Alchip Technologies, Limited (3661) Q4 FY2025 Earnings Call Transcript & Summary

Dear investors, analysts and portfolio managers, this is Daniel Wang, CFO of Alchip Technologies. Welcome to our fourth quarter '25 earnings call. And thank you for your patience. We will start the meeting immediately. So for reference, as a routine, it is safe harbor disclaimer. And this meeting will be in English. If you need Chinese presentation slides, you can please go to the MOPS, go ahead and download the Chinese version. And you can write down your questions through Teams message function, and we will answer them accordingly. During the Q&A session, you can use the raise hand function, we will comfortably answer your questions. This video and audio content of the meeting will upload to MOPS. It takes about 2 to 3 hours to upload the file. So if you want to review the meeting, you can go to MOPS for the video and audio content about, like, 5:00. So the first part will be the message from our CEO, Johnny Shen.

Good afternoon, ladies and gentlemen. I'm Johnny Shen, CEO and Chairman of Alchip Technologies. Thanks for joining our investor conference today. We truly appreciate the opportunity to share our Q4 financial results and provide the update on our business outlook going forward. Our first quarter revenue come in below the plan. Primary reason is due to lower-than-expected production revenue. However, our service business recorded modest growth. As a result, the revenue reached $153 million with a net income of $47.8 million and EPS TWD 18.29, which is slightly higher than Q1 and the highest quarterly EPS for the year. The more detailed financial breakdown and analysis will be presented by CFO in the later section. Now, let me recap few highlights of the last year. 2025 was not a present year for us, especially given the most of our major AI-related players delivered exceptional performance in terms of revenue, earnings and share price. In contrast, Alchip's revenue declined significantly with net income and EPS also decreasing by approximately 10% to 15% year-over-year. The primary reason is the decline -- for this decline is due to lack of production revenue as we miss one product generation from our #1 customer. The most important highlight for last year is that we successfully regained the position with this customer. The new design was taped out last year and chip now is ready for high-volume production starting from Q2 this year. In addition, design activity for the next generation has already begun. We expect to complete the design by Q3 this year, positioning us well for the next phase of growth. Another highlight of last year is our automotive business in China. While the design and prototyping stage progressed smoothly, we did encounter some geopolitical-related challenge during the process. These issues has been fully clarified and resolved, and mass production has already begun. In addition, development for next generation product has already started, further strengthen our position in this segment and supporting continuously growth in automotive market. So other than these two high volume production, we have also achieved several important milestone with the emerging AI-related customer, primarily in U.S. market, leveraging the latest technology like N2 or N3. This engagement representing promising opportunity to lay foundation of future growth in advanced AI silicon platform. Another highlight is was improvement of our gross margin, although the revenue declined significantly due to lack of production revenue, but our NRE remained very strong, contributing about 7 percentage points improvement in gross margin compared to the year before, reflecting our different business mix and improve of supplier relationship. Ecosystem partners, like I mentioned before, starting from Q2 -- starting from N2 process, sorry, the HPC design become very complicated. Yeah, single design require integrating multiple directs and diverse solution. Unlike most of our competitor, follow a captive kind of solution, syncing for higher margin. Alchip is promoting an open ecosystem strategy, collaborating with a broad range of partner to develop comprehensive and efficient solution. Our ecosystem network has expanded substantially in the recent quarter with many partners eager to get in. We firmly believe the future of AI industry will lie on ecosystem, instead of a captive solution. Geopolitical risk management, we continue to diversify our business and design resource beyond China. In 2025, less than 8% of the total revenue contribute from China. Yeah, to strengthen our global engineering capability, we have launched a very aggressive hiring plan, focused the region outside of China. Now, our Japan office can accommodate more than 250 employee, while our Malaysia and Vietnam office now hosts over 120 engineer combined. And in conclusion, we expect Q1 revenue remain at a similar level. Beginning in Q2, we anticipate a strong ramp as a key production revenue resumes. For the full year, we have a confidence in delivering significant revenue growth, even compared to our peak year in 2024. Looking ahead, we remain very optimistic in long-term outlook for AI market. From this year to 2029, we expect our growth momentum remain in line with the industry leader and competitors. We are confidence in our ability to outperform the overall market CAGR in high growth HPC and AI market. Thank you very much.

