Third Bridge专家称AI数据中心功率引线框需求强劲但未披露公司级新增订单

[[00:01:39]]
How have lead frame shipments changed across different industries over the past 2-3 years? Are there any key trends that you see?

[[00:02:00]]
For end of 2024 until 2025, actually, it's a very bad year. The demand is dropped very badly until next period introduced AI to the world. I think start from there, the whole market start to get excited. Not only AI industry, but I think the EV automotive market in Europe also start to increase. I think overall, we are seeing the European automotive market has increased about 5-10% for the past few months. I think they are still growing. For AI, definitely, because of the data centre, they need a lot of lead frame, especially power lead frame, QFN. Yes, the market is booming or you assigned to each. I believe this market, for automotive, I think in the past, they are very badly because China automotive, especially EV car, export their product out to the whole world. The business of European and US automotive industry is very bad.

For those consumer products, I think consumer products also affected by China. The appliance is very cheap and we are having a hard time at the moment. Now, it looks like when AI is growing, it also help the automotive industry grow. Compared to last year, I think the market is booming out at least about 15%. For AI, I think they are ramping much more than that. The problem is for AI qualification time, we still do not sure how long is the exact qualification time. Automotive, that is two years. 1-2 years, there's a standard. For AI, we think they are longer than consumer but shorter than automotive. I think some of the products start mass production in the past few months, so we still cannot get a very exact picture how is it.

[[00:05:31]]
In the past 6-12 months, have you seen any trends in AI data centre power lead frame shipments? Have you experienced any volatility in the market so far?

[[00:05:57]]
You mean AI product? Yes, they are booming, especially in those QFN power like TO, Dpak, Tpak maybe. They are growing because for AI, especially data centre, they need a lot of power lead frame.

[[00:07:05]]
Have you experienced shipments for AI product falling sequentially for more than two quarters over the past 2-3 years? Does that happen?

[[00:07:38]]
Maybe you can say just for past one year because 2-3 years, the growth is not so high as today because when Nvidia, I think we now mostly are doing Nvidia business, either to the sub con or the main con. They have a lot of different QFN design. This needs a very precise plating line and also the packing technology to support. Some of the sub con, I understand they just opened a new line to support Nvidia products, so yes, the growth is quite rapid.

[[00:08:42]]
Where do you think we are in the cycle today for AI data centre power lead frame shipments?

[[00:08:54]]
For AI data centre, I think they are using Toll, they are using QFN, they are using Toll and some of the TO package. These are sometimes using thick copper material. The problem is the raw material market might not able to support because they are not prepared such a high ramp in the past. Yes, material is an issue. The demand is there. Unlike traditional lead frame, which is one standard IC equal to one lead frame, but one of the AI-powered lead frame, they might need a lot of lead frame to support for one system.

[[00:10:12]]
I believe that AI products have more cyclicality, but some believe there's no cyclicality, and it's more of a structural change growing demand for AI lead frames. What do you think?

[[00:10:52]]
You mean, is it for AI data centre power, where do you think we are in the cycle today, is it?

[[00:11:06]]
Yes.

[[00:11:08]]
I think we are still at a very early stage offering because they changed very rapidly and they are still a very big room to grow.

[[00:11:25]]
What do you expect for the demand of lead frames, especially for AI and data centres, in 2-3 years? How much do you think shipments or volume will increase on a yearly basis for AI products?

[[00:11:57]]
Data centre, I think everyone is working into the data centre. Data centre is supposed to be run quite rapid unless, but there is also a concern for the environmental friendly and power concern in the country like just take an example my hometown in Malaysia. We have a lot of data centres in Johor Bahru because they are next to Singapore. When they have a lot of data centres, the problem is we're always having shortage of water or no electricity. This is all the impact. Yes, they can grow very rapidly, but the problem is when they grow, is there any NGO to do a protest, say, "You are un-environmental-friendly, you affected the residents' life"? In the other words, because AI, we are growing well in this era, so they won't be stopped. They were still growing because you need the data centre, whether they are in the sea, now we understand they can build in the sea. Whether they are in the sea or somewhere else, but they still need to grow because we are in this era and everyone work with AI and there's no choice. There's a trend.

[[00:13:45]]
What sort of CAGR do you expect for lead frame shipments over the next 2-3 years?

[[00:14:02]]
For the next 2-3 years, I think it will be grow about maybe 30% or more, I think, just for AI server power.

