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

Hi, everyone. My name is Francois-Xavier Bouvignies from UBS, head of the Tech Hardware team at UBS. Today, we're happy to have Skip Miller, head of IR at ASML; and Pete Convertito, head of IR, North America at ASML. So thank you both for being us -- with us today. So we're going to have a couple of questions, and unfortunately, we won't be live. But I have a lot of questions anyway. So hopefully, it will still be helpful.

And maybe, Skip, if we can -- and Pete, if we can start with the first one, with regard to the demand, simply when we look at Logic and DRAM, how do you see the demand going into 2022? And if you can also separate deep UV and EUV will be also interesting to get a sense of the outlook there, please.

Yes. All right. Yes. So thanks, Francois, for having us. And let me start with the different market segment demands and maybe talk about EUV and DUV in particular. First off, on Logic. Demand continues to be very strong. I think as we keep talking about as part of this ongoing digital transformation and this broadening application space. I think we've talked about secular growth drivers 5G as kind of this enabling pipeline. You talk AI, artificial intelligence, you have high-performance compute. Peter in our Investor Day talked about this distributed computing and connecting all these things and how the number of connected devices is really exploding. Each of these obviously require a lot of advanced technology. But also, I think if you look at the digital infrastructure, as we call it, the things that connect so that you think about sensors, you think about controllers, power management, RF, all those type of applications, drive a lot of mature. So we're seeing this demand in Logic strong, both in the advanced and mature nodes. And again, this is driving strong deep EV demand. The strong deep EV demand goes to both advanced and mature. And then on the advance, you also have EUV, which I'll comment on more. So that's Logic. Memory, I think if you look at what's the end market drivers for Memory, you think of things like servers and smartphones. Those are the primary drivers. Obviously, there's also a long list of other applications that just like Logic, if they need Logic, high-performance Logic, they typically need high-performance Memory, so they go together. Customers' indications are next year that the bit growth will be, let's say, around the high teens. And if you look at the litho tool utilization for a Memory today, it's very high. So customers obviously will need to add capacity in addition to make the technology transitions to meet this higher bit growth or this bit growth, you say, in general, for next year. I would say the other thing with respect to Memory, Logic, we continue to see the growing layers of EUV on future nodes, and you're seeing maybe stronger advanced nodes than we did, say, just a few years ago. Those 2 things combined are driving the EUV demand in Logic. And on Memory, we're starting to see customers put EUV into volume manufacturing. And all 3 of the Memory DRAM customers have all communicated their plans to put EUV into their product road maps. And therefore, that will help drive EUV in the Memory side of the segment for the upcoming years. So yes, in summary, I think we expect continued growth next year for both Logic and Memory. We haven't put specific numbers or percentages on those yet. That will come as we get into next year. On EUV, we're still planning for capacity around 55 systems next year in 2022. As a reminder, those will all be D systems, so the higher productivity machines there. DUV, first, maybe a comment about this year and that is that we had to utilize or we call our safety stock or think of it as your buffer inventory to produce a significant increase of output due to the higher-than-expected demand as we went into the year. As you know, we saw every quarter kept getting stronger and stronger. So we had to really drastically increase our output. So using that safety stock in this year, we have to build that back up. Now we're increasing our capacity. As we communicated during our Investor Day, we're planning to increase the deep UV unit output around 1.5x when you take into account the productivity goes closer to 2x. And that will do between 2020 and 2025. So next year, we'll have to build up that buffer inventory. And we think right now, we believe we should be able to at least reach around our 2021 shipment output next year for DUV. So that kind of summarizes the the markets and again healthy growth next year on both fronts and DUV and EUV in terms of what we're planning for capacity next year.

Okay. And I mean in light of this strong demand, you got also impacted by the shortage to some extent, so like you disclosed last quarter. Can you give us a bit of a sense of where you are today and when you will see the recovery and the catch-up from the impact that you saw in Q3, Q4?

