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

Thank you very much for being here. I'm Francois Bouvignies from the tech hardware team in Europe. And today, we're happy to have a Skip Miller Head of Investor Relations at ASML. You can scan -- in front of you, you have a paper where you can scan and ask questions. I will receive them in the second iPad have today, and I will be happy to pass that to Skip. We're going to talk about 2023 outlook first going through the different business units, and then we'll try to look a bit beyond 2023. If that's okay with you, Skip, and talk about the long term. So thank you for being here.

So maybe let's talk about 2023, as we know, '22 is already behind. Can you talk about the outlook you see for your different business units as you look at 2023 whether it's memory versus logic and maybe DUV versus EUV would be great to start with?

All right. First off, thank you, Francois. Thank you, everyone. Yes -- so 2023, we look at that, obviously, there's -- as we sit here today, there's a lot of macro uncertainties you say with respect to the inflation situation, what's going to happen around recessionary fears, geopolitics. So there's a lot of number of items there that creates some uncertainty. But if we look at our business and our backlog, especially we talked about October being around EUR 38 billion, so a strong backlog covering us for next year, we're looking -- talking to our customers and looking where we're sitting today with demand still significantly exceeding what we can supply. We're looking at next year, increasing output, both EUV and DUV. And we've talked about shipping 375 or more DPV systems by next year. Within that, those DPV shipments, we talked roughly 25% of those systems would be immersion, the remaining being dry. And then on EUV, we talked about shipping 60 or more EUV systems next year. So that's where we are with respect to our EUV and DUV business. And we didn't break it out by memory and logic, maybe something we'll talk about more in January. And you have to keep in mind that other dynamic at play here is this whole revenue recognition with fast shipment, which maybe we can talk about later on, but that's another piece of the revenue, which we currently have, EUR 2.2 billion coming out of 2022 into '23, and depending on how fast shipments evolve, we may have some coming out of Q4 next year unless we're able to resolve some of this accounting issues with respect to revenue recognition.

Great. It's good start. So you said summarize, you expect to grow next year with DUV, EUV. Now when we look at the market as a wafer fap equipment, you had this question quite often have it as well. it's more looking like down 20% for many expectation out there. So it looks like ASML is gaining a lot of market share out of this spending in terms of wafer fab equipment. So can you explain maybe the drivers of how you are disconnected maybe from the general market? And what it means for beyond 2023, looking at the fact that you can grow in a down market?

Yes. So first off, I think the comments I'm making with respect to the 2023, taking -- looking at it from a backlog perspective of demand supply, I think that gives us some confidence that we really need to focus on the supply. Now of course, as I said, the things can happen depending on your view with respect to recession on will it be a deep ugly recession, then obviously, it will be impacted if it's more moderate, then there's reasons why you can argue that litho may be slightly different in that respect. And I think it really has to do with the fact that the lead times are longest first off from the point where you place in order to actually receive a machine or the longest of the -- some of the other process equipment. And the second is the actual install qualification time for the time you actually ship it to actually it's running wafers are quite long. So you put those 2 together, our customers have to look at it from a perspective if I'm expecting a -- if they do first off view, there's a recession and they view there's a recovery by the back of the year, early 2024, then they need to have machines in place in 2023 such as ready to capture that increase in demand when it recovers, if and when, again, depending back to your recessionary comments. So I think that's the fundamental difference, say, from some of the other process equipment. And also keep in mind that it's been, over the past years, or at least a couple of years, it's been a bit of a challenge to get the machines that they want. So they've been looking to get these machines. So they haven't really received all of the demand, that the full demand that they have, we've limited in what we are able to actually book. And so that's actually restricted the amount of demand out there or the amount of supply that they have with respect to what they'd really like. And I think to those 2 factors, long lead times, longer lead times, both in the ship and the qualification, but also the restricted capability they've had to actually receive machines over the past year plus will likely push you in a way where they're going to take the machines, if they view the recession to be more of a moderate cycle going through here.

