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

It's time. Dear investors, analysts and fund managers welcome to AIchip Technologies Fourth quarter 2024 earnings call. I'm Daniel. And this call will be hosted by me, Daniel and our CEO, Johnny Shen, as always. So next page. Okay. As usual, it's safe harbor disclaimer and the next page. Okay. This meeting will be in English. So if you need Chinese presentation slides, please go to the MLPS to download the Chinese version. And you can write down your questions through Teams message function. Sorry, I didn't change the application there. And/or you can use the raise hand function through the teams application for the Q&A session. The host will unmute you for your questions. And this video and audio content of the meeting will upload to the MOPS, which is [indiscernible] about 3 hours later after the meeting closed. Next page. Okay. First of all, I want to invite Johnny, our CEO, to start with a very short AIchip introduction.

Okay. Good afternoon, ladies and gentlemen. I'm Johnny Shen, President and CEO of AIchip Technologies. Once again, thank you for joining our investor conference meeting, we appreciate the opportunity to share our 2024 result and provide the guidance for future business outlook. So briefly, a company update, we founded -- the company is founded in 2003 last year, we've been achieving market cap more than USD 9 billion since we found the company, we've being successfully take out more than 600 design. And within 600 design, this more than 60 are FinFET and also more than 18 are CoWos related design. Our last year revenue achieved $1.62 billion, majority of our revenues, the right contributed by HPC and AI, where TSMC, 3DFabric member and also VCA member. Our annual capacity is the 20-30 tape-out every year. Our market focus is HPC, AI, networking and automotive. Okay. Yes, allow me to give everybody a quick update for our Q4 and also the last year summary. We are pleased to announce our Q4 financial results with revenue reaching $404 million, representing approximately 39% year-over-year growth and 12% Q-o-Q decline. Our net income came in at $57 million with an EPS TWD 23. Both markets record high for the company. For 2024, our full year revenue totaled $1.62 billion with net income of $201 million, EPS TWD 81.3, all historical high for the company. We consider 2024 is one of the best year for AIchip. For the first time, we have surpassed $1 billion in revenue, achieving over $200 million in net income. We are proud to be world's #1 pure ASIC service provider and #2, AI ASIC provider by revenue. More than 90% of revenue directly from HPC and AI applications. In addition, we have successfully completed approximately 30 tape-outs primarily using leading-edge technology and has made significant progress in penetrating in ADAS market. Regarding the capital market, Alchip maintained one of the highest liquidity level in the industry. Despite strong revenue growth, our share performance last year is weak due to various vector. The stock experienced high volatility through the year, ultimately ended flat. The detailed breakdown comparison will be presented by CFO, Daniel in the later section. As for the future business outlook, our major mass production line is currently in a transition period. The previous generation is approaching end of life, while the next generation is set to ramp up starting from next year. We had high expectation for our IDM customer and also ADAS production this year. Initial volume forecast was very promising with the potential to offset the transition gap, even drive the revenue growth. However, post production are facing challenges. Our IDM customer has revised their production demand multiple times and ADAS customer revenue will be delayed 2 to 3 months due to a recent incident. As a matter of fact, our mass production revenue forecast has dropped significantly compared to previous earnings call. Regarding the design opportunity, we are seeing increasing demand for HPC and AI-related business in North America, utilizing the leading -edge technology like N5, N3 or even N2. We have already secured several design wins and remain confident to further grow our NRE revenue this year. Additionally, there's a potential upside in the cryptocurrency sector. The revenue opportunity could be significant provided we can address the dependency such as wafer capacity limitation and also the prepayment requirements. For the geopolitical risk management, we have successfully diversified our business beyond China into other regions. In 2024, only around 10% of our revenue are originated from China. However, we remain committed to support business in China provided they are financially healthy and compliant with all the related regulation. Regarding our workforce expansion, we have launched aggressive hiring plan to strengthen our engineering support resource outside of China, including Japan, Taiwan, Malaysia and Vietnam. Our Japan office has expanded to fit up to 150 employees and our Malaysia office already staffed more than 25 engineers. We plan to grow our Southeast Asia workforce to 120 employees by the end of this year. It is a strategic expansion plan to provide more flexible and cost-effective solution to meet our customers' dynamic requirement in the global market. In conclusion, our revenue faced some challenges this year. We anticipate 10% to 20% year-over-year decline, but our percentage gross margin expect to be improved, driven by higher production margin and also greater contribution from NRE revenue. Despite the revenue adjustment, our earnings forecast remain consistent with last year. So overall, we remain highly optimistic about the future AI market from 2025 to 2027, like other industry leader, we are confident to achieve 40% to 50% compound annual growth rate. Thank you very much.

