these different parameters, we have to start from scratch to design a cold plate solution that works for that specific customer. Now, keep in mind, when we design thermal solutions, we design it for the entire system. We take a system view of thermal. Therefore, you know, then comes in consideration of how many ASIC or GPUs there are to CPUs. Are there other heat-producing components that sit around the ASIC chip that we can also use the cold plate to cool down? Do we use a heat pipe? Do we use what kind of manifolds do we use?
The entire combination becomes very, very different when you consider it on a system level, even though the underlying principles are similar. In terms of understanding our value added on top of hardware manufacturing for each customer, I think the difference between working with NVIDIA and ASIC lies in NVIDIA ultimately are still a hardware company. They sell chips. They are very, very good at designing hardware, designing systems, and that gives them the ability to also have more insight onto thermal. When we work with NVIDIA, it's often the case where they're more like the architect and we're like the civil engineer. They tell us what overall structure they wanted, and we help them turn that into a reality.
When we work with ASIC, on the other hand, we act both as the architect and the civil engineer, where we're helping them design from scratch what it should look like, how the heat should flow, how the tubing should flow, and how the air should flow. There's quite a fundamental difference between NVIDIA and ASIC on the value that we deliver to them. Just to touch on your final point on the risk of design standardization, this is only a risk for NVIDIA because they make hardware. For NVIDIA, when they vertically integrate, they naturally think about thermal. For ASIC customers, any of the Hyperscalers, when they think about vertical integration, they're not thinking hardware. They're thinking, how can they get their models to also vertically integrate into the application layer.
There's a fundamental difference between, who our customers are when we're working with ASIC versus with NVIDIA that cause how we work with them to be different and the value that we provide to them.
Okay.
To answer that just succinctly, there's not really a risk of design standardization for ASIC customers.
Another question is related to, do we see any, like, price competition here as we are going to increase the exposure to China?
Thank you, thank you, David, for the question on exposure to China and price competition. Actually, we have quite a special position in China, and thermal in China is quite particular in the sense of they have very, very good thermal manufacturers, but they don't have very, very good thermal design capabilities. What this means is, just like when we're working with ASIC customers where we have to design what that architecture looks like, for our Chinese customers, we're seeing the same. They really value our design capabilities, and that's why we aren't, at this current moment, seeing severe price competition in China for our AI products. Now, naturally, when AI and AI infrastructure becomes more of a commodity, we may start to see this grow as a risk.
Since China is still very early in their AI build-out, we don't expect this to be a significant risk, at least for this and the next several years.
Good. Next question comes from Brittany. The question is related to will the China ASIC growth be stronger than that in, from U.S., next year?
It's hard for us to know what the ramp schedules are for each of our customers as they all have their own considerations. We are proud to be first source for most of our Chinese CSP customers, including the larger CSP players. Whether or not they will be ramping is more dependent on them, and we have little control over it. What can be said is we will be ready for when they want to ramp.
Okay. The other question comes from Brian. Brian has two questions. Number one is that last year, we saw strong revenue seasonality in Q3 and Q4. Given this year, we are super exciting about second half, do we potentially see stronger seasonality versus last year, the model transition period? The second question is related to do we see different margin profile for us between ASIC players and NVIDIA?
To answer the first question, it's an interesting question, right? Because last year we saw a strong revenue seasonality in the second half of, well, 2025 was driven by two things really. Number one, it's the smoother ramp of GB300, but also the fact that GB200 wasn't ramping in a very smooth manner, which actually led to Q1 and Q2 last year to be weaker than what we expected in 2024, and also weaker than what the investment community generally expected, right? That's why, you know, Q3 and Q4 looked particularly strong. It's hard for us to, at this current moment, have a good estimate of what how strong or weak seasonality will be for the second half of this year, but I can share a few things.
Number one, we are seeing ramp for, from our customer and forecast from our customers being steeper, but also the first half of this year is not as weak as the first half of last year. These two things push against each other and we can't make a real solid comment at this current moment. And we'll have more to share closer to the time. I think before I answer the second question, the other thing to add is Vera Rubin, the transition into Vera Rubin is not like the transition into GB300. The transition into GB300 was smooth because GB300 practically a 90% similar architecture to GB200. That's what enabled it to run very smoothly.
Rubin is, although on the, on face value is still Oberon, but a lot of how, a lot of ways of working have changed, and we don't see any problems on our end, but we do have to see if once we put everything together, will everything work as intended? It still remains to be a risk going into the second half of this year. Now, on the second question, do we see different margin profiles for us between ASIC players and NVIDIA?
We do see that our ASIC customers and ASIC projects have higher margins than NVIDIA, and once again, this is due to our ability to deliver thermal, more thermal architecture, more thermal design to these customers while we're working with them, versus NVIDIA, who have a very, very strong system team, and they have their own thermal team as well.
Right. At the earlier session for on-site meeting, I think we emphasized that AVC, its position as a thermal and a mechanical solution, vendor. We have been going through a lot of thermal solution related. Maybe we spend a little bit time, mechanical solution related. First of all, for this year, for VR, generation, we are going to see inner manifold and also rack manifold to be adapted. How do we see our business opportunity there?
