You can hear me. Good afternoon, and welcome to Besi's Capital Markets Day, or Investor Day. Whatever the name is, we had many discussions in the past. How should we call this? Capital Markets Day was not the best because then people expect an update for the quarter. Well, this is not an update for the quarter. This is a longer-term perspective on where we believe this industry is heading, and especially where Besi is heading. So, no update on the quarter. I hope I don't disappoint you, but anyway, that's how the rules of the game are. But before I start, I'd like to thank everyone taking the time and coming here. This time it's a bit more formal in a hotel. The last two or three years we did it in Radfeld, where you also could see machines.
I had some reactions from people who said, "Fantastic, it's here." But many also said to me, "Next time we want to have it in Radfeld again. We want to see machines." So anyway, so next time, 26, I promise you, we will organize it again that you will see the most beautiful machines. And especially at that time, you should see the 50-nanometer machine, the Generation 2. But anyway, more about that in a few slides. So what have we prepared for today? Is this working? Yes? Well, the schedule. Strategic overview I will start with. Market trends by Chris, Chris Scanlan, who most of you have met, I think.
Then Peter, the head of Submicron, SVP Die Attach Submicron. Then we have a short break. And after that, Christoph, who is head of Die Attach, our conventional world, as we sometimes refer to, and Jim Wang. So what's driving that?
And then I will share with you the update of our strategic plan. This morning you saw in a press release that we have put the bar significantly higher. But we tried to convince you in this presentation that that is an outcome of a very thorough investigation with our customers in particular, and also the help of outsiders in our TAB, our Technology Advisory Board. So understanding the markets, the progress, and what is to be expected has led to an updated model. But we'll get to that. So let me start with the strategic overview. This picture doesn't change. I think I don't change either. So anyway, AI expansion drives the continued advanced packaging adoption. I think we all agree on that. So we see that in every sense, every single day.
Huge investments in software, in models, in a world which is completely different than our current data world. And that simply drives enormous demand. Demand mostly focused on cloud and moving to edge computing, accelerated advanced packaging innovation. So if you look at this world, and that is very fascinating, if I compare what's happening right now, and let's say in the past five years, since advanced packaging became ever more in the view of all of our customers the most critical thing, with device geometry moving down below three nanometers, how do we connect? And how can we use that interconnect to build more effective performance and less heat? And that is driving for, let's say, the preparation of an enormous change in society, which is well underway. And that is where advanced packaging is ever more critical.
Then if we look at this particular point, how are semiconductor devices connected? You may well know that for many, many years, 50, 60 years coming up, the world of semiconductor assembly was a world of components. Individual components and customers developed circuits. And those circuits consisted of all kinds of components, standardized in the old days in 19-inch racks, computers. But then with more and more portable devices in the past 30 years, ever smaller geometries led to ever smaller interconnects using substrates, using whatever interposers. And now that is more and more done on chip level itself. And that is creating the unique opportunities, which then force us to build ever smaller geometries. So with the hybrid bonders from 150-100 nanometers, now down to 100.
As you may know, we will ship the first 50-nanometer machine by the end of September, October, as a test vehicle to a leading customer in Taiwan. That opens up again a next generation in ever smaller circuitry. In addition to that, we have the new photonics coming in to these technologies, offering another source of energy in steering those devices. That is where we've been involved for the past over 10 years already. Now you finally see that coming into mainstream applications. A significant growth driver also going forward. The never-resolved debate yet, HBM4, HBM5, where is the inroad of hybrid bonding versus the today used reflow processes, whether it is mass reflow or whether it's TC, thermal compression? Where is the crossover? Because on the one hand, TC is a known technology and a known cost.
Hybrid is more expensive, but better performance. On these criteria, there's much more on the performance side. Chris will take you through the latest views in the industry. That is very important to understand where hybrid is kicking in. At the same time, a TC roadmap for many applications which are in those technologies and then not so critical. Next generation cores, LNR structures, evaluating TCB the same, TC Next, as we call it, next generation flip chip and MMA. Christoph will share some more about that in his presentation. Anyway, some key elements which lead to ultimately a model which we have shared, as said this morning in a press release. Simply to understand, if you are talking about packaging, well, you need first front end.
So if we look at this list again, who is building what and where, it's a very impressive list, but that tells you a simple message. The world is preparing for an enormous next round. And that next round in all corners of the world here, in the U.S., in Asia, in Korea, in Japan, in Europe, in India, and simply name it, both front end and back end facilities, all in preparation of this AI adoption in the world. And if you look at the impact of that, you can see here on the left side, 2024, last year, and an estimate how this will develop in the next four, five years. And Besi's portfolio is clearly focused on those growth areas. And today already, 70% of our revenue is from advanced packaging applications. 50% equipment revenue from advanced die placement, so below seven micron accuracy.
50% is related, what we expect from the AI-related applications. We are entering into that new fluxless TC market. Anyway, significant growth in the next years, a compound expected of 18.9% and other advanced packaging of about 7.8%. More about that in a few slides. If I look here at the entire market for hybrid bonding and TC fluxless, it's estimated to be a market size in 2030 of about EUR 2.1 billion. That is our estimate, but also compared to those in the industry independently who try to size that market. If you look in the box, hybrid bonding estimated to be the largest segment by 2030, clearly with the transition expected for HBM stacking, so 20 stacks. That market we estimate as a size of about 350 systems by that time. That we call that the mid case.
The same for the hybrid bonding part. The TC part is a bit smaller than that, is about 100 units by 2030. If you take an average selling price, it's not hard to calculate. That delivers you in a mid case model, roughly this total market size. Just to share with you the latest data on market shares on 2024 die attach overall, and then the advanced die placement, you see that we have again made progress focusing on the forefront, and that should offer us also the opportunities to extend the growth in the years to come. This is data from TechInsights from last week, by the way. To summarize our strategy on seven points, they're all pretty clear.
In the last many years, we have shared with you exactly that focus to maintain always on the forefront of development, but not only at the forefront of development, but also in the mainstream. In the mainstream, pick the winners in the three categories. We have communication, we have data centers, computing, and we have automotive, and that's a never-ending growth model for all those applications, and some years there's more growth in one and sometimes in the other, but if you maintain your focus on the right winners, then typically that gives you a higher growth and a higher return than the overall industry, and for that, we need to continue to increase our R&D spending simply because of complexity and also because of different applications, so if you look at the entire scope of packaging solutions, it's very easy to imagine that over the years it only increases.
Now, with these 2.5D modules, 3D modules, ever more variety. And that variety also requires ever more R&D solutions. Are your systems able to handle multiple dies and with different accuracies and different processes? Also, the materials are constantly under review. Different materials used and tested. And with things getting ever smaller, like you see here, the 50-nanometer accuracy, that comes close to the world of litho 20 years ago. So anyway, enormous challenges, and that increases the R&D spend. That we are in a cyclical world, every one of you knows that. And we always like to show this slide, which is a very simple history from 2006 onwards. We see a pattern of four, sometimes three years of cycles driven always by new end market applications. For many of the past, this was high-end smartphones.
Now it is very much focused on how data is being processed, data centers, etc. And you see in between, you see the average, but then in the yellow, you see the gross margins. And the gross margins typically tell you the success of the choices which companies make, and not only choices on end customers, pick winners, but especially how you build the machines. Many years ago, we understood that your profit is in the manufacturing of your systems. And in a cyclical world, it's not just that. It is also being able to scale your business at the time needed, both up and down. And that is what you nicely see in this slide. And if you look at the very right in the peak, you see 64.3% gross margin right now.
And what you saw this morning, we have upped that target for the next five years simply because we believe that the current generation of systems is again more complex. And we have a lead in many ways. And that should translate in building our systems more effectively. Supply chain focus, more experience is helping to create ever higher returns. Next is Chris. So with that, questions. If you have an immediate one, you can ask. But otherwise, at the end of the presentation. Thank you.
Hello. My name is Chris Scanlan. For the next 30 minutes or so, I'd like to share with you what we see as the long-term technology and market trends and how that affects our business and our products strategy. Yeah, starting with the overall semiconductor market, yeah, the good news is we still see long-term healthy growth rate in the market.
The growth rate over time, on average, has been pretty consistent, yeah. Let's say for the last 30 years and into the future. Only what changes over time is really what are the applications, what are the new technologies and devices that are driving the growth, and clearly, what we see for the future is AI is going to be driving growth for the foreseeable future, and if you look at what that means, there's really kind of three categories of devices that are driving the growth. Early on, it's really the infrastructure build-out, so the data center components, the GPUs and CPUs and switches and other things that you need to build the infrastructure for these systems, then we'll see the current devices that we all have, our cell phones, our PCs, even automobiles, becoming AI-enabled.
That's why you see the non-AI trend sort of going down, because the existing things that we're using today are going to be enabled by AI, and then in the future, things that are really kind of native AI devices like AR glasses and so on, so this is what's driving the growth in the future, and this is what's driving our roadmap, and you already kind of see this in our numbers. Last year, 43% of our revenue was from computing. This is really unusual for us. Usually, it's less than 30%. It's usually behind mobile, but this was really driven a lot by the, yeah, this whole infrastructure build-out, right, so a lot of capacity going into CPOs and other advanced technologies like transceivers and things like that required to build these AI systems, and this is kind of the initial phase, like I said before.
But we anticipate in the future, all the end markets, mobile, automotive, and industrial, will have many applications taking advantage of this technology and driving growth in those markets as well. But there's another kind of macro trend that's also driving our product strategy, and that's the slowing of Moore's Law. And what does that mean? Well, simply put, the transistor density that the fabs are able to achieve in their process has been slowing. So you really saw an inflection point somewhere around 2017, where the growth rate and the number of transistors per unit area started to decrease. And it became more complicated, or it is becoming more complicated to increase the density and the performance of transistors. So you can't any longer just kind of scale them in two dimensions.
In order to make gains, we have to go in three dimensions to FinFET, maybe to gate all around kind of transistors, backside power delivery technologies in the fab. In the future, the complementary FET is coming. And what this means is that the transistor or the front-end process is becoming much more complicated, many more steps with a lot more cost. And you can see that in the second chart. The result of that is the cost per transistor, which in Moore's Law was reducing every generation, is starting to increase in every generation, so going the wrong direction. And then also, as a result of the slowing in the transistor density, we see die sizes rapidly increasing. So this data I plotted here is from NVIDIA devices. So every data point here is that generation flagship GPU die size from NVIDIA.
You saw also around 2017, they started bumping up against the reticle limit. What is the reticle limit? That's the maximum size chip that you can pattern using advanced lithography today. And finally, with this current Blackwell generation, they broke through that limit. They could no longer implement advanced design on a single chip. So now we're talking two chips, the next generation, four chips, plus some additional IO chips. So simply, we're seeing kind of an explosion in die area. So what does that result in? That means people have to start thinking about how to integrate multiple chips in a package and what kind of strategies can they use to optimize those kind of designs. You can already see examples of this. So these are the kind of the high-end devices from all the major chipset suppliers for high-performance computing.
