Okay, as usual, this is a safe harbor disclaimer and next page, so this meeting will be in English. If you need Chinese presentation slides, please go to the MOPS, [Foreign language], to download the Chinese version, and Johnny and I are all mandatory speakers. If you are asking, if you want to ask questions in Chinese, you are so welcome, and for the participants, please write down your questions. If you don't want to ask a question orally, you can write down your question through the team's message function, and if you are going to ask questions during the Q&A session, please use the raise hand function of the team. This video and the audio content for this meeting will upload to MOPS about three hours later after the meeting, so we welcome our CEO, Johnny Shen, to start the meeting. Okay, good afternoon, ladies and gentlemen.
This is Johnny Shen, President and CEO of Alchip Technologies. Yeah, once again, thank you for joining the investor conference meeting. Yeah, we appreciate the opportunity to share our Q3 results and provide guidance for future business outlook. Let me briefly update our company again. The company is founded in 2003, IPO in 2014. The current market cap is approximately $5 billion. Since we founded the company, we've been successfully implementing more than 600 designs. Among 600, 60 of them are FinFET, and 18 of them are CoWoS related designs. Our current employees are about 600 people. Revenue, last year, TWD 978 million. So this year, the first three quarters, we have achieved more than TWD 1.1 billion already. 90% of our revenue is coming from HPC and AI. We are TSMC 3D Fabric and the VCA member, and our current capacity is that we've been tape-out.
We can handle 20 to 30 tape-outs annually. Our market focus is HPC, networking, and automotive. Okay, allow me to give everybody a Q3 update. For Q3, we are pleased to announce another record-breaking quarter, with revenue reaching TWD 460 million, net income TWD 55.6 million, and EPS standing at TWD 22.46. All these numbers are historical highs for the company. The detailed breakdown and comparison will be presented by CFO Daniel in the later section. For future business outlook, Q4 is a peak season for NRE. We plan to tape out several designs this quarter. We have also secured many design wins across AI, networking, or cryptocurrency applications, mainly in the North America region, leveraging the leading-edge technology at N5, N4, and N3. So in terms of mass production, demand remains strong. Even with some adjustments, the revenue outlooks remain reasonable.
We anticipate only a single-digit reduction in top-line revenue in near-term quarters. However, thanks to high NRE forecasts, we expect earnings growth and also improvement in percentage gross margin as well. In terms of geopolitical risk management, we've been successfully diversifying our business beyond China to other regions. In 2024, only around 10% of revenue originated from the China region. However, our business direction in China remains unchanged. We continuously believe in and support businesses in China as long as they are financially healthy and comply with rules and regulations. To emphasize again, Alchip has not conducted any business directly or indirectly with any company on the Entity List. All our customers are bound by three-way NDA among Alchip, foundry partners and customers themselves. DD and KYC, due diligence, and know-your-customer processes are strictly enforced before engaging in any business.
In terms of workforce, we have initiated a very aggressive hiring plan to bolster our engineering resources outside of China, such as Japan, Taiwan, Malaysia, and Vietnam. Currently, we have 100 engineers working in Japan. Our Malaysia office already staffed more than 25 engineers. The workforce in Southeast Asia is planned to be approximately 60 people by the end of this year. This strategic expansion plan is a desire to offer more flexible and cost-effective solutions to meet our customers' dynamic requirements in the global market. As for conclusion, we are confident that our business is in an excellent stage. Likely, Q4 is going to be another earnings-breaking quarter. Overall, we anticipate a strong completion for this year. Thank you very much.
Okay, here are the quarter-detailed P&L numbers. The total revenue, as you may know, reached TWD 459.7 million, which is a 9.2% quarter-on-quarter growth and a 91.1% year-on-year growth.
As for the operating income, we recorded TWD 58.4 million for the Q3 , which is a 14.1% quarter-on-quarter growth and an 82.5% year-on-year growth. And for the net income, the net income for the Q3 is TWD 55.6 million, which is a 12.8% quarter-on-quarter and a 97.7% year-on-year growth. For the EPS, translating into TWD, it is TWD 22.46 for our Q3 earnings per share. Next page. Here is our breakdown by application. You can see that for the Q3 , HPC dominates our quarterly revenue breakdown. It accounts for 97% of our total revenue. The rest, the consumer networking niche, accounts for about 3% of our total revenue, which indicates very strong demand from the HPC, especially AI side. For this year, Q1 to Q3 , the HPC-related revenue accounted for 94% of our total revenue, and the other three applications combined account for about 6%.
Next page. And for the process note, again, I'm very proudly saying that among our peers, Alchip should be the leader for the process node technology. In the Q3 , for the 7 nm, it accounts for 73% of our total revenue. And if you combine the 7 nm, 5 nm, and the more advanced process nodes, combined accounts for 96% of our total revenue. And for this year, for 7 nm and the more advanced process nodes, which accounts for the revenue related to the nodes, accounted for 97% of our total revenue. Next page. This is the geographic breakdown. For the Q3 , the North American revenue accounted for 87% of our total revenue. And here is a reminder that starting from the Q3 , one of our major customers shipped from the European account to the North American account.
