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

Perfect. Well, welcome, everyone. I'm Alex Duval, tech analyst for Europe. I'm delighted to be here with CEO of ASML, Peter Wennink. Thanks very much, everyone, for joining. We're going to start off with some Q&A from me, and then we're going to open it up to the floor for further questions. So maybe I can kick it off with EUV. Sounds a logical play to start, and in particular, 2021.

Yes.

You talked about the new sort of range of tools, and you're thinking about that. And can you talk a bit about what sort of underpins that from a demand perspective? And what's giving you the confidence to talk about that range of tools for 2021?

Yes, I think, 2021 is, we'll see a continuation of what we see in 2020. We have to -- most important, if you think about EUV and the future, you have to really go back to 2018 where we turned a corner on EUV. And the relevance of it was that it created a level of confidence for most of our customers, I think, all of our customers in the logic space to say EUV is real. And it effectively meant that our customers and their customers started to think of a road map that would extend into the next decade using EUV, yes? It will be the driver of high-power compute, which would drive all the things where we're currently thinking of, AI, high-power compute, 5G, artificial intelligence. That, I think, was a really important moment in time. So that meant customers started to look at their road map. And if we really look at their road map, they shrink road map, the node to node road map. Over the last 2 years, it actually has, I would say, if anything, it has accelerated. So it means that 2020 and 2021 is just a reflection of that road map and so is logic. And it's going to be industry 7, the first 5-nanometer output in that time frame. And that will, of course, when you have high-power compute, you need to think that the derivative of high-power compute is the memory space. You need the high-power computing, the high performance memory. It also meant that the memory makers and the DRAM makers, in particular, were going to look at EUV as a way to extend the road map. And that's happening as we speak. So 2021, it's going to see memory and logic. 2020, primarily logic, but we'll see in the second half of the year. The first, you could see -- you could say that the pilot production -- or no, it's not pilot production. It's the learning of using EUV in 1z DRAM, but it's really the reflection of EUV becoming mature and actually driving the industry way into the next decade on the shrink road maps.

Makes total sense. You touched on memory there. What's the key delta that's really made people put those orders? We've seen news flow around Samsung and so forth. What sort of change in the last year on that front for you to get those significant orders in the door on that?

Yes, I think, when we -- a few years back when we started to think about EUV and DRAM, it was a bit of a question mark whether we would ever see EUV in memory and in DRAM, particularly. I think what the productivity level because DRAM is very much about cost. So the first way we looked at it was we need to provide industry with a machine that has enough productivity so we can reduce the cost per die. I think when we look at the productivity road map, I think there's a good reason why DRAM customers are now interested at using EUV, but it's all we got. I think what we've found out in the discussion with our customers, it's not only the productivity and the cost. It's really the performance of a device -- a DRAM device using EUV basically because it replaces multiple patterning layers. And we use multiple masks to do one layer. It created a better yield, better performing die, the distribution of high-performing die across the wafer is better. It'll effectively yield. What we're now seeing is that customer are saying, well, while in the current node, they're really agnostic. When it comes to using DPV or EUV, they don't care. It's these elements, yield, it's potentially cost that would drive them to choose for one or the other. But if you look at 1a, for instance, we now have some clear indications that we think customers are designing their DRAM just specifically for use with EUV, which actually means that we believe that the leading customers in the space field has a competitive advantage in using EUV because of the specific design features that it will create for the next-generation DRAM. And that's what we will see. We will see going forward in 2021. We'll definitely see a ramp, one-off are coming. I think we see 1z preliminary production for a limited number of wafers by the second half of this year. So it's also -- it will all come.

Very clear. So the demand is clearly there. When you think about this 45 to 50 units, so I think you've talked about the ability to supply and that being really the kind of main determining factor as we look out to 2021. Can you talk a bit about your level of confidence there? And what are the things that need to happen within ASML but also at suppliers? And how that's progressing?

Yes. The -- I think we said it also on the call. I mean, in terms of square footage manufacturing space, we're not expanding anything. So how do you get more machines out? It's reducing the cycle time. And the cycle time reduction is a learning curve. So basically, it's -- if you can think about this machine, a highly complex machine, when we did the first of this series, our focus was on getting out a machine that worked. So you could say in our procedures to build that tool, you're building up safety upon safety just to make sure that you can get a tool out there that actually works. Now when you start reproducing that tool in certain volume, you're starting to take the procedures out and you're starting to -- you use your R&D power to actually consolidate procedures and to basically learn to do things faster. That's what it is. I think we're looking at the cycle time reduction in our factory. So next -- by the end of the year, early next year or anywhere between 14 weeks -- 14 to 16 weeks, we're now at around 20, coming down from 26. We used to -- we're out of 18 at the end of last year, but we have a new modular vessel, which is the big vessel in which we create the EUV plasma, which is a new module. It's a big module that stepped up the cycle time with 2 weeks. It will go down. I think by the end of the year, early next year, we'll be at that 14-week period, so roughly a quarter. That will be sufficient for the 40 to 45 units. Now the question is, can we do the same thing in the supply chain? Now ZEISS has always mentioned, there's a big bottleneck. I think ZEISS will be there, yes? But there are other ones like up to 10 critical suppliers that need to ramp up their production capability. We have our own people there that actually help our suppliers to make sure that their cycle time reduction plan get executed. So where are we out today? I think we'll get there, 45 to 50 units. We've taken all the measures. I think our learning curves in die home and in our factory will get us certainly there, and we put a lot of effort in making sure that our supplies can fall.