So you see AMD GPU and CPU for advanced data center applications. You see NVIDIA's Blackwell GPU, their new network switch platform, Intel's AI accelerator, Gaudi, and their new Clearwater Forest CPU. So all these are using multi-chiplet designs. AMD was kind of the first to employ these chiplet strategies. The other thing that you see is they're all, at least in some of their products in their portfolio, already using hybrid bonding. So as I mentioned, AMD was really the first of the suppliers to really adopt this chiplet architecture. And you see on the bottom kind of what that looks like. So they use the chiplet architecture to break the CPU device. They use the same strategy on their GPU. But in this example, in the CPU, they implement the CPU cores and some embedded S-RAM in a single smaller chip.
So really just the part of the chip design that really can take advantage of the most advanced node. Why do they do that? Because the most advanced node is very expensive and getting more and more expensive over time. They take the rest of the design, the IO functionality, some networking functionality, and communication functions. They put that in a trailing node, and they implement that as a separate chip. Then what they can do with that is they can take that CPU chip, and as you see on the bottom, they can combine it in different ways, different numbers of them combined with different IO chips to create different products for different markets. So it becomes a very scalable and cost-effective solution. So they were the first to kind of implement this, but we see now other suppliers adopting the same kind of approach.
What does that mean for Besi? Well, if you look back even like seven, eight years ago, all these chips were single SoCs that were attached to a substrate using a flip chip bonder, using a relatively relaxed bond pad pitch that required, yeah, at that time an advanced bonder. But nowadays, we have chips that have dozens of smaller chiplets integrated into a single package. So there's more die attach placement steps. And also, because these multiple chiplets are meant to function as really kind of a system-on-chip equivalent, the interconnect density between them is very high. That means that the accuracy required to place these chips is very high as well. So these are very, let's just say, expensive machines that are required, and there's a lot more of them. So effectively, the capital intensity is higher with chiplets, and that's simply good for Besi.
So let's look at some of the processes that are required to build these systems. So I showed you six examples in a previous slide. And this is kind of a picture of what those packages look like internally, generalized. Okay. And what are the kinds of interconnect technologies and processes that are needed? Well, the most advanced process that we have in our toolkit is hybrid bonding. So hybrid bonding is used, first of all, to create the SoIC or 3DIC. So this is a three-dimensional chiplet that is actually fabricated by the wafer fabs or foundries. It functions, and it looks just like any other wafer, but inside it has a three-dimensional stack of functional blocks that are created by hybrid bonding. Hybrid bonding can also be used in high-bandwidth memory, and I'll describe the roadmap for that. We anticipate with HBM4E, it will be adopted.
And we're also seeing adoption in co-packaged optics, where TSMC is already implementing hybrid bonding to combine electronic and photonic chips together to create these photonic chiplets. Thermal compression bonding is also needed in these systems. So chip-to-wafer thermal compression bonding is used today for high-bandwidth memory. And it can also be used for things like embedding 3D bridges and interposer wafers and for assembling these complex wafer-level assemblies. We still have quite some volume and standard flip chip, but these are now becoming more focused on wafer-level assembly. So all these packages are fabricated on wafer-level interposers. So we need more accurate wafer-level flip chip equipment with capability to bond components from wafers, from different kinds of formats like tape and reel, all together into a complex system, and also to place things like embedded bridges in the substrate. And so we have advanced technologies for that as well.
And then on the bottom, we have all the other assembly technologies that are needed, like advanced flip chip, like photonic assembly to place these photonic components. These modules get very big. So these CPO modules are now over 70 mm and going to 110 mm in the future. Those have to be placed using a complex flip chip process. And finally, encapsulation. So at the end of the day, you have to encapsulate all these things using molding. And if you look at our product strategy at Besi, basically, our strategy is to offer the most advanced and most capable equipment for all these steps in order to provide a full solution for these AI applications, starting with hybrid bonding, where we already have a strong market position. Richard mentioned the 50-nanometer bonder on the roadmap, and I think Peter will share our detailed roadmap with you later.
Thermal compression bonding, we have just introduced a, we think, leading capability with the best quality, the best accuracy, and the best productivity in the market. We have a new flip chip bonder in development that will be introduced in Q1, which is really laser-focused on providing three or four times higher productivity for CPOs, and then a range of other products that are servicing the other applications that I discussed. So, as I mentioned, hybrid bonding is really the most advanced interconnect in our portfolio. And why do we think that it's the technology of choice for advanced logic integration? The performance benefits are obvious. We've talked about this before. Hybrid bonding is the only technology that allows a wafer fab to really integrate 3D interconnect in the fabrication process itself instead of in an assembly process later on.
That's because hybrid bonding doesn't involve any materials that are not compatible with the fab process. It provides the highest interconnect density, highest speed, highest bandwidth density, and most importantly, the energy-efficient performance of the transistors is increased by a factor of 100. But what's not so obvious is that the cost per interconnect is actually lower as well. If you compare it to the next best alternative, which is fluxless thermal compression bonding, just on the basis of interconnect density, the cost per interconnect is a factor of 10 lower, actually, with hybrid bonding. So for high interconnect density designs, this is really the best way to achieve a cost-effective design. That said, TCB will still be used because many applications simply don't require the highest interconnect density available. So I want to turn now to the new applications that we see emerging in the market for hybrid bonding.
Intel continues to expand their use of hybrid bonding. So they announced the Clearwater Forest last year. They're now ramping production, and that's becoming very successful. The Foveros Direct is now part of their design kit in the foundry, and we expect to see more designs from them in the future as well. Then we also see new customers coming. So in December, Broadcom made an announcement that they're adopting TSMC's SoIC process in their custom logic business. So why is that significant? So Broadcom has a pretty significant custom ASIC business. They work with Google, they work with other hyperscalers, and they help them to develop accelerators and other custom ASICs for their own internal use in data centers. And they announced that they're going to be employing hybrid bonding, SoIC, in the design of some of those devices, with production starting already in early 2026.
So that means we should see multiple hyperscaler kind of customers adopting this. There's also been a lot of media reports about the M5 adopting hybrid bonding. We can't really confirm this, but I think last year I presented a roadmap where I explained why that kind of makes sense, because these previous generation M-series processors, I think going back all the way to the M2, are already kind of bumping up against this reticle size limit. So another emerging use case is co-packaged optics. What is co-packaged optics? This is really the idea of bringing the fiber that is used for high-speed communication. Typically, you would see these fibers plugged into the server, right at the edge of the server board. But then you have to, it's kind of like fiber to the home, kind of a similar concept.
When we went from fiber to the neighborhood to fiber to the home, we realized higher speeds, and the same thing is kind of true in the network. The idea is to bring fiber directly to the edge of the switch or the GPU device. In order to do that, you have to have something to connect that fiber to that will take that signal and light form and convert it to electricity so that it can communicate with the switch or the GPU. And NVIDIA made an announcement in March that they've adopted a technology from TSMC using hybrid bonding to create these photonic elements.
So what they do is they use hybrid bonding to attach the electronic chip to the photonic chip, and then they mount a plurality of those, about 36 in this example, around the switch in order to create the backplane and the bandwidth that they need to communicate with the switch. So that's at least 36 hybrid bonding steps per switch in this particular case. So we anticipate this will be a big driver of growth in the future. And there are other varieties of these photonic chiplets. Broadcom, for example, has a different technology that uses not hybrid, but advanced thermal compression and flip chip. So it's an exciting growth area as well. But we have to talk also about hybrid bonding in TCB and HBM. And where is Besi anticipating intercepting this market space? Because we haven't been so involved in HBM up until now.
But with our new TC Next machine, which Peter will describe to you, again, best accuracy, best advanced process control capability for high quality, and most importantly, highest productivity fluxless process, we anticipate penetrating an HBM4. We've already announced orders from major customers. And with hybrid bonding, we anticipate first adoption in HBM4E, which we should be able to take advantage of as well. If you look at the roadmap, there's really kind of three technical drivers for this. The first we've talked about before, it's just the number of die that we can fit vertically within the height budget that we have. So with microbump or TCB, we can realistically go up to about 16 and no more. After that, which we expect with HBM5, we have no choice but to go with hybrid bonding.
At the same time, the interconnect pitch is reducing with every generation as well, increasing the accuracy requirements. That's why our thermal compression bonder is becoming interesting for customers. And also hybrid bonding will be needed for that reason as well. But another key reason that we maybe haven't talked about yet is the base die technology. What does that mean? So the way these things are built, we have a stack of many memory chips, but they're all placed onto a base wafer. And that base wafer has a logic function, helps the memory to communicate with the processor sitting next to it. Up until now, that's been really fabricated by the memory suppliers themselves. But starting with HBM4, we anticipate that these base die will be fabricated by foundries like TSMC using advanced logic nodes. And once that happens, there's going to be a lot of customization.
The end user will be able to influence the design of this base die, and they're going to want to put more compute functionality or different functions into that base die. There was a paper by TSMC just two weeks ago where they talked about what that means for the interconnect. Without boring you with the details, basically they concluded that you can't do this without hybrid bonding because you're going to add thermal dissipation into that base die because of the additional functionality in the base die. With microbump, you simply cannot get the heat out effectively enough. You end up burning up that bottom chip. Because of this, hybrid bonding is anticipated to be required for these advanced HBM devices. That timing is about the same as HBM4E, HBM5. We also see growth opportunities in smartphones.
So the smartphone market itself is not growing rapidly, as you know. However, AI-enabled phones are growing quite rapidly, and that's going to cause changes in the architecture of the application processor package in particular. So if you look at how those are assembled today, they're single chip packages, still SoC. But AI is going to increase the demands on the processor as well as the memory capacity and the bandwidth of the memory interface, driving these customers to go to chiplets. So this will be kind of a once-in-a-decade technology transition for this kind of application. And what we anticipate happening is that all the customers are going to go to a chip-to-wafer flip chip kind of construction, driving new kinds of interconnect compared to what they have today. And in the future, hybrid bonding to create a 3D chiplet to also be integrated into the same format.
This data on the lower right is Yole's prediction for when that will happen. It could be 2027, 2028, not sure, but that's what we anticipate happening as well. So that's a little bit about our view of the future for technology. Again, AI is really driving growth in our business and in our forecast for the future, hybrid bonding, advanced TCB, advanced flip chip. With our product portfolio, with developments focused on these applications and enabling the best capability for these applications, we think we're well positioned for the future. That's what I'd like to share with you today. Thank you for listening. I'd like to hand it over now to Peter.
Thank you, Chris. Welcome, everybody, to our this year's Capital Markets Day. My name is Peter Wiedner.
I'm the business responsible for the submicron business, and I have been with Besi for over 20 years now. Now, when we at Besi say submicron business, what is that? Submicron business, in our case, is everything what covers hybrid bonding, all the machines, all the future roadmaps, as well as everything what we are doing for thermal compression bonding, which you can see in the middle of the slide here. But we also do have actually a product line which is doing lid attach, so the lid, which is the cap that you put on these advanced packages on top in order to seal the package and also to get the heat out of the package, which is getting more and more important. And all these areas are covered within that product portfolio.