You can see the difference is mainly from the shipping of the customers' categorization. For this year, North America dominates our revenue breakdown, accounted for 85% of our total revenue from the Q1 to the Q3 , thanks to the strong shipment to the AI-related ASIC or chip to our customers. For the business review, our CEO just mentioned we had record high quarterly revenue and profits. The reason behind the strong revenue was because, first of all, the strong shipment of our 7 nm AI ASIC actually exceeded our previous estimation, which is the major contributor to our record high revenue. Secondly, the ramping up of the 5 nm AI accelerator to our North America IDM customers is another driver for the Q3 revenue growth.
For the profit side, profit hits the record high, and the gross margin slightly improved from what we had in the Q2 . The Q3 gross margin was 19.5%, improved mildly from the Q2 . The slightly gross margin improvement attributes to, first of all, we have higher NRE revenue numbers than the Q2 . For the production mix, I'm not saying the revenue mix, I'm saying the production mix, by adding the 5 nm AI accelerator business, which enjoys slightly better margins than our 7 nm AI ASIC business. We consider these two factors as the main reason for our gross margin improvement. The non-operating income came mainly from the interest income. We don't have too many complicated investments or the other FX risk exposures.
So the numbers on the non-operating income side are mainly the interest income we get from our deposit and the fixed income investments. Next page. And for the business outlook, I will do it very quick, and I believe many of you will ask related questions regarding the outlook. First of all, for the Q4 , we believe the Q4 revenue will head up slightly. Given, first of all, the 7 nm AI ASIC shipment is expected to go down slightly in the Q4 , and the 5 nm AI accelerator shipment to keep on ramping up. But by combining the decline and the ramping up, we expect the production revenue to go down slightly in the Q4 compared to the Q3 this year.
Our CEO mentioned the NRE, or they say design revenue, will be pretty strong in the Q4 , which means the profit margin, in general, will be higher than what we had in the Q3 . And for the next year, I know many of the investors paid a lot of attention to our 2025 outlook. For now, briefly, for the first half of 2025, we will consider it stable. And for next year as a whole, we see very strong NRE project pipeline coming in because of, first of all, the process node technology migration is still ongoing. We expect many 3 nm projects to kick off in late Q4 this year or early quarter next year.
Starting from the Q1 next year, 2025, the 5 nm AI chip to the North American IDMs will become the major revenue contributor to Alchip, which means it will replace the 7 nm AI ASIC as the number one revenue contributor for this chip. Because we expect strong revenue growth in 2025, for now, I will say the profit margin, especially the gross margin side, will reasonably improve year on year. That's the session from the company for today's meeting. We will go into the next session, Q&A. As I mentioned, if you want to ask questions, you can go through the message board or raise your hand, and we will unmute you for the questions. Thank you. Hans, please. Hans, can you hear me? Hi. Hans. Yeah. I can hear you.
Okay. Okay. Yes. Thanks, Johnny and Daniel, for the detailed remarks. Thanks for taking my questions. I guess to start off, I was wondering, based on your guidance, that revenue will drop by single digits quarter on quarter. Could you clarify if it is for the whole company or just for the turnkey business? I was wondering how we should think about your NRE business when you talk about the tape-out revenue contribution in Q4 across 5 nm to 3 nm. I have a follow-up question on Q4 . Thank you. Okay.
First of all, for the Q4 , as we mentioned, the NRE revenue will grow seasonal demand, which is pretty strong, which means the production revenue is actually declined in percentage more than the total revenue decline in percentage. Right? So what we guided that the slightly, or they say probably high single digit quarter on quarter decline, it means the total revenue.
Okay. Yes. That's pretty clear. And then I think just based on the business assumptions you provided, could you share your view on the revenue mix between NRE and also production in Q4 ? And while we should think about gross margins, will you be back to 20% plus levels in Q4 ? And operating margins, will you actually be higher quarter on quarter?
Thanks. Hass, this is where we cannot give you numbers in the conference call due to the regulation. But since you can see our Q3 gross margin is 19.5%. So if we will have improvement from this margin, it, of course, means above 20%. Right?
Okay. Yeah. But I mean that if you can provide a rough breakdown between your NRE mix versus turnkey for Q3 and also Q4 . Thanks.
I will say the NRE revenue likely accounts for 20%-30% of our total revenue in Q4 .
Okay. In Q4 . And then in Q3 , was it around high teens to low 20%? We could assume that.
Around 20%.
Okay. Yeah. And then I think shifting gear to 2025 outlook, it seems that you are still pretty confident on the U.S. IDM customer contribution, which is somewhat in contradiction with your customers' comment as well as the news reporting recently. So could you provide us some more updates on that part of the business, especially considering that for the second half of this year, the ramp seems to be weaker than expected? What gives you the confidence that you will still be the major contributor next year? Thank you.
