Transcription
I think a sensible place to start is why now? You've been thinking about a US listing for some time, but why put SKH Highix in front of US investors now?
Well, we've been waiting for that for a long, long time, and since I acquired that Heinix about 15 years ago. So, this is one of our milestones. Well, we have to go to access the US capital market. So now, dream comes true. So people are asking why now? But well, this is kind of the earliest moment. Well, we can have the earliest moment possible.
26.5 billion is a lot of money, but in the context of SK's commitment domestically in Korea and internationally, not so big. Can we expect multiple returns to ADRs? The sale of multiple rounds of ADRs?
Well, yes, I'm expecting we can deliver those things too. So, uh, we try to expand our ADR size and also better that actually requires a better return. So, once we have a better return, then there is more demand. So, first thing is that well, we have to keep the stock price stable and, well, hopefully in the long run that we can have upside potentials. And I see that well, this moment, so, and this is the AI era. So, this AI era is actually spiking a lot of demand for the memory. So, this is kind of time for us.
Is this something near-term? Did you, with the leadership group and your CFO, say, "We will need to do this with regularity," or you haven't discussed that yet?
Well, to tell you the truth, yeah, we've been discussing that one, but I cannot just say that in detail yet.
I appreciate that, Jamon. Jay, look, the US listing gives SK a new source of equity capital, but what about the broader US capital markets? Do we see SK Group, SKH Highix as well, become an issuer of global bonds, global debt?
Well, that's possible, but I guess right now, well, we have enough cash for that. The SK Highix, some other company could actually be issuing another bond and well, yeah, access the capital market.
Chairman Chay, right. Bear with me. I'm going to get this out for our audience around the world. This is why we're all here, right? [clears throat] This, you can't even see it, tiny HBM4E high bandwidth memory, 12 layers in this case, right? The whole world knows about this. [laughter] How severe and how tight is the supply right now of HBM? And more importantly, how long is that going to last? Do you have a sort of a year where you see the supply situation changing?
Well, if you really understand the all the requirement and the condition for the HBM memory type, and you have to understand the AI side first. So, people know that the large language model.
Yes.
Yes. But that large language model is a very important thing. But I guess in the future, so, the more we use AI, and AI actually requires more inference phase.
Yes.
But I see that memory, the mean is KB caching. So, well, it generates a lot of KB caching due to keeping the performance of the LLM. Then you should have a large volume of KB caching, and you have to actually, this is kind of memory. In people that as the same thing, we store memory in our brain. And yeah, AI needs that kind of KB caching. That memory has to store somewhere. So, if you have a better size of the memory, then it has better performance, as simple as that. And that memory cycle is well, it's turning to well, it's a different, different cycle. I shouldn't say that the cycle is different momentum. So, and now back then, in the old days, the memory chip is, memory chip demand is actually depends on the number of people, number of hardware sets.
That was a consumer electronics story, in part.
Yes.
Boom and bust.
Yeah. Yeah. Well, mostly, well, you add that to your own smartphone, but you don't need tens of different types of smartphone. You only use just one, like a PC.
Right.
So, the demand is actually limited by the number of people. But in the AI era, it's different. You may use your own AI agent in probably a few years from now. Then you may have depends over hundreds of different types of your own agents. So, every agent needs a lot of memories because they create a lot of KV caching. So, you have to actually store somewhere that actually needs a huge amount of memory. Well, that's the AI era. So, I guess that memory business is turning into a different stage.
You still call it a cycle, though. That's the demand side. A reason to do the ADRs is to raise capital, in part to boost capacity. Your industry peers, Micron, had said very recently they have line of sight to supply improving 2028, the other side of 2027, in part because of the capacity that is coming online. What does SKH Heinix see?
Well, we didn't talk to our customers. So, where we just announced that a few months ago, so we're going to double up our capacity within five years.
Yes.
So, uh, that means that what we've been doing that business for more than 25 years. And then, well, with the next five years, we're going to double up our whole capacity. This, well, people see that that's unrealistic and probably oversupply, and people worry about it. But my customers said that's not enough. We need more. So, double up is not enough. Actually, what they want us is, yeah, five times, six times. So, the more is the better because that well, demand will be exponentially grow in the future.
Based on that customer demand. What do they see then? If they are the crystal ball, the lens to the future, they believe that their level of demand will stay where it is 2029, 2030. How long does supply tightness respond to that?
