Alchip Technologies Ltd

TWSE:3661

Earnings Call Analysis

Record Growth with Strong Future Outlook

In Q1, the company achieved record-breaking revenue of $334 million with a net income of $39 million, but lower gross margins due to postponed NRE design. Despite share price declining from TWD 4,500 to below TWD 3,000, robust AI and data center demand is expected. Growth in 7-nanometer and advanced node AI ASIC shipments led to a 77.4% year-over-year revenue increase. Future prospects look strong with significant gains projected from AI and custom ASIC markets. The company remains committed to navigating capacity challenges and geopolitical risks with strategic workforce expansions across Japan, Taiwan, Malaysia, and Vietnam.

Record-Breaking Performance

The company celebrated another record-breaking quarter with revenues reaching $334 million. This marks an impressive 77.4% year-on-year growth. Additionally, net income stood at $39 million, resulting in an earnings per share (EPS) of TWD 15.8, setting new historical highs for the company.

Challenges with Gross Margin

Despite the stellar revenue numbers, the company faced challenges impacting its gross margin. Postponements in some design NREs to future quarters caused a higher production to NRE ratio. As a result of this shift, the gross margin was lower compared to the previous period.

Optimistic AI Market Outlook

The AI and data center markets are expected to grow substantially. The company projects a 50% Compound Annual Growth Rate (CAGR) for the AI-related market over the next five years. This optimism is underpinned by anticipated growth in custom silicon, or ASICs, which are expected to double their market share from the current 10% to 20%. The company positions itself as a leading ASIC service provider, driven by innovation and collaboration.

Share Price Decline and Investor Expectations

The company's stock price experienced a significant drop, falling from over TWD 4,500 to below TWD 3,000 since the previous earnings call. This decline occurred despite meeting all guidance figures. Management is actively working on strategies to meet heightened investor expectations, including diversifying business concentration and improving gross margins. They are progressing in business opportunities and partner relationships but call for investor patience due to the complexities involved.

Strong Future Business Outlook

Demand for AI and data center production remains robust, with substantial revenue growth anticipated throughout the year. Already, significant demand for the next year has been confirmed. The company has secured multiple design wins with leading-edge technologies such as N5, N4, and N3. A strong increase in NRE revenue is expected for the current year.

Geopolitical Risk Management and Workforce Expansion

To mitigate geopolitical risks, the company has diversified its business operations beyond China. Less than 15% of Q1 revenue originated from China, and the company continues to support financially healthy and compliant businesses in the region. Aggressive hiring plans have also been initiated to strengthen engineering resources in Japan, Taiwan, Malaysia, and Vietnam.

Guidance and Expectations

For 2024, the company expects continued sequential growth in revenue for both the second and third quarters. However, Q4 may see a slight decline in production revenue, typically a high season for NRE. For 2025, management remains confident about growth, bolstered by forecasts from IDM customers for their 5-nanometer chips.

Earnings Call Transcript 2024-Q1

[Daniel Wang, executive]

Good afternoon. This is Daniel Wang of Alchip. Thank you for your interest in our company, and thank you for your participation in our first quarter 2024 Investor Conference. We will start the meeting by our CEO speaking and the financials and then the business review and outlook and then the Q&A session.

Next page. This page is for the safe harbor disclaimer as usual. This meeting will be in English, if you need a Chinese presentation slides, please go to the MOPS to download the Chinese version. You should be able to download it already.

[Operator Instructions] And this video and audio content will upload to MOPS [Foreign Language] about 2 to 3 hours after this meeting.

Okay. Let's start with a very brief about what is Alchip. Johnny, please.

[Johnny Shen, executive]

All right. Good afternoon, ladies and gentlemen. I'm Johnny Shen, President and CEO of Alchip Technologies. Thank you for joining the investor conference meeting. We appreciate the opportunity to share our Q1 results and provide guidance for our future business outlook.

A brief company update. Yes, our company is founded in 2003, we went public in 2014. We maintain -- since we founded the company, we've been a success story, tapeout of 500 design Within 500, more than 55 are FinFET. And current headcount, it will be over 600.

Our last year revenue, $978 million. In terms of revenue, 80% of our revenue is contributed by HPC, AI area, we're a TSMC VCA member and also TSMC 3DFabric Alliance. Our current capacity we maintain 20 to 30 tapeout every year. Market focus is HPC, AI, automotive, agents. Whenever you need leading-edge technology, which is largely scale design.

Okay. Let me provide a quick update for the Q1. For Q1, we are pleased to announce another record-breaking quarter, with the revenue reaching $334 million, net income of $39 million and EPS stand at TWD 15.8. This number marked historical high for the company. However, some design NRE has been postponed to the future quarters, resulting higher production to NRE ratio and causing lower gross margin. The detailed quarterly breakdown in comparison will be presented by CFO, Daniel, in a later section.

Let me elaborate a bit about current AI market through comprehensive market analysis and the endorsement of both user and also suppliers. It's evident that potential AI and data center-related markets are very optimistic. TSMC recent earnings call and technology symposium further reinforce this sentiment. AI stand out as the primary segment experienced significant growth. In contrast, other market segments are either flat, showing minimum growth or even declining. The CAGR for data center, AI-related market projected to be 50% in the next five years.

