BE Semiconductor Industries N.V. (BESI) Earnings Call Transcript & Summary

Okay. Good day, everybody, and thank you for joining this webinar. For those who don't know me, my name is Nigel van Putten. I cover Besi here at MS. I don't think that Mr. Richard Blickman needs much of an introduction as he is the President and CEO of Besi and has been in that position since May 1995. Welcome, Mr. Blickman.

Good afternoon, everyone.

In addition...

Or good morning.

Yes, evening as well. In addition, we have Senior Vice President of Finance, Leon Verweijen; and the VP of Corporate Development, Mr. Edmond Franco, on the panel as well. So thank you all for joining. And the format today is that of a fireside chat. I have some questions prepared, but you in the audience can submit questions as well through the Q&A box at the bottom of your screen. I will receive those, condense those a little bit, so we're able to cover as much ground as possible. Please do take into account that we have quite a large audience today, so I'd like the questions to be quite specific but not too detail at the same time. All right. Again, lots of ground to cover. We have about an hour. So I'd like to structure the call in 3 parts: maybe to start off with a review of the year, including some of the important milestones, strategies and progress; then into current trading, the order book, the outlook, et cetera; and I want to leave plenty of time to talk about technology and maybe even some stuff around hybrid bonding as well. So with that, I'll turn over to you -- turn the call over to you, Richard. Perhaps you can talk -- walk us through 2022 a little bit.

Well, thank you very much, Nigel, for the introduction. I'm happy to focus on some highlights of the 2022 calendar year. If you look back, it's been a very important year for the development of our products and in particular, the breakthrough of hybrid bonding tools. We shipped in total 35 machines, of which about 25 are used in production, which is basically from scratch in 2021. So a major breakthrough and very substantial volumes and used in chiplets, which have been marketed as of September 2021. And from that date, both volumes and specific designs have increased steadily. And that has led to more orders. Q4 was a very important quarter with 11 additional systems and then 3 more in the early part of Q1, which then totals 14 machines. What also happened in the hybrid development is, as we have explained in the Capital Markets Day but also in our quarterly reviews, the objective to develop an automated line which supports the -- simply the quality of hybrid bonding process by eliminating more particle generation because that line is fully enclosed, and that line consists of cleaning tools supplied by AMAT plus the line itself and basically connects the bonders to that line. And very important, the first line has been set up at a customer in Taiwan, the major customer for hybrid bonding so far. And that should become the workhorse for the industry as this technology further is adopted in an ever broader sense. At the same time, integrated lines have been ordered in the U.S. and several will be installed this year. So '23 starts to look like a continuation of '22, setting up the basic capabilities for hybrid bonding adoption, both in Taiwan, which started, let's say, 1.5 years ago, and further developing of this capability onshore in the U.S. So summarizing, one can say that '22, again, has been a major breakthrough year for this technology. It is used for chip design architecture below 10 nanometers. Today, supposedly, it is used in 5- and 7-nanometer design environment. And in the future, it's intended certainly and it has to because it's the only way you can connect devices once the industry moves to node sciences below 5 nanometers to 3 and potential also 2 nanometers. And that is then made feasible by further designing this Generation [ 1, 2 ], which allows 150 nanometers to go down to 100 nanometers. And for that purpose, we will ship the first tool capable of 100 nanometers around April time frame. So then in '23, that will be further developed and using for these ever smaller node sizes. So then you would ask the rest of '22. Well, if you look at Besi's revenue, it's reduced by, you could say, only 5% versus '21 in a year where the downturn really took shape since Q2. The overall demand for equipment reduced gradually Q3, Q4. So with a major downturn now as it looks for, let's say, the conventional business, also lower demand for high-end smartphone applications. The only bright spots were this hybrid bonding development and also a continuation for applications in the automotive arena. So '22 marks a year where we have benefited from early introduction, adoption or recognition of the need for a new process called hybrid bonding, and that has helped us so far in this down cycle. Our margins improved versus '21 simply because of very tight cost control. We also reduced the flexible head count in operations simply because we built less machines in the second half. And we were also supported by a strong U.S. dollar in that sense. So all in all, the year '22 was for Besi very successful with, you could say, only a small revenue decline. Margins improved. And what also is very important to note, in Q4, our orders increased by 40% -- 40-plus percent versus Q3, which also demonstrates that our equipment is at the forefront of technology. And once there are applications, which will become, let's say, used in mainstream applications, we see that translated into increased orders. We also mentioned that these orders are typical for forefront technology. So they will use some more time to be manufactured. So the effect of that increased orders we will see in Q2 mainly and also partly in Q3. We guided the first quarter down between 0% and 10% versus Q4. Remember, we're in the midst of a downturn, and margins to stay roughly at the same level as in Q3, so between 61% and 63%. And our operating expenses will be similar to slightly below, apart from the onetime cost related to our incentive program, which we mainly take in Q1. So that is in a summary what has been 2022 and also a slight outlook for this first quarter of '23.

