Alchip Technologies, Limited (3661) Earnings Call Transcript & Summary

[Audio Gap] We'll upload to MOPS, [ both video and audio ] content about a couple of hours after the finish of the meeting. So it takes time for uploading. So please be patient if you need it. So this is the message from our CEO, Johnny Shen.

Okay. Good afternoon, ladies and gentlemen. I'm Johnny Shen, Chairman and CEO of Alchip Technologies. Once again, thank you for joining our investor conference today. We truly appreciate the opportunity to share our Q3 results and provide an update for our business outlook going forward. For Q3, let me summarize a bit. Our third quarter revenue came in below plan, primarily is due to the planned change of our IDM customer and also capacity shortage for our cryptocurrency business. As a result, the revenue reached $223 million and net income $44.3 million and EPS is TWD 16.4. The detailed financial breakdown and analysis will be presented by CFO, Daniel, in the following section. There are a few highlights for last quarter. Number one, we made a significant progress on our -- with our #1 most important customer. The N3 design we tape-out was successfully verified. A substantial number of wafer order already placed to the major supplier. The project remains on track for high-volume production starting from Q2 next year with no change in our annual revenue forecast. In addition, the design activity for next-generation product have already begun. We expect to recognize a portion of NRE revenue this year. Number two, regarding our automotive business, our end customer is highly satisfied with the chip's performance, has already placed substantial wafer orders. The chip was deployed across all their product line, and customer has even decided to integrate 2 chips into their high-end model. Production is scheduled to begin late this quarter, and we expect this product will become one of our top revenue maker starting from next year. Meanwhile, the next generation is already underway, ensuring the revenue continuity and strong business momentum through 2028. Number three, regarding our gross margin improvement, although our revenue -- Q3 revenue declined significantly due to lack of production revenue, but our NRE business remains strong, contributing to a 6 percentage point improvement in gross margin compared to previous quarter. Despite 20% decrease in total revenue, our net income and EPS was slightly higher than last quarter, reflecting our different business mix and disciplined cost management. Looking ahead, our Q4 is a traditional peak season for NRE. We are confident of achieving even higher gross margin this quarter. Number four, regarding the ecosystem partners, starting from N2 process node and beyond. HPC capacity has increased dramatically with a single design often integrated multi-die and diverse solution. Unlike most of our competitors follow a captive model, sync for higher margin, Alchip is promoting an open ecosystem strategy, collaborating with cross range -- with broad range of partners to develop comprehensive and efficient solution. Our ecosystem network has expanded substantially in recent quarter with more partners eager to engage in co-development. We firmly believe the future ASIC industry lie on the open system, not in a closed captive system. As for the geopolitical risk management, we continuously diversify our business and design resource beyond China to mitigate the geopolitical risk. In Q3, less than 8% of total revenue are originated from China. To strengthen our global engineering capability, we have launched a very aggressive hiring plan focused on the region outside of China. Our new Japan office can accommodate up to 200 employees, while our Malaysia and Vietnam office now host 100 engineers combined. We plan to expand our Southeast Asia workforce to around 120 employees by the end of this year. Conclusion, we anticipate a further revenue reduction this quarter. However, our gross margin percentage is expected to improve again, driven by higher contribution from NRE revenue. Despite the revenue decline, our quarterly earnings are projected to reach the highest level of the year. Looking ahead, we remain highly optimistic about the AI market. From 2026 to 2029, we expect our growth to stay in with the industry leading partners and competitors. We are confident in our ability to outperform the average market CAGR in a growth -- in a high-growth HPC application. Thank you very much.