And what means you think that they don't pull in or they don't over order because, obviously, nobody knows what's going to happen in 2024, but we know there are a lot of little tools. So what makes you confident that you won't have like this big drop after big in 2023?

Yes. So I think from a -- looking beyond, sorry, I didn't fully give you beyond '23, but I think we actually -- last month at our Investor Day, we talked about '25 and 2030, and if you look out, you see the continued growth in terms of demand. And we actually went into that not only in units, what we're seeing with respect to our units, but we also talked about the different markets. And we can talk about that in a bit more detail. But I think in our view 2025 will continue showing strength in that respect. So then the question just becomes what happens if '23 delivered or respective then you have 2024. Obviously, we haven't guided anything out in '24. So I don't want to get too far in front our cells, let's get to '23 first. But I think with the significant demand there, the fundamental secular drivers longer term, I think that gives us confidence that we see the trend heading to where we talked about last month in 2025 and beyond.

Okay. That's clear. On China, it's also a very common question in this industry. Can you summarize what the impact of the new restriction is for your business and your backlog in the business? Just [ any thought ] would be great.

Yes. So first off, maybe just to level set with respect to the facts of where we are with China. So China with respect to export controls, we're not able to ship EUV. That's been in place for a few years now, so nothing new there. And then the most recent communication from the U.S. government with respect to export controls did not change with respect to litho. So -- which is why we said -- that we saw we've very limited impact with respect to our view because no change litho, meaning we'll continue to ship DPV systems into China. Now we did though quantify and said we had an indirect -- potential indirect impact. By that, we mean that if these Chinese customers that are -- have a targeted technology are unable to secure non-litho equipment for these particular technologies, will they take litho? And that's a question mark. And so because this is a question mark, we defined -- quantified in our backlog, what would be in this category in the, particularly, in DPV. And we quantify that as 5% of our backlog. Our backlog being EUR 38 billion. So that quantify the number. But again, that depends on their decision and whether they'll take those litho tools in spite of the current new changes to some of the other non-litho areas as it relates to export controls.

Okay. So since the new restriction has been in place, you don't see any impact on demand so far for your litho?

Correct?

Okay. So let's move on to the DUV, EUV and so base management for 2023. Just a quick zoom there. You said in the past, first with DUV that you were under shipping the demand by 40% in the last few quarters. How is it today with maybe the supply getting better demand may be falling in some areas, how is it comparing today that with maybe a few quarters ago?

Yes. So I think just over the past -- like we talked about in October, there has been some communication around memory customers saying, hey, 2023 CapEx, maybe some changes there. We've seen some adjustments in terms of timing from those customers. However, we've also seen other customers in -- move in to those changes in time. So in other words, if some of the customers want to push their time later. We've had other customers backfill in terms of demand. So fundamentally, the demand, not any fundamental change. We still are operating with DPV demand and around the 40% above what we're planning to ship next year. So still, again, very strong in that respect. And again, we'll have to -- EUV is the same. We've talked about demand higher. We haven't put a percentage on it. It's also higher than we're planning from a demand perspective versus supply.

Okay. So you don't see any cancellation. I guess it's not something very common, the cancellation in your industry, small pushout, right? So despite what we see from Micron publicly about this CapEx down, SK Hynix, you still don't see an impact. Do you expect any impact from this? Or...

Well, I think so it's -- again, primarily, we're talking about DPV because I think they were pretty clear not to get into customer-specific comments. But I think from a -- to your earlier discussion we had is that they're going to hold especially your strategic type investments that are going to be about future technology, which EUV falls into that category are not likely to touch those slots. DPV may be a bit different if you're going to look at capacity if they're changing -- planning to push capacity a bit to the right. But you have the ability with machines, customer A going to customer B, we can reslot, reallocate to a different customer. And currently, the demand in some of these other areas is still quite strong. And so the desire to get a machine earlier versus later is still there. And so some of these movements in time are being quickly backfilled by other customers. That's the situation we're in today.