Despite certain product companies like NVIDIA will continue to dominate the data center-related market. But in emerging trends within this landscape is increasing momentum for custom silicon, aka ASIC. The ratio between ASIC and standard product is expected to rise from 10% to 20% within the next few years. In the other words, the ASIC market CAGR will be even higher than the standard product. However, there are only a few proven ASIC provider in the industry. Alchip probably stands as one of the premier ASIC service provider, and #1 as a pure ASIC service provider. We are committed to never making any product compete with our customers.

Our focus in collaboration and innovation ensure we can sustain the market leadership and continuously capture the opportunity in this dynamic market.

Let me talk about a little bit our share price. Our share price experienced a significant decline since previous earnings call, dropping from above TWD 4,500 to below TWD 3,000. The decline occurred despite our consistent record-breaking performance and our ability to meet the guidance provided in each of our earning calls. It's clear investors' expectations are even higher than what we have achieved.

The management has fully embraced this message. We are tirelessly working to meet the heightened expectation. Our effort, including diversify our business concentration, improving the percentage gross margin to ensure sustained growth. Fortunately, we are making progress in the right direction in terms of business opportunity and partner relationship. However, we also wish for patience from investors as we navigate complexity of winning multibillion dollar opportunity. There are numerous steps involved from engagement to come from winning.

Let me talk about the future business outlook. Demand for mass production in AI and data center remained robust. We anticipate substantial production revenue growth for the entire year. Looking ahead for the next year, we have already received a significant amount of demand. We are confident to further grow our company as long as we can carefully manage the capacity-related challenges.

In terms of a new business opportunity, we have to secure several design win from AI-related companies utilizing leading-edge technology like N5, N4 or even N3. Additionally, we have worked very closely with major CSP company offer custom silicon, custom ASIC solution in order to diversify their business dependency. Overall, we anticipate much stronger NRE revenue growth for this year.

As for the geopolitical risk management, we have successfully diversified our business consolidation beyond China to other regions. In Q1, less than 15% of the revenue originated from China region but our future China direction remain unchanged. We're continuously believing and support business in China as long as they are financially healthy and complies with the rule and regulations.

In terms of our workforce, we have initiated a very aggressive hiring plan to bolster our engineers supporting resource in other regions, such like Japan, Taiwan, Malaysia and Vietnam. Currently, we have around 80 people working in our Japan office. Our Malaysia office already staffed more than 20 engineers. The workforce in Southeast Asia is planned to be around 60 people by the end of this year. The strategic expansion is designed to offer more flexible and cost-effective solution to meet our customers' dynamic requirement in the global market.

As a conclusion, we are confident our business is in an excellent stage. We are anticipating strong revenue growth this year. 2024 is going to be another record-breaking year for Alchip. Thank you.

[Daniel Wang, executive]

Okay. This page is for the first quarter P&L results. For the first quarter, the revenue is $333.5 million which represents 14.6% quarter-on-quarter and 77.4% year-on-year growth, respectively. And for operating income for the first quarter, the operating income of $38.5 million, which indicates 2.2% decline, but 84.1% year-on-year growth due to the lower gross margin compared to last quarter. And the net income for the first quarter is almost $39 million which represents 10% quarter-on-quarter growth and 104% year-on-year growth. For the first quarter this year, the EPS is TWD 15.8.

Next page. This page is the revenue breakdown by application. You can see the first quarter, the HPC dominates our total revenue. 93% of our total revenue contributed by the HPC area and most of them are AI applications. And the others like niche, networking and the consumer accounts for a single-digit percentage of our total revenue in the first quarter. And in the past two to three years, you can see HPC dominates our revenue breakdown with 80%, 90% of our total revenue in past years.

Next page. For the revenue mix by process node, we are very proudly saying that we should be the leading ASIC provider within the industry in terms of the process technology. In the first quarter this year, 7-nanometer or more advanced process node-related revenue accounted for 94% of our total revenue in the first quarter, similar thing for 2023. So we expect the 7-, 5-, or 4- or 3-nanometer will keep on increasing in the near future.

Next page. For the geographic breakdown for the first quarter this year, the revenue came from North American market, accounted for 77% of our total revenue, while Japan accounted only for 1%, and the Asia Pacific accounted for 15% which means our exposure to China market in the first quarter is relatively low to the previous quarters. And the revenue from the other regions accounted for 6% of our total revenue in the first quarter 2024.

Next page. For the first quarter business review, the revenue actually hiked as we expected. Obviously, this growing revenue contributed by the soaring AI chip shipment to the North American customer, especially our biggest customer. The first quarter revenue went 14.6% quarter-on-quarter and 77.4% year-on-year, respectively. And among the first quarter '24 revenue, the 7-nanometer AI ASIC shipment remains the biggest contribution, accounting for more than half of our total revenue in the first quarter this year.

And for the profit margin side, the gross margin in first quarter this year was at 19%, given the seasonal weakness of the NRE revenue. Usually in our business, the first quarter is the low season for NRE. And the fourth quarter is usually the high season for the design revenue.

