Thank you for you all to join our second quarter investor conference meeting. This meeting will be in English.

If you need Chinese presentation slides, you can go to MOPS, which is 公開資訊觀測站, to download the Chinese presentation materials. If you want to ask questions in Chinese, no problem for us.

Secondly, if you don't want to ask question orally, you can also write down your question through Teams message function. For the Q&A session, please use the raise hand function, and we will call your name and for your questions.

This video and the audio content of the meeting will upload to MOPS, 公開資訊觀測站, about 2 to 3 hours after the meeting. If you cannot download the video or audio content immediately, please be patient.

I think this meeting will start with our CEO's words to investors. Johnny?

Please. Good afternoon, ladies and gentlemen.

I'm Johnny Shen, Chairman and CEO of Alchip Technologies. Thank you for joining our investors conference today.

We appreciate the opportunity to share our Q2 financial results and provide an update on our business outlook. Q2 summary.

Our second-quarter revenue came in line with the plan. Revenue increased significantly quarter-over-quarter due to higher production revenue while the NRE remained very strong.

As a result, total revenue reached $242 million with a net income $51.8 million and EPS $20.01 NT. A more detailed financial breakdown and analysis will be presented by CFO Daniel in the following section.

There are a few highlights worth mentioning last quarter. First, our N3 AI accelerator entered mass production and began shipping in late June.

The program is under early ramp-up stage, but even with a month's revenue contribution, it drove our quarter-to-quarter revenue growth up more than 80%. We just announced a few days ago our July revenue reached a record high for the company, reflecting the continuous production ramp for this program.

Meanwhile, the next-generation design also achieved a major milestone and entered the final implementation phase. We remain on track to tape out by the end of this year.

In addition to our AI accelerator business, our other production program also performing well. The automotive ADAS program in China continues to see strong demand and is now our second largest revenue contributor for this year.

We highly confident to sustain this business and further expanding this business as we already secured the next two generation of design from this customer. The first next generation remained on track for tape out in early Q4 this year.

We have also two Arm-based CPU program in mass production now. Shipment volume increased meaningfully in recent months, supported by ongoing global CPU supply shortage.

In addition, several consumer programs are contributing steadily to our production revenue, including gaming application in Japan, smart device applications in both U.S. and China.

This program continues to run smoothly, provide a healthy level of business diversification. For future business outlook, overall, like the industry leader and some of our competitors, we remain very optimistic about the long-term outlook for AI market, particularly is in ASIC segment.

We believe our competitive position are even stronger than most of our peers because we continue to serve as the sole design partner for our end customer. In addition, our end customer maintained one of the highest CapEx level in the industry, creating a very big SAM, serviceable available market, for Alchip.

On the execution side, we are largely resolved most of key constraints that previously limited our growth, including engineering resource, manufacturing capacity, production allocation, and working capital. As a result, we believe we are well positioned to catch the significant growth opportunity ahead.

Looking ahead, we expect our business enter a new phase of growth beginning of this quarter. Product revenue will expand substantially, and we anticipate delivering the record high revenue and earnings in Q3, followed by another sequential increase in Q4 as our major product program continues to ramp.

More importantly, we believe this is not simply a single-quarter or single-year's growth story. Based on the production program already in place, together with our next-generation design pipeline, we believe this growth momentum will sustain for at least the next 3 to 4 years.

For conclusion, I would like to emphasize once again, we believe the most challenging period for the company is now behind us. We regained our growth momentum, are well positioned to capture large share of the rapidly growing AI ASIC market.

The industry continuing to move toward the COT model, exactly as we anticipate for many years. We believe this secular trend play directly to Alchip's strengths.

Based on ongoing customer engagement and business pipeline, we are highly confident in securing another major second hyperscaler CSP customer in the near future. Looking ahead, for the near term, we remain very optimistic about the long-term AI opportunity.

From this year to 2029, we expect our growth trajectory to remain in line with our key customer and industry-leading players. We believe Alchip well positioned to deliver sustainable growth and create long-term value for our shareholder.

Thank you.

Thanks to Johnny. This page we will start the financial part from this page.

For our second quarter, please be noted that the number will be in U.S. dollars.