Yes. So just maybe as a summary of what we said in our Q4, we experienced some issues regarding material shortages in our supply chain, but also we experienced some issues with respect to start-up of our new logistics center. Both of these taking into account resulted in a delay of start of the assembly of some of these machines. And so it delayed revenue from Q4 of this year into next year. We're working obviously on both of those fronts, the Logistics Center continue to get that, bring that online and making progress of a plan there. And then on the supply chain, we continue to monitor. Obviously, there's -- we're not the only industry that's been managed around these supply chain issues. And of course, that risk is still there. So you can't say it's gone in this environment, but we are practically working with our supply chain partners regarding I'll say, more efficient communication. When they have a chip shortage, escalating obviously, to the ASML team and so we can help and provide the proper prioritization and minimize any delay. So that activities are ongoing, and we'll give updates as we go through this quarter into next year of anything new on that front and how we're progressing on each of them.

And one of the impact of the shortage, and we have a lot of discussion with investors is like you may have double or triple ordering in the supply chain across different segments. And also for you, in a way, it may be an impact. And on top of the shortage, you have the geopolitical tension with China. You have the pandemic, creating some supply chain issues. So how do you assess this risk of double or triple ordering, which is for your own tools, but also down the supply chain? How do you assess the risk, so you don't overbuild capacity, if you...

Yes. So I think if you look at the demand environment out there, and today, the demand is here and we're supplying to, obviously, not to that full level. If you then look at that and break that out a bit by customers, and we get back to talking about our advanced customers versus the more mature nodes. I think on the advanced nodes, if you really look at the advanced nodes that take say the majority of our immersion and also our EUV and you could say it makes up the majority of our revenue. Those customers can find to, say, 10 or less in terms of total customers. So we can have those discussions, very comprehensive, very routine communication with our customers, I mean, technology through capacity planning, engineering level to senior executive and CEO, CFO and CTO level. So you get a very in-depth understanding of what they're doing from each of these key customers. And I think, therefore, the risk on the advanced nodes and the advanced -- the customers that produce those advanced nodes, the risk of double ordering are pretty low. Now if you look on the more mature technology, you can almost think of it as going back to when the industry was, let's say, 30 so years ago, when you had a high number of customers, you have 20, 30, 50 customers, that we're all competing for the same space and same end application, same business. And I think in that scenario or in that space, think today. There's lots of customers, and you can't properly track every little piece and making sure they're not chasing the same business and there'd be some double ordering in there. So there's possible that exists. But why at least in the near term, if you look at '22 and going into '23, I think the -- back to the story here in terms of demand versus supply, demand is here and supply here. And we say that gap for that older technology that drive business is probably the most significant on the mature or the dry technology for us, dry litho technology compared to, say, the emergent and the EUV space. So even if there's an error bar around that demand that would be to take out some of the double ordering, you still have a ways to go before you get the supply. So I think that from that perspective, we don't see a major concern in terms of double ordering in 2022. Now as you get closer and that supply/demand gap gets smaller, then obviously, you need to be more attuned to making sure that that's okay. But I think that the near term, that's not a major concern as it relates to the double ordering. And I think that's true in the supply chain, if you look at the components that we're making, we're ramping our capacity so that's needed. And again, some of the error bar could exist, but we'll manage as we get closer to the -- when the demand and suppliers are more tight in that respect.

Okay, makes sense. On the -- so since you talk about the capacity and supply, you touched upon the EUV capacity of 55 tools that you expect for next year. Can you give us the road map that you have for [indiscernible] 2025 for EUV capacity and also deep UV, given all the demand, what the capacity you're building to cope with this strong demand?

