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

Hi, everyone. I'm Alex Duval. I lead the European tech auto SMEs and gaming research team here at Goldman Sachs in London. Delighted to be joined today by Christophe Fouquet, Head of EUV, extreme ultraviolet lithography, at ASML. Before we get started, I'd like to state some disclosures. I'd like to state that this conversation is not intended for the media and is off the record. We are required to make certain disclosures and public appearances about Goldman Sachs' relationships with companies that we discuss. The disclosures relate to investment banking relationships, compensation received or 1% or more ownership. We are prepared to read aloud disclosures for any issue upon request. However, these disclosures are available in our most recent reports available to you as clients on our firm portals.

Great. So with that completed, thank you so much again, Christophe, for joining us. And I'd like to just kick off with a high-level question really to sort of set the scene for EUV. For those who maybe are more on the general side, could you give a quick description for 2 or 3 minutes summarizing to investors, who might be new to your business, what role -- what your role at ASML is and frame the importance of EUV to the evolution of the semiconductor and global electronic industries? I appreciate a big question, but glad if we could get some perspectives to kick off.

Okay. So I'll try to be short. So I'm Christophe Fouquet, indeed. I've been in ASML for the last 15 years. And today, I'm a member of the Board of Management of ASML and responsible for EUV. And the fact that EUV responsibility reside in the Board of Management of ASML was just a sign of how important this product was for us and for the industry. A few words about EUV. I think you know that ASML, for many years, has been focusing on delivering very advanced lithography scanners to the industry in order to allow our customer to deliver, basically, the most advanced chips. EUV was a very important step in this mission. Very important because it took us quite a bit after immersion to come with a new wavelength, which was basically capable to provide customer with smaller resolution. So EUV, what does EUV do? Just offer smaller capability to our customer, ability to shrink, ability to basically print smaller features. This is very important. EUV, it's in the net. This is about the wavelength of the tool. Going to EUV, we went from 193 nanometer wavelength to 13.5, which basically allowed us to make a very significant step in resolution. You have seen most probably that customers have been adopted, as a result, EUV very aggressively in the last 2, 3 years. It took us many years to develop the product, of course, because of the challenging technology, especially in relation to the source. But once we were successful with [indiscernible] product, we have seen major, major pool for -- by our customer for EUV.

Great. Well, thank you so much for setting the scene, Christophe. And I'd like to kick off by talking about sort of EUV and its applications as a digital enabler. You recently announced that ASML is examining the feasibility to expand the EUV production capacity to about 90 units, which is obviously up from the 70 units you talked about before and that's for 2025. And I think as part of that, ASML has referenced strong structural trends related to digitalization. So what do you think are the sort of secular drivers here that underpin your confidence in that sort of robust multiyear demand picture for EUV? And how does digitalization related to some of these applications like 5G, AI, high-performance compute translates into demand for your tools?

Well, I think some of the applications you just mentioned are requiring basically the most advanced chips I was talking about. And any company who is engaged in AI, in 5G is really driving the very best possible product because the competition there is also pretty high. So EUV is by default serving those advanced application. We have seen a major growth in the demand for those applications and we expect this to last. We've been a bit surprised, to be honest, in the last, I would say, 12 to 18 months on how strongly the demand was increasing. I think there's been a bit of a COVID-19 effect, some kind of acceleration of the appetite for all this technology. So those applications, the one that you see in the news every day, are really at the core of EUV demand. On top of this, there's a few more things. One thing that you may also be aware of is that most chip makers are trying to improve energy performance for their devices. This is to basically extend battery life, reduce power consumption. ESG is becoming a big topic for also those companies. And in order to reduce, basically, power consumption, you typically need to use less transistor of the chips you are going to make and this leads for the demand for larger chips. And if you look, basically, at some of the latest chips created by some of our customer, we see also there that they are becoming bigger. And bigger chips means for us more demand for silicon and, therefore, more demand for [ tool ]. So the combination of those 2 is driving what we believe is a structural increase on the demand for EUV. The number 90, by the way, is already validated by what we call bottoms-up demand by our customer. We even anticipate more demand moving forward. So we already see signs that this may not be the last step. This is happening despite the fact that we also increased productivity. You know that we have made several stake on EUV. So our tools are becoming faster. But this has not been sufficient, basically, to meet all the demand. So our customers are already looking at, indeed, 90 systems per year, maybe even more in the future. So that's something we believe to be structural.