Yes. Thank you for that great introduction. I think there's actually one topic you haven't addressed but I know is important for both you and most investors on the line here. That's capital allocation. So even though top line was down, R&D up 48% of the year, EUR 416.3 million returned to shareholders. And actually, my question would be, it seems you're on track to do even more in terms of both growing R&D but also returning cash to shareholders this year, talking about the share buyback and the already announced dividend proposal.

Yes. Well, we have a very simple, let's say, rule in the company developed. We maintained 20% of our revenue in net cash simply to be prepared to invest in R&D in particular, but also if faced with real downturns, we are in the position to healthy develop through these down cycles. So anything above 20%, we simply distribute to shareholders either with dividends or share buybacks. So last year marked an important year in terms of share buybacks. We increased our program significantly to EUR 300 million, which then, as you said, totaled both the dividend of EUR 3.33 per share over the year '21 paid out in May '22, and in addition to that, the share buyback. So in total, EUR 416 million. And then still, if you look at our net cash, things have not changed in that sense. So this year, again, we will propose at the AGM a slightly lower dividend of EUR 2.85 and at the same time, a continuation of our share buyback program, which is until basically end of September, October this year. And if things continuously do, we will certainly extend that program further. So simply based on the rule of thumb, 20% net cash and 20% of revenue, which is what we maintain as a baseline.

Excellent. In the meantime, trying to consolidate some of the questions that are coming in. But first looking into the March quarter, you already alluded to the guidance. I phrase it a little bit different. Like you also said FTE, especially in Asia, temp were down quite significantly. Into the current quarter, do you see that number going down further? Or is it here -- stabilizing here or maybe increasing already a little bit in sort of preparation of what's to come?

Well, the latter. So if we look at the order intake, what I mentioned in Q4, an order intake of around EUR 180 million. That brings us again to a higher revenue levels, remember, EUR 137 million in Q4. So that means you need more hands to assemble those machines. And also, there are machines, let's say, in a higher category. But the key question is, of course, all of us don't know. Are we really at the bottom or are we reaching a bottom maybe in the next quarter? That's a big question for everyone. But again, Besi's model is very flexible. We have this flexible workforce, which works excellent, I can say. So whatever we need to do, we can organize very rapidly.

Excellent. Then maybe on to the backlog of orders, quite high, also seeing a bit of a change in the seasonality of ordering already in the third quarter, more in the fourth. Is that continuing into the first and maybe second or -- because if I'm correct, the shipment is still due in the second quarter. So most of the smartphone seasonality is still there. However, the orders were placed earlier. So how does that look sort of where we stand today, end of February? Is that continuing still?

