ASML Holding N.V. (ASML) Earnings Call Transcript & Summary

Very helpful. I'd like to also talk about average selling prices. Clearly, there was some discussion about potentially getting higher ASPs to help compensate for inflationary impacts. Where are we on that? To what extent is there more upside to sort of come through from that? And equally, given the sort of weakening macro environment we've seen, is there any sort of downside risk on the selling prices?

Yes. So Roger Dassen talked about that just quite a bit over the last couple of quarters. We saw some impact last year as we -- with headwinds on our margins as inflation crept into our costs. Some of those costs, we don't add value to. So for instance, the shipping and transportation costs. And that's an area where we have been going back to our customers and saying, "Hey, it's an area we don't add value. It's pretty much passed along. Can you help with that?" And we're seeing some success in that in improved ASPs. So we've seen some of that, and we'll probably see a bit more going forward. And Roger comment about that. In terms of gross margins, we think we've neutralized that for this year in terms of impact on gross margins as long as inflation doesn't continue to go up.

Great. And I think another topic we've had questions on a bit is on sort of lead times. So if you think across EUV and DUV, the argument, I guess, a few quarters ago, was that these lead times are so long that it insulates you to quite a significant degree from macro risks because the customer knows they have to wait a long time to get the tool. And therefore, they don't want to go to the back of the queue. I think the numbers quoted were 1 year on DUV, 2 years on EUV. To what extent does that dynamic still hold? To what extent has it perhaps changed? And to what degree does it still help insulate a bit from risks?

Sure. So litho is still quite a long-lead time item in the fab compared to the other equipment. Pre-pandemic, if you came to us and hadn't given us any forecast for tools or hadn't included a forecast for incremental tools that you wanted, it was about 6 to 9 months for a deep UV tool, 6 months on the, say, a dry tool and more like 9 months for an immersion tool and 12 to 15 months on a low-NA EUV tool. Through the pandemic, that grew as supply chain grew and our capacity was well below what demand was out there. And so you had low-NA that was more like 18 months to 2 years across the board. Deep UV was settling at more like a year if you hadn't given us an advanced forecast. And now those, as we've worked through supply chain issues and increased our capacity, those have come down not to pandemic levels, but as Peter -- or pre-pandemic levels. But as Peter Wennink put it, the supply chain probably throughout the summer will get back to in the shadow of ZEISS, which is our longest -- traditionally, our longest-lead time supplier on optics, so getting much better. So there's still a couple of quarters where customers put orders in to fill out next year. But it's still a long-lead time item. So from the insulating factor standpoint, you're still a year plus on low-NA EUV tools and somewhere between 9 months and a year on deep UV. So if you think this is -- if you're a customer and you think the supply-demand gets back in balance in a quarter or 2, you're not going to get out of line for, say, an EUV tool. You're going to want to put orders in. Because what you're putting orders in, if it's second half of next year delivery, it's about outputting wafers in the second half of 2025. So that's why we say things like technology tools are insulated from this a bit. And on the deep-UV tools, if they think it's a back-end balance within a quarter, they'll put orders. And if they think it's a couple of quarters out, they may hold off a bit to see where they're at. But still relatively insulated.

Thanks for the sense, Pete. I'd like to now talk a bit longer term. We've talked a lot about low-NA. Let's talk about high-NA. So could you give us the latest snapshot on the situation on high-NA orders? If we look at your 2025 plan, your 2030 plan, you obviously updated those relatively recently. How much of the high-NA unit targets within that is actually covered by current backlog? And how should we be thinking about the high-NA orders going forward?

Sure. Well, we certainly saw -- well, we started booking high-NA tools, the EXE5000, which will be the first version of it, development tools for customers several years ago. We initially booked the -- I think we talked about 4 tools to 3 customers. We booked a couple more since then. But last year, we started booking orders for what would be the production of high-volume manufacturing tool of the EXE:5200. And we booked several of those. I think we even commented that at least every customer has at least one in the backlog of the 5200s in both logic and memory. So we have a good backlog of those tools. We'll start shipping the EXE5000 late this year, first tools. That 5200 doesn't really start to go out until late next year into 2025. If you look at our modeling that we showed at Investor Day, I think we show a high scenario revenue of 5 systems in 2025. We'll start shipping those tools again late this year and the 5200s in 2025. So you'll see the revenue from those in there, but we'll plan to ship more than that. It will take a while to revenue these tools as it's a new platform, and we have to ship them and install them in our customers' fabs and prove they meet spec before we get revenue on those as with any new platform. So I think we're working towards those numbers. We talked about growing the capacity to 20 units in the 2027 time frame at our Capital Markets Day, and we're still working towards that plan.

