Applied Materials, Inc. (AMAT) Earnings Call Transcript & Summary

Thank you for coming, everyone. I'm Stacy Rasgon. I cover the U.S. semiconductor and semiconductor capital equipment space here at Bernstein. And it's my true honor to have our guest here today, Gary Dickerson, the President and CEO of Applied Materials. And I'm thrilled to have -- you may not know, Gary was supposed to be here last year, and unfortunately, he got stuck circling New York in bad weather for several hours and had to get diverted to Boston. And he missed it. So he very kindly flew out a day early today, and so I'm thrilled to have him here. I wanted to mention, before we start, if you want to ask a question or have a question asked, on the inside cover of your program, there's a QR code that you can scan. That will take you to our question-and-answer format, it's called Pigeonhole, and you can put your questions in there, and we'll leave time at the room -- leave room at the end for those. So semicap has really been top of mind for many of my clients lately, especially as the strength that we've seen over the last couple of years now turns into the first kind of real down cycle we've had since 2019 for the industry, at least, I should say. Even amid all -- look, there are, of course, near-term questions, memory spending trajectory, the impact of export controls and the surprising strength right now of trailing node build-outs. I'm also getting people, though, increasingly looking at the longer-term potential of the industry as they start to view it much more in secular terms in addition to the purely cyclical. And I hope we get a lot of chance to talk about all of that today. So it gives me great pleasure to welcome Gary. Thank you so much for being here today.

Thank you, Stacy. There's another part to that story. Actually, my flight had mechanical problems. So there was an initial delay of 3 hours before we took off, and I was circling South of New York, waiting for the thunderstorms to clear, which never cleared before we ran out of fuel. We had to go to Boston. So anyway that was...

I'm thrilled to have you here out today, though. So thank you.

Yes.

I wanted to start out maybe -- I talked about the industry, WFE, like being down. And you're clearly seeing that in areas of your business. We know memory spending just broadly is horrendous right now. And it's hard to imagine it could get any worse, frankly, given where it is. And we know leading edge has [ some issues ]. But that trailing node piece is very strong. And it's actually -- I mean, your results right now are diverging from the industry. So the industry -- your revenue is not really down, in fact, growing.

That's right. Our last 3 quarters were the highest non-GAAP EPS in the history of the company.

So I mean how do we parse that? I get a lot of questions around the sustainability of that ICAPS -- because you suggest that you think it can continue even in the next [indiscernible] and beyond.

We do. Yes, yes.

Talk a little bit about like what's driving that? Where is the demand coming from? Regionally, people are worried about China. And just what can you tell us about that divergence between what you're seeing in, I guess, for the broader industry, like the trends we're seeing there?

Yes. So if we look at overall wafer fab equipment, we formed our ICAPS group, IoT, communication, auto, power, sensors, a little over 4 years ago. So when we looked at the overall market, our view we've communicated this, is we think foundry logic will be sustainably 2/3 of the wafer fab equipment spending, memory, about 1/3 going forward. And really balanced between...

That makes it a little different, by the way. It used to be 60-40...

Or even 55-45 [ when they asked ]. But I think memory -- foundry logic, I think, has been bigger than memory in the last 21 years, except for 3 years, and 2 of those years were when you had the 2D to 3D NAND conversion. But again, it is more weighted towards foundry logic. And certainly, this year, you're going to see that more than 2/3, and we think next year, foundry logic will remain strong. So we formed our ICAPS group a little over 4 years ago. We brought great technologists from all across Applied Materials just to focus on those segments. If you think about automotive and power electronics or compound semi and industrial automation and sensor technologies that are proliferating everywhere with edge computing, we saw those markets that we thought those would be very high CAGR market. Our share, if you look at Applied share, we're super strong in this race for leadership and high-performance logic. We're -- they're even stronger, but also very strong in this ICAPS market. DRAM, we're the largest process equipment company in DRAM. NAND is not quite as strong for us, in mid-teens. And packaging is another one where we have tremendous strength. But ICAPS, we saw that. We have very focused teams enabling those technology inflections. And we believe those markets will remain strong. Certainly, they're strong here in '23. We look -- going forward, we don't believe that the growth rates -- you'll see the same growth rates that you've seen, at least near term, what you've seen over the last couple of years. But we think those markets are going to remain very healthy. When you think about chips incentives -- so there's about $400 billion we're tracking over the next several years, that's mix between leading edge and ICAPS. And you think about ICAPS, ICAPS really ties to very large vertical markets and really the foundation of competition in many of those different vertical markets. So automotive or again, industrial automation, some of those areas, the fastest-growing ICAPS markets this year are North America, Europe and Japan. So we see those markets continuing to grow at a healthy rate. China is the largest market, but not the fastest-growing market. And certainly, China is very, very focused on ICAPS. Again, that ties to a number of different big verticals in China. And so again, we think that spending also will remain healthy.