Okay. Thank you, Johnny. And for this page, this is the details about our third quarter P&L. The numbers are very straightforward. For the third quarter this year, we record almost USD 223 million, and it represents 25% quarter-on-quarter decline and quite significant 51.5% year-on-year decline. I will address the reasons behind it later. So as a result, the operating income in the third quarter is $38.8 million compared to last quarter, which is a 3.2% quarter-on-quarter increase, while the year-on-year decline is 33.6%. And for the third quarter net income, we record USD 44.2 million, translating into TWD 16.4 EPS. The net income represents 3.1% quarter-on-quarter up and 20.3% year-on-year decline. And this is the revenue breakdown as usual. You can see that the HPC remains the majority revenue contribution for our revenue. The reason for the consumer product to our revenue in percentage going up is because the production contribution from Japan's consumer electronic projects. So as usual, we are focusing on the HPC, especially AI-related applications for our future. And for the process node breakdown, yes, like always, Alchip has been proud to be the leader within the most leading-edge process node technology. So in the third quarter, we have combined close to 90% in revenue exposed to 7-nanometer, 5-nanometer, 3-nanometer and the 2-nanometer those so-called most leading-edge process nodes and only around 10% of our total revenue exposed to 16-nanometer, 12-nanometer or the more legacy process nodes. I believe this revenue breakdown in process nodes remains the top tier within our industry. For the geographic breakdown, like I mentioned, for Japan, because of the mass production of the consumer product to Japanese entertainment end customer. The third quarter revenue exposure to Japan went up to 17%. In the meantime, the majority of our revenue exposed to North America -- still exposed to North American region, accounting for 74% of our total revenue in third quarter. And for the Asia Pacific, which includes China and Taiwan and almost no Southeast Asia exposure accounted -- combined accounted for 7% of our total revenue. So for the third quarter numbers, as just mentioned, the revenue declined. The main reason is our 5-nanometer accelerator chip. The shipment has tapering off because of -- the end of the life cycle actually happened in September. The 5-nanometer AI accelerator shipment, which shifted to the U.S. IDMs ended its life cycle in late quarter. And in the same time, we don't have too much production revenue contribution from the other projects. However, because of the higher NRE percentage in contribution, our profit margin improved quite a bit. The third quarter '25 gross margin improved to 28% from 21% a quarter ago. The NRE revenue contributed around 30% to 40% of our total sales in third quarter. And like Johnny mentioned, we have done a good job in the OpEx cost and expense control. And another reason for a relatively low operating expense is the less employee option amortization expense, which we expect to steadily go down a little bit in the coming couple of quarters. And I would say the increase in NRE revenue in third quarter is because mainly to the strong project inflow from -- mainly from the North America region. And in the same time, we do tape-out several big milestones for some important projects. Okay. This page is for the outlook for the fourth quarter this year and 2026. We expect our NRE revenue to further grow sequentially, which means we are expecting better profitability in fourth quarter. And we see the process node migration continues, and we expect higher NRE contribution in percentage also in absolute value in fourth quarter. And we expect the 2-nanometer projects to begin revenue contribution in fourth quarter. Actually, we already record some revenue for the 2-nanometers already. And for the revenue outlook next year, for 2026, we expect our total revenue to jump up. Let me starting from the fourth quarter. Fourth quarter revenue in all will remain weak, unlimited project to significantly contribute production revenue to us. It is a fact. As we mentioned, as the 5-nanometer AI accelerator project enter the end of the life cycle, in the fourth quarter, we don't have projects to contribute significant production revenue to us. And another reason for the weak fourth quarter revenue expectation is because the ADAS project, previously, we were expecting to begin its contribution in the middle of fourth quarter. However, because of the test -- longer-than-expected testing process, right now, we are expecting the revenue contribution from the ADAS project starting from very late fourth quarter, which is one of the reasons behind a weak fourth quarter. However, like I said, we expect revenue to level up significantly starting from the second quarter next year because of the shipment started for our most important 3-nanometer accelerator project to kick in, in the second quarter next year. Okay. I guess we addressed many points for the investors. And we are going into the Q&A session.

[Operator Instructions] Okay. Laura, please? Laura, you can unmute your speaker, your microphone. Okay. There could be some technical problem for Laura. So Charlie, please, you can unmute your microphone for questions.

Thanks Johnny and Daniel and also your comments about the results and the outlook. So first of all, I wanted to know how we can probably model the 3-nanometer or the future 2-nanometer revenue contribution from the major customer because I seem to sense that there could be sort of a business model change to customer on tool. So I think for this, right, not just the sell side, but also buy side, we do have some hard time to predict your future revenue. So just if I -- for example, in the past, we can cross check from supply chain to get a sense about the potential of the chip volume. And based on the ASP, we can derive our top line gross profit and then your EPS. Can you give us some guidance how we are going to model your future profits or revenue properly?

Okay. I'll go first, and Johnny will give you a broader overall picture. Actually, we cannot comment on specific projects. But since it is the most important project from us, I would like to provide some clues for your projection. First of all, based on our experience and the 5-nanometer project of this product, this 3-nanometer revenue definitely could be a multibillion business to us. And for next year, I want to emphasize, probably you may hear about that we told that the shipment likely to kick off in the first quarter. And right now, we shifted to the second quarter. However, no matter it started -- it will start in the first quarter or second quarter, the scale -- the yearly scale for this project remain unchanged. To us, next year, I will still expect this is a $1 billion business. And Johnny, go ahead.

