Good afternoon, everyone. Welcome to Accton's 2024 Annual Conference Call. My name is Derrick Yang . I'm the Tech Power Analyst at Morgan Stanley. So today it's our honor to have Accton's senior management with us to discuss the 3Q results, the 2025 industry outlook, and some other industry dynamics. And with us today, we have the CEO, Jun Shi, CFO, Fanny Chen, and Spokesman, Edward Lin. So we have a few, let me pass it to them for the opening remarks.
[Foreign language] . Good afternoon, ladies and gentlemen. Welcome to Accton 2024 Investor Conference. This is Edward, Accton's Spokesman. Today we are hosting our conference with the support from Morgan Stanley. For all global investor consideration, today's presentation discussion will be proceeded in English and also will be recorded. Thank you for your cooperation. Thanks. For today's conference, Accton Management Team, we have our CEO, Jun Shi, our CFO, Fanny, and myself. The agenda is as follows. First, our CFO will go through our financial results with more background explanations. Secondary, our CEO will present our corporate highlights and future outlook for you to better understand our strategies. For the final part, we will proceed to Q&A to go through your questions and suggestions. Thanks. As we are - oh, sorry.
Before our presentation, we'd like to remind everybody, today's discussion may contain the forward-looking statements that are subject to significant risks and uncertainties, and this may cause actual results to differ from those contained in such statements. Please ensure you have fully acknowledged the safe harbor notice on this page. Thanks. As we proceed, our conference online, we'll put the photos of management team here for your reference. Thank you. Next. First, I want to turn the call to our CFO, Fanny, please.
Thank you, Ed. Good afternoon, everyone. Let's start with the financial highlights for the third quarter 2024. The third quarter revenue was NT 28.90 billion. As compared with the Q2, the growth rate is 15%. Three-quarters revenue was NT 71.4 billion, and 15% growth rate year-over-year. The third quarter gross margin was 19.9%. It's a 1.3 percentage point decrease compared with Q2. The major reason is the product mix. The operating expense was NT 2.1 billion. It's near the second quarter. Operating margin in the third quarter was 12.3%, a 0.2 percentage point decrease compared to the second quarter. The third quarter net income was NT 2.65 billion. EPS was NT 4.75, and the three-quarters EPS was NT 13.39. Next page. About the balance sheet. End of Q3, our cash balance is NT 12.8 billion.
Inventory was TWD 17.8 billion, an 8% increase compared with Q2, mainly due to the increase of the half -finished goods. And Q3's inventory includes TWD 8.3 billion finished goods and TWD 7.7 billion raw materials. Let's move to the key financial ratios. Q3 net debt ratio is 57.98%. The cash turnover days decreased from 8.93 days to 42.8 days, the major due to the account receivable turnover days reduced and the inventory turnover days reduced. Next page. This is the revenue by segment. Our major business comes from the switch. It's 58% of total revenue for the three quarters of this year. The network application product revenue is 34% of total revenue for the three quarters of this year, increased 74% year-over-year. The Metro Access switch revenue is 5% of total revenue for the three quarters of this year. Next page. The revenue by region.
Our major revenue contribution from America is 72% of total revenue for the three quarters of this year. Revenue from Asia and the Europe area increased. The year-over-year increase rate for Asia is 28%, and Europe is 1%. I will now turn to send over to Ed.
Thank you. Thank you, Fanny. Now I would like to turn the call to our CEO, Jun. Please go ahead.
All right. Thank you, everyone. I try to focus on the conversation around three topics. Firstly, I try to give a recap around 2024 and give a little bit of an outlook view about the end of the year. Then focus on our investments for the future for simple growth. Last but not the least, definitely I will touch the number, give the outlook around the 2025 business. So before I talk about anything about numbers, I try to share a view about what we have done in the past year. There are six major focus areas at investment that are definitely beyond just the product itself. As a company, we have been restructuring and reshaping quite a much of the company strategy really towards a technology and a service company.
So, secondly, since I joined the company about a year and a half ago, last year and this year as well, we have been structuring quite a much of the organization and formalized the business unit to get ready to scale for further growth. And actually, it turned out very well this year as evidence as proof. It's turned out to be a very growth engine for us to scale our business based on this operating infrastructure. The next one is really definitely everything about business growth, right? Since the years before, we went through the post-COVID and then market coming down to a new area for the growth, there was some of the challenge for sure, as I mentioned last year. I will recap some of the discussion I had last year exactly this time. So we have been focused on the four major areas.
Number one, stabilize some of the business and then continue to drive our field farming of the existing business to drive further growth. And meanwhile, hunting for the new logo and the new business development, new partnership, and a new service model. And to drive the organization and the expression around the growth, definitely how to drive the execution will be the key, right? A strategy without execution is nothing. And from the execution perspective, we really focus on the efficiency from operation. We drive a lot of transparency, transparency in terms of the communication, the management, and operation. And set up the clear ownership and definitely define the accountability, right?
Eventually, to enable, for example, the business unit, all the tasks, all the task force, whatever the owner, right, really have an accountable delivery as a key, turn out to be a supporting force to turn the reality into the reality. Meanwhile, as you, if you have been following Accton for a long time, in general, in the past, we don't do much of the marketing. And this year, we invested quite a much. We even hired a global head of the marketing, and we turned into more of the presence in the market, not simply because of branding development by itself. It's really about repositioning to fit into the new market demand and the new market opportunities, and including our new service model and operation model change for the new revenue generations. Okay?