So just the China piece is the biggest, but it's smaller than the other ones combined?

It is. It is smaller than that. It's less than 50%. And it's not the fastest growing.

And the other ones are growing faster.

Yes.

Okay. Got it. And then -- you've talked about, again, even in the next year, I think what you said on the call was you think ICAPS is probably sustainable.

We do.

I know leading edge right now is weaker, but you have gate-all-around and other things that are -- we'll talk about that in a minute, where there's growth. And then like I said, memories, frankly, for you and most of your peers, are so bad right now. It's hard to believe at least it could get worse, right? Is it not a positive setup? It sounds like reasonably -- at least for you, like a reasonably positive setup into next year, assuming these trends play out. I mean, is that kind of how you're viewing things? I'm not asking you for a forecast...

No, no. But I think ICAPS, again, real strength there. And we have opportunity to grow even more in ICAPS. In leading edge foundry/logic, there's this race that's happening with all of those different leading edge companies. They're all investing a significant amount for 3-nanometer, 2-nanometer, A14, A10, all of those different inflections. So if you look at -- and then of course, we talk -- look at AI and those monster GPU chips that fill up the whole radical field. That drives leading edge, that drives demand for that capacity. But -- and then for Applied, those kind of inflections are great because more and more is from materials-enabled inflections. So everybody has a smartphone here today. Your processor in your smartphone has 15 billion transistors, 60 miles of wiring. There are layers in that chip that are 4 atoms thick, 25 materials combined more than 100 different ways. So it's all of this materials innovation and 1,700 steps. Applied has more of those steps than anyone else. And more and more, if you look at what [ Mark Lewis ] talked about on energy-efficient computing road map, a lot of that's right in the sweet spot for Applied Materials. If you look at DRAM. DRAM, you're moving to high-bandwidth memory more and more and more. And so you're moving to logic-like processes in the periphery. That's -- those are all of our leadership positions. We've talked about, in one of the master classes that Mike put on, capacitor scaling. We have a very strong position there. Conductor etch, about 50% share of that market in DRAM. Packaging is even a stronger position for us. Again, we have very broad positions, very large share. That's about $1 billion market. That's also resilient. You're seeing healthy packaging investment even here in '23, into '24 and an opportunity to double going forward. So yes, I think the setup for us is great.

Got it. One last question on ICAPS, and I want to talk about some of these other drivers. So you talked about forming this group 4 years ago, and it's clearly an area that you've been focusing on. This is not new. What's different or special developing a tool for that space versus like the leading edge? Is it something that's hard to do? Or is it more around like building the customer relationship or materials inflection?

For sure. For sure. If you're building compound semiconductor device, and we have many innovations, implant is a big part of that. You have implant -- when you're trying to put the dopant into that type of a substrate, you go to over 500 degrees centigrade to get -- to make that work or CMOS image sensors. There's so many innovations that are happening that are really, really, really critical innovations across that whole space. More than 20 major products have been introduced by Applied since we formed the ICAPS group 4 years ago. So definitely, there's a tremendous amount of innovation.

And I guess it's not just 200-millimeter. There's 300-millimeter as well.

Oh, absolutely, yes. Absolutely, yes.

Got it. Got it. Have you ever talked about like that split?

I don't know that we've -- have we...

[ 58% ] of AGS is 200. The rest is [indiscernible].

I'm wondering the split of ICAPS between 200 and 300.

We can't do that because we said ICAPS is more than half of logic/foundry.

Got it. Okay. So I'll work it out after. I want to move a little bit to leading edge now. And there's a number of inflection points that are coming down that I think -- gate-all-around...

Backside power distribution.

New tools that you developed. Everyone is getting very excited about the Sculpta tool that...

Oh, that's a great one, too.

Let's talk about gate-all-around first. So I think it was $1 billion of incremental per 100,000 wafer starts per month.

That's right. Great memory.

Okay. What's driving that inflection? What types of processes are driving that upside...

Yes. So there's -- we have, in terms of process equipment, close to 50% share of FinFET. And we've said we'll gain 5 points of total spending for that transistor module and gate-all-around. So there are new materials. And again, you've got 1,700 steps. So there are many steps involved in creating those different structures. So there's materials, there's integrated materials also where we have a material that we're creating, and then we combine that with interface engineering selective removal type of a process. So some of them are combination under vacuum because again, when you go to 4 atoms thick, if you go to air, you oxidize and you damage the electrical property. There are double...