Well, first of all, we must bear in mind, we've had nearly 3 years of major supply chain issues partly due to COVID, partly due to an enormous increased demand for semiconductor equipment across the board. And that's also why certain times backlogs have built differently than from pre-COVID times. Pre-COVID, we had run rates per quarter typically for all of our machines. So you could very much see in the order intake in the previous quarter how the next quarter revenue would develop. But because of COVID, supply chains have lengthened for -- and still for certain components that there are longer lead times than we had in the past. But also the systems, the -- simply the design of the systems has changed. Hybrid bonders, they take about 6 to 9 months, depends a bit on the configuration. But there are also other equipments which have been developed for next-generation application. So we look at a slightly different pattern of lead times, so internal manufacturing lead times. And that reflects then, of course, in when the backlog will be fully depleted and the new orders coming in. So I think it's fair to say that we are gradually going back to a more normalized pattern before COVID, but the equipment we now supply, for a large extent, has a longer manufacturing lead time. How order development, and that's what I said earlier as well, don't forget, we are in a downturn. There are some thoughts on -- that we may well be in a bottom. There are some others who think it may be a bit longer. Let me give an example about the Chinese subcontractors. What is, of course, in a way, disappointing is that with China opening up in December, there was a clear thought that the industry would pick up again shortly thereafter. And then Chinese New Year is a usual, let's say, a moment when orders are picked up for the rest of the year. This time, they're disappointed. Some people think it may take a bit longer. So orders may well come in, in the second quarter for, yes, an improvement in the second half of this year. We commented on the call to some questions about the utilization rate at Chinese subcons, and we shared that our own system of counting that utilization still tells us that we are not at a point where there's imminent shortage of manufacturing equipment. It has improved. But -- so my answer summarized is simply that we're still in, as we said also in the press release, in a downturn. And the question is will -- are we hitting bottom or will it take a bit longer?

Understood. Got it. I think it's time to switch gears into some technology. And yes, we'll get to hybrid bonding very soon. We have quite some questions coming in already. I do want to touch upon TCB first. I think there was form factor saying they see quite some movement in what they said, x86 client compute chiplets, CPU. We all know what we're talking about here. But then there's also perhaps a foundry that may or may not adopt TCB as well. Just curious to see -- I mean, as you said in the call, you're shipping the first chip to wafer to -- in April. What is the outlook there? How many applications are you targeting? And what do you think will be sort of the ramp into the current year?

Well, first of all, you have hybrid bonding, you have thermal compression bonding, TCB, and you have flip too. In our deck, there's this beautiful slide which explains a certain design where different technologies are used, so for many product road maps. And you can see for certain critical, especially microcontrollers. And once you start to stack memory on top of that, that hybrid bonding is the technology that has to be used because of the shorter distances. And then for the broader, let's say, total design, you may need to use TCB. And then, ultimately, the entire package, you can do with flip chip. And in -- yes, let's say, many of the designs of the major road maps, you are starting to see that combination. And with the shipment of the first TCB [ 2 ], we will be able to supply both to the industry and, of course, the flip chip for the entire package already today. And that total solution is used by a major U.S. customer. And also in the foundry world, it's recognized that, that completes the offering to those customers who are using the -- and those are the fabless companies. So with the shipment of the first tool, after qualification of about 6 months, you can expect in the second half of this year orders for this first application where this is designed for. And that opens up a major expansion program in '24, '25 and onwards. So it's a simple completion of a product offering. We started with TCB back in 2012, where we shipped numerous machines for stacking, at that time, DRAMs in memory packs. And that technology is still used but in limited volumes. We then switched to hybrid bonding full speed because that was a direct requirement from a very major customer and now several. And then on top of that, we have this next-generation TCB. So simply following the roadmap in the high-end CPU and memory applications.

Excellent. Very, very interesting. We have ways to go from here, but I do suggest we start covering hybrid bonding. Maybe my introduction would be, it seems that in the past, packaging was quite specific to the chip. I think there's over a dozen type of chip packaging, and the specifics could be triple digit even if you took the configurations, et cetera. But then hybrid bonding seems to be a very common process across multiple applications, more like a [indiscernible] process. So my question would be what do these applications share, CPU, GPU, ASICs, DRAM even, smartphone processors? Why are they all looking to hybrid bonding?