For the first quarter, NRE revenue accounted for only 15% to 20% of our total revenue, which is the main reason for the low gross margin. And for the nonoperating income, you may notice that the nonoperating income this quarter is quite high. It was mainly due to the interest income given the high U.S. dollar deposit rates on FX and the cash in hand in the company.

Next page is the 2024 business outlook. We believe the AI computing demand will remain robust. We see the trend for the major cloud service providers to develop their own ASIC. This trend will keep on going. We believe Alchip will remain one of the top beneficiaries from this trend as we have been proven one of the leaders in providing customer ASICs.

And in addition to the cloud service providers, the North American unicorn startups are also aggressive in introducing AI chips. It provides more design opportunity to physical design suppliers like us.

And for the growth outlook, we are not allowed to go with the numbers. But for the growth this year, I will say in addition to the current ASIC business to the North American supplier, 7-nanometer and the 5-nanometer AI chip, to the North American IDM customers will also contribute in our 2024 total revenue. And we believe the growth this year will be strong. And the growing momentum will extend it to even the further future.

The NRE pipeline of the company remains promising. We are doing the design for multiple 5-nanometer or even 3-nanometer projects within the AI-related applications. And we maintain our strong growth geared towards the full year top line performance, and of course, the bottom line.

And I guess that's the presentation part for our first quarter investor conference, and we like to go into the Q&A session.

[Daniel Wang, executive]

And the first one is Haas from UBS.

[Haas Liu, analyst]

My first question would be regarding your second half outlook. Last time you expect the business to see sequential growth through 2024. So could you discuss on your latest expectation for 3Q and also fourth quarter with some projects start to ramp production, while the others could be in transition, as you just mentioned?

[Daniel Wang, executive]

Can you say that again. I'm not quite catching your question.

[Haas Liu, analyst]

Okay. Yes. What I mean is that if are you still seeing sequential growth throughout second half of this year?

[Daniel Wang, executive]

Okay. For the whole year, I would say we expect sequential growth for the second quarter and the third quarter. And for the fourth quarter, it is usually the high season for NRE. But the production and revenue contribution may decline a little bit in the fourth quarter. So for now, I'm not so -- I cannot give you some narrow guidance about the movement of the quarterly revenue.

I would say for second and the third quarter, I can say we will enjoy quarter-on-quarter for the growth. But for the first quarter, I would say maybe flat. That's the best guess. But to be honest, the NRE revenue -- there will always be surprise and disappointments happens all the time. So I cannot give you a very narrow guidance for the fourth quarter.

[Haas Liu, analyst]

Sure, that is very helpful. So just regarding the fourth quarter production revenue decline, could you elaborate a little bit more about the detail because I think your IDM customer recently on his earnings call is saying that the new project on 5-nanometer should start ramping through second half. So if the fourth quarter for your production business, the wind down of the 7-nanometer project is actually a negative offset.

[Daniel Wang, executive]

I think I already told in our last year's year-end earnings call that the major product we are currently shipping to our North American customers will gradually phase out starting in fourth quarter. And the last year for this product will be a generation migration year. So I think that's the major reason for fourth quarter mass production decline quarter-on-quarter a little bit.

And for the IDM customers, we will ramp up the 5-nanometer product in the second half of this year. So I would say the fourth quarter right now, our guidance is like a collective conclusion for these 2 products.

[Haas Liu, analyst]

That's very helpful and clear. Just a quick follow-up to my first question before I ask my second question. Looking to your outlook for 2025, what should we think about the growth trajectory versus your long-term guidance for 30% to 40% plus on CAGR growth? And if we just think in a more conservative way, how should we think about the wind down of your hyperscaler project on 7-nanometer? Will the contribution from that project based on the current order visibility be lower than the level in 2022?

[Daniel Wang, executive]

I think it's not possible for us to give the guidance for some particular product that precisely. I will say it this way. Usually, when the product is into the generation migration, the shipment will decline for sure. But the magnitude, I would say right now, it's only May 3. The magnitude right now, we are not that clear about the magnitude in 2025. But I want to emphasize, for the growth outlook in 2025, we are currently still confident that we will still have certain growth for the top line in 2025 because of the forecast we received from the IDM customer for its 5-nanometer chip.

[Haas Liu, analyst]

So my second question will be related on the competitive landscape. And with your U.S. peers recently being more proactive in providing the detail on their customer base and potential private wins, could you also share your potential wins in the next half year, and longer term, with your hyperscaler customers?

[Johnny Shen, executive]

Okay. Let me take this question. I think -- yes. I know a few of our competitors are very aggressive as rather a series of earnings call or even an AI Day also. But yes, to be honest, the business still is very big. Like we -- like Daniel also mentioned about entire market analysis, major CSP, they all plan to do the ASIC chip. So that's a lot of opportunity.

Recently, as you can tell, before, we are talking about the inference and training chip. Now the CSP are even more aggressive. They try to do their own CPU as well. So we have to -- to answer your question, we have to receive a lot of inquiry to make the custom silicon for most of our major cloud service provider. The opportunity is a lot, to be honestly, our management team and also supporting team are working very aggressive, and they also approach us almost daily meeting.

So I think I'm not saying we win it immediately but to be straight, a lot of opportunity for us to digest and some of them, the decision will be made in a very short period.

