Thank you very much for everybody dialing in and your interest in Infineon and how we are powering AI. As we just heard, we will be talking about the area of powering from the grid to the core. From an Infineon perspective, this covers the two main segment application areas of AI data center, as well as the power infrastructure. Of course, now with the massive AI infrastructure buildouts, this is now translating into multi-year CapEx expansion. Recently, we talked about our AI business really from last year going up to EUR 700 million, and with this year doubling our revenues to an ambitious EUR 1.5 billion in FY 2026. The addressable market, which we will talk about no doubt as we walk through the presentation, is at the end of the decade, EUR 8 billion-EUR 12 billion.
Now, why are we so excited about talking about powering AI from grid to core? It's because we have an abundance of opportunities and addressable markets, thanks to our system understanding, our talented organization, the technical communities, and the customer intimacy. Of course, with that, the quality aspects. What we're going to do today, Peter and I will share the material. Peter will head up the grid aspects, and then I will be further talking about going into from the grid to the rack, the core, and then a brief introduction on the physical AI. With this, we're looking forward to giving you a glimpse, and we will ensure in the Q&A we can answer your questions. With this, over to you, Peter.
Yeah, thank you very much, Adam, and also welcome from my side. Why grid to core? Because we realized that the demand for electricity due to the upcoming demand from data centers leads to a significantly increased demand for electricity generation and transportation. We are a big player in the area of semiconductors for transmission and electricity generation. Just here as a teaser, we need the electricity fast, right? The demanding customers are having huge requirements with respect to additional electricity being generated. Looking into the lead times and into the available sources of electricity, this picture shall send a message with respect to cost and also lead time. Renewables, solar, wind, solar + storage, and gas are most likely the players, especially for the U.S. market, who are very competitive and quickly available.
Now, looking into how the data center of the future will look like, you have the connection line, high- voltage DC grid, and then you come to the data center side. There are some very interesting architectures which now evolve and have a high additional demand for leading semiconductor solutions. First to mention is the solid-state transformer, because converting high voltage to lower voltage levels required by the consuming data center side to additional transformer as we speak are in short supply. There are solutions where we, together with our customers, work on replacing traditional transformers by solid-state-based solutions. I will come to a second what this means in more detail. The second topic with the evolution of the data center is that data center goes DC grid. With DC grids, you need specific protection features, as for example, circuit breakers.
The traditional forms of circuit breakers do not fulfill with respect to tripping time, the advanced requirements, and therefore also the solid-state circuit breaker is a very attractive solution for our customers, which generates additional market size for Infineon and our industry. Later on, looking into the IT racks, I will hand over to Adam to explain more about the details. Before this, we do have the UPS, as in the past, but is now being complemented by energy storage, battery-based energy storage systems. Let me start with the latter one at the moment. That is a market that we know pretty well already exists today, and it's not only being used for data centers, of course, but as of today, it's being used for grid stabilization on the utility scale, and of course, also in private households for buffering and energy down in the basement.
Infineon offers a very comprehensive set of semiconductors ranging from power, via control and connectivity to sensing solutions. This market is expected to grow with a very solid CAGR of around 30% in the years to come. We are engaged with all big players in those ecosystems, ESS inverter companies, system integrators, and of course, also battery OEMs. The content, the semiconductor per MW, is estimated to be a little bit above EUR 2,000 on average. That is available. That is now growing nicely. Now we come to a topic which is very fascinating for technical people like me, and that is replacement of the traditional transformer by a solid-state solution. On the left-hand side, just to exemplify what we're talking about, you see the traditional transformer. Semi content for transformer is, as we speak, zero.
It's a piece of metal, of course, also a metal piece of art consisting of metal, wood, and oil inside. The total transformer market is estimated to be around EUR 15 billion. What is the beauty of the semiconductor solution? With semiconductor solution, we transform the passive transformer to an active device. Higher switching frequency gives the capability to significantly reduce the footprint, making it much, much smaller and lighter, and most important, making it quickly available. We provide semiconductor solutions like, for example, our EasyPACK module based on silicon carbide, which has a very high blocking voltage, up to 3.3 kV, which eases the construction and realizes a very compact design. We are collaborating with quite a couple of customers. Two are mentioned at the bottom line. What is very remarkable is the fast evolution of customer interest and the market.