And if you go to our website last year, including this year, you will see a lot of messaging, much more than I see in a way for the messaging, positioning, and it could be more relevant. Definitely, it will be helpful for the lead generation. The last but not the least one is all back to people, right? So my job is, to be honest, we're only looking to the few areas. The first one definitely will be measured by the number. The second one is really about the culture of the company. The last but not the least one is about the people. And one of the missions for me is really try to grow the next generation, younger generation of the leadership, make them accountable, let them drive the decision, take the ownership.
That is a different way or in a way to support the business growth from an organization perspective. So with those efforts in place, this is a picture as I explain and walk through to many of you, to publicly, to our customers. We call it a four-pillar strategy, right? Quickly recap. If you look into the right bottom, that was the ODM business. That's our bread and butter, our core business in the past 35, 36 years. And if you move up to the top right, that is the technology investment, not only through the technology partnership. We have a small venture. We do the technology invest quite a much as well. As you may have paid attention, even the news went out yesterday. We did some technology investment into some technology components. We tried to acquire quite a much of the technology to enable our future growth, right?
And then move all the way to the laptop. That's really about the JDM of the business. We leverage a lot of cutting-edge technology to apply our leadership from technology business perspective, work very closely with hyperscaler in particular. And we have a different bar for the quality design criteria because the magnitude about the deployment is at a very different level. And then we apply those acquired know-how and advantage onto our white box because open networking is always our belief to drive the innovation with both hardware and software. So if you look into this like a flywheel as an innovation cycle for us, not only just for technology, but from go-to-market business as well. By leveraging our white box for open networking, we can help dramatically reduce our customers' time to market for the technology adoption.
Meanwhile, we still offer very flexible ODM and JDM models to help our customers for the longer-term investment as well. It's quite flexible in a way. This now paid out very well, right? This is I feel very pleased through the past one year execution. This one, we locked quite a decent new logos, especially to drive the business growth, as I will share some more details in the later slides. We definitely need to hunt some new business. We call it a big whale, right? It's quite successful. I see we acquired quite a few top brands in the market. That is where hold up the growth for our future. Now, this is a picture actually I showed last year. It's just a little bit different format.
On the left top or top left, if you really try to show the picture, if you recall for whoever attended the session last year, that was a projection about the network equipment providers. About the trajectory last year, we projected it could follow different paths, right? Some of these network equipment providers actually continue to go through the struggle. Some really start to see the bottom and start to pull it back, and some pull back early, some pull back late, and some of them are still struggling. So it's quite a mix. This comes down to the product mix as Fanny mentioned from the financial report perspective. At the left bottom, it's really about a projection. Last year, we made about AI/ML of the trend.
From the market perspective, we're looking to full customer mix from open networking, hyperscaler, Network-as-a-Service, and network equipment providers. We are quite, I would say, happy in a way, not necessarily talk more about our understanding of the market and the confidence level of our execution. We set up quite a few strategies last year. It turned out it's compliant with what we have projected very well. That's in line to demonstrate our performance from the execution level in line with the market. If you look into the market perspective, there's a quick takeaway here is anywhere from our customer base, if anyone catches the AI wave, right? Anywhere, whoever catches the AI wave, we see the growth from them.
Most of the people, I cannot see 100%, but most of those people, if they meet the AI wave, especially from the telco area or some enterprise area, SMB area, still the struggle is still over there, but definitely, there are some good signs from some of the sectors to start to show the pullback a little bit. There's a little bit of confidence, especially from Q3 across some of the market segments, but from our perspective, I think because we made several of the strategic decisions last year from the investment, turning out our business is keep growing. However, if you do recall in January, February in Q1, because the transition from the pullback of the AI/ML from that adoption cycle versus the drop from the enterprise or telco areas, Q1, we indeed saw a dip.
Then we pull back all the way from market, then we keep the growth all the way until right now. Not only about the growth, as we expand our market segments at acquisition as well. Besides, right, whatever the new segment, I call it emerging AI. This emerging AI is not only about everyone talking about everything in the spotlight about a large language model or ChatGPT, this training. Actually, there are a lot of new use cases. The flying and the radar, there are quite a many of the excited movement over there start to adopt the AI/ML, not only the training, but the influence into the market segment, into, for example, the retail or manufacturing or the other locations as well.
And from other than this new area, the Network-as-a-Service data segment actually is expanding because it becomes more enriched in a way beyond networking. Like a Security -as -a -Service, like a Network-as-a-Service, or even like a Cloud-based Campus -as -a -Service, there are quite a few of the areas. We acquired quite a few leading logos from there. So that's why you see the color code. I just try to demonstrate every single sector so we see the growth. And in addition to that, as I mentioned, the brand development definitely is something I put some investment over there to demonstrate our leadership, right? The picture here is we just took recently. We had a good presence at OCP in the Bay Area in the United States.
And if you're looking to the far left, that is our 51.2T switch, Tomahawk 5-based switch. And we have our own software, open networking software, SONiC on it. If you pay attention to the picture on the left side, there's a rack, right? On that rack, actually, we put quite a few of the new technology demonstrations. Actually, there are three racks over there, and the people block it. There are quite a few new technologies other than just the traditional classic high-speed switch. We do demonstrate, for example, the new technology from CXL for memory pooling and for GPU cluster. And we have our own GPU server over there as well. I will show you in the next piece of the picture on the next few pages. And we showed our liquid cooling rack-based solution. But overall, we focused on a premium design, high quality.