For the second quarter P&L, the revenue in second quarter is $241.7 million, which is 82.6% quarter-on-quarter growth. However, the year-on-year it is a decline for 18.7%.

For operating income, I think which is quite promising, our second quarter operating income is at $52.1 million, which is 20.5% 20. 20.5% quarter-on-quarter growth and 38.7% year-on-year growth.

For our net income in the second quarter, it is $51.8 million, which is 14.9% quarter-on-quarter growth and 20.7% year-on-year growth. For the second quarter, the EPS is $20.01 NT.

This is the breakdown by application. You can see the second quarter's breakdown is pretty similar with what we did in the first quarter: 50% from HPC, 1% from niche market, 16% from networking, and 32% from consumer.

The reason for 32% for consumer in the second quarter and 36% for the first quarter is because we currently put our autonomous driving chip business into the consumer. We are internally considering to recategorize the autonomous driving chip business into probably the niche or HPC categories in order to better fit its original purpose of the chip.

For the process node, because the N3 accelerator started to kick off the production in late May last quarter, you can see the 3nm, 2nm process node revenue contribution grow to 47% in the second quarter from first quarter's 21%. For the contribution, you may see again the same reason, the N3 accelerator's production in place is the major reason for our 51% revenue contribution from North American region in the second quarter from 23% in the first quarter.

For the second business review, actually there are many parts is a repeat from what we said previously. Again, I want to reemphasize the numbers again: the second-quarter revenue $242 million.

For the second quarter gross margin, although it went down on increasing portion of production revenue, however, I would say this way, it is a little bit better than we expected previously. We think although the gross margin went down, the operating margin improves on year-on-year if you do the year-on-year comparison.

For our business model, we do believe as the revenue scale keeps on going up, even the gross margin goes down because of the increasing portion of the production revenue, the operating margin will keep on growing given our quite good operating leverage. You may notice that the second quarter OpEx went up.

If you do the quarter-on-quarter comparison, it is quite significant. The main reason is because first of all, the headcount expansion and the salary adjustments of our company.

Secondly, we do see the expense increase for our EDA tool licensing and our server usage because nowadays the design in leading edge process node requires quite significant amount of server time. For the future, we do believe for the third quarter and the fourth quarter, the operating expense would be pretty similar with the second quarter.

We are very confident that for the yearly OpEx growth, it is still in control. Secondly, like I said, the N3 accelerator business kicks in starting in late May.

We didn't have too much revenue in May from this N3 accelerator. However, as the N3 as this N3 accelerator business keep on ramping up, we are going to see increasing revenue sequentially in the following months from this product.

As for the autonomous driving chip, I would say overall the shipment remains very stable. I would say the demands seems a little bit better than what we delivered to the investors before.

We are currently very optimistic for the current autonomous driving chip shipment. The next-generation project is going to tape out probably late this quarter or early next quarter.

I think this concludes our presentation. We are going to enter the Q&A session later.

Thank you all. Okay.

I noticed that Haas raised the hand very early. Haas, please, you can unmute by yourself.

Yes, thank you Johnny and Daniel for the great results. I would like to ask questions just regarding how we should expect your gross margin and operating margin trend in the next couple of quarters as the Trainium3 or your largest hyperscaler's ASIC starts to ramp in a meaningful way.

Because what we can see is already gross margins declining from 50% to 35%, operating margins down from 33% in first quarter to 22% in second quarter. What should we expect in the next few quarters when the business starts to scale up with your production business ramp, your gross margins and also operating margin target in the next few quarters?

Thanks.

Okay. I'll give you some numbers for it.

For our gross margin, yes, it went down to 30-something percent from 50% in the first quarter. I would say this way, since we only have 1 month high-production revenue in the second quarter, based on our calculation, I would say for the third quarter and the fourth quarter, the gross margin, the blended gross margin will most likely be about 20-something percent.

For the whole year, this year, I think our blended gross margin should be in the low to mid-teens low to mid-20s. I would say this way, as the N3 keep on shipping, the NRE to the production revenue ratio most likely be in 10% to 20% for NRE and 80% to 90% for production.

