Good morning, good afternoon, ladies and gentlemen, and welcome to Besi's quarterly conference call and audio webcast to discuss the company's 2024 fourth quarter and full year results. You can register for the conference call or log in to the audio webcast via Besi's website at www.besi.com. Joining us today are Mr. Richard Blickman, Chief Executive Officer, and Mrs. Andrea Kopp, Senior Vice President of Finance. Currently, all participants are in a listen-only mode. Later, we will conduct a question-and-answer session, and instructions will follow at that time. As a reminder, ladies and gentlemen, this conference is being recorded and cannot be reproduced in a whole or in part without written permission from the company. I'd like to remind everyone that today's call management will be making forward-looking statements. All statements, other than statements of historical facts, may be forward-looking statements.
Forward-looking statements reflect Besi's current views and assumptions regarding future events, many of which are by nature inherently uncertain and beyond Besi's control. Actual results may differ materially from those in the forward-looking statements due to various risks and uncertainties, including but not limited to factors that are discussed in the company's most recent periodic and current reports, filed with the AFM. Such forward-looking statements, including guidance provided during today's call, speak only as of this date, and Besi does not intend to update them in light of the new information or future developments, nor does Besi undertake any obligation to update the forward-looking statements. I would now like to turn the call over to Mr. Richard Blickman.
Thank you. For today's call, we'd like to review the key highlights for our fourth quarter, and year ended December 31st, 2024, and update you on the market, our strategy, and the outlook. First, some overall thoughts on our performance. Besi's fourth quarter results were adversely affected by ongoing weakness in mainstream assembly markets, seasonal influences, and lower demand for hybrid bonding and photonics applications as customers digested capacity added in 2024. Revenue of EUR 153.4 million was down 2% versus Q3 last year and 3.9% versus Q4 2023, primarily due to lower demand for automotive applications, partially offset by increased hybrid bonding shipments. Orders of EUR 121.9 million decreased by 19.7% versus the third quarter last year and 26.7% versus the fourth quarter in 2023, due to lower bookings for hybrid bonding, photonics, and mainstream assembly applications.
Hybrid bonding and photonics orders have fluctuated on a quarterly basis due to the timing by customers of new device introductions and related capacity additions for these emerging applications. Our operating income in the fourth quarter 2024 decreased by 8.2% versus the third quarter last year, primarily due to lower revenue and a 0.7% gross margin decrease from adverse forex movements. Q4 2024 net income of EUR 59.3 million increased 26.7% versus the third quarter last year and 8% versus fourth quarter 2023, due to net tax benefits realized from an upward revaluation of deferred tax assets.
Besi's business development in 2024 reflected contrasting growth trends for AI and mainstream assembly equipment markets. For the year, revenue grew by approximately 5% to reach EUR 607.5 million due to significantly higher demand by computing end-user markets, particularly for AI-related hybrid bonding and photonics applications. Similarly, orders of EUR 586.7 million increased by 7%.
As a result, orders for AI applications grew to represent approximately 50% of our total orders in 2024. We continue to navigate an extended downturn at industry-leading levels of profitability. Besi achieved gross operating and net margins of 65.2%, 32.2%, and 30%, respectively, in 2024. Gross margins increased slightly versus 2023 due to a more favorable advanced packaging product mix, which were partially offset by unfavorable net forex effects, particularly in the second half of the year. Net income rose 2.8% versus 2023, primarily due to higher revenue and gross margins realized and a net tax benefit of EUR 18.2 million. Such favorable influences were partially offset by a significant increase in R&D spending and higher share-based compensation expense. Besi's revenue from computing end markets has grown significantly over the past three years due to growth of hybrid bonding, photonics, and other AI-related 2.5D, 3D assembly applications.
In combination with an extended mainstream downturn, revenue from computing end markets grew to 43% in 2024 versus 24% in 2023. Shareholders were rewarded for their investment in Besi as we distributed a total of EUR 251.3 million in 2024 that represented approximately 41% of total revenue. During the year, we paid a cash dividend of EUR 171.5 million and initiated a new EUR 100 million share repurchase program in September 2024. In total, we repurchased EUR 79.8 million of our shares in 2024, which increased our total shares held in treasury to 1.8 million shares, or 2.3% of shares outstanding. Besi ended the year with a solid liquidity base consisting of cash, cash equivalents, and deposits, aggregating to EUR 672.3 million and a net cash of EUR 143.8 million, increases of 62.6% and 27.3%, respectively, versus year-end 2023.