Thanks, Shannon. Once again, AVC is proud to be total thermal and mechanical solutions providers. It's important for us to provide the entire solution because each of these components work with each other, and therefore having that whole view gives us the opportunity to design something that works perfectly. That's why we continue to expand our presence, not only in the liquid cooling modules, but also in the tray manifolds, the rack manifolds, and in the rack and chassis. We're still very excited about our opportunity here as we continue to deliver more and more solutions and components for our customers.
Yeah. You also mentioned about the chassis. How about the growth magnitude for the chassis and also the rack, growth this year? Do we expect to gain a few more clients for the rack, offering?
Yeah. I think as Bill showed and displayed earlier, we're seeing very, very good growth in our chassis and in our mechanical business, and we're once again continuing to expand our presence there. There are very good other competitors in this space, and we're excited to work with our customers and work with the supply chain to deliver more. In terms of rack, our current constraint with rack, although we are doing more, the current constraint with rack is it's a two-meter product. It takes up a lot of space. We are aggressively expanding, we have to prioritize the components that will be most profitable and have the largest revenue generation for us.
While racks are a good business opportunity and we are expanding and working on it, if you look at it from per square meter of factory space, production value, it's understandable that we have prioritized other components first.
Okay. Maybe for the other part of liquid cooling solution would be CDU. What's our latest progress for our CDU development?
Yeah. For the CDUs, I think before CDUs became standalone racks, they used to be part of the AI server itself. It would sit either at the top or at the bottom.
From that point, we've already delivered solutions, customized solutions for select hyperscalers. This remains to be a part of our business that we're working on. There are a lot of very strong companies in this space as this is more a system integration solution. It's a rack-level solution. We are making good progress on it, we hope that this will also become a good revenue stream for us going, not immediately, but in a few years' time.
Okay. Maybe also, I think, this probably does not been mentioned, at the earlier section, but, I just also want to follow up here is regarding our smartphone VC business momentum. What's our expectation for the revenue contribution from these, specific VC projects for smartphone this year? Do we see potential upgrade for this specific project?
While I'm not able to comment on specific customers, we are seeing that this product type is expanding to more products in this year. It is a product that we continue to expand in our Wuhan factory, and we are, you know, excited to continue to serve our customers in this way. Its growth and its revenue contribution will still not be that of our AI servers, it will be a meaningful contribution to our 2026 year.
You mean, we are going to see a few more models also adopt this solution? Is that?
Yeah, models, products, product archetypes, you know, phones, tablets is also something that we're looking at right now.
Mm-hmm. I see. Perfect. Then maybe lastly, maybe a bit more update on capacity plan and also our CapEx budget this year.
CapEx budget.
Okay. Our CapEx budget this year is TWD 15 billion. Next year will be TWD 17 billion. Yeah.
Our capacity expansion, as we've shared with our investors last time, we are looking to increase our liquid cooling module capacity from 200,000 units a month to 1 million units a month.
We are also expanding our other components. Not, you know, we're not expecting to 5x all of our capacity. We are 5x-ing our liquid cooling module capacity, but we're also expanding in our chassis, in our manifolds and in other components as well, in Vietnam and in China.
Perfect. Maybe we pause it here for a few seconds to see if any questions from the line. Okay. Maybe Matthew would like to share with us some closing remark.
Oh.
Oh, sorry, very, there is a question just come in. It's from Kay. If you do five times your capacity, will your revenue for cold plates increase by five times or less than that? Just to help his modeling.
Fair enough. I'd love to give you a direct answer to help with your modeling, Curly. On the aggregate sense, it's hard to make a comment because some of the liquid cooling modules that we make going forwards will be more expensive. Some of them will be less expensive compared to the GB300s that we're doing right now. There's no sure way to answer it, and it's hard for us to know for now, given just the sheer number of different architectures that are coming out in the next generation.
Perfect. Before we end the session, maybe Matthew will be able to give us some closing remark.
Yeah. I think once again, thank you, Shannon and Morgan Stanley, as well as thanks to everyone who joined online today. I think when we look forwards in the next few years of thermal, our growth will be closely tied to AI and AI deployment. What's really exciting for us is that our hyperscaler customers, as we see AI become more and more useful, we also see the growing types of architecture to deliver this AI value to the end customer and the end users. Now, last year, the focus was really on just the chips. This year, we're starting to see NVIDIA as well as ASIC customers describe their AI architecture and infrastructure as more of an ecosystem.
I think the clearest proof of this is that we've started to see that divergence between training and inference infrastructures, as well as for agents which now run predominantly on CPUs. For us, this presents us with that additional opportunity to deliver the best thermal solutions to different architectures. I think what I'd like to just mention here is from this point onwards, thermal becomes a problem not just on the chip level, but more extensively on how do you solve heat on a system on an architecture level. This continues to be what AVC excels in, and we're really excited to see what we can deliver with our customers in the next few years.
Great. It's very exciting and very promising, and we expect AVC to deliver strong performance for the coming years. Thank you very much all for joining us today, and thank you Matthew and Bill for your time and also your insight with us. Hopefully we will get you guys up again next time. We all see you next time. Thank you.
All right. Thanks, everyone. Thank you.
Thank you, everyone.