Well, the easiest way to make this, let's say, clear is that if you look at this industry over the past 60 years, it has been an industry with individual components, whether that's processor or whether that's memory or whether it's power devices. And the industry design circuits combining these components and mostly on a printed circuit board. And this chiplet architecture allows that on chip level, you connect multiple devices. So you eliminate the interposer and simply, you go from single devices to integrated devices. And that is what the industry already several years ago has presented as the way forward. So it's not brand new, but it is certainly a next step in integration on chip level. And because the simple fact that below 10 nanometers, the interconnects using flip chip or TCB on individual devices is not feasible below a certain distance between IOs, the fact that you then have to move to chiplets using hybrid bonding, is driving this technology into mainstream application. So this writing on the wall has become, I would say, 7 years ago, yes, by major forces and one of them is, of course, in Taiwan, TSMC, who are still the world to better prepare that this is, in the end, the only way to create these circuits if you move down to node sizes below 10 nanometers. And that is what has driven our R&D programs and simply focused on those customers to be ready in time and to enable this technology. And one of the important things with that is, of course, the partnership with Applied Materials. AMAT, one of the largest ones in this industry, is very close to the design tables of these key customers. And they have recognized which we then call the advanced packaging, that has become the major end goal as opposed to just supplying equipment to produce these devices. So that has brought us together about 3.5 years ago now. And that simply confirms how the industry is changing from an individual building block or device environment to an integrated directly connected, which has to be done in front-end environment because of particles and cleanliness of those processes. So that's in a nutshell where we are in the very high end at this moment. So up till about a year ago, there was definitely other voices in the industry who would say, well, this is very nice, but it's very expensive because that has to be a front end. These machines have to be very accurate. So because of that accuracy also, the ASPs go up. So if we can stretch it further using the older technologies, larger die sizes, but that immediately has an impact on the speed of this circuit. And so the chiplet architecture which is enabled by using this hybrid technology simply proves to be able in the end to even to stretch Moore's Law in the same node size, expanding the size of the CPUs by simply connecting CPUs designed in multiple devices. So again -- and that's where we started. '22 has been a major year in bringing this technology from the lab into mainstream applications. And that will simply continue.

Yes, definitely with more and more chips being manufactured using hybrid bonding. Question here but also something that I picked up on the call was there seems to be an acceleration in the number of clients that are interested in the technology at the moment. What customers do you see adopting this technology next? I think AMD has been quite public that they pretty much see packaging and hybrid bonding as key in their road map going forward in pretty much all their chips. They recently announced a new MI300 using -- a different type of using hybrid bonding but seems quite significant. But besides AMD, maybe more in terms of the domains that we should be thinking where to next?

Well, easy, of course, the first adopters are the top 5. So if we look at the forefront of this technology, and it's very well-known who are the top 5. We, of course, have Taiwan; we have in the U.S., 2 major ones; and we have Korea. And then what is very important is there's, of course, these high-end fabless companies, making CPUs for -- especially for the mobile space. And then also as a clear proof of adoption is that now also the subcontractors in the assembly space are starting to prepare for setting up capabilities using hybrid bonding technology. So in a year's time, that has changed significantly from, you could say, only 2 who are in the development phase and one had in production for mainstream applications that now is further expanding in a more broader sense. But as we mentioned also in the call day before yesterday, it proves, as just said, capabilities to be set up. But in the end, you need end applications. So you need the road maps of the, let's say, Intel's, AMD's, you referred to, but also many others. And you're in the sort of a chicken and an egg situation. But then what's very encouraging is, of course, setting up these capabilities is very expensive, capital-intensive. So yes, the positive message is that that's setting up these capabilities is supporting demand ultimately by customers which have made those commitments. So that's in a broader sense an answer to your statements.

Yes. A couple of follow-ups on that. Starting with the big U.S. logic IDM, I think as you call them, deciding on 3 major families. Could you give us a little bit color what we should think about?