More than 4 atoms that you grow.

Yes. But you have double-digit numbers of selective removal steps. We have the majority of those steps with all those leading companies. You have material modification, many thermal processes, where you're modifying the structure, the material structure. That's a multibillion-dollar business for us. You have eBeam. eBeam, we pretty much doubled our PDC business from 2000 -- close to doubling 2000 to 2022. We're the leader in eBeam with 50% share. We have innovative new cold field emission electron optics. And that technology is really important when you're trying to optimize all of these structures, being able to see -- that high resolution enables you to build those structures better. And it has a lot to do with R&D velocity. So again, we have broad exposure across that whole module. And again, that's why we have such strength.

How far out are you looking when you're looking to develop stuff? I mean, is it -- are you looking out 10 years, like more? Like what is it?

I'd say through the end of the decade. So we're out pretty much 4 technology nodes. So that's -- and again, I think -- one of the other things we did, ICAPS was 4 years ago. 5 years ago, we built this Integrated Materials Solutions team. And -- so when you're starting to get all of this complexity, co-optimizing all of those processes together become more and more important. So we have world-class integration capability. And so we're really working together with customers when they're developing those 4 technology nodes out for the transistors, the wiring for capacitors, for any of those kinds of things. And I would say that's a big shift in Applied in the last 5 years. We're in with those integration teams. We're so deep, and we can see what they're developing many years in the future. And that's also -- we can talk maybe about EPIC later.

I was going to -- it's on my list here, but that might be a good segue into the EPIC Center. So this is not just -- it's not like the -- you have the Maydan Center for a while. So this is much more.

It's much bigger, 4x bigger. But you think about the innovation process and really our focus is innovating the way we innovate. And so today, we have the Maydan Technology Center. So we have applied all of these innovations that enable those -- all of those steps and all of those new device inflections. We also have the most -- a full-flow packaging lab in Singapore, so you can develop new packaging structures. You're developing the new chip structures in that Maydan Technology Center. So customers have been coming to the Maydan Technology Center in Silicon Valley. This concept, and we have -- when we had the event, we had all the innovators from across the industry. We had the R&D leader from TSMC and the person driving 5 nodes in 4 years from Intel and Samsung, the President and Chief Technology Officer, [indiscernible], all of them. All of them there. ICAPS customers. We had leading EDA companies, leading fabless companies. We had the top universities. So this platform -- again, there's a great opportunity for accelerating time to innovation, innovation success, right? Placing R&D teams, they're in our labs because the first place they're going to have access to those innovative technologies is in our lab. And so really getting first access or early access can take a process that is kind of slow in serial and move some of those steps in parallel and just incredible enthusiasm, Stacy, on that. And then for a customer, if you can shorten your time to a node, the R&D efficiency is massive besides being in a better competitive position. So the economics of this for Applied and for the customer is tremendous.

So does this work from the customer standpoint? Is it they have like dedicated private space within the facility?

Dedicated private space, they'll have dedicated teams there. So we'll have that with customers. And the universities -- the superpower of the United States is our university system. Many of the technical leaders in the industry, even if they were born outside the United States, they were educated in the United States. There's a lot of opportunities for us to take that superpower and amplify it. So customers are super excited about that.

Got it. How does this get monetized? Has that been decided yet? I mean are they paying you to work there? Or is there technology sharing? I don't know.

Well, I think the economics are strong. Certainly for the customers, the R&D efficiency and even -- Saturday, I had lunch with one of our top customers. Tremendous opportunity. If you like the old way, you're running evaluation wafers, then you have people flying back and forth. And then finally, you decide to put a tool on-site. That takes a lot of time. If you have that co-optimization happening, again, there's a lot of opportunity for optimization, powerful economics for our customers, powerful for Applied because we're designed in and we're already deeper connected in that ecosystem that anybody from an overall breadth standpoint and integration standpoint. But this takes it to a whole new level. And what we've said is on the economics, we're still committed to the 16% OpEx percentage of revenue. And we will -- this will be, I believe, very economically efficient. We haven't disclosed all of the aspects of how all that comes together. But the bottom line is that we're still committed to the 16%.

I get it. And then the CapEx, it was $4 billion by 2030, right? And that was a gross number. So...

Yes. That would be any incentives from government or customer investments or any of those things.