Our yield is quite high, and it's quite recognized by many of the people. Our 51.2T, right, from a 400G and 800G adoption perspective, definitely showed quite a much of the deployment. Still, the market keeps growing, and we definitely want to work hard to create more market share over there. Again, right, the design, the quality, definitely will be our core competence over there. Continue to help our customers to drive the reduce the time to market with more value. You can think beyond just networking the skill. That's why from our position perspective, we try to introduce ourselves as a game changer really from both networking and computing infrastructure together. As many last year, I met many of the customers and the investors and the partners. Everyone definitely is asking, has been asking, right, which direction we are heading for.
We should try to give you a very high-level, 10,000 ft level of a picture about our investment. On the left side, right, that is maybe a very high-level picture about the AI cluster. If you look into the top, there's a switch. That's our bread and butter. We have been doing that one in the past 30, 35 years. And we have our open networking software, SONiC, as well. And then we formalized our small company. It's 100% owned by ourselves. We call it GoldiLink. We developed our own 800G optical transceiver, right? We don't intend to try to turn ourselves to be an optical transceiver company. That is not a point.
But acquiring the capability, especially focused on the high-speed, like 800G or even above, the optical transceiver will help enable us to grab the market quickly with our switch because we developed our switch with high quality, much faster compared to many of our competitors. So that six to nine months lead time for us is very critical. We cannot simply wait for people to come. Let's not wait for the market to be ready to get optics. We definitely want to enable a total solution over there. The next one is about accelerator, right? Accelerator definitely has been one of the major contributions to our overall business. But this will be one of the areas very critical from the AI cluster perspective. Then I think early this year, we start to launch our own GPU.
As many of you know, we have been building the UBB or OAM for our ODM or JDM customers, but given the trend from the market, it has a very clear sign. The line between the computer and switching right now is blurry, right? We don't intend to try to become just another server company. There definitely is a good incubation for us to drive the knowledge and the technology, and definitely, we need to make ourselves ready into that market. There are quite a much of the learning, for sure, right? Because simply from the server versus the switching, there's still quite a much of the difference from the manufacturing, the production, and even the good market perspective, and given a lot of new involvement, right, given the GPU, definitely.
But beyond that one, looking to a lot of good momentum about the new technology from, for example, the CXL or PCIe next generation. Our investment into several of the startup companies as a partner to drive the server business. Definitely, this one is one of the directions we are investing into. The last, not the least one, right? To formalize the cluster, definitely we need to think about interconnect, right? The interconnect from optical interconnectivity or copper or hybrid. Meanwhile, it's really about a back-end network from AI cluster perspective about interconnect. That is one of the areas for us to look into the investment as total solution as well. To put everything together, the market trend and the signal is quite clear as we demonstrate at OCP. We definitely look into a rack-based solution, right? Given the power, the deployability, right?
Given all the liquid cooling. From the deployability perspective, the whole market is moving to the rack-based solution. We have been investing in this one from level 10, level 11 to ease the adoption and deployment for the AI/ML cluster. So the takeaway here is definitely we are investing and we will continue to do so from an end-to-end perspective rather than just focusing on one single component of the whole solution to drive the adoption and help our customers to adopt the new technology quickly to generate a revenue return. And going beyond, right? Simply just to get a solution is still not just to pile up a bunch of the hardware components over there. We need to stitch everything together from the underlying networking and the computing infrastructure. And the top is about a network operating system, right? Because we drive the open networking.
And we partner a lot with third-party our partners about network operating systems together. And then above, definitely we want to look into as a solution. We want to make it easy as an out-of-the-box deployable solution to our customer. Again, customer can run the application. That is above. We don't look into the application right now, but just to try to show you the focus from the solution perspective really about the networking infrastructure, computing infrastructure, the network operating system, and the solution play. Not only try to build everything by ourselves because we are a very strong believer about open networking, deep aggregation. So we will own whatever is a core competence to ourselves, and then we partner with our technology partners. All right. Now let's take a moment to look into the numbers. This is a picture I shared exactly. I didn't touch anything, right?
I didn't change anything. This is a picture I presented last year. We project from 2023 to 2024 as a double-digit growth. We still hold that outlook as a projection. And then around 2024, as I showed towards 2025, it follows what's happening exactly follows what we projected. That's why we are quite happy in a way about our predictability forecast and our understanding of the market. Now we're moving to 2025. It's really as our projection going to that rank. I didn't change anything. By the way, don't waste your time trying to measure how much of the picture I put over there. I just randomly draw a picture to drop in the middle. Okay? The guidance and direction is definitely we continue to hold a double-digit growth for next year. Okay?
And the rebound, the new business rolled in, the acquisition of new logos has been happening exactly like we have been planning. And at the beginning of the year or end of last year when I came over here, back then we did a 2024 planning about this new customer as a target. I cannot see that we 100% close them, but it's very decent in a way. It achieves the most. I cannot see all of our expectations. Okay? And then that's why from a 2025 perspective, the growth, we continue to hold our cautious, but definitely optimistic view about the growth. However, the market definitely has a dynamic, right? The dynamic majority is being driven by the market from the expectation. I believe almost everyone worldwide has been asking the same question: how long does the AI hype cycle will continue to last, right?