Okay. I'll answer first, and Johnny will do a more detailed answer later. For the next year, I knew that there were a lot of rumors or sayings about the IDM customer to cut the orders. It is not 100% true. However, we do see that the customer to trim their 2025 demand. And some of the orders are right now putting on hold. But for the rumoring like 40%, 50%, 70%, or even cut orders, it is not true. Based on the current orders POs to TSMC, and which I indicate it is the worst-case scenario, we will still have pretty decent revenue contribution from this customer. And that's why I said for next year, for the production or even for the overall revenue, this chip, this 5 nm AI chip will become the number one revenue contributor to us.
Okay. Let me try to add more. Yeah. To be honest, yes, we did receive some instruction to put a certain amount of the order on hold. Yeah. Hopefully, the customer can recover. But maybe there's still the inventory level goes up. So some amount is just put on hold. But the current situation, we do some detailed calculation. I think overall, the demand in hand is not that bad. But of course, from this Earnings Call to previous Earnings Call, I think things change very dynamic. If I remember right, we have a very high confidence to further grow our company to a certain percentage, 20%-30%. But right now, I think the situation changed a bit. I can say we still have a chance to grow our company, the entire revenue. But overall, like I mentioned before, the NRE is quite promising next year.
So we have a higher confidence to further grow our company in terms of earning, in terms of revenue. Yeah. Hopefully, we can be flat or a little bit better than this year.
Okay. Yes.
That's all.
Yeah. If the order continually put on hold. Okay. Hass, please leave some opportunity to others. We will circle back to you later.
Okay? Okay. Sure. Thank you.
Thank you, Hass. And Robert, please. Robert, you.
Yeah. Daniel, Johnny, can you hear me?
Yeah.
Yeah. Thanks. So I want to follow up on this IDM project. So I think you previously talked about if things go smooth, it can be as big as the 7 nm US CSP project revenue size this year. So in the better case scenario, what would be the revenue contribution now after this round of order cuts? And similarly, for this project into 2026, how should we think about the revenue trajectory? Because the AI product roadmap from this customer appears to be pretty unclear, and some of the customers are holding back on adopting this program.
I'll say it will be the same. I guess most people understand that we are under strict NDA, so what we can say is for the AI accelerator market, we do secure some projects, and one of the most important projects is the 3 nm projects we are working on right now, and for this project, the schedule is so far so good. Hopefully, and we have confidence that this 3 nm AI accelerator will likely go into mass production in very early 2026. That's what we can deliver to you the most.
Oh, sorry. I was asking about the U.S. IDM project.
Okay. For US IDM project, I will say 5 nm, as I mentioned, is the major revenue contributor to us next year. And for the project beyond that, the same thing. I will say the next generation to the current 5 nm AI chip, supposedly 3 nm, if the project stays in TSMC, we have pretty good chance to still be the ASIC partner with this customer. But if the customer, as you may know, that it is an IDM player, if the customer wants to bring the project back into their own fab, I will say the chance for us to support the project will be smaller. But for now, in our view, we see if a 3 nm AI chip wants to be competitive in the market, TSMC is a better choice.
Got it. Got it. And moving on to the US CSP, the 3 nm. I think you shared some likes. And could you help us understand the nature of this project? Do you think it's still only one winner taking the project, or it could eventually be the split between you and the other competitor? Because the reason I'm asking is about your comparison to be provocative about retaining the project after what they have on the 5 nm right now. So can you help us understand what's the nature of this?
Robert, if I answer the question, which means I am indicating somebody else's business, right? No. Do you think it's still you winning all? Is that still you? It will be 3 nm project, as I mentioned. But I cannot confirm with you we are the only one or we are one of the multiple. I cannot do that.
Okay. Okay. Yeah. I have a final one on your NRE. So I think, yeah. So first, congratulations on your recent 3 nm project win with the AI startup, I think, last week. So looking into 2025, how should we think about the NRE growth? I think you mentioned it could be strong. And could it be above the trend that you guided before, like 20%-30%? Can it be above this number?
Yes. Let me try to cover that. So technology migration, I think, is still ongoing. Yeah. Recently, I see before majority of our design are N7 or N5. Now, the new RFQ we receive is mainly from N3. Yeah. As you know, the N3 NRE, in terms of everything, is much, much higher than N5. So I think based on current demand and also we calculate the NRE number, the next year, the NRE growth is quite significant.
Yeah.
Some customers, for example, the biggest customers, they're even thinking about beyond N3. They are thinking about N2 now. Yeah. In terms of NRE, in terms of resource, I think the cost for design will be much higher than before. Even we maintain the same capacity or a little bit less, the NRE increase is quite significant.
Okay. Thanks, Johnny. Thanks, Daniel.
Okay. All right. And next one, Charlie, please. Charlie Chen, Morgan Stanley, I guess. Charlie, you got the news. Oh, yes.
Okay. Yeah. Thanks for taking my question. And also, congrats for very, very good Q1 outlook. Yeah. So my first question is a little bit mid to long-term outlook. Right? First of all, to engage with the future 3 nm or 2 nm project, do you think you need to consider adding some front-end design engineers or there's a kind of CTO level of people? Would you consider to do some spec'ing type of partnership with your customers?