Having that said, let's have a look, actually, how this product portfolio fits into, let's say, the chart of interconnect technologies. So you see that's a generic chart, nothing new. You have seen that before, by the way. So where you see from wire bonding on the very left side all the way to hybrid bonding on the right side, where actually the driver has always been to get interconnect technologies that allow smaller bump pitches and smaller interconnect pitches. And semiconductor, high-end semiconductor has always followed or even was the driver of these roadmaps also in the past because they needed ever smaller bump pitches and higher interconnect count. And for that reason, so the high-end is always on the very right side of the chart.
It obviously is here with hybrid, undisputed, and Chris already explained a lot about hybrid and why it's used and where it's used. There is actually right next to it, so actually not below the 10 micron because there's hybrid, but above the 10 micron, there is actually an arena which is also interesting and that I want to point out. That arena is roughly between 10 and 25 micron, which we call the TC Fluxless Arena. That's a special playground for a special version of TC technology. Note, this is not the whole TC world because the standard, the mainstream TC is, as you also can see on the chart here, more in the 40, 35 micron bump pitch arena, which has been there for a longer time already, so nothing new.
But actually this specific area here that goes between 10 and 20 roughly, that's really an area which was until recently more or less a gap where there were no applications and also no technologies for that interconnect because hybrid did kind of a jumpstart and left that gap open because they started with 9-micron. But this area is getting more and more interesting for a lot of applications, no matter logic and memory and everything else. And that's the area which will be filled with thermal compression, but once again, not with a standard thermal compression, but with the so-called fluxless thermal compression. I'll get to that on the next slide. And these are the areas, these TC fluxless and hybrid areas, that's where Besi has set the strategy on, and that's what we are targeting to cover with our product portfolio.
Now, talking about product portfolio, let's start actually with the TC machine, the TC fluxless machine, which we call TC Next. Yeah, what do you need if you want to cover this specific gap that I've pointed out? First of all, because the interconnect pitch is extremely small, you need a very high accurate machine, higher than the normal TC, which is not easy because thinking on the TC is a hot process, a lot of force, so quite a complex thing to achieve, and we have designed this machine from scratch in order to achieve this, and we are specifying our machines currently with 0.7 micron accuracy, which is best in class and which is perfectly suitable for these types of bump pitches.
The other area, and that's the reason why it's called fluxless, is normally with TC, you're using flux to overcome certain issues with oxide layers, which are on top of the bumps. And you cannot do that anymore with these small bump pitches. And for that reason, you need alternative technologies, which we call fluxless. And so you need oxide reduction and avoid reoxidation and things like that. So you need a lot of either plasma treatment or gas treatment and things like that. And that's for sure also all integrated in that machine. But it's not only integrated, it's also integrated in a way that the consumption of gases, because it always goes along with consumption of gases, is extremely small, so the consumption, and that makes it a very efficient and cost-effective machine in the production environment for the production cost.
Chris mentioned, actually, the quality. So, typically, the more right you are on the other chart that we have seen, the more expensive the die are. So, customers want to have the highest yield possible because every yield loss is quite a big impact on their balance sheet. And so, process control, or some customers call it defense, but it's the same, just two different names. We call it process control, real-time process control. That's a key feature for all these high-end machines on the right side, so also for these TC machines here. And that's not easy during the process, during the joint that you are just forming to control if everything is going well physically. But we have come up with methodologies to do that, and we have implemented that and even patented.
For that reason, it's not only that we have a highly accurate machine that has all the features that are needed, we also can really do a real-time process control, which is once again supporting the yield. If you combine all that with a high UPH, depending on the TC process, once again, there are different flavors, but up to 2,000 an hour, that really becomes an incredibly great benchmark for cost of ownership and the quality and to really utilize that in that space that I've pointed out. That's what we are offering for the market today and where we have been starting selling this machine this year. Now, let's go to our other area, our stronghold area hybrid. It's definitely our stronghold.
So our machine, which is in the market today Chameleo machine with the 100 nanometer accuracy, is the standard for high-volume production today. And that is reason enough for us not to sit back and relax, but to double our efforts because if you look at the roadmaps of our customers, and once again, Chris told you a lot of technical things, what's going on and the direction this is heading, they say, well, we want even higher accuracy because our next logic generation will even have smaller bump pitches. That's what the logic guys say. On the other hand, the memory guys, where also there was a chart from Chris, they never will go down below 10 micron with bump pitch. So they are not keen on the accuracy.
They are, because they have so many dies, they have to stack and even more so with every generation, every new generation, they are much more interested in the cost of ownership and in the speed of the machine. And for that reason, we are following these two major roadmap line items, logic and memory. And the machine that Richard mentioned, where we will have a lead customer, where we will ship end of this year, the official launch of this machine will then be next year in 2026 for the broad audience. This machine, this is what we call Hybrid G2, that will cover both of the demands that we are seeing. So on one hand, this will be then accuracy-wise a 50-nanometer machine, which allows for high-volume manufacturing bump pitches down to 3 micron according to the roadmap of logic customers and leaders.
But on the other hand, it also allows to increase the speed up to 3,000 for the HBM world, where the accuracy is not that relevant, and you can do with a little bit less accuracy, and you can set up the machine in a way that you rather optimize it for speed instead of accuracy. So we have both in one machine, in one offering, and that is a perfect match in order to continue the success story that we have started with our first two generations. But even that is not enough because if we look at the memory, not the memory, the logic customers, they all say, "I want to go down to 1 micron bump pitch or even below 0.8." These are some numbers that you might hear around or read.
And for that reason, my pathfinding team is already working on alignment concepts for the generation thereafter for 25-nanometer because only then you really have a good setup for going to one-micron bump pitch or 0.8, something in that arena. So we are working already on that as well, which we'll then implement in the generation to come. And at the same time, because once again, logic is not the only, HBM memory will be one of also a huge market segment within that whole hybrid. We are also working on making the speed even better and going to 5,000. So that is all what we are doing right now, and well, not all.
That's all what we are, a huge effort that we are putting into hybrid because only if we as a machine supplier can offer machines with that qualities, the roadmap of our customers come true, and only then the adoption of hybrid is coming true. While we're talking about the adoption of hybrid, I want, before we go to our next chapter, which will be the markets, stop for one slide at a perception which for me is always very painful, actually. You have heard from Chris, actually, that, yeah, there's a lot of technical reasons. And they are all physically, chemically very sound and very good to explain. But then I'm hearing very often, "Yeah, but hybrid bonding is too expensive, so it will never make it to a real mass production. And this perception is simply wrong." And why is it wrong?
Because it's not only that you have the technical benefits. Once again, Chris mentioned it, it's the same table. I don't read it out to you a second time. But you need to think if you talk about too expensive, not only about the comparison, what is a hybrid bond versus a TC bond? Yeah, a hybrid bond itself for one chip is more expensive than a TC bond. That's true. By the way, it's not because of the die bonder. It's more because of the wafer preparation before. But anyways, our customers and end customers, they think in total cost of ownership.
If I can use two chips in that whole assembly, which are cheaper because they have a higher node or a cheaper node, and then the interconnect is a little bit more expensive, it still works out for them as a package as a total cost of ownership. If you take it even a step further, especially, for example, for the reason of the energy efficiency performance, and people are talking about building nuclear power plants next to an AI data center, you can imagine how much energy that is, how much money that is on running the data center. If you can make chips which for the same compute power simply take less energy, even if the chip is a little bit more expensive in the end, that's a huge cost saving. You need to think that whole from a total cost of ownership perspective.
And then you see that it really is a big benefit, and it also works out from a financial point of view. With that, let's go to a quick deep dive to a few application areas that we are covering. And let me give you a little bit of an update what, in our opinion, has happened and what has changed since our last year's meeting that we had. Let's start with the big driver, the AI logic. There, actually, we definitely can say that the interconnect AI use cases, they have clearly solidified. So on one hand, the usage of AI is accelerating, so all of us. I don't know who has ChatGPT open right now on his laptop here. I'm sure some have or other brands doing the same. So actually, the use cases are getting much more. That means it needs more data centers.
So the demand on chips is increasing. So that's one thing which is good for us. And once again, it should be energy efficient. That's the big topic of all conferences that I've visited the last year. It was all about energy efficiency, compute power. And then there was another trend, and Chris mentioned it already a bit, but I want to tell it a second time because that was really interesting. We all know the chip suppliers from whatever, you name it, AMD, Intel, NVIDIA for sure in that arena. But now, actually, the hyperscalers, the ones who are doing the data centers, they start to design their own chips because they want to have it tailor-made for their specific AI algorithms. And then they are using a Broadcom or a GlobalFoundries or somebody else to design their chips and perhaps TSMC to manufacture their chips.
The supply chain is getting much more complex. The interesting part is all the ones who are now starting their own designs, if you look at what they are doing right now, they are all using hybrid for the AI chips. There is not a single one not using it. And so that's also a good signal that everything is solidifying. And we are not the only ones seeing that because if you look at TSMC, who is by far the biggest manufacturer, actually, for SoIC in today's world, they have announced that in 2026 they are opening up in Taiwan a second factory for SoIC. And they for sure, with that huge investment of such a factory, they do it for very good reasons. So you see the demand is there. The designs are all going that direction, and the manufacturers are increasing their capacity.
So I think that's a good, all these three pillars showing a good, saying, a solidifying sequence in that market. And for that reason, in our own model, that's a detail that we have not shown last year. For that reason, we also have increased the forecast for the next five years by 20%. Now, no AI data center without HBM memory for sure. That goes along. I think most everything what is also written here, Chris said already, he gave you a much more complete overview when we think HBM is kicking in. So I have just here, once again for you, pulled out the, yeah, also our model. So what we see, so here we see a bigger growth rate, 33% in the next coming years, simply because also here the inflection point when HBM should be kicking in, it will be kicking in, is getting more solidified.
The ones who are following us since years might remember two years back, everybody was about, there was big excitement about, oh, HBM memory goes hybrid, and then last year there was a pushback, oh, everything is a year delayed, and these type of messages. What I can say for the last year, for the last 12 months, actually the programs at all the main manufacturers of HBM memory towards hybrid have solidified, and the R&D activities on their sides to make it work and get the processes right and prepare it for mass production have all been accelerated, so actually, I would be delighted to show you some cross-sections and samples here, which I unfortunately cannot do because we have NDAs, but I tell you, the 16 upstacks in HBM, they really look great for me as a technician, at least.
Then there is another topic, which once again is AI, and that's the co-packaged optics. Because in the data centers so far, the short interconnects were done with copper. But also here, because of the energy usage and the bandwidth that needs to be increased in between the individual compute blades in the data center, also that is going now to optical link. And that brought on the co-packaged optics, where then in the co-packaged optics package, there is a hybrid step involved. Because you don't want then the interconnect between the optical and the electrical area to be the bottleneck once again. So you need a high-speed interconnect there. And also here, if you look at the market, Chris already mentioned NVIDIA, the announcement of their Quantum-X and Spectrum-X.