And I guess it's fair to say, I mean, this does not seem to be something that anybody else could do. They don't have your scale and they don't have your breadth of...

That's exactly right. I think that, that scale and breadth is a big deal because when you think about -- let's say we have, any point in time, 40 innovations that we're driving across the industry. The first place you're going to have access to them is in our labs. So that breadth is going to be a very big deal. And we do have partners. It's like on the hybrid bonding, we have the partnership with Besi, which is a big inflection in packaging. So it won't only be applied tools in that innovation center. So you have to have that complete innovation flow. And typically, what customers do and what we do, you have short loops. So you have -- if I'm building a gate-all-around, if I'm building a 3D DRAM or any of these different types of structures, I have these maybe 20 steps or some number of steps, the first place I look at electric results is on those test vehicles. That's where I get -- I downselect the innovations. And once I see that, then -- most of the time, what happens is customers want those tools way before they're high-volume ready. They want it in their labs, and they want to be developing as fast as possible.

So now they can do that here.

They can do that here, and it saves time. Again -- and it's also, again, more capital efficient.

Yes. No, I like it. I like it. I want to ask about a couple of these innovations you've talked about. And maybe we start with hybrid bonding and then I want to talk a little bit about Sculpta. So hybrid bonding, so just for the audience, like -- and backside power, what is that? And what are you doing -- I know you have the partnership with Besi. What are the contributions that each of you bring into that partnership?

So I mentioned these integrated platforms. Again, the one integrated platform was 7 -- ALD PVD, CVD, CMP, all of those things. That's a multibillion-dollar business for us. So we're good at integrating these technologies together. So in the case with Besi, we're taking for die-to-wafer or die-to-die bonding for chiplets, we're taking their bonder, combined with all of those adjacent steps on the surface preparation and all of those different kinds of things, putting that into a front-end platform with -- as you have tighter densities, particles become a big issue. We'll have metrology and inspection integrated into that platform. So this is a big deal. Our leading customers are the first ones pulling the hardest. Some of them already have that integrated platform. And by the way, even before they had the platform, they went to Singapore in our integrated packaging lab. Same thing like this innovation center in Silicon Valley, they went there to have -- start their development even before they had the tools. So this is a really great innovation for heterogeneous integration and chiplets.

Got it. Got it. How does it play into the whole backside power argument? So Intel is moving toward backside power and the rest [indiscernible] will be moving there.

So backside power is where you take the power lines, you have power and signal lines, you take the power lines from the front of the wafer and you put them on the back of the wafer. So that opens up, up to about 30% chip area. And companies like Jensen, he'll pack everything he can into that area once you open up some area there. But...

Or put transistors there.

Yes. So it's 3% area savings with no shrink. So it's incredible innovation. And then also, you can enable better power and performance. Now for Applied, our share of wiring, whether it's in the current interconnect schemes on the front of the wafer, in packaging or in backside power, is extremely high. So for the customers, this is a very important innovation for them in their road map, but also for us. Again, we're right in the sweet spot with all of the technologies enabling that backside power.

Got it. Got it. Let's talk about Sculpta. I thought this was -- so this was -- I've been told it's not an etch precisely, but it looks like a slanted etch sort of. And you're sort of using it to stretch out features to avoid certain EUV lithography steps, double-patterning lithography, which is very expensive. So I got a lot of calls that day. People were wondering, is this the end of EUV? Does this kill ASML? And my view is no. We still need EUV. But like why do we need this? Like what is it doing? And what problem is it trying to solve?

Well, I think -- so basically, what we're doing is again, from a directional perspective, you take a feature and you remove material in one direction. So you can certainly work with multiple EUV steps to create very tightly packed features. In this case, what you do is you take those features, and you start removing material in either direction to bring them closer together with higher density.

You're going to stretch them out.

Yes. So one of the things they talk about is that chip-to-chip spacing. So with this technology, you can accomplish that same thing without having to go through the multiple EUV steps. So the economics are powerful, and you can also have very high precision in how closely I can pack that spacing. So for Applied, again, we're using technology -- actually, it was originally developed when I was CEO of Varian. So this is a technology we developed a while ago. But yes, it's -- that's really what we're doing there. So the economics are powerful. And then also from a performance standpoint, you can get that spacing very, very precise.

Got it. And you gave some numbers for opportunity, I think, on this. And that was -- it was avoided -- that will also avoid some other action gap steps, right? If you're using double patterning, you use more etch and -- what was the economics of this for you?

I think it's $250 million per layer basically. So again...