Yes. So in terms of how much revenue we are going to get and some of the -- I also agree some of the business maybe the customer decide to handle by themselves due to some reason. But honestly, that's a customer's decision. For us, we just make sure the growth rate meet our expectation. I think that's already prior agreed with the customer that a certain portion of our revenue that definitely will be allocated to us with upside. Yes, if we have some prior agreed percentage, unfortunately, we cannot share, but it will ensure companies to grow and sustainable -- revenue will be sustainable in the future. And as you know, the N3 capacity is so tight and customers try to get more capacity. If any additional allocation they can receive from TSMC, we will share some of them as well. So I think we are -- even though we didn't get the full allocation, but our customers make sure the growth rate for our customers satisfied for both sides. And also when they have an upside, we will also enjoy it. I think that's an agreement on the current generation and also the generation gap.

I see. So can we get a sense of the so-called potential EPS contribution? Because no matter how hard we try to do the modeling, right, the biggest unknown is always the allocation. And that allocation seems to be very fluid based on what you just described, right? So I think some of the U.S. company when they provide revenue or gross margin guidance, sometimes they also provide the EPS range, right? So I'm not sure if management can also do that for us.

Okay. Charlie, unfortunately, we are listed in TSE. So we cannot provide the numbers. The authority, the TSE will be mad about that. And another thing is, actually, we usually -- we cannot comment the EPS or the things on a particular project. So I would like to offer you some clues for the production, for our 7-nanometer project, the 2 generations ago, we guide the gross margin for the product is about 10% to 15%. And we are pretty confident for the 3-nanometer project to be in the same range or better.

Okay. Yes, put it this way, it just came to me 10 seconds ago. Maybe let's assume the TAM of the ASIC would be USD 30 billion next year. Do you think you can get like 10% of market share next year? And your industry peer, right, they said 2028, the TAM would be USD 50 billion. Do you think you can get a market share of like 10% to 15% as well?

Charlie, don't push us with the numbers in this call. Yes, it is -- yes, we -- like I said, it is definitely a multibillion-dollar business for the project in generation. And since the life cycle of the product is expected to be in 2026 and 2027. So you can -- I guess you can easily get a ballpark range for next year's revenue contribution from this project.

Okay. Charlie, I know there's a certain answer. Yes, a lot of stuff is not clear at this moment. But I think we believe this customer, we've been working with them for a long time, whoever make more contribution to them, I think you will enjoy, I think, more outcome. I think that's for sure. Yes, in terms of EPS and those kind of things, I think once the project start to kick off after you do some calculation, I think on the given quarter, during the production you will figure out. But right now, I think it's too early for us to tell, and we'll be wrong almost every time to share some future numbers.

Okay. Okay. We will be patient. It's just from a little bit goodwill because for investors, they don't like uncertainty, right? Just my very, very honest opinion but we hopefully...

Yes. I also agree.

Yes. But we appreciate it and hopefully, we can get your updates next time when that project enters mass production. Yes, it's a long discussion. Probably I go back to the queue. I do have some other questions, but I will leave it to other speakers first.

Okay. Laura, you can unmute your microphone. Laura? So it seems Laura still has some difficult technical issue. [Operator Instructions] Okay, next question. So Charlie, please?

I will continue. Yes. So I just want to know kind of progress of other new projects. I think one is the 2-nanometer project you said you just booked some revenue. I believe there's similar types of chip. But in terms of the 2-nanometer mass production timing, is there any change? I think a quarter ago, you kind of talked about the revenue in the coming 4 years, right? So I'm not sure when should we start to incorporate the 2-nanometer mass production revenue in your future revenue forecast?

Okay. Actually, for the 2-nanometer project, because it's still times for -- from tape-out. So we don't talk about the potential production scale yet with the customer. However, based on the current industry practice and the outlook for those users' future CapEx, we do believe that the production scale for each generation will be bigger and bigger. And for the scheduling, I would say the cadence for the project for this customer is usually every 2 years. So I guess it's quite easy to predict the potential production time for the 2-nanometer project. However, I have to say that everybody wants to have the project go smoothly and quickly. Everybody wants to get the chip as soon as possible. So we will try our best and the customer will try their best. And of course, the whole supply chain will try their best in order to get the chip on time.