But also Broadcom announced the new generation, the Tomahawk 6, where they also said that's, by the way, not optical only. So they have them both versions, copper and optical ones. But also there, they are jumping on that train right now. And once again, TSMC, as one of the big, big players in our world, they have developed the package type, which can be really manufactured on a high-quality level and then on an industrial scale, which they call COUPE. That's just their name for the product. And yeah, they are not only using that for NVIDIA, but they have that as a general offering. So it's very easy now, actually, for other companies who are also playing in that area to jump on the same wagon. And we talked about that, but we didn't have that in our model last year.
So we added that in, actually, for this year's model in order to, because it's there now. So it was a very fast turnaround development to get these co-packaged optics, actually, into real life. Enough about, let's say, really the data centers. Let's get a little bit more to the consumer side, where you have the PCs, the laptops, the smartphones. And they are actually also there with the edge AI applications. We definitely do see hybrid processors coming up. And by the way, don't forget, the very first hybrid architecture which was on the market was an AMD Ryzen processor, which was not an AI chip, which was a computer chip, so a PC chip, a gaming chip. So it's not really not totally new. And Chris mentioned already, there is a lot of announcements about the next M5 family that is being launched end of this year.
So it will be very interesting to see then with all the speculations around what is really coming up. But yeah, that's just perhaps four months ahead of us. And then we will see if they also do a split and have for the high end of their M5 family different technologies for interconnect and for the mid and the low end. Because that's what we are expecting in general, that this market will split really up in the high end, which will also use then the high-end technologies like hybrid, a mid section which will grow heavily in TC. And there will be for sure a very low-end section remaining, which will further use mass reflow flip chip like in the past. That will not go away. You always have also the trailing products, and they will use simply the old technology.
And that will stay for a long time. But for us, that meant in our model that we saw a slight uptick on our hybrid forecast. But in general, what we added here is actually the TC forecast, because especially in that arena, we see the benefit of using TC and still these small bump pitches between 10 and 20, 25 micrometer, that area that I have pointed out on my second slide. And so that's really a nice growing area for this technology. And by the way, watch out, Kirin. Kirin is also a processor family or an application processor family for the non-Apple world, a big one being used in whatever smartphones and tablets. So that's also something to watch out on the long run, what is happening there. And last on the consumer, I want to get to this very exciting area of AR and VR glasses.
I guess you all know they're existing already since a few years. These big and bulky glasses, perhaps you had the opportunity to test it somewhere already, where you have these AR and VR contents, and then after 10 minutes of testing, you're happy that you can take it off because they are so inconvenient that you don't want to wear it, and only in a few industrial areas, they are really used today, so that's not a broad market. However, companies like Google, Samsung, but also others, they didn't stop and said, well, we will make that happen, but we'll make the same technology in a format which is much more resembling our glasses, our optical glasses that a lot of us are wearing here, and you see the pictures, so they're getting closer and closer.
Really, there has also been a huge R&D boost in these companies in the last year. They were testing completely new assembly strategies and methodologies to get that done. Luckily, even so, everybody is doing it a little bit different. Everybody's trying a little bit of a different path. All of them either include a hybrid step or a fusion bonding step. That means we are really involved in a lot, even more than I've pointed out here on the slide, of companies developing then these light engines and the displays for these glasses, which is another area of growth for the future. Now, you could say when you look at the numbers here, well, 25 machines over five years, that's not an exciting number. Agree. But think of it that way.
We have been conservative because we have no idea on how this market develops. If that's really taking off, like the smartphones in the 2000s, that market can be way bigger than that, so that's really a conservative number from our side, and all these things in all these different areas, we are doing together with AMAT, where we have a very good and very long-lasting relationship and collaboration more than five years now. And it looks like it will be much more years than that, where actually together with AMAT, who is supplying all the preparation steps beforehand and a much broader offering than Besi has, together we can overcome all the difficulties for the customer that he has when he wants to implement hybrid from scratch, and that is for sure also helping significantly to help to accelerate the bonding, the adoption, and the growth of hybrid bonding.
And with that, actually, I come to my summary slides. And I'm glad that this year, first time, I can also show a slide, same style like you're used to, but not for hybrid bonding in this case, but also for TC bonding for our TC fluxless. So with the current model that we did the first time now this year for TC as well, for TC fluxless, we are coming, depending on we take low mid or high case, up to 350 to 600 machines. Once again, I think that's a repeat now. We see it mainly in memory and in logic, while perhaps the memory sales is more front-loaded in the near future and then getting a little bit less because hybrid is taking more of the portion in the later years.
While on the other hand, as I said, especially in the consumer logic, we see a big continuous growth rate in that arena. And that leads me to my final slide, which we see how we see the market potential for hybrid bonding. And if you compare it to our last year's model, what I said also on one of the other slides, we see simply all these cases solidifying. And for that reason, and we do the precise calculation, we can also see that our mid and our low case have increased by 7% if you compare it to our last year's slide. And once again, the low case because simply these logic applications are much more confirmed. And yeah, I don't read it out to you here. Once again, Chris said it, I said it already.
So it's AMD and Intel anyway, Broadcom coming on board, high-end CPUs coming on board, then the AI devices from the hyperscalers. So that's really a solid base. And then we can add to this low case, the memory. And in comparison to last year, we also added the co-packaged optics in here because we didn't have that in our mid case last year, which also helped to increase that. And it's not only helping, it's already there. NVIDIA is doing it. So I think nobody's doubting that. And that leaves then all the consumer things like the one, the last one that I showed you, the smart glasses, but it was just one example. For sure, there are more markets than the five that I've shown you here. And for the high case.
So all in all, we are very happy and we are very positive that our forecasts given last year are not going down. They are going even further up with the developments, and they are not going up. They are also solidifying if you look at the business cases behind that. And I think with that, not I think I am at the end, but before we go to Christoph, we have planned a quick break so that everybody can grab a coffee. I think it was planned for 15 minutes. So if we start once again at 2:25 P.M. Thank you.
So they say. Hello? Yes, it works. So please take a seat, ladies and gentlemen. We're only halfway, and the most important part is still to come. So, Christoph, stage is yours.
Hello again, welcome back, and good afternoon. I'm Christoph Scheiring, Senior Vice President, Die Attach.
With 20 years of experience in Besi, I am leading the mainstream Die Attach business. I have the pleasure today to walk you through our plans and the roadmap of the mainstream Die Attach business. Our market, the semiconductor market in general, is in exciting times, as you all know. Why is that? Simply because we are in a growing market across all segments. You can see that in the bottom chart. From computing to wireless communication, consumer, automotive, et cetera. Not only that, the growth is even accelerating going forward if we compare with the recent history. What that means to our addressable market can be seen on the chart on the top. From today, around EUR 700 million, this market is expected to more than double in the years to come till 2030.
With a CAGR, as you can see here, of about 15%, and even more so in the advanced portion of the market. Basically, the core of what Besi is doing, the growth, is even bigger than that. Where is that growth coming from? I think after all the presentations we have seen today already, it is clear there is only one answer to that question. It is artificial intelligence. Artificial intelligence for us means basically three growth vectors. The first one is very clear, imminent today. This is the data centers, the expansion in data centers because of all this training and also inference workloads. That at the end means a lot of electronic content from high performance computing to memory to photonics, all the way to power management ICs. That's one.
The second one is that this ever increasing need, in performance is driving advanced packaging. And again, advanced packaging, what Besi is all about. And the third one is the edge AI. So those are the devices, that come along with the usage of these large language models. And that create basically the entry point for the consumer, for industrial and applications to these, AI digital world. So now I have a quick look at the Besi portfolio, our current portfolio that you can see here. And it's indeed a leading, mainstream die attach equipment portfolio. And I start on the top left side with our high volume performer, the MMA machine, our 2200 Evo. So this machine has great, versatility, hence used in many different applications. The most relevant ones are the ones listed here.
At this moment, it is photonics and co-packaged optics because of the accuracy capabilities of that machine. The other one clearly since years already smartphones, but also the interposer attach, which requires large die flip chip capability and other leading process capabilities. And that's the core of that platform. Sorry, I go on with the next one here. This is the proven and, I should say, industry recognized flip chip leader, our flip chip product lineup where we have very high-end capabilities for chip to wafer, mainly on the accuracy side. We have highly productive solutions for substrate assembly from BGA, CSP, lead frame. And we have recently added a dedicated tailored solution also for this advanced camera market based on this platform in combination with the already available camera solutions on our MMA platform.
Next one on the list here is our most advanced epoxy die bonder. So that's really the die bonder of choice if it comes to accuracy and UPH. And our really strong point is all the applications that require stringent epoxy control, for instance, power and sensor applications in the mobile as well as in the automotive space. And finally, our leading edge soft solder equipment platform, which provides really unique soft solder process capabilities such as very low oxygen level, gas consumption. Peter mentioned it before for his machines. Also here that is a key criteria. And this is what our equipment differentiates from competitors. And also very important, we offer on that platform unique process capabilities called diffusion soldering. So a combination of soft solder and diffusion soldering.
Now, from that generic portfolio, I'd like to go a little bit into application details for some of the key applications we are supporting and explain here the Besi approach the market first, a bit the Besi approach and also the progress we made. I start with the advanced packaging market for obvious reasons. What we see is that this ever increasing push for performance, for efficiency and also footprint or miniaturization, that is creating a lot of momentum for advanced packaging. If you look at the Yole chart here, they describe really an explosion in the market about a factor of three of growth until 2030 of these 2.5D logic packages. These new package architectures on the other hand require capabilities, new flip chip capabilities to support those chiplets that need to be placed chiplets, chiplet structures as well as multi-chip structures.
What you see in the middle of the slide here is a cross-section of a typical HPC application. Chris was going into details here. What's a very common approach is that we see logic and HBM dies to be placed on the interposer, which we are doing with our 8800 Cameo platform. We have dedicated solutions there. In a subsequent step, that whole pre-assembled subsystem need then to be flipped and again placed onto the package substrate. This is one of the core applications we are supporting with our 2200 Evo Advanced because of its large flip chip, large die flip chip capability. You see here up to 110 by 110 millimeter, just one component. We work basically with all the key players in the industry. You see them listed here.
We are particularly successful in the recent months with the Chinese players. I can say, with JCET, SJ Semi , Huatian, TFME and others. Building on that success in China, we are now working, engaging with TSMC and working on a solution we call 8800 Cameo Flex, which is not only offering, and that's the reason why we call it Flex, a flexible approach in terms of material handling, but the ultimate, yeah, differentiator of this machine is the productivity itself. Chris was talking about a factor three to four. Remains to be seen whether we can prove that number, but the message here is that new machine is really pushing the limits of productivity and by far exceeding the incumbent solution TSMC is using. Another application or market I'd like to talk about is the photonics market.
Here, yeah, it's all about bandwidth and energy consumption that is driving the need for optical communication, especially in data centers. And if I talk about optical communication, the device behind is transceivers, optical transceivers. For those transceivers, you see here, a chart from LightCounting talking about the growth of those transceivers, which is tremendous in the next years. And this is all basically fueled or driven, by data centers, AI, driven data centers. Another nice data point I can show you is Besi internal data where we looked into specifically the number of die bonding steps required to assemble such a state-of-the-art, AI server, this NVIDIA NVL72 you see here.