Okay. This is the numbers, and it is pretty straightforward. I guess you -- you can find our revenue numbers easily from the public information. I'll state it very quick again. For the fourth quarter last year, the revenue is USD 404.1 million, which is a 12.1% quarter-on-quarter decline and 38.8% year-on-year growth. And for the operating income, we reached USD 54.2 million in fourth quarter last year, which indicates a 7% quarter-on-quarter decline and a 37.6% year-on-year growth. And for the net income part, we record USD 56.9 million net income, and it represents 2.5% quarter-on-quarter growth and 60.7% year-on-year growth. The fourth quarter last year, the EPS will be TWD 23. For the revenue and for the reason behind better profit with quarter-on-quarter decline revenue, I will address the reasoning later. Next page. This is the comparison for 2023 and 2024. For 2024, the total revenue is about USD 1.6 billion, which is 65.4% year-on-year growth and the operating income is USD 202.3 million, which is 67.6% year-on-year growth. And for net income in 2024, we record USD 200.7 million, which is 88.4% year-on-year growth. For the EPS for the whole year last year is TWD 81.3 based on the concurrent shares outstanding and the FX. Next page. And for the revenue breakdown, as usual, for the application, you can see HPC and AI represents dominant percentage of our total revenue. For last fourth quarter, the HPC and AI-related application accounts for 92% of our total revenue. And for 2024 as a whole, the HPC and AI-related revenue accounted for 93% of our total revenue. Next page. Okay. For the Process Node breakdown, you can see for the fourth quarter last year, the 5-nanometer or advanced process node is gradually growing. In fourth quarter last year, these Process Nodes account for 45% of our total revenue from 23% in third quarter last year. And for the 7-nanometer -- for the revenue in the 7-nanometer Process Node, it accounted for 51% of our total revenue. And I would say the main reason behind is because our major products, the 7-nanometer AI accelerator is gradually phasing out starting from the fourth quarter last year. And for the 2024 as a whole, for the 7-nanometer higher, I didn't put the 5-nanometer category into the yearly comparison in order to be more consistent with 2022 and 2023. So for 7-nanometer and the more advanced nodes, the revenue come from those nodes accounted for 96% of our total revenue. I do believe for the Process Node technology, Alchip is still the industry leader among all the ASIC vendors. Next page. For the regional breakdown, like Johnny mentioned, we are gradually shifting our focus to the North American market. For last quarter, fourth quarter '24, the revenue come from North American region accounted for 90% of our total revenue, while the Asia Pacific region contributed about 8% of our total revenue in last quarter. For Asia Pacific, it includes Taiwan, mainly Taiwan and China. So for 2024 as a whole, the revenue from the North American region already accounted for 86% of our total revenue, while the revenue contribution from the Asia Pacific region accounted for only 11% of the total. Next page. Okay. For the fourth quarter and the 2024 business review. As I mentioned, the fourth quarter revenue declined by 12% quarter-on-quarter, and it's mainly due to the tapering of shipment of the 7-nanometer AI chip to North American cloud service provider customer. And like I said, despite the weakening sales in fourth quarter, the net profit in the same period came at USD 57 million, up 3% quarter-on-quarter and 61% year-on-year, respectively, because of higher gross margin. I think the better fourth quarter gross margin is mainly -- due mainly to the more favorable revenue mix. Like Johnny mentioned, we have a little bit better percentage mix for the NRE. And for the production, the product, the major contributor in the production revenue -- for the production revenue has better gross margin than our 7-nanometer AI accelerator products. For 2024 as a whole, I would say it is a very strong year in Alchip's history. The 7-nanometer AI chip shipment to the cloud service providers is still the major contributor. And as I mentioned, the net income outgrow the sales growth. It represents that we have pretty good operation gearing and better nonoperating income. And the nonoperating income last year contributed mainly by the higher interest income compared to the previous years. And I'm proudly say that the both top and the bottom line in 2024 are our record high numbers. Okay, next page. For the outlook in the first quarter and 2025, I would say for the first quarter '25, we would like to post a seasonal decline because, first of all, the seasonality of the NRE revenue. Usually, the fourth quarter is the high season for our NRE and the first quarter is a low season. Secondly, the shipment of the 7-nanometer AI accelerator will keep on going down while because of the end of the product life cycle. And the 5-nanometer AI accelerators to the IDM customers is kind of slow due to customers' special situation. I guess many of you already acknowledge the whole -- the situation right now our IDM customer is facing. And we expect a quite 2025 sales performance because, first of all, the phasing out of the 7-nanometer AI accelerator shipment. Secondly, the soft demand for the 5-nanometer AI accelerator to the IDM customers. And for the number-wise, though we were won by TSE, many times, we cannot release the guidance in numbers. But roughly for now, we expect our revenue performance in 2025 will be declined 10% to 20% year-on-year. And the demand and the pipeline remains very strong. We see strong project pipeline for AI and HPC and networking-related projects, coupled with the Process Node migration from 7-nanometer, 5-nanometer to 5-nanometer and 3-nanometer, and we expect multiple 2-nanometer projects to kick in this year. And we see more favorable sales mix, which will result in, I would say, pretty good gross margin improvement in 2025. And under these circumstances, we don't -- I don't consider the bottom line performance will be -- will be worse as the sales performance this year. Okay. I guess that's the whole part of the company presentation, and we will go into the Q&A session.