Okay. Let me also add some information. First of all, let me clarify, we have multiple wins for 2-nanometer, but I definitely know which project you are pointing to. So I think the -- just like Daniel mentioned, the scale for the -- for each of generation, I think that increased exponentially, I think like before even for the given N2 project. So I think we have -- we are under the huge pressure to complete this design as soon as possible. But to be -- I think both of us and also our customers are in sync, and we will provide the best service and to achieve the -- in order to control the schedule and to make sure the product go to production as soon as possible. But for us, to be straight, unlike the current generation because we don't have N5, so we have a huge expectation for N3, we are waiting for N3 to go to production in order to fulfill the gap. Like this year, there's a huge revenue gap. But in the future, this kind of stuff will not happen. If next generation happens sooner, we'll be very, very appreciated. We'll be very happy. But even though any production labors, it doesn't hurt our revenue too much because in usual case, when the next generation due to some reason, slow down a bit, the current generation volume will increase. I think the best thing for us is we have to continue. I think that's very important there.

Right, right. So are we confirm to win this 2-nanometer project as a big customer? There just have been several debate, right? How do we convince investors that you win this? And second question is that you talked about generation on generation, the revenue scale would be much bigger, right? But after changing to circular COT business model, I feel like the narrative should be on whether your profit -- your total profit to earn on a 2-nanometer project will be bigger than 3-nanometer. Yes. Can you comment on both? Yes.

Okay. How to convince. Honestly, I don't know how to convince. But what...

Okay. I guess that's our job.

Yes. Honestly, from our point of view, we just do our job. Actually, like I always talk to the investors for project with such scale and complexity, actually, in order to catch up with the cadence of the customer, a lot of engineering work is already there. So I don't know if it is true, but it is happening. So that's why we mentioned, actually, we already have 2-nanometer related revenue in our P&L in the fourth quarter this year and also profit scale.

Profit -- yes, profit scale. Yes, profit scale.

Profit scale, I would say the pricing pressure is always there. If you -- where customers choose, definitely, you want the suppliers to provide some cost cut every generation when the volume keeps on rising. However, we will try to defend the pricing pressure by showing our value to the design, by showing our value to the product and most importantly, by showing our value to our suppliers, which is the manufacturing, the OSAT and the packaging. So all in all, by combining those reasons, I would assume even there is pricing pressure for the next generation, we can mitigate the pressure to a certain degree by working with the suppliers.

Okay. Yes, I feel like 2-nanometer is much more complex and profit scale is much bigger. I think you deserve higher profits generation by generation. So -- anyway, so that's my second question, and I will hand over back to you.

[ Jeffrey ] from Macquarie.

So you mentioned the big accelerator customer, you mentioned the ADAS customer. In the past, you've also talked about networking projects. Can you maybe give a little bit more details on geometry, scale and timing of some of those bigger networking-related projects?

Actually, we do several projects within the networking application with a couple of North American networking players. And I would say for the smaller customers, the production scale is not that significant, honestly, because it is in a quite special application. But for the second customer, I would say it is a series of project change. And for the first project, it will go into the production phase next year. However, we do not expect very significant volume from the first one. But for the second one, which is going into the design, already, we are expecting the shipment volume to be significant. But the production scheduling is most likely to happen in 2027.

Yes. Jeffrey, I think the networking-related application, we all know the chip size are much smaller. But I think customers also adopt the leading-edge technology like N3. And we expect the revenue will be a certain number, but it's not happened to be a $1 billion scale, I think, for sure. But I think to be honest, if it's a few years ago, we still consider it's a very, very good project for us. But right now, I think because of the other -- most of the investors, I think, has a very high expectation for our growth rate. I think networking only play a small portion of the revenue starting from next year.

Okay. Great. And my last question, your IDM customer, any future projects do you expect the next couple of years? Or you think that's done?

Okay. For the IDM customer for now, honestly, no. But our concept is the same. For doing the AI-related chips, especially the accelerator or let's say GPU, if anyone wants to have a competitive edge, they have to find the best manufacturer to produce their chip. For now, it is TSMC. So for this IDM customer, if they want to do AI chip and with high commitment, I would say the better way for them is to do the manufacturing in TSMC as well in order to compete with their competitors. So if this IDM wants to do the project in TSMC, I think we have a pretty good chance to play the same role like we did in the previous generations of chip.

Yes. If they decide to use their own fab, I think the chance for us to win is very slim. And we probably do not want to touch.

Okay. Next is Patrick Bob [indiscernible].

I just wanted to ask about the N3 project. So whilst you can't give us guidance on the absolute size, have you seen that size be increased in the last few months or any changes in the potential size of that project?

The size of the production scale.