That's super interesting that 90 may not be the limit there. And I guess you referenced sort of bottom-up modeling and -- we've, obviously, seen announcements of a large amount of spend on a multiyear basis from some of your sort of biggest customers in Taiwan, Korea, U.S., et cetera. I think you referenced some of the structural drivers there. But what do you see as the kind of drivers for their capital investment? How do you see that kind of underpinned that large amount of investment there? And secondly, you referenced sort of COVID and the pandemic and the spending there. And to what degree is there kind of spend that's happening that's actually been structurally accelerated versus to what extent should we be thinking that there's been a pull forward that could unwind in the next couple of years?

Yes. So I think what you're asking, Alex, is a bit what do we see as a structural growth, which I think is what I tried to explain. And there, I think we are convinced that, over time, we will see an increased demand. So we don't look at the peak. We look really at a growth that will continue over time, and this is a feeling that our customers share. And of course, around that, we also believe that you may have some cycle. One of it, indeed, could be related to the fact that today, we see very fierce competition between our customers. So everyone is investing very heavily, especially in foundries. But there, if you have more foundries customer, you by default create a need for more capacity. And this doesn't go away because you cannot be in a foundry business and not have capacity. So it could be that the utilization overall is not as high as it used to be when there is only 1 foundry customer, for example, but everyone needs to have some capacity in order to be able to compete. So mostly the competition we see could be heading about 10%, 15% on the demand on the short term. And this demand, you could in theory think, will go away, but it will only go away if one of those customers goes away. As long as the customers are there, they will all want to continue to build their own capacity and that, as a result, creates a bit of an efficiency. On the COVID-19, so I told you it did accelerate the demand for application. I think there, we don't really see that going away. The only question, of course, is related to the macro economy we are all looking at. And you can never, of course, know exactly what the overall consumer demand will be. It depends a lot on also some of the external factor we see, inflation, et cetera, et cetera. So we could have some bump in the road, and a lot of people try to already forecast or speculate when this may happen. This may happen, but still, on the long term, we truly believe that the demand will continue to be there.

Understood. That makes sense, sort of macro backdrop, but with those sort of secular overlays. And, I guess, coming back to this question about what your sort of key customers will do, can you talk a bit about what you're hearing from those key partners in terms of future technology road maps and willingness to use EUV at smaller nodes? Maybe you could elaborate a bit on some of the sort of comments from key partners and sort of their sort of longer term plans in order to ensure that they get the capacity. Obviously, it looks like there have been some developments on the memory side and the logic side. So it would be great to get an overview there.

Yes. So I think on logic, I talked a little bit about it. So a very competitive market right now between, I would say, 3 main players. And I think one of their biggest concern today is, indeed, to maybe not get all the capacity they need. So that's 1 element on EUV. So -- we don't see them slowing down. I think we see them continuing to apply a lot of pressure on us to deliver system. When it comes to their road map, I think if you look at the logic road map, there is a pretty good view of what can happen for the next 10, 15 years. So the transistor structure, I think, has a very clear road map. And this also includes scaling, which means that when we look at the future, we are also in very advanced discussion about how to improve the current 0.33 NA EUV machine. But as you are aware as well, we are also in very advanced discussion on how to introduce and ramp the next platform, which is the High-NA tool. So pretty good visibility on the road map, I would say, alignment, their commitment to the long term and major, again, concern focus on capacity. When it comes to memory, I think the EUV adoption in memory has been stronger than we initially thought. And the reason for that is that EUV is being used more than just for scaling. When we think lithography, we typically think about scaling, right? I did it myself in the introduction, I talked about resolution improvement. But what we have seen with DRAM customer is that EUV can be used to simplify dramatically the process and, therefore, create some more yield. I think our customer, Samsung, for example, have been very, very public about it. So increased demand more than we expected. And then also on the long-term road map, there's been many discussions about 3D DRAM versus 2D scaling. We believe today that the 3D DRAM will come most probably at some point, but 2D scaling most probably has at least 5 to 10 years of very, very strong, I would say, and aggressive road map from our customer. And as a result, we're also very much engaged with DRAM customer on both the extension of the current EUV platform, but also High-NA as you may have also read in some of our quarterly publication. So EUV, when we talk with our customers, EUV is really a very important component of their road map. And I will say, if I compare what we know today versus what we thought 4, 5 years ago, the reliance on EUV and future utilization is most probably quite higher than we expected.