Okay. Yeah. Let me try to answer this. So for front-end design, that would be categorized into many fields. Yeah. Using our number one customer's example, they don't need any front-end support at all. But for other CSP, yeah, potentially like Google, Meta, they may need some front-end support from their side. But mainly on the subsystem side. Yeah. For example, using the UCIe HBM D2D interface, they need some help on the subsystem to ripping out the IP, do the verification, and even do the system level bring in. Yeah. For that area, we do some investment working with a close partner. And also, we also are working with the front-end partner. But internally, we don't have too much intention to build a front-end team. Yeah.
Because the front-end, as I mentioned before, that involves architecture, RTL, and that will infringe the neutral position. I don't want to touch customer's architecture. I don't want to get involved for their spec design. That's why I can take many projects.
Yeah.
If we involve too deep, later on, I think that our neutral position, I think, will have some challenge on that portion. So to answer you again, some customers do need some front-end support on the subsystem. We have a partner. We're working with a partner closely.
Yeah.
Try to take this kind of project.
Gotcha. Thanks. And I know you kind of have some front-end design partners in China, right? For example, for China Auto project. But for the U.S. project, do you consider to work with some other defensive company like just randomly? I didn't hear anything like Alphaw ave. It seems like also one of TSMC's growing important partner. Do you think that kind of collaboration with some U.S. company for front-end design is possible?
I see. Yeah. We don't, again, yeah, in China, we have a partner. In Singapore, we have a partner. In U.S., I think Synopsys is our front-end partner as well. And also, they're doing a lot of subsystem with us already. Yeah. The company you mentioned about, yeah, we don't, again, actually, personally, I also have a tight relationship with them. Yeah. If a business is making sense, I don't mind to work with them as well.
I see. Thank you, and my second question is about a little bit details about your 3 nm project wins. So of course, you have one major focus, right? But can you give us more detail about the numbers of 3 nm project wins? Who are those companies? And I guess investors are very, very clear whether or when you can win another CSP customer. Can you comment on those fronts?
Okay. For the most important project we are having right now, and I will say the earliest one for getting into the mass production and in large scale is the 3 nm AI accelerator to the CSPs in North America. In addition to that, we have multiple 3 nm project wins from North American customers, such as networking customers, who is also a listed quite scalable company in North America and a major networking equipment chip players in North America and Europe. So there are multiple opportunities. To be honest, for the pipeline, it's quite fruitful for now from the North American, no matter if it is AI or networking or the networking used in AI data center.
For the other CSPs, we are currently engaging with discussion or, let's say, compete for a couple of projects with the industry peers. Of course, the strongest competitor is Broadcom. So we do have winning chance. But right now, I would like to be conservative by saying if we want some projects, we will talk to the investors. But for now, we are competing.
Yeah. As you know, most of our CSP projects in North America are dominated by Broadcom. Yeah. Broadcom is a very respectable competitor. But somehow, most of our customers do have some concern,
yeah,
to only use Broadcom. So they are open to a lot of opportunity. Before, it's very difficult for a company like us receiving RFQ from the CSP. But right now, I think whatever they have a project ready, yeah, we have a chance to compete.
I can say that. I see. Thank you. It's perfect. So I will be back to the queue. If I can answer some second-round question, I will do it later. Thank you.
Okay. Next one is Chao Ren Chen. Please unmute. Chao Ren Chen, are you there? Can you hear us? Yeah. Please.
Yeah. Sorry. So I had a couple of questions, some of which you kind of already answered. But just to go through those. So first, you said for the U.S. IDM customer, if their 3 nm project is still done at TSMC, there is a good chance you will participate in that, correct? And to be competitive, most likely, they would do that at TSMC rather than insourcing that again.
That's correct.
My understanding is right, right? And that's scheduled for late 2025. Is that correct?
That's our view. That's not the view. That's not the view of our customer. That's our view.
My second question, which Charlie was touching upon, is the new 3 nm or even, I don't know, 2 nm customers. You mentioned you're competing with the other designers for the CSPs. You mentioned the networking customers. Is that going to be a switch chip? When you talk about networking customers, are you working with them on their networking switch chips, or what would that be specifically? It's not an AI accelerator, correct?
Okay. Let me first clarify that. First of all, what we are saying, we are competing with projects from CSP, which means the non-existing customer. And for the existing customer, we will do the projects generation by generation. And for the networking projects, I will let Johnny to elaborate it a little bit.
But to be honest, because all the big customers, especially the big US customers, the confidentiality requirements are quite high. I understand. I understand. Yeah. Yeah. So there are many opportunities, I think, for CSPs. Our team is very busy to prepare the proposal and working with them. Yes. One of them is networking switch. Yes. You are right.
Okay. Got it. So if I look at 2026, 2027, who is likely to be the third or, I don't know, could be second largest customer? Would it be a second CSP? Would it be the networking partner? Or could it be, I don't know, some auto guy, right? Because I don't want to name names here, but could it be a Chinese auto company? Could it be a US auto company? Who's likely to be the next big customer after our US CSP, existing CSP, and existing IDM? It sounds like it's going to be a networking customer.