Super helpful. And just maybe as a follow-up, could you help put in context a bit their applicability to logic versus memory? How should we be expecting the order trends there?

Yes. So I think in our backlog, EUV in total is high-70%, low-80% logic in the remainder of memory. And I think it's -- that's representative of EUV in general.

Very clear. I think we've got 15 minutes left. So I'd like to leave some time at the end for audience questions, but maybe just one more from me. How should we think about scope for high-NA to cannibalize low-NA over time? How will the dynamic play out? Because obviously, you're getting high productivity. You're getting all these extra bells and whistles and efficiency from high-NA. So these are very expensive tools. So how should we think about how they coexist with low-NA as they ramp up towards 20 tools, et cetera, per year?

Sure, sure. So just like with -- as EUV came in for deep UV, it's there to mitigate the need to go to multi-pass patterning. Frankly, on EUV, we were late to it when we initially thought it would go in. So multi-pass patterning grew, which allowed our customers to continue to shrink, but not in a cost-effective manner. If you went back to one of our Investor Days, when we started seeing EUV actually going to volume manufacturing, we thought we'd see some cannibalization of the immersion business as it ramped. But frankly, we didn't. We're shipping just as much now, if not more immersion tools, because the nodes that it applies to continue to grow longer and stronger. But certainly, it did cannibalize some of that on a specific application when you go from one node to the next. On low-NA to high-NA, it's there to negate the need to go to multi-pass patterning on low-NA EUV. So it will be used on the additional critical layers that would need multi-pass patterning if you didn't have high-NA. So I don't think we see a lot of cannibalization there. You'd see more low-NA if high-NA weren't there. But as it comes in, you'll see low-NA kind of stay on the layers it's on, and the more critical layers will move to high-NA. So yes, some, but not as much as you saw as the transition from immersion to low-NA because we were a bit delayed on that and multi-pass patterning grew quite a bit.

Great. Maybe one final question for me before we go to the audience questions. If we look across the semi cap landscape, we've got some of these new trends, chiplets and new materials and also gate-all-around, And sometimes, we get asked by investors, what does that mean for EUV? What does it mean for ASML? Can you help put that in context a little bit?

Sure. So we see gate-all-around as a good thing. The reason they go to gate-all-around is because if you don't, if you try to shrink FinFET, the performance of the tools or the power consumption, it gets worse. So you need to go to a different architecture just as they did going from planar to FinFET. FinFET allowed shrink to continue and performance to continue to improve. We see the same thing happening when you go from FinFET to gate-all-around. You're going to gate-all-around because you want to allow performance to improve, and frankly, it allows shrink to continue. So we think the transition will be very similar to the transition from planar to FinFET, which took place at the 20-nanometer node and [ signs all point ] to the 16. At 20, they were a planar. At 16, they brought FinFET in. There really wasn't any shrink because it's highly risky to do shrink and an architecture change at the same time. So they introduced 16 -- FinFET on 16, put an amount of capacity in that, let them pipe clean out that process. And once that was working, then you go to a sub node of 14 and 12 where you have more shrink and frankly, required more litho. But we consider that all 1 node, meaning 20, 16, 14, 12. We think you'll see the same thing at gate-all-around, whether -- regardless of what node you're doing at. The first incarnation, you'll bring gate-all-around, and you probably won't really shrink. But after you pipe clean that initial introduction of it, then that enables shrink, and we'll see more litho or more EUV as you do sub nodes of that. So in general, we'll see more EUV for the whole node. And we've talked about, in general, you see about 30% more litho spend per logic node as you go forward, and DRAM is about 20%. NAND, it's about 10% to 15% spend per node. So gate-all-around, we see as a positive. Certainly, materials benefit from that. And our CTO, in the past, has talked about there's 4 ways to continue to improve cost on the semiconductor road map. There's shrink, which we're the shepherds of, if you will, and certainly going to gate-all-around enables more shrink to happen. So we see that as a positive. There's material. So you need those materials for gate-all-around, so that's a positive. Packaging is a positive. Things like chiplets make the ability to do this more affordable. So anything that improves cost or affordability for our customers has the ability to spur demand. So we see that as a positive as well in [ last year's ] architecture, which we just talked about. So we see any of these that are -- improve the cost per function equation for our customers as beneficial for all of us in the equipment business.