100,000 -- is that 100,000 wafer starts normalized? Okay.

Yes. So I think, again, very, very, very powerful economics. And this technology is a new technology in the industry, this directional removal of materials. So we have these initial applications around pattern shaping. But there's a lot of different potential applications with this kind of -- this is a new tool in the tool kit.

Okay. Is it all logic? Or is it logic and memory? Or...

No, it can be utilized across multiple customers. It is being utilized across multiple customers.

Got it. Anything else exciting you want to tell us about today in terms of what you're working on?

There's so many things. I would just say this whole area of materials-enabled scaling, Stacy, when you think about all of these inflections in every one of these different device road maps going forward, more and more is going to come from materials innovation. And this is why we formed this IMS group 5 years ago. If you asked me what I'm most excited about, that's what I'm most excited about because it's very clear, and even TSMC, when they talk about energy-efficient computing, they talk about all those materials inflections. They talk about design technology co-optimization, which is basically, how do I drive scaling through all of these materials innovations. So we're the broadest with this innovation center that we're going to be in place. That's going to put us even a stronger position. And the ability to co-optimize or integrate under vacuum, that pull is stronger because those layers are so thin.

Got it. I want to ask you a higher-level question. So I want to go back to WFE. I think first, just at a high level, like -- so I think a few years ago, we were all talking about a potential sort of $50 billion kind of normalized industry WFE. And I mean, look, now near-term worries aside, we did 90 or 95 last year. It might have been 100 without the supply constraints last year maybe. This year is a downturn. Okay. We'd probably still do 70, 75, even in a downturn. And the capital intensity, people always look at like WFE divided by revenue. It is going up. And so I guess a couple of things. Can you just talk about just the drivers of that increased capital intensity, number one. And then number two, is there a limit to how high it could go? Because ultimately, somebody has to pay for this. In a perfect world that somebody is going to be the end customer. But like what are your thoughts on that? Like how much higher can capital intensity go given the drivers that we know are actually starting to take it up today?

Yes. So I think capital intensity is 15%, 16%, somewhere in that zip code. So if you look at kind of what's happened in the industry, you have these eras of computing. So you had PCs and you're waiting for the operating system upgrade and so you have the volatility there. Then you went to mobile social media. Everybody in this room has that device with them. And so it's more pervasive computing. Then you go to AI and the technology transformation of every industry. So that's going to be -- and every era is at least a 10x jump relative to pervasiveness of computing. So we see that happening for sure. And then again, if you look at energy-efficient computing. So if we don't drive innovation in the industry, obviously, the cost doesn't work. You're going to potentially consume 15% of the world's power when you have $0.5 trillion or $1 trillion edge connected devices and the data centers. So these are really powerful. I think the economics at the foundation of competition, semiconductors driving the biggest inflection of our lifetime is going to happen. Pervasiveness of content is going to increase. And then you look at how you get there. It's hard. It's hard to drive the innovation. The 2D Moore's law scaling really ended several years ago. So now you have to -- and the great thing about the industry is you always find some new ways to innovate. So that's one thing. So I think you're going to see that in the leading edge continue. The other thing you saw in ICAPS, if you go back more than a decade ago, you had to move to foundry. So you had lots of used equipment that was out there in the marketplace. And so capital intensity there was in the single digits or even mid-single digits. So no more used equipment, your 200 to 300 transition, which gave you 2.3x the number of chips per wafer, not going to happen going forward.

No 450?

No 450. I haven't heard that -- nobody's talked about that in many, many years. So I think that, that's going to keep the capital intensity in a very healthy range.

Do you think it could go up from that 15% to 16%? Like, is it plausible? I always wondered, like, everybody talked about $1 trillion in semis by 2030 or 2035 or whenever. Is it plausible we could have a $1 trillion industry with 20% capital intensity and have a $200 billion -- it was 20% -- more than 20% back in the days, right?

Yes. I think capital intensity will remain healthy. Where it's at in that zip code, I think that's the right zip code of where it will be. And I think more and more will move to materials-enabled innovations. If you think about what's happening in ICAPS, it's all materials innovations. If you think about high-performance logic, this road map for energy-efficient computing with design technology co-optimization, if you think about packaging with chiplets and heterogeneous integration, all of that stuff, 3D DRAM sometime in the future. Those are all enabled by materials. So I think the capital intensity can remain healthy for sure, and more could be moving to the sweet spot for Applied.

Got it. Let's talk about one of those like big demand drivers. So like every company has got to be an AI company now. Tell me the AMAT to AI story.