Based on this scenario, the reasonable blended gross margin would be in 20-something, lower 20s. That's the outlook for our gross margin.

Secondly, as I mentioned, our business model enjoys pretty good operating leverage, when our scale reach certain point, I'm targeting our operating margin to be in high teens. Am I answering your question, Haas?

Yes, that's very clear. Thanks, Daniel.

I think that gross margin guidance is actually way better compared to your previous Trainium1 project when you started to ramp. Would you be able to provide some more detail regarding how you are able to lift the gross margins while you are ramping your production business to 80% to 90% in second half this year and even into 2027?

Thank you.

Okay. I would say for the pricing, for the gross margin, for the margin to a given project, I cannot disclose into the detail into the narratives.

I would say, first of all, I think the customer started to appreciate our value. We are providing much better value generation by generation to our customer.

That's one reason. Secondly, our position within our suppliers, within foundry, within the OSAT is getting better and better.

From customer, from supplier, we do have the room to stretch our gross margin a bit. Thank you.

Yeah.

What Daniel say is exactly right. I think, Haas, to answer your question, I think we have the confidence to outperform the gross margin compared to our first high-volume production.

I think this statement is true.

Okay. Yeah.

Then just relatedly on Trainium on your follow-on projects, which you said to be tape out in later this year or early next year, I was wondering if you could just provide some more detail on the scheduling for the engineering verification and also the time frame for mass production for that project, and whether the customers would further appreciate your work and then your gross margins for that project could be staying at similar level or even better compared with the current ramping chip

set. Haas, I think the design schedule for this design I think it's very sensitive.

I don't think customer will allow for us to share anything. I can tell you right the design is unlike before.

The design is heterogeneous. It's like a chiplet design.

One design require multiple tape-out, all tape-out is not going to happen in the same time. As the turnaround time for N2 will be much longer than Q3.

It allow us to tape out Q3 later, and also they have interposer tape-out, a lot of tape-out add up all together. I think overall customer are very satisfied with current status.

As I just mentioned before, the current target still by the end of this year. We strive to meet this target, and production similar will be by the end of next year.

That's the current plan.

Just a quick follow-up before jumping back to the queue is that so it sounds like you are going to be the one in charge of the next-generation chip set as well. Would you be able to discuss about the opportunity, the content opportunity and also the volume opportunity there since your customers recently also mentioned they are going to be more aggressively selling the chip set to the external customers in addition to for their own use?

Thank you.

The continuity I think is the most important. Like I mentioned before, I think right now the relationship between us and the customer, we can say it's getting better and better.

I didn't sense any urgency for this customer to pursue other choices. Again, the precise schedule, it really depends on the manufacturing availability, and most likely it will be the A-series.

Also customer need to decide for the IO chiplet, for the memory chiplet, which technology they're going to use. I think prior customer make a decision, we going to have another test chip similar to the previous one to do test chip first.

I think that happened on the previous generation, the test chip scope we are under discussion now.

Got it. Thank you so much.

I'll be back in the queue.

Okay. Goku, please.

Goku? You can unmute.

Let's go back to Goku later. Charlie, please, you can unmute by yourself.

Hello?

Hello. Charlie, please.

Yes. Okay.

Also congratulations for good results and also looking forward for the next-generation Trainium for tape-out. First question is really about this year.

Previously company provided some expected revenue range, right? We said that Trainium3, I feel like the mass production is kind of slightly ahead of schedule starting from late May.

Would company revise up the original guidance for the Trainium3 revenue contribution this year?

I would say for this year, Trainium3 is pretty much the contribution from Trainium3 is pretty much as we mentioned to investors. Because as you may know, the lead time for this product is very long, actually at current stage we almost place all the orders of this project this year.

I won't change the guidance for the N3 project contribution this year. For next year, yes, we do have chance.

Next year, you mean a chance to grow or chance to revise up your original targets?

The chance to revise up the previous assumption for next year's shipment and the revenue contribution of this N3 project, for sure.

Okay. What was the original target for next year's growth for this project?

Charlie, I cannot say the number here. That's the requirement by authority.

Sorry. Yeah.

Okay. Sure.