Well, they are families very well known. And they have, as we mentioned, what we've been told, decided on 3 major applications. We are not allowed to share specifics, which we never do of any of our customers. But in a broader sense, one can guess which ones they are. But the important for us is, is this mainstream, what equipment do they need and then ultimately, how many machines do they need. And if we look at it from that angle, we simply ship machines, we build machines for applications, yes? It still fits into our model where we have set up capabilities to build 15 -- depends which ones, so between 12 and 18 machines per month. But that translates into about 200 machines per year. That should simply fulfill the capacity in the next 3 years or something must dramatically accelerate and then we are able in the 6 to 9 months to expand our production capability significantly. We have set up in Malaysia the high-volume assembly of these machines in a cleanroom 6 environment, and we've also prepared the rest of that facility that we can expand when needed. And we've added a new facility close by, which then enables for the conventional business also to once the tide turns and the cycle picks up again that we are able to build significantly more machines than we did in the last up cycle. So anyway, and customers come regularly to evaluate our capabilities, test our supply chains because they have to be sure that once they accept an order from their customers that Besi can deliver. So that's in a nutshell where we are today.

Got it. And maybe very specific, has this have to do maybe with the generation of tools? Are they ordering the second or you've got a 1-plus, kind of 2.0 type tools? Is that also why maybe there's an acceleration now with these customers deciding, okay, we're going to use this tool for this process as we see this capability? Or is it still actually on the tools you've already shipped to the market?

It's still Generation 1. As we mentioned in I think also in October after the third quarter but certainly at the Capital Markets Day, this first generation is designed and capable to place accurately down to 120 nanometers. It's used today around 200 nanometers. And so you have some way to further make these machines more tight. The 100 nanometers, as explained, will be shipped in early April, and it will take at least the rest of this year or maybe somewhat earlier, but to really bring this machine to full acceptance. So it will be used as of 2024. And probably that is for the note Generation 3 nanometers but above that. So 7, 5, that is the sweet spot today where the major volume is expected and also the first chiplet families can be done with this Generation 1. Generation 1 plus or, let's say, the 100 nanometers is more expected '24 onwards, and then for those applications which are based on the 3-nanometer design world.

Yes. Very clear. April first shipment, to be delivered next year. That seems too -- well, let's just say it's a little bit earlier than expected. But then again, the question would be, how is your competitive position today? Because it seems that if you are already moving to a Gen 2, well, I don't think there's a competitive product in the market yet. It's difficult to say your lead is expanding because it was already -- yes, there wasn't a competitor. But it does seem that you are moving ahead quicker than competition is even getting towards the starting line. Is that a good way to look at it?

Yes. You could -- well, let me answer it differently.

It was a little bit leading indeed. So...

Well, a year ago, we were the only one who offered really production-capable systems. And that is -- today is still the case. But what happened in the meantime is several high-end die attach platforms are, let's say, offered to the industry to also be developed for below 10-nanometer design. So accuracy is around 200 nanometers, and the industry ultimately wants to have a choice. So today, we offer these machines and we have the largest, by far, installed base. But ideally, the industry wants to have a choice. How far competitors are lagging, that's so -- it's difficult to tell. Our concept is very well proven. It is also proven to be ready for the next generation. So with that baseline, we expect that we can further extend our lead. That's at least our objective. But one also has to note, this is not easy. This is very difficult technology, not only accuracy, but also this whole particle -- 0 particles. Otherwise, you cannot create this connection. That means these machines are pretty complicated, and we have great help from the partnership with AMAT in how to build those machines for those front-end environments.

And maybe -- because if you say there's competition down to 200 nanometers, are those considered hybrid bonders or maybe very accurate, call it, fluxless TCB tools that can to a large extent do similar maybe in a different way?

No. There's -- as we mentioned several times, there's a major Japanese competitor who was also invited 7 years ago in Taiwan to develop this technology. But their concept is simply not able because we have a dual-bonded design. And so the output of that machine is significantly lower. So it's more from that angle. TCB is a different process. It's also different, the accuracies. So the system we will ship has an accuracy of 1 micron, and that's the most advanced in the TCB applications. So that's not yet nanometers. So that's a different situation.

That's good to -- but then to fully understand what you're saying, there is competition in hybrid bonding. However, based on especially throughput but also accuracy, you don't feel threatened. Is that a good way to summarize it?