Okay. Great. And when you look at the mix within DPV, as always the logic and memory, we started to talk about but logic has been outperforming meaningfully and even more in 2023 by the look of it. So how sustainable, what is driving this outperformance from a fundamental perspective of logic foundry. Is there any fundamental reason behind? Or is it just a cyclical element? So how can you justify because you are -- maybe you can remind us your exposure to logic versus memory that -- so why it's important?

Yes. So I think if you look at the past few years, and you look at our revenue, system revenue, you see that our logic revenue -- logic versus memory split is around 70% or so, plus or minus, in the past few years. So clearly, a stronger percentage going towards logic versus memory. Combination there. I think first off, if you look at the number of layers, look at the mix of layers and therefore, the litho spend on a per wafer basis, it's the highest from a litho intensity perspective on logic compared to DRAM or NAND. I think that's one. The other is that, again, when you talk logic, there's a broader, we talked about these advanced and mature nodes. Mature falls, a lot of that case, into that category as well. So you're also not only talking about the advanced growing. If you looked at our Investor Day material, we talked about future nodes, node-on-node growth, in logic around 30%, whereas memory, it's 10% to 20%, node-on-node. So future nodes are growing not as aggressive in terms of spend per node as they are in logic and the absolute number is higher in logic. So that gives you a stronger demand there in logic. The other piece, as I mentioned, being the mature versus advance, you only have this long tail of advance that goes into -- sorry, long tail of mature that goes into the logic bucket, but also these new applications that we're talking about, secular ones around some of the server space today, but also the AI, the 5G that we've talked about, those things are all going into the advanced logic. Yes, you need memory with it. But it's going to be a logic being a primary secular driver. And I think that combined -- those 2 combined is really what's fueling the strength in our logic space, and we expect that will continue going forward, as we talked about in our Investor Day material.

Okay. Great. That's clear. Maybe moving on to EUV. So you mentioned 60 tools capacity, we can say shipments for next year probably versus 55 maybe in 2022. How is the demand versus capacity on the EUV part today versus maybe 3 months ago? Do you see any change? Because we saw some weakness in the smartphone area and maybe on high compute side of things. So do you see any impact on your EUV business?

We weren't so specific, so our 60 plan or more for 2023 is still plan, as we talked about earlier. But the -- we weren't so specific exactly where the demand -- how much higher. We said it was higher than what we could supply, but we didn't say how much higher. I think that's something that I talked to our earlier point that technology transitions are something that's quite strategic. So customers are not looking to move or change their view on the EUV machines because it is focused on new technology, but also the lead times being 12 to 18 months out. We're not going to move those and go to the back of the line. So I think that combination is keeping that demand pretty stable as it relates to EUV. And therefore, again, I think our focus is really going to be around next year as it is in DPV on how we can maintain, manage a higher number of units out really focused on the supply side.

And you mentioned your backlog of EUR 38 billion, I mean, how much is the EUV versus the DPV out of this backlog?

EUV is roughly EUR 20 billion of that, of the EUR 38 billion, so EUR 17 billion of that is in the non-EUV piece of it.

Okay. And deferred revenue is something that impacted your business and mainly EUV in 2022. How should we think about deferred revenues and maybe you can do a quick summary of what it is? And how should we think about going forward, '23, are we going to see the end of it or is it like practice that will continue?

