Yes, today we have Executive Director Lee Se-cheol from Citigroup Global Markets Securities with us. How are you, Director? Hello. Right now, we are at a turning point in the memory semiconductor market. Well, the stock prices reflect that. Yes, the stock prices have indeed fallen significantly, and there are many concerns in the market about macroeconomic conditions and oversupply. It seems that many people are viewing the situation quite negatively. Recently, some securities firms have been discussing winter-related topics, and those discussions have been ongoing for about a week. However, Micron's performance report has come out, and it looks very good. Micron's results were strong, and of course, this is based on their second-quarter performance, or rather mid-year results, when we look at it from our perspective, and considering their expected performance for the second half of the year.

The analogy is that the market has come out higher than expected, so stock prices have surged again, and it seems to have reflected on the domestic companies' stock prices as well. However, the issues regarding winter and such discussions were actually about next year. So, since everything seemed fine until this year, the views on how next year will unfold seem to be quite divided. To cut to the chase, will it decline next year? We believe that the second half of this year will indeed weaken, as the legacy numbers are quite poor. So, to conclude, next year will see significant decoupling in supply, the market, and among companies. Yes, we call it decoupling. The premium is likely to remain good.

It seems that the legacy market will continue to struggle. The difficulties with legacy products stem from the IT products we have been using, particularly consumer devices, which are not selling very well. Additionally, the diesel market is still quite saturated. The surplus inventory from 2022 and 2023 is still present in the channels. So, is that still available? Yes, that inventory is still there. Companies purchased it anticipating a turnaround, but when the market was expected to improve in 2023, particularly with the rise of AI and premium markets, the increase may not have been as significant as expected. As a result, much of that inventory remains in storage. Therefore, the outlook for legacy products is not good, and those who view it negatively expect that situation to persist until next year.

It seems that you are viewing it negatively now. However, I don't think it will go that way. What do you think about the legacy? Are we talking about DDR4 and similar products? DDR4 and now the high-end DDR5 products, those that fall below 5600, are considered legacy and are often discussed. I see, decoupling is a good thing, but there aren't any premium products available, so they are not selling well. The legacy products are somewhat less favorable, right? Yes, that's correct. But if the higher-end products do well, won't the production and supply of the lower-end ones decrease? That's right. Therefore, we believe that after the second quarter of next year, the third quarter, meaning the second half of the year, will be good. The reason is that companies will reduce the production of legacy products. For example, investments will be made to transition DDR4 to DDR5, which will increase the share of DDR5, while the share of DDR4 will decrease.

This year, the demand mix for DD4 is about 40%, while DD5 is around 60%. Next year, it is expected that the demand for DD5 will rise to about 75%, while DD4 will decrease to around 25%. What about the supply mix? It seems that the supply will follow that trend as well, likely reducing more DD4. Over time, even if DD4 is a legacy product, a reduction in supply will inevitably lead to changes in demand and pricing, both from a price perspective and a supply-demand perspective. There are discussions suggesting that we should have transitioned to DD5 sooner, meaning we should have switched production lines earlier. Some believe that this transition was delayed. Is that true? We haven't been able to manage the transition smoothly, and also DD5...

In a way, it's true that as we move into HBM, the supply of DDR5 has decreased. It seems that the market may have overlooked that aspect a bit. Since the second half of last year, the market conditions have improved consistently, right? The reason for this improvement is largely due to the demand from servers and AI accelerators, wouldn't you agree? That's correct. So, while the market portion is better, the rise of premium products has led to a corresponding movement in the market. Previously, companies had reduced their production of DDR4 RAM, and that had some positive effects. With the rising demand for DDR5 and HBM, the overall sentiment regarding demand has shifted positively. I see. Now, Micron's fiscal year last year...

When looking at the fourth quarter performance based on the standards, the server segment has significantly increased its share to an unprecedented level. The company has provided guidance that this trend will continue into the next fiscal year. Now, should we separate the AI accelerator market from the general server market? There are some challenges in viewing them separately since they are interrelated. The reason the demand for general servers is rising is that we are transitioning from DDR4 to DDR5. Hyper-scalers, for instance, weren't using it much until last year, but they have now started to adopt DDR5 in earnest. With the CPUs now available, it is also related to AI servers. The need for DDR5 arises because using AI requires not only HBM but also servers.

