Good afternoon, everyone. I am Steve Barger with KeyBanc. I cover semicap equipment and cyclical industrials, and I'm here today with Kevin Engel, who is the EVP and COO of Amkor, and Kevin, to start, could you just give the audience a brief history of Amkor, what it does, maybe your evolution with the company, and then, just because it's topical, maybe a quick comment on tariffs and how you're positioned for it?
Yeah, good afternoon. First, let me just start with Amkor in general. Amkor is an OSAT outsourced assembly and test company. We're focused on more of the advanced packages in the industry. We service really all the markets, from communications, automotive, industrial, and consumer. Amkor is over 55 years old, so a relatively old company. We've been growing over time. A really exciting opportunity for us. Myself, I've been with Amkor for over 25 years. I came through an acquisition from a smaller company called Unitive Electronics and just been continuing with the company to grow over time. Prior to the COO role, I managed our business units.
Very good. Most common question I've gotten post-earnings is the socket transition. Can you just talk about how that evolved, what happened, when you expect to restart that?
OK. Yeah, so if we go back in time and you think about the life cycle of the mobile market, specifically the phone market, really, it's a year kind of cycle, so every year, you're going through a new device change. You're working with the customer to really rebid for that next-generation product. You have to maintain a hungry footprint to make sure you're winning the next-level socket, so that cycle is normal. This was a little bit different in that the customer wanted to migrate to a different architecture, and with that architecture change, basically, there were discussions around us supporting the future architecture or the legacy architecture, and in the end, we agreed to move forward with the future architecture, so that left us out for that one socket. It was not a surprise. We knew it was coming.
Basically, the next socket we'll start ramping this year.
So you've already gotten a commitment from the customer, and you're highly confident that that comes in the back half.
Yeah, so again, we've started development on that product last year. We are where we would expect to be from a qualification perspective. We're in the process of mobilizing our manufacturing infrastructure to support it. So yes, we have high confidence.
Great. And over the past year or two, there has been a lot of variability between smartphone forecasts and what we've actually seen. Are there any other design changes or issues that could affect the year?
There's always things that can affect it, but in general, what we see today, again, is because of that socket opportunity loss, the first half will be a little bit more muted than we would typically see, and then the second half, we expect that typical seasonality, but to be a little bit accelerated based on that first half trough, and then if you look at the overall mobile market in general, it feels about normal from what we can see.
So does that extend to the Android environment, or can you tell us what you're seeing there on the premium tier?
Yeah, so for Android, same type of dynamic. There's definitely a shift towards more premium-tier phones. And that's an area that we participate pretty heavily in. So if you look at our customers that are supporting those launches across that platform, there's a good bit of optimism going into next year or this year. And we saw that even towards the tail end of last year.
Do you have any early thoughts on the AI-enabled phones? Are they going to drive an upgrade cycle, or do we expect that they'll just get that functionality over time and we'll see normal replacement cycles, whatever that means?
Yeah, good question. So for me, I think everybody's waiting for a user experience that's really going to drive that refresh cycle. So if you look at the phones today, different phones have a different level of AI built into them. But in general, until the user experience really shows something that benefits them to go off and buy your next $1,000 phone, I think you need to see that. But to me, that will happen. It's just a matter of is that 2025 or 2026. But I think that trend will continue.
I want to ask about system-in-package. I think right now it primarily is for wearables and smartphones. Do you expect that technology will spread into auto, industrial, and other consumer applications?
Yeah, so if you think about what SiP is really good for, it's condensing a lot of functionality in a very small package. So it's really good for hearables, wearables, that type of area, and then some of the, in general, any kind of consumer market, and then also for phones where, again, very small footprint is needed. If you look in the other areas, so for automotive, we do have some SiP in that automotive space today already.
To me, then you start thinking about just chiplet kind of infrastructure in general, because SiP versus some of the other chiplet-type platforms, either multi-chip modules or even more advanced packages, they're all taking integration from multiple chips and putting it in one package. So what I see for those other areas is different types of chiplet architectures, whether it's flip-chip CSP or multi-chip modules.
Yeah, we'll talk more about automotive in a minute, but I guess just to extend that to Arm-based PCs, any comments there?
Yeah, Arm-based PCs, I mean, potentially some opportunities for SiP in some of the Wi-Fi-type modules. But I think the trends today are more towards other package types.