Also Johnny, you mentioned about Alchip's fits better for customer needs for COT business model, right? I think yeah, besides your current a big customer, there seems to be another big CSP are trying to consider in the COT, right?

What do you think is going to happen? Meaning the same, for example, TPU design, you cut into several compute die, IO die, and everyone can have their parts, or it will be several different SKU, right?

Just like today, right, you have a TPU v8T, v8i, and whatever SKU go for COT. What is a kind of likely development for customers' COT efforts?

I think the COT is already happened. To be honestly, our end customer, you can consider is one of the typical COT model.

They're doing the front-end design, doing architecture. They try to find design partner like us, foundry partner like TSMC, and we work together and find out the testing and packaging partner together.

I think another customer, I think Google eventually is already doing that. Unlike the original pure ASIC model doing with the Broadcom, Broadcom doing everything and Broadcom dictate the unit price.

The COT doing reversely. They try to figure out the cost of each element and add together and plus a margin then to providing to the service provider.

I think Google has so many different generation. They already try to minimize the dependency from the Broadcom type of pure ASIC model.

Eventually I think all CSP, Google included, will going toward the COT model when the volume continuously increase. I think that's unavoidable.

We see a lot of inquiry yeah from this account or even from other accounts. To answer your question, I think it's already happened.

Okay. Is it likely that one project, you have a two design service on the same products, or it will be kind of different SKU and customer is looking for one design service for one SKU?

Charlie, I'm guessing you are referring if it is a possibility for like Alchip responsible for the physical design of the compute and the other company responsible for the IO or

Yes, exactly.

On. I'll let Johnny comment on that.

I think sooner or later that's the right trend. Because each design contains so many elements, yeah, from customer perspective, yeah, they want to diversify the dependency.

You're right. There's some discussion, one of the potential customer discussed with us.

We can do compute and working with other people to do the IO chiplet. At the end, we can provide KGD type of model back to customer when they can do the integration by theirself or the ASIC house like us can do the final integration.

Many variety different models under discussion. I think the main purpose for customer is a cost saving.

With all those being the case or the facts, back to this core questions, right, why Alchip can be a better choice for customers' COT business model? If I can name some Asian competitors, one of them very aggressive in pricing.

I think they got one CPU projects, right, at a very low margin. You have another Asia peer since have their own proprietary SerDes IP help this customers to migrate to PCIe, right?

What's Alchip's value proposition in this COT business model? Okay.

I think the design capability, schedule control, and quality control is our gain. You'll hardly find any service provider doing more efficient than us.

If you look at the track record, Alchip pretty much is the one tape out the most CoWoS design in the industry. Talk about the design turnaround time and customer satisfaction, I have a confidence to outperform as the most of our competitor.

Also the pricing, yeah, even our customer consider we have a very competitive price. I don't think in the short term any ASIC vendor can come closer to us.

If a customer capable to do everything by themselves, they only require a service provider who doing a manufacturing only type of business, then I agree with you, so many competition. This kind of model will be very bloody.

For us, it's not what we are concentrating. I still thinking about we need to provide the value to our customer in order to make this business sustainable and more sticky.

If I'm only doing a manufacturing service, I don't think we provide too much value. At the end will be the pricing competition, yeah, would be allowed very huge.

One thing for me, Charlie. I will say by comparing Alchip with the industry peers, I understand that the other companies probably addressing the IP or another company also address their excellency in 2.5D advanced packaging.

Please note, until today in Taiwan or in Asia, not too many company have the track record to do the CoWoS production or design. I would say until today, Alchip probably is the only one in Asia who has the experience of doing the advanced 2.5D or even 3D packaging.

If you are talking about the expertise within the design and the advanced packaging, I would say Alchip probably should be ahead of anyone in Asia.

Just a very very last appendiced question. You seem to suggest that you're very good in CoWoS, right, but what if the next-generation TPU might still want to stick with PCIe, would that change the dynamic?

Then I'll go back to the queue. Thank you.

Okay. For EMIB, actually I talk with many investors.

For EMIB, we had experience. For our 7nm accelerator whose high-volume shipment in 2024, actually the last few batches of the 7nm accelerator, we applied EMIB.