Okay, this page is the fourth quarter P&L. The numbers just like Johnny mentioned, for the Q4 last year, we record $152.7 million, which represents 31.5% quarter-on-quarter decrease and a 62.2% year-on-year decrease. But given the improvement of the growth margin, our operating income last quarter was $39.1 million, which is 1% year-on-year -- quarter-on-quarter growth and 27.9% year-on-year decline. For the net income, last quarter, we recorded $47.9 million for the net income, translating into EPS of TWD 18.3, which is 8.1% quarter-on-quarter growth and 16% year-on-year decline. And for yearly numbers, like Johnny mentioned, due to lower production revenue contribution last year, our revenue for last year is $991.9 million, which is a 39% year-on-year decline. However, given the gross margin improvement, our operating income last year was $160.9 million, which is a 20% year-on-year decline. The net income was $179.5 million, which is an 11% year-on-year decline. So the full year EPS for last year 2025 was TWD 69.2. In the next page is the revenue breakdown by applications. You can see that the HPC/AI related revenue still accounts for the majority of our revenue, quarterly or yearly. For last quarter, the AI/HPC revenue accounts for 67% of our total revenue, while the niche market, networking, and the consumer accounts for 4%, 7%, and 22% respectively. The yearly breakdown, for last year, 83% of our revenue contributed by the HPC/AI category. And the other sectors such as the niche market, networking, consumer, accounts for about 15%-16% of our total revenue last year. And for the Process Node, we can still proudly say we are the industry leader in terms of the Process Node technology. For last quarter, 3-nanometer and the 2-nanometer combined accounted for 36% of our total revenue, while 5-nanometer and the 7-nano accounts for 43% of our total revenue. Combined with the 3-nanometer, 7-nanometer total at nearly 80% of our total revenue last quarter. And for 2025 as a whole, 3-nanometer and the 2-nanometer revenue accounts for 14% of our total revenue, while 7-nanometer and 5-nanometer accounts for the majority, which is 73% of our total revenue last year. For the regional breakdown for last quarter, North America is still the major market for us, which accounted for 50% of our total revenue. While the Asia Pacific accounts for 18% of the total revenue. The revenue from Japan accounts for 6% and the others account for 26% of our total revenue. For last year as a whole, North America remains the majority, accounting for 78% of our total revenue last year. And the other three areas, including the Japan, Asia Pacific and the others account for 8%, 8% and 6% respectively. For the number review for last quarter, like Johnny mentioned, despite the weaker revenue last quarter, the high gross margin made it the most profitable quarter last year. The Q4 revenue declined 31% quarter-on-quarter, due mainly to lack of the production revenue. Although the sales is not strong, the gross margin last quarter reached 42%, which makes our net income the highest among all quarters in 4Q '25. Number wise, we consider last year is a temporary hiccup prior to returning to a long term growth. Because, like I mentioned, we lost a generation of the major product, but we regained the next generation last year, which makes our revenue has the gap in 2025. However, since we kicked off -- we already kicked off the 3-nanometer accelerator last year, and then we are targeting to start the shipments in the second quarter this year. We think for this year, the overall revenue, we will enjoy a very good growth. Beyond this year, since we already secured the 2-nanometer project, which will lead a relatively long term growth outlook for Alchip Technologies in the next four to five years. And this page is for the outlook in the first quarter in 2026. The fourth quarter, revenue wise, remains sluggish, to be honest. I would say the first quarter this year will be pretty similar to last quarter. Production net revenue is that we don't have too many production revenue. The NRE, although from the full year perspective, the NRE will be strong. However, the first quarter usually is the lowest quarter for our NRE revenue performance. So for the first quarter, we are expecting the overall P&L will be pretty similar to the fourth quarter last year. And for this year as a whole, first of all, we emphasized many times that our 3-nanometer AI accelerator will start contributing to our revenue in second quarter. The NRE demand and the pipeline projects remained very strong. We see multiple projects in both leading-edge Process Node to kick off this year, starting actually from the first quarter through the whole year. The majority of the projects, those AI HPC related projects, come from the North American market. And we do expect multiple 2-nanometer project to kick in, in the following days of this year, which will bring in strong momentum to our NRE growth. The last one is we are expecting the most important 2-nanometer accelerator project to tapeout by the end of this year, which will ensure Alchip's pretty good long-term growth outlook. I guess that's our presentation part of today's earnings call.