You're talking about from a standpoint of how we're implementing AI? Or are you talking about...

Well, both actually because you are, right? So we haven't even talked about the services and the services business yet. But -- so now, I guess, number one, just on the equipment side, talk a little bit about where you see AI driving equipment. One of the big questions I get is, obviously, Jensen's out there trying to make chips as big as a radical -- as many of them as he can make. And like what does it actually mean for leading edge? Because like right now, the -- it's not a lot of wafers, right?

It's not a lot of wafers today. So I think that when I talk to the CEOs from our customers, I actually ask them the same question, what do you guys think about it? They see it as incremental. The -- as you said, those GPUs, I talked about 15 billion transistors in a smartphone processor. You have almost 100 billion in a GPU. So those are big chips. They're really always driving the leading edge. So big chips are good for wafer fab equipment. And then also, all of that innovation, that's also good for wafer fab equipment. So I would say near term, there's some incremental demand. I think you have to really think out more beyond the next year on what is the impact. And certainly, again, I think -- I've talked to CEOs that think some of that technology will become more pervasive across many different kinds of devices. And it's still early, I think, for people trying to size how big those inflections. I do think it will be meaningful going forward. And certainly, that's going to be at the foundation of competition for many devices in many industries. So I don't know the exact size, but certainly positive -- super positive for us.

Got it. Maybe that's a good segue -- like, let's talk about how you're using AI yourself. And maybe it's a good segue into your services business broadly. So...

So services for us is growing this year. So if you look at our services business, about 85% is service and spares, 15% roughly is 200-millimeter equipment for ICAPS. And that 85% in service and spares, over 60% is long-term agreements. So that gives us resilience in really any kind of a market. And the length of those agreements are about 2.6 years. Renewal rates are above 90%. And you think about complexity in these tools. As that complexity increases, it's a great thing for us for our service business because the first thing you got to do is you've got to qualify those tools in R&D. So you have opportunities there. Then I qualify a tool and a chamber, I have to ramp all of those chambers into high-volume manufacturing. So people talk about a golden chamber making them all golden. So that is hundreds of thousands, maybe up to 1 million pairs of chambers every year that you're ramping. And again, that's where a lot of the sensors and metrology and also just -- so I have an epi tool, I want to ramp 20 into high-volume manufacturing or CVD chambers, I want to ramp a platform of 5 CVD chambers, all of those chambers into high-volume manufacturing. So all of those, when I run a wafer, I want them all to come out with high yield. So they all need to be matched. They all need to produce the same electrical performance regardless of which tool or which chamber they're running through. So that's where there's enormous opportunities for us, and we're developing new metrology capabilities. That's also great for our eBeam business. We have thousands of tools and chambers that are connected remotely. So when you think about all of that technology that we're building into our systems, there's enormous opportunities. So our service business, growing this year. And we've said that it's a little over $1.4 billion. And we said that longer term, we're on track to grow low double-digit rates. So that becomes meaningful. You're getting close to $6 billion run rate. So if you're growing in double digits, that's going to add some stable strong growth to the...

So that's a function of the installed base growth as well as, like, I guess, the services content per tool...

Services content, revenue per tool. And again, certainly, having those long-term agreements puts us in deeper relationships with our customers. So again, that's a strategy also. If you look back maybe 10 years ago, that number of agreements -- the percentage of agreements would be maybe 30%, something like that. So we've grown that percentage of agreements a significant amount. But that's been, again, another conscious strategy that we've been driving inside Applied.

Got it. And where does AI play into all of this? We've talked a little bit about like what you've done in the services business. I think in the past, you've also talked about things you've been doing on the metrology side and some new...

Oh, absolutely. So again, we're the leader in eBeam technology. So when you're developing these new processes, eBeam is a very crucial technology. So -- and when you're trying to optimize any of these tools, there's maybe 100 different knobs you're trying to optimize. So what we're able to do is come up with new imaging technologies, so we can look at residual germanium for gate-all-around or the width of a nanosheet or -- into a super high aspect ratio structure without cutting the wafer. And doing that, you can do it orders of magnitude faster. So that fingerprinting and then being able to look in that multiparameter space is where AI can come in to help you dial in, you want the biggest window because that gives you the highest yield possible. So we're using AI in a number of different areas within Applied, in R&D and also in our service business. An opportunity -- one of the things we're looking at is also driving productivity in the company. So we have an opportunity to double Applied going forward. So we're looking at...

Double what?