Yes.

Which project you mentioned about the...

Yes, which project you mentioned, sorry?

The N3 project next year.

N3. Okay. Actually, for this project, starting from the design, early design phase, the customer keep on telling us the scale of the production is a must. And we keep on hearing the additional thing, additional scale all the way through the design phase. However, as you may know, that the wafer capacity, the wafer and the CoWos right now in TSMC, especially for 3-nanometer is very, very, very tight. So like Johnny mentioned earlier, if we can or the customer can get more wafer allocation, there could be upside. So the scale really depends on the wafer.

Yes. For this product line, fortunately, the CoWoS doesn't play too much important role in terms of capacity because they're using different CoWoS solution. The major challenge is N3 wafers.

Okay. Got it. And so when do you anticipate finding out if they can secure more N3 wafers? And how much more could they secure? How big an increase could you see to these projects?

Actually, for the next year, we already have the numbers. But in the same time, since the demand is better, so both parties are talking with TSMC in order to get extra wafer allocation. There's -- in usual case, there may be some upside in the past, TSMC reserve little capacity for the emerging usage and maybe some of their existing customer change in plan. So that's a sort of upside.

Got it. And will the upside that could be unlocked there be proportional to you? So if they could increase wafers 20%, would you get 20% higher revenue? Or does this change in business model mean it's not as proportional?

Patrick, I know you want to quantify the numbers. But honestly, for now, we don't have an idea.

But as long as the customer has an upside, we will enjoy the upside.

Okay. Great. And then looking at your kind of other customers and potential new customers, has there been much progress in -- with other hyperscalers or with other large customers?

Okay. Yes. So the definition of the large customer, you probably mentioned about other CSP, right? We do -- we encounter so many customers right now whoever doing the accelerator in the industry. I think right now, we have a chance to win even for the CSP. To answer your question, if you ask me, do we have -- do we already have any significant win? Unfortunately, the answer is no. We don't have any other CSP at this moment. But to be very straight with you, every time when they have a big project on the accelerator side or on the CPU side, we have a shot. I think just like I mentioned before, sooner or later, I think most of CSP will adopt COT, this kind of open solution. I think they really like to work our business model to make everything so transparent.

Okay. Let me try Laura again. Laura, you can unmute your microphone. Okay, it works. Yes, we have to talk to her...

I just told Daniel, we have to pick up your phone before this earning call ends.

So you can hear me clearly right now, right?

Yes, very crystal clear.

Appreciate to help me come back. Johnny, you just mentioned that your open ecosystems business model will help you to engage more of the opportunity when it's moved to 2-nanometer project. So may I ask that what would be -- more details can you share with us on the business model? You work with a third party? And how would that impact your gross margin and also the business scale for the future AI accelerators? And before you answer me, can I clarify your 3-nanometer project business model is pretty much the same as your previous 7-nanometer one, that's correct?

Yes. Yes. Yes, the model is pretty much the same. Yes. So basically, the open ecosystem, what we propose to our customer, I think number one is the transparency. The customer know all the costs, including wafer, including IP, including our design service and also our margin. We agreed with a certain margin to add up to their cost. So this is, we call, the COT model. And both parties agree with a certain percentage margin. I think that's a prior agreement. The upside for us, yes, as you know, the volume increased drastically generation over generation. So in the other word, even agreement between us and the customer, the number is fixed. They negotiate with all the pricing by themselves. But sometimes because of a volume accumulation, we get -- we will get additional support from the supplier side. In this industry, I think the supplier play a very fair role. Whoever place more order, they can enjoy better pricing. I think that's an upside. If we accumulate more customer together, we will have more benefit from supplier directly. It doesn't conflict with the pricing, which customer negotiate with the supplier directly. And for the N2 technology or beyond, the design becomes so complicated, it's not like -- it's no longer a homogeneous type of one single tape-out and produce a chip. One design require multiple tape-out, require a variety of IP and also the solution. For example, we may work with some chiplet, I/O chiplet, chip provider. We can do some integration with them. We may work with a CPO vendor. We may work with different IP vendors. So I think I don't believe one single company, no matter how big they are, they can provide most comprehensive and total solution in the future technology, because different solution has different expertise, has different company to focus on that area. So we just try to put everybody together to provide the most comprehensive and cost-effective solution. So right now, we have multiple IP partner and PCB partner and I/O chiplet partner and also CPO partners.