[Operator Instructions] We will answer those questions accordingly. Thank you. [indiscernible] please.

It is great to see the business is coming to the inflection. I would like to start from your major hyperscaler AI accelerator projects. You mentioned we should see significant growth from 2Q this year and also into second half to drive a solid growth this year and also next year. I'm just wondering how should we think about the linearity for the project ramp on a quarterly basis? or to be more specific, we think...

Your voice is a little bit breaking down.

Okay. Yes. I just want to be more specific that how should we think about your revenue headwind for first half versus second half of this year? That's my first question.

Okay. I would say it will be very imbalanced for the first half. I would say the distribution will be quite, I would say probably 80% of our total revenue this year will be concentrated into the second half.

Yes. That includes NRE and also production business, right, both of the business?

Yes. The total revenue. The total revenue standpoint.

The total revenue, production revenue become very big, the NRE began to play a very insignificant role for total revenue.

Okay. Yes. And just on that, how should we think about the margin profile because you are reramping your production business quite significantly again. And should we expect that trend to persist into 2027, especially with your customer recently signing a 2 gigawatt [ power ] supplying to OpenAI powered by the ASIC.

I would say our customers' business has nothing to do with our margin. For the margin guidance, our previous guidance remains unchanged. I don't consider 2025 is a good benchmark for 2026 since in 2025, we don't have a very big production. So 2024 is actually a better benchmark for 2026. And our guidance remains the same. Our gross margin will be higher than what we had in 2024.

Okay. And yes, I mean, just on the 2 gigawatt computing power signed between Amazon and also OpenAI last week or two weeks ago, I was wondering if you are seeing incremental orders from your customers on that kind of agreement? And/or on the supply chain side, are you able to mitigate through the supply shortage of 3-millimeter CoWoS as well as substrates this year? And how are you going to just manage that.

We don't comment [indiscernible] or actions by those hyperscalers. I can only say for the substrate shortage. I would say this way, we communicate frequently and intensively with our suppliers, including the substrate supplier. And since this project is very important, a very good application from a very good end customer. So the substrate supplier has very high commitment to the project. So even though there is a shortage -- potential shortage for the substrate, based on the current information we received from those vendors, the impact will be very, very limited.

Yes. And also, as you know, the [indiscernible] capacity is very, very high. So even though you see some upside from somewhere, I think unlikely it happened in this year. So right now, the N3 capacity is fully booked in the industry.

Okay. At least that is a good reason for 2027 for the incremental demand, yes.

So we will come back to you later. Okay, Charlie, please? Charlie, you can unmute your microphone. Charlie, are you there? Okay, he seems to have problems -- okay, sorry.

So my first question is also about this year's revenue trend, because you just mentioned that the second half, you have like 80% revenue concentration. So can I also get another hint whether your single quarter revenue can reach like USD 1 billion in this year?

That, we cannot give such precise guidance because of the regulation. I would say for this year, that's possible.

Okay, okay. Yes. I wanted to provide some more ground for you to come from the potential target, like a quarterly revenue to be USD 1 billion. So I want to follow up whether your team can really squeeze out more outputs from your testing program? So that's why you see kind of additional revenue from the major 3-nanometer projects.

I would say it is not that related to the testing broker because like Shyang mentioned, for this year, actually the wafer capacity is very, very tight. And it is also because the [indiscernible] for the production for this 3-nanometer accelerator is very, very long. So for now, we already placed the wafer orders for like November. So the room for the expected numbers to change is limited.

Okay. Okay. Yes, because last time, you just mentioned that through your wafers in a testing program, there could be additional outputs...

Yes. We are still trying the last month. And the end of this year, a month revenue for a single month is also quite significant.