Our net cash position benefited from the conversion into equity of EUR 129.1 million of Besi's 2017 and 2020 convertible notes during the year. In July, we strengthened our liquidity position further via the offering of EUR 350 million of 4.5% senior notes. Next, I'd like to speak a little bit about the current market environment and our strategy. The timing and trajectory of a mainstream assembly market upturn is difficult to predict at present. Leading analysts, such as TechInsights, expect a strong rebound in 2025 and 2026 of 21% and 31%, respectively. However, the assembly market still suffers from post-pandemic excess capacity, which has taken more than two years to approach equilibrium levels. Semiconductor unit growth and capacity utilization rates have improved since 2022, but at a less rapid rate than previously anticipated by analysts.
That being said, we believe it likely that a mainstream assembly recovery will begin in the second half of 2025. Its trajectory will depend on demand trends in each of our end markets and the ultimate course of global trade restrictions. Investments in Besi's future growth continued in 2024, as reflected in higher development spending and a planned expansion of our advanced packaging production capacity in 2025. We increased R&D spending by 31.7% this year to offer customers a complete portfolio of leading-edge assembly solutions for next-generation 2.5D and 3D architectures. In addition, progress continued on our hybrid bonding agenda, as revenue approximately tripled versus 2023 and orders more than doubled. In addition, adoption increased from nine to 15 customers during Q4 2024. Some notable hybrid bonding bookings included the first order from a Japanese semiconductor producer and from a Korean IDM for advanced logic applications.
Now, a few words about our guidance. We entered the year 2025 with cautious optimism based on strong momentum in our advanced die placement solution for AI applications, partially offset by ongoing weakness in mainstream automotive, smartphone, industrial, and Chinese end-user markets. We believe that the commercial viability of our hybrid bonding process technology has now been confirmed by some of the industry's leading players and research institutes. Significant incremental adoption is anticipated to occur over the next three years as the technology is increasingly used in HBM 4 and 5 memory stacks, ASIC logic devices, silicon photonics, co-packaged optics, and consumer mobile computing applications. For Q1 2025, we forecast that revenue will decrease between 0% and 10% versus the fourth quarter last year and for gross margins to remain in the range between 63% and 65%.
Aggregate operating expenses are forecasted to rise 10%-20% versus Q4 levels, primarily due to higher strategic consulting costs. This concludes my prepared remarks. Before we begin Q&A, I would like to remind everyone to limit your questions to two at a time so all participants have an opportunity to ask questions. Operator?
Thank you, sir. Ladies and gentlemen, if you wish to ask a question, please signal by pressing star one on your telephone keypad. If you wish to cancel your request, please press star two. And please make sure the function is switched off to allow your signal to reach our equipment. Again, it is star one to ask a question. And our first question is from Alexander Duval from Goldman Sachs. Please go ahead. The line is open.
Yes, thank you very much for the question. Richard, you mentioned that one of the downward impacts on orders in the quarter was related to hybrid bonding, and obviously, 2024 guidance on revenues was a bit below consensus. Could you help us understand to what extent was that a result of a change in the backdrop on hybrid bonding at a market or customer level, or was it more to do with normal quarterly lumpiness within hybrid bonding? And secondly, as a follow-up, it sounds like you're calling for a rebound in the second half on the mainstream part of your business, which some call the legacy part of your business. We have seen a number of analog semis players talking about improvements in future quarters. So what's your level of confidence or visibility on the mainstream part of your business outside of hybrid bonding improving towards the second half?
Many thanks.
Excellent. Thanks, Alex. Hybrid bonding orders, your qualification of a lumpiness nature is typically how this occurs. So if you go back over the past three years and you look at the orders for hybrid bonders we received, that also was characterized by orders in certain quarters and then delivery and then again orders. So the question can be also formulated a bit more. What do we expect going forward? This typical nature is expected to continue. So key is to look at the adoption. We mentioned in the year 2024, we increased the total customers using hybrid bonding in development or in early applications from nine to 15, so we added six. And last quarter, very critical, also the entry into Korea and also Japan. That will take some time before that is in real production.