If the customer wants EMIB, we can support it without a problem.

Thank you.

Thank you, Charlie. The next one, Goku, please.

You can unmute.

Hi. Daniel, can you hear me?

Yes, I can hear you clearly.

All right. Thank you.

First question for your follow-on 2nm project for your current big basic customer, I think as Johnny mentioned, it is a heterogeneous kind of design project. Could we talk a little bit more about Alchip's level of involvement?

Are you also involved in the IO die process or that will come from external vendors? It looks like there are multiple IO interconnect options.

Secondly, for the final packaging integration, including the IO die, compute die, etc., would Alchip be the primary backend vendor for all of these interconnect options, or the final packaging vendor would be different for different interconnect options?

Okay. Goku, your statement is correct.

Like I mentioned before, we are the sole design partner for this design. As if they're any party doing the IO die or whatever the chiplet die, they need to work with the main die.

All the interconnect need to line up. Unless another party doing everything by themselves.

Otherwise, whoever working on the IO die need to work with us. I think that's a common sense.

Right now, schedule so tight. We don't even have a time to deliver the original scheme.

No way this customer can working with other vendor to do any alternative solution at this moment. You're asking for the possibility if our chip tape out and go after the prototyping, are they consider about using other IO chiplet to apply for CPO or for other solution?

That's a possible, but definitely not at this moment. Both of us do not have any bandwidth to taking care any other approach.

The design I think in the final implementation phase, all the resource are very concentrate on the current implementation.

Understood. Johnny, if I may clarify a little bit, the final package design including 2.5D packaging etc.

will all be still Alchip, not really going out to

Correct. You're correct.

All right. Compute die, back-end design and final packaging, IO die could come from other vendors depending upon the interconnect option.

No, everything is done by us.

Oh, okay. Even interconnect IO die will be done by Alchip now?

You are right.

That is interesting. Thank you.

Secondly, on this new major hyperscaler customer that you now are sounding a lot more confident, how does this business model fit? Are you, like you mentioned earlier to Charlie's question, are you happy to be a KGD provider in this business model for the compute die or you think that you can have a chance to engage in the packaging, the final packaging design as well?

Goku, if you are referring to the production only, which means in the industry we call it GDS-in, if the projects comes from the CSPs, we will support the project. We will compete for the project.

However, we want to provide value. The GDS-in project usually the value we can provide it to the customer is low, it could be a very bloody battlefield for us.

We will consider the GDS-in project to be a knocking-door project for us to penetrate into we call it TD1, design plus production project. If you are referring to we are sharing a accelerator's different part with others, yes, we are interested in as long as we can provide physical design value or others to our customers.

Understood. Just to follow up on that, I think earlier Johnny mentioned the packaging capability which obviously Alchip is much better than most of the companies.

Let's say that SerDes IP still is proprietary from one of your competitors, are you able to still apply the packaging side? Is that kind of a supply chain possible where the SerDes can still come from a proprietary vendor or typically the SerDes needs to be third-party SerDes if you want to do the kind of more disaggregated COT kind of model?

We still believe many third-party IP vendors still working on the SerDes. We all know that the reason people consider some competitors can sustain project because of IP.

I think we've been working with our customer very closely. We are looking for variety of solutions.

We can source from the IP, do the integration by ourselves. We're even consider about using the KGD model like you mentioned about.

I think starting from the next generation and generation beyond, I think there will become a lot of new working models. I believe it's not only happened to our customer, but also happen to other CSP.

As if you look at those kind of expensive IP, most of the high-speed interface IP is all implemented in the IO chiplet. The area contribute the most expensive part still the compute die.

Starting from last generation, compute die is full reticle. IO may be 1/3, maybe 1/4 of the size.

I think maybe customer is willing to pay higher margin for the IO chiplet. For the compute die, I think go to the COT is the market trend, because there will be very limited dependency.

Pure logic architecture maybe only UCIe for the interface. There are very variety of business models.

I think all of them we can adopt.

That's very comprehensive. Thanks, Johnny.

Just one last clarification, for your N2 project that will go into production end of next year, will that be using 400-gig SerDes, or it is lower speed than that?