Thanks for the additional color. Yes. And also, you mentioned that you're going to tape out the 2-nanometer by the end of this year, right? So I'm not sure is there any schedule pulling, because I know the demand from your major customer and also your customers' customers, right, the OpenAI deal seems to suggest very long-term demand for that chip. But our [ checks ] also suggest that the [ 23 ] system performance is not that ideal, right? So I'm not sure if you can have this kind of a conclusion that 2-nanometer need to pull in. And also, can we get your confirmation that whether second half next year, we are going to see 2-nanometer [indiscernible] revenue?

Okay. Charlie, first of all, I would say our customer is very satisfied with the performance of the 3-nanometer accelerator we designed. So I haven't heard any complaint about the performance for design by the customer. And for the schedule, I would say the customer always want the chip to be -- to take up quicker, trying to bring the project schedule. It is a norm in the industry, which can lead our customers more buffer for a range, the timing, the presentation, everything. So the quicker, the better. However, it is about engineering. So we try our best to satisfy the customer. For now, I would say we are not committed to do that to meet the schedule, but we will try our best. For now, everything is on schedule. We are very happy with it. That's the current situation driving the 2-nanometer project.

Yes. For N2, even you call 2-nanometer, but you also contain other tape-out in, other technology like N3. So this is design unlike before, it's very complicated. It requires multiple tape-out and also the verification stage will be longer. So if we can complete the design by this year, I think that's much to achieve. But if we can do that, I think the customer will be very happy to achieve. They are shooting for -- yes, you are right, they are shooting for prototyping potential schedule by near the end of next year. That's they schedule. That's [indiscernible].

So last one I will be back to the queue. The CPO adoption timing, right? So I think Broadcom CEO said that the CPO adoption for ASIC would be very late. But I'm not sure what's your observation? Would there be CPU adoption in your N2 generation or your 1.4-nanometer generation? And what's your take on MediaTek stacking in your optical IO die partner Ayar Lab?. I know you want to promote this open ecosystem strategy, right? But why would your industry peer want to stake in this Ayar Lab?

Okay. Yes, for -- when will be the CPO-related product go to production, I think the situation changed a bit. Before people thinking about whatever the speed over 400 gig, it has to be go to optical. But recently, NVIDIA just announced this kind of bidirectional service. I think that we can say that still one generation. So unlikely, the N2-related design will adopt -- fully adopt the CPO. Most likely, I think will be the next, next generation. I think the similar situation to our customers, I think they don't have a current plan yet. But in terms of CPO potential, yes, I still very optimistic. It's design dependency also manufacturing design dependency. This is beyond 400 gig. Beyond 400 gig I think the CPO is a must to adopt. So -- and comment on the MediaTek invest, one of our partners, I don't think there's any competition between us and MediaTek. We're still winning. We were still doing the tape-out for this particular customer. During the design, I don't think NVIDIA MediaTek has involved for design activity. The probably -- they have a partnership for other strategic reasons.

Got it. So let's say, if you continue to win 1.4 nanometer and then they will adopt the optical I/O die for the CPO, UBL chip provides the design service?

Charlie, we cannot count upon particular project here. I can only tell you, first of all, you probably can see the press release from IR, not only MediaTek, Alchip Technologies is also the investor for this series [indiscernible]. And we have been partner with Ayar Labs for long term and Ayar is our partner and these also our customers. We are working together on some projects. So in the future, we think these two companies [indiscernible] and Ayar will keep up close relationship [indiscernible]. And the same thing, I would say the same thing for MediaTek. MediaTek invest in Ayar, and I would say MediaTek will also be [indiscernible] in the ecosystem for sure. And Laura, please, with Citigroup.

Yes. Can you hear me?

Yes. Yes.

My question is also about the future chip designs. As Johnny mentioned that it's becoming more complicated with different kind of chiplet designs going forward. So I'm just wondering that for L chips, you will still handle the 4 chips integration in the future or you might also kind of do the partially design and tape out and work with other partners potentially like MediaTek or other peers in the industry?

Yes. I think our preference, if we can handle customer has related resource to prepare on the architecture, I don't think for back-end implementation, we have any limitation. But if the customer are willing to adopt the commercially available I/O chiplet, we can also do integration. But for right now, most of the design we are doing for N2 chiplet is all done by us doing back end. But eventually, if available solution appear, yes, I think that's also a possible case.