That's more in development, early development, but still a major benchmark for us. That's how it will continue, but so far, it's following what we shared many times, the trajectory of adoption over time, so first, logic, and you can say in 2024 that has definitely made solid progress. Key is this year to understand, will the adoption occur in stacking memories? Will there be certain HBM4 versions which are stacked? Well, we'll see whether that happens in the quarters to come, and then the chiplet architecture used for many other applications. We also see that in progress in the development centers, both in the Center of Excellence in Singapore, where we develop these processes on an ongoing basis with Applied Materials, also in our own labs.
So there's a lot which has changed to the positive in the year 2024, albeit that orders in Q4 were simply somewhat lower because of the typical nature. Your second question, will the tide turn in 2025, is, of course, a very, very important question on everyone's mind. So far, the customers in the mainstream have reported numbers in, let's say, a careful way. Also, some customers have announced that they reduce headcount, mostly related to automotive, is our view. On the other hand, we've seen positive signs of recovery in certain areas, which all have to do with AI and high-end computing. So maybe the TechInsights model is correct. We will also find that out in the next month. But if you look at the length of this downturn, which we also explained earlier, it is atypical long.
Probably it's because the effect of COVID has been very significant in all areas of semiconductor application. The world believing that we would move into a virtual world altogether, that hasn't happened, and that has changed demand, and that's what we're still hearing is being, let's say, inventories are finally coming down somewhat. Utilization rates, which is a critical factor at some contractors, are moving well into the 80%, and those are typical areas when capacity expansion is considered, so that's in a nutshell what we see. Apart from, of course, independent analysts like TechInsights, who have always forecast, but be careful, 2024, there was also this expectation of a tide change, which didn't happen in the end, so we always look carefully at what is happening at our customers.
Thank you very much.
Our next question is from Didier Scemama from Bank of America. Please go ahead.
Yes, good afternoon. Thank you, Richard, for taking my question. So I think you kindly gave us on your press release the revenue growth in hybrid bonding in 2024. I think you said that your revenues tripled. I was wondering if you could give us your, maybe not guidance, because I know you don't want to give numbers and it's very hard, but what would you say, how would you handicap 2025 ultimately for hybrid bonding? Would you say that we are going to see another very strong growth this year or something a bit more muted just to help us a bit calibrating a little bit the models for 2025? And then on the flip chip or mainstream business, would you say that, first of all, the Q4 revenues are, if you could give us a historical perspective, where we are in that Q4 revenues relative to previous downturns?
Are we currently in line with what we've seen perhaps in a normal downturn, or is it worse than a normal downturn? Just that we understand the magnitude of the recovery whenever it comes.
Excellent. Thanks, Didier. You said it already, 2025. We're in early days still, February. We see, as I mentioned earlier, significant continuation of adoption. In the logic space, there are several key programs being unfolded from the major U.S. customer, and in Taiwan, it's also continuing. So one could expect that there will be more demand for capacity. We are, of course, in daily contact with ramping up a new installed base in the U.S., further improving ramping also, but on a less volume scale. In Taiwan, we're now close to 15 machines up and running every single day. So that looks very well. Then the key is to understand in the next two, three months, will we see HBM, let's say, adopting hybrid bonding that could add to that? Then there are some other devices which can be characterized as chiplets.
Very solid development programs are carried out to understand, yeah, of course, the yields achievable and the cost compared to using mass reflow type of interconnect. Will that in total bring the year to a higher level than 2024 is too early to tell. But there's definitely confirmation of adoption in every sense. So that looks positive. Your second question is?
No, just if you could give us some perspective of effectively the depth of the downturn in the mainstream assembly equipment revenue. Are we in a sort of level that we've seen in, I don't know, 2012, 2015, or is it even worse than that?
No, no. Oh, sorry. I just wanted to answer that, but if you look at Besi, if you look at our revenue levels, if you look at the characteristic, we dropped down in 2017, 2019, or 2018, 2019 much deeper than what we have, let's say, had so far. Same in the 2015 downturn, and the reason for that is very simple because on these new technologies, and whether it's hybrid or photonics or, yeah, other applications, that has helped us to offset that. The rule of thumb in this industry is typically at the end of a downturn, the demand is at its lowest, and you will see at a certain point a turnaround if that is the same as we've had in the past, that all of a sudden the industry realizes we need to expand, and that is then a broader expansion recovery.