[[00:14:28]]
You mentioned there's 30% or more growth on a yearly basis in lead frame shipments. Considering production and supply, do you think there's space to manufacture enough lead frames to match the demand?

[[00:14:51]]
They have to because in business-wise, we will have a short-term shortage of material sometimes. People will still reserve the capacity and expand the capacity to support when there's a demand. Just like what I told you just now, some of the sub con, we understand that they are just built the line to prepare for this AI booming.

[[00:15:29]]
Let's say demand is 100. How much manufacturing capacity do you think there is currently?

[[00:15:44]]
AI have a lot of different process because they are more on the high precision and unique tech plating process. If you ask me to 100, I think we need more than that because not every lead frame supplier can build the product for AI. There is some requirement and it also depends on what is your operator scale. Otherwise, your loss also be very high.

[[00:16:25]]
Is AI product demand diversified across enough customers and applications, or is the demand focused on limited customers?

[[00:17:00]]
No, a lot of customers, just like what I told you, not everyone can do. It will depend on the (audio distorts 17.15) productivity. The customer, I think at this moment, you go everywhere, more or less they will tell you they have AI products. None of our customers will tell you they don't have it. I think AI is growing and everyone want to get a piece of the cake, just like last time, automotive. Automotive, you need some capacity to build the automotive product, just like AI. It's the same.

[[00:17:58]]
Are there any specific lead frame suppliers that should be highlighted, perhaps with increasing order volumes?

[[00:18:13]]
A lot in the market, a lot. Even in Malaysia. Let's talk about the lead frame supplier. Lead frame supplier, not many people can do because, as what I say, it's due to the technology. They need the capacity. Some of the customers like China lead frame supplier, they are more on this low-end product, so they are not in this competition. Then let's go to the customer. The lead frame customer, so who buy lead frame, yes, there's a lot in the market and they are actually expand their capacity to prepare for the booming. I think, as what you know, one of them is Broadcom. Broadcom is support, I think, the GPU for Nvidia. They are quite famous.

[[00:19:47]]
You mentioned that the lead frame market is growing at 30% or more in terms of volume. The overall pie is growing. Does that mean that lead frame manufacturers' shipments are equally distributed to suppliers? Are order increases concentrated in a few specific companies?

[[00:20:48]]
No, because some of the supplier… You have to understand it's not only AI is booming. Automotive also booming as well. Automotive is only also booming. Not only for AI, so some of the supplier, they maybe don't have the capacity to support AI, but they're able to support automotive. The overall booming maybe is about 10-15%. Compared to AI, AI we're getting more volume, but you have a lot of production line. Some of the line you are running maybe SOIC, which is high volume but low margin, so because we have a lot of line, we are not shifting all our line to run only AI. Overall, yes, they are increasing, orders is increasing but not only for AI. It's also for other industrial like maybe industrial and those Internet of Things, maybe automotive, not necessarily it's only AI. I think the three companies you listed out there, they're supposed to be benefit from AI.

[[00:22:31]]
Focusing on AI product, who benefits the most? Is it more equally distributed?

[[00:22:46]]
I think it's not equally.

[[00:22:50]]
How could the overall increase in shipment volume affect overall volume share? Which company do you think will take more order intake compared to others?

[[00:23:12]]
I think Shinko and Enomoto may be taking more orders of AI than CWTC because CWTC, they are famous on high-volume products. I'm not sure whether the new expansion having any new machinery, which is able to support the AI needs. According to what I understand, they are actually more competitive in price with higher volume. They are mostly on those maybe SSOP, SOIC product, which is high-volume and competitive price. Enomoto, I think they are in niche market. Actually, I don't compete with Enomoto directly. I think they are more on niche market. They support Philippines with very cheap price because they are having the raw material support by I think it might be (inaudible 24.35). They have a raw material supplier to support them. In this three company, Shinko is having a high technology. They not only lead frame, they're also having substrate, so using more substrate than lead frame. I think Shinko is in a very good position.

[[00:25:05]]
Could you briefly walk us through the cost structure of lead frames, particularly AI products? How has it been changing over the past one year after the AI boom?