Whether and how the 2025 will show the trajectory about the growth. We definitely see the AI/ML continue will be the driving force for 2025. However, around the second year, next year, definitely there will be mixed signals over there given quite a much of the uncertainty, given the global geopolitical situation, given especially like if you remember or see Stanley presented quite a much of the larger of our market customers' contribution comes from North America and Asia, right? The geopolitical situation for us to really force us or drive us in a way to stay agile, to be flexible, and to be adaptive. And this is on the left hand. This is our new manufacturing location in Zhubei in Taiwan. It's pretty much up and running. We'll have an opening ceremony very soon.
On the right hand, just to try to show you a picture from our global expansion and the strategy to basically position ourselves as a global footprint. And we have right now we have a 500%, well, I see still 100% owned our own manufacturing facility. On the right bottom to show, actually, we are working on the number seven, but let's be six-to-seven of the joint venture approach for the manufacturing of the servers. Not only as a manufacturing capability, but help our partners to really deal with the geopolitical situations in every single continent because if you're really looking to carefully about every single country's icon over there, if you open the map, right, every single continent as an entry country or the gateway over there, we have our presence. But net-net, right? We try to stay agile, adaptive. It will not be easy.
2025, definitely there will be a lot of exciting moments upcoming. The AI/ML is still there, but the challenge definitely, as everyone can see, we will stay in a way for our commitment for the investment, drive the growth, and definitely give the contribution back to our shareholders. Okay? That is all I have.
Thank you, Jun. This is the end of our presentation. Now we can move on the Q&A section. Derrick, please.
Sure. Thanks, Jun and Edward. Right now, we will start our Q&A session. If you would like to ask a question, please use the raise hand button, and we will read out your name. Please remember to unmute yourself to receive.
Our first question comes from Paul from Cathay. You can unmute yourself and ask your question. Thank you.
Okay. Hello, hello. I'm Paul from Cathay, and there's some questions to ask. So before I ask the question, I would like to thank you for taking my question. And the presentation is crystal clear for us to understand the company's operation status and the future strategies. So we are very impressed by the growth of AI accelerators and achieving such operating results in a complex and competitive market. And it's not easy, as we know. Also, congratulations for Accton to achieving this new record -high performance this quarter under the leadership of the outstanding management team. So I still have some questions for the management team, and I am eager to learn more about the AI accelerators and their influence on our business. So first, how does Accton evaluate the development of the AI accelerator in the new generation data center infrastructure?
The second question is about what potential growth does this present for Accton's future operations? These are my questions. Thank you.
Regarding the first question, right, how Accton evaluates the AI, I would say, evolutions for the AI infrastructure, so the DC infrastructure. The data center in general, I see, has a classic traditional, I would say, workload of the data center. The AI DC or AI data center, that's where the AI cluster, there's a different momentum over there. From what we have been experiencing and contributing over there, the AI data center or AI cluster contribution, we indeed, I would say, take advantage and benefit from quite a much of the investment from there from the large customers. Definitely for training and some of the instruments over there, we participated quite heavily into many of these customers. The customers, not only from the hyperscaler, Fortune 500, but there are quite a many of the newcomers who are providing the GPU -as -a -Service.
It's quite different from the, I would say, the classic data center for the IT workload. Okay? I guess that one may not be 100% accurate, but that's the definition.
It's overgrowth.
Yeah. So by looking to from their perspective, if you see the picture I draw on my slide, so end-to-end AI cluster from the server all the way to the connectivity, from the cable to the switch, and including the software together. Actually, we had a very early customer adoption. We moved very fast, grabbing some opportunity much early in a way. Some of our earliest AI/ML deployments went back all the way to last year, right? And again, we start to experience good adoption, but definitely it's still in the middle. The market is still growing. The investment is still quite much. And actually, most of the challenges people are facing, other than just how to build up the infrastructure, as everyone has been talking, I will not talk too much about the power, the cooling, all the others.
To be honest, it's really about a shortage of the GPU, right? But we definitely have been working closely with our partner and our customers to continue to drive our innovation into the AI-based data center infrastructure. But we are not ignoring the traditional data center over there. It's still growing. It's just the amount of the budget spending between our customers to where you put them all because the GPU cost is quite expensive. Now, this is where it has come down to your second question about a future growth level. I would say it's really about the lifespan, about the outlook, how far ahead looking to the future. 2025 definitely is a year we see a very strong signal and a clear forecast view from our sales and from our customer and partners. 2025 will still continue to be a strong year for AI/ML.
However, trending towards the second next year, that's come down to a lot of expectation or different opinion about how this return will look like, given people's confidence level about the continuous investment into this heavy-loaded infrastructure. Because right now, as everyone knows, right, the returns versus the investment still come down to the point as people expect it. But no one, I would say, a few people there really try to miss this investment cycle and a growing cycle. But we definitely see this quite a much of sustainable growth engine upcoming in the actually, we are still in the middle as an early stage. We do have a high hope, but given the risk from the non-technology or business perspective, given the geolocation, geopolitical situation globally, that introduces definitely quite a much of the uncertainty we have to watch very carefully. So hopefully, I addressed your questions.
Yes, yes. It's a very clear answer. Thank you. It's my question. Thank you.