Yes. Yes. So first off, maybe again talk a bit about demand and capacity on each of these. Regarding demand, first for Logic, I think EUV, the increased EUV demand is driven by 2 things: one, increasing EUV layers on future nodes. And at the same time, the customers are seeing stronger demand for the more advanced nodes. I think that combined is driving the strong EUV demand for Logic. I mentioned that earlier. On Memory, there's all 3 now, as I mentioned, all 3 DRAM customers now have communicated plans to implement EUV in the road map, and this is -- they have a -- If you look at the layers, the initial adoption is on a few layers and then they scale the number of layers in time. And I think the other thing to keep in mind when you think about EUV into Memory is that you say, okay, it's a fewer number of layers maybe than Logic, but you have a lot more wafers out there. So there's a lot of DRAM wafers. And that, call it, say, the vast majority of those total, let's say, ballpark 1.4 million wafers starts a month. The vast majority of those in time will be converted to the most advanced nodes and therefore EUV. So it's more of a wafer-driven story than, say, a layer-driven story in time. Regarding for that's all demand and then for the EUV capacity side of this, we communicated a plan at our Investor Day this year, where we plan to increase EUV, if you look at a unit capacity increase over 2x from 2020, time frame 2020 to 2025. And then the other piece, which is also critical in terms of providing our customers with capability and capacity has to do with our EUV product road map and our increase in our productivity and time. And so as you know, today, we -- or last this year, we transitioned from a C to a D, which gave a 15% to 20% productivity increase somewhere in the middle of next year, we've shown on our road map that we had transitioned from D to an E, which will give another boost somewhere in the range of 200 wafers an hour or more, so to have another step-up in productivity. And when you map that productivity road map into the capacity plan, and put these together, call that effective wafer capacity increase, then you see on the order of greater than 3x increase in effective wafer capacity for EUV between 2020 and in 2025. Regarding DUV, similar, I think I made to earlier question, is that we are expecting from a unit perspective to be around 1.5x increase, 2020 to 2025. And then when you take into account the productivity road map for DUV, that number from a capacity -- effective wafer capacity increase on DUV would be on the order of 2x. So both those were a driver capacity plans. Now what's -- what are we doing in that arena. And that's between now and 2025. There are really 3 things that we're utilizing to improve the capacity. One, cycle time. So if we can drive the cycle time down on each of our factories. And we talk -- when we talk about these capacities, we're talking about our factory -- factory or factories as well as our supply chain partners and their factories. And so the first being cycle time, just how long it takes to get a component or a machine through the factory. The second is if you have the existing space, you just need to add more equipment or more people. That's another step. And then the third is if the buildings full, you need to build new buildings. And each of those have different time horizons, the shortest being cycle time improvements, obviously, going to level where you need to add new buildings. And so each of those in our -- if you look at our entire supply chain, there's each of these activities that are going on and each of them have different time scales. And so we're driving that to meet this capacity increase that we talked about at Investor Day. And we'll continue to have discussions with our customers to make sure that, that meets their demand in time.

Okay. Makes sense. So you are doing all 3 now at the moment.

Yes, correct. That's correct.

Okay. Good. Maybe from a high-level perspective, you laid out your 2025 plan and your 2030 kind of a hint, I would say, on 2030, I mean less detail, which is kind of normal given the long horizon. But just starting to the 2025, I guess you need to assume some wafer fab equipments, [indiscernible]. And when you look at lithography, I mean, it's trading at a very high level versus history. Of course, EUV is ramping aggressively, but I just wanted to have your thoughts around how do you see the wafer fab equipment moving out to 2025 because I'm sure it's something as well you need to think for your forecast. And also the intensity that -- do you expect this intensity to remain flat or increase over time with EUV, the D to C to E level that you described. Just to get a sense of how lithography and ASML is playing in this wafer fab equipment market.