Goku, we cannot answer that.

We know the answer inside out, but sorry, we cannot.

No worries. Thank you.

I'll go back to the

queue. Okay.

Next one, Laura, you can unmute your microphone.

Yes, hi. Good afternoon.

Can you hear me?

Yes.

Yes, hi. Thank you, Johnny and Daniel.

I just want to follow up. I hear that, Johnny, you mentioned that CPU you got the two projects.

Can you share more with us what kind of a spec and the schedule? Would that relate to the new customers or existing major customers?

In the earnings call, we tend to not disclose the names.

Sure.

Laura, I can share with you one for server CPU, the other for kind of a consumer-type CPU. I think both of them are doing very well.

It's Arm-based CPU.

Sure. Also maybe from technology perspective, can you also share with us process node or going forward, are you also seen that these projects on Arm-based CPU may also require advanced packaging?

Thus Alchip may have a better chance to get more of that kind of business?

I believe so. The current generation I can share with you is N5.

It's not so advanced. Nowadays I think the CPU design, to be honestly, the design complexity is much lower than TPU.

If you look at the CPU floor plan and also the interconnect, it's pretty much each generation about the same, similar to application processor. For customer have enough design resource, most of customer can doing by themselves.

For the hyperscaler, if you look at the current generation, this is still 2D implementation. Most likely the CPU opportunity will be the GDS-in only.

Because customer do not need so powerful design resource to do the implementation. CPU nowadays also consider about using 3D.

If 3D technology is become popular, I think it will be our strength to implement the CPU. Our end customer also discussed with us on this area as well.

Sure. Thank you.

My second question is also about our 2nm of the next-generation AI accelerator. I think, Johnny, you mentioned many times that the design complexity is substantially increasing and the die size is also getting much bigger.

Can we assume

Die size is the same. It's a full reticle.

Oh, okay. Oh, reticle.

Reticle is getting much bigger. Can we assuming that for the next-generation, the 2nm projects we are doing right now, we'll see that the multiple ASPs comparing to this generation?

How should we look at the potential revenue scale in say 2, 3 years perspective? Also because of the capacity, supply chain management is getting more important.

Is there any restriction in terms of the capacity or even working capital requirement to secure the next few generations AI accelerator?

Okay. Laura, I cannot talk to you into the specs or floor plan of the next-generation N2 accelerator, but I can provide you some clues.

Nowadays starting from the N3 accelerator, the die size is into the full reticle. The die size for N2, N3 remains the same at the same full reticle, 800 mm².

However, the die the number of the compute die are different generation by generation. The the the the architecture of the chips, generation by generation, are different, like 2.5D or 3D or the chiplet architecture, something like that.

For sure, for the N2 accelerator we are designing, the pricing would be much much expensive more expensive than the N3, the N3 one. The volume of the chip in shipping would definitely be higher than the N3.

No matter the price or the potential shipment will be both of the both of them will be bigger than the previous generation, which is N3 currently shipping.

Yep. Sure.

Can you give us a rough idea how you see that this potential in terms of the business scale, say in maybe 2, 3 years?

I would say if TPU is to be told, the serviceable market scale is $80 billion, $100 billion, I would say probably for the N2 project, the scale could be at the similar range.

Okay. Following question is that in terms the working capital requirement and the supply chain's capacity, demand, and also since the market is getting so big, we also seen that more of the participants in this business always these customers?

For capacity, I would say this way. As we mentioned, it is a COT model.

Actually our customer receive very, very good support by the suppliers such as TSMC for both wafer and packaging. For now, for the next-generation N2, I believe in 2027, 2028, N2 capacity will also be very tight.

However, we are very confident that our customer will get really, really good support from TSMC. Also in terms of competition, yeah, like I mentioned many times, we have a confidence, high confidence, yeah, to sustain the design position for this account.

I think because the relationship, we've been doing for multiple generation, and customer did try other vendor then come back. They start to realize how reliable we are.

Nowadays as it's literally only two company produce the good ASIC volume. Our competitor, whoever claim they have a high volume, they all share the part from the other customer.

For our end customer, we are the sole design partner at least.

Sure. Thanks very much.