They provide support, and that started to come together. Meanwhile, the legacy demand is declining, so it's a bit of a downturn. In that case, the decoupling will occur because, anyway, there is high demand for high-end products, and they are sold out, while some low-end products will still have a lot of inventory left. How do you see the demand for regular phones and PCs? Well, I think the demand for conventional servers, conventional PCs, and smartphones won't be very good. It seems to be the case right now. However, next year, we can expect some improvements with new products like AI PCs and AI smartphones, which will have an increased adoption rate. Additionally, the components used in those products will be high-performance products.

Therefore, I believe that this will also drive additional demand. For example, if we consider memory capacities of 8GB, 12GB, or 16GB, the increase in capacity will be significant. If the capacity increases by 50% or 100%, that will definitely have a considerable impact on demand. However, it seems that there is still a lack of confidence in the market regarding this aspect. One thing that is certain for next year is that we believe the HBM DL5 will consume a significant portion of the capacity. Additionally, HBM products are changing frequently, with HBM 2 currently available this year and HBM 3 set to become prominent next year. In the second half of the year, HBM 4 will also be released, which will include 12GB and 16GB options. Furthermore, in the following years, HBM 42 will also be introduced.

There will be changes, and because there are continuous changes in the products, it seems that there will be parts that consume a lot of capacity. How do you view NAND? NAND is also in a decoupling situation. Mobile is not doing well right now; the mobile segment is quite weak. The demand for PCs is also not good, but the demand for ESD is okay. Oh, really? The demand for QC ESD continues to be steady, especially for servers. Yes, that continues to grow. You mentioned that NAND prices will dive next year. So, when looking at the overall memory market, do you expect profits to increase next year compared to this year? Yes, I think overall profits will increase compared to this year. I believe prices will also rise. Yes, so you have a completely different view, right? Yes, we have downgraded like everyone else, but we are still maintaining a mid-level outlook.

It seems like a cycle. It's a mid-cycle, but due to some legacy issues, there are some corrections. After this, I think it will go back up again. If there hadn't been any positive news regarding AI or HBM, or any movement in the market, what would have happened next year? We would have naturally proceeded with the next steps because the current demand for legacy products is not good, and there is a lot of inventory for DD4. The reason we haven't moved on to the next stage is that there isn't much inventory for Diesel 5 in the market right now. Next year's new demand is all looking for DD5. So, I'm observing a decoupling. If there is no change from DD4 to DD5 in the current timeline, and it remains just a commodity, then other houses might see negative trends as well.

There seems to be no choice but to proceed. However, the current situation is not like that. If NVIDIA had not been able to create the HBM ecosystem in the second half of last year and early this year, it would have been very difficult up until this year. This is because the existing demand is not good. New demand is emerging, especially with AI, and as a result, the premium market has expanded. Throughout my 25 years in the semiconductor industry, I have seen that every year, the prices of new products have dropped, and the costs have decreased as well. However, with the transition to HBM, the prices have actually increased. This is a significant change; it has been an enormous transformation. If the market had not exploded, I can't even bear to think about it.

It must have been really tough for you, especially since 2023 has been quite challenging. It likely took a lot of time to digest everything that happened. I wanted to ask about the past; back in 2006, 2007, and even 2008, there were times when the bit growth increased by 100% or 120%. In some years, it rose by 60% or 70%. Now, the bit growth is only around the mid-10% range, but there's still an oversupply. What do you think is the reason for this? Even last year and the year before, when things were difficult, companies in the DRAM sector saw bit growth around 20%. Why has the supply increased so much? From my perspective, the industry still seems to have a strong perception of memory as a commodity.

I think there are still thoughts about lowering costs and increasing market share. It seems that the market is still somewhat ambiguous, even though we are in a situation where we can raise prices. It has become an oligopoly, but it's still a bit unclear since there are global companies involved. For instance, there are companies in Taiwan and China, and these factors might be influencing the competition among firms.

Currently, the CXMT company in China is facing significant losses, but their capacity is reportedly around 100,000 to 150,000 wafers. They are said to be producing RAM at the end of the 10-nanometer process. What do you think the overall impact of that company on the market will be? While their capacity is large, it may not have a significant effect overall.

I don't think the overall market yield will be less than 1%. Is that due to the yield? Yes, the yield is not good. However, in the case of DD 4, it is expected to rise to around 3% because almost all production is now DD 4. But if the market is around 3%, it will still have some impact on supply and demand. Especially in the Chinese market, this will be significant. If the market is around 20-30%, the effect will be even greater. Looking only at the Chinese market, the proportion is around 99.8%, which indicates a substantial effect. So, when we focus solely on the Chinese market, there could be a significant volume coverage effect. How do you view YMTX? Well, the situation there seems better than YMTX's. Really? Yes, on the contrary, it seems that RAM will take quite a bit more time. It looks like it will take a long time. Meanwhile, the regulations have been a bit...