OK. Let's switch to computing. All through last year, we talked about tripling or quadrupling capacity for 2.5D. On the call, you talked about lower-than-expected demand due to China export restrictions and shifting product timelines. Can you start with the China restrictions? What's happened most recently?
Yeah, so obviously an ever-changing environment in that space, but for us, what we saw was we had several customers that were looking at ramping products either in 2.5D or other applications, and as some of those restrictions came in place towards the tail end of last year, that created an environment where they were pausing some of those product launches, so that was one dynamic to having a slightly muted Q1 in that 2.5D space, and then the other one was this transition from the GPU from the previous generation to the most current generation, and that transition happened a little bit faster than we or the customer anticipated.
While we still have a good pipeline moving forward for that legacy product or older product, and then overlay with that some of these newer products that were forecasted to ramp the beginning of this year, some of that's just been pushed a little bit further out.
So that's a good point on the shifting timelines. Have product rollout dynamics changed in your view?
Related specifically for the China market, maybe. But I would say in general, no. I mean, we're ramping our first SWIFT product, which is organic RDL rather than silicon interposer products. That's a first ramp of that device, which is really exciting for us. And we have another device coming online later in the year.
Is that what Giel talked about on the call in terms of bridge-based packaging? And what does that mean from a design standpoint?
That's the next generation. We have the 2.5D, and I'll correlate that to some of TSMC's terms. That would be like CoWoS-S. Then we have our SWIFT technology or organic RDL, which is like CoWoS-R. Then the bridges are more CoWoS-L. We would call that S-Connect. For us, that's a development activity. We would expect to be ramping that next year.
Understood. It's been a little over a month since the DeepSeek news. Everything we've talked about so far has been kind of before that. Has DeepSeek changed how you think about industry dynamics?
So to me, it comes down to where is AI going into the future, whether it's related to training or inferencing. In general, if you're able to drive down the cost, the computing cost over time for the customer or the consumer interface, as that continues to go down, to me, that just accelerates the adoption rate. And again, once you have some of these user interfaces that really drive a benefit to the consumer, having that at a lower cost, I think, is just going to accelerate that. So I think whether you're doing that compute on a very high-end, very expensive part or maybe a slightly less expensive part is still going to drive additional packaging and benefit Amkor.
Yeah, and it seems like DeepSeek was trained on non-restricted GPUs. Does that create more of an opportunity in China for you around AI?
We have to follow all of the restrictions out there around that. But again, we are an approved vendor related to the BIS restrictions. So I think that's an area that's benefiting us. Obviously, we have a facility in China that we continue to look for opportunities there to support the China local market. So we'll see how that develops and if additional restrictions are put in place.
When you think about AI in the near term versus the medium term, how do you expect revenue will evolve from traditional data center GPUs to more edge devices like PC and smartphones? And do you have a timeline for when you think inferencing could overtake DC?
Yeah, so that's an interesting one for me. We constantly talk about when an edge device is really called an AI device. If you take a phone today, it may have some AI functionality. Do you count all that revenue in AI, or do you not? So I think for everybody to kind of think through what does that look like is going to take some time. But I think the trend towards edge compute, definitely more towards an inferencing side, reduce that latency going back to the data center. And obviously, the data center guys, if they can push some of that compute into the edge device, that benefits them as well. So I think that will happen. As far as when is that transition point to where inference is higher, I think that's still yet to be seen.
Understood. Your primary OSAT competitor has also expanded CoWoS capacity. How do you differentiate against competitors in advanced packaging?
So to me, there's really several dynamics there. I mean, first, quality and execution are key. These are very large devices, very expensive. If you have yield issues, then that's a major issue. So I think that's definitely an area that we're focused on is the execution and the quality. Then you have to have the whole suite of technologies. We have full-flow technologies where we're taking the front-end wafers and going through all the packaging. And we have partial flows like the on-substrate portion or test opportunities. So that full suite, we really offer the customer what type of flow they need. And then finally is our geographical footprint in that for these types of packages, we're the only company outside of Taiwan and China that can support that. So if a customer is looking for some level of diversity, then they're definitely drawn to Amkor.
Given some of the near-term demand concerns, can you maintain pricing through a period like this?
Yeah, pricing is competitive, but we feel reasonable, so there's definitely an area there, and then if you look again at our focus more on the higher end on advanced technologies, the amount of players in that space is more limited, and so the pricing pressure is a little more contained. If you're really in that commodity space, then the pricing pressure is a lot more aggressive.