The interesting point is that there are obviously some compute and memory, die attached steps behind such a device, but the number of die attached steps required to assemble the transceiver is even exceeding both the compute and memory space. Very interesting. Looking at the details of a transceiver now in the middle of the chart here, you see how such a transceiver looks like. A transceiver basically is a optoelectronic converter that translates electrical signals into photons and vice versa, right? Yeah, a typical format can be seen in the middle called SFP, small form-factor pluggable. That's really an industry standard, quite low cost, quite widely adopted and distributed meanwhile. Inside such a transceiver, you see a number of optical devices, a TOSA, a ROSA, receiver optical subassembly.
You find a laser, you find a photodiode, you find other optical elements. All those need to be assembled together accurately and in a way that the signal transmission is being optimized and good. That is where Besi equipment is the number one player, especially for the market where the accuracy goes down to three micrometer. This is what our Evo Advanced can do at this moment. We are working with all the key guys you can see there. We are at this moment engaged with some of them to further improve accuracy and not losing productivity. That's really the key and that's our strong point in order to address parts of that market that we can currently not address that are addressed by suppliers that use active alignment in order to achieve these higher accuracies.
But with that, they typically lose productivity on their equipment. And there is an area where we believe that with a higher accuracy, but the same productivity on the machine, we can grab part of that light green transceiver area, which is currently not addressed. Lastly, I would like to quickly talk about edge AI. So, edge AI devices basically create the link between the user and the digital world of AI, obviously. The verdict is not out there yet how the final edge AI device will look like. Three I'm listing here and those will certainly be ones that are relevant and we will be seeing in the future. Gen AI we talked about already. So those Gen AI phones are expected to kick in quickly.
The shift from normal smartphones into AI smartphones, smartphones that can run at least part of the model on the phone. Those phones are typically richer in content, have a more powerful AP, have more memory, have better cameras, important point for us. Also those cameras keep on developing further. We have qualified our machines now for the next step, which is mechanical aperture, which we will be seeing soon in high volume. And next steps like, yeah, stacked lens stacking, we expect to see anytime soon. Then, we go further on to cars. Cars are also making massive use of AI, AI content. This way they develop from, yeah, cars that have assistance, smart assistance, to really autonomous vehicles.
What that in the end means is that not only the electronic content of a car will increase, but also this electronic content will be pushed to very powerful advanced packages. The last one, AR, VR. Yeah, those devices will basically change how we interact with the digital world. What it means, you have seen also on one of the earlier slides. I think Peter had it in his presentation. They will be based on hardware that is highly integrated, that is very efficient in power consumption. That needs to be powerful as well in order to have intelligence running on the device. Again, that drives the need for advanced packaging.
Bottom line, AI, AI investments, AI growth can be translated into architectures, advanced packaging architectures that are really the core of what Besi equipment and especially die attach equipment can do already today and certainly also tomorrow. And if we look at a number of machines here for our Evo, it is the smartphone cameras where we are strong. It is power modules for automotive as well as for industrial applications. It is these AR, VR devices we are engaged. We go to our 2100 platform, which is very strong in these mobile PAs, mobile sensors, as I indicated already, as well as power management systems all the way to the 8800 platform, which is used today for a large variety of processors, everything from APs to GPUs, CPUs, et cetera, and even used in DRAM.
Important also to mention again this new development in the pipeline, an 8800 based CIS solution for smartphone, camera applications, which will boost the productivity of our current, smartphone assembly lines. In summary, AI clearly is driving, advanced packaging needs. That is obviously a lot, helping us and favoring the Besi equipment. Because of that, addressable market is expected to grow significantly. In our models, we expect more than a doubling until 2030. Besi products are favorably positioned in that growing market with some of the highest, growing segments like CoWoS, like photonics, like mobile, and, the, automotive EV and autonomous driving application.
In addition, we do have developments in the pipeline very near to launch 2025, 2026, which will help us to solidify our position in the market, accelerate the growth, and hopefully also increase the market share. That's all I had to present today. Thank you. And with this, I hand over to our CEO, Richard Blickman.
Thank you, thank you, Christoph. So let's see where we are. I can also see it here. Well, it's very helpful. What we have done in the past many, many years, we have always had the difficulty of making choices. Choices is the core of life. So instinctively, we focus on the winners, as has been said, communication, the computing, data, processing, and automotive has been our application fields forever.
If you simply digest what has just been presented to you in advanced packaging, in all the challenges ahead of us, shrinking chiplets in whatever way photonics, what choices should we make to simply increase the return on capital ever further? Because that is the only reason why you are here and me too. So we have understood many years ago that taking that, in a way with help from outside, that you test your strategy, including your customers, that that is extremely helpful. That's what has taken us to where we are today. So once again, we decided with all these variables, with TCB, with hybrid, with modules and name it, what choices should we make?
So in a process of 16 weeks with our senior management, in total about 40 people, our customers, and also some key stakeholders, which is also in the supply chain, we have rewritten our Strategy 2029 because we count in five years. And whether it's 29 or 30, it is five years down the road. But in any case, that's what we shared in the press release in terms of revenue, but also the margin potential, the cost structure. So that should give us some guidance in the choices we think are the right choices today. And that should support growth above the market expectation and position Besi in an ever stronger position going forward. But that's not all. The world moves ever faster.
So also anticipating on faster demand from customers, higher flexibility, further consolidation in the industry, all those elements are part of that test, which I simply call a test, which results in a new, and you've only seen the revenue, but we've detailed that down to every product, to the organization, what is required in resources and resources in capital. So just to share with you some of that background. If we look at the opportunities and the ones we selected and with the key customers, that shows a trend if that happens along the lines which are considered today in an enormous market opportunity. We expect, if we do that right, to outperform once again that market. You see in the box certain key messages which confirm that growth is what the world expects. That has led to an update of our model.
Going from the EUR 1 billion +++, which we have set ourselves a couple of years ago and confirmed in the addressable market, we're not far off the 40%, but that is what it should be. To repeat this for as long as I can, this market today, and you've seen that again in all of its applications, is not a market share game. It is a margin game because the margin potential offers you survival in the future and access to capital, cost of capital, and that is more important than only market share as long as this industry continues to grow in complexity. That should result then in higher margins, higher operating results. Of course, we've set ourselves the targets to work on a better world. The targets for zero emissions and also renewable sources. What are those key drivers?
If we split the old model we see here on the left, we see the new model, the EUR 1.5 billion-EUR 1.9 billion. There we list them, the hybrid bonding opportunity with the details as explained extensively by Peter, same for TC Next, and then the mainstream advanced two and a half D explained by Peter, by Christoph in every detail. What we haven't said yet is that the last one, the spares and service with the industry moving to ever smaller geometries and also partly into front end, the service requirements from customers are at a significant higher level than what we have experienced so far in the assembly world. Number one, 24, 7, also ongoing process support, and that only increases the value of the service and the support you deliver, so that's also a significant driver for higher returns.
Those key goals, gross margin also supported not only by product position, but further cost reduction. One of the biggest enemies in the world is cost. And Besi forever focused on cost, our supply chain. On purpose, we haven't spent much time on that in this presentation because we want to show you the market opportunities, but equally is the cost. If you look at this from the perspective that there is an enormous imbalance because of the cyclical demand for our equipment, that offers you an enormous opportunity to work constantly on better cost structures. And that was part of the strategic review as well. Taking that EUR 15 million-EUR 30 million cost out theoretically over the next years, partly by selecting different suppliers, also more focusing on inventory.
Inventory is one of the biggest enemies apart from cost, is because of the uncertain market outlook that you are working with a suboptimal inventory. And there's a lot to be gained. Then we look at the OpEx, of course, the investments in R&D and in service support, as explained in the earlier slide. And then operations expand our presence in Southeast Asia. As you may know, we have set up the first operation in Vietnam three years ago. It will be expanded in Vietnam. And also the next question is, what do we do in India? India is predicted to be the strongest growth area. We have five significant customers who are setting up their businesses locally in India, which offers also for the longer term significant growth opportunities.
The last but not least, the sustainability, as we mentioned before, the net zero and also the 100% global energy needs from renewable sources. Our liquidity position has always been strong. We have used the cyclicality to finance our business in the most optimum way in the beginning through equity. After that, since 2005 with convertibles and last year, first time a note. That with a simple baseline, 10% of revenue in net cash should enable us to fund our growth in the coming years based on our own balance sheet and maybe at some point increased with debt rather than issuing stock. At the same time, with that 10% level, anything above that 10% net cash from revenue, we distribute to shareholders. As you, some of you will definitely know, we have distributed EUR 2.2 billion in capital to shareholders since 2011.
And we will simply continue with this policy. That brings me to the last slide. Assembly market at major growth inflection points. I hope you're more than before you came here convinced that this is driving our business and offering enormous opportunities. Besi is at that forefront with the winners. And that should offer us again going forward also a broader application field. Hybrid bonding is certainly gaining enormous traction. What hasn't been said by my two colleagues, we have over 100 hybrid bonders installed at 16 customers, number 17 coming up. So that tells you the world is working on this technology ever more convincing. And at the same time, our TC Next anticipates to have exactly that offering, which completes not only the Mass Reflow TC and hybrid bonding, enabling customers to build the entire device architecture with our tools.
Strategic plan updates has positioned us with ever more confidence for the next five years. So long-term growth substantially above what the market expects and offering us also the higher returns, as explained earlier, due to somewhat higher margins and also our cost control. And that leads to attractive capital allocation policy maintained. So that's it for today. We now would like to offer you the opportunity to ask whatever questions you would like to ask. For that, my colleagues, Chris, Peter, Christoph, join me because I don't know all the answers and you certainly do. So I will try to organize these questions a little bit, but anyway, who has the first question?
Madeleine, you're first. Yes, thank you. I'm Madeleine Jenkins from UBS. Yeah, let me give you the microphone. No, no, don't shout. Hello.
Thank you. I'm Madeleine Jenkins UBS. My first question is just on the Gen 2 tool. What ASP are you expecting for that? And also just for it to clarify, is it 3000 UPH and 50 nanometer accuracy or is there kind of a relationship where you'd have less accuracy for more throughput? And then I have a second, if that's okay.
Actually, that's a relationship because you don't need both at the same time. So it's, as you said, it's a trade-off in between, but with that, you can focus on both areas with one machine. And that's good. Yeah, I think the ASP will go up another 20% or so because, yeah, high accuracy is for sure much increased effort in the machine building.
Okay, thank you. And then my second, you mentioned the new TSMC Fab next year. What sort of kind of total capacity would that be and when are you expecting orders for it and also what applications you're expecting it to use? Thank you.
The Fab will come online according to the TSMC plans in Q1. So that means order intake is already considered for this year because of the lead time of the machines, obviously. And sorry, what did I miss now?
What's kind of the total capacity and also what applications do you expect that to be for?
Well, I'm not sure if I can. I think what the capacity is, you need to reference to TSMC statements. I definitely can tell that it is bigger than the first one that they have fully occupied now.
And for the application, it's simply their extension, you know, so that's on one hand products that they are doing, which are getting more run rate on one side, but also you heard me saying that there is a lot of new products in development and Chris mentioned, for example, a Broadcom device, which was officially announced. So it's a mix of more volume of the same, but also new products adding.