Yes. So when we talk about delayed revenues, it's really tied to fast shipments. So maybe some of you have been involved with this and understand it, but others not familiar with it. So maybe I'll start with an explanation of what is a fast shipment versus a standard shipment. So standard shipment when we build our machines, you integrate, obviously, the hardware, you build up these machines. And then in our factory in Veldhoven we run a set a series of tests. And they set a test, we call it -- and our factory call it factory acceptance tests. Long list of tests. But once we complete those tests, that, in turn, we get sign off, that triggers revenue and then we ship the machine or we ship the machine, then we trigger revenue. Then the machine goes to our customer site. It then gets reassembled, facilities qualified, so on and so forth. Then we rerun a set of tests at the site, we call it site acceptance test. They're basically mirror the type of tests that we run in our factory. And then we'd lease the machine to production. And so in fast ship scenario, first off, why are we going to do a fast ship. We're looking to save cycle time. Customers want the machine quicker. Obviously, we're in a supply-constrained environment. So we're looking at what we could do at our factory to try to save a couple of weeks. And so we said, okay, if we skip some of these tests on our factory, we looked at how to test, we looked at the test list and said, which one of these tests are reasonable to pass, meaning we're not expecting a problem and therefore, if we skip them, we will not be exporting problems in the field. So we identified a number of tests that we could skip, that would save us you could say, 2 to 3 weeks, 2 to 4 weeks of cycle time in our factory. And then that would get to our customers sooner in production, and therefore, we get more machines out of our ship from our factory over the course of the year. Now we did this. The good news is that by the time you ship the machine to our customers qualify these under fast shipment, we found that the time from shipment to qualification was very similar, meaning we weren't exporting problems. But because you weren't recognized or completing all these tests, sorry, you were not recognizing revenue at shipment. The accountants want to see the full tests run. So you had to wait for the site, which is roughly 12 to 14 weeks later. So it's a whole another quarter. So therefore, you have this continual movement or delay of revenue until you receive site acceptance on these machines. Now like as I said, the good news is that we are actually showing a consistent pattern from shipment to qualification, similar to our standard test. So we would like to go back to our accountants and say, hey, can we recognize revenue at shipment. By the way, both the standard and the fast have the same cash flow, but it's all about revenue. So the question here on revenue is, can we get revenue recognition match shipment even when we run the fast shipment because we've demonstrated through a number of months of data that we are delivering the same type of performance under a fast shipment versus a standard shipment. That involves customers. They have to say we're willing to sign off to do this. And the accountants have to say we're willing to do this as well. We're not there yet. We're still working on it. But we do have, again, a plan of working through the customers and the accountants. And hopefully, we have something here in the not-too-distant future where we'll have revenue at shipment, on fast shipments because we're likely to continue to do these going forward. Why? Well, because we save it a couple of weeks, customers get the machine earlier, and we're not exporting problems to the field. So that's our plan going forward, but we're not there today.

Okay. So 2 to 4 weeks out of 20, right?

Somewhere in that range, yes.

Great. So one last question for '23 is in installed base management, something that we don't have that many questions, but it's a very strong business model as well in there. So can you talk about the drivers and outlook for 2023 on this part?

Yes. So installed base is a combination of service revenue and upgrade revenue. That's what goes into our what we call service -- or sorry, installed base business. The service piece grows in a pretty, call it, linear fashion as you ship machines, the more machines you have, you get service from those machines. Again, EUV is starting to contribute in a more meaningful way in time. So that's driving the service side of it. The upgrades business is a bit more lumpy. And as you may recall, we talked about over 2022, we're running at very high utilization levels, business is very good. Demand is very strong from our customers with respect to their wafer output. And so to do some of these upgrades, you have to take -- if it's hardware especially, you have to take the machine down. If it's a software upgrade, you also have to take the machine down, but you can recover it in say, hours as opposed to days. So in 2022, we did a lot of software upgrades same with 2021. Because they were -- did not want to take machine down for a day. Okay, maybe hours even there, they were hesitant to do, but they were able to take it down and do the upgrades. Now if we get some you can say, pressure off the utilization, if you will, or some utilization level reduction. Maybe they'll be more willing to take the machines down over the course of next year and do some of these upgrades. But that's still to be determined. So we'll have to wait and see how that upgrade business -- the upgrade business will evolve next year. There's opportunities. There's plenty in the pipeline for our customers to upgrade, especially the hardware upgrades, but we do need to get the machines -- availability to machines to do those upgrades. So next year, I don't know if 10% or more. But we'll talk a bit about probably more of that detail in January. But it's really going to be the upgrade part that's the core -- the part we'll have to assess better for the year as we get a little closer to 2023.