[Haas Liu, analyst]

Okay. Yes. I think for your competitors, they actually mentioned they have a project win and also a new customer win. And so I think both of them actually mentioned they have 3 hyperscalers out of the top 4. So do you think it is going to be the case for you in the next 6 to 9 months that you will be able to secure another 1 or 2 major hyperscaler project wins? And what should we think about your relationship with the existing hyperscaler customer?

[Daniel Wang, executive]

Okay. Haas, let me answer this way. To be honest, we are not allowed to elaborate who we win, who we lose or -- but we have very good opportunity to win in the CoWoS projects. So you are asking about the narratives about competition. And I think because we knew that the market is focusing on the chip using in the data center infrastructure, if you are talking about only the business, I can tell you we already win projects from those hyperscalers.

But to be honest, let me say it this way, when hyperscalers are looking for ASIC providers in the market, they don't have too many choices. Broadcom, Marvell, Alchip, MediaTek, GUC, probably you can only have these 5, can provide the most leading-edge process node and the large-scale design ASIC to date. So obviously, and logically, we received multiple IPs from multiple hyperscalers in North America. That is what I can say the most.

[Johnny Shen, executive]

Okay. Haas, let me elaborate a little bit. Just like Daniel mentioned, what's the definition for winning? If you're asking very specific, do we win the next-generation AI or something, I think that's a lot of cooperation we need to do.

For example, we are doing a few test chip for some potential customers to full blown like 3D concept and also the IP. So do we consider that as winning or not. But I think without this approval stage, is very difficult for a customer to make a commitment immediately. That's it for us.

Starting from 2-nanometer, the design becomes very complicated. Each of the design requires at least 3 people tapeout. Yes, as you know, due to our technology migration limitation, even the SRAM cannot do the further scale anymore. So the main chip, the computer ended up continuously go for 2-nanometer, 1.4. But I know that is most likely will be stayed at 5-nanometer. SRAM die is going to stay in the 3-nanometer. So it's a lot of proven concept in order to win both.

Currently, I do have a high confidence to secure those kind of projects because the design is just so complicated. It's an entry barrier because high-end provider is almost impossible for unproven service providers who plan they can do ASIC to penetrate in, I think, it's almost impossible. The only handful of the service provider that has this kind of capability and track record, pretty much it will be the Broadcom, Marvell and us. The rest of them, I think they don't have enough track record. It's very difficult for CSP to choose them.

The reason CSP want to do ASIC, they try to reduce the dependency from standard product like NVIDIA. NVIDIA is very expensive. ASIC solution already is going to provide a significant amount of savings, but choosing the unproven creator to save more, I don't think -- for the CSP, they will consider to do that. Yes, I hope that answers your question.

[Haas Liu, analyst]

Yes, and maybe I can squeeze one more...

[Daniel Wang, executive]

No, no, no. Give some time to others.

Robert, please. Robert from JPMorgan.

[Robert Hsu, analyst]

So my first question would be on the largest U.S. customer program. Previously, you seem to be pretty confident about your 2026 growth. So can you share a little bit about how are we tracking in terms of the design milestone. You say that you've been doing some tests, probably maybe for this customer. Any indication on the table schedule. Yes, just give a thought on that.

[Daniel Wang, executive]

Okay. Robert, as always, I have been pretty consistent with my view that 2025, I'm kind of conservative because I keep on mentioning it's the generation migration year for our major products. But for 2026, I am very confident that it could be a very promising and hyper growth year for Alchip because we will have the new AI ASIC product to the North American service provider for sure.

And we will -- the current 5-nanometer AI chip to the IDM customer, we will extended it to 2025 with considerable shipment volume -- to 2026. And in 2026, we will have full year contribution from the autonomous driving chip to our China customers. Not even counting in the potential win or revenue contribution from other big customers. We will have a very good 2026.

I think this guidance already tell you some clue about the scheduling. For the detailed scheduling, I'm sorry that we are not allowed to disclose.

[Robert Hsu, analyst]

Okay. So speaking about your IDM customers. So it looks like the demand is quite strong. How should we think about the revenue contribution from this IDM customer, it looks like it will become the largest customer for 2025. And how should we think about the production gross margin for this IDM customer?

[Daniel Wang, executive]

Okay. For sure, this customers probably will become the biggest customer for us in 2025. And the gross margin is a little bit better than what you should know but not much better, a little bit better. Yes, that's the status for this customer.

[Robert Hsu, analyst]

And currently, what's your view on the, say, the CoWoS deployment rate for this customer and probably share a little bit detail about what's your base case scenario, if we can secure probably 15%, 16% of this PO, what will revenue be in 2025?

[Daniel Wang, executive]

Yes, the revenue we can get, or let's say the chip we can ship, depends highly on the CoWoS capacity allocation. And currently, I'm personally optimistic because Alchip is becoming more and more important to TSMC, especially for the CoWoS side. I think Johnny can elaborate much better than I do.

[Johnny Shen, executive]

Okay. Yes, I think we already delivered a very clear message to TSMC. We have to gain the support for a specific customer next year. So far, so good. We will get a reasonable feedback. So that's why the customers are willing to place their top backorder even without the CoWoS commitment. So that's why we can, unlike before, there will be no commitment, but right now, the top line already placed.