If we are talking about 2026, I would say if we win, if we win the networking project, the contribution from the chip could be quite significant.
Okay. Right. Sure. Yeah.
They have a chance to be top three, I would say.
Okay. Okay. Got it. Yeah.
For next year, I think IDM will be number one customer. Number two, pretty much, will be original number one. Then number three, I think many customers have a chance to be that spot.
Okay. Understood. And my last question, if you don't mind. So regarding the old number one, next year's number two, for their 2026, when will we know with higher certainty? So when can we know that they've made the decision on their next generation? When do we get more visibility on their next-gen AI ASIC on 2026? Whether it will be done with us or with another competitor?
Sorry. The next generation, you are meaning 2 nm?
Yes. Yes.
I would say very soon. Okay. Yeah. The design will be kicked off, yeah, for sure, will be in 2025.
Okay. But so 3 nm, just to be clear here, I may have some confusion. There are 3 nm went to another competitor, right? And then we're trying to win back the 2 nm. And you're telling me sorry. Yeah.
No, we are doing the 3 nm AI accelerator right now. And we are expecting the chip to be in mass production in early 2026.
Okay. Got it. So they're 5 nm. Okay. So last year was 7 nm. 5 went to another guy. Okay. So would it be too early to ask your view on 2026, now that you know that 3 nm is kicking off in 2026?
A little bit too early, but I can assure you that it is a very promising year for Alchip.
Understood. On the back of all those we just discussed, right? The IDM customer, your view is if they want to be competitive, they should do TSMC partnering with us. So on top of that, on top of us starting the 3 nm with the big CSP customer and future, not future, but potentially more networking projects, potentially to add one or two more CSPs and other stuff, right? So 2026 looks promising.
Yeah. I would say everything looks good in 2026. Even without new winning, the 2026 will be very promising. But we will continuously try to win more customers.
I see. I see. I see. I see. Understood. Understood. Okay. So we should really look past 2025. Understood. Okay. Thank you. That's all my questions. Thank you so much.
Thank you. And the next one is KGI Lucas. Please unmute.
Hello. Can you hear me?
Yes.
Okay. Thank you for taking my question. I have one question on your contract liabilities. I have noticed that your contract liabilities started from TWD 3.6 billion in Q2 .
Lucas. Lucas. Your voice is breaking up. We cannot understand you.
There now? Yes. Oh, much better. Much better. Okay. So my first question, I have noticed that your contract liabilities have surged around TWD 7 billion in this year. And could you clarify how much of this increase is attributed to the production revenue commitment for the U.S. IDM customers and five AI accelerator projects for next year? Thank you.
Can you repeat your question again in terms of revenue breakdown?
Yeah. No, go ahead. It's about the contract liabilities because the company has recognized a huge amount of contract liabilities this year.
Okay. The contract liability is mainly for the shipment for the five nm AI chip to the IDM customer. So starting from now, or this is as the shipment is ramping up, the contract liability will gradually decrease when we ship the chip to customer.
Got it. So is this commitment or provision, as we may refer to it, noncancelable?
Yes, it is not cancelable. Got it. Okay. So if the demand of the N5 AI accelerator were to fall maybe below customers' expectation and production were reduced, would the revenue recognition be based on the proportion of the production, or is it still noncancelable?
So this is all the contract liabilities we recognized earlier can be transferred to our revenue. Okay. The basics of our business model is doing the service business. So actually, the chip is only for the customer, and it can be used by the customer only. So almost every contract, the customer is mandated to take the chips.
Sure. Got it. So I only have one follow-up question for the 3 nm AI ASIC from the U.S. CSP customers that we have won. Are there still two packaging versions that utilize the same top die? And maybe another way to ask, have we secured both the training and inference projects?
Sorry, I cannot answer you this question. It's related to the chip design. Yeah. As I said, we are under strict NDA. We cannot disclose the details about the chip.
Okay. Got it. That's all my question. I'll be back to the queue. Thank you.
I guess from Macquarie, please unmute your speaker.
Great. Thanks, Johnny. Daniel. First question. Your current large CSP customer, your mass production volume was very high for, I would say, over two years. Do you think the 3 nm project, when it ramps up, will also be that long? Or do you think the projects will get shorter mass production life cycles in the future?
Yeah. In terms of revenue number, N3 definitely will be much higher than N7. But in terms of production cycle, I do some calculation, at minimum , like one and a half year. I think average maybe two years.
Okay. And second question, any update on your major ADAS customer for next year in terms of ramp-up and how that's going?
Yeah. Everything goes smoothly. We'll be on schedule. We will start the shipment in the Q3 next year.
Okay. Thank you. And my last question, your IDM customer, assuming their next generation ramps up roughly what you think, how much mass production timeframe do you think the 5 nm will be? If it's just ramping up from Q3 this year, do you think that will actually have revenues in 2026 on mass production or close to end of life?