Yeah. Can we talk a little bit about panel-level packaging? We're hearing a little bit more about that. Maybe give everyone just an overview of what it is, why it matters, and do you think it'll primarily be for AI and HPC?
Okay, so panel, interestingly, I think of what most of the processes today, it's on a 300-mm wafer, so obviously round, and when you think about high-end devices, these larger GPUs and switches and things, you're going to very large structures that you can only fit maybe less than 10 of them on a wafer, so you lose a lot of scale when you have a round thing, so going to panel, now you're moving to square and then potentially also scaling that to much larger. For Amkor, since 2020, we've been working on a 650 by 650 panel, so you can imagine that enables you to have four 300-mm wafers, or you can do four 300 by 300 panels on that one panel, or you can do one full-size panel at the same time.
So it enables you to get many more devices through the process equipment at one time and manage that, so it's a cost benefit. The challenge is always yield. On these large panels with very fine line of spacing, you have to have that yield. So that's kind of one of the pivot points. So we have a platform there. We've been working on that for quite some time for lower density applications and then doing development on high-density applications. So we have multiple customer engagements there. I think it's an exciting opportunity. Definitely will help drive yield and cost in the future, but it has challenges related to yield. So there's still work to be done there.
Any idea of timeline towards more mainstream manufacturing? Is this years away?
I would say two to five years away.
Okay. If AI demand slows or macro conditions worsen, just thinking about Amkor's $850 million CapEx plan, would you view that as more discretionary in this year, just given how you're thinking about future growth, or would you consider adjusting that?
Yeah, so first, I think we need to kind of bucket that CapEx spend, so that's about 70% is related to equipment capacity. If we look at those buckets in there, that's our advanced package types where we're increasing our capabilities and growing going into this year. Then we have our SiP platform where those investments are more for specific device ramps. And then we have test opportunities. So those are the three main buckets.
S o could we pause or push those out? Yes. I think at this point, we see those as strategic investments, and we wouldn't do that, so another 20% is related to our facilities. About 5%-10% is for our U.S. manufacturing site. Then we have other expansions in Portugal to support the European market, as well as some expansions in Korea and other locations, so those could be moved around depending on overall situation. And then about 5% that's left is related to R&D and quality.
Understood. I have a few more questions on CapEx, but I want to make sure we have time for automotive. Very challenging market. Have you ever experienced a downturn this prolonged before?
This was long, and the way I view it is if you go back to COVID, the auto industry was very heavily affected by COVID. They were very aggressive to move from more of a just-in-time to a just-in-case type model. I think that set the stage to where there was a significant amount of inventory built up, then they were the last really to turn off the knob, and so now we're in that dynamic to where we need to continue to burn off this inventory level, and that's going to take some time, so that's a challenging dynamic.
So when you think about the prior cycle, it ultimately was just getting through the inventory in the channel before things flip back to positive? Or how do you see that evolving?
Yeah, so we think about it from two perspectives. We have advanced, which are more advanced packages that go into ADAS functionality, infotainment, connected cars, those types of things. That business is doing better. And then we have the more mainstream and legacy type packages, which are things like microcontrollers, things like that, where there's a mix of IDMs. There's a mix of other companies. And in that market, there's still a significant inventory buildup there.
When we see the more mainstream products start to reaccelerate, what does that mean for corporate margins?
So it will boost the margins. And I think when you look overall at our utilization levels, because the automotive industry has continued to have this trough, we have multiple factories that are significantly underutilized. So as we fill those factories back up, that will definitely be a tailwind in the margin.
As you look across the spectrum of vehicles from low tier to premium, how much more runway is there for semiconductor content increases? Are we hitting the upper limit in any platform?
So, any platform. So, I think if we think about the combustion engine, our typical semiconductor content is less than $1,000, maybe in the $700. Obviously, that can range depending on ADAS and infotainment functionality and things like that. But if you look at a fully autonomous electric vehicle, you're more in the $2,000 range. So, I think those trends towards more connected cars, more digitalization in the cockpit, EVs, even hybrids, autonomous driving, all those things will continue to drive semiconductor in the car.
Would that be more as we think more about Tesla's full self-driving or Waymo, is that more dollar content, just more expensive chips, or is it unit content across the vehicle?
It's a little bit of both. So definitely, you think about just that when we look at the number of components in a car, as you add in more of that functionality, you're definitely adding more chips as well as more advanced chips as well.