Whatever the order we receive, we will try to get the CoWos capacity for them. But looking forward, the reason I'm so optimistic because TSMC, in terms of CoWos capacity, also improved quite a bit. And also since the cutback already plays, if we cannot get the CoWos capacity, fully CoWos capacity next year, the revenue still will go on to the year after. The business is a [ rail ] business. So working on the CoWos capacity, I think it's right now, one of the #1 priority for the company.

[Robert Hsu, analyst]

Okay. My last question will be on your strategy on the project selection. So previously, you seem to be less interested in bidding for the production 3 or so-called production-only basis. Have you changed your strategy on this where you still will focus on the production 1, production 2 going forward?

[Johnny Shen, executive]

No. Our strategy remain the same. More than 90% of our design, I think high 90% of our design is the businesses from design to tapeout and production. We do start to open a door slightly if the customers are up. As you know, for major customers like a cloud service provider, we try to get into their door, become their supplier list as soon as possible.

Maybe production-only in design will be the shortcut. If not the case, we will consider to get in. Once we get in, we were working with their design team. Trust me, all of them need the design support. Like I mentioned before, even the CSP customer, they're only doing one tapeout a year or one tapeout every other year.

So in terms of design know-how, schedule control and also the technology limitation, we have a lot of room for them to play with each other. So once we get into the door, we have the confidence to kind of step up and eventually will be the design funnel for that.

The reason I'm not focused to production-only-related design because I consider this kind of relationship will not be sticky. Yes, we need to -- our company's major competence is to provide the design service. So we believe we can do in the best design compared to most of the service providers.

[Daniel Wang, executive]

Charlie, Morgan Stanley, please.

[Charlie Chan, analyst]

So my first question is about your future sort of strategic partnership, meaning the chip design will become much more complex. On the other hand, you see some of your new several competitors, right? They went through the Arm Total Design to access to your CSP customers. So do you have any kind of new thoughts about your future partnership with those IP or EDA vendor?

[Johnny Shen, executive]

Yes, let me take this question, Charlie. Yes, the major supplier, we consider many, for example, foundry partner, IP partner, front-end partner and packaging and testing as well. For foundry partner, we don't need to elaborate, I think only majority, more than 99% of our business' co-dependency.

The IP right now, we are working with a very tight relationship to a specific IP vendor. In fact, they are the, in terms of business, they are #1 in the world. We have a very close relationship to this IP vendor. And also, we need each other. As you know for the leading-edge technology that, just as you mentioned, become very complicated and to be honest, majority of leading-edge technology like an end-to-user, they all have their own IP. They are standard product company, I say most of our company, they have their own IP.

The leading-edge technology IP can -- the major customer will be the [indiscernible] and cloud service providers that make us complement each other. The IP house needs a ASIC provider in order to get into the door of the cloud service provider. We need the IP solution in order to make our solution more comprehensive in order to compete with the other competitor in the [indiscernible]. So we're going to have a very high relationship. And in fact, there will be some exclusive announcement that will happen very soon. So IP house, I don't worry about too much. Eventually, we're going to have a #1 IP provider sitting right behind us to win the CSP customer.

Also for the other front end or OSAT kind of IP vendor, we also have a few progress. For example, the software IP, yes, we're going to have some relationship with them as well similar to some other service providers. And again, in terms of support, we are complementary to each other. We don't have any -- we are not having any competition. So eventually, we can win together. OSAT and also the packaging, since our volume start to surge 2 years ago, we get more and more better deals from the OSAT. So partnership is one of the areas we are focused for the future years.

[Charlie Chan, analyst]

I see. My follow-up question is to examine your recent project win/loss. So not trying to be picky, right? I just want to understand why or the company decided to walk away from some business because on the recent -- [ you made a news floor ] announcement, I think Marvell, they got AWS in French at 2.5 and maybe Google's CPU and it seems like they have a Microsoft sort of AI accelerator right? And Japan, social [indiscernible] seems to [indiscernible], et cetera.

Yes. So can you explain why your competitors got those projects and you didn't? And if not, what are you planning for the future?

[Daniel Wang, executive]

Okay. Charlie, let me answer you this way, we cannot say the name of the customer or say the name of the project, it's policy. And I do believe the connection for the project name and the supplier is newer research and your industry contract, I don't believe Marvell say directly, saying which projects to whom, which projects to whom. I can only say that there are many projects in the market for competition. And for generation by generation, there will be like the process node, there will be a major node, there will be a half node.

And of course, we want to keep on doing project with existing customers, especially those big ones. And as Johnny mentioned, unlike the other peers within our industry, we won't speak very loud saying, okay, we received some testing project or IP project or block project, those type of projects to the outsiders because the contribution may be minor. Yes, that's what I can extend -- my statement to you.

[Johnny Shen, executive]

Let me add some color for this. The ASIC business are very big. Last year, total ASIC business, I think, is around [ $4 billion ]. People predict 4 years later, this market will grow to more than [ $70 billion ]. So the common growth rate is extremely high. To be honestly, I don't intend to winning all, it's impossible. Some people, some service provider, they will try to find an end goal, try to get in. I think that's -- I totally welcome competition. Without competition, I think the company cannot grow. So I still consider Broadcom and Marvell will be my true competitor. I don't worry about other too much.