Yeah. Maybe the second question I can ask is about the 800G network switch. Given that we are in the middle of this transition and the migration seems to be a bit slower than expected back in early this year. So Jun, could you share with us your view on the current status of this 800G network switch migration and what we expect that to play out in 2025? And also, a smaller question is indeed during this kind of technology transition, there could be opportunities for share among different players. So I was wondering whether you could share with us about the potential share allocation at our major customers or at other major hyperscalers.
Okay. Just give me a second. I'm making notes. Okay. So firstly, I would say when you look into 800G, when people have expectations, always think about it, follow the pattern about 100G, 40G- 100G, 100G- 400G, 400G- 800G. Very unfortunately, this kind of methodology does not apply. That's our opinion. The reason is the 800G, I will not say 100%, but the majority are driven by the AI/ML. Because if you just think for a second, just like the previous question I tried to address, there's a traditional classic data center primary driven one by the hyperscaler, but it's still about a non-AI/ML workload. So they follow the trajectory about generation by generation, depends on the new technology, new speed coming in on top of the existing infrastructure. That is a migration cycle.
However, when we move to 800G of the switch, actually, to be honest, right, because 800G GPU, like a Blackwell, is just coming around the corner all last year, it's really about a 400G. Even for 800G of the switch, people use the DAC cable, like about a 400G adoption because of GPU size. And if you're thinking for a second, right, because almost the majority of the deployments for AI/ML for 800G switch are for AI/ML, not for classic IT workload or the regular data center. So that migration cycle, if you just apply to whatever the trajectory in the past, does not work out very well. So that's why if people say, "Hey, I expect the 800G trajectory should follow the patterns of 100G- 400G," unfortunately, it's not for another reason.
Because if you look into every single IP, every single enterprise company, your budget every single year pretty much is fixed, more or less, and AI/ML investment is quite expensive, to be honest, right? If you take a majority of the share of the investment into AI/ML, the remaining of the budget from the customer perspective will continue to drive the regular data center investment, which really drives the previous trajectory from 10G, 40G, 100G. It's a different pattern, so these two patterns right now, it's not coming to a thin trajectory to say, "Hey, how much of the enterprise adoption infrastructure really under the pressure to move from 100G to 400G and 400G- 800G?" No, that's not an investment focus right now, and the majority of the people really do not have the need or demand much to really look into 800G or 51.2T kind of switch.
It's still too much, right? That's it. Hopefully, I addressed the first question. The second one is definitely market is shifting, and that's a trigger, the opportunity and the challenge for every one of us. So it's really about the strategic decision and attitude in a way, say, "Hey, we take the chance to win the new market and grab the new share from it." Because eventually, even the AI/ML cluster, it's still from technology perspective, they still have quite a much of the leverage can be applied back to the data center, traditional or classic data center as well. So that's definitely an opportunity for us to get more shares from the market. And from our perspective, as I mentioned, from our customer acquisition, the big whale hunting, I will not share too much of the details.
But overall, from our acquisition from hyperscaler, from Fortune 500, from this new player, GPU-as-a-Service, it all has a very different growth. And we do believe our 51.2T effectively for the 800G is in the leading position based on the customer feedback and based on even customer turnover from our competitors back to us. So I think, yeah, this is an opportunity to us.
Great. Thank you, Jun.
Our next question comes from Edward Yang. Please unmute yourself.
Can you hear me?
Yes, yes.
Yeah, perfect. Yeah, appreciate it. Thank you for taking the time to take my questions. I have two, but I'll start with the first one. Just kind of looking at your presentation, and thank you for that. It kind of feels as though there's a lot of growth opportunity, right? There's a lot of revenue growth opportunity. And my question is more kind of on the cost side. Maybe if you can kind of talk a little bit of how your strategy will affect kind of the cost. And then the second thing that I kind of wanted to also talk about is also on the cost side. It kind of feels as though there's a lot of cost inflation going on right now, right? Whether it's if you move into AI, naturally, there's shortages there.
So the cost is probably going to be relatively high for some of these components. At the same time, geopolitical factors, that probably also plays factors quite a bit into the cost equation as well. And then lastly, utilities, we know that that has gone up as well. So I'm just curious as to how you can kind of take this into consideration with regards to your strategy, one, and then how would you tackle some of these cost inflation challenges?
All right. Thank you. I tried to address your question more from a product point of view rather than from a component point of view. Because about a new product adoption or new solution adoption at a very early stage, it always follows the cost curve when you enter into a new market, develop a new, bring a new product to the market. The cost curve at initial point is always higher. It's always in that way. However, whenever you have a new technology trying to address the same issue existing before, the new technology always can dramatically reduce the cost per bit or cost per whatever the measurement over there. But we are now, from AI/ML in particular, at the early stage right now at this moment, right?
That's why the cost factor, if I look into this one, whether it's under AI/ML or not under AI/ML, to be honest, because I did a lot of product before, but still in my different role, I don't think it's dramatically different from when you open a new market, the cost curve. We definitely have a different challenge to manage, right? At the component level, you mentioned quite a few factors from a geopolitical situation. Yes, if you put it into that factor, that is a big deal, right? Because globalization in the past decades, actually, everyone has been benefiting from it. That one is a cost factor we need to consider. But if you just consider it as a cost factor anyhow, due to whatever uncertainty introduced to drive the component level or cost level, it's just in a way for us, just maybe we take it for granted.