Double revenue, double profit, all of those things. So we're looking at revenue per headcount or dollar per task when we're doing that matching of chambers. What are we doing today? There's huge opportunity, Stacy, through technology to do those tasks much more effectively and much more productively. So again, those are things that we're driving really huge focus in Applied.

Got it. Sounds exciting.

It's fun, yes. It's fun.

I want to ask about something maybe less exciting, China. China is exciting, I guess. Talk about the impact of the export controls. I know you had some lost revenue. Maybe you revised that a little bit more recently. Most of the companies have been basically taking like, call it, a more careful read of the regulations over there. Just what has the impact of that been for you?

So we communicated that the range was in $1.5 billion to $2.5 billion. And then we've received more clarification on some of those regulations. So we're tracking more towards the low end of that range right now. There's about $1.5 billion to $2.5 billion, tracking towards the low end.

Okay. Got it. It was trying to get more and more decoupled. So clearly, like we've -- the U.S. has probably stymied their ability to really build a leading edge ecosystem in China. So they're going to have to double and triple down probably on trailing node on ICAPS. And right now, that's clearly been a positive for you, and you've got strength from China. There are some local Chinese companies that do build tools, and they are much more on the lagging edge because of the technology requirements. Is that something that we do need to think about over the long term as China has to double and redouble their efforts in this -- are those kinds of players -- the growth of a homegrown ecosystem at risk?

I think that, again, if you look at 1,700 steps in building these chips and the interaction between all of the optimization of those combined steps, you can make progress on noncritical types of applications. But more and more, the complexity is increasing. So the optimization across that whole flow, I think, is more important than ever. And I think for Applied, that's a very strong market. We have the strongest position across that entire flow. And working with leading edge customers, whether it's leading edge customers in the edge computing ICAPS or high performance, is really, really, really important. So always, what's happening is we're innovating. What's competitive today is not competitive 3 and 5 years from now across that whole space, not just critical also and noncritical. There's tremendous innovation that's happening, integration into these platforms that's happening. So being able to shoot the head of the duck is really important. We know where the duck is going. And again, we have such deep relationships with our customers. And you're co-innovating constantly. If you don't have access to those leading edge customers, it's almost like trying to shoot the head of the duck in the dark.

Does that help you in your ICAPS business, the work with the leading edge customers?

Oh, for sure.

So you pull down, like, innovations, I guess...

Oh, for sure. It's all co-optimization. We're innovating on structures. We're -- you know the way the industry works. I mean you're going forward at an incredibly high innovation pace. And constantly, the companies that win are the companies that collaborate the best and the fastest. That happens in ICAPS the same way. So I think that for us, in those -- that's an enormous advantage. I think it's even more important going forward.

Got it. Gary, we're down to about 8 minutes. We've got a few questions from the audience [indiscernible] lightning round.

Okay. Let's go.

All right. Some of these are a little near term, too. But you're seeing some order pushouts and cancellations for leading edge foundry/logic. Has the surge in generative AI caused those orders to reaccelerate or those order pushouts to decelerate?

What's the question again?

Sorry, let me ask again. You said you're seeing some order pushouts and cancellations for leading edge foundry/logic. But we've got a surge of demand for generative AI. Is it causing any of those orders to reaccelerate or the order pushouts to decelerate?

Yes. What I would say is the magnitude of those changes are really, either way, not significant changes, Stacy. And what I would say on the leading edge, again, the main thing is you have this race for leadership between these different companies. When you're building capacity -- I was with one of the largest customers -- CEOs recently. He said they're thinking about what's going to happen 18 months out. And certainly, if you look at 3-nanometer, that's going to be a big node, a lot of demand. Next year, you'll start seeing more meaningful spending on the gate-all-around.

Next year, it starts?

'24, yes. So you'll start seeing more of that spending coming in. I think for us, we look at the -- we haven't given specific numbers, but we believe that, that's going to remain healthy. That leading edge spending will remain healthy.

Okay. Got it. Are we hitting a limit at 4 atoms wide?

I think there's so much opportunity for innovation. It's amazing. It's really amazing. I think there's a clear path through the end of the decade. Again, we're working 4 technology nodes out. And yes, I think very, very, very clear path on how we're going to innovate going forward.

Okay. I've got a question I'm going to...

I don't know about 2 decades out. But at least we're pretty far out there and have high confidence.

I'd love to know, by the way, like all the stuff you're working on, how much of it actually finds its way to a product? And how much of it eventually gets scrapped because it didn't work...