So I would characterize this in a similar pattern, although longer. So there was more to digest. But then in some more clarity, what keeps us going in every downturn is not only new development of technologies, but also always new type of devices. So also in the conventional space, especially power devices, although automotive is very thin, there are many new power devices being developed, but also other package types, different configurations to create, yeah, a lower cost simply through higher density. So there's a lot of development going on. And that's why overall, if you do that successful, our margins are much higher than in previous downturns. To maintain 65% and even an increase from the peak in 2021 simply tells us that on all applications, there's a very positive business to be achieved.
Thank you. And just a quick follow-up and a precision or clarification. You said that one of your customers could decide to use hybrid bonding for HBM4. Is that from a market leader or the market leader, or is it from one of the two followers?
If we care, who am I to characterize our customers? We know very well that our engagement with the U.S. memory manufacturer has been very close forever. We also have a deep customer, both technologies in development, so both TC, TC next, and the hybrid bonder already since it's now three years. We're very much engaged in that sense, and that is, yeah, likely the first to decide. We're also in very close development, as we announced proudly, the first customer in Korea, although the first orders were logic. Memory should follow. There's this key next step on HBM4, but certainly on HBM5, who is able to deliver that? There's a lot of development going on, which looks positive for us because we're able to fulfill the requirements. Time will tell, and often you have to be a bit patient.
Very well understood. Thank you, Richard.
Thanks, Didier.
Thank you. The next question is from Adithya Metuku from HSBC. Please go ahead.
Yeah, good afternoon, guys. Thank you for taking my questions. Richard, everything you said on hybrid bonding sounds very positive. But in terms of the articles I've seen recently, the Korean press was talking about Hynix not needing hybrid bonding for 20-layer HBM. TSM on their last call said that SoIC will not be used in smartphones anytime soon. Intel pushed out their Clearwater product from 2025 to 2026. AMD, which is the biggest user of hybrid bonding, removed its AI GPU guidance. So I'm struggling to kind of square the circle here. I just wondered, do any of these data points change how you think about the demand trajectory for your hybrid bonding tools? That's my first question, and I've got a follow-up.
As we answered also to the previous question, the inroad for memory is, let's say, undefined yet. So you have the conventional way is with a reflow process, either a mass reflow process or a TC process. They are all facing certain limitations, which can be partly overcome, like this height restriction having been lifted for HBM4. But on the other hand, well-publicized data that the performance of those devices do not improve with a larger height. So the heat issues always remain. So we have not heard the definite statements that hybrid bonding will not be used. At the same time, for TC next, there's certainly with our platform many opportunities to build devices using TC. So our fate is not dependent upon hybrid bonding alone. The same goes for flip chip.
So we have this beautiful offering of mass reflow, TC, and hybrid bonding, and time will tell. These statements made by customers often vary over time. They sometimes are linked to a certain program. We've not heard that certain programs have been stalled for a long period. We've even heard the contrary that other products will be using hybrid bonding as well. So that is in a positive sense for hybrid bonding. But again, it's not only hybrid bonding.
Understood. And as a follow-up, I just wonder if the biggest HBM vendor doesn't adopt hybrid bonding and goes with an advanced version of their existing process, but some of the smaller vendors adopt it. Can the smaller vendors be cost competitive as hybrid bonding requires better clean rooms and so on? And if they can't be, then will they have any traction with their product? How do you think about that if the main player in HBM does not adopt, but one of the smaller players does?
That's a good question. And to formulate that in my words, and I said it already, we all know that connecting devices with a reflow process is, let's say, from a performance perspective, less than having a direct copper-to-copper bond. And that's why the world, since five, six years, and we started with the Taiwanese company now nine years ago, is developing this process to get away from reflow. As you mentioned, to do this copper-to-copper, you need clean room. You need zero particles. From a cost point of view, it's more expensive. You also need more accurate placement because the reflow has a self-alignment nature. That's all well known. But there are certain criteria which is also well known is heat. So the heat dissipation is much more difficult with a reflow process than with a direct copper-to-copper.