[[00:25:40]]
For standard lead frame, actually, we don't have much changes because the structure is majority from material. I think material is about 50% and production costs around 20-30% and maybe some of the annual loss. This structure is a standard structure, but some of the company, they are running in very high volume like CWTC. Their material costs may be cheaper because they are running that non-stop, so their production cost will be lower, much lower. The cost structure may be different. Maybe material will be 50-70% and balance will be the production cost because they are running in high volume. They need to stop their production line every time. They don't have a production cost losses for some products like AI and power. We talk about QFN. For QFN, they need a very precise machinery and you need skilled workers. In this way, the material cost is not at 50%, maybe it's 35-40% and balance will be a production cost annual loss because they need a high precision. Maybe some of the lead frame, they need to run two plating type or they are running 2-3x operating process, so their processing cost is very high. For the power, we are taking longer time to stamp or etch because the material is thicker. Lead frame costing structure is depending to the product design and the process requirement.

[[00:28:18]]
What are your thoughts on unit economics? How much does the lead frame cost, both standard and high-end frames, including AI products?

[[00:28:51]]
How much is the cost for high-end product? QFN, it maybe can up to USD 6-7 per frame for AI product. Yes. QFN we usually talk about frame, how much per frame. It depends on the process because, for example, one of the QFN hybrid, this one, they need the PPF plating and also the AG plating. They need to go through at least two plating process, so their processing cost is very high.

[[00:29:57]]
How much does a standard lead frame cost? You mentioned that CWTC products are a bit cheaper. How much might they cost?

[[00:30:16]]
Actually, I not heard anything in the market they provide for hybrid for the AI QFN. I'm not sure. The standard or maybe say like this, their price are about 30% cheaper than Shinko.

[[00:30:54]]
You mentioned that QFN [quad flat no-lead] is USD 6-7 per frame, and CWTC is 30% cheaper than Shinko. Roughly how much does CWTC cost?

[[00:31:05]]
I think you can't compare CWTC with AI QFN because I don't hear in the market, they are doing this.

[[00:31:24]]
Has there been any increase in the USD 6-7 per lead frame price tag for QFN [quad flat no-lead] over the past 12 months?

[[00:31:36]]
No, because the price is almost there and the increase is only based on LME, but our pricing will have an adder. When LME increased, the unit price also increased. The LME dropped, the price also dropped.

[[00:31:57]]
In the next 2-3 years, how much do you expect the QFN price to increase?

[[00:32:25]]
I think they were still stable because this QFN not only use in the AI, they also use in automotive. For past few years until today, the price range is almost there. They won't change much.

[[00:32:41]]
I assume material prices are quite volatile. How do you expect this to impact lead frame prices?

[[00:33:12]]
The copper material, actually, they can be starting sudden because commodity price are based on the market situation. When there's a war, the prices ramp out rapidly. Just like recently, silver prices ramp up from USD 30 up to USD 80 in a very short time because the Iran-Israel war started. It's hard to say whether the price is there or not, but the structure should be the same. The pricing structure will not change, but LME, yes. LME metal has changed for those commodity product. I think the commodity product prices will be different, but the lead frame structure, it won't be much difference.

[[00:34:27]]
You mentioned that lead frame prices will be stable. What about production cost efficiency? Would you expect any decrease in production cost if we see more efficient production lines or product activities in the coming 2-3 years?

[[00:34:54]]
I don't think so because, firstly, the lead frame itself design is high precision. They also are very, how to say, it's not complicated. It's more on the production process. You need to get the skilled workers and skilled workers, when you train, maybe he will change the company later, so you need to get more salary to employ another one. I think the cost is there. I don't think they will become cheaper. This is my idea of AI.

[[00:35:48]]
I assume that labour costs will increase. Would you agree?

[[00:35:56]]
Yes, labour cost will increase. You also need a new machine. Just a lot of suppliers don't invest in the past few years. Capacity is not enough to support today's market.

[[00:36:19]]
Your outlook seems to be that machine installations and labour costs will increase, but won't increase lead frame prices. Is that your view?

[[00:36:42]]
What I say is the lead frame costing structure is there. The cost structure is affected by the LME because the copper price… I'm not sure how much you understand about lead frame costing structure. The lead frame cost structure, actually, one part of what we call it the pricing is LME price. LME price like copper, silver, palladium, gold, you can put it aside without a consideration. In the processing costs, I think they were getting higher, even though the product efficiency is improved. That means your cost will be lower down, if your product efficiency is improved. You don't forget, when this product efficiency improved, you still have the investment in the new machine, the investment in the labour costs. Actually, they are more likely not much changes in overall.