It wasn't much, so Land doesn't use Yubi. In that regard, it seems that Land can be used even if its performance drops. Therefore, if we compare the two companies, it seems that the situation for Biz is a bit better. Regarding the current memory market, in the case of YMTC, it seems to be somewhat better off, but CXMT is actually important for RAM, right? Its revenue share is significant, so could this company pose a threat in the long term, say three to five years from now? As you mentioned, in order to increase market share, they might engage in strategies like producing at a loss, but I’m not sure how that will play out. There are also perspectives that suggest there could be some impact. One of the threatening aspects could be the pricing.

It's a company that isn't sensitive to that, right? And since it doesn't significantly impact costs and profits, that's actually the part that is most concerning. In fact, there seems to be about a five-year gap. In terms of products, a five-year difference is quite significant. The gap still remains substantial, and in the case of DJ5, I think there will be some difficulties due to the lack of certain processes. Also, for premium products, high performance is required, but it seems that entering the market is still somewhat challenging. In the past, the rankings for memory were completely established, with the technology nodes developed, production timelines set, and market shares clearly defined.

When I looked at it, Samsung seemed overwhelmingly dominant, but can we still say that now? If we look at it by company, in the premium market, as you know, HBM has actually come in first with Hynix. That already shows a difference, doesn't it? Yes, that's right. The competitiveness of individual DRAM dies, or cost competitiveness, seems to be similar among global companies. Yes, it does seem similar, but in terms of product performance, as mentioned earlier, HBM is currently led by Hynix. We have to say that Samsung has caught up a lot. Yes, they have caught up significantly. In fact, do you think there is a possibility of widening the gap in the future? The company will certainly try hard to do so.

I think there might be some concerns. Anyway, recently there have been quite a few changes in management, so we will need to keep an eye on that. For now, in the IT sector, as I mentioned earlier, whether the gap narrows or widens, the pattern takes some time to develop. To catch up, a tremendous amount of effort is required. These are the challenges Samsung will have to consider moving forward. Currently, both Hynix and Micron have stated that they are completely sold out for next year's HBM. Does this mean they have contracts for a fixed quantity, or have they just set prices? As far as I know, it seems like the prices are fixed. There is also some discussion among certain securities firms about HBM supply and demand. I heard that when Samsung entered the market...

Yes, yes. But in my view, it is true that the volume is indeed increasing. The supply is going up, but HBM itself is somewhat semi-customized. Also, in the case of HM4, the customization will be quite significant. Each company using it has different specifications and uses different base types, so it will be quite complicated. I think supply chain management will become very challenging. In such cases, the prices usually cannot fluctuate like commodities. That makes sense. Additionally, as I mentioned earlier, the products are continuously changing. The CBI is coming out now, and in the first half of the year, HM4 will also be released. They are producing 1,024 and 248 units, right? So there will be initial demand issues, and the complexity of the products will increase. Therefore, I believe HBM will still maintain a premium next year.

Markets will be quite tight, and within HBM, the entry-level products may see some price stabilization. Yes, that's right. Last time, Executive Director Lee Se-cheol came out and gave a lecture about memory customers. For those who want to know a bit about the past history he mentioned, I'll share a link above, and you can check that out. In any case, you believe that next year, the profit margins will improve slightly compared to this year, right? Yes, it varies by company, but premium non-memory companies will likely have prices higher than expected compared to other companies. On the other hand, companies with a high proportion of commodity markets will be somewhat affected. Overall, this is quite different from past patterns. In the past, it was just...

Commodity, then it's just that commodities move, so it was really simple. So now, with that simple logic, data is coming out, and there are reports about downturns and such. I also applied that simple logic and thought, Isn't this how it would be if I take a shot now? Because recently, diesel prices have also dropped a bit. So, there are many who think that the signal is quite large, and there are those who view the second half as very unfavorable. However, when I look at it, as I mentioned before, there are many products becoming semi-custom and it's getting harder to create premium products. It's like a foundry run. Additionally, orders have already been placed, and products with fixed quantities and prices have already started to emerge in the premium market.