Okay, thank you.
Thank you. No, thanks very much. Maybe a question for Richard, just taking a step back. I think the message today is to tell us, look, HB is real. It's a bit bigger than we thought by 2030, but we also had the race is in TCB, which was not the case until now. Is that the first question, is that the right way to think about it?
Related to that, when I think about 2025 and 2026, right now we are in a downturn when it comes to flip chip because of smartphone pieces, et cetera, automotive. But obviously you are telling us that things are going to turn. I guess my question is, when do we start to see the benefits of this better strategic position in TCB and in hybrid bonding? Is that a second half story? Is that early next year? I have got a follow-up.
Thanks. First question. Let me answer it in this way. I think we shared that also in the last updates. We were asked eight, nine years ago to engage in the development of hybrid bonding. In Taiwan, our key customer explained that this will be needed at whatever point at bond pad pitches and technology. We had to make the choice.
So again, choices: do you engage in that? Do you see that realistic? Or do you hold off? Because probably it will be later, like many things in the industry. Based on the information we had at that time, we deliberately chose to engage on hybrid bonding first. We had already a significant position in mass reflow flip chip. We were at that time already recognized as certainly the leader. And TC, we at that time, we started with TC far longer ago in 2012, stacking memories already at that time. So we thought it's better to be two to four generations ahead first and secure that. And at the time that you have a further stretch in reflow, that there will be opportunities in between the final transition point where the industry is forced to go from a reflow process to a hybrid bonding process.
This was a very deliberate choice and a roadmap we have seen forever. You could argue, as Peter also said, that last year with JEDEC standards allowing a higher stack of memories that stretched that into a 16-layer reflow device. But then again, as also is explained, the industry is very hard working on what to do reflow and what to do with hybrid bonding. There's a battle going on between the three, especially the two in Korea. What is the advantage? And Chris had it in his model and Peter as well. I invite you to look at that. From a cost point of view, there's definitely an advantage switching to hybrid. But anyway, the complete offering gives us, our customers, the flexibility to use the different technologies in every application in the optimum way.
So for us, it unfolds as it was expected at the time we made the decision to engage in hybrid. Your second question, are we in at this moment, is the tide turning? As I said in the beginning, this is not an update of our market, but I'm, yeah, let's say everyone reads it every single day. There's clearly the conviction that the worst of the cycle is behind us. We see announcements from several directions that should turn. If it does, according to what TechInsights expects, 2026 should be a significant growth year again. It also depends, of course, on what the world is reorganizing right now, what the impact of that will be. But the message is, Besi is more than ready for that. We have the right products. We have the right organization structure.
Our operating model is ever more, let's say, able to adjust to increasing demand. What gives me the confidence, if you look at the margin structure today, where revenue is certainly 30% below what we had in peaks in 2021, that despite a 30%-40% drop in revenue, still the margins go up. Despite the inflation, that means that we are well prepared for the next round.
Thank you. That was a long answer. Just the second question.
So it's a long question.
My follow-up is on HBM. Do you think 2026 is the year where you start to get revenues from HBM in a meaningful way?
Well, as it looks right now, there's definitely the possibility. There are two major customers working on that. So yeah, that looks very promising. Thank you. Thanks, DDA. Thank you. Please.
Yeah, thanks. Sandeep Deshpande at JPMorgan.
Quickly on your new model that you presented on hybrid bonding, is the market share assumptions that you're making for the whatever future period that you're talking about, is it the same as that overall company 40% market share, or is in the hybrid bonding specifically that your market share overall will be higher? And associated with that, I just want to quickly talk about the competitive environment in this space, given that some of the customers that you are targeting tend to use some local suppliers, which you're never ever sure whether exactly where they are with using the leading supplier, which is yourself, versus the local supplier. So how do you see that playing out as such?
Well, excellent. Your first assessment is correct. This overall share, which we didn't detail, but the 40% is an average.
We have on hybrid bonding. We have high up 80%, 90%, whatever. And then we have with certain packaging. We are at about 20%. Plating again up in the 70%. The average about 40%. But anyway, we expect that. But I said that also very clearly. It is not a market share game. It is definitely a gross margin, which tells you the value of your product to the market. So that is what drives our business. Your second question helped me again a little bit.
So my second question was on the competitive environment.
Yeah, the competitive environment. Of course, in technology, which is expected to grow significantly, you will have everyone anxious to be part of that above-average growth in a sector. And we recognize that many of our, let's say, die attached competitors who built beautiful machines also have discovered that this hybrid bonding is a unique opportunity.
And then you have the local for local. So in Korea, as you're probably referring to, but also in Japan. What was very special in Japan, we announced that we had that order from a major Japanese company in Q4, which was for us a surprise because we had expected that one of the Japanese competitors would be more likely the choice, but in the end, we were chosen. We've had in the history in Korea, and that goes back 30 years, many times that for new technologies, we were engaged, and at some point, it was shared simply because the dependency on a supplier from Europe is seen as a risk. But that's what it is. And then you have Asian competitors. You have China in particular, and also strategy in China to buy from Chinese evermore. But that has also not been different in the past.
And that's the beauty about technology. So if you have something to offer, which offers a higher reliability, a lower cost of ownership, and the world is small, our world is very concise. So you have that opportunity each time. But how it will be in the next five years is, of course, a wonderful challenge. But it's not one who can conquer the world in that sense.
Just following up on that competitive environment, because hybrid bonding is more like a front-end technology than the typical packaging back-end technologies, given your early leadership, does that give you an advantage? Because in typical front-end technologies, once you've conquered the market, you've conquered the market.
Yeah, what should be definitely clear here, and that's enormously supported by our partnership with Applied Materials. The Center of Excellence in Singapore from Applied has offered us jointly the industry offering to the industry process development from the very start. That experience is vital and exactly to how you state that, to engage customers because at the final stage, and especially these 2.5D modules, whether from NVIDIA or somebody else, they sell for $50,000 and more $80,000 a piece. Any mistake is 100% loss. That experience is enormous, but you should never expect that that gives you a ticket for the rest of your life because technology changes. Each time it's again a challenge. The basic foundation is unique, I would say, compared to anyone else in this space. We're working hard on that.
The commitment of us to Applied and Applied to the industry in advanced packaging is enormous. So that should give us an excellent position for the next round again. Thank you. Next.
Yes, thank you. Ruben Devos from Kepler Cheuvreux. I had one question around actually on pricing. I think you've talked about the market forecast, 350 units, EUR 1.2 billion market size for around EUR 3.5 million ASP. That's for the market, of course. And I'm curious about how Besi looks at it relative to that market average. I'm not looking for a quantification, but rather I was interested in your tools, obviously the Gen 2 is first, but then if you deliver on the full roadmap until 2030 with 25 nanometer placement accuracy and the 5,000 units per hour. Yeah, how do you think about, you said it's a trade-off, right?
So how versatile will your tools be, not only looking at, let's say, the logic side and memory, but then also consumer logic and CPO? How should we think about the price differentiations across these applications?
Well, it's a little bit more complex, actually, because as I've shown you, in the next generation, it still will be one machine with, yeah, the capabilities as explained. But we think that in the generation thereafter, it will split up because it doesn't make sense, technically speaking, to have then a 25 nanometer alignment accuracy, which is really difficult to achieve. And it's definitely getting more expensive, and these machines will get more expensive for sure on the ASP. And then to cross-utilize that for a segment which is more focused on cost of ownership and not on accuracy.
So we think that after that, that will split up in two models, which are then really streamlined towards the respective end market. And for sure, as these end markets are both using hybrid but are very different in character, we definitely also do expect then also that the, yeah, the ASPs are drifting apart significantly, actually. So yeah, for sure, we do some modeling around that, but definitely that's not an easy answer to say, yeah, that's the way it's going. But think of the EUV and everything and the history of that one and look at that pricing development that they had. Every step, they got more accurate.
Okay, that's very helpful. Thanks. And then just could you remind me? I think you talked about sort of the preparation stages being more costly potentially than the actual bonder, given your partnership with Applied Materials.
Could you remind us, when the order comes in, how that value is split?
Well, I cannot give you a value split, but I can explain. It's not that, let's say, the preparation itself. It's simply totally different because applying bumps, whatever style and methodologies that are being used is simply different from the whole process step. It needs different types of process steps compared to preparing a wafer that it's perfectly flat, that the roughness of the surface does not exceed a certain limit, plus a lot of other aspects that you need to consider. It's not about that an individual step is more expensive. It's simply different steps. If you add all these different steps up in the preparation, bumping versus preparing a flat wafer, yeah, then the flat wafer is more expensive in the end. That's what was my message here, actually.
And it doesn't matter. It's not somebody, another brand or whatever, cheaper. So it's simply different.
Okay. All right. And then just my second question was on the mobile opportunity for hybrid bonding. I had the sense that throughout the presentation, you were quite bullish on that opportunity. I think you've now sort of captured it in the high-end case. Just curious what needs to come together to make you maybe a bit more optimistic, such that it would land within the mid case. Like what are some of the factors in the market that you need to see materialize?
Well, the answer is very simple. If you look at the roadmaps of all these different segments and applications and also the one that you are pointing out, that's the one which is in the whole time scale, the one which is further to the back of the time scale.
So that means the projects are not as mature as of today. It's easier that somebody makes a change, that things are shifting around. So it's simply the visibility is not as good as for many of the other projects. And so if then the question is, yeah, what makes us more better visibility? And that will come as time is passing by. So it's not a matter of that we are not convinced that this is coming. It's just from a timely manner to predict that with more precision. It's quite difficult today because once again, it's one of the later aspects that are coming online.
All right. Thanks a lot.
Thanks, Marc Hesselink. First question on chiplets. It's clearly a big opportunity. And you also stressed in the, I think it was on slide 16, a CAGR for the market even beyond the 2030 period of 48%.
How can you link that to your addressable market growth with chiplets? I mean, it can be really a multiple of bonding steps. What kind of visibility do you have on how many different chiplets you will use in one package in the end and how many bonding steps will be addressable for you?
Yeah, this is something that we in our strategy development that Richard described, dug into a lot of detail on by application, by customer, by device type. And it varies quite a lot. So for example, today, some of the products that I showed you from AMD have an excess of 30 or 40 bonding steps for the most advanced ones. Whereas the application processor that we just discussed, which is not yet using chiplets, will use a much lower number, maybe somewhere between two to five, depending on how the chiplet architecture is implemented.
So we do have quite some visibility. We haven't disclosed all that detail here, but we use that in order to develop our models that we shared with you.
And that visibility is based on your client. Your client shared that information with you.
Yes, absolutely. We have roadmap discussions with not only the folks that buy machines from us, but also the end users who are designing these chips. Yeah.
Okay, thanks. And the second question is on the TC fluxless. Clearly, I think a big opportunity. Also I shared a lot more detail on that. Linked to that, the order that you earlier disclosed, the EUR 20 million order, how crucial was that? Was that the breakthrough volume order for a key client? Or how do you link that to your much more positive stance on TC right now?