Yeah. Like I mentioned before, yeah, there's a certain order put on hold. But before, we expect to see the production for the whole year. But I don't know. They still have a chance to restrain the holding. Yeah. Before, we were thinking about this production well less than two years because their next generation is coming late. So we expect without the big issue, the lifetime for this particular project will be longer.
Okay. Great. Thank you very much.
You're welcome. Okay. I guess it circles back to us. Can you unmute your speaker?
Okay. Yes. Can you hear me now? Yes. Okay. I think just two quick follow-ups for the NRE versus turnkey mix next year. Since this year, it is pretty much like 20% from NRE, and not until Q4 this year, it is going to be back to 20%-30% of your total sales. What should we think about the NRE mix as percentage of your total sales looking into 2025? And what is your initial view or take for the NRE contribution as percentage of your sales in 2026, if you could share based on your current forecast?
Thank you. Okay. I'm lazy. I haven't done the budgeting, detailed budgeting. But based on the current picture, I would say the next year, NRE will account for about 20%-30% of our total revenue. The NRE revenue, 20%-30%. Yeah. So based on my calculation, I didn't do very detailed as well. But the next year, in terms of NRE weight, yeah, will be higher than this year. But we don't know about 2026. As you know, 2026 production revenue ramping up very quick, even though NRE also ramping up quick. But I'm still thinking about the production improvement will be faster than NRE. That's for sure.
Sure. Sure. And then regarding your IDM customers' business contribution next year, you discussed about the contract liability will go down, quote-unquote, "throughout 2025" when the chipset starts the shipments. But I was just wondering if the customer can actually delay the chipset shipments, and that will actually push all your revenue contribution schedule as well. Thank you.
If we are talking about the possibility, I would say anything is possible. But to us, to the supplier, as a supplier, and for TSMC as a foundry partner, I will say everyone wants the shipment to be on schedule or even to accelerate the shipment for the current project. Because TSMC is running a very tight schedule for its wafer capacity, especially for the 5 nm and the CoWoS capacity. We want it to be on schedule.
Yeah. Let me also add some input for this. Whatever the order we are receiving from this customer, the wafer process already happened. So there's a separate way to put on hold. In usual case, people do some functional changes to put it on the contact level and waiting for the metal change, and most of people just process the order way, so everything is in wafer process now. If they put anything on hold, there will be consequences. Put a hold on contact, there will be storage fee. Put beyond metal is useless because most of the major cost is on the device, so if the wafer is already in the metal process, most of the customers will decide to take the inventory part.
Yeah, because not much money they can get it back, even they cancel the order, so everything is in the wafer process. I think the chance for them to cancel the entire order, I think, will not be too high, and given the fact, they are working very hard to get the CoWoS capacity from TSMC. Yeah, last year, both Alchip and customers are working very hard to get this capacity. I think this is very precious. I believe and hope everything will proceed accordingly.
Okay. Yes. And I think lastly, with my follow-up questions, I think just regarding the CoWoS capacity that you have been securing for 2025, how should we think about the scale compared with 2024? And for the next 3 nm hyperscaler project you are going to about to ramp from late 2025 or early 2026, what should we think about the technology of CoWoS that the chiplet will be adopting and whether you are actually already negotiating with the foundry supply chain regarding that project ramp? Thank you.
Actually, we already have projects in production that is chiplet. So if you are asking me about the chiplet technology, sure, we will keep on supporting customers who is the needs for that. Looking forward, I would say most of the advanced AI accelerators will be in chiplet technology. What's your other question?
Yeah. I was just wondering, compared to 2024, for the CoWoS supply you secure from your foundry partners, could you tell me a little bit about the stronger trend for 2025, even with your IDM customers putting some of the orders on hold? Thank you.
Oh, okay. For the North American CSP customer, I will say it's really special, and TSMC gives their best support to the customers. So for the future capacity, although the scale of the production will be huge in 2026, we are confident, and the customers are confident to get very good support from TSMC's CoWoS capacity. Right.
Yeah. For 2025, so far, I think we have much, much less constraint from CoWoS than 2024. But 2026 will be another challenging year for us because of volume increase a lot. Many customers ready for production.
Okay. Yeah. Thank you so much. That's all my follow-ups. I'll be back in the queue.
All right. Thank you, Hoss. Okay. Robert.
Hey, Daniel. Thanks. So I just have a follow-up on this IDM customer. So how much percentage does the customer have to prepay? And if I look at your contract liability, I think it's around TWD 300 million. And then if we assume probably 70% of that, or you asked for 70% of prepayment, should we assume that this project should contribute around TWD 500 million or so next year?
Robert, we cannot talk about the numbers because it's related to our customer. So
I just want to know how much is actually in the pocket and how much is up in the air?
How much is in our pocket? You can see the contract liability. Based on that amount, you still have to put something more because that's for wafer and some others. To finish your chip, they will still need some money to do the testing, something like the old testing.