Very clear. Thank you.

Thank you. Next one, Jerry from UBS.

Jerry, please, you can unmute your speaker.

Hi, can you hear me?

Oh, yes. Please.

Oh, yeah, thank you. Thank you for taking my question.

Just want to follow up on previously I think Laura and also others have asked about the N2 projects. Because apparently Johnny and Daniel mentioned that the ASP could be pretty high compared with N3.

My question is that how will this impact your margin going forward? Because N2 N3 projects mentioned that the gross margin blended can still maintain somewhere around the low 20s.

As we're going to N2, if the ASP of the wafer or the overall design further increases, how should we think about the margin once we move on to N2?

We are just into the commercial terms discussion with customer. I cannot give you the answer.

Overall, the feeling right now is we can feel that the customer really, really appreciates our design effort. It is a very, very difficult design as Johnny mentioned many times.

With the value we provided to the customer and I would say we kind of have the consensus with the customer that no matter you call it ASIC model or COT model, we are in partnership kind of relationships. Customer kind of agree with that when customer success, we succeed.

We follow this kind of spirit. We don't worry too much about the gross margin for the next generation.

For the working capital, we do have ways to talk with the customer, to talk with the supplier. At the same time, after the GDR we offered like 1 month, 2 months ago, our balance sheet is getting better.

With all of those things, we think we don't worry too much, at least at current stage, about the working capital for the N2 accelerator.

For gross margin thing, we just described the current-generation design is N3 is much more complicated than our previous design, our gross margin improved a bit. For the next generation, the design complexity I think it's increased a lot.

I believe we can maintain or improve the gross margin accordingly. I think the gross margin thing can change very dynamically.

I think it depends on the competition. It depends on the overall size.

I still believe Alchip provide the most reasonable margin compared to most of our competitors.

Got it. That's pretty clear.

Then follow-up question on I think you mentioned about another CSP customer. Based on your discussion, can you give us some indication about the likely time frame for this business to kick in?

Because apparently the customer is already working with U.S. and also the Taiwanese fabless.

Their some product is already ramping or for this year, maybe for next year is pretty much finalized. I'm just wondering if you're going to get involved, when should we start to see this to contribute on your P&L?

Jerry, again, we cannot discuss the schedules here. Those are really, really sensitive things.

I really want to provide you some thinking. If those CSPs want to do COT in order to reduce their cost, actually they have to be capable.

Everybody knows right now that for the major four CSPs in North America, who are capable and who are uncapable of doing their own front-end design. If a CSP is capable of doing their own front-end design, when they do for the physical design partner, obviously they will talk to almost everyone who are capable to offer these kind of services in the market.

Actually we are one of them. Before you can consider NRE plays a very important role during entire project budget.

Nowadays you can see all the CapEx, the NRE become totally insignificant. Customer are willing to have a multiple supplier.

I think that's their strategy. Once they have a capability to do the chip, they're going to invite more and more vendor.

I think that's what they do. Everybody has a piece of that company.

That's a reality. That's what they plan to do.

Okay.

Jerry, because we have really limited time, we want to go for the other investors for the question. Thank you.

Thank you so much.

Next one, Lucas, KGI, please, you can unmute your microphone.

Hi, can you hear me?

Yes, we can hear you.

Thanks, Johnny. Thanks, Daniel, for taking my question.

The first question is that given that revenue concentration from your largest customer might exceed maybe 60% or 70% of total revenue, does Alchip have maybe specific strategies in place to mitigate single customer risk?

Okay. Lucas, let me give you a very quick answer.

The reality is if we want to diversify this customer, there's only one customer we are going to pursue. Otherwise, no any given customers in the world can diversify this customer.

I guess you knew who is the potential customer for diversification, and we are working really hard on it. I think all our competitors, we are facing the same question and same challenge.

Starting from next year, you will find out for the ASIC revenue, majority of the revenue coming from one customer. For me, just like what Daniel said, if you want to diversify our business concentration, we have to win another.

I'm thinking about this way. To sustain the current customer, I think it's the most important task.

That's why I keep on emphasizing our position is much stronger than most of our competitor because the relationship. I think I have a confidence to continuously provide our service.