[Operator Instructions] So the first one will be UBS, Annie. Please unmute your microphone. Annie, UBS Annie Chen, you have to mute your microphone. Okay. Probably some technical problems for Annie. Charlie, please. Morgan Stanley, Charlie.

Can you hear me okay? Hello?

Yes. We can hear you clearly.

So first of all, maybe just want to go through your key assumptions for this year's revenue guidance. I'm wondering if you include the 3-nanometer production revenue in this year's guidance when you provided a 10%-20% decline. This is the first question.

No. Actually, based on the schedule we are having right now, the revenue contribution, it is possible we may kick off the mass production close to the end of this year. However, we -- first of all, we want to be conservative. Secondly, even we can have the production contribution from this product. The revenue could be insignificant because of the scheduling.

Okay. So if you anticipate you're going to do the mass production, is it fair to say you have already tape out this 3-nanometer chip?

Okay. Sorry, I cannot disclose the details of the project based on customers' request. But like I said, usually, if you want to have a 3-nanometer AI accelerator into production in early 2026, you need to tape out at least in the first quarter 2025. So based on the schedule, you can guess the situation for the chip right now. So all the scheduling right now is on track.

Okay. Got you. And also, Johnny, I appreciate you provide a long-term CAGR guidance. I feel like your business will have a big fluctuation. So providing a long-term CAGR, I think is a very good news for long-term investors. But what's the trajectory? First of all, when you said 40%, 50% CAGR, is that also TSMC's 5-year CAGR or is a 3-year CAGR? And any implication to 2026 revenue growth?

Okay. Let me try to cover that. So basically, the similar number, I think, presented by -- shared by the market leader, like the T-Foundry you mentioned about and also the respectful competitor in U.S., Broadcom also mentioned about they also have a very aggressive plan. And for example, they mentioned about 2027 will be $45 billion from the ASIC market. I don't have a precise number shared by our customer, but we do receive some information because based on current ASIC revenue percentage weight, our customers play a very important role. Yes, they have confidence to maintain the same percentage contribution in terms of total ASIC market contribution. So based on this effect, I think 40% to 50% compound, I think, is highly achievable.

Okay. And last one, and then I will be back to the queue. I'm a little bit worried about the circular design house wide list. So I think some of your competitors or industry peers are in the design house wireless, right? So I'm not sure if that would impact your future engagement with Chinese customers who want to do like AI or high-performance computing chip.

Okay. Charlie, for the wireless, honestly, our legal department is still working on the wireless. But I have to say, for now, first of all, we still have about 1 year base period. Secondly, because the Alchip has always been compliant with all the regulations, so at least for my personal thinking, we should be fine to obtain the qualification on the wireless in the future. So far, we don't see great difficulty of applying the list.

Why did you stop -- I think some customers did mention, but right now, it's not a showstopper because we still have a grace period until April 2026. And yes, in fact, this wireless list just been announced, I think a couple of months ago while the Biden administration, I think the last stage of the Biden administration. For U.S. customer, we don't have any concern. For China customer, yes, because of that, I think we need to apply the wireless. Based on current interaction from the legal department, I don't think we have too much difficulty to get on the wireless.

Okay. Laura, Citi, please.

Can you hear me?

Yes.

My question is also about the long-term growth outlook. I think we obviously quite confidence about -- after the transition into next year. So I'm just wondering, aside from your existing customers, any other potential big projects you are working on? I recall last time, Johnny, you mentioned that several other projects such like the networking part and also some of the staff companies. Can you share with us more of the progress right now? What's the process now they are? And when is the -- like the schedule we can expect will have better visibility or contribution?

All right. Yes, let me try to answer first, maybe Daniel can add more color later. So first of all, in order to further grow our company, winning more customer may not be -- yes, winning more customers, I think, may not achieve our growth rate. I think the key is to secure our current customer and winning their next generation. That's the #1 priority as a company. If we maintain this design way and winning their next generation, yes, for sure, we're going to have a 40% to 50% growth rate. So the other project will be the upside. But given the fact, other than this account, we did do well for other accounts. As I mentioned before, even last year, we've been taping out more than 30 designs and this year, we see more and more opportunity in N5, N3 or even N2. Some of the -- whoever has the funding to do the N2 are a very big company. Even they are startup, they are famous start-up company or they are a well-known company. They have -- that's why they have a funding to do the N2 design. So unfortunately, we cannot disclose too much detail about the big account penetration, but I can -- yes, I can mention -- all I can mention is that whoever they have a design opportunity, we will have a chance to compete. To be honestly, there's a very limited proven ASIC vendor, maybe less than 3. So whatever the IFQ release, Alchip will be always has a chance to compete with others.