Yeah. Yeah. Let me try to answer this way. We already started production for this particular customer for a while. Yeah. So far, everything they are paying, the credit is so good. They are paying everything on time. Yeah. I don't think that's a yeah. Personally, I also do a lot of channel checking on their funding situation. I think so far, I get a positive feedback. The total amount of number compared to the whole company, I think it's not that significant.
Okay. Got it. Got it. And secondly, I think it sounds pretty promising on the networking ASIC, especially for the switch. So I just want to know the customer's thought process. So what has prompted them to suddenly move toward this customer ASIC model for the switch IC? Is that to get a better cost or better performance? And how should we think about the SerDes IP support? Because I think the incumbent, one of the key core components is the SerDes.
Robert, it is a complicated question for the consideration of why the customer is choosing us as the silicon supplier. There are multiple considerations. The engineering.
Oh, sorry. Sorry. Sorry, Daniel. I was asking about the networking ASIC. So what are the.
Yes. Yes. Yes. The same thing for the networking ASIC. I would say the logic is pretty much the same with the ASIC, the CSPs. Let's assume this way. If the networking chip is used in the AI data center and the so-called turnkey solution providers need to charge a lot of the margin with very high price, probably the customer will consider to look for other suppliers with more flexible pricing and more flexible engineering support. Given the potential shipment volume could be huge, 1%-2% difference in cost is a significant absolute amount of money.
Yeah. Talking about a specific IP like SerDes, yeah, of course, Broadcom has a good reputation on that area. I think, again, it's a very respectful company. Yeah, but right now, many other companies have a SerDes solution, and some of them, for example, Synopsys has been proven for different generations from our number one customer. And also, many other IP providers have this kind of solution. Yeah.
We're working with most of them to try to provide the most competitive solution, yeah, to minimize their Broadcom dependency.
Got it. Got it. Yeah. That's all I have. Thanks, John. Thanks, Daniel.
All right. Thank you. And Charlie, yeah, please.
Thanks for taking my follow-up question. So we talked about the worst-case scenario for the US IDM customer. But what is the base case then? Meaning, if based on the current TSMC's wafer orders, what would be the implied 2025 revenue growth?
Yeah. It's difficult to. I always say the revenue from this customer will hopefully account for still over one-third of all total revenue next year.
That's a base case, right?
That's the base case.
Okay. Okay. Yeah. That's fair. Yeah.
Because I totally agree with you that it is not ideal for their customers to cancel wafer orders at TSMC if they want to pursue sort of a long-term partnership, right? And there are not many kind of valuable assets for that group. So yeah, I think it's good to manage investment situations, but it's also good to share what is the base case. So that's very, very helpful. And another follow-up is really the 3 nm NRE. Do you think 2025, we are going to see any sort of turnkey revenue from those 3 nm projects other than a major CSP? 3 nm production, let me see. Yeah. Because our production number is very high. It's dictated by only few accounts. But even though some projects go to the prototyping stage, or I think they start to produce some MP revenue, but compared to overall number, it's still insignificant.
Yeah. I believe because some of the projects will be tape-out, whatever the project tape-out before the first half, we expect to see some prototyping revenue, production revenue on Q4. There will be a few on the pipeline, yes. But it's not as significant as other major projects. Okay. So I would assume those 3 nm projects outside of the major CSP should be more 2026 revenue contribution as well. It will be 2026. Yes.
Okay. Would those also need CoWoS capacity?
Yeah. If it's AI-related, yeah, definitely they need CoWoS capacity. Yeah. We do have a few, many startup AI accelerator customers. They need CoWoS capacity. Yes.
Oh, okay. How about the so-called switch networking ASIC? Would they need the CoWoS as well?
No. That one is not. They don't need CoWoS.
Oh, I see. Thank you. And last one is the geopolitical risk. I know you kind of handle this very carefully. But recently, you said the news, right, about your foundry partner and some China company. It's kind of escalated. So I'm wondering if your product to China customers incidentally used by companies in the Entity List would that still impact your business or operation in China?
Yeah. I would say no. First of all, our China exposure is currently really low. You can imagine this way. Let's take the recent news as an example. In order to cover Huawei, not too many companies can be the so-called agents to do that. Secondly, we do know because we have been in China for many, many years, actually, we knew the customers and the market. So I would say it doesn't make sense for us to take a risk to be so-called the agent, as an Entity List customer. Yeah. It doesn't worth it. Right.
The overall revenue from China is insignificant. I think looking forward in the future, our emphasis point is auto. I think the automotive right now is, I think, still politically insensitive at this moment.
Yeah. Got it. And last one, are you comfortable to talk about your potential or existing Japan consumer opportunity or projects?
We do have a good win in Japan. And the production will be in 2025. But to be honest, it is not within the HPC area. So the contribution for the revenue scale as we have currently, the revenue contribution is, I won't say not good. I will say how to describe it. It's good contribution from a consumer project, but cannot compare to the scale of the AI accelerators. Yeah. Definitely will be number one customer in Japan.