It's our life, right? We have to manage it through, consider all the factors. It's just maybe a different practice. I will call it maybe the old wine, oh, sorry, old wine in a new bottle or the other way. But I don't want to enter at least the amount of challenge about the cost. The cost down, the cost optimization, the design for cost definitely has been the core competency we have been introducing into our premium design. But from an overall practice perspective, it's just part of our challenge daily life we have to manage through anyhow. The AI/ML just introduced more of the factor from limited supply because the GPU over there, and meanwhile, more new material, high grid level of the PCB, and then power, the cooling, it's new, right?
So it's a new factor in today's context compared to 10 years ago, for example, when we introduced another new technology. It's just a different dimension of the issues we have to address. I think it's just an everyday challenge. Hopefully, I did share my thoughts to you, but I don't know whether I addressed your question.
No, it's very, very, very, very good. I guess just one follow-up for me, and I'll get back into the queue, is also kind of just if you can talk around kind of longer-term kind of growth framework. I know you already kind of gave your thoughts on 2025, right? But apparently, one of your networking peers also just kind of just had just concluded the call. And it kind of feels as though the visibility is actually quite solid. I mean, it kind of feels as though your peer is kind of looking even into more of a team type of growth into 2026.
So, I just kind of wanted to kind of talk around that, maybe get your view as to how, whether or maybe talk around the pushes and pulls that might allow you to maybe even reach or maybe even outperform that type of kind of growth beyond maybe 2025, maybe even doing teens again in 2026, 2027. Just kind of trying to get an idea of your longer-term kind of growth framework.
So a new factor in the AI/ML compared to the past 10 years or 20 years, we have been in the market so long. The new factor AI/ML introduced compared to the previous generation actually really thanks to the GPU. I think NVIDIA has done a good job. We really keep the pace about every year trying to deliver a generation of the silicon that pushes up the whole cycle from the generation of the new technology comes to the market of going much faster than before. I mean, the iteration of the new technology.
Even this iteration of the new technology coming in, one of the different factors or the result or the pattern definitely is different than from before, just building a switch because the life cycle from the switch, from the adoption, easily will take two years, three years, stabilize, then kind of go into life cycle, right? But even the AI/ML, the life cycle is much shorter. The iteration of the AI engine, the inference, the training, actually is running much faster. So that's why year-by-year, let's say looking to 2024, look back in 2023, 2024, and then look into 2025, we definitely will foresee another generation of the split or the new technology coming in because the 800G GPU is coming, right? And then the new deployment. So then it falls back into a normal pattern.
So basically, whenever you go through a product transition, the previous generation of the product starts to sink down, right? Going down, the new generation starts to pick it up. And the new generation almost does double, triple, whatever the x amount of the capacity grows over there. So that means from the material, from the supply perspective, it will drive a different curve. So that means the transition period introduced, and that transition period, the impact will definitely be bigger than the previous technology because the amount of capacity, bandwidth, connectivity we are talking about is different magnitude than before. So that means one month or one quarter of this transition, the impact can be amplified compared to before because before may take about a half year, six to nine months to do the transition.
Very, very clear. Thank you. Thank you so much, Jun. Thank you, Fanny. Thank you, Edward.
Thank you.
Our next question comes from Carrie from Citi. Please unmute yourself. Thank you.
Hi. Sorry. Is this for me, Alex? That's Alex who's on call. Can you hear me?
Yes. Yeah, yeah. Alex, I can hear you.
Oh, sorry. Okay. Thanks a lot for taking my question. Hi, Fanny and Shi. Hi, Derrick. Maybe one question I would have is if you could comment a bit more on the customer diversification in general, maybe start with as far as you can talk about in the AI accelerator space, if you can put a bit any numbers on how many customers you could gain or maybe how well the new facility in Zhubei is kind of booked. And also on the switch side, how you see your progress in growing business with, say, the less prominent hyperscale customers or potential to add new. That would be one of the major questions I have.
All right. Thank you. Customer diversification, I mentioned a little bit in my slides. As I mentioned, we kind of grow from four major customer mix, right? Open networking, hyperscaler, networking providers, and the Network-as-a-Service. Right now, we definitely grow the area about the AI as a when I mentioned about AI, it's not only about, as I mentioned, right? It's not only about the switch itself, right? Definitely, switch is one of the cores of our business. But at appliance level, you saw the growth from our appliance that we're really from accelerator. One of the patterns or the thinking we need to, at least from our investment operation perspective, pay attention is what is different from before is in the past, we can sell a switch. We can sell a transceiver. We sell a cable or we sell a software.
Then people put them together. And then someone will put together SI or VAR or whoever. But the AI/ML buying pattern is people want to buy end-to-end. It's one because the time to them is very important because the GPU is very important. No one wants to leave the GPU sitting idle over there because every single second about the depreciation of the GPU is very expensive. So that's why people always want to try to get an end-to-end solution quickly. That turns out to be a new buying pattern if you want to buy everything, right? From the server switch and the connectivity and the software and quickly roll it out and then start running the workload. That's why from the accelerator, you mentioned about the accelerator, the switch side, the progress, there are a couple. There are a couple, right?