Laura, like Johnny mentioned, I would say this market is very special, especially for the Cloud Service, AI-related chips market. I guess all the investors acknowledged there are only 4 big fishes, the Meta, AWS, Google and Microsoft. We -- and there are really limited numbers of the suppliers, Broadcom, Marvell, Alchip, MediaTek, probably GUC. So whatever there is a new project, I think the customer will release /the RFQ to the vendors I just mentioned. But to compete for those projects, for now, honestly, I will say Broadcom will still be the strongest among all the suppliers in terms of the front-end capability, IP readiness and the back-end capability. However, we also do think Alchip is very competitive in competing projects. First of all, we are more flexible. Secondly, we are pretty good in pricing. And trust me, in the future, pricing will be a very big issue for this market space. So what we are trying to do now is to be technology ready. First of all, there will be many, many new technologies will be into the 3-nanometer and 2-nanometer Process Node projects, such as the 3D-IC, such as IO chiplet solution or probably in the near future, CPO. So for now, I would say the advanced -- the ability for -- the ability and experience in advanced packaging will be Alchip's advantage in competing with the future projects. Secondly, we are already prepared a lot of things for the 3D-IC kind of technology and the IO chiplet solution. So for me, I don't want to release very optimistic guidance or hope to the investors because for the ASIC business, winning is everything, lose is nothing. So as long as we have a very good position, I believe we eventually will win some projects from other customers in addition to the existing one.

I'm asking that because we see that the ASIC every generation candidate is also getting kind of shortened. We know we are adding more R&D resource, but -- and also, we have quite strong confidence on the next generation next year, 3-nanometer project. But some of the long-term investors, they also care more about that what will be the next one. So yes, that's the reason I'm asking that other than the existing biggest customers, what would support our better visibilities into maybe 2027 onwards?

Okay. Okay. First of all, for the cadence of the chip introduction, I would say right now, the ASIC, the ASIC is catching up with the ASSP, you may see the current GB200 is in 4-nanometer. And our customers will have the 3-nanometer product by the end of this year. So I think the speed for ASIC related AI accelerators is catching up. Secondly, I think for the future, like you said, 2027 and onwards, like Johnny mentioned, to execute the current 3-nanometer project in perfect shape will be job #1 for us. Secondly, to win the next-generation accelerators from the existing customer is the job 2. For now, we are highly confident we can be the generation-by-generation partner with our major customer. And next one, Robert, JPM.

Johnny, Daniel, can you hear me?

Hello, yes.

I don't want to delve into the long-term growth outlook because when I look at lease for CSP customers, I think there's only one CSP pursuing the engagement model, which is outsourcing the back-end design to you guys, right? The other three is basically the outsourcing the full stack design. So that requires some front-end design capabilities. So what's your strategy here to tap into the other 3? Because as far as I think compared to other bigger U.S. companies. I think the brand design generic volume is much stronger. Are we going to beef up our fund and design capabilities to tap into other three customers?

Yes. Let me try to address this. Front end, I think you can categorize the front end in many field. So basically, our strategy and principle remain the same. We will not touch customers' architecture. We are the service provider. We will not get involved on the spec. We don't want to compete with -- we will not compete with our customer. But in terms of IP integration, either a scale out or scale up. We do have a plan to establish the resource in order to connect SoC, which is a GPU XPU to the networking part. We do have this kind of plan to expand our resource either through investment, we try to hire more people or we partner with some partners. Yes, we do have this kind of requirement. We do receive this kind of requirement and also have this kind of opportunity. And some of the CSP and many start-up company has this kind of requirement. Our strategy remains the same. We will not get involved of the architect. But for front-end integration to connect all the IP together, UCIe, HBM, I think we can -- we will have a solution for our customers to use. And fortunately, our #1 customer, I think their strategy remains the same, the cooperation between us and them, I think, is working very well for current generation also for the next generation or beyond.

Yes. Also, I want to follow up on this. I think you touched upon some of my question. So is there any possibility of this key customer to change the engagement model? Like, for example, they are now at 3-nanometer and in next generation, they'll migrate to 2-nanometer or A16. Will they be, I mean, requiring some front-end design support, RTL support from U.S. vendors for that? Or do you think that they can still handle the front-end design for even the next few generations?

Yes. I'm not much sure about any specific customer. If their design, next-generation design remain homogeneous. I think we don't -- they don't have such kind of a requirement. If they think about chiplet, and yes, we may -- both of us need to do something. So that's why before taking the next project, there's a lot of homework we need to do, a lot of trial, including test chip. Yes, both companies need to work together. And yes, we need to do a little bit more investment in order to support the chiplet design. I think they also understand. Yes, you're right, to answer your question. If the next design go to chiplet and there's more connectivity integration kind of a design work need to be done.

Okay. Rob, any further question?

Yes, I'll go back to the queue.

And if you have any questions, please use the Raise Hand function. We will -- and then we will call you and you can unmute your microphone, and we will answer your questions. Okay. Allison, HSBC, please.

Johnny, Daniel, can you hear me?

Yes, please.

I just want to have a quick follow-up. I noticed that our OpEx number increased like 17% Y-o-Y last year. I'm just kind of curious, how should we look into like 2025 OpEx level given more engineering expansion overseas.