That's great. Really appreciate that, Christophe, for sort of taking us through the latest on the logic side and the memory side and, in particular, highlighting the higher-than-expected demand on the DRAM side, looking back at the last period. Maybe if we can come on to your sort of pricing model and the sort of economic side of things, we've obviously seen, over the years, that ASML has been able progressively to increase the EUV ASPs, as you've improved productivity, but also a number of other factors, which you mentioned. Could you sort of perhaps explain a bit your value-based pricing model? And perhaps related to that, to what extent are you seeing an increased or reduced willingness from customers to pay higher ASPs as you were able to deliver better productivity and better features?

Yes. So I don't know if you've ever made the exercise. But if you look at our ASP over time and you compare it to our productivity road map, there is almost a one-to-one correlation, which means that for many, many years, the appetite for productivity is extremely high and the reason for that is, of course, productivity is a major efficiency gain for our customer, right? You get a lot more output with only a fraction of the tool you need. So I think we don't see the appetite going down. In fact, our customers are driving us to continue to increase our productivity. We have a very strong road map when it comes to the source power of EUV, which is at the core, of course, of productivity, the more photon we can get, the faster we can move. And for us, we also see, as a result, you mentioned it profitability increase because the same happened for us. We can basically ship more value while spending less money on parts, on manufacturing, on service, et cetera, et cetera. So this is also why our profitability on EUV -- this has been true in the past on other platforms. I've been following quite a bit our productivity improvement. So this is something we expect to continue to be honest. And I think as long as the appetite of our customer is strong on productivity, we will improve our system. And I would say we do expect to continue to see similar benefit out of it. Maybe one more thing on productivity because we talked about the demand, right? And I told you we have to do 90 machines despite the fact, for example, that our next generation of EUV 0.33, the NXE:3800E would be 20%, 25% faster than the previous one. And so productivity at this point of time is also helping us with capacity enormously. And I think this also helps our customers get again a lot more output in the same square feet area. So I would say productivity is -- value is almost at an all-time high because not only we bring, of course, more out of the same tool, but in the recent months, this really can help us with capacity.

Really appreciate that and giving us a sort of historical recap, but also pointing to the 3800E. And I guess the other one we've had questions about is the, if I remember rightly, the 4000F. So if we move to the sort of F model, can you talk a little bit as to your expectations there? Clearly, that should be an improvement, again, in terms of productivity, overlay and then presumably sort of trending into ASP as well? So what do you expect to see there? And if we think kind of longer term, you've obviously talked about High-NA and a big ASP there, but should we expect that Low-NA will continue to have these sort of increasing ASPs as well?

Yes. I think the 4000F, we call it F because we also expect to make indeed a productivity step. So it's -- we currently plan to ship the first one around 2025. So it's a bit far away. We do expect to improve the throughput because we expect to make a step on the power. We don't know exactly how much yet, most probably, again, somewhere between 10% and 20%. So we're still doing some work on this, but this is something we'll do for the 4000. And to your other question, yes, I think we will continue to do that on the 0.33 NA platform as long as we have ideas to get more wafer out per system. And right now, we do have, again, ideas for quite a few years. So you should expect a few more systems that will provide more productivity, better overlay, you mentioned that as well, on 0.33 moving forward. High-NA. So High-NA, we will follow a bit the same logic. So when you introduce a new platform, you're a bit careful. So the 5000 system is a system which we ship, basically, in order to get the learning on High-NA as quickly as possible. Then the EXE:5200 will come with much higher productivity, and this is the tool that we think most customers will use in high-volume manufacturing. And then moving forward, if we look at the High-NA platform, if we look at the 0.33 platform, the source is common and whatever gain in power and, therefore, productivity we can get on 1 platform, we will also get on the other one. So we will have those 2 products running in parallel. And on both of them, because we have developed them with a very high level of commonality, we will continue to improve both overlay and productivity. So this is a bit the motto of our road map most probably for at least the next 10 years, and we have a pretty good idea of what we could do there. We still have to validate all of that, but we are not lacking ideas.

That's great. And again, maybe looking backwards and looking forward, you made steady improvement on EUV gross margin and that now seems to be not too far off the group average. What should we be expecting in future in terms of your gross margin progression for the sort of upcoming EUV tools? You obviously referenced that the productivity could get better, so one would think ASP better, therefore, gross margin should be better. And then how should we be thinking about the sort of key improvements you can make as far as cost reduction goes? Because there's always been this story about sort of scaling up and getting cost down effectively per unit.