[[00:38:27]]
Could you provide an overview of lead frame material suppliers in the market right now? How has the landscape changed over the past three years?

[[00:39:04]]
I think it won't be much changes because power lead frame, maybe you need 0.5mm material and 0.5mm usually not many people use unless you need to stamp a power lead frame. The market may not have it now, but they will catch up very fast because they can run the OT, they can extend their production line or their production hour, just to build out the inventory. I don't think inventory having any but shipment may have problem. Inventory, no problem, but shipment may have problems. Shipment may have problems because you know now is the Iran-Israel war. The vessel in this area is not as many as what in the past because some of the vessels may still get stuck in the Strait of Hormuz. Every time we need to book the vessel, I think now it's getting better than last two months. It's so difficult to get the vessel. You have to book earlier but you can't necessarily get the vessel at the schedule you wanted. Now, we're still having this problem, but it's not as serious as past two months because of the freight, everything. I believe some of the vessel might not able to cover the expenses or they have no cash to run the business. This might be a problem.

[[00:41:03]]
Overall, who are the primary suppliers of the key materials used in lead frame products?

[[00:41:14]]
They are using C194… I think the raw material is not an issue because they can just build out the inventory. They can just build up the inventory in a very short time. I don't think raw material is an issue. We are using very common C194 or C7025. It's a common material. I don't think a lot of suppliers can supply. I don't think there is a big issue.

[[00:41:51]]
Would you say there are no companies that specialise in these key materials?

[[00:42:00]]
No, no.

[[00:42:05]]
Are there no difficulties in procuring these key materials?

[[00:42:22]]
Maybe it's a short time, but it won't be a long term.

[[00:42:29]]
Could you quantify 'short time'?

[[00:42:36]]
I think it's about three months, they can cover already. 3-6 months, they can cover already. They can build out the inventory already. I don't think there is any problem because now our common lead time for raw material is 3-6 months. I don't see any problem on that.

[[00:43:04]]
You mentioned the lead time of lead frame raw materials. How long is the lead frame manufacturing lead time currently? Has it extended or shrunk recently?

[[00:43:25]]
I think most of the supplier capacity is already used at around 80%. There's still room to go. CWTC, I understand they are just expanding their KL plant. They should have lots of new line, either moving from China or local. I think the market should be able to support, but it's not all the supplier able to support this AI product, as I mentioned to you earlier. Someone may benefit from that and someone may be not.

[[00:44:19]]
How long is the lead time for the lead frame manufacturing process?

[[00:44:28]]
Lead frame manufacturing. If you are asking QFN, I think it's about 10 weeks already. Market now, even one of the Taiwanese competitor told me, they inform their customers six months, they accept. For QFN, the lead time is quite long now. I think yes, 3-6 months.

[[00:45:08]]
What about the 10 weeks?

[[00:45:11]]
About 10-12 weeks are minimum.

[[00:45:17]]
Would you expect this lead time to continue to expand?

[[00:45:31]]
No, because those already expanded their line. They are just waiting for customer to qualify the parts, so they can run the new line. CWTC, they are doing that. When customers qualify their new line, they can move parts to the new line. The lead time will be getting better.

[[00:45:58]]
What is the current capacity utilisation across different lines, including stamping, etching or plating? What are the bottlenecks that you see, given AI-driven demand?

[[00:46:58]]
You mean how many products are utilised for AI, how many products are running for others?

[[00:47:05]]
Yes.

[[00:47:07]]
I think it's about 10-15% because the AI product, some of them are still under qualification. It's not really into mass production for everything yet.

[[00:47:34]]
What about the lead frame manufacturer operating ratio within factories?

[[00:47:59]]
The frame manufacturer, I think it's around 10-15%. It's not that much. It depends on the product. I think overall not more than 10% because they're still having other products to run. AI is just the beginning. AI product is still at the very beginning stage. They are still not fully utilised in the production line. I think 10% is very high expectation already.

[[00:48:40]]
You mentioned that 10-50% is the operational rate for AI product. Is that right?

[[00:48:51]]
I think not more than 10%, sorry because at this moment, the AI product is not booming so much yet. Yes, the demand is very high, but some of the parts are still under qualification. Maybe until next year, yes, but at this moment, I think it's still not so much.

[[00:49:28]]
How might capacity utilisation shift in the coming 2-3 years? You mentioned that some products are under qualification. Taking that into account, how far do you think the rate could go?