In terms of business opportunity, and trust me, we are working very close to the target customer. Everybody wants us to win. Yes, I don't think we are behind, honestly. Capturing everything would be very little revenue contribution right now because our AI revenue is so high. And without a significant win, you cannot see any total revenue growth, but we will continuously -- to work with the customer, even the test chip, IP test chip, 3D test chip, we will do that.

But I think based on the current NDA, even though we are confirm winning, as a customer, we cannot share any information. So I think wait until people see the number, we will start to realize the winning. Just like what Daniel mentioned before, 2026 is going to be an excellent year for Alchip. I have the confidence to satisfy everybody, by 2025 in the transition period, if we manage it carefully, we still anticipate a certain degree of growth on top of this year.

[Charlie Chan, analyst]

So just a quick follow-up. So you mentioned several times about a test chip, right? So how long does it usually take for customers to evaluate and confirm the kind of official project?

[Johnny Shen, executive]

Yes, 2 questions. Like I mentioned before, right now, the current generation, if it's N3, most likely it will be a homogeneous approach. But starting from N2, that will be the heterogenous kind of approach. You want design to require many tapeouts so does the test chip.

Yes, the 2-nanometer and those kind of test chips, I think it fully depends on the TSMC's drop of schedule, it's not commercially available outside. So there will be a lot of dependency and trigger. But I think -- and also the effort to work with lots of customers. Just like I mentioned, say, in the very beginning, there are many, many necessary steps from 0 to 1. The winning is -- it takes a long time to win a multibillion dollar kind of business, it's not that easy.

[Daniel Wang, executive]

Jeff, I believe, from Macquarie.

[Jeffrey Ohlweiler, analyst]

[Technical Difficulty]

[Daniel Wang, executive]

Jeff, your voice is really blurred. We cannot even hear you. I guess it's because of your signal.

[Jeffrey Ohlweiler, analyst]

You mentioned the ADAS order from China, can you give us more specifics on -- Okay, I'll start from...

[Daniel Wang, executive]

Right now it's good. Hello, Jeff?

[Jeffrey Ohlweiler, analyst]

Yes. So you mentioned China ADAS order, can you -- I'm trying to get some color on the ADAS China order mass production schedule?

[Daniel Wang, executive]

Okay. We are not allowed to disclose the specific schedule. I can only say the projects are ongoing. And we will tapeout within -- in 2024, and probably '25 will be a kickoff for the production. But the contribution period in '25 would not be so considerable, will not be long, simply speaking.

[Jeffrey Ohlweiler, analyst]

Okay. And just last follow-up question. Any other color on any meaningful China-related orders and/or auto-related orders?

[Daniel Wang, executive]

Yes. Our China policy remains the same. We are not refusing business opportunities in China market. However, for some projects, we intend not to take. Those projects are sensitive projects, or projects from sensitive customer or projects may have the concern violating the BIS restrictions. So -- but to be honestly, for now, not any given China customer can deliver the revenue contribution scale similar with the service provider customer in North America. So currently, our focus will still be in the North American market.

[Johnny Shen, executive]

Okay. Let me add some color to the China business. As Daniel mentioned before, there are many customers. Before, we have a lot of opportunity, people working on the GPU AI-related business. But unfortunately, most of our customer has been slowed down by the new regulation. So China, our focus remains the same, whoever come to us, if they comply with all the regulation using leading-edge technology, they are financially healthy, we will take. The market focus will be automotive. We're thinking about automotive areas currently still politically insensitive.

And also another area is the CPU. As you know, CPU, the IO speed and also total compute power is much less than the GPU-related, AI-related chip. So I think recently, CPU-related opportunity in China, I think it's booming. In fact, we already won a very sexy one, 5-nanometer, they will start to have production in short period because more and more institutional has hesitate to use the commercial CPU solution from the U.S. provider. So that makes way of a lot of room for the China CPU maker. So we try to take a project because the NRE from China is still very reasonable, and in terms of business and production in CPU and also automotive. I hope that answers you question, Jeff.

[Daniel Wang, executive]

Okay. Haas, please.

[Haas Liu, analyst]

I just have 2 quick follow-up questions for the next-generation hyperscaler project ramping in 2026 you just mentioned. What should we think about the annual contribution from that project? Because the way I think of it is if we only consider the wafer price difference between 7-nanometer versus 3-nanometer and everything else equal, it should provide you at least 70% to 80% higher annual sales contribution for the new projects versus the 7-nanometer one. Is that a fair assumption to make?

[Daniel Wang, executive]

Haas, have I ever mentioned it is 3-nanometer? I guess now. Just kidding. For the potential shipment volume, I'll let Johnny elaborate. It is quite amazing.

[Johnny Shen, executive]

I'm not talking about any specific customer, but in usual case, 3-nanometer design versus 7-nanometer design versus 3- versus 5-, I think each of the generation is [ set ] of wafer pricing in different. But by [ part ] it is also different.

For 7-nanometer, in usual case, it's a 1:2 HBM ratio. For 5-nanometer, since the computer is shrinking, so within the package, they can afford to place like a more large die. For 3-nanometer eventually even the unit price compared to 7-, I'm not surprised in the 5x or even more. So even with the same amount of shipment, the total revenue contribution will be totally different. So I hope that answers your question.