Okay. So before moving to the long term, I have received 2 questions actually from the audience. So I will ask them before going to the beyond 2023. So I'm just going to read out loud, TSMC tripling spend in Arizona, does it mean some double ordering of equipment in both Taiwan and the U.S. in order to minimize potential yield issues in the U.S.?

Yes. So I think the -- TSMC is just an example, but I think, again, you look at what's going on in Intel and Samsung and other customers, as it relates to spending in the U.S., and it's again, what we've put in this topic of technological sovereignty. By that, we mean that there's clearly a desire of different governments across the globe to onshore or reshore some of the semiconductor manufacturing in their own countries. We see that with the CHIPS Act in the U.S., a similar type of act in the EU, where they're looking to bring the fabs being built back in this case in the U.S. in Arizona. And I think, yes, that will create some inefficiency, and clearly the most efficient scenario you can come up with is one customer in one fab, building all wafers. Every time you move away from that, you create some inefficiency. And so this example, building in Arizona, you can say, that the technological sovereignty by design being a technological sovereignty initiative. Governments want the capacity on their shores that will create some inefficiency. I think that's, again, by design. However, I think our customers are going to be smart with how they do this and how they manage their demand and where they choose to build. So they'll look at what they need in terms of output, and they will look across the globe and where they have their different fabs and they'll decide accordingly. But I think, again, by design, these technological sovereignty initiatives will create some efficiency as we talked about last month at our Investor Day.

The second question is, Intel has been investing a lot recently in next-generation nodes. How important is Intel for growth rates for ASML? Is there a risk that if Intel moderate its spending, it will have a significant impact on your growth rate?

Yes, I think, well, Intel is clearly an important customer of ASML. So I think it's -- we want to continue to see all of our customers succeed and do well. And I think continue to drive innovation. I think that's critical going forward. One thing I always keep reminding, as you always have -- when we talked about our scenarios, it's a demand-driven scenario. We didn't talk about a particular customer building certain wafers or certain regions or within the -- across the global building but more about where the end market. So the end markets are the driver. Where it's being built. Obviously, that's not as important. However, we do want to see all of our customers succeed, not only continue to drive the business but also innovation going forward.

Okay. Great. So maybe moving on to the long term. I mean nearly 2 CMDs in almost a 6 months' time, so which is not very common.

12 months. A little...

12 months. Yes, that's right. That's right, 12 months, I'm exaggerating a bit. So you increased your targets, I mean, in a very short period of time, let's say, and EUR 12 billion incremental revenues by 2023 and EUR 8 billion 2025, if my math is correct, hopefully. Can you explain that drivers of this incremental revenues versus 1 year ago? And I think you mentioned that, for example, 42% of that -- of this EUR 8 billion incremental is coming from mature nodes. So can you elaborate more what is going on there to justify such increase on the mature node first?