Yes. So given the complexity of the chip designs and how chips will provide this kind of system integration and silicon modulation, et cetera. So can we assume that the gross margin in that kind of type business model will be higher than your current business model? [indiscernible]

Hopefully, I think that customer is very smart. If we will be providing more -- higher contribution to their share, I think that we deserve higher margin. But in fact, our margin expectation even increased a bit, still much less than our current competitor in U.S.

Yes, of course. Okay.

Okay. Next one, Gokul please. You can unmute.

So when we come to the COT business model, the one question that a lot of investors ask us as well is like is it the final kind of stopping point for the customer? Or will they move forward to kind of control everything themselves? Because your main customer also has some CPU projects where they do it pretty much direct to the foundry itself. So I just wanted to understand your thought process in terms of how you kind of make sure that your value is getting captured and the COT model with the open structure kind of continues to remain and the customer does not want to kind of move to like kind of internalizing everything themselves?

Gokul, I assume you are asking, is there any possibility for customer in-sourcing the production, right?

Yes. It's not just production, but I mean they are already doing COT. So is it the natural progression for them to do everything themselves in maybe not in the next or the next generation, but just wanted to understand how you kind of keep your value still.

I will say this way. First of all, if you are talking about in in-sourcing the physical design, we don't consider customer has very good incentive to do so because unlike the CPU and networking, the accelerator design -- the physical design of an accelerator is very, very difficult and complicated. No need to mention about the process node. So for them to build up a capable team to in-source the physical design for such complicated chip, what is the benefit they can get. And they have to take the risk for delay and mistake, multiple tape-out and a much longer design turnaround time. So from our point of view, we don't see the in-housing happen in -- I won't say in the short time. Even in the mid- to long term, I don't consider it will happen.

Yes, I agree. I think to answer your question, if you're asking me about the possibility, of course, anything is possible. If a customer can do the better job than us, yes, of course, if I'm a customer, I was doing everything by myself in order to save the cost. But nowadays, I emphasize so many times, design becomes so complicated, each of the player in the supply chain should focus on their own area. The reason we can tape-out everything on time because we're doing this kind of design at least like 10 to 20 design every year. If a customer decides to do by themselves, they're only doing the one design every other year. So the knowledge accumulation and those kind of things at the end, I believe customers will make it happen, but likely, the schedule will be delayed. Right now, I think the each of generation only less than 2 years or a little bit over 1.5 years. If the delay like a previous generation just happened, delayed for 6 months, the market is not going to be waiting for you. You lose the market, you try to catch up on the next generation. It's not worthy to take this kind of risk. And also different region has a different working mentality. I just joke with my customer, even I send my team to your company, do you have a confidence to make them work through the weekend, work through the holiday, work through the Christmas? I think different company has a different DNA and mentality. If a customer try to go beyond that, I think most likely, they will suffer a bit.

Understood. That is clear. So second question, since you talk about the Alchip moving to a bit more of a kind of collector of all the IPs and kind of offering it in an open kind of environment to customers. Could you talk a little bit about networking and chip-to-chip interconnect? I think right now, it feels like scale-up networking becomes as important, if not to some extent, even more important than the ASIC itself. You have the partnership with NVIDIA on NVLink Fusion and you are working with Astera Labs also on some of the other interconnect stuff. But at least observing from the market, it feels like that piece of the IP seems to be still very proprietary. The open source seems to be -- at least if you look at what happened in OCP and all the developments, it felt like open source is definitely behind the curve compared to proprietary technology from NVIDIA or Broadcom. So how do you see this is evolving when you talk to all the customers? And what does Alchip's -- what could be Alchip's role here, including any progress you are seeing on the NVLink Fusion?

Okay. For the customer silicon or just say ASIC, we are trying to provide an open ecosystem to customer. The reasons are, first of all, by providing open ecosystem, customer has much, much better room to save some cost because in the future, you definitely know that those future CapEx plans is crazy. Any percentage saving is a big money. We do believe with -- by providing an open ecosystem to customers, customer can enjoy much better pricing or cost saving from the ecosystem. And secondly, like Johnny mentioned, we always believe that everyone should do what they are good at. We are a physical designer. We do physical design. And if the customer needs relevant related IP, we can provide IP -- our IP partners to them. And for CTO, the same thing. For I/O chiplet, the same thing. So by providing that, we believe we can have pretty strong competitive edge to our competitors. I would like to mention one thing. 20 years ago, actually, TSMC is providing an ecosystem and win out the whole market from IBM. For now, Qualcomm, Marvell, MediaTek probably can offer everything, just like IBM did 20, 30 years ago.