[Haas Liu, analyst]

Okay. Yes. Got it. And could you discuss more detail about the hyperscaler CPU opportunity you just mentioned? When do we start seeing the potential contribution from CPU projects? And how should we think about the potential business scale relative to your existing AI accelerator projects? What I would like to focus on is just on your hyperscaler CPU projects. We don't need to discuss about the China CPU project, you provided a lot of detail already.

[Johnny Shen, executive]

Yes, the China CPU project is totally different compared to the AI CPU. Let me talk about -- the China CPU application, mainly for desktop, laptop because a lot of institution are trying to reduce the dependency from the Intel, AMD type of a company. Yes, I think that's the China CPU.

In terms of AI CPU, most of service providers in addition to GPU, they need that CPU to run. But the amount would not be as much the inference chip. Before, the CPU solution is all x86, x87 kind of approach. But right now, since the proven track record from one of the CSP they use own-based kind of approach, saving a lot of cost and more CSP tried to catch up. So the difference in own-based CPU, that free up a lot of room for the ASIC solution.

[Haas Liu, analyst]

[Foreign Language]

[Johnny Shen, executive]

The design complexity and the unit price for CPU will be much less than GPU as you know. CPUs, they are using the DDR kind of solution, they don't need the HBM at this moment. Yes. So the unit price will be low. They're still using the high-speed DDR interface. And in terms of logic, most of the logic core will be using the CPU core, for example, [ R ] or even [ Raspi ]. They can do this kind of a CPU core.

[Haas Liu, analyst]

That's very clear. And when could we at the earliest time frame, start seeing some of the contribution from these hyperscaler CPU projects?

[Johnny Shen, executive]

I cannot give you a date. But everything is -- or maybe sooner than you would expect anyway.

[Daniel Wang, executive]

Okay. Robert, please.

[Robert Hsu, analyst]

I just want to follow up on the N3 and N2 and also, you're doing more test chips. So does that mean that in the future, we're not going to see first time right, in terms of the physical design going forward? And how should we think about the NRE activity. So is it a common practice for customers to release 2 design contracts or release 2 test chip projects to different suppliers and eventually go with one into mass production?

[Daniel Wang, executive]

Okay. Robert, for that question, I can guarantee you in the future, in the future process nodes, especially for the AI field, the technology will become much more complicated than what we have right now. There are many different technologies you may love in the chip design. I guess, Johnny can elaborate some new technologies to you.

[Johnny Shen, executive]

Like I mentioned before, starting from N2, a lot of scaling, its contribution will be very insignificant. For compute die, I think it makes sense if we want to generate more compute power and also save the I/O, you put the compute die using the leading-edge technology. But for the analog and mixed signal I/O interface, it doesn't make any sense due to the further scaling.

Currently, I think based on current analysis, the 5-nanometer will be suitable and no less than 3-nanometer. Unfortunately, the 6T SRAM, usually the cache, are also facing the scaling challenge. So most likely, they will all stay at 3-nanometer using hybrid bond to connect with the CoWoS. So in order to save the I/O and most leading-edge technology, they will only serve at the computer die and put the analog mixed signal and also the SRAM, which is the area suffered the high-yield loss put in relatively more mainstream technology.

Yes. So eventually, one design require multiple tapeout. So in terms of NRE will be quite significant to complete one design. The good thing is we're going to receive a very high NRE if we would design. The drawback thing is I truly believe, the number for player are getting less and less. Not too many companies can afford this kind of approach.

Early stage of, Daniel mentioned about, we have a lot of AI start-up company in U.S. but also those startup companies is no longer a pure startup. They're called unicorn or even [ unidragon ] kind of a start-up because they invest directly by the CSP. And they intend to put it outside. If everything proven, they may consider to acquire this company. This kind of a pure startup company, fund raising, finish the design, try to find customer is not going to happen in the future stages. It's very difficult. So NRE will be very, very good. And business opportunity will be getting less, entry barrier were getting higher and higher.

[Robert Hsu, analyst]

Just want to follow on that. So it looks like it will become a chiplet architecture starting from N2, and will you still be targeting at the core compute dies or you'll be responsible for all the chiplet, computer I/O, SRAM, et cetera? Because I'm a little bit concerned because if the different triplets are designed by different suppliers, how should we think about the responsibility if one of the triplets become -- having some defect?

[Johnny Shen, executive]

Yes, IDM will be the one design. Yes, just a multiple tapeout, and we use hybrid bond. We use TSV to connect all together. At the end, it's one chip. We just separated I/O and SRAM into a different technology, but we have to connect everything together, sign off the timing, sign off all the senior integrity, that is impossible for a different company to do different piece and somebody has to do stitch. Otherwise, the timing. SRAM, [ it should come to logic ] otherwise it's useless. So does the other interface.

To separate 2 different technology mainly is to save the I/O and also save the overall cost, the NRE will increase, but unit price will reduce drastically.

[Robert Hsu, analyst]

And just to confirm, these trends, it is only starting from N2, not from N3, right?