Yes. So first off, as you know, we typically don't talk WFE because we're -- we have only a piece of that being the litho. But we did provide some growth numbers during our Investor Day with respect to the different segments and looking at through semi CapEx to litho. And based on this, you say, current view on the end market, demand and contribution of third-party analysis, we provide a semi CapEx, CAGR between -- and we provide a period of 2017 to 2025 of around 5.9%. Now at our 2018 Investor Day, this CAGR was around 3.5% for the same period. So you see definitely that strengthening in terms of semi CapEx demand, which, again, I think is a strong as a reflection of the stronger end market demand that we talked about, the strength as well as the mature nodes, both, I think, are stronger than what we saw in 2018. If you then look at the CAGR on the litho CapEx front for the period 2017 to 2025, it was just under 14% in the update that we provided in our 2021 Investor Day. And when you compare that to our view in 2018, it was around 7.5%. So again, you see that strengthening as you get more layers, more litho intensity in time. So what does this mean in terms of -- and we also provided some directional alley in terms of the spend node-on-node, where you see Logic and Memory all stronger node-on-node in time. That obviously means more layers on a given node, but it's still cost effective in the sense of driving lower cost per share [ in ] transistor. But you roll all this up and it looks like that it translates to a growth in litho intensity as a percentage of WFE moving from, say, the low to mid-20s to around 25% by 2025. So that's, again, that's the -- that's our update that we provided at least tried to address some of the questions around WFE that we don't talk too much on a daily basis.

Okay. No, that makes sense. We have a lot of question on that as well. So that's why I asked. Maybe moving on to a different subject. It's something that is very common in the press and a lot of attention is China and the geopolitical tension as we are all aware of. One recent article is talking about maybe SK Hynix, having some issue with deliveries in China for EUV. So can you explain the issue here, what -- how it can impact you as well? I mean with that regard. And the export license as well that -- Is there something that you need to ship in China. Can you give us an idea of how does it work exactly? And do you need to apply for one for delivery at your customers even if it's not Chinese.

Yes. So I think, first off, I'd say that nothing major new on the China front. And that for EUV, we -- as it's been the case for now quite some time is that we need an export license from the Dutch government to ship EUV into China. This is still waiting on approval. So we're currently not able to ship EUV systems to China. I think that's nothing new on that front. For DUV, which is the our business in China. And frankly, it's where the nodes that are currently operating in China, they're not at a point of needing EUV. So for deep UV, we continue to ship systems to China, immersion and dry, and that's the strive in the rig our customers. So I don't think there's anything major news on the China geopolitical front with respect to how we do business. We continue to operate as is and have a large number of systems in China today.

Okay. Okay. Maybe on the more long-term and technology side. There are 2 big debates that we have and we have with investors at the moment. One of them is around gate-all-around and how it could impact ASML in longer term? So maybe can you describe quickly and simply for the audience, I may say what it is and how it can impact lithography intensity, I mean, how is it impacting ASML for everybody to understand?

Yes. Okay. So first off, gate-all-around, gate-all-around, it's a device architecture. And it basically is a way of better -- enabling better control out of the gate as the industry scales to small and larger features. So the evolution is we used to have a gate that set on a - we called it planar. And so it was only being controlled really on one side of the gate, the bottom side gate. We then went to FinFET that wrap that went around the i.e., the fin that cover additional sides, all 3 sides of the gate. And now with gate-all-around, you wrap completely around the gate, and then you chop this gate up, which is why you call someone will call it [ Ribbon vets ], some call it nano sheets because there's multiple levels that are being wrapped. Again, it's all about, as you continue to shrink, you need to maintain control and gate-all-around provides us. Just like FinFET enable the continuation of shrink from planar, gate-all-around is doing the same thing. As publicly announced by our customers, that gate-all-around. It's first planning to start next year. One customer said we'll start with 2022 use the gate-all-around. Others will plan shortly after that. and they all have slightly different node names. But it will enable shrink for future nodes. And it's similar, again, if you look at FinFET, which was introduced in the 16-, 14-nanometer time frame, it basically enabled a shrink down to, call it, the 3-nanometer node, the 5- to 3-nanometer node, depending on the customer's a [ nomenclature ]. And along that way, we saw this growing litho intensity as we continue to shrink down from the 16, 14 down to 3. The -- what you -- if you look at gate-all-around in terms of what it means for litho, I think there's a couple of things you'll make. In general, what it does, it allows, as I said, the continuation of shrink, which means more aggressive layers and more layers in time, which means higher litho intensity in time. At the given node, at the node, which customers make the transition, it's a pretty major change in terms of risk with the device architecture change. So customers may choose to be less aggressive with their shrinks at that particular transition. So once they verify the device architecture, is successful and working and they've implemented correctly, then they can resume the more aggressive shrink. So at a given transition, they may choose to mitigate risk by being less aggressive with their shrink. Beyond that, though, customers resume to the typical aggressive shrink and in growing with the intensity of time. And so in summary, I would say gate-all-around is good for the industry, good for ASML in general, that helps ensure our future shrink road map by enabling device performance at the smaller features.