And that directly then results in performance improvement, less power consumption. So the whole equation, whether that, let's say, brings the cost of hybrid bonding to a similar level as of a reflow, also depends very much on the speed of our equipment. So the throughput so far we achieve is 1,500 typical, and then for certain devices up to 1,800 UPH. And we're working on a second-generation hybrid bonder, which is one more accurate. So that should address the next step, 50 nanometer as opposed to 100, which is generation one plus the core today. That 50 prototype will be delivered in the second half of this year, September, October timeframe. But that is also a much more stable designed system, which allows a much faster operation. And that reduces the overall cost. So those factors are impacting how technology advances. So the world is not going in reverse.
The world is always going forward. So the question may also be, can you develop reflow processes which are more optimum and producing less heat and needing less current? And the world is also developing that. And on our TC next, that's one of the key advantages with fluxless processes, either using plasma or something else. But the key is performance and power consumption. So on all the applications in high-end computing, data centers, that's what it's all about. And if we are able with the second generation to really bring that throughput up, that also is bridging the gap in cost to a large extent.
Understood. And would it be fair to say that the fact that you're developing a TC process means that you're kind of hedging your bets between hybrid bonding adoption and TC adoption?
No. We developed the first TC system also for memory, by the way, already 14 years ago with that major U.S. customer. Don't forget, we started in mass reflow 25 years ago, 26 years ago. So the way from mass reflow to TC and then to hybrid, you can say there was a period that we focused to bring this hybrid bonding to market simply based on the demand from the largest semiconductor producer in the world. But that completed our offering. So it's not the other way around. And we're very happy that we have this complete portfolio that addresses basically the complete demand for high-end interconnect technology.
Understood. Thank you.
Thank you. As a reminder to ask a question, please signal by pressing star one, and we kindly remind you to limit your questions to two questions per analyst to allow everyone to ask their questions. Thank you. Our next question is from Madeleine Jenkins from UBS. Please go ahead.
Hi. Thanks for taking my question. The first is just on hybrid bonding. And I was just wondering if you were getting continued interest from OSAT and sort of more generally what applications would you expect them to be using this technology for? Thank you.
Yes. The answer is yes. As you may recall, we have delivered the first hybrid bonder to a major OSAT, the largest, for a specific end customer program. And it is well published that the world will move from IDM, particular development, and also foundry to OSAT as soon as the process is more defined. And as I answered to the earlier question, how do you bring down cost is also to move the interconnect into an environment which is in a way less front-end directly developed. So they are preparing the big ones for the hybrid bonding offerings. And we are very much engaged in that.
But also, if you look at our offering, connecting the hybrid bonding to an automated line, which is required for the preparation of the bonding services, so cleaning, but also preparing that for an ideal copper-to-copper interconnect process is, let's say, bringing this into a subcontractor environment.
Great. Thanks. And then just quickly, kind of what do you expect for photonics demand this year? Are you expecting to grow as strongly as last year? And also, sort of how do you see photonics ramp up from hybrid bonding in the next few years? That would be great. Thank you.
More and more we see the combination. In these modules, not only current by electricity, but also light, co-packaged optics is under great development and focus. We have been engaged in that development also for over a decade, which initially was more on the connector side, but then it moved into co-packaged optics. Still early days. If you listen to the customers, their expectation is pretty, let's say, solid on this development. We're in a very good position.
Great. Thanks.
And we will now take our next question from Charles Shi from Needham and Company. Please go ahead. Oh, apologies. One moment, please. Charles, please go ahead. Your line is open. Charles Shi, Needham and Company. Please go ahead. Your line is open.
Charles, you're on mute.
You're on mute. Charles, we can't hear you at this time. I will remove your line for your question from the queue. You may signal again. And we will now take our next question from Andrew Gardiner from Citi. Please go ahead.
Good afternoon. Thanks very much for taking the question. Just one quick clarification on the hybrid bonding growth you reported for last year. Richard, you described it as a tripling. Is it fair to say that that was all unit-driven, or was there much of a change in terms of the spec and price of those tools? And then secondly, one on the cost side, you're increasing OpEx markedly in the first quarter, and you're citing strategic consulting costs. Can you give us any sense to what they're looking at? You probably don't want to say too much, but maybe is it a growth question? Is it a cost question? And is it a one-off, or will it continue through 2025?