But definitely, our customer is kind of quite a balanced mix. Some of the for the switch here, we serve the hyperscalers. They pretty much are the driving force for the solution. But for enterprise or GPU-as-a-Service, we definitely offer them a solution level play. That is, I would say, a shift or you mentioned about a new space as a buying and offering as different from before. From the cost per day level, to be honest, I don't have anything in my mind I can directly share to you, but there's tons of study and research over there. We just ride on top of that curve and wait. And from the growing business perspective, definitely, we have been growing across all the dimensions I shared, right? From hyperscaler, from even including the ODM business, and there's a newcomer, and there's a new emerging business.
Definitely, it's a positive, right? That is hopefully I share the view as you are looking for.
Yeah. Thanks a lot. Just maybe specifically on the new facility in Taiwan, can you say if you have basically sold out the capacity already to customers or how that looks like?
I really cannot share that number, but I can tell you it's a very highly occupied.
Okay. Thanks a lot, and just one follow-up also on the switch side, if I may, so I think an important factor for you to kind of grow market share in the switch space and hyperscale customers is your software capability, so how do you see kind of the momentum there or the traction for basically gaining potentially new customers or new business on the switch side?
So I just want to make sure it's clear, right? For anything we drive to, there are many strategies, but there are two clear strategies. Firstly, we do open networking, okay? And then we drive disaggregation. That is our belief. This is the value we bring in. And we support our customer, the networking equipment provider through the ODM, JDM. We support the hyperscalers through the JDM as well. And then we provide our own open networking white box through the open networking software, right? Everything we drive is open networking because we have many different customer forms. We definitely don't want to compete with them. We partner with them. The partnership is very key. I want to make sure that way it's clear. Secondly, from the switch perspective, from the market share, we don't have exactly the same number coming back. It's still in the middle, right?
Hopefully, year-end, we'll get a more clear picture from the market. Actually, I hope our analysts can even have a better view to share. But we overall think positive about our competence level from our switch into the market. And as I just demonstrated and showed that our OCP presence over there, right? Definitely, our competence is over there. We do believe. It's not a belief as we just think about ourselves. It's a lot of feedback, positive from our customers with evidence from our large customers' deployment, adoption. It has been in production and for large cluster deployments. And quite many still in the early stage about a POC and a pilot. So I would see that, yeah, it's good to set up the foundation and the growth for next year.
Excellent. Thanks very much for the call. If I may, the other question would be before I move back into the queue just quickly on the margins trend and outlook, if you could comment a bit how the switch upgrade trend and the new product introduction from AI cards, basically how that impacts the margin trends, maybe initial dilution versus kind of uplift potential further out and timing, and also these new strategic pushes that you mentioned and how far that has kind of a material impact on the margins. Yeah. Thanks so much.
Thank you. So as I just briefly mentioned, right, the buying pattern, deploying pattern, and even the life cycle for AI/ML has been driven primarily because it's by the GPU cycles, right? Because the pace about NVIDIA to deliver their generation by generation of the GPU, the pace definitely is much faster than a traditional or classic enterprise-based switch fabric in terms of the different generations. And the following pretty much is a common pattern, right? It's from the product point of view, every single generation, when you start to turn to a shifting or switch between the generations, the cost curve, the margin curve definitely ride in a different pattern. And if GPU is really happening, to be honest, this curve is really the past year or two, right?
A lot of patterns right now are still being formalized in a way, but whatever we can see is the margin from the amortization from one generation of the cost curve is running much faster compared to the other non-AI/ML-based ones because the pace for the iteration is much faster. But on the other hand, it brings a downside because you haven't got enough chance to try to monetize from the volume perspective, and then the next generation starts coming up. So it turned out to be a pressure not only from the product level, from our supply chain level, because how to amortize the investment into every single one generation in general. In the past, you take a year, right? It's a long curve try to get a return back, recover costs. But right now, the cycle is much compressed. That introduced a pattern about a margin curve pattern.
It's quite different, okay? But the essence of the nature about the margin shift is not much different, to be honest. It's just kind of condensing everything, fast switch over. Then you have to rethink about the cost curve. And now, given the margin erosion perspective, especially from the, I would say, from the deployability and even from the supply chain perspective, the curve, it's just kind of much more condensed in a short frame. But if you pull them out, zooming them out, it's still the same learning. Just everything we got to do faster, think faster, and react faster. But the design philosophy perspective, how to catch up the wave, keep up the pace, that is one of our advantages because how to keep up the pace with multiple iterations so fast, that is the competence actually we continue to hold.
And then from a margin perspective, the falling to our expectation, that's not much of a surprise. But definitely, we need to work harder in a way, try to monetize and recover the cost, right? But that's life .
Okay. Got it. Thank you, but basically, for next year, we should expect margins rather more under pressure from this kind of initial compressed curve rather than move up with higher ASPs of new products?
If I only had one, I would only have one business about AI/ML. I would say that's the answer. However, we have a very balanced product mix and customer mix. Especially from product perspective, the strategy continues the strategy that Timothy and I mentioned last year. We will not put everything into one single basket. It's not because this year everything's AI/ML, then we rush everything to AI/ML because we have quite as much of the balance over there for enterprise and still for data center of the investment. It's still growing. Don't get me wrong. It's just not grow that fast just like the AI/ML that occurred, but it's still a very healthy business.