It seems that the foundry will move. We need to equip ourselves to recommend the spirit of customer service again. Yes, we can also receive the volume. It seems that the days of just producing and waiting for volume, like in the past, are over. If we still do that, it will likely become a very difficult environment. Micron, as they close the fourth quarter based on their fiscal year, recorded hundreds of millions of dollars in revenue for the 2024 fiscal year. They didn't specify the exact revenue amount, but I think in the market, they are talking about around 300 to 400 million dollars. They mentioned they would achieve around 300 to 450 million dollars, but I'm not sure if they actually reached that. It seems they did achieve something similar. Until that point, Hynix was taking the whole market, but next year, Micron is entering the scene, and of course, the overall market is expanding.

There may be some factors to consider, but when I asked about the impact of Samsung entering the market, they said, Since we have already completed our plans for next year, there will be no short-term effects. However, thinking about it from another angle, it raises the question of whether there is no room for latecomers to enter. What do you think about this? Recently, there has been competition in HBM samples and a lot of discussion about who gets there first. The company that establishes itself first has an advantage because it becomes the de facto standard. If a latecomer wants to enter, they have to adapt their products to that standard. However, if the company goes in first and aligns with the market needs, they are likely to succeed.

아무리 성능이 좋아도 들어갈 때, A 회사의 스펙이 유사하게 만들어 줘야 합니다. 왜냐하면 이미 AI 칩은 AI 회사에 최적화되어 있기 때문입니다. 이 때문에 제가 커스터머라는 얘기가 이런 겁니다. 어느 정도 최적화가 되었기 때문에 비슷한 회사가 들어올 때 더 어렵습니다. 사실, 먼저 들어온 회사가 더 유리하고, 뒤에서 오는 회사가 오히려 힘들어지는 경우가 더 큽니다. 삼성은 지금까지 사실 메모리 쪽에서 누구를 쫓아가 본 적이 별로 없기 때문에, 남이 먼저 만들어 놓은 것에 맞추는 것도 쉽지 않습니다. 커머디티 선점 효과가 있는데, 이 경우 H 선점 효과가 더 크다고 보셔야 합니다. 왜냐하면 커머디티 뒤에서 오더라도 비용과 가격만 맞으면 스펙은 다 도달할 수 있기 때문에 들어가는 데 문제가 없고, 그다음에 모듈화 되어 있잖아요. 실서 모듈화된 것도 이렇게 끼면 됩니다.

This is an embedded part, and it has been optimized for that, which causes a lot of difficulties for the companies that come afterward. Currently, regarding HBM, based on the bit growth rate, it seems to be around 6% this year and next year. If we consider the bit growth quantity, then in terms of revenue, it's usually about four times more expensive, so it would be over 20%. This year, it might be around 10%, and next year around 20%. Yes, I see. In terms of the overall DRAM sales share, yes, during the actual announcement, Hynix mentioned that the HBM sales share exceeded 20% in the second quarter. I see. Summarizing what you mentioned, it’s also very important to look at who will first receive the core and the revenue for HBM 4. Who will be the first to commercialize and release HBM 4 products? That will also be a key factor in the H4 market's de facto standard.

This is a part that can become a standard, and the companies that follow are really struggling to adapt to that. In fact, from what I've heard, the memory companies are saying that if winter is really coming next year, they need to seriously reconsider expansions and such. However, we don't see much movement in that direction right now. A few days ago, we released a report and adjusted our CapEx slightly. We originally projected it to be in the mid-40% range, but we lowered it to the high 30% range. What does that mean? The big CapEx investment for next year, regarding equipment investment, was initially expected to be about 45% for DRAM, but we lowered it to around 39%. The reason this number seems significant is that it all pertains to HBM.

Actually, there are many investments in HM, but I am considering whether we could adjust some of the commodity investments. This is because there are concerns about commodities in the market. The investment amount for next year is expected to be about 40% compared to this year. Typically, this can vary greatly depending on the cycle. This year, it seems like we were around the mid-30% range. For next year, we originally projected the mid-40% range, but we have lowered it to the high-30% range. We might be able to reduce some of the legacy investments. It seems possible to adjust the commodities a bit. Thank you, Mr. Jeong, for discussing this today. I look forward to seeing you again. Thank you. Thank you.

프리미엄 메모리 수요 급증, 레거시 제품은 고전...디커플링 전망

디일렉(THE ELEC)은
반도체와 디스플레이, 배터리 등
각종 전자부품 업계의 소식을
전문으로 다루는 인터넷 신문사입니다.

공식 홈페이지 http://www.thelec.kr
문의 thelec@thelec.kr

전자부품 전문 미디어 디일렉