[[00:49:50]]
I think in 3-5 years' time, maybe it will reach about 30%.

[[00:50:08]]
What is the operational rate for products that are not for AI, such as automotives?

[[00:50:25]]
Every lead frame supplier having their own, their main products. Basically, in the market, maybe 15% is for automotive and others will be consumer and industrial.

[[00:51:18]]
Could you give us an overview of Shinko Electric, CWTC and Enomoto? What are their competitive dynamics in terms of lead frames?

[[00:51:40]]
Actually, I'm quite surprised you will compare those three companies because it doesn't look like the main competitor of Shinko because Shinko is quite strong, they have substrate and also they have lead frame. Shinko is more high technology. They provide a good quality. The price, I think, is higher if compared to CWTC. Shinko also expanding their KL plant to support the Southeast Asia customer. I think Shinko is having more demand in the automotive and AI products. CWTC actually, to be frank, I never heard from my customer they compete on these AI products. They are actually more on the competitive price and with high volume. I think they are more flexible. They are expanding their capacity in KL also to support the SEA customer, just like Shinko. For Enomoto, I think most of their customers, they are quite strong in Philippines because they have a plant in Philippines. I think they are only doing stamping, is it? I didn't hear they are having etching frame. I think they are focused on the niche market. They are strong in providing a good quality of power product. They are having a high accuracy stamping and good quality as well. Actually, I'm just curious why you take this three as a competitive lead frame supplier because it looks like they are focused on a different area.

[[00:54:04]]
I'm looking for a broad understanding of lead frames. Which companies do you believe have a great value or product positioning in AI-specific products?

[[00:54:37]]
Shinko will be my main competitor.

[[00:54:42]]
How much does Shinko dominate the lead frame market in terms of share?

[[00:54:58]]
I can't remember what is our range together with Shinko, but I think Mitsui is the first in the lead frame supply and Shinko is the second. The market share, I can't remember. There's a semiconductor ranging every year. I think CWTC is second or is third or fourth. If you mentioned about percentage, sorry, I already can't remember.

[[00:55:38]]
What lead frame weakness do you see in Shinko? Do you think the company's weakness resides in design, sales strategies or capacity? How would you evaluate it?

[[00:56:14]]
I think compared to some of the lead frame companies, Shinko, they are just spending their fund in Malaysia. Unlike other lead frame companies, they have a lot of plants everywhere in China, in Taiwan, in Malaysia, in Singapore, in Philippines. Japanese, sometimes they're having a language problem when communicating with a customer. Some of our customers, even the orders is big, they are not going to use those. Having a communication problem with them, a very big customer in Malaysia, I asked them why some Japanese suppliers don't use them, they have a heavy communication problem. They don't use. I think this is one of the weaknesses. Shinko, I'm not sure because I think they are choosing customer. This is my feeling. I'm not sure because when some of the customers, they're having a less product, they also complain to me that Shinko asked them to find other suppliers because the lead time is very long. They give a very long lead time at a very high price. I think they are also focusing their customer list. They will take those customers they wanted and maybe some of the low-value customer, they might not interested. This is my feeling, but I cannot confirm whether this is right or wrong.

[[00:58:20]]
In the coming 2-3 years, how do you expect key buying factors to change, especially for AI products? How might that reshape competitive dynamics among companies, particularly for Shinko?

[[00:58:53]]
I think Shinko is a very good position in this AI era because they are supplying the substrate. AI customer, they will need substrate more than this lead frame, so I think they are in a very good position. They have a stamping, can support the power. They have a production line. They can support the substrate and also the lead frame. I think they maybe will have a very good future in the next 3-5 years compared to the competitors, but now, they are more concentration on the substrate, instead of lead frame.

[[01:00:04]]
Could you provide us with your overall lead frame market outlook, especially for AI products, over the next three years?

[[01:00:28]]
For the power lead frames, they will be just automotive, they're not able to change so rapidly. For the others, AI design, the pattern might change very rapid because just a few months ago, we talked about GPU and now we are talking about ASIS, I think. The world is changing. AI also changed very fast. When you invest in this AI product, you also have to take note what is the product life cycle. It may be very fast.

[[01:01:22]]
The time has come, so we're going to end this Interview. Again, thank you so much for joining this Interview and see you next time.

[[01:01:31]]
Thanks.

Transcription ends at 01:01:35 of the recorded material