Sure. So in the near term, maybe in like a two years perspective, so far, our project is still fully handled by ourselves, right?

Yes. Yes. Even the customer decide to get the I/O chiplet by themselves, but that will be the [indiscernible] similar model to HPN, yes, we will take the die to the final integration and also the production.

That's very clear. And also, my second question is that aside from your biggest customers in U.S. right now, are you also working with other CSPs for next-generation AI accelerators, and which we may see potential contribution in two years?

Yes, of course, we talk with everyone. But just like the scheduling we are talking about the 3-nanometer or 2-nanometer. For now, actually, if you want to have revenue -- production revenue contribution, probably you have to win now and you can get revenue after 1.5 and 2 years. So I would say this way, the trend I talked to investors two years three years ago is actually realizing. First of all, we told two years, three years ago that the CSPs, especially those big ones, they have very strong incentive to go for ASIC, it is realizing, right? And another trend I would say we are also seeing the trend is ongoing right now is those CSPs are trying to lower their cost, because the CapEx is getting unrealistically tough. Any savings to them will be quite significant. So we still believe under this trend as long as we are capable of doing those [indiscernible] design and meet the customers' needs. We -- our position is very good.

Yes. So given that we see various different kind of like inference or training perspective, now your products are mainly focused on training or big model, the potential projects you are now engaging with your customers, are they like a new market or focus you are looking for, which may contribute in the next two, three years?

We do have some applications, but some customers we consider will be a jackpot to us in next two, three years. We do have those projects.

Okay. And Jeffrey -- Jerry Su please.

Can you hear me?

Yes.

Yes.

Okay. So my first question is want to understand that I think for the current ASIC design, the chips are getting more complicated. I think previously, you have discussed that your customer adopting a COD business model. So might be although Alchip do all the design service, but the actual wafer start could be a split between you and your customer. Can you give more color on what is you are seeing for this year's project and also perhaps for the N2 as well? How should we expect the allocation on this front? That's my first question.

Okay. Like I said, we are not allowed to talk about specific project into details. I would say, again, like we talked to the investors in previous meeting, we -- when we talk with the customers, there are many considerations. First of all, the gross margin. Secondly, the scale we can handle for the production revenue because as I mentioned, the production turnaround time is very long. So the working capital requirement is high. So we will evaluate every element of the project in order to talk with the customer for a proper business model. That's what I can push to the boundaries for answering your question. Further detail probably I may not be able to share.

Also the allocation thing.

Yes, yes.

Okay, I got it. Then just to follow up on the 2-nanometer. I think in the prepared remarks, you mentioned that you have multiple other projects. So aside from your largest customer, can you also comment a little bit about those other projects? What are they going -- what are the -- can customers? And when should we expect to see revenue contribution?

Actually, various type of customers. We understand that the N2 project, the N2 process node will be -- it's a very, very expensive project. However, since in the North American market, for some emerging accounts, they are funding very, very well, because all the capital in the markets [ go ] to the AI sector as long as they have strong team and the innovative ideas, probably to do a 2-nanometer project is not a problem for them. So for our 2-nanometer [ traditional ] projects, we have emerging accounts and we also have some traditional networking companies in North America. [ F4 ], N2, I think obviously, the use are very limited, either it's all AI-related application, either in the networking or in the accelerator side. I hope we answered your question.

What is the rough timing to see revenue contribution for these emerging accounts for traditional networking companies?

Actually, the revenue is what, already there. The NRE kick off already there.

So that means that our mass production could happen after your main customers take off? Should we -- is that the correct interpretation?

Yes.

Okay. Got it. And lastly, on the automotive side, can you also give us some color on how should we expect for revenue this year? And also perhaps into next year as well.

Sorry, because actually, the [indiscernible] worn specifically asking not to say numbers in the earnings call. So I can just tell you like we described for this year, we are shooting for high revenue performance. And the potential for AI accelerators growth is quite significant. And our CEO, Johnny also mentioned we are expecting our growth to match up with the industry growth or even better.

Yes. For all the business, I think -- yes, for sure, that will be our #2 customer in terms of revenue.