Okay. Makes sense. And what kind of time frame should we look at in terms of introduction?

The first one -- first customers announced the plan for 2022. And you have another customer in '23 another in '24. I think that's the 3 time horizons for each.

Okay. Okay. And another one is 3D DRAM. So it's also doing a lot of attention. So can you talk about the impact as well on the lithography intensity, same kind of question and the timing on 3D DRAM? And yes, how it can impact ASML?

Yes. So first off, talk maybe about what we today call it 2D DRAM. And I think customers continue to find ways to shrink and I would say, have gone way beyond what was expected just a few years ago. I can recall many debates about getting below 20-nanometers. We'll never get below 20-nanometers, and we'll never get into the 15. And so there is this continual -- it reminds me a bit of Moore's Law that's always at the end, and then you find lots of innovative ways that we continue to extend this capability. So I think that's what's going on today with the 2D DRAM. And if you look at that, customers have publicly talked about extending their 2D DRAM nodes now sub-10 and even some talking about going sub-5 nanometers over the next 10 years. And as they continue to do this, they clearly now have said that they will utilize EUV. So it's required EUV to do this. And all 3 now major DRAM customers have talked about -- I've talked about announced plans where they will plan for EUV on their road map and the upcoming nodes and they're buying EUV systems to support that. You could say 3D DRAM or call it, next-generation memory, but next-generation memory going in 3 dimensions is nothing new. Customers have been talking about and looking at different options for these next-generation memory for years. I can recall discussions in the late '90s, early 2000, and about going to MRAM. That was going to take off in volume production next year, right around the corner. Well, here we are years later, still talking about it. There's other ones we talk about with ReRAM, there's PCRAM. I mean until Micron talked about Crosspoint. Those are all 3D DRAM or 3D memory-type alternatives that as you go down in the shrink of 2D, and at some point, you say, we'll reach an end. These are looking at these alternatives, and they're going to continue to evaluate these alternatives. I don't think anything today identified as the solution. In the meantime, the 3D DRAM keeps extending, taking us probably closer to 2030 with the number of nodes that they've identified. And I also think maybe to say not only are they looking to use EUV Low-NA to extend this 2D, but they're also, as mentioned on our October quarterly call, we're also seeing a very significant engagement with our Memory customers around High-NA. And so that they're not only planning to extend with EUV, but also see plans an opportunity for High-NA to take it further. So in the end, we listen to our customers. I think if the ask the customer is best when how this will happen, but they're making very large investments in EUV with clear plans to shrink DRAM in the future.

It's good that you talk about EUV and High-NA actually because it was my next question. And on EUV specifically, I mean, it's going to be one of the biggest growth drivers in the next 4 years with all this ramp and profitability as well on many fronts. Can you remind us the the number of layers that you expect on the EUV in the next 5 to 10 years, just -- and by node, just to explain and understand the road map of ASML for EUV?