Well, it was definitely an important order.
That's also the reason why we did a press announcement around it. When you have a new machine, it always starts the same way. Customers first want to evaluate the machine, and you're selling R&D machines typically because customers want to develop their package. And for sure, we are doing that as well. But there is then always one lead customer who gets it started for high volume manufacturing. And that's for sure always in the introduction phase of a new machine or a new machine generation, a key ingredient to your success that you get the ball rolling.
But also for you, that was the key proof that indeed you took that step.
Okay. Thanks.
Thank you.
Martin Marandon from Oddo BHF. The first question in the logic segment, maybe could you help us understand how correlated is hybrid bonding adoption and the node transition in that segment?
Notably, I'm thinking about the transition to 1.6 nanometer chips because TSMC is talking about potentially bonding 1.6 nanometer chips on 2 nanometer chips. And do you see this as kind of the next inflection point for logic? And in that case, what do you think would be the application for this?
Yeah, I mean, clearly, I showed you a chart that showed the increase in wafer price with each advanced node. The 1.6 was not on that chart, but that's probably somewhere in the range of $45,000 per wafer from what we hear. So simply as we go up that curve in terms of cost per wafer for each advanced node, the motivation to adopt chiplets and hybrid bonding will increase. So in that sense, yes, there's a correlation.
But that doesn't mean that people won't use hybrid bonding also for the applications that don't require that most advanced logic. And exactly what is the attach rate of hybrid bonding for the 1.6 node? Difficult to say, but definitely increasing as the wafer cost increases with these advanced nodes.
Okay, thank you very much. And the next one is on TCB fluxless for memory. I mean, you had orders for five systems in Q2. I'm just wondering what does this mean because five systems is more than testing. So does it mean that this customer is more likely to go through the TCB route for HBM4E or not necessarily?
Not necessarily because currently all the guys, they are developing in parallel for 4E already, the pathfinding groups on five because they have so the R&D times is only one year, so they could not manage to develop in it.
So they have all these different lined up in parallel. And yeah, typically what they're also doing in these days is that they are on some generation simply in the R&D work. They work on both. One group is getting saying, "You show me how you do it on a TC machine and you show me how you do it on a hybrid machine." So actually, they are not putting all their eggs in one basket, actually. So there's not a clear answer to your question, as they are doing everything in parallel.
Okay, thank you very much.
Thank you.
Thank you very much. Christoph, I've got a question for you. I haven't been able to look back at the slides, but smartphone seems to be a big driver of your expectations.
Now, we've seen strong smartphone growth maybe 10 years ago, and more recently it was about adding more cameras benefiting your CIS business. So what gives you the confidence that smartphone as an end market will continue to be growing or exciting?
I mean, to be very clear, the absolute number of smartphones is growing, but not at a significant rate, right? What we see is two things, or it all boils down to these Gen AI phones I have shown in my slides. It's basically that the richness, the features, the functionalities of an average smartphone will increase, and that, for instance, we see on the camera side where we continue to see developments in functionality to get these units, these smartphone cameras closer to SLR.
I have indicated the apertures, the mechanical apertures that will be replacing the digital function, which you already have available, but is not delivering the quality level, the anticipated quality level. That is a function we are seeing near term. We do see beyond that next steps, next additional requirements that will come. You have to see it, the overall number of smartphones might stay at a certain level or slightly increasing only. If you are able to introduce a new technology, that basically means at that point when it's being introduced, that brings you from around zero to not 100% because it always goes in steps first from the high-end or starting at the very high-end model and then deployed further down.
But those are steps we still see, and we still believe in that smartphone market to be a significant driver for us, yeah?
I mean, would it be fair to say that it's more of a replacement cycle of your tools that's now needed rather than more features being added? Because I haven't seen any for a couple of years, and it seems that the direction is more towards AI CPU application processor capabilities.
So I would not argue it's a replacement cycle of tools. Our tools are being upgraded regularly for the latest and greatest processes. And although you do not see much as an end user on the camera functionality on your smartphone, there is a lot going on in the background with ever-growing CMOS sensors, adding additional lenses.
So those machines are all utilized, are continuously being upgraded, and new functionalities that come up will require new installations. That is what we see.
Got it. And maybe a question for the other two gentlemen on HBM. I think helpful, Peter, you're saying we're working on their NDA with customers and also the TSMC paper by you, Chris. But just during the presentation, so much questions incoming in terms of it's still pathfinding towards 20-layer HBM. So what happens if the JEDEC standard gets relaxed again? What happens if 20 high is not needed because we just go to 16 and add additional real estate around the GPUs or XPUs? Or, for example, introduce instead of a 20 high, a 2x10 type contraption? Just in terms of how you've sort of countered those pushbacks and also how you've sort of translated that in your modeling.
Because I do think out of the 500 AI, the bulk is related to HBM, is that correct? The 500 cumulative for AI that you presented? Yeah. I thought that included HBM.
No, actually, I was showing two slides. One was for HBM and one was for AI logic.
Got it. Okay,
so that was differentiated in my slides.
Okay.
I think it would be good to have actually our presentation from last year because there we had a little bit more on the HBM side, a little bit more about really technology backgrounds, and that would be fitting perfectly to your question because the JEDEC standard alone is not the criteria, and Chris mentioned that also the data that JEDEC standard, the height is one aspect, but it's also the aspect on how much bandwidth do you get, whatever the stack height is actually.
How fast can you read and write the data of the memory? That's one thing. There definitely also you get the benefit from hybrid bonding because the electric interconnect is quality-wise a better one. For that reason, it allows you to run the memory stack on a higher bandwidth. The other aspect that I have mentioned, I guess Chris as well, is simply you're getting more and more heat problems. Also there, because you don't have this layer that you normally have on every TC, no matter which TC technology you're using, you have always the layer in between the two dies, which has to be filled with something. Otherwise, you have issues with the reliability. That's also at the same time a kind of a thermal isolation layer. That's really not helping.
For that reason also there, if you can put chip on chip, you get these 20% better thermal performance that I had in one of my slides actually. Also Chris had it. There is enough really other technical reasons which are simply about the performance and the reliability and the cooling problematic that you're getting more and more in all these AI centers which are still driving towards hybrid. In one word, yeah, sure. In one word before I hand over to Chris, I'm not afraid of the scenario that you are asking.
Yeah, I agree with all that. I just wanted to add a couple of points. On the JEDEC height spec, in my view, it's unlikely that they'll be able to increase it any further.
Basically, what they did in the first step is they had a safety margin in the height spec related to the difference in height between the logic chip that sits next to the memory and the memory itself. So that after they grind the surface of the whole assembly, after they put them next to each other on the substrate, they can create a perfectly flat surface there, which they can then mount their complicated thermal solution on top of. And what they did is they increased now to basically the same height as a full-thickness piece of silicon. And it seems very difficult to go any further than that unless you completely change the architecture of how you build the package and the thermal solution. So in my view, I'm not so worried about that one.
Peter hit on the thermal topic that we also talked about, but on the side-by-side case, think about what you're doing. You're taking whatever it is, 4 nanometer, 3 nanometer TSMC die. You're now making it twice as big and twice as expensive. And you're mounting memory on top of that. You're going to want to maximize what you can do with that chip in two ways. One is put more compute power in that chip, which is increasing the heat dissipation, which drives the need for the thermal performance. And then also, I'm skeptical that you'll want to really reduce the number of memory chips that you stack on top of that very expensive piece of logic. So of course, that could happen. And if it does, then people will buy TCB machines from us, yeah.
Thank you.
Yeah, Rob from Deutsche Bank.
Just two questions, if I may. On HBM adoption of hybrid bonding, should we expect a phase of pilot production orders of like 10-15 units, or will there be suddenly orders of 50+? Just how we anticipate how this rollout should look?
Well, that would be the perfect question to Samsung and SK Hynix what they are planning there. So in the end, what you are asking is depending on which types of their whole portfolio. It's not one HBM. They have a whole portfolio of different versions of HBMs.
If they decide to go for, yeah, let's say one version of it with a low volume first just to test it and make sure everything is working, or if they are more bold because they might be also pressured from their own market position that they want to make a leap in order to outperform their competition because that's also a big topic in that arena, as you know. And they might take a more powerful decision and say, "Well, let's go all in." And then it's perhaps even the number 50 is too low. That's really a question that, yeah, is hard to answer at this point in time if you are not a Samsung manager or a Hynix manager.
Okay, great. And then can you just remind us of your monthly capacity in hybrid bonding and TCB?
I think in the past, if I remember, you talked about going to 15-20 units a month. Have you revised that capacity plan for both TCB and hybrid bonding? Thank you.
Actually, we have taken all kinds of measures in our production facility in Malaysia that we are capable to increase or double and triple that number in a timeframe of one quarter maximum. And so we have, for example, made space available, which is already empty and waiting, so to speak, for that. So we took that place, took all kinds of other measures. So we are more, our mindset is always more that we need to be prepared when it comes in the right way. So all these preparation measures have been taken.
Henrik Munk Nielsen, 2Xideas. If we think about your new target model relative to last year, have you changed in absolute numbers your thinking of your more trailing edge part of your portfolio?
Once again, what changed? Just a question.
Have you changed your expectations towards the contribution from your trailing edge equipment?
We have left, we have assumed the, and we don't call it trailing edge, we call it the engine one, because trailing would mean that you would have maybe lower margin, but that's not the case. We have left at more or less EUR 1 billion. So remember, with the old model, it was EUR 1 billion+++ , and the increase is in the hybrid bonding and in the TC, so your question is an excellent question. What we also should have said is that we update this model every year.
So looking forward to 2030 is a long way. Probably if the tide turns next year, you look differently and also the independent research companies look differently at the industry. But we try always to be conservative in that sense. Of course, in the new technologies, and we've said that it depends a lot on, yeah, and the previous questions hit the nail on the head. What happens if things delay? Yeah, we also have the reflow technologies. So there's no exact science in these forecasts. But a longer answer to your question, the conventional part does not maintain that significant growth.
Okay, then just a quick follow-up on ASPs. I think I understood that you said ASPs would go up by 20%. That seems a little bit low to me given the productivity gains you get as well as the accuracy gains.
I.e., you're giving away some of the value added or did I misunderstand?
We don't give anything away. If we would be a company giving anything away, you would not see the margins that we are achieving. And you have seen our margin targets have increased. So we are well aligned with our margin targets.
Yeah, quick couple of follow-ups, easy questions. So on the new, I think it's AP7, the new TSMC advanced packaging fab, do you think that fab would be big enough to accommodate also smartphones or is that another fab after that? And then secondly, on the ASPs and margins, just to follow up to the previous question, TC Next, I think on the press release, we were all a bit surprised by the ASPs. Is there a service element in there? And all right, so it is what it is.
So should we assume then therefore TC Next is even more profitable than hybrid bonding?
Well, what most people underestimate when they think about apps for high-end thermal compression machines is that really you have to make a real big step in terms of machine design and machine building when you want to have a thermal compression machine which has an accuracy below one micron. Because as I mentioned it in a side sentence, because you have to apply temperature and force at the same time, and then achieving at the same time the accuracy is really physically a very difficult challenge.