And Johnny, please. [indiscernible], please.

I just have two quick follow-up questions. I think the first one is just regarding your value add on the chiplet on 2-nanometer project because it seems that there are going to be multiple takeouts on different nodes. So I understand that on the back-end side, it's going to be pretty complicated. But the reality is that there are going to be more of the chiplets or the tiles coming from the other part of the design service partners or ecosystems. So I'm just wondering if you could share the margins profile for the upcoming 2-nanometer projects, whether it is going to be as high as the one that you are going to ramp pretty soon in the following quarters?

Various type of form is possible for the 2-nanometer process in general. We cannot have a given project. For example, we can do the [ Top die ] physical design and we can do the integration for the [indiscernible] type of business model. Of course, we can also do the physical design for the I/O die and do the integration for I/O die and Top die. So any kind of the business model, we will be kind of...

Yes. So for I/O die, I think so far, like I mentioned before, most of the integration and production are handled by us. And also in terms of size and the die price, I/O die is much cheaper than -- I mean the computer. And if in the future, with more I/O die, play more significant role, then the model, I think, will be similar to HBM. We get the I/O die, if it happen, we get the I/O die in and we count out this portion of the revenue is similar to HBM and maintain our gross margin.

Okay. Yes, that's pretty clear. And then just a follow-up on the CPO, and I will be back in the queue. I saw that you and Ayar Labs at TSMC's OIP symposium last December that you actually would provide EIC and also the networking switch to integrate with Ayar Labs [ PIC ]. Would you be able to comment about this which nodes that you are running on? And second thing is that it seems that the networking switch that you are able to provide to integrate with Arya Labs PIC is actually just -- it's actually not simple at all or the die size -- yes, I was just wondering regarding the die size for that networking switch and also the EIC, whether it is actually going to be another significant opportunity for you outside of the core AI accelerator business.

First of all, your voice is breaking up a little bit. I try my best to get your question. For I, please consider it is a partnership, just pure partnership. Arya Lab is our customer and we do project with them. As long as for what kind of project EIC, PIC whatever, we cannot disclose this kind of information in earnings call. And we -- both companies management committed to each other to the future design for those chips requiring high-speed transmission. So yes, I'm sorry that we cannot disclose such detail for a given company or a given project. Okay, and next one will be Charlie. Charlie, please?

Okay. I'm fine if you want to get new caller to ask question or maybe I do it quickly, okay? So yes. So first of all, I remember you sort of have some involvement in LPU design. And I think the company recently was a so-called acquired hire by NVIDIA. So I'm not sure if customers still want to take out that LPU. And also, I'm wondering because that design include SRAM, right? So I'm not sure is that the same foundry for the SRAM production.

Okay, for the project we are doing with [indiscernible]. Currently, it is a cost. So I would say the decision will be made by it's new owner, NVIDIA. If they want to continue the business with us, we are more than happy to so. If they want to do it by themselves, of course, they pay for our work, they already pay. So we cannot work on that. So I cannot give you a firm answer, but the current status is the project is in the cost.

Yes. As you know, the supplier relationship between us and NVIDIA are different. So yes, it's kind of debating situation who's handling if they go.

I see. And also another very, very big U.S. customer opportunity I'm referring to the 2-nanometer U.S. automotive and potential robot chip design, right? So is that a reasonable target for our Alchip sometime in 2028 or 2029?

Of course, any big company in U.S. with big volume is our target. We are shooting for, I would say, everyone. Actually, we are engaging with almost everyone in North American market like Microsoft, Google, those kind of CSPs or other big names. We are happy to do business with them for sure.

Okay. But if you look at the current so-called 3-nanometer project, I think is by your industry peer, GUC. Do you believe there is a level of so-called level playing ground when you compete for 2-nanometer?

I would say GUC is a respectable competitor and peer for us. I think the market is very big, and we compete, yes.

Okay. And last one, I think it's a little bit more strategic discussion. So Johnny, I wanted to get your view because recently, as you know, that the AI already started to transform or even disrupt the software industry. So some of your EDA partners like Cadence, Synopsys, they got some doubts from investors whether this is also disrupted by the AI. So I'm not sure to Alchip, right? It's kind of AI automation for design flow, validation, whatsoever. Is there a long-term positive or negative to Alchip? I think positive argument is that your productivity will increase, but negative is that whoever maybe not with a very long experience, they can use AI tools to design the chip for customers.