Well, you're always aware of only the paranoid survives. So ASM has announced that they are working on a beautiful concept, and they are offering that to customers to codevelop. Maybe something comes out of that or -- yes. So never expect that nothing will change you.

No, of course. But I think the -- my view was that as you move to Generation 2, competition is actually still trying to even ship an initial product, except maybe the Japanese competitor who has its own difficulty. So I guess that would still be an okay way to summarize that. That's helpful. Plenty of questions on AI. Most language models seem to be quite memory-intensive. I was looking at some of the presentations or articles, et cetera, and there does seem to be 2 key areas, which is both there can never be enough SRAM, it seems, for CPUs and GPUs can definitely benefit from having a lot of near -- DRAM near, so call it HBMs, et cetera. Now I know this is quite recent, the hype, but have you already seen some acceleration? I think the Koreans memory makers have come up quite a bit during the call on today. Do you see that movement there accelerating for memory as well?

Yes. Well, also a long road map which has been presented to us in the last 3 years, I would say. And that's very easy to understand. So since about a year, we're assembling those memory devices, either stacked or on top of CPUs, in the center of excellence in Singapore for the major U.S. memory manufacturer. And also that is now being extended in programs for the major Korean memory manufacturer. So along a very, yes, let's say, clear roadmap that is happening. And -- but I should add to that, that's not simply just adding memory. Memory are much thinner than CPUs, yes? So there's another complexity in really making this happen. So it will take some time before those next steps are really mainstream. But everyone knows that, and we're fully engaged in making that happen.

Got it. And then talking about roadmaps. I think a lot of investors are curious to hear more about smartphones application processors and the possible adoption of hybrid bonding there. Now the major smartphone clients seems to already move to M3 without the use of hybrid bonding. So why would they in a future process include hybrid bonding, considering that maybe M2 seems a couple of years away? So you've hinted that maybe '24, '25 as a logical envelope for introduction. What would drive that? Why would that -- has that not yet happened? And why is that increasingly likely next year and the year thereafter?

The simplest answer is yield. With hybrid bonding and a chiplet design capability, you can achieve a higher yield and for that reason, a lower cost. And that is why this technology is being tested, being developed this moment. And hybrid will find its way into 2024. I think it's a bit early. Because time flies, we're now already close to March '23, but certainly, thereafter. So that is what is driving the adoption of this technology.

Got it. That's yield on the front-end process. Of course, we've talked also about yield in the hybrid bonding process, up from 20% at introduction to 80% by October and moving up. When should we expect full maturation of the process...

In the course of this year, we're already in the 90s, but there's sometimes -- one of the issues, to have a specific example, is the flatness of these chips. And we cannot fully control that. That is how the material is delivered, yes? So yes, that's one of the issues. The machine itself is more and more, production 100% yield-capable. But there are also other issues, simply particles. We mentioned particles. How can you control that you have several particles on wafers, on specific dies going into this process. And that will take more time. And one of the answers to that is this integrated line consent. So are we -- how fast are we able to bring that line into full production? And that will take the remainder of this year to make that happen. And there's significant commitment in setting up this first production line as we speak. And the same will happen in the U.S. with 2 lines in the course of this year. So those are the major issues facing further mainstream adoption of this technology.

Got it. I'm not muted. Excellent. A couple of specific questions in terms of -- you've mentioned inflation this year, I think also into next year and the impact on margins. Now I think you used to comment that hybrid bonders would have an above corporate average margin, which progress, et cetera. How is that looking today? Is that still the case?

Well, the best way to look at it is the margins in 2022. Imagine, we shipped all of the other 35 machines, of which 25 are used in production. They passed all tests. The customer paid -- pays the bills, yes? And we have significant support, I wanted to say armies, certainly in the first half of last year and then less because the machines become more reliable. But then margins of well over 60% is showing that we both have some handle on cost. We have that on inflation. And that's the way how to do it. What could impact the margin negatively is, of course, a rapid decline of the U.S. dollar, a more rapid inflation. It doesn't look like now, but we immediately act in increased costs.