Okay. There are a couple of onetime things -- onetime items in our last year's operating expense. So we see very limited operating expense expansion in 2025. In other words, I would say you can project single-digit growth for our operating expense this year.

Very clear. So I think my next question will be given we have many cash on hand right now, and this also leads to a higher interest income. I'm just kind of wondering, when do we expect this cash level will significantly decline. Will this be like same quarter next year, second half next year? I'm kind of curious when is the real timing for the cash level to decline significantly?

Okay. I knew we have a pretty good cash position. However, in order to handle the coming production needs in 2026, which I have to -- I want to mention again, the scale of the production could be really, really significant. So the cash position I would say probably will start to go down in the second half this year in order to be prepared for the production in 2026. And we will see the terms among Alchip and the customer and the vendor, especially TSMC in order to get a more clear picture for the cash flow while we do the 3-nanometer production plan. May I answer your question?

Yes, very clear.

And any further questions from you, Allison?

That's it. I'll go back to the queue.

Sure. And Charlie, please.

Thanks for taking my further questions. So I guess there are 2 or 3 follow-ups. First of all, Johnny, you mentioned there's IP for future [indiscernible] design, right? So I wanted to ask a very specific technology CPO because to me, CPO is more than just an IP, right? It's EIC, PIC design and also the CPO module, right? And if you look at the global sensors provider, Broadcom is very ahead in this technology. Marvell claim, they will provide the solution to ASIC design soon. So my question is that for your coming 2-nanometer competition, is CPO being considered? And even it's not the coming 2-nanometer, let's say, the coming 2 to 3 years, right, if you need to get a CPO technology, whether you really have a real partner?

Yes. To be straight, the inquiry for CPO potential customer right now, we have a few, but not too many. CPO. We do have 2 vendors willing to work with us very closely. In fact, some of them, we are even thinking about investment. I think within a short period, maybe you will see some announcement made by us on the CPO kind of approach. We all know increase the bandwidth and also taking care of the heat problem. I think that definitely will be the next-generation consideration. CPO and also SOW, I think we all have some solution. We may not be like Broadcom, we can do a lot of test chip, but we do have a few customers are willing to do this kind of study and R&D along with some start-up company as well. I think on CPO stuff, we are pretty much pay attention and we'll have a plan in place very soon.

Well, Charlie, no, I cannot disclose the NIM of the company we are trying to work with. This company is considered leading the optical solution companies in Taiwan. And most likely, we will have a plan for what do you say, the module, the solution of the modular solution of the chip and in the module very, very soon.

Okay. Yes, I don't mean to dig in too much detail. Actually, my intent is really to get a sense in the coming 2-nanometer project competition, whether CPO is a consideration. So it seems like CPO is more for the future project, but for the coming 2-nanometer competition, so what would be the critical points for the major customers to consider whether they want to award to you or your U.S. competitor?

Okay? So I think for N2 technology, most likely due to the optical size limitation only and most of people were using the chiplet solution. And in order to reserve for the compute die, the major top die for the -- only for the compute area. The rest of even the analog mixed signal and also the SRAM, they may put in another mask. So those kind of integration, I think it will be very challenged. And the interface, either they are using their own proprietary solution or UCIe. And there's many area we need to -- we need to prove it. So we do have -- we do win some design, N3 and N2 chiplet-related design. We are pretty much will be the first few company has a chiplet production soon. And in addition to that, this year, we also have a plan to work with the foundry and to make our own test chip, either it's N2+N3 or N3+N5. So this is a test chip and also along with the customers' IP, this plan already in place. So I think future technology in addition to the form factor and also the heat solution like a CPO or SOW, I think the chiplet is the most important area. We do the -- we are focused to prepare R&D around that.

I see. Yes. So that leads to my next question. Pricing. So Daniel, you talked about pricing is one of your key advantage. So I think two advances it could be both side. It means that you're competitive, which is good, but it also means your long-term gross margin profile. So as a CFO, can you give us some sense about your long-term target of gross margin given it seems to be more aggressive interim pricing. And if you look at the operating margin, right, I believe there's a more fair kind of metric to evaluate your long-term profitability. So can you comment on both fronts, gross margin and operating margin for long run?