I suppose put another way, OpEx for you guys over the last few years has traditionally peaked in the first quarter and then been down a little bit through the remaining quarters of the year. Is that something we should expect again for 2025? Thank you.
Excellent questions. Thank you. Number one, it is unit and price increase, so remember, we started initially with the generation one, 150-200 nanometers. We introduced two years ago the generation one plus, which brought the accuracy down to 100 nanometers, and the bulk of the orders are for 100 nanometers, and the price for generation one plus is about EUR 500,000 higher than for generation one, so not, let's say, impacted by price pressure. Your second question, consulting cost, there are three, well, major parts. One is the support we have from, let's say, for technology, 50 nanometer, but then we have to move down to 20 nanometers, and we have support from the infrastructure in the industry to understand better how we can develop the right solutions.
And then we use on a two- to three-year basis outside consulting help to align with our key customers, the winners, the development roadmaps for the next three to five years. So with all happening in this industry, with the AI in particular, development and demand attached to that, we have engaged again in a very close with key customers program defining our strategy. And that will always result in a new model. And that model, once we're done, we will share with you. And that's how we also have developed the current model. And that is simply carried out from mid-December until mid-April. So once we're done, we will share the outcome.
Thank you, Richard.
Thank you, and we will now move to our next question from Ruben Devos from Kepler Cheuvreux. Please go ahead.
Yes. Good afternoon. Thanks for taking the question. I just had one on sort of summarizing what has been said on hybrid bonding and the 15 customers, right? So could you sort of give a high-level overview of where each of them are in the adoption curve? I understood the new Japanese and Korean customer is basically R&D. You talked about the U.S. logic and Taiwan logic. That's mostly for volume. The OSAT, you also mentioned a major OSAT customer, not yet in volume, I understood. But yeah, you've got 15 in total. So very helpful to sort of have a high-level overview of where each of them are in terms of R&D, pilot line, follow-on capacity orders. That's very helpful. Thanks.
Thanks, Ruben. One is proudly in your country and not so far Imec. So we have development centers in Europe engaged. So one in Belgium and one in southern France. Also close cooperation with a German development center. And then we have in the U.S. universities. And we have one major customer. A second one starting from a low level. And then, of course, Taiwan, which was the first, where we have our largest installed base. All logic, also in the U.S. That major customer, logic device, Clearwater Forest. Also Diamond Rapids is supposed to be hybrid bonded. In Korea, also logic, but also, yeah, close to adoption or early adoption in memory. U.S. memory as well. And then Taiwan logic. It's well publicized. It's a major program of the Japanese industry with leading-edge technology below two nanometers, widely publicized.
That's amazing that they have selected our hybrid bonding over the systems available from Japanese suppliers and also Korean. That's in a nutshell. If you add it all up, you're at total 15.
Okay. Thank you. That's very helpful. Okay. Well, and thinking about maybe competition then, because I think in your prepared comments, you said that even though you already have 15 customers, it's still in the early stages, but it's gaining a lot of importance across a growing range of applications. How do you assess the competitive landscape at this stage? I mean, in Logic, it appears that you're clearly leading, and you've been supplying hybrid bonders since 2021. So yeah, it would be great to hear your perspective on the development efforts in the industry.
Today still, the major focus is on reflow processes. As you know, in flip chip, we're a world leader for a very long time. TC, the landscape is a bit more diverse, and some have certainly a larger installed base on previous technology levels. The market is open for the next, so the below 20-micron bond pad pitch in particular, also different processes. The fluxless processes using either plasma or some chemical. Hybrid bonding, also with the continued adoption, competitors are announcing also systems aiming at that growing market. At the levels of 100-nanometer, also integrated lines where we have a much better process overall control, that gives us a significant lead. Also, in the strategic exercise we are currently engaged in, that is also one of the major parts.
How do our customers, because they're involved in this exercise, what are the main directions to be expected in the next three years? Industry, of course, likes competition. There's certainly, in this very broad spectrum of interconnect devices, whether it's the Logics, where you have many different types, the 2.5D, 3D, then also the memory stacks, which was alluded to earlier. They're different processes. The three major ones have all different processes. And that also offers different equipment solutions or requires different equipment solutions. So that's in a broad brush how this industry will look like in the near-term future.