Super helpful, Pete. I'd like to go to audience questions. If anyone has a question, be most grateful if you could raise your hand.

At the beginning of your remarks, you said that in your long-term model, you maybe did not correctly think about the die size. And die sizes were a bit bigger than what you initially thought in advanced logic. But what I'm interested in is also trying to understand how you came up with your long-term $1 trillion model for 2030? And in particular, how you think about volume versus price of semiconductors? I think over the past 2 years, we've seen a very strong ASP for some semiconductor areas. And I was wondering if in your long-term model, you assume normalization of price growth, price declines versus volume? If you had any thoughts around that.

Yes, it's a good question. So our -- again, our long-term model is built off again, to your point, a third-party forecast. And I think even in that Investor Day slide deck, if you look at it, there's a couple of different ones that are anywhere from $1 trillion to $1.2 trillion-or-so semi end market in that time frame. And we use that and apply things like die size and bit growth and our knowledge of -- which are -- we have pretty good knowledge on layer stack and road maps and can calculate a number of tools from that. If our CEO Peter Wennink was here, he'd say the industry in total, us and forecasters, always get that long-term model wrong to the conservative side because we're always thinking of the current drivers, not the future drivers, right? So in 2014, we did an Investor Day, and we made some assumption about node sizes. But did anybody know in 2014 what a Ring doorbell or a Nest thermostat was? Probably not. They were thinking about things like PCs and mobile phones and saw that flattening and were concerned that maybe we were overestimating a mid-market scenario. But frankly, they weren't thinking about some of these things nor were we. But so he'd say, we probably, if anything, are always wrong to the conservative side. In terms of ASPs, I'm sure our marketing, people that do the modeling, made rational decisions about ASPs. They're not using peak ASPs, but kind of long-term average ASPs in there, so -- but I'd say our CTO and our CEO and CFO would say, if anything, the industry always gets it wrong to the conservative side.

Great. I think we've got 2 more minutes of this off-record discussion. So I don't know if anyone else has a question.

Pete, EUV installed base revenue is based on some operating variables at the customer sites. How have those trended versus expectation? And do you expect the same algorithm, which, again, we don't know exactly, but do you expect the same terms on future shipments or an adjustment in installed base revs?

Yes. So we talked about our Installed Base Management model as we went to a different type of model on EUV, a more output-based model. In our deep UV, we talked about roughly 2% of the ASP of the tool in revenue on what has been a break-fix model on deep UV or parts and labor type of model. But under that model, we can meet the contractual obligations that we signed up to. But there could still be long downs in that. If a tool goes down and a field service engineer goes in and say he needs a part, maybe he has it on hand and he can get it back up and running quick. But maybe you need something out of Wilton or Veldhoven or San Diego, and it's a day or 2 or maybe a week. That's a long down, and that's an unpredictable output for our customers. So on EUV, we went to more of an output-based model where you could -- it's -- we agree on, say, a price per -- or service cost per wafer and agree on an output they want to get, and we're responsible for maintaining the tool and providing good uptime and availability of the tool. And if we exceed the agreed number of wafers, we get paid for that. If we short them, they pay less. The win-win is they get more predictable output, and we get paid for the value we're offering in service. And we think we can get about 5% of the ASP of a tool per year under that type of model. And that's after the tool comes out of warranty, so roughly 18 months. Some tools are 1 year. Some are 2 years but if you say 18 months, that probably captures it. We've seen EUV service business grow over the last couple of years, in line with what we've expected in terms of revenue. I think Roger talked about the margins being coming out of 2021, roughly, say, 20 -- high 20s percent gross margin on that business, and it grew to probably 30% last year, and that grows towards corporate margins in the 2025 time frame. So I think we're seeing the revenue we expected, and we see margins continue to grow as more tools are installed per location, and we continue to work on the cost of parts for that as well.

Great. Well, Pete, thank you very much for a very illuminating discussion. I think we're out of time.

Thanks for having me.

Thanks, everyone, for joining, and speak very soon.

Great.