Yes. So I think first, we talk about what were the fundamental end market drivers with respect to what we saw in Capital Markets Day '21 versus '22. And then secondly, anything additional, which is back to the sovereignty piece and foundry competition. So first, on the end markets. I think if you look at it, probably the biggest increase in terms of growth, in terms of wafer demand came from the mature market segments, which for us is greater than 28-nanometer logic. And if you look at that, the primary market segments were around industrial and automotive. And we actually, in the Investor Day material, we went into that in a bit more detail and the automotive is a combination of electrification as well as the whole automation at us that's evolving here in the automotive sector. We already had it in 2021 at a healthy growth rate, but not even -- again, we continue to underestimate I think that market segment. Industrial, it's a bit broader you had -- we not only have the whole automation, robotics, but also we even put a slide in around electrification, that whole grids, everything from moving from an ICE or say more of a -- not ICE but more of a coal-driven society to the renewables around solar and wind. But also taking all the way through a smart grid and eventually into the automotive. So those 2 match up at the end there. But that was primarily driving, you could say, the more mature market segment. On the advanced and maybe put numbers on these. In Capital Markets Day 2021, we said that with -- if you roll up all these end market demand, it requires 505,000 wafer starts a month a year. In Capital Markets Day 2022, we saw that number is around 780,000 wafer a month a year. Of that, that incremental increase, roughly 180 of that came from mature. The remainder of that came from advanced logic. In memory, we basically left unchanged. On the advanced side, there are really 2 primary drivers. First, on the end market segment, the end market segments that drove -- is primarily driven out of servers and AR, VR. Those were the 2 end market segments that the biggest impact, let's say, on the advanced logic demand. And then the other piece of it came from increasing die size. And increasing die size is due to, as you increase performance, but you want to do this at a lower power or manage power or, let's say, manage power in general. You have to increase the die size to do that. And that's not in itself a new phenomenon, but it's been going on since 2005, which we showed last year in Capital Markets Day. Martin, our CTO, showed it as in the Dennard scale. But it's the number of applications that are adopting this, that are driving -- and driving up the wafer demand there on that front. So those are 2 key drivers in the advanced segment. So if you put those 2 together, you're at 780,000 wafer starts a month a year. And then we said on top of that, we said there's roughly another 150,000 wafer starts a month per year, what we call technological sovereignty piece, which we were just talking about earlier, which is government's desire to bring technology onshore as it relates to semiconductors. And then the other piece of that was foundry competition. And meaning you have now 3 foundry at the leading edge as in logic. There's 3 different customers that are driving the foundry logic demand there. That creates some additional efficiency as well. So you roll those all up. That obviously gave us a stronger demand than what we saw a year ago, and we talked about capacity. That means we have to drive our capacity in DPV up towards 600 by '25, '26-time frame. And EUV, we talked a year ago about maybe you have 70 EUV units by 2025. We now need to have 90 million by 2025, '26-time frame. And then on High-NA, we mentioned that we should be somewhere around 20 or more by '27, '28-time frame.

Great. I will talk about High-NA, but I just received a question from [indiscernible]. What are you thought on litho's share as a percentage of total wafer fab equipment for the next 5 years and why?

Well, first off, we don't talk about WFE [indiscernible] ASML. Let me explain why? And I'll say then, but I think we've given you everything, we're -- as you say percentage of WFE where does litho fall? Well, I think we've given you the numerator, and we haven't given you the denominator. Why we haven't given any denominator because we don't know the non-litho piece of it. So we've given you -- basically 2023, we've given you this year, we've given '25 and we've given you 2030 data points. So from that respect, I think put in that -- you can put those in the numerator based on your assumptions on where you think the market will be. We've given you a range, a low market, a high market and where we see our revenue landing. We've given market shares that are assumed in that. So you can calculate litho. So I think with that data, you then need to see to figure out what the denominator is going to be. And I think you need more than just litho to do that. And so you'd have to get that data from others to determine. But you can create scenarios. We talked about litho as a percentage of semi revenue because that's something that we did talk about with semi revenue going from EUR 1 trillion to EUR 1.3 trillion by 2030. We gave you scenarios for '25 and '30 for our revenue. And it looks like if you just look at the third-party data, litho versus semi revenue that the assumed third-party number has it on the lower end of our scenarios based on the lines we put on that graph, which is there as a slide. If you haven't seen it, I encourage you to look at our website under Investor Relations, and you can see the Investor Day material and Roger Dassen's slide about litho intensity, you can see what I mean. And so if there's -- if we move to the mid to high end of our range, obviously, there's opportunity to show a trend, continue to grow in litho-intensity. But we don't talk a lot internal with respect to WFE. But we will give you all of litho of what we know our numbers, and you can do the math on the rest.

Great. That's helpful. Then you have a good idea on the EUV demand and the ramp going forward. After EUV low-NA you have the High-NA. Can you talk about the demand for High-NA and when are we going to see it? And also, what's going to be after High-NA? I mean, because one of the questions we have for many long-term investors is what's next. And I think one of the highlight of your Capital Markets Day, you are more open to talk about the High-NA potentially. But maybe you can provide more details of what's the road map basically?