Okay. Sure. First of all, the reason why I said pricing will become more and more important in the future is because the revenue generated from AI chip, especially from those cloud service providers to the ASIC vendors is getting more and more significant, which means if the suppliers -- when the suppliers try to maintain the same gross margin, the gross profit the vendors can have is double, triple, quadrupled in the future. But however, in the same time, like Robert mentioned, I do think to control the front-end design by their own is the target by most of the cloud service provider customers. And based on the logic thinking, like the customer is trying to do more and more of the NRE, while the suppliers providing less and less, if we assume that the supplier is providing the front end and the back end previously. But in the same time, as the revenue increase, the profit margin, they can get more if they're trying to maintain the same gross margin. So I think that's a challenging thing for the ASIC vendors within the cloud service providers market. So for us, I would say, for the gross margin side, last year is a good benchmark, which means when we have a very, very significant scale of production, the margin for last year is the margin we are trying to maintain. It is also difficult to us because the same thing happens on us as well. So -- but fortunately, we can argue with the customers saying that we are -- we probably will be the cheapest among the capable vendors for the AI accelerator. So I would say our target is to try to maintain the gross margin. But to be honestly, it's a tough environment to negotiate with customers with such big bargaining power. And in the same time, we want to leverage our customer and our revenue in order to get better pricing from the suppliers, from the foundry side. So all in all, I would say last year's gross margin will be a good benchmark. As for the operating income side, honestly, I don't know because the revenue change could be very, very significant. And as I mentioned, our operating leverage is quite good. So as long as we can have a very good revenue growth, you can see continuously operating margin improvement in the future, especially in 2026 and 2027.

Yes. Great. Thanks, Daniel. So yes, I truly believe that operating margin probably is a better way to measure your profitability. I also understand that. Yes, the big giant customer, $1 billion of revenue per project, I think margin could be fair. So since we are talking about this type of $1 billion or above revenue project. Can you give us some preview about the 3-nanometer life cycle revenue margin profile, whether that includes HBM or not? Are you comfortable to talk about 3-nanometer long-term revenue contribution right now?

Okay. Charlie, like I mention before, yes, based on the other people's market analysis, they've been very aggressive on the overall market, including ASIC, for 2027, people thinking about it will be USD 45 billion ASIC revenue. I also mentioned beginning of the speech, our customer has confidence to maintain or even increase the total weight about the ASIC revenue. So if you think about this USD 45 billion for ASIC revenue, you can quickly calculate the percentage of that particular customer that now pretty much will be the revenue target we have. But that's why Daniel mentioned about -- even though we have a very high cash flow, but we still worry about the future production. As you know, there is a half year gap between minimum half year gap, assuming capacity issue, we need to place the order before delivery. Yes. First of all, I'm very optimistic, yes, for this particular design.

Yes, that's a really helpful guidance in a way to think. So if you think about what we've just said, right? So USD 45 billion TAM in 2027. And assume that big customer a 10% global market share in ASIC, would that mean like USD 4.5 billion in 2027 per year?

I don't think that customer only take a 10%, more than that. I talk about -- but anyway, so yes, you are also asking a very good question about the HBM. Yes, we all know HBM is a commodity stuff. And I think it's kind of open. If you want to have HBM, you probably need to sacrifice more percentage gross margin. But if you consider about the net and also the top line, HBM can contribute some. But if you talk about the percentage, if we recognize more HBM revenue, I think the percentage will further go down. So this is kind of trade-off, which they put into consideration.

Okay. And the next one, Haas, please unmute your microphone.

So my first question is still back to the addressable market that you could capture in the AI market where you think it is going to grow to USD 45 billion and will be growing at a CAGR of 40% to 50% in the next few years. I was just wondering if you could share a little bit more color between GPU versus ASIC and within ASIC, which part of the segment you are seeing will be seeing the fastest growth or as a percentage of the total ASIC market, which part are you -- how much of the contribution is going to come from the high-performance computing or cloud AI-related chipset? I think this is my first question regarding the high level for your opportunity in the longer term?

Okay. Let me elaborate a little bit deeper. For the existing project, we do expect very significant production scale. It also represents the quite significant revenue contribution from this chip. As for the details about the production, does it include the HBM or not? It is still under discussion. HBM is a commodity. Actually, I think the customer doesn't care who handle it. They can handle it by themselves or we can handle the HBM and the customer pay us probably some handling fee for it. So I can say for now is everything is under discussion. And as for the scale in 2026 and 2027, I will say we expect year-on-year growth for 2027 as for the revenue for the project. And for the other opportunities among the cloud service providers marketplace, I will say, in addition to the accelerators, we do believe, first of all, the next-generation AI CPU could provide some opportunity to us because the next-generation AI CPU could apply the advanced packaging, which, as I mentioned, we consider the experience within the advanced packaging is one of our advantages. Secondly, we did a lot of preparation and we took several projects for networking application. The reason behind is we believe in the future, within the AI data center infrastructure, networking-related application could be the target by customers to do the ASIC. In order to be better prepared, we have to create the track record, the technology in order to compete for within this marketplace. So this area we consider is a correct direction for Alchip. So all in all, I would say the ASIC market, especially within the cloud service providers, AI ASIC market, the market size is expanding very quickly. And in the same time, the supplier -- the capable suppliers actually will be remain similar for maybe next couple of years. Like I said, Broadcom, Marvell, Alchip, MediaTek, probably GUC or maybe a little bit Social Next. We really think we have good position to compete for those increasing number of projects, no matter it is accelerator or CPU or networking chips.

Yes. Daniel. That's pretty clear. And I think just doing the math, on an annual basis, you can take pretty much like 30 projects per annum. And each of them, if we consider all of them migrating next couple of years?