Well, nobody expected. I don't know what's going to happen in the next years. But I'm saying that well, I have some confidence that the demand will grow and our supply capacity will never catch up. So, I don't know how long. But well, if I think it's going to be until that our society, human society has set some settlement with AGI, then well, I guess that's kind of normalized at the old cycles. But anyhow, until then, well, we need a lot of memories.
You're with us live on Bloomberg Television and Bloomberg Radio worldwide, and we're in conversation with SK Group Chairman Chay. And we're talking about a record for SKH Heinix in terms of US ADRs. Does, um, take me back to 2012.
Mhm.
SK acquires Heinix. At the time, people questioned that decision. What, what did they not see? What did they miss that you saw about the potential of that memory market?
That's kind of a bold decision, and it's, I think that I made a really risky bet in a sense of financially, because that the company was really suffering from deep debt and also had to fight for its own survival. Because at, well, 15 years ago, that company was under control by the banks. It's simply a chapter 11. And, uh, yes, but in Korean sense, what we were doing, the banks is a workout program. So, but they were doing that for more than 10 years, you know, the workout programs. That actually means that nobody really wants that portfolio because it's too risky. Reason why is, well, memory business is cyclical, and that actually suffers from uncontrollable cyclical business cycle. And it requires every year multi-billion dollar capex, no matter how. So, even though you don't make money, but that business requires machines and expansion and adapting new technologies and buying new equipment. And they have to spend huge capex. That's why nobody really wants that kind of dangerous business.
But you did.
But yes, I see a little bit differently because that well, if you see from a different angle, this memory manufacturing is the most sophisticated manufacturing business. So, if you really do it, then you can do anything.
Like, if you crack the technology, the process?
Yes. And also, so, the whole process of that manufacturing, as a, it actually, wow, once you make those little chips, but chip inside that 380 billion transistors in it.
Yes.
So, that's the business. So, how we're going to create those kind of things? And it's a small, yeah, small size, you know, silicon things. And we actually put it in the, well, huge size of the world, the transistors. That's a state-of-the-art technology. So, once you're doing that, I can manage anything, and well, I can actually challenge anything. Also, in the digital world, create a lot of memory in the future.
Right.
So, that's you saw that.
Yes. I just bet on that. And well, people needed the memory.
Yeah.
You talked about managing the process, and the idea that, you know, at that time in 2012, debt was a very big consideration in the market. It was boom and bust cyclical. How much has this, you've been asked about this a lot, but the idea of the long-term agreement with customers changed that history?
Right. But there is a supply shortage, and every customer wants to have to secure their own supply and cover up to their own demand. So, they're asking us that well, we need some kind of long-term contract. It's not suggested by us, and well, they actually require that.
It's them that's making the proposal, not SK?
Not SK.
So, they are more willing.
Mhm. Well, we also love to do it because that it's not a cyclical business anymore. If we have some long-term agreement, then well, even though it's a downturn, we still maintain our volumes and a certain level of price. So, that actually gives us a different momentum, and that actually created another opportunity for us to change some of our business model.
What about SK's commitments? Where are you going to prioritize your investment, you know, domestically versus here in America, not just with the proceeds from this US share sale, but generally with your balance sheet?
Generally, outside of Highix, the most important matter is the AI data center business. So, we announced almost a $1 trillion investment in the next 10 years to set up AI data centers. My target is about 15 gigawatts in Korea, but I guess about 5 gigawatts outside of Korea. So, I'm looking for this, whatever area, including the US side. And, um, yeah, if there is a good chance, and the problem is that the one main thing is power. So, yeah, power and the price of electricity is a very important matter for these AI data centers. But once we build that AI data centers, then well, it can actually have synergy between the memory chip and this infrastructure matter. So, that's what I'm doing in the future.
We're live on Bloomberg Television and on Bloomberg Radio, and this is Bloomberg Tech in conversation with SK Group Chairman Chay. You talked about the data center, the issues of the data center. This is an SK Group approach.
Yes.
So, you're talking about SK Group looking more broadly at the AI stack, not just memory.
Yes.
Where else will you focus your attentions?
Uh, well, a company like SK Telecom, and well, they are doing their study on the, or putting their resources on their own LLM, also the applications, and the physical AI side.
Yes. Robotics.
Yes. And the healthcare side. Well, there's a bunch of opportunities. And what we've been actually trying to pursue all those business opportunities in the future.