Got it. Any significant progress you are observing on NVLink Fusion, given it's already about 6 months since you signed the partnership?

Yes. NVLink Fusion is a very good technology, and we are more than happy to work with NVIDIA for that. But in the end, it is customers' decision. It's a good technology and many of our customers are interesting with this technology.

Yes. So saving has many dimensions. Sometimes a quick turnaround time is a saving. Sometimes the total power consumption is a saving. We do have a few -- many inquiry on NVLink Fusion cooperation, and we work with NVIDIA very closely. They even assign the people when the customer come in, we co-visit customer to promote the solution altogether. Hopefully, we have -- in the near term, we have some good news to share.

Okay. Okay. Understood. Understood. That is clear. And just a follow-up on a different question, I think, from a different analyst. On the second customer in CSP, Johnny, in your mind, what is the main challenge for you to get the second customer? Is it the customer coming to have the same kind of capability as your first customer that they can easily embrace your open ecosystem? Is that the bigger challenge? Or is there some other factor that is the bottleneck to potentially win the second CSP customer?

I think the most challenge for us is we have to change customers' mentality and perception a bit. And I think when we talk to the potential customer, I think they are kind of convinced which area they should focus and which kind of area they should outsource. But it takes some time for other -- I think our current customer is different. They already have a design team even before everything. They have a core team prior to winning this business. And the other CSP doing reversely, they already have a business, they try to build team. They hire a lot of guru from each of company like IC company, NVIDIA, AMD, everywhere. So they have a different mindset, they have a different mentality. I think we have to -- working with them, I'm not saying educate. We have to work with them and tell them what's the right way to adopt the COT model. Yes, it takes time. But every time when we have some engagement with them, I think they are lessened, and we are -- I think we have a shot and some capability they have to develop. And sometimes more people doesn't mean they can do better job. I think efficiency and also the one team with a good leader, I think they can help us to do a much better job. So yes, to answer your question, yes, that's a challenge because we try to work with the pure CSP and work with pure system company to do COT. I think that's still a lot of lesson learned for both parties.

Okay. And Charlie, please.

Sure. So I remember a couple of quarters ago, you sort of shared some chip price -- not chip price, right? The ASP is like 4x to 5x versus the 7-nanometer generation, meaning your 3-nanometer chip ASP. So do you still hold the same assumption? Because management just said that in coming months, right, you probably will have a better visibility for the wafer allocation. So can we also get a little bit color about the ASP assumption?

Yes. Unfortunately, Charlie, we cannot talk about the pricing. Yes, it's a hard requirement by the customer.

Right. I think recently, you can see a lot of news from that player. Yes, they just have a concall. They have -- pretty much has a high confidence to use their own chip, eventually replace their current supplier. I think the next generation and also the generation behind, I think the plan doesn't change. We expect a very, very big volume and progress on N3 and beyond.

Okay. Okay. Great. Yes. So the next question is a little bit harsh. Yes, because for your recent quarter revenue mix, part of that is due to you cannot secure a 3-nanometer wafer for a crypto customer. So how investors should be convinced that they can -- you can secure 3-nanometer wafers for your major customer next year? And also another tougher question is that your local peer, which is also your foundry subsidiary, right? Recently, it seems to be okay to secure those wafer source for crypto. Can you comment about what should we think about your circular foundry partnership?

Okay. First of all, crypto project is totally different from accelerator. They have huge difference for the priority within TSMC. So for crypto business in TSMC, the priority is low. So when you try to secure wafers from TSMC, especially for the 3-nanometer, it is really, really difficult. And I do believe that our peer is also facing difficulties of getting 3-nanometer wafers for the crypto customers. And for the accelerators, it's a totally different thing. First of all, the customer is what TSMC wants and the application is what TSMC wants. And fortunately, the CoWos of the chip is in less crowded process. So for the 3-nanometer accelerator project, for now, yes, the customers do want more wafer. However, the current allocation is quite good.

Okay. Sure. So which means that for next year's 3-nanometer for AI accelerators, it will be totally different case versus the crypto.

Yes, yes, yes.

Okay. Okay. Great. Yes. And last one is the Arm Total Design. I think following Gokul's question about NVLink Fusion, Astera Labs partnership, all good, right? But I remember you also have this Arm Total Design circular partnership. So I'm wondering if you consider to be more aggressive in Arm Total Design because lots of CSP customers, they wanted to do their CPU as well. And also, again, referring to your industry peer, very, very low margin, but actually it's very, very good for their revenue momentum. And I believe it should be still positive to the EPS. So Johnny, how do we think about if there will be future opportunities coming for this kind of Arm-based CPU, how should we handle this kind of opportunity?