No, I think the process node migration is gradually moving process. Some of the customers may move fast, some may move slower. But to us, we do see increasing demand from the 2-nanometer side because many of the customers show their interest in moving into 2-nanometer Process Node. However, like Johnny mentioned, for the 2-nanometer Process Node, it is a whole new ball game from FinFET to GAA and there could be a lot of chips applying 3D-IC or the I/O chiplet solution. So the whole size of the NRE, the value of an NRE will increase significantly by moving the process from 3 to 2.

Exactly. Yes. So that's what I mean is that the blended NRE content per project is going to be moving up very significantly. And I will assume that your NRE business, if you are able to still take 20 to 30 projects per annum on an annual basis, your NRE business opportunity in the long term could actually reach USD 1 billion to USD 1.5 billion approximately, which is the same scale as you are going to make this year on a total company basis. And plus you will have the optionality from the turnkey business, including the hyperscalers startup [indiscernible] . So I was just thinking if that is the correct way to think about that your recurring business is going to be this level like USD 1 billion to USD 1.5 billion going forward on annual basis driven by the NRE migration.

Haas, actually, your voice is breaking up. I can barely understand your meaning I'm trying to guess what you are asking. I guess is how to keep the NOI growth when the Process Node is moving and every project needs more design engineering capacity. Is that your question?

Similar.

Yes. And the other thing -- you can hear me clearly now? Sorry, I was just want to make sure.

No, your voice is breaking up.

Yes. I mean this place probably needs to improve its infrastructure on the cloud AI and also networking going forward. But I was just wondering if it is fair to assume that your NRE business run rate on an annual basis will be growing to USD 1 billion to USD 1.5 billion going forward as you are seeing more projects migrating to the advanced sales and each of the project will be significantly higher compared to the previous generation.

Yes. Okay, Haas I'm sorry to guess your question.

Okay.

Yes. I think your question is related to NRE growth rate. Yes, we are very optimistic for NRE growth momentum, just like you mentioned before.

Yes. It's okay.

Starting -- yes, if you compare the N3 and N2 NRE, I think the increasing rate will be very significant. Given the fact the mask itself is more expensive, N2, if they are using the chiplet solution, one design require multiple tape-out. And also, you're thinking about each of the mask need to do many simulation before tape-out, the hardware usage will be drastically increased. So that's also contribute a big NRE. Just for your information, before we're doing one N5 tape-out, maybe just require like a 200 server for N3, N2, the peak time may use over 1,000 server. Server is also very expensive. If you consider multiple tape-out on N2 and also including the hardware and also the design fee, the NRE increasing rate will be very, very big.

Yes. Yes. Can you hear me now?

Yes, go ahead.

Okay. Yes, that's good. So yes, I was just thinking because your NRE dollar content per project will grow significantly to the advanced nodes to the previous generation nodes. And if you have the capability to take on like 20 to 30 projects per annum, and each of the project will require multiple tape-outs, as you just mentioned, I would think if your NRE business run rate could actually be growing to USD 1 billion to USD 1.5 billion on an annual basis going forward versus what we have been seeing that your NRE business was good probably at around USD 300 million to USD 400 million in the past few years.

Yes. We are looking forward to break USD 1 billion range on NRE, but it may not -- it will not happen this year for sure because yes, we do have a capacity right now tapeout like a 20 to 30 design. But to be straight, if for N2, we may consider one design equivalent to 3 to 4 tape-out, like we did before. But because that one design, if it's a multiple mask needed, the resources will also increase a lot. And resource will become a limitation for us to take more N2 project. But to be honestly, N2 player is not too many. It's only a few company, either it's a CSP or well-established company or very, very full funding start-up may consider about N2. Otherwise, I think N2 is one -- if they want to do the one design, I think USD 200 million is kind of minimum, including everything.

Okay. Haas, I will suggest you to write down your question because your voice is breaking up severely.

If I can ask one more. If you can hear the question.

Okay, please.

Yes. For the 2-nanometer projects on the ASIC side, when do you see the earliest timing of the mask production from the industry perspective and also for your own business perspective?

Haas. Honestly, I cannot hear your question clearly. Can you write down your question?

Okay. Yes, I'll do that.

And Gabriel. Gabriel, please. You can unmute your microphone.

Can you hear me clear?

Yes.

I just have one question, and I just want to switch the gear to China and ADAS. As you know, in China right now, especially this year, I think BYD tried to push ADAS and intelligence to beverage per car. So I think the ADAS market definitely increased a lot. So are you seeing more inquiries from China about the ADAS ASIC chip projects? I know you guys are working with deal over right now. But is any other guys are seeing more interest from China to do that auto ADAS? And how do you think about the long-term TAM for this market? Are you seeing like overseas auto companies also trying to do their auto ADAS ASIC? And do we have the capacity to catch up that opportunity given we are very shorthanded in AI projects?