That means also if you go, let's say, from a traditional TC machine, which is perhaps more built to support, let's say, a 40 or 35 micron bump pitch, and you go down to the area where we want to play, the machine gets really disproportionately more complex, and for that reason also more expensive actually. So don't compare machines in that range, in that TC range with prices of conventional TC machines. You have to compare them much more with prices of hybrid machines actually.
It's even more expensive by the looks of it.
Yeah, but that's also due to the fact that actually, and well, I did not point that out, it's too much of a detail actually normally for a presentation like that, that this machine that you have seen also on our picture is a kind of a dual machine.
You can have a single and a dual machine. For high volume manufacturing, you are selling a dual machine. It is two cells connected together, which is then also driving the price up for sure. But for R&D purposes, you would only buy a single cell, for example. That also makes a difference. It is really a lot in the details. I think the first question was about 87 actually.
Is 87 really for logic devices or AI and networking and things like that? Or do you think that there is enough space to also accommodate smartphones?
Well, once again, I am not the planner of TSMC, unfortunately. But if I look at the end market and what is coming up on the AI logic side, then my personal guess would be that they can easily fill it without any smartphone applications.
Got it. Maybe last question. On the gross margin uplift, so obviously you are a very frugal company. I think Richard likes it that way, so I'm amazed that you're finding a few cost savings in COGS. She's great. But are the gross margin uplift? Is that a function of the mix, e.g., TC Next, more hybrid bonding? Or can you give us a sense of the bridge with cost savings versus mix? And revenues, of course.
As we said, it's always about choices. Where do you engage, so at that forefront, are potentially always the highest margins. Number two, which hasn't been said, the development of our TC Next has also taken seven years, so we started with, on the horizon, this below 20-micron bond pad pitch, down to below 10, and Peter explained it very quickly.
But if you look in detail, so next year, you're most welcome also in between, by the way, to have a look at these machines. That complexity, together with the application, the accuracies we achieve, plus the fluxless process, and for the gases you need, absolutely a minimum consumption compared to the rest of the world, which is very difficult. Yeah, that offers you the higher margins. And the industry will definitely move in that direction. There's no roadmap which says, "Well, we're finished."
Perfect. Just looking for those orders to come in.
Yeah, we too. We'll celebrate together. Next question behind you.
Marta Bruska, 2Xideas. Firstly, you just mentioned that the outlook, the market outlook excluding TCB and hybrid bonding has not changed, remained more or less EUR 1 billion. So the upgrade has to do with TCB and hybrid bonding. But that more or less implies that you expect that market to double in size. If I look at the outlook statement you provided for hybrid bonding, you also increased the low end and the midpoint with 7%. So I'm a bit puzzled. I can't reconcile that. Could you give some further explanation?
You shouldn't be puzzled. If you look in the beautiful slide going back to 2006, the typical cycle offers you doubling your revenue from your lowest point. Yeah, so if you make that analysis, so the conventional business today, which is EUR 600 million minus the hybrid bonding revenue, which you can calculate on the back of a napkin, and if you subtract that from the EUR 600 million and you double that amount, you're exactly at the EUR 1 billion. But that is a huge step. We can say, "Well, we've done this many times." But also take 2019, a nice example.
Our revenue in 2019 was EUR 356 million, 356. And in the peak of the cycle 2021, it was EUR 749 million, 749. So double that. And I can go back to previous cycles in a similar way. But this time, I hope I'm right, if you look at the margins today, it is the entire product range, so also for conventional. Otherwise, you never reach mid-60s. So in a downturn period where revenue is more than 30% below the peak, that should give you the opportunity to grow in that next cycle in a similar way as what we have done in previous cycles. And we have tested that with looking at all of the opportunities in a 16-week process.
But your question is a perfect question, which is also on our mind because then you have to be prepared for that next round so you can put a nice number out there. But can you build those machines and can you deliver in time? So do you have the infrastructure? Bricks and mortar is easy. But your supply chain, do you have your supply chain? Because once the industry turns, it turns for everyone. So you have to test that very carefully. Then are you able to use those suppliers to the need you have? And our philosophy has been for many, many years dual sourcing and sometimes triple sourcing. But that all determines whether you are able to double that from the downturn trough levels. I hope this answers your question.
Not completely because I agree with the market excluding TCB and HP that it will double.
But if you just look at, you've increased your revenue from EUR 1.3 billion more or less to midpoint EUR 1.7 billion, the billion remains the same. However, the market for TCB and hybrid bonding is increased by EUR 300 million to, let's say, EUR 700 million. But the outlook provided for hybrid bonding was only increased by in volume cumulative by 7% for the low and midpoint. And to double, so in contrast to a doubling revenue and only 7% increase of the market, that's a gap I can't understand.
Now I think I got where your question's coming from. You have to consider two things, and one question was around that already. This is an outlook for the next five years, and the next five years will also change the models significantly, showing the roadmap. And that also changes the ASP.
Don't underestimate the effect of increasing ASPs because that also gives you a hint why this is going and why that's happening. And also don't forget one aspect which has nothing to do with the hybrid, but also with the TC that we have added on our modeling actually that we did not have in the last year. That's also bringing a portion, I think, of the gap that you are seeing in your mind.
Yeah, hi, Shahar Cohen and Lucid Capital. Trying to understand better the GenAI opportunity. What is assumed in your mid or high end in terms of GenAI penetration, in terms of phone kind of ballpark? And whether do you see a clear line of sight to that? Because let's assume, I don't know, 20% will be 2027, the iPhone 2027, they need to start preparing for that.
So, I think you should have a clear line of sight for that. Is that the case in the GenAI phones?
You're talking about the GenAI phones? Exactly. Okay. The GenAI phones, really the phones are considered in, if you talk about the hybrid portion of that, okay, only not other technologies, then the phones are included not in the mid case and also not in the low case. They are considered in the high case actually. And the high case we did leave unchanged compared to last year. And the reason, once again, there was already another question that, yeah, the visibility is simply at this point in time still difficult for this type of GenAI application process or forecast. We have something because we are convinced it's coming.
But to make a big revision of that based on the information that we have is, yeah, it's not the right thing to do.
What's the percentage of penetration you assumed in the high end? Like 10% of the phones, 20% ballpark?
Just need to think actually. It's a very detailed question. I don't have the numbers off the top of my head. Yeah, I think currently we have a rather conservative number, which is in the ballpark that you have been mentioning.
All right. And another question about the operational leverage. So fantastic job in maintaining cost. Let's say the rosy scenario happening. What kind of leverage should we expect in the operating? Let's say you increase the 30% of 50% revenue. What's the operating leverage you will have to maintain?
Look at the model because it specified the operating income in the model. You see, it's going to 50, which is. But we did reach 44 in the last big cycle.
Thank you.
Next.
Thanks for taking my questions. Nabeel Aziz is from Redburn Atlantic. I just had one on co-packaged optics. So could you just help us with the shaping of the adoption of hybrid bonding into co-packaged optics? Because NVIDIA talked about shipping Spectrum-X next year with co-packaged optics. So what's the sort of shape of hybrid bonding adoption that we should think about there? Thanks.
Yeah, I think that's
what you want to address.
Yeah. Yeah, that's exactly right. So NVIDIA is the first adopter with their Spectrum-X and one other network switch device. And you're correct. They announced that they'll start volume production in 2026. So that'll be the first production. I think in the presentation, we show a ramp profile. And that's basically our view.
So it's going to start out relatively small. But I think in Peter's section, he also showed a kind of a cumulative estimate for the number of machines that we anticipate will be required through 2030, basically following the profile that we saw there. So it'll start with the network switch devices. I think in one instance, they have 36 of these hybrid bonded chiplets per switch, and another instance or 16. But later on, before 2030, definitely we'll see also these hybrid bonded chiplets sitting next to the GPU as well on the same substrate there. So that'll increase the demand as well when that happens. Yeah.
I'm afraid that brings us to the, oh no, one more question. Now, if there's more than one, we have to start all over again because then our presentation was just one.
Maybe you talked a lot about servers and smartphones as well. I was wondering about the opportunity in PCs. Where do we stand in the adoption cycle? And how big do you think is the opportunity for PCs?
Yeah. That's definitely in PCs. It's definitely a good opportunity. And there the adoption cycle, we are seeing much more clearer than on the smartphone APs. And I think it was stated in one of my slides. Perhaps I didn't spell it out, but we see that the adoption cycle for high-end computing, high-end computers will start actually end of this year.
Okay, thank you very much.
And let's not forget AMD with its very first product. They started it already. Perhaps not the biggest volume with the gaming computer, but you could also argue it has started already. But if we argue that way, then I would say it's getting more traction end of this year.
Yeah, just a clarification. So I'm just looking, we don't have them, but I've got some people looking at them for me. So it looks like looking at the chart of 2024 versus 2025 of hybrid bonding, base case, bull case, looks like you pushed out 2025, 2026 for hybrid bonding. Is that fair? And it looks like because you've got TCB that wasn't in there before, it's been replaced by TCB. But I need a bit of a magnifier just to be clear. Is that the idea?
No, that's not the idea actually. For sure, we did update our model and some things are pushing out and some things are moving in. So for example, the co-packaged optics business was moving in. We took the best assumptions that we knew, for example, when the volume of the HBM is coming in for the hybrid now in that case, but also then for the TC. So yeah, things were shifting around and it gives a little bit of a new shape.
It looks like the new chart is like more parabolic or is the other one was a bit more like.
I think you should confirm that the adoption continued above expectation.
Right. Well. I'm trying to catch you out.
You can always. No, but I said that at some point.
If still you see more customers buying and engaging in hybrid bonding, then that means that at some point, because they all make their assessment of performance and cost, etc., and also the activity in the center of excellence, as long as that all is increasing, in this stage, it is not a matter how many bonders you sell. Of course, in terms of looking at earnings per share, etc., but the key is that how solid is that adoption? And I think it's fair to say that the adoption has continued. We would have wished maybe more bonders sold, maybe also the inroad for high-end smartphones. But that debate goes back, I think, two, three years. We have always said 27. Many have thought this to be earlier. One of them, Qualcomm, they all have pointed to earlier adoption, which has not happened. So I'm just sharing with you.
Yeah, it's not an exact, but the key is, is there further investment and is that investment increasing?
No, I'm not trying to argue you've changed, etc. I'm just trying to understand is the idea that hybrid bonding adoption is happening, but it's a bit further away. And in the meantime, it's been replaced by TCB has had, let's say.
Yes. And the answer is yes, because reflow, and you can go back even to wire bond, the industry will use a conventional process as long as they can because of the reliability and the cost, etc. So in that, yeah, let's say never-ending evaluation, you could say that some people expected hybrid bonding faster. One of the discussions was the JEDEC standard. But that did not improve the performance, which is also, let's say, confirmed in every sense.
Perfect. Thank you.
Thanks, DDA. Thank you all. Any last question? Anyone desperate to, but you know where to reach us? It's been a great pleasure and an honor once again that you've made the effort to come here. Thank you so much and always welcome. Thank you.