An issue here in America and in Korea is talent. There is a deficit of skilled workers in the chip industry in America, and a company like Micron knows that there is a lot of talented people in Korea. How are you going to manage that? And again, your investment plans for here in the USA?
Well, to secure that, bring that more talent is well, ADR is one thing, one good option.
Yeah, options.
Yeah, options. And well, ADR gives also some kind of financial play options, like well, even stock options, then well, we could actually be using that of the ADR directly.
Right.
So, it actually attracted a lot of people's, a lot of talent to our group. See the ADRs is also about public perception of SK.
Right.
That's another thing. And perception matters. So, well, also, well, we have to spend a huge amount of money on recruiting a lot of good talent. And I'm, I don't really care about the talent. It could be US, Japan, well, yeah, it's everywhere. So, I try to recruit much more talent. And that's kind of important. And I'm asking our CEOs to, well, your job is to secure talent in the future. Let's just do it. And well, we're going to create a lot of jobs. And well, we announced several trillion dollar investment plan for the future. So, we need a lot of talent, right?
Chairman Chay, what is the biggest technological advantage that SK has built?
H, or on the memory side? Or?
On the memory side. Okay.
Process design.
I guess that well, you cannot just say just one thing, but what's the most important thing?
Well, one team. So, high, well, I really focused on that we could bring that just one company, one team. So, this is a simple business model, memory manufacturing. So, we can actually emphasize that we are just one team. And so, I think that one team made a really one big historical achievement now. So, that's what I'm saying, and that's the most important matter. Yeah, you can say that there were some, what, design capacities and well, process technology. This is right.
But that's, or working with TSMC on the base die, which the others do not do?
Also, well, we don't actually compete with any customers. That's another.
You do not compete with any customers?
And you feel that's important?
Yes, that's important. And that actually gives it a lot of room to actually partner in the future. So, that's how we can actually maintain all those partnerships.
Bigger picture. Mhm. Can you catch up with Samsung?
Oh, that's I don't want to answer those kind of questions. So, uh, I, I don't, well, my main goal is that well, I have to make our stakeholders happy.
Yes.
Well, in order to that, there's a certain part is competition, but I'm not really want to head up. Well, that's my main goal.
That's difficult though, Chairman, because the shareholders love the profit.
Mhm.
The customers want more capacity, more supply. Those seem to be competing ideals.
That's what we call that the stakeholders. So, it's not only the shareholders. So, shareholders want great returns, right? And the customer wants cheap product and enough supply. So, um, but well, I cannot just satisfy every time every stakeholder, but well, we have a certain balance. So, you know, between that stakeholder and what I'm really see that well, I also have to make my members, members of the company, well, they have to be happy. So, a happy company makes a happy wizard. So, that's simple.
Uh, I, I need to say, Chairman Chay, the ADRs. One stakeholder, the US investor. The ADRs indicated to open $177 per ADR. Of course, the ADRs were priced at $149. The memory market is also evolving. So, while supply is tight, there's a very broad look at other technologies. I'm very interested about SK's, uh, Heinix's approach to NAND and flash memory, and the role you see that technology playing going forward.
Right. So, it's a different type of memory storage. So, DRAM has its own functions, but that's instant things, like what people just remember certain numbers. Well, if you just prepared for your exam, then you had to recall good memory for those certain numbers and certain targets. But you also need a lot of other memories. And when you were young, then you had some memories. But well, that memory needs not just in the brain, you probably saw that somewhere. But like, well, AI, AI needs a lot of storage. So, DRAM has the instant response, and instant response type of memories, and you should store there. But if you have a more big size, and we call that cold data. So, if it's cold data, then you could actually store in the.
And where is the innovation? Where, sorry to interrupt you, Chairman Chay, where is the innovation on that other side?
Oh, there's a lot of technologies. And the problem is that well, in NAND, it actually, well, response is needed a lot of time, that means latency. So, in order to shooting that what that latency, well, we need a lot of special technology on that. And well, probably, well, what we call that KB caching, key-value caching. And well, that cache has to store somewhere. So, that's kind of a pyramid hierarchy of the whole memory storage systems. And we need all those evolution of those memory things.
And SK will invest into that technology?
Of course. That's our main business. And well, we're going to, we're going to make it. And we could actually deliver some other business model, even not just as a memory producer. So, we could actually, memory services, like memory as a service. Oh, you know, in the future, that's another area. So, we could actually focus on.
Talk a little bit more about memory as a service.
Oh.
The basic of how that would work.