Okay. Thanks a lot, Richard.
Thank you. And we have back now Charles Shi from Needham & Company. Please go ahead. Your line is open.
Thanks. My line dropped halfway through. Forgive me if I ask a question that was already asked. Richard, I want to start with a comment on 50% of the order came from AI-related applications in 2024. I think this is the second time you mentioned about that 50% number. I want to clarify, what do you define as AI-related? Because every company seems to do it a little bit differently, but if you provide some clarity on that, thanks.
We define it as being related to high-end computing data center applications, 2.5D, 3D, Logic, of course. So all those devices in that high-end is what we determine as AI. Probably you could define that market even broader, but for us, it's at that very high end.
So what kind of products are you supplying into those end applications at least for 2024?
Hybrid bonding. Also, let's say photonics products, certain flip chip. Don't forget the 2.5D, the OS on substrate. Those are all fitting into that qualification.
Okay. So among these three, right, the hybrid bonding, photonics, and some of the 2.5D OS applications, how would you rank order the size? I mean, within that 50% of the total orders, which one were the largest? Which one were the smallest?
The largest were, yeah, of course, they were all batch orders in the NVIDIA supply chain. Definitely that. Well, I would say the largest is, of course, the hybrid bonding orders in total. Also, the specific one in Q2, the 26 at that time, EUR 60 million. We're just short of that. But if you add that up, you come to EUR 300 million.
Got it. So hopefully that is counted as one question. Can I ask a second one?
You can also ask a third.
Thanks, Richard. Then the next question really is about co-packaged optics. We know that your leading U.S. customer has been working on this for a long time, probably I would argue probably longer time than the leading Taiwan customer. But the U.S. customer seems to be using more of the TCB in some of the bonding steps. I believe that was the EIC bonded onto PIC, that specific step. But there has been rumors maybe the Taiwan customer will use hybrid bonding, but I wonder what's your view here between TCB and hybrid bonding for that particular application? What do you see, especially in Taiwan?
Yeah. The hybrid bonding, I mean, don't forget that in Taiwan so far is the most experienced. And that makes it more easy to test those applications. So to answer that question, we see the most traction at this moment outside of Taiwan, a bit less in the U.S. Maybe that also has to do with other factors at that customer at this moment, but we don't know. But you're absolutely right. It started in the U.S. already more than a decade ago. But anyway, that's as much as we can share.
Got it. Thanks, Richard.
Thank you. We are now going to move on to our next question from Martin Marandon from ODDO BHF. Please go ahead.
Hi. Thanks for taking my question. My first question is on TCB. Maybe can you give us an indication of what do you think are the advantages of Besi's TC next fluxless equipment over the competition? And notably explain the advantage because I think there is a debate at the moment. So the advantage of doing plasma cleaning to remove the oxidation layer versus one of your competitors using formic acid.
Yeah. Well, the situation currently is as follows. First of all, our TC next has been developed for the next step with tighter accuracy. So below 20 micron bond bed pitch, that is where the key advantage over today's competition is very evident. Also with more and more qualifications having been performed. Then on the fluxless discussion, formic acid, as we understand from customers, is a second-best solution because you still have issues in applying that toxic at the same time cleaning question as well. Plasma is, when it is finally developed in the right applications, a much preferred solution. So the world is working on both. And our TC next has the ability to include the plasma heads into the bonding area. And that has a major advantage. And there's a high interest currently and many development programs ongoing. And that should become more clear in the near-term future.
Okay. Thank you very much. And my second question is on the new hybrid bonding system. So can you maybe make a quick update on the new system coming with 15 nanometer accuracy? And what is the customer demand there? So do all advanced logic customers have expressed strong interest in getting the 15 nanometer tool when it will be released, or do some customers will stay at 100 nanometer for now?
Well, that's also a very key question. This development is driven by the large Taiwan customer. And their roadmap shows the requirement for 50 nanometer accuracy as of 2027. So if we are able to deliver the development tool by October, September, October is the planning, then the qualification in the next year, and that should be available. And probably that is tied to the below 2 nanometer chip design generation. Others are facing, yeah, that same hurdle. So 100 is expected not to be sufficient once we go down below the 2 nanometer design structures. So that is also on the roadmap of all the leading-edge logic developers.