All right. Yes, I think because of a successful tape-out and also the chip working very well, thus presenting a very good example for -- open the door for us for ADAS-related business. But when we're taking the business from China, we need -- we will be cautious. ADAS, we consider only maybe 2 to 3 car makers will produce the volume we expect. So we consider we already won one of the home run, but we do have some plan to penetrate other car makers and I think we have a good expectation on the production revenue produced by this customer we already won. And also, they have a next generation will be kicked off very soon. And in addition to that, we did consider to take maybe one more. To be honestly, in terms of working resource in China, yes, we have more people, we have more resources to take a China business. And unfortunately, ADAS-related application right now, I think is not getting stuck by the major foundry or based on current regulation, we don't have -- we don't encounter any problem to take this project. Before I mentioned about this project is being slowed down by 2 to 3 months due to an incident. And fortunately, this restriction is already removed. After 3 months, the design can go to production now.

Right. So are you guys potentially working with like BYD or in China or like any other big names with volumes? And how do you see the long-term TAM of this market? Have you guys trying to...

Right. Yes. Like I mentioned before, if we want to take one more ADAS-related customer, you got to be big name. Unfortunately, we cannot share too much at this moment.

Yes. Gabriel for your question, I would say China is already the area that those carmakers are incentivized to do the ASIC. You can compare the China auto market and the non-China auto market. You barely see automakers outside of China trying to do the ASIC for their car. However, in China, I think the incentives are multiple for those automakers. First of all, they are competing fiercely. They knew they need to provide differentiation, the technology features to the customers. And therefore, it is an incentive for them to do the ASIC. To do an ASIC is very expensive, so we are expecting the top ranking automakers to do it. Obviously, the auto is a big fish and BYD and Nio and [indiscernible]. In China, we are relatively very confident once the customer tries to do an ASIC, at least they will talk to us because probably speaking, we are pretty famous in China market. So we don't consider it is a really good problem. As long as the projects out there, we will talk with the potential customers. And let's take a final question from Robert because of the time constraint. We still have other things needs to accomplish. So Robert, please.

Yes. I just want to go back to the key CSP customer. So I just want to understand the life cycle revenue for this 3-nanometer, maybe not the absolute number or maybe we can compare the revenue potential versus the 7-nanometer that you have been -- you are running, how size we are looking at? This is my first question.

Yes. Based on the generation, based on their trend and also the conference they present every other year, they will do a new design. Of course, if the current design performance and selling situation are better, the life will be longer in the opposite way, if we're not doing well, the lifetime cycle will be shorter. Our previous generation, I think it did in production over 2 years.

No, no, 3 years.

Over 2 years. So you're almost 3 years, I know. And yes, for N3, I have a high expectation. I mean current, I think, minimum 2 years.

And as for scale, I knew that you all want to know the scale of the revenue. I would say -- I would suggest to think this way. The trajectory of the generation by generation projects of this customer will apply the same trend to the 3-nanometer project. Okay. Please?

Yes. And I also want to move to the next generation after 3-nano. Will it migrate to 2-nanometer or is it possible to stay on 3-nanometer. And what's the status of this project? Has the customer started the RTL design already, and when will the RFQ be issued and decision -- the vendor decision to be made? And Johnny, you talked multiple times about the chipset design. Will we see multiple suppliers on this next generation project at 3-nano or you think that even if the win is so weak?

Robert, you are going too deep.

So the architecture portion, I think customer is not finalized yet for sure, their front-end team is working on this. And we will -- they will make a vendor selection for sure within this year. So a lot of homework we need to do, and we are working with them with a different variety of topology, I think because I consider about the yield, consider about the cost, considering about the potential risk, some possibility go to chiplet and some possibility remain homogeneous. I think we are working with this particular customer closely.

It's okay, thanks. Haas to write a question.

Okay, sure.

First, on 2-nanometer, when would you expect the earliest timing of production contribution. Second, if we did -- if we did not take it wrong, your sales this year will drop by 10% to 20% this year, but earnings will be flat. Could you provide more details on the NRE production mix and your target gross margin this year. For your first question for the 2-nanometer project, I would say what we can say right now is we are highly confident that we can maintain the partnership, relationship with the customer. We always want to do generation-by-generation projects with our customers. And there are some signals about the possibility of winning or losing. For now, the whole situation is pretty good. And secondly, yes, our revenue, we are expecting to drop by 10% to 20% this year 2025. However, as I mentioned, with the improving gross margin and coupled with our operation gearing, we don't think the net income, the net profit will go down in the same range with our revenue, which means the net profit will be go down like less than 10%. That's what we are assuming right now. But for now, it's only February, and you are asking me about almost all the P&L items. I'm sorry that I cannot give you a very firm answer. I can just tell you the trend and the direction and for the NRE and the production mix, I would say, for now, I am expecting the NRE to account for like something 30% -- around 30% of our total revenue this year. And I guess that's it. Thank you all for your participation in our fourth quarter '24 earnings call. And many of you, I will see you soon. And most of you will see me and Johnny in the next earnings call probably within 2 months or 3 months. Thank you.

All right. Thank you very much.

Thank you.