Well, it's an interesting thing. So one of the things that we're focused on in the innovation center is working with universities. So this is another big opportunity. So a lot of times, they have old reactors or old tools. They don't have the whole innovation flow. So we're going to work on innovation networks with those universities on those verticals. And we worked with one university -- so getting back to your question, we had 10 innovations and working very, very closely with them. 4 of them in a 4-year period of time made it into chips. So -- yes, and that's all the way from kind of fundamental research and concept to -- which is -- the ultimate thing is does it get into a chip or not. So I think it's reasonable success rate. And then once you go past that initial concept, the success rate goes up a fair amount. But the time to innovation is also absolutely crucial.

Got it. Customers have increased the duration assumptions for your tools. Is this more of an opportunity for your services business than it is a headwind for your tools business?

So duration -- I'm sorry.

Is they're using them longer. So I think the question is, is it a tailwind because you got to service them, so that's good? Or is it a negative because they're using them longer so...

So what happens is that every customer -- they're constantly ramping new technology nodes. They're constantly ramping new technologies. So all of that complexity that you see is increasing, whether they're new tools or existing tools, and that creates opportunities. Again, all of that -- again, that's baked into our low double-digit growth for service. If anything, I think there's more opportunity for us. Again, some of the technologies that we're developing, I think, can enable us to create more value for customers and drive better outcomes at a faster pace.

Got it. So we had really strong memory spending last couple of years. We had $40 billion in WFE in '21 and '22, and we might do $20 billion this year, maybe. Presumably, you can't stay here forever. It's got to -- does it get back to $40 billion any time -- like, any time soon?

Well, first of all, I don't know if I would agree with $40 billion. But let's don't...

[ I'm used to ] Gartner. I know you guys -- I don't know if you guys use [ anyone ]. Like...

Anyway, I think it's a little bit less...

It was a lot, and now it's a lot less. And like, does it ever get back to a lot?

Yes. So I think the -- what I would say is longer term, we think 2/3 foundry/logic, 1/3 memory. And that's different than the way it was before. It was not weighted as much to foundry/logic. So we think that's the kind of normalized run rate. I think it's when -- I think the question, Stacy, is when does it recover and the magnitude of the recovery? I think those are fair questions. And at least talking to customers, I think they're more optimistic on DRAM recovering before NAND recovers. But I do think those are fair questions. When and what is the magnitude? Is it going to be a little bit or...

Do you have an answer?

I do, but I can't tell you what it is.

Got it. Update on competition you're seeing in the epitaxy and ALD markets?

Yes.

That's the question.

Okay. No, I think -- look, as I mentioned, in gate-all-around, we have high confidence. We have about half of that flow. So we can see that very, very, very clearly. We're on track for what we've discussed before, $1 billion incremental opportunity for Applied. Most of those critical steps are with Applied, most of the critical steps in deposition and, as I said, the majority of the steps in selective removal. I know a lot of people will talk about share gains, but we have high confidence, Stacy, in our position and continuing to grow going forward.

Got it. With expensive high-end chips leading sales and growth in semis, could this pressure or lower the CapEx percent of WFE in coming years for non-litho? I guess what that means is well, these chips cost so much, it drives a lot of growth, but you don't need as much CapEx to drive that revenue, I guess.

Well, I think on -- I believe -- and again, you can look at what customers are talking about publicly. I think the percentage of spending with materials-enabled technologies are going to increase. So if you look at the road maps, a lot of the innovations around the new materials, the new structures are in the sweet spot for Applied. So I think the semis, the growth rates are going to be solid going forward. Capital intensity is going to go up. So the equipment will grow a little bit faster. And I think Applied is in a real sweet spot relative to materials-enabled scaling.

So you sort of like previewed my last question, which is always you got a room full of folks here, we got a minute or 2 left. Why should they buy your stock? And I know -- you didn't say PPACt, like this whole...

Performance, power, area, cost and time, which is the speed of innovation. But...

Why should they buy your stock?

Yes. I think we're in a great position. I think semiconductors are the engine that will power the biggest transformation of our lifetimes. And the innovation is -- capital intensity is going to be high, and the materials-enabled scaling with Applied puts us in a super position. And the other thing -- so I think that part of it, whether it's in leading edge foundry/logic, ICAPS, DRAM, we said were very strong. Packaging is another big area that we have strength. So very strong. And I think our breadth and our ability to co-optimize is even going to be more important in the future. And then I'd say the other thing is our services business, as we've talked about, double-digit compound annual growth rate, high percentage of agreements, tremendous opportunities with technology innovations to drive that even faster.

Got it. Gary, I can keep this going for hours, but I think with that, we'll call it a day. So thank you so much.

Thank you so much.