Yes, I think that this is a very big question. Actually, personally, I am the AI believer. I think sooner or later, most of the work can be replaced by AI gradually. I think -- but right now, we are also working with all the EDA vendor very closely at Synopsys and Cadence. There will also improve quite a bit using this kind of inference. But what I understand is currently AI is still a statistic approach. If you have enough reference on the specific [ process ] node, you can -- they can anticipate all the timing problem, electrical problems in order to shorten the kind of long time. But for most leading-edge technology, there's no database yet. They don't have this kind of reference. So from a statistical point of view, AI doesn't work. So that's why including us and also our competitors, when they are doing a leading-edge technology like N2 plus multiple N3, they need more than 100 people. But if you go back to like a 28-nanometer, 40-nanometer or 60-nanometer, in usual case, within 10 people, we can do most of the work. I think AI is the power getting bigger and bigger. Maybe ultimately, they can do something. But right now, for the most leading-edge technology, including us, including our customer and industry peers, they need more people. [indiscernible].

Yes. That's super helpful. Yes, so I think even back-end design, there are several stages like place routing. And as you said, right, there's no database. You need to -- based on customers PDK, based on your experience. But for some sort of steps like validation, I think, is a pretty labor-intensive, very tedious. Would that kind of a chip as [ station ] be gradually migrate to the AI?

Yes. I think for some point work, we can see the AI really help. And also most of our work is done by using the server type of CPU, some specific case, and we are working with the vendor to use the GPU. As you know, the GPU performance for the point task is much, much faster than CPU. So those kind of improvement can reduce not only the design resource, but the design turnaround time. I think that's also very important. So AI is gradually get involved for all design activity almost everywhere.

Because of the limited time, we answer the last three questions, and I would say, after that, we will conclude our earnings call this time. Mr. [indiscernible]?

I just wanted to clarify, OpenAI and Amazon have signed like a 2 gigawatt worth Trainium order, right? 2 gigawatts is a lot of capacity. So you're saying 2026, there's not a lot of room for orders to change. But I want to clarify whether there can be upside to what you guys are forecasting for 2027? Or should I interpret this 2 gigawatt upside will come in Trianium 4. That's my first question. yes.

First of all, I'm sorry that we don't answer the pacific project or pacific company. I will say this way, for our major product, the 3-nanometer accelerator chip, even we place the order today, the chip will come out in very late this year. So any incident happens or any upside surprise happens, most likely will be next year, 2027. However, for 2027, I cannot give investors a number because -- we are not -- the allocation is variable. The demand is variable. And those industry dynamics is not very clear yet. So I cannot give you a very firm answer for 2027. But definitely, we will enjoy a pretty good growth in 2027 for this project for sure.

Right. I understand that there will be growth. What I'm trying to understand is that relative to like what you thought like, say, 5, I don't know, 5 days or 10 days back prior to this announcement, in your mind, is the upside higher in 2027? And I think related to that question is NVIDIA and Broadcom are saying they have locked up all capacity, memory, ABS substrate, front-end capacity for their requirements till 2028, right? Your peer, Marvell is also saying that. So I'm just -- like even if there is upside, will you be able to capture it? I just wanted your sense on this.

Again, we cannot comment on specific news or events. Honestly, I would say for your question you're supposed to ask AWS, OpenAI, not us, right?

I know, but like it really depends on whether you guys will be able to secure the supply necessary, right? You are getting upside in orders, but then you have to ready the supply chain, where you have to place orders at TSMC, you have to coordinate the supply chain, right? There is a lot of shortages all across the supply chain, right, in every -- so I'm just trying to figure out whether -- even if there is upside, will you guys be able to see it because every one of your peers is saying they have locked down capacity, they have locked down supply. So I'm trying to understand whether you guys have also done it? And will you be able to enjoy this OpenAI upside, right?