Beginning of the year, I was known of roughly five startups in the U.S. who were working on the topic. Now, as the year comes to its end, this number of startups quadrupled, adding to the established incumbent players which are working on those solutions sometime since many, many years. The dynamic here is very high. The most important topic, it's the real new market for us as a semiconductor provider. Second, as mentioned, with respect to the DC grid, we need to protect the devices, the electric devices behind. We know about the huge value of a GPU, the final computing core, and the traditional electromechanical circuit breaker, which is shown in the picture on the left-hand side, has a reaction time in the order of milliseconds. Now, seeing on the bottom, the market for industrial electromechanical circuit breakers, you see it's a multi-billion market.
Overall circuit breaker market is significantly higher. That is only the industrial portion. Now, again, the beauty of the semiconductor solution? We provide a solution which has three orders of magnitude faster switching time. First, players and customers adopting our semiconductor solution, you see in the lower right corner, Siemens allowed us to share that we are in their devices, and we do believe into a market which grows a bit south of EUR 1 billion until 2030, as indicated on the lower left-hand side. The interesting topic is about also semi-solutions which are perfectly meeting the requirements of the solution. Just again, to explain a bit, the circuit breaker has not the intention to switch. The circuit breaker is always on, but if it is required, it has to react pretty fast.
While I just explained to you the value proposition of semiconductors going down to reaction times three orders of magnitude faster compared to the electromechanical device, there is another solution besides the traditional MOSFET, which is perfectly suited for such kind of application. That is now why we reintroduce on silicon carbide the JFET. The JFET is a very competitive device because it is perfectly suited for this kind of application. While we speak, we are in qualification of our first 750 V and 1,200 V blocking voltage devices, which have an industry benchmark RDS on, which is in the area of 1.5 or 2.3 milliohm per device.
The beauty of us being a portfolio player is, of course, that we are not only capable of offering this device as a single switch in a discrete package, as it is shown up there, but we have the ability to combine it with our module expertise. By paralleling this device, you can drive down the resistance even to significantly lower values. 10 devices in parallel with 1,200 V blocking voltage, and then you are at 0.23 milliohm, or you can think about stacking those devices and then achieving very much higher blocking voltages. That is a compelling value proposition for DC grids as for the AI data center, but of course, also for other applications, as it is indicated by the picture in the lower right corner. For example, battery main switch disconnect for trucks, buses, and even considerations for the automotive industry are very relevant.
The topic of DC grids is not only compelling for AI data centers, but of course, also for future industrial applications in the manufacturing environment. That is now exactly the point in time where we both meet with respect to the business responsibility. We can, of course, also protect exactly those devices we are now talking about in the second half of our presentation, and here I hand over to Adam.
Fantastic. Thank you very much, Peter. As you can see from the grid update, there is of course a growing SAM, service available market, for Infineon. Now we are going to move into the rack. What we wanted to show is really the evolution of what is happening in the industry and, more to the point, how Infineon is at the table with the lead GPU, TPU, ASIC providers in shaping the future of these data centers. Today, as we know, the racks are roughly at the power level, give or take, of 125 kW. For that, we can service around $15,000 of content for Infineon. Moving to the Northstar, as I call it, on the furthest right, you will see this hybrid microgrid really going, playing into what Peter was alluding to, moving to an 800 V high voltage DC architecture.
Now, within Infineon, our technical teams have been working with the customer base, the ecosystem, to really validate, can we possibly get to this greater than 1 MW per rack? The answer to that is yes. Of course, in order for us to get there from a technical perspective, from an ecosystem, from a reliability perspective, we also understand that there needs to be a stepping stone. In the middle, you see this three-phase VDC power cycle, where again, Infineon has an abundance of opportunities to really service these markets, not only in the car for the IT payload, but also in the power delivery and the backup power side. As you can see on the bottom, the Infineon content goes up significantly. As the power goes up, the Infineon opportunity goes up.
Now, what we'd like to also show is a little bit of this stepping stone, right? Talking about the three-phase power sidecar. Here we are seeing this will, of course, accelerate the compute, but also accelerates the content growth. We wanted to show you this because in the top side of this, the top of this graph, you can see here the server rack itself, talking about the computer blades plus the switch trays. As you move forward into the motherboard, then you've got the components like the intermediate bus converter, the high voltage, what we call the first stage. We've also got the GPU board on the second stage, where Infineon also has significant share opportunities in. At the bottom here, we talk about the power sidecar. Again, that is full of opportunities for Infineon.
What I'd like to do now is really walk through some of those examples. I must stress now, due to time and to allow Q&A, we can't show everything today, but we wanted to give you a flavor to let you understand that there are amazing opportunities for power management thanks to our system understanding and how we're going to be addressing these markets. Let's first of all talk about the PSUs. Of course, when we were here last year and the year before, we were talking about these solutions. At the time, we were talking about the 3.3, 5.2 kW, the 8 kW designs, and then the 12 kW designs. Now what you see on this is an evolution up to a staggering 30 kW three-phase PFC opportunities, where we are now working to get the technology evaluation boards for our end customers.