Can you maybe provide a little bit more sort of context around the dollar impact on the gross margin? Let's say, 1.06 today, 1.15 perhaps. Is that then sort of a 100, 200 bps impact there? How should we think about that?

No, no, no. It -- there used to be rules of thumb. Today, it's different because our manufacturing costs are very global. And we've seen that in -- yes, throughout '22. You could say that 10% is something like 2.5%, 3% impact both ways. But that was the situation last year. It's difficult to forecast with all the different elements. We also have Swiss franc. We also have euro cost. We have Chinese renminbi cost. So if you look at the total margin or the total cost structure of Besi, that -- you have to look at it at that moment in time. But I would say 10%, 2.5% is a very good rule today.

Got it. So we still have some time. I just want to go back maybe on the adoption roadmaps of customers. You've mentioned the big 5, I think the IDMs plus 1 fabless. On the sort of more specialized producers maybe for hyperscalers, their internal programs, et cetera, are you also engaging with those type of customers for, I guess, the ASIC products also tying into what seems to be the AI had that's to come?

Well, it's mostly the institute you have in Europe, in Belgium, you have iMac. You have [indiscernible] in Germany. You have in the U.S. institutes. You have in Taiwan an institute, E3. They're all setting up those capabilities and which is simply supporting the -- along the whole supply chain the, let's say, basics, starting at universities, yes, supported also by the TPx programs in the U.S., in Europe. So expecting that this technology becomes the basic technology for the next 2 decades. And then you have the early adopters, and they use the front-end infrastructure. And today, this is Taiwan. And IDM capabilities setting up a dose, which we referred to earlier in this call.

Yes. And you also referred earlier that capacity looks pretty good. 200 tools that could give you revenue above -- depending on the generation, EUR 400 million or EUR 500 million. That seems to be tracking a little bit ahead of maybe the cumulative numbers that you provided at the Capital Markets Day. Is that a fair view? I know it's sort of...

Well, we said the same at the Capital Markets Day. But then we still were in very early phase. It becomes more 6 months, 7 months down the road and further adoption. But as I said, we can increase that number pretty much within a 6- to 9-month time frame. But the key is to understand is '23. As we said, '22-'23 are the adoption years and the first rounds in end applications. '24, '25, '26 will be the further rollouts. AI models predicted by the TSMC, but also by -- in the U.S. by Intel, which have big numbers. They build those fabs for advanced packaging. And they do that with underlying roadmaps because it's very capital-intensive. And that all fits into that model, which we explained.

Right. So the model doesn't change, but -- the numbers maybe don't change. But I do have a sense that we're -- like you said, I don't know how many months down the line. You're certainly not less enthusiastic of making maybe the upper end of those numbers.

Well, in '25, '26, certainly. Something really must happen, either the technology that we're overexcited that this all works, yes? But the proof of concept, I think, is already made. But you have to be careful with these new technologies, and we keep saying that what's different this time is that the adoption is simply following the road maps. And usually, there's a delay because of whatever reasons. Remember EUV for a long time. And that has not happened so far.

Got it. And help is coming from the integrated tools that could certainly further help [ contamination ] effects, et cetera. So that's something to watch. Okay. Maybe the final question then, how should we think about 2023 breakthrough year in terms of orders. It seems a bit lumpy 11 fourth quarter, maybe 3 so far this quarter, IDM potentially coming into the second half. Would you argue that there might be an expectation of, not orders yet, but an expectation of this accelerating into the second half of the year, both there but also elsewhere?

Well, we should see an acceleration or we should see some programs being ordered in the next quarters to fit those road maps because there's a manufacturing lead time of 6 to 9 months. And then there's the insulation, the start-up, the training, all the support, yes? So that is what we are anticipating, and we are ready for that in our supply chains. But we will report as that comes.

Excellent. I look forward to that. I think with that, we're even slightly overrunning the hour. So I want to thank you, Richard and the rest of the Besi team for joining us today. Same goes to the audience. I hope it was insightful. And thank you very much.