Right. I think a couple of weeks ago, we just announced we joined Arm Total Design service. I think we work with Arm very closely, and they think about -- even though they say this program is close before, but they kind of open and reinvite us in for some reason. There are some CPU-related design opportunity we are using Arm. I think focus on the design is our #1 priority. We try to do design along with the production in order to provide value and also to get a higher margin. I think that's our #1 priority. We do have some opportunity in U.S. and also in Japan using Arm as a server type of CPU. To answer your question, if any design -- any customer come to us to do the production only type of business. First of all, we will evaluate the margin. If the margin are within the range, it doesn't fall out of our bottom line, we will take. But it's -- I think I agree with you in terms of EPS, in terms of risk, it definitely helps. But the drawback side, they will have a certain impact on the gross margin as well. So if it's a CSP type of customer come to us, most likely, we will try to participate because we try to win their design and eventually try to provide more value on the design side. If it's other business, other smaller customers come to us for production only type of business with low margin, most likely we will reject.

Okay. Because of the time constraint, we took 2 more questions, one from Laura, one from Edison. So Laura, please.

Just very quickly, Johnny, you did mention that the constraint of 3-nanometer wafer supply and yet for the CoWoS part, I'm just wondering, do you already work with OSAT partners or still mostly at TSMC on the CoWoS packaging?

Okay. Laura, actually, the bottleneck for the 3-nanometer project, right now, the bottleneck is actually wafer but not in CoWoS. So CoWoS has a different time, right? The specific time we are using, the capacity is not the bottleneck to provide the chip. The top die wafer will be the area, major shortage.

Yes, sure. So I'm just wondering that do you already kind of secure enough CoWoS supply at TSMC or for the CoWoS, you also have kind of a backup plan to work with the OSAT partner, thus there is a relatively less constraint?

Actually, recently we learned the CoWoS is secured. And Edison?

Can you hear me clearly?

Yes.

Yes. Just a quick one. I recall in 7-nanometer generation, we were facing some of the ABF substrate shortage issue. So we increased the suppliers. So I think as ABF now is getting tight again, my question is, can we get enough ABF substrate supply this time? And will this impact our 3-nanometer project revenue scale if we cannot secure enough ABF substrate? Because you have already talked about 3-nanometer capacity, cost capacity. So I just want to know your view about the ABF substrate situation.

Okay. Let me explain to you. Actually, there are numbers. The numbers based on the secured capacity for wafer, CoWos and of course, the substrate testing, everything. So there are numbers because we are entering November now. What we are trying to say is if there is any upside from wafer because the demand is getting higher, if we can have or the customer can have more wafer allocation from TSMC, definitely because of the [indiscernible] of the customer. I don't think the other parts of the supply chain will be a very big problem. Yes, I think there's a big lesson for us and also for our end customer right now for -- except the top die for HBM, for substrate, I think they have a multiple source in order to diversify the risk.

Okay. Got it. Can I ask another question or we need to...

Sure, sure, sure. Please.

Yes. I think my second question is about your major competition and also on the 2-nanometer project because we're kind of hearing that your major competitor is also considering lower the ASP or margin in order to seek more opportunities from your largest customer now on 2-nanometer project. So I just want to know that if they lower the ASP, will this impact our current relationship with our largest customer? Or do you think the ecosystem we are building now is strong enough to compete even though they consider to lower the ASP? That's my last question.

Okay. Edison, let me make sure the competitor -- major competitor, are you referring to in U.S.?

U.S., yes. Yes. U.S...

Okay. First of all, we try not to comment on other companies. But for the competition, I would say this way, first of all, design capability and ability are our strengths. We want to win because of our design. And our existing customer knows it very well. They understand inside out about our design capability and the value we can provide to our customers. And on the other hand, it's the consideration of the cost. I don't think there is any chance that the U.S.-based competitors can compete with us on pricing. So for the future generation, first of all, like I -- like Johnny mentioned or like I mentioned, for a project with such complete -- such scale, design scale and the complexity, the engineering is already there. And for the other thing, I don't know. And I really don't want to comment on it because it was like we are trying to -- we really don't want to say things on other companies. Okay. And the time is late. And I guess this concludes our third quarter earnings call. Thank you for your participation. See you again in our next earnings call. Thank you.

All right. Thank you very much. Thank you, Daniel.