Well, it's a kind of right now, it's a kind of imaginary thing and idea. Yeah. Yes. Just idea. So, but I think that our whole AI trend, the direction is goes there, because that memory is a kind of real bottleneck, and we have to solve it. And some of the specialists actually bring that certain type of solution to the each customers and each application. So, it's not just that one solution could actually cover all those problems. It's not going to happen. But as an influencing market, and that influencing market has its own characteristic. That characteristic, and we have to develop certain adjustments for that stack of the memories. And how to do it, and we need some special software together. But anyhow, well, we could actually serve as a kind of services rather than just here, memory, and you just use it. But we just fixed all kinds of different memory, memory systems. So, once we made it, and we could actually deliver as a kind of service rather than just produce it as a commodity.
It's an interesting look at the future. We are live on Bloomberg Television and Radio, speaking to SK Group Chairman Chay. In October, when you spoke to Bloomberg, Sheran, you said, "Everyone asks me about AI bubble, AI bubble, AI bubble." And you said at the time, "It's about the cost of AI."
The cost of AI is too high.
Yeah. Has that changed since October? And what is the trajectory of the cost of AI?
Yeah. Well, cost of AI, as well, you could actually say that token for dollar, dollar per token basis. Yeah. Anyhow, the we have to lower down that token cost.
But do you see it coming down?
Yes, actually, well, within three years, actually, token cost is about 1/5 or 1/10th. So, that's evolutions. And well, there are a lot of players just contributed to those kind of token cost is way down. So, even my memory company makes some solution for how we can actually lower down that token cost. That's our challenge. And well, our direction is goes there. So, that's kind of an ecosystem for AI. And well, right now, the cost is too high. But somehow, our technology, you know, down the road, probably five years, 10 years, and we actually digested all those problems. And it could actually be everybody's AI, and everybody could actually use their own agent in the future. So, that's that's the direction I see. But no AI bubble.
AI bubble? And well, when you see that the AI bubble, and this real world, and the stock market, stock market always, I see that there's some overshooting, and well, as too hot, too cold. So, there might be that world, there is a lot of bubble for the stock price itself. But that just reflects, that doesn't actually reflect the whole real world. But the trend-wise, if it's a long, then two years, three years down the road, then you could actually reflect the old real world. That's what I see. And you can say that AI bubble, but I see that in the stock market, but this AI technology is real. You use it. And well, actually, the problem is, well, this AI thing is that right now is a little immature, imperfect yet.
It's the early stage.
Mhm. Because it's the early stage, it's kind of, I said that it's only a four-year-old baby. But when you become a college boy and something else, then it has much more mature. So, it could actually help us a lot of productivity. And yeah, we just rely on those things. And this is kind of a race. And right now, we see that some AI trial.
Chairman Chay, it's historic today.
Right. Thank you.
Literally making history with the ADRs. But the focus on America, SK's relationship with America. What is your relationship like with the Trump administration, and how are you working with them to bring more of SKH Heinix to this country?
Well, not only just SK Heinix, and we've been already, SK already investing more than $35 billion in the US side. And not many people have noticed those things, but we do have bio business, we do have battery business. And he just announced a new fab in the Indiana side. And also SK Telecom made a lot of investment in AI startups. So, we already invested in the $35 billion, but I see that well, my plan is much bigger number. So.
Bigger than 35?
Yeah, much, much bigger than $35 billion. So, I'm planning to, and this is one milestone. We access the US capital market, and we have a lot of options. And also, we had the responsibility of keeping the share price. And well, one good thing is that we made the right investment, and so we kept our own value.
China is an important market.
China.
Driver of AI.
Yeah. China AI, as well, we have to actually notice that China's approach is way different than the US side.
How so?
Hm. They have a lot of disadvantage on the AI side. But anyhow, they put in a lot of other resources. The number one is the human force.
Human capital.
Yeah, human capital. And the number of people, number of engineers. So, annually, well, they had STEM students from majoring in STEM, and it's more than a million students coming out every year. So, they're looking for jobs. And mostly they apply to the tech area. When you say tech area, it's probably 50% is AI side. So, they are AI as a well, as a, yeah, they're using a different strategy. So, I don't see any kind of visible result yet. But there are some signs, and they're making cheaper ways. And they try to create cheaper tokens. Token cost compatibility is a, yes, that's kind of an important matter. And the US focused on well, top niche market, and well, quality of tokens. But China is focused on other ways. So, I don't know the result, but well, I see that what if they're using a different strategy.