That's all. Thank you very much, Richard.
Thank you, Martin.
Thank you. And our next question is from Martin Jungfleisch from BNP Paribas Exane. Please go ahead.
Yeah. Thanks for taking my question. Good afternoon. Just on the hybrid bonding orders, you mentioned that these have doubled in 2024 compared to 2023. And I guess this would imply orders of around 15 tools in Q4. Is that a correct assumption? And can you also disclose the backlog of hybrid bonders as of Q4? Is it around 40 tools to be shipped this year still? That's the first question, please.
We don't disclose on individual numbers. We have shared with you because it's always, let's say, difficult to tie them down to the delivery date. You are involved in the discussion about orders and delivery. The comment we made is over the whole year 2024, the order intake doubled. We are shipping, as we mentioned in the prepared notes, currently the last part of the major order placed in June and several smaller orders which have been entered in Q3, and those in Q4 will be delivered in Q2. That's the picture as it is.
Okay. Thank you. And maybe a second one. There's been some news from the U.S. restrictions on Chinese OSATs. So TSMC apparently won't ship orders to Chinese clients using Chinese OSATs unless they're on the whitelist. Just wondering if you can provide some color on it. What have you heard from this matter? And what would you say are the short and long-term implications from that for you?
We also heard this message. Already for quite some time, the Chinese OSATs are very careful to remain in the white zone, as you call it. That's also the intention of the restrictions. We are, as we have also indicated in the past many times, very careful to follow strictly the U.S. trade regulations. That may result in more business growth outside of China. That's at least the intention. Time will tell.
Okay. Got it. Thank you.
Thank you. And to the hour. And we'll now take our next question from Timm Schulze-Melander from Redburn Atlantic. Please go ahead.
Hi there, Richard. Thanks for squeezing me in. I had a two-part question, please, which is really about co-packaged optics and the extent to which R&D is shared across hybrid bonding, TCB, and co-packaged optics, so the first part of the question was maybe you could just share a little bit more details about the number of tools that you have addressing co-packaged optics, the sort of ASPs and the roadmap and the challenges, just kind of how similar or different is that to hybrid bonding, and then on R&D programs and looking at 2025 with the Gen2 tool, you've got a lot on your plate, presumably also development of TC next as well as co-packaged optics, and so my question was just, is the sort of precision and vibration development one vector that applies across all of those products, or is the R&D duplicative rather than shared? Thank you.
Well, the first answer is clearly the objective is always to have a shared, we call that common modules, common platforms. And the best example is that the Generation 1 and 1 + is a common platform to the CHAMEO, the highest end of flip chip. So the next step, Generation 2, is a different basic design. And to achieve, let's say, the key objectives to achieve is more stability. More stability because precision and placement is introducing the never-ending battle between accuracy and dynamic behavior resulting in stability issues. We intend to use that platform once successful, much broader for the developments, not only in hybrid, but also in TC as a next platform. So that's a philosophy we have been using for two decades. What's being used today varies.
So the industry is always looking at the lowest cost option, and that is directly linked to the accuracy requirements. So less accuracy allows to use the multi-module attach or the CHAMEO 8800. And those are very sound platforms already over many years. But their accuracy stops at about 1 micron. So well away from 50 nanometers. But you don't need 50 nanometers for the co-packaged optics in the foreseeable future. Probably 1, but maybe, yeah, down to 150, 200 nanometers is what is required. But we will share at the next Capital Markets Day more details about those applications. A lot of it is already in the last update from June. But that's a very interesting, always, and a perfect question. So you want to.
Great meeting. Thanks.
Use your R&D investment as broad as possible to have the best return on investment. Excellent. Thanks, Tim.
Thank you. And this is all the time we had today for questions. With this, I'd like to hand the call back over to Mr. Blickman for closing remarks. Over to you, sir.
Thank you all for listening in and the perfect questions. I can assume you have more questions and don't hesitate to contact us directly. It will be a pleasure. Thank you all. Bye-bye.
Thank you. This concludes today's conference call. Thank you for your participation, ladies and gentlemen. You may now disconnect.