Honestly, I don't think those so-called lock-up is a real thing for 2027 or 2028. The current -- like I said, the current capacity for 2-nanometer and 3-nanometer will be very, very high in 2027. It's not the negotiation or the talk between the customer and the TSMC is not there yet. How can they lock the capacity. So I'm sorry, I cannot give you an answer for that. I'm sorry, we have to speed up -- if you have further questions, please welcome to mail me or call me any time you want. Okay. Next one is Jennifer [indiscernible].

I have 2 questions. So first of all, about the CPU projects for the CSPs. They are now becoming quite large in size, even though the margin could be possibly lower. So do you think at this large scale, it makes sense for Alchip to start engaging in some of these CPU projects? And do we have any potential progress on this? This is my first question

Honestly, we are not ruling out to take this kind of so-called production only project. However, we do worry about the margin can go very, very low. So it is not a very easy equation because we have to consider a lot of the elements into it. For example, if we want to take project with such low margin, the purpose is not trying to make money. The purpose is trying to break through into the supply chain. However, we still think that through providing value, it is the best way to secure a customer. To do the production only business is not that valuable to customers and it usually ends up with a pricing war -- so I cannot give you a firm answer, but I can only say we are not taking those projects. However, it will be a lot of influence.

Yes. Understood. So another follow-up is on the networking customer. So you have the 2-nanometer project with the U.S. networking customer you mentioned before. Could it be possible to become the second biggest customer of yours and say, 2027 or 2028 once the 2-nanometer tapeout goes into production? Can you give us some color on that?

It's hard to judge because we do have high expectations for these projects because the customer is probably [ big name ]. However, we also don't rule out the [indiscernible] to become very big, because in the AI era, as long as your product is very innovative and being very efficient, [indiscernible] could go up very quickly. So of course, we consider the contribution of the [ networking ] project, we are currently doing can bring us some revenue.

The revenue mainly happen in or maybe 2029 in your -- for the 2-nanometer production, do you think it will maybe possibly come in 2028 or maybe longer in 2029.

Yes, 2028, definitely. The 2-nanometer project will be in production in 2028. And the last one would be for Laura from Citigroup.

Just a quick one, actually quick two. I'm not sure if Alchip has done some study of Intel's EMIB packaging. As we see that some of the U.S. CSP, they kind of been asked or encouraged to do more of the production in the United States. And we see some of the CSP, they are also kind of working with Intel's on the EMIB packaging. So I'm not sure if there's also the possibility that Alchip to work up with Intel on their advanced packaging process.

Yes, it depends on customer decision. And actually, we did, and did for the packaging before. So we are not -- we are -- actually we are familiar with this EMIB.

Yes. So do you see any like progress or any schedule you potentially aiming for?

Yes, the previous design we're using the EMIB, I think it is already in production for a couple, two years ago, two years ago. Yes, we are pretty familiar with the team. As you know, you can consider Intel is our customer, right? They acquired one of our customers. We've been working with the them entire infrastructure team very closely. But again, the decision is still made by customer. But we don't do a thing. We are okay to work with Intel for the EMIB.

Yes. As long as TSMC, CoWoS capacity is not an issue. And I think not too many customers decide to design the big portion using EMIB.

Yes. Understood. Yes. And the next question is more like housekeeping. We understand you already kind of explained that Q1 will be similar to Q4 in terms of the revenue, and there will be less NRE in Q1. But can you give us kind of indication about like the gross margin because Q4 was really strong in gross margin. So should we expect that the Q1 or first half margin to back to like a Q3 level last year? Or how should we think about that this year's gross margin.

I would say the first quarter will be very similar. So the gross margin will be also high, I think. So I would say for the first quarter this year, last quarter will be a very [indiscernible]. And for the second quarter, it's kind of a mix because the revenue is picking up because of the contribution from the 3-nanometer production. So the growth is mainly because of the production, which is a lower gross margin business compared to our [ NIE ] -- and obviously, the second quarter's gross margin will be lower to the first quarter, but however, still higher than the third quarter for the second half. That's what we see. I cannot give you the numbers. And I guess it concludes today's earnings call. And thank you very much for participating into our fourth quarter '25 earnings call and see you next time.

Thank you very much. Thank you.