That's very clear. So clear progression in terms of that learning curve. I guess, another progression we've seen on EUV is the average selling prices as you deliver more value. Obviously, we're on the 3400C now. That's been a very important tool. But as we move forward in the next year or 2, how should we think about that evolving? And to what degree can ASP continue to progress going forward?

Yes, as you've seen, the progression of ASP from this B version to the C version was about a 35% increase in productivity, about a 30% increase in price. So there's a direct relationship between the productivity, the output and the price, which is also true for DPV. It's not new for EUV. It's also -- we've done that for DPV. And we need to provide our customers with that value. And if we do, then we feel we're entitled to part of that game. That's the same thing. So next year, we'll come out with a D version that will not be 35% higher product. It will probably be something like 10% to 15%, but that's, again, in that area, you will see the price increase. And it's not only the productivity, it also has better lithographic performance. So overlay, which is very critical, if you shrink the minimum feature size of the chip, the overlay, that's the positioning accuracy of the subsequent layers of these structures on top of each other, of course, also becomes more critical. So overlay, in EUV and in DPV is a very important value driver for customers because it is a -- it has a direct correlation to yields. So these things all drive value, we need to prove it, yes, that value is there. I think we can charge it.

Very clear. We've talked about the progression of selling prices on EUV. Can you also talk a bit about gross margins because that's obviously an important value driver in the next year or 2? Obviously, prices will feed into that, but maybe some of the other key features that will be feeding into that.

Yes. I think when you look at gross margins, it will be -- let's start this year. I mean, this year, we guided 46% to 47% in Q1. Of course, in our Capital Markets Day 2018, we had a target of 50% this year. That was under the assumption that both memory and logic were in a moderate market scenario. In -- I think in memory, we're not in a moderate market scenario. We're in a down cycle scenario. So that means that for the first half of this year, the mix will be different. The second half of the year, we would expect the memory market to rebound. And the memory market is still very much DPV, which is our bread and butter product, with a better margin profile. It's more mature. So it also means that throughout this year, we think towards the second half of the year, you will see the margin reaching that 50% level because we do believe that the memory market will come back in the second half. That's one. Secondly, we also are introducing new tool types in the second half where on the DPV tool type, it's a 2050. Specifically, a little bit higher productivity, but more importantly, overlay is very important. It's a better price. So that's a better margin mix. And third, and absolutely not least, is EUV service. In all of our EUV service, we've built an EUV service infrastructure over the last couple of years, with virtually no income because they were in warranty. And to actually train an EUV engineer to be able to work on its own, it takes about 2 years. So we hired all those people 2 years ago. And basically, we're getting now the first wafer output. And as you know, we're getting pay per wafer. That means that when we start seeing the wafer output because of the first high-volume manufacturing happens as we speak, you will also see that for the first time, we will see EUV service revenue coming in, which by -- in the second half of the year, we will start to be profitable on the EUV service, which we never were until today or until that time. We always were loss-making because hardly any revenue and a big infrastructure. So it's the mix in DPV, driven by the memory return, introduction of new tool types and the EUV service revenue, we will start seeing -- bringing us to 50% margin in the second half of the year.

Talked about, obviously, the memory recovery in the second half. What are the data points that underpin your confidence on that? What are you seeing at the moment that we should be aware of?

Well, when we think about the memory recovery, this memory downturn is a bit different than we saw before in previous downturns. In previous downturns, when the market was in maybe the state of overcapacity, what memory customers did as they were producing basically a commodity-type product, they were ramping their output as much as they could because the more wafer is out, the more bits out to lower the cost per bit and lower the cost per chip. And it was a race to the bottom, the ones that hit the lowest cost per bit first, they were still making a little bit of a catch money and they're gaining market share. That has not happened this time. This time what we saw is that across the board, memory makers started to turn tools off. So basically creating some idle capacity, whereby managing the capacity output meant that you try to manage the supply-demand situation. It's a bit more proactive, which actually means that we can follow -- for the industry, we can follow the output out of our tools because some tools were switched off and we can see that. So we saw actually a reduction of that output in 2019, bottoming out in Q4 of last year. Then we saw the return of that utilization. Now if you follow that slope and you extrapolate that slope and absent any macroeconomic disasters, if you can follow the slope, you'd see, in Q2, we'll probably hit this equilibrium, a little bit faster for 3D NAND than for DRAM. But then around that time frame, you should also see the effect of pricing because once you still have underutilization, pricing doesn't necessarily move. And that also means that if you follow that trend, then this equilibrium will be reached probably middle of the year or somewhere in Q2, then you need to add capacity in the second half of the year. And that gives us, I guess, it's not a confidence, but it gives us the indication that we're on the right track. And we're preparing for it. Lots of people ask the question, look, how are you going to ship? You have a order lead time and are you going to only ship in 2021? No, because we see this. So we are taking some risk, some supply chain risk, you could say. So if order plunges and based on a certain assumption of bit growth in the -- throughout this year, we can guess, well, how much capacity we would need to add for the remainder of the year, those lenses we ordered, so we have them. That means we can hopefully deliver pretty fast when those orders come in.

Very clear. And if we think a bit sort of longer term, again, about EUV, obviously you've got those orders. Ecosystem is going to scale up. How do you think about the broader EUV ecosystem and the progress there? To what degree do you see any bottlenecks for any of the future nodes that we should be aware of? And then if we think...

It is assuming.

And even in the broader ecosystem.

In the broader ecosystem, okay.

And then secondly, if we look even further out, how important is High-NA to your customers' long-term road maps? And philosophically, should we think of that as more an evolution or a revolution in terms of what you need to do?