We know each other for all levels, I think that's one of our value. To diversify, to be honestly, I have so many tape-outs, about 20 tape-outs a year in terms of customer count.

We can say we're pretty much diversified. If we talk about the revenue, I think it's no chance.

Only two company produced the highest ASIC revenue.

Got it. Very helpful.

I think the industry is also observing a growing demand for custom ASIC among Chinese automakers and IC design firms specializing in autonomous driving and maybe smart cockpits. With applications even expanding into robotics.

Could you share your perspective on this trend and how the company is strategically positioned to capture these opportunities?

Literally every single EV carmaker in China think about the robotics. Robotics can be all kind of form.

I think we discussed with our end customer the human-type of robot or industry-type of robot. I think you can discuss the chip is about the same.

The beautiful thing is the design we are doing, first of all, they put in the car, eventually they can put on the robot. I think our customer and also their competitor are thinking about this as well.

Can we move to the other because we really have a limited time?

Thank you.

If you have further question, please feel free to call me directly.

Thank you, Lucas.

Sure, thank you.

Next one according to the team sequence, Charlie, please.

I can let James go first. If we have time, I can do the second round.

Thank you.

Then James Roda, you can unmute your microphone. Thank you.

Thank you, guys. Yes, can you hear me?

Yes.

Just in terms of next two generations chip for the major customer, how do you think of mix shift between your product and the GPU? How can it change?

Then a follow-up question which is sort of related, how does a customer see total TCO of the ASIC chip versus the GPU? Thank you.

I would say for me, I cannot give you a very detailed answer. Based on our knowledge, the customer is working on increasing the percentage of their own ASIC usage for their AI data center.

I cannot offer you the exact percentage or number of the chips or number of the machines, but we knew the trend that the ASIC usage is getting up because of the cost.

Yep, and that's a public information. Their CEO also mentioned in the conference the saving is quite significant, like a 3 to 4x saving compared to their current supplier.

James, do you have a further question?

Yeah, 3 to 4x, is that on what basis? That's chip-level basis?

Which means from the cost perspective, to get the similar TCO, you'll pay only 25% to 30% 30-something percent of the money by comparing with you paying for the GPU.

Yeah. Do you have a sense of how that translates at a rack level or full TCO to serve tokens?

No, we are just doing the chip design. We don't

We do some rough calculation for fun, but it's not presentable. Sorry about that.

Probably you have to ask those system providers.

Thank you.

Okay. For the last one, Charlie, thank you.

Please unmute and questions.

Great. Thanks, Daniel.

Just to clarify the Q&A from Laura, you mentioned that the 2nm TAM could be $80 to $100 billion. Was that your comments regarding the scale of these projects?

Can you repeat again?

The scale of the 2nm, no matter volume or revenue, did you provide any comments on that one?

Honestly, we don't know the scale because we don't have the information yet from the customer. Even we do, we cannot share with you today.

If you really want a number, because the N2 project, we can feel the ambition from the customer. If the customer's competitor or peer has $80 billion for its market scale, I don't think our customer will be very, very far away from the number.

The overall CapEx number, I think they still maintain the highest in the industry.

Now I got your point. Is that the life-cycle TAM or it's a annual TAM can be that much?

It is a fair question, but I don't have an answer.

Okay.

Honestly, we don't have the position to answer this because it is customer's decision and the strategy.

Right.

In terms of spec, in terms of schedule, I think our customer has a very high confidence.

Right. How about your expected allocation?

I'm not sure if I get this right, but I feel like your production allocation, given Marvell, Qualcomm all get into provide this production service, I guess your 3nm production market share is probably 30% to 40%?

Charlie, we are just entering the discussion with the customer.

Right.

The customer will definitely listen to the audio video content of this earnings call.

Let's

do further discussions later, not in the earnings call.

Right. Okay.

That's exactly what we are focused, and I hope to deliver a very good news in the near future.

We look forward to that.

Thank you. Thanks for everyone joining our second quarter earnings call.

We do appreciate your participation. Thank you, and we will see you all again probably 3, 4 months later.

Thank you.

Thank you very much.

Thank you.