Yeah. I know your company is very, very successful in AI, US, maybe China auto. But why are you winning a Japan consumer project? And it seems like it was handled by another design service. So can you give us some background why you are winning this kind of a project?
I think this customer used to be our dream customer. Yeah. We've been continuously working with them. Finally, they grant the project to us.
Yeah.
And we do the similar application like 10 years ago.
Yeah.
We have some experience. And during that time, that was our number one customer. So we have a base in that area. But anyway, right now, the revenue contribution is reasonable, but it's not good enough to further grow our company in terms of a significant percentage.
Okay. Okay. Got it.
But you said very good customer. Very good customer.
Okay. Congrats. And famous. Yeah. Congrats on day one. Thank you.
Okay. Thank you. And we have. It's about time. So we want to finish the question through the message box. So from Mike Yang, Bank of America, question one is, can you update us about the tape-out schedule of the 3 nm project you mentioned? And the second question is, if the biggest US IDM customer does not pull the inventory from you, what will be the minimum revenue that you can recognize in 2025? Should we refer to the country liability number? So I guess I just covered. I kind of covered these two questions. For the scheduling, we cannot talk too much. So you can, Mike. I guess you know the whole scheduling of a typical HPC or AI chip in the leading-edge process now.
As we said, that the 3 nm project will be in mass production in early 2026, which means the tape-out will be in the first half, yeah, at least the first half, 2025, much earlier. Yeah. Anyway. For the U.S. IDM customer, the country liability number is a reference, but it is not exactly the same. As I mentioned, you have to still pay some others for the completed chip. For the inventory, I want to emphasize we are providing services, so we don't bear the inventory risk. The customer has to pick up the inventory. From Britney Lin, he asked about, can you share more about the margins and the profitability of NRE pipeline and the competitive landscape, especially with the CSPs? NRE timing, is it more second half 2025 based, and how much comes from the ARM total design platform versus direct RF2? Thank you.
For your question, I would say it is really big questions. The competition landscape with the North American CSP, I would say currently, mainly, we are competing with Broadcom. And Broadcom is a very respectable competitor. And we will try; actually, we will try our best to win through better engineering, better support, more flexibility for our pricing. We sincerely believe we have a chance to win some projects from other cloud service providers other than our existing customers. And for the margins and the profitability for the NRE, usually, the NRE is relatively the same. We can get about 40% gross margin from the NRE revenue. But it really depends on the design content and what is the application and what is the process now. And the numbers come from the ARM total design platform.
I'm not too familiar with it. I'm not understanding your meaning, but I will assume saying customer asks us to do a design by using the ARM IP. I would say for now, no. For ARM, I want to emphasize that ARM is providing IP. It's the biggest architecture IP provider in the world. But for ARM's architecture IP, it has more relations with the front-end designer. We are providing back-end design service. So for quite a long time, we don't have too many direct business relationships with ARM. You can keep on hearing that we are working with Synopsys or Cadence. That's the reason. For Brian Yang, EFGAM, his question is, as the industry advances towards 224G, several of your peers have emphasized SerDes IP as a key competitive advantage.
Considering that Alchip primarily relies on third-party licensed IP, how does management think about Alchip's ability to sustain its competitive edge in winning key AI HPC projects? Additionally, what implications might this have for you for the margin profile? Okay. Let me try to answer this question. 224G SerDes; it is a very important IP to sustain the HPC AI-related application. Using our number one competitor, Broadcom, as an example, the reason everybody considers to use their IP is because their IP is proven. They are already proven on their own product. Yeah. The old CSP customers are very conservative. They are not willing to use any unproven IP. The IP needs to be silicon-ready, silicon-proven, and production-proven, and also volume-proven, yeah, in order for them to use.
So many of our smaller competitors claim they have an IP, but to be honest, I don't think they can win. Yeah. Even the company like Synopsys, we can consider it's the number one physical analog signal IP in the world. Not too many people are there to use without proven. So our strategy, I think, is like this. We are working with a smaller IP vendor like Synopsys, Credo, or others. Proven their IP from the startup company. We have so many startup companies in the pipeline. In usual cases, startup companies go even faster than CSP. That's how they survive. That's how they establish a company. Once we've proven on the startup company or smaller company, then we can bring this solution back to the CSP. Yeah. So we already have several wins working with a partner using 224G SerDes of 5 nm and also 3 nm.
So the solution, after silicon comes back, we can propose to the CSP. So that's our strategy. The company like us, even we spent a lot of effort, spent a lot of time to make an IP solution by ourselves. After two years, when the design is done, I don't think we have a chance to win CSP directly. Yeah. It takes a lot of effort, and I don't think that's a good investment. Working with a partner is the right approach. Nowadays, it's also very complicated. Everybody should focus on the area which you are more capable of. Yeah. I don't want to, and also, I don't want to compete with our IP partner. The so far relationship, I think, is very good. Okay. Thank you. I guess it's about time, and we already answered the questions from the message board.
Thank you for your participation in our Q3 E arnings Call. Thank you.
All right. Thank you very much. Thank you. Thank you for your support. Thank you.