Yes. So we actually showed at our Investor Day, first off, if you kind of look at where we are, let's say, today for Logic, you have somewhere, let's say, north of 10 layer EUV layers in Logic today. And then if you look at the next node, major node after what's being produced today, and again, that's not an industry 7 or 5-nanometer foundry, that type of a node name. You end up going 20-plus in terms of that -- the next node. So that's kind of the next step that was communicated in terms of the EUV Low-NA. After that, now the next major node, again, it gets a bit -- not all customers at alignment. Some may choose to do High-NA and others may choose to do some multi-pattern with EUV Low-NA. So that's kind of the evolution of EUV for Logic. In terms of DRAM, today, I think -- well, first off, initially, customers usually start with a few layers in learning, and you're seeing some of the customers in learning and others are already in volume manufacturing. The one that's in volume manufacturing, you talked about having 5 layers today in EUV. And we see, as you go through future nodes, we see opportunities to that to expand towards 8 EUV layers for the DRAM nodes. And again, the High-NA comes in there at some point in time as well. And that's all being aligned with our customers in terms of timing. But that's the general flow in terms of the layer growth on EUV for both Logic and Memory.

Okay, makes sense. And for High-NA, I mean, when do you think -- When do you expect the adoption -- mass adoption of High-NA? And is it going to be a strong ramp from year 1? Or do you expect like a very linear or gradual increase because anyway you have EUV. So it's going to take a bit of time?

Yes. So in terms of High-NA, I think again, maybe why people do or doing High-NA, I think start there and say the value High-NAs obviously in the reduction of process complexity. So as I mentioned earlier on these different nodes, if you look at the 3-nanometer below, customers started to utilize double-patterning EUV. How do you remove that out of the most critical areas to go back to signal exposure, utilize High-NA. So that's the value. What it does in terms of the value proposition it provides when you say what is reduction and process complexity mean, it helps translate to less defects. So pure defects that can be imaging-related bridging defects which translates to improvement in yield, and that comes through, obviously, higher contrast that you get with the High-NA. Lower cost, if you take away some of these process steps both litho and non-litho and replace it with a single exposure that drives lower cost. And then obviously, the other one being a short cycle time, fewer steps, fewer cycle time. And that is a double benefit. I guess wafers out the fab faster, but it also the longer the wafers are in the fab, the more potential exposure you could get to defects, particles and stuff of that nature. So there is a secondary benefit there. We provided some of this detailed additional analysis, updated analysis at our Investor Day material. Actually quantify some of this stuff, but that's the value proposition. In terms of timing, our High-NA, we will -- we have orders for the initial 4 machines that we took back in 2018. Those machines will do start shipping in 2023. And then volume manufacturing, we're targeting for the 2025 time frame. So you could say as you start with this HVM ramp, if you looked at the unit numbers that we provided at Investor Day for 2025, it was around 5 machines. Now we expect to ship more than that in 2025, but those are the ones that we expect to -- 5 although it will be the ones we revenue recognize. And then if you go beyond 2025, obviously, that number will move up in time such that we're will ramp. I don't know if I could say exactly like Low-NA or to the exact same numbers, but you will get an increase in the unit count in time on an annual basis based on the increasing layers as well as the increasing wafer starts and time on these advanced nodes.

And maybe it's like a good question, but the capacity for High-NA and Low-NA, I mean, is it the same capacity, you use the same kind of capacity or as a totally different one?

So if you go to Veldhoven and you look at the buildings that have been built, I mean quite a lot since [ COVID ], but I think the other -- like I said, I haven't been there for almost 2 years. And it was amazing all the progress in terms of the building and stuff while we were all being locked down, construction continued, and that's something of the video we start our Investor Day off with kind of highlighted that. But it will be nice we can resume in-person engagements. Again, you guys can come over and see all the progress at Veldhoven. But in Veldhoven, the actual building that we manufacture High-NA versus Low-NA, the building itself, obviously, the infrastructure is different for High-NA in the sense that it's a bigger pedestal, much bigger machine needs more supporting more significant infrastructure around the ability for the weight, the crane structure, so on and so forth. So in that respect, they're different in the final assembly, you can say, at our facility. However, in the supply chain, there are a lot of similarities. So there's a lot of similarities that we can utilize there. So I think it's a combination to answer your question of -- there are areas where we have common parts supply and then there's areas where in the supply chain, i.e., the optics are different, obviously, between High-NA and Low-NA.