Yes. So High-NA, so just to calibrate, today we talked about the machines we're shipping today, some may refer to them as low-NA, it's 0.33. That's the number of the numerical aperture. That's what NA means. And think of it as basically the size of the lens. When you go to High-NA, it's 0.55 NA. And those are the machines that we'll be shipping in volume in the '25, '26, I think we'll start towards the end of next year with our first machines, but in volume in '25, '26 looks to be when both logic, but also memory will be adopting High-NA in their production and their technology going forward. And again, I think you look at it why High-NA. High-NA -- without High-NA, you would go to multi-pattern, low-NA. And so it's all about High-NA reducing the process complexity, reducing cycle time, improving yield by reducing the process complexity, like Low-NA replaced for multi padding immersion. So that's the whole purpose time frame. Again, '25-time frame in real volume, '25-, '26-time frame in volume production. And you'll probably see some type of layer, we talked about layers in our Investor Day where they'll start any way where you'll start ramping in a similar way like we did with Low-NA, the most critical layers and continue to adopt layers going forward, such that you'll see as you go '25 to 2030, you'll see a mix of low-NA and High-NA units in time. We're starting in the process of integrating a machine in the factory as we speak. Some of you were over in Veldhoven last month, I went and saw the machine and again, making progress, I think, on the program as planned. And again, that will take us out into 2030s. But then the next question comes, okay, what is beyond High-NA?

Greedy question.

Out in the 2030-time frame. And I think if you look at it from a litho perspective, you say, what are your options? Well, the way you make smaller lines, either change the wavelength of light because the equation -- the Rayleigh's equation is, wavelength over the numerical aperture. So if you lower the wavelength, you get smaller lines or if you increase the numerical aperture, which is the dominator, you also can increase -- decrease the feature size. So having gone through a -- just recently a wavelength change, what you normally try to do is push the numerical aperture, like we have in other wavelengths from KrF to ArF and now EUV. So it would be logical to say, wouldn't you then expand your numerical aperture? And I think that's a logical conclusion and why there's been a lot of speculation around Hyper-NA. But as Martin said, our CTO, at Investor Day, we have a technology, but it's more about you need to make sure it's done in a very cost-effective way. And so that's where they're all focused. We'll continue to research in High-NA and look for the right timing in the 2030s when we need to have the next numerical aperture. And then the focus will have to be on driving it in a cost-effective way, such that it brings value to our customers.

Okay. Great. And the last question because we're running out of time. Can you -- you mentioned the product road map. What does it mean for the ISP road map within this and ultimately your gross margin trajectory?

Yes. So if you look at -- we roughly deliver products, our Low-NA products, for example, on EUV on a 2-year cadence. And if you look at the EUV machines that we're shipping today, the 3600D, we're in the EUR 160 million, EUR 165 million ASP, in that range, which we've talked about for a number of quarters over the past few years. With the 3800E, we plan to start shipping that machine in the second half of next year. The productivity on that machine starts out around 195 and then goes to 220. When you compare that to 3600D, which is around 160 wafers an hour, that's what 35% more in terms of a productivity gain. So it has a potential to have a healthy increase in ASP. We haven't specifically talked about the number, so I don't want to say anything yet there. But I think there's an opportunity there. And with those, when you see we do these model changes that usually also offers us the opportunity to improve margins. So I think that's the next step. And again, if you then look out beyond 2023, you'll see in 2025-time frame, on the road map that we communicated that we bring the next models out in time. So again, with higher performance, higher ASP opportunity to further drive the margins on Low-NA. And then High-NA, again, we'll start off, as I mentioned, towards the end of next year shipping the EXE5000. We mentioned ASP around EUR 270 million. That will then transition to EXE:5200 in the '25-time frame that will be significantly above EUR 300 million, which is what we said at our Investor Day. So -- and again, there will be models that we'll follow on in time, that will offer opportunity for not only ASP but margin improvements.

Great. Thank you. We are running out of time already. Thank you, Skip. Thank you all for listening.

Thank you, Francois. Thank you all.