And yes, some of the market is still, as I mentioned, like telco, for example, or as you can see our portfolio from access aggregation from the metro area. It's still going through some of the struggles on the market, but we will not give up of them. In the longer-term perspective, you always need to keep the balance about the investment because the market will not only go into one direction. There always will be just self-correction here or there. That's why from the margin, it's more compounded together, not only about one single dimension just by AI/ML, but definitely AI/ML contribution right now is much higher, have a high weight of the business contribution. But we will not only look into or present that one single factor to show our margin.
The other businesses would mitigate the margin pressure from AI/ML next year?
No. Well, that's why it's not an easy answer. It's a multi-dimension answer because remember, as I mentioned, I didn't touch too much about this year because last year when I mentioned about the new business rolled in, when we start to win a deal, start to build a system, start to help our customer to, let's say, buy enough time to ramp up the business, it always takes time, right? In general, it will take more than, let's say, 12-18 months from there. That's why the new business coming in, some will come in too early, some will come in too late. It's always in a dynamic situation. I don't think this is an easy answer about a yes or no answer.
Okay. Got it. Thank you so much. Thank you.
Did you mind our time? We will take our last question coming from Kylie Tai. Please kindly unmute yourself.
Thank you. Hi, Jun and Fanny and Edward. Thank you for taking my question. So my question, number one, will be on Trump's election. So after he was getting elected as U.S. president earlier this week, there have been talks about potential increase in tariffs or other strategic moves to increase production in the U.S. So what's our thoughts about the potential impact to the industry and our capacity planning? And what's our production capacity breakdown by region as of now and going forward?
I think this is a tough question because everyone right now has their own crystal ball to try to because it's still early, right? This is just happening, and from whatever we learned years ago, the last term, and through the past year, always the geopolitical situation from the deglobalization perspective, the training actually has been very clear. Right now, with the new term coming in from the Trump election, everyone, I believe, not only us as a company, but from the different countries, even whatever, not only from, let's say, Asia, China, but from Europe perspective, it's quite a dynamic. We cannot just plan for plan, right? Definitely, we need to make our own judgments.
As you can see from our agility and our agile adoption from our thinking, it's always try to drive the diversification into different locations because we have our global presence and have a global business for sure. Now, given not only the tariff conversation, but about a restriction about the GPU or from China or from Europe perspective, the tax one, I leave to Fanny from our funding to look into that one. But from the business strategy perspective, it's definitely we need to consider given this further accelerated deglobalization and this new business model and a new barrier between the countries. We have been thinking about this one. I believe many of the people just don't wait until the election is down to make decisions, right?
We have been planning in the past, even now, always try to stay agile in a way, try to cope with the situation more proactively, and it's multiple dimensions of the consideration so it's not only about the tax itself. Definitely from the product level, solution level, customer level, service model perspective, there are a lot of challenges, but it's always coming with a lot of new opportunities so we stay positive, but yeah, we have to stay very cautious in a way about thinking about where to invest, but we continue to invest into new technologies. That is the key driver for us to stay as a technology company, but have to be flexible enough to deal with the global situation.
Thank you. Very clear. If I may follow up on your earlier comments on 800G adoption, you mentioned that it should be mainly for AI/ML applications, not for general data centers? So could you give us more color on the adoption rate for 800G in next year and going forward? And is there early indication of 1.6T adoption maybe two years later? If you have any color of that?
So the 800G adoption, actually, because there's an AI market trend overall and AI-powered sales. For our 51.2T of the system for the 800G solution, we brought our solution to market quite early, far earlier than many of our customers. We have our first customer shipment last year with our 800G solution already. And if you're looking at the adoption this year, definitely majority being driven by, I would say, the GPU side, right? Because given the AI/ML clusters, the topology, deployment, we definitely see the curve. And this year, still, other than the hyperscaler, other than many of the people see the spotlight from this large cluster, but there are a lot actually under the radar about the smaller clusters. And for that one, quite many still in the pilot, still in the POC. So we do believe next year the 800G will continue.
And given this is a pattern not only in enterprise or the new player, but at the hyperscaler too. So 800G, definitely, I would think next year will take the lead from the ramp-up and the growth perspective. And now talk about 1.6T as a next generation. As everyone knows, if you're familiar with the technology from 51.2T, the next one is a double, right? We have been working on that one for a while already, okay? Definitely dependency about the silicon timeline. We are not building chips, right? We use chips from our partners, the technology providers, right? There's a dependency over there. But definitely, we are at the execution stage working on it. And then from the adoption level, to be honest, this is a follow. I do believe it will follow the adoption cycle about 800G and 400G. That's the dependency on the GPU, right?
So if the GPU, the iteration of the new generation keeps the pace in that one in a similar way, I think from the adoption, from the switching line perspective, will follow a similar pattern. But timing-wise, just like any new product introduction, it's just quite a much of the dependency. It's really some end-to-end control many have dependency with our technology providers. But the momentum and development and investment definitely is trending to that direction. I think it will only go higher.
Clear. Thank you.
So for the interest of time, we will conclude the Q&A session here. And Jun, would you like to make any closing remarks?
All right. I just want to thank you. Thanks for all of your attention and consideration about Accton, and always appreciate your support, so hopefully we can all enjoy this journey and experience another good year for next year, all right? Thank you for your support again.
Okay, so we will conclude the call here. Thank you, everyone, for joining. Thank you, Jun, Fanny, and Edward, for talking.