Yes. So that's a good point. So I think the cost down continues and it comes with the learning. I think we've made major costs down, of course, on EUV in the first couple of years of the ramp because we rationalized our manufacturing activity, rationalized service and this continue to improve. But at the same time, over time, of course, the improvement on costs, you see year-over-year become more difficult. I think that's something we have seen on the DPV platform. So this means also that the biggest contribution to potential margin improvement will remain the value we continue to bring to our customer. And there, it's very difficult to give you a number. I know you're looking for the exact number, but I will stick to the principle I gave you, which is if we continue to be able to improve productivity by 10%, 15% every generation, we do expect some margin improvement as a result of that, of course, as long as the cost of the technology makes sense for our customer. But I think there also we most really look at -- for EUV quite a few years where we should be able to continue to improve the margin of the tool.

Really appreciate that. And maybe going back to the sort of capacity plan, the sort of 90 tools, clearly, that's sort of something you're going to execute over a multiyear basis. Could you give a bit of color, Christophe, on what you see as the kind of greatest bottlenecks to that happening? Obviously, it's a very complicated type of technology. Obviously, there are complications of the world more broadly right now. So all the particular components where there could be more sort of difficulties in terms of the ramp-up, what are the kind of moving parts that need to fall into place for your -- for you to be able to execute on that kind of capacity plan?

The biggest challenge to ramp with technology like EUV is lead time, to be honest. And to go from 40, 50, 60 to 90, there's nothing really difficult in it. It's all about building a clean room, getting the same process tool we use today to get more capacity. So there's no real operational challenge, if you want. The biggest challenge we have in this ramp is really the lead time because some of the equipment needed for EUV can take more than a year, sometimes 2 years, basically, to be ready. And I think it's a bit also one of the learning we had in the last few months. We don't need to plan, of course, always for capacity that would be a lot higher than the customer demand. That's why we talk about 90. But on top of planning for 90, we also take steps to make sure that the lead time for the next number, if there were to be a next number, will be a lot shorter. So we talk a lot about the 90 because this is also the number I wanted looking for, looking at the market. But I think what we have done, and it's also very important, in the last few months, is to make sure that if we had to go 1 day beyond 90, the lead time to do that will be a lot shorter so that we can react to our customer demand a lot faster. So no real, if you want, technical challenge. No real operational challenge. I think today lead time, because the technology is indeed very unique, and that's something that, of course, we have means to reduce by investing very small amount of money on the most critical either space or equipment we need to further ramp if needed.

Really appreciate that. And also the color about the lead time if you were to go beyond 90 machines. If we go back to High-NA, I think you sort of touched on this and I think you sort of commented ASML that you received multiple orders for the [ 5200 ] High-NA machine this quarter. Clearly, these machines are significantly more expensive than Low-NA machines. But I think the discussion was about potentially having a need for 20 of these machines or capacity to produce 20 in a given year. So can you remind us a bit about the sort of benefits there that customers are getting? Your latest view in terms of what that will drive in terms of advantages as to the granularity of the lithography, the productivity and perhaps also the chip performance? And of course, related to that, it seems that High-NA really obviates the need for multiple pulsed patterning on sort of Low-NA, but it would be helpful to get a recap there and any other benefits I'm not mentioning.