Yeah. Shifting to consumer, in terms of devices in the market, can we just talk about the market share that you touched today in your addressable market? And I'm just trying to think about how much room there is for share gains versus the platforms you're on, just kind of riding cycles.
Yeah, so we feel good about where our share is. I agree with you. If you look at kind of industry projections, it's low single-digit type growth projected for 2025. I think it's also important to remember that Amkor is playing more in the premium tier aspect of the wearable market. So we feel pretty good about that position. Similar to the mobile market we talked about before, you're constantly making sure you're hungry for the next generation device. So I think there's opportunities there, and we continue to work on that.
Is it similar to other parts of your business where you're in the prior design cycle or the forward design cycle and you see what that roadmap looks like?
You typically do, and if you're not seeing that roadmap, something's wrong, so that's definitely an area that we stay very focused on.
I just had a question on AR and VR. How important is that to the future of the consumer segment, and maybe, are you seeing a proliferation of applications?
So we're in multiple models there. And the unit volume is just very low. So I mean, it's not something for me to be too excited about. But the application space is interesting. I think it needs to continue to grow. I don't think they'll ever get used to seeing somebody on an airplane wearing one of those things. But in general, I think there's an opportunity there. We've heard about some applications to where you can do remote training in factories, things like that. So I think there's a space, but I don't know that the unit volume is going to be something that's going to be a huge driver for us.
Yeah. And you have a lot of capacity coming online in Vietnam. That business is typically more margin constrained because of the material content. What is the normal mix that you hope to maintain there going forward? Or can you talk about how you're going to balance the plants?
Yeah, so I view it slightly differently. I mean, we started with SiP and memory. And they're both ramping right now. So I think that's a win, and they're going to ramp significantly. But if I look forward in Vietnam, we have one module of four built. That one module is already very large. So eventually, this will be a huge facility. And we will bring in additional technologies. And we'll bring in incrementally more advanced technologies. So I would see that factory in the future more as kind of a standard type footprint of what we would see in some of our other locations.
Got it. Capital allocation, can you remind us of the preferred ranking of capital allocation and how you approach that?
Yeah, so first, we invest in the business. I mean, that's number one. Second, I would say investment in R&D to make sure we're supporting our future generations, then we look at optimization of our borrowing. After that, special or strategic investments. That's definitely an area that we spend some time on, and then finally, returning to our shareholders in the form of dividends, and obviously, we have stated some targets around what that looks like.
You did do a special in November, which I think was your first. What was the investor response? What was the reaction to that?
It's been mixed, so I think some investors would like us to look more either at increasing the standard dividend. Others would like us to look at buybacks, so I think those are all things that we're talking about internally to see if we change that strategy.
You talked a little question in the back.
Yeah.
Question. So you have a partnership with TSMC in Arizona. TSMC obviously made headlines the last 24 hours. Partner investment in the U.S. Do you see that partnership growing, or is that kind of a one-off, highly specific customer that it is what it is, and you need to expand to 2022 to enhance the partnership?
Yeah. I mean, so to me, the way I look at it is as part of the CHIPS Act, they move forward with three fabs and they're still in that build cycle. So I think the second fab is getting close and then third to be done. This new recent announcement was three additional fabs on top of that. Six fabs is a lot of output in the U.S. So in our discussions with them, the collaboration is still there. We see that more as an acceleration of the opportunity. So we're excited about that moving forward and excited about that for the U.S. as well.
I think for you, full rate production is planned for 2028.
Correct.
Could you accelerate that, if it makes sense?
That we are looking at, it is not easy related to permitting and build structures, but that's definitely something that we're exploring today.
On a relative basis, that'll be your most automated plant? Or can you talk some about the investments and how you're planning to meet the demand levels?
Yeah, it'll definitely be very automated. I mean, we want to balance, obviously, the U.S. cost structure. So that's an important component to managing that. Our Korea facility is also very automated, but we would expect the U.S. to be a little bit more than that.
We just have one minute left. Maybe if you want to wrap up, what do you view as the biggest opportunity, just given the end market set that you face right now?
I think that depends short term or long term. I think short term, the whole AI data center trends that we've talked about are very important. Long term, I think all these automotive, consumer, data center, all of those are going to drive a lot of growth for us, and we're really excited about it.
Very good. Thanks very much.
All right. Thank you.