[Johnny Shen, executive]

It also depends. A lot of people use N3 plus N2, a lot of people using 3 plus 5 and 3 plus 2. So definitely what happen on N2, at some point, it will be earlier. We don't know. For your information in China, people using this kind of technology, even for the mainstream technology like 16 or 7, because are facing challenges, their foundry technology, I think, is saturated, so they have to use this kind of chiplet kind of solution even for the mainstream technology.

For TSMC and also the 3-nanometer, based on my understanding, right now, only one customer has a production chip using this kind of a chiplet solution. For CSP, up to 5-nanometer or 3-nanometer still mainly homogeneous type of approach based on my understanding.

[Daniel Wang, executive]

Okay. Charlie, Please.

[Charlie Chan, analyst]

So also on that 2-nanometer chiplets, it sounds like a very, very complex, several tapeouts. So I would assume this project may also take like 80 or even 100 engineers to work on that project, if you win the project. So Johnny, how are you going to plan for that? Have you reserved like those head counts if you will be winning the projects?

[Johnny Shen, executive]

Yes. Very good question. Our head count will grow along with the design complexity, we certainly maintained at least 20 tapeout a year. Before, we only need about 50 people who can do a 5-nanometer, 7-nanometer tapeout. Our 3-nanometer right now, we need 80 people. Eventually for 2-nanometer to handle multiple tapeout, I think hundred people put chip, the design is highly possible because the I/O, SRAM and also the especially the I/O.

In usual case, if they say, stick with more mainstream technology, the die cycle will be much shorter. You have the 3-nanometer, 4-nanometer and we need -- so much we can manage our resources well. I hope we shouldn't have too many people. I hope 100 people [ can scale doing one SRAM ]. Right now, the test chip we are going to do after this, we can kind of determine how many people we need, and we have to [ charge ] our customer accordingly because the resource right now is hard to estimate.

[Charlie Chan, analyst]

I just want to make sure you leave some flexibility. Yes, because it takes time to hire and even train engineers. So yes, anyway, so with your success and 2-nanometer win.

And my next follow-up is about your 5-nanometer. So -- or Daniel, can you give us some breakdown of 5-nanometer revenue in 1Q? And based on the current customer's order, what is the 5-nanometer revenue breakdown? I think your 7-nanometer below, that portion is just too high, it's at 94%. So maybe going forward, we can get some more details.

[Daniel Wang, executive]

Give me a second. 5-nanometer is about 10% to 15%.

[Charlie Chan, analyst]

In 1Q?

[Daniel Wang, executive]

Yes.

[Charlie Chan, analyst]

And so what would they be by the end of this year? Actually, let me also risk my rear-view question. I really want to know your 5-nanometer project pipeline. I think this question could be asked before, like which type or vertical or flows of 5-nanometer project in your NRE pipeline.

[Daniel Wang, executive]

Okay. I would say for the 5-nanometer, for the revenue contribution from 5-nanometer project, it will increase quite significantly starting from the third quarter because of the shipment to IDM customer. And 5-nanometer definitely will become the major contributor nodes for 2025. But I'm lazy, I haven't done the 2025 forecast numbers. So I can't give you a number for that.

[Johnny Shen, executive]

Yes, Charlie, also in terms of NRE, right now, we have a very little 7-nanometer design. It's almost nothing. So the design we are taking, either 5-, or 3-nanometer. So I think 5-nanometer NRE contribution is also [ very large ].

[Daniel Wang, executive]

Yes.

[Charlie Chan, analyst]

So because you mentioned several start-up company, right? Any names you can disclose or some hints about the product?

[Johnny Shen, executive]

No. They are unicorn type of start-ups.

[Charlie Chan, analyst]

So just search for unicorns. Okay. Yes. And last one, also just a confirmation, right? I think you talked about CSPs and CPU opportunity. So may I confirm your future opportunity as CSP includes both CPU and AI accelerators?

[Daniel Wang, executive]

Yes.

[Johnny Shen, executive]

Yes. Actually, many people knock at our door, now we are projecting both, [ angle isn't ] quite suitable. I think CPU, eventually, the volume is also quite significant.

[Charlie Chan, analyst]

Circling back to the previous question. So does attending the Arm Total Design's a requirements for you to win the CPU projects?

[Johnny Shen, executive]

Not really the -- in usual case, I think our customer has a capability to do all the architecture and sourcing all the IP from the software IP provider. But for some customers, I think they don't have architect-related capability. They are using, for example, some platform already established by Arm, I think it's kind of helpful. So that's why we don't -- against to join this associate.

[Daniel Wang, executive]

Okay. Paul Hsu from Taiwan Life asks 2 questions. The first one is does the company consider any share buyback to respond to the current pullback. Second, any chance to collect the bad debt that happened last year?

For your first question, we consider a little bit, but we don't want to do this too regularly because if you buy back the shares purely because of the share price downfall without any accident or surprises or events like we had for FEITIAN, it is not a good habit to do so. And for collecting bad debt happened in the last year, we do collect a little bit but not in full, in the first quarter.

Okay. Is there any question? Because right now, it's 3:40 already, we take one more question and we can finish today's earnings call. Is there any question? If not, thank you for your participation, and thank you for your interest in Alchip. Thank you.

[Johnny Shen, executive]

Thank you very much.