No, I think the biggest benefit is what you described. So again, High-NA is also in the [ NIM ]. So resolution is about wavelengths. That is also about the numerical aperture of the tool. By increasing the numerical aperture of the tool, we improve the resolution, which basically either allow customers to scale more, so getting to even smaller pitch, or allow them, you mentioned it already, to avoid too many multi-patterning steps. So when it comes to the value of High-NA, I always say, if you understood the value going from DPV to [indiscernible], thanks again to the better resolution, less defect, all those parameters that our customers have published, the value of High-NA is exactly the same. We're going to make another step, which will simplify the process again and bring yield benefit, extend the road map of our customer, et cetera, et cetera. So that's the exact same type of benefit as when we introduced EUV. The biggest difference with the previous introduction was that when we put 0.33 NA on the market, we had major challenges with the source -- the EUV source. And what we like very much about High-NA and what our customers like about High-NA is that in this case, we're going to use the exact same source, meaning that whatever we have learned will be applicable to High-NA. So we expect a far smoother introduction, if you want, with High-NA, a bit like when we moved from the XT platform to NXT platform on immersion. So that's -- and I mentioned that because the maturity of the product we introduced is also a big part of the value because the more mature, the faster it is for our customer to introduce it. The number of High-NA, too, now you mentioned 20, so that's always difficult to estimate. I mean you have seen with 0.33 NA that we were wrong because we -- we start. We have a certain assumption of how many layers are going to be used with the new product, and that's always a fraction of the total layers and, typically, the more advanced the product, the less layers. So we have a lot more KrF layer than immersion layer and a lot more immersion layer than EUV. That's always true. So there's always a bit of less layer. But the reason for that is that customers are optimizing basically their cost and only use the most advanced tool and the most complicated layer. So we do prepare for 22 layer, you mentioned that. But again, learning from what's happened in the last few months, you understand that if the overall market was to grow, the number 20, which we have given to the market also a few years ago will scale in the same extent. So we're also looking at -- we're also trying to prepare ourselves for more than that, again, with a much shorter lead time. So we don't invest anything today, but we make sure that if we are successful with the High-NA introduction and the volume was to increase faster than we expected, we could react also fairly quickly.

That's great. And a question we get from investors really is about the extent to which High-NA tools could end up cannibalizing effectively demand for Low-NA, so regular EUV machines. I wondered if you could provide a perspective on that. Does it even make sense to think about cannibalization? Does it matter for ASML? I don't want to lead the witness here, but curious on your view.

No, it's always depending on how you look at it. If I'm [indiscernible] certain process stack and replace some multi-patterning 0.33 EUV with High-NA EUV, it means that those layers are not going to be 0.33 anymore. So this is very special stack. You have more High-NA tool and a bit less 0.33 tools, right? This is just what's happened when you replace 1 layer with a more advanced tool, you take away from the less advanced tool. But now when you have this discussion and look at it this way, what you always miss is the fact that, at the same time, volume goes up. And we have seen that for DPV, for example, where we never fully took into account the fact that despite this transition from DPV to EUV, the volume will grow so fast that the total number of DPV tool today we ship is higher in fact than it was a few years ago. Now will it be exactly the same for 0.33 EUV? It's always difficult to predict. But what I'm trying to say is that we expect that some of the, what you call cannibalization, which is a transition of a layer from one tool the other, when this happens, some of this will be compensated by additional volume on demand, basically, from our customers. So we don't really see at least or don't really plan on our side to necessarily reduce dramatically our capacity on 0.33 EUV when High-NA comes to the market.

Got it. And going back to first quarter results, you talked about sort of ASML having received High-NA orders from 2 memory customers. I think we've seen some broadening over time. So I guess just from a sort of technological standpoint on the memory side, what's driving increased adoption of these kind of quite expensive tools by memory customers? How should we be thinking about that?

Well, I think -- I could refer to maybe the Micron investor presentation because that's only a few weeks old, where I think Micron nicely showed a bit the order of what they had in mind for the road map. And I think one thing that was clear there is that they were still counting on extending 2D for quite a bit, and some of it with the help of High-NA. So I think that's a bit what we see is our DRAM customer. I think they are counting on 2D device for, I would say, at least the next 10 years. There's a lot of discussion about 3D DRAM transition. I think there will be a lot of work down there, but the complexity is pretty high. And also the cost benefit is tricky because in order to move to 3D, you need to have already enough stack to be able to compete cost-wise with wherever the 2D roadmap would be. So I think the commitment of our DRAM customer to High-NA, which I think has been strong in the last few months, is most probably the sign of their commitment to 2D scaling for still quite a few years.

That's great. And I guess another question that we get from investors is just on sort of layer counts. Obviously, when you last did a forecast for 2025, you talked about a certain layer count, generally speaking, for example, on EUV on the logic side, what you've got feeling in terms of sort of how that's tracking and sort of the comfort level of -- and desire to maybe have more layers and so on from customers as we look out a few years? And then, I guess, as we sort of evolve from Low-NA to High-NA after 2025, what will that do, I guess, in terms of layer counts? How should we be thinking about that?