Okay. Makes sense. Moving on maybe to installed base management. I mean, it seems to be more and more focus for you and also an attractive growth profile, I mean, even outgrowing your product in a very long-term in a way because it's great news. So can you remind us what kind of growth you expect? And how does it work really in terms of fee per - 2 per year? How do you see it progressing with the years? And just to explain why is it going to be one of the important growth drivers as well in the next few years?

Yes. So yes, maybe just to explain. So on our deep UV machines today, they're supported with kind of, you could say, a break-fix model, meaning we put parts and people. And when the machine goes down, there's a contract with a person. And parts go with it. On EUV, it's more of a, you could say, a win-win or value sharing model. And we move to that model, that value-based service model and linked it to a wafer output with EUV. So the more wafers you produce, the better value or higher value to our customer and then we share that value. So we try to find ways to drive up not only the availability, but a good way for us out of the machine and then do it in a cost-effective way. So we all win in that scenario. In terms of what it means in revenue, we expect the EUV service revenue to be around 5% to 6% of the tool ASP or average selling price, regarding revenue per year once the system is out of warranty, which typically is a 1- or 2-year warranty. So that's what will be the recurring and it'll become as more and more of these machines are shipped into the field, our ramp into volume manufacturing, you're getting more and more wafers out, i.e., higher revenue in time. and then they will be coming out of warranty, so that will become a more compelling and contribute more significantly to our growing installed base revenue in time.

Okay. Okay. That's clear. And maybe one last question from me. It's a bit like we talked a lot about your growth and prospects. Maybe one quick question on the profitability side, which is also, of course, important. So a question I have often is why ASML has lower margins, like especially the gross margin versus some peers when arguably, you have one of the best position in the market, right, if we look at EUV and High-NA. So when some competitors have 60% plus gross margin and you are in the 50s. Do you have any plan, I mean, to close that gap? And how do you plan, especially to close that gap?

Yes. So maybe first, talk about what we communicated with respect to our Investor Day in terms of where we are. I mean the plan that we provided in 2025 is that we would expect to be around 54% to 56% gross margin by 2025. And then we also talked about R&D and SG&A in the 3-6 $3.6 billion range for the R&D and then the $1 billion range for SG&A. And so when you look at those 2, you say, okay, well, that it really is about an operating margin story. So if you take the -- some of our -- if you look at our R&D and our SG&A relative to our peers, you see they have a higher R&D as a percentage of sales and a higher SG&A as a percentage of sales. And that's because a lot of our R&D is in our supply chain. And therefore, the R&D costs from our suppliers come through in the cost of goods. And that's where, I would say, we should probably talk more, highlight more the story around operating margins versus gross margins. And when you do that perspective and you see we're moving towards this 40% number, which, again, I think is similar to if you look at the different peers. And then there's opportunity, as we mentioned in our Investor Day, for us to grow these margins in time. As we bring new products online, we bring more value to our customer. We share in the value there. That has the opportunity, obviously, to improve our value, but value equals the ASP of our systems. And if we can do that cost effectively, then obviously, there's opportunity for margin as well. So I think one is putting it in perspective of operating margins versus gross margins. And the other is that we do have, and we're showing, obviously, from where we are today in the low 50s to getting to the mid upper 50s as we get to 2025, I think there's an opportunity there to grow that. But again, I think when you do a peer comparison, it's more relevant to look at the operating margin.

Okay. Okay. That's clear. Thank you very much, Skip, I think we are running out of time. Thank you, Pete, as well for your time, and thank you all for watching.

All right. Thank you, Francois. Happy holidays.