Well, I think if you look at the High-NA introduction, typically, we -- people used to think about our tools or our road map as a succession of cliff, right, where we have 1 tool and then we get a new tool and the old tools disappear and the new tools take over. I think it's very far from reality. I think customers are building up very flexible road map, which will allow them, basically, to insert new technology very rationally. So if you look at layer count, when High-NA is being inserted in production, for example, for logic around 2025-'26, I would expect customers to start with 1 or 2 layers. Two reasons for that. First one is because they use that for their own learning, they reduce the risk. Second, because this also corresponds a bit to the capacity we can offer on High-NA in the first few years. We also cannot follow up this kind of very abrupt ramp. And then for logic, there's still a lot of debate on the road map, but the node after that could be anywhere between 5 to 7 layers depending on the customer. Then we see that climbing further or a bit like we have seen with 0.33. But when we ask customers about what's happening in 2000 -- mostly now in 2028-'29, they don't really know. So you can rely a bit on the historical trend, I think, because history has been repeated itself. So we have no doubt that the number of layers will grow. Our experience with the last EUV tool was that we always calculate for scaling. Like I said before, we typically miss all the other reasons like process integration simplification because we are not expert for that in ASML. So I will say, usually, there is always a bit of upside at customer based on the opportunity they see with new product in their process compared to, for example, what we told you at the Investor Day. So it's very difficult to -- the prediction we made for '29 or even beyond that because that's the type of -- question you're asking me are more based on our experience than a real understanding of what customer would do there. I think the -- honestly, the EUV developers, they don't know themselves.

That's very clear, Christophe. And as we go into last 5 or 6 minutes, I'd like to just put to you a couple of questions we've also had from investors. Firstly, on gate-all-around, there's been some recent news flow that a Korean player has achieved perhaps lower-than-expected process yield performance on 3-nanometer gate-all-around. Is that significant? Could that have an impact on EUV adoption by ASML's key customers? And then if we think about geopolitics, ASML has talked about a desire to build out capacity to levels in 2025, which are well beyond current abilities in terms of EUV units and there's been talk of secular drivers, but also geopolitically driven demand. So what has sort of tangibly made ASML more confident on the geopolitical side of things being an important driver? And how important -- how significant is that driver relative to some of the other factors you've mentioned?

Yes. So maybe I'll start with gate-all-around. So there, the -- I think the understanding that this will come in the road map is pretty clear. So we have factored in this transition in all our forecasts already for quite a few years. So you will get a lot of press around that because first results would come in, but we see our customer very solidly moving forward this step on the transistor. And we don't expect any real, I would say, noise for us around that. The geopolitics, so I could go back to some of the comments I made in the first part of the discussion where I told you we see, for logic, and I think that's where most of the geopolitics come in, quite fierce competition between, I would say, Samsung, Intel, TSMC. I think geopolitics is part of that. I think that -- if you look at the latest Intel presentation, I think, they are clearly making geopolitics an argument in presenting themselves as the only company that can be in any occasion. That most really could happen with the other players. So this, I would say, almost amplify this competition. But when it comes to the effect, it's still part of the 10%, 15% extra demand I was talking about before. It's -- you see everyone putting fab everywhere so that they are ready to take the business if the business comes. And like I said before, as a result, they will see some inefficiency. So this could have an impact for our customer costs because they, at the end, buy tool that they may or may not use depending on who wins the market. For us, yes, the capacity we put on the ground and, of course, whatever the utilization is, those tools will be gone. So I think you could mix a bit the geopolitics element on EUV with, what I call, the fierce competition before. I think the 2 are very much combined right now because the geopolitics also drive some of the competition we see.

Great. And Christophe, I see we're down to the last 2 minutes. Another question I received from investors is just around China. You talked about geopolitics. To what extent will China allow to get EUV at some point? Clearly not at the moment. Could there be a scenario where they could eventually be allowed to get EUV, for example, several nodes down the line? To what extent does that matter for ASML?

Yes. I think for the foreseeable future, I think that -- I don't see EUV going to China soon because I think that the current position taken by the U.S. administration is pretty strong. And I think there is a real willingness to block China to have access to this technology. So on the foreseeable feature, we don't expect EUV to go to China. Now of course, you could ask yourself what's the time at some point in the U.S. may or may not allowed in terms of technology delay there is very speculative. So I won't -- of course, I won't say EUV will never be in China, but I think this will take quite a few years, at least based on the discussion we're having today.

Great. Well, Christophe, I think we're out of time. We'd like to thank you very much, indeed. Always greatly appreciate the dialogue. It's been a very wide-ranging conversation spanning drivers of EUV, High-NA and also geopolitics. Really appreciate your time as ever and look forward to speaking soon.

You're very welcome. Thank you very much for the invitation.

Thanks, again.