Peter and I, we are tied at the hip between our divisions and our technical communities to make sure that we come up with a compelling solution for our customer base. You will see here we're moving away from just the single-phase PSUs, more into the three-phase PSUs as the evolution requires as the power goes up. The takeaway of this slide is also Infineon is one of the few companies in the world that really has a play within not just leading-edge silicon technology, but the silicon carbide, as well as gallium nitride. Very clear speaking, these designs will need wide bandgap materials. There's no doubt about it.
By us working with our ecosystem and our customers, they have told us in order to get the efficiency at the right size, which translates to density, you need wide bandgap, meaning silicon carbide, meaning gallium nitride. I also wanted to talk a little bit about the advancements in the architecture. As we advance, there are now new opportunities for protection. Protection is very, very critical because, of course, it's essential to get high performance in the computing environments, but also monitoring to reduce disruptions and to, of course, maximize the server uptime. These protection ICs effectively come in the area of the mid voltage, for example, in the 48 V smart EFUs families. Plus, at the end of the day, for the high voltage solutions, which then service the opportunities for the architectures for 400 V as well as 800 V hot swap.
Here we have also got reference designs. As you can see here, you have the SAM projections. Whether this is conservative or not, we will have to see the adoption rate. We can see now that the 48 V will stay for some time. Now you see the introduction of the value of the 400 V as well as 800 V, especially as the AI power consumption significantly increases. The next topic we want to talk about is the broad portfolio of high voltage as well as medium voltage intermediate bus converters. Again, these are required more and more. This topic is a topic that Infineon has, quite frankly, been asked to address from some of our customers. We have traditionally had discrete offerings at the bottom, and we see those, of course, proliferating as the content grows and the power requirement grows.
At the same time, we've also now been asked to do certain module designs because of understanding the design, but mainly on the areas to provide quality and reliability at a system level. This topic, again, is an area that is showing significant CAGR growth, as you can see here, + 50% on already a number that is growing. We feel now that we've got a dedicated team focused on these IBC solutions on the high voltage as well as the mid voltage. Let me go now from these IBCs, and we talk about converting the high voltage DC to the GPU supply voltage. This is another great opportunity for Infineon. There's a number of topologies out there in the industry. We have the three-stage approach, which takes the 800 V to the 50 V.
You can then step down to the 12 V or the 6 V. Again, even in that three-stage approach, there are a number of ways that you can do this from an architecture point of view. The reason why we're showing this slide, look at the gallium nitride play. You see the GaN in this green circle where we are now seeing customers literally telling Infineon, "We need GaN for this solution." On top of that, we have the two-stage approach where, of course, you go from 800 V straight down to 12 V. We will have to see how the industry adopts and which ones of these topologies that they take. Of course, there are risk-rewards on those topologies. Again, as you can see, we have silicon solutions. As well, on the high voltage, we have a GaN solution as well.
The final statement, if you look at the three-phase approach from the 800 V to 50 V down to 12 V, we've also introduced to the industry an interleaved buck topology, which we believe is now starting to proliferate with a very strong pipeline, and we believe will go into production very soon. The magic, the second stage. This is a lot of the analysts are, of course, following very closely because this is the solutions that go very close to the GPU, TPU, or the ASICs. I didn't want to get too technical, but just at a flight height, there's three different ways that we see these markets developing. We believe that we have got solutions for all of these ways. I'd like to give a little bit of a glimpse and an introduction to those. First, you have the lateral, the discrete solution.
On the right-hand side, you see the XPU, which is a GPU, TPU, or ASIC, whatever you may wish to call it. You then have the substrate with the motherboard. Effectively, you put down now your discrete solution down next to effectively the substrate and the GPU. That, of course, benefits on multiple areas. It is definitely the lowest cost, but of course, there are trade-offs on ultimate efficiency. Other customers we have been working with, we have actually started to develop the backside vertical module. These are the vertical power delivery modules that we are shipping today already in the industry. These are the customers that recognize if you take the vertical power module and put it underneath the GPU, of course, then you simplify the motherboard from the design perspective, but you also increase the power density.
Of course, ultimately bringing the efficiency to the next level. On top of that, we do not stop there. We also recognize SiVR is going to be a trend in the industry, and we are already developing the next solutions that actually go into the substrate. Now, from our portfolio perspective, we have, of course, got an abundance of modules today that are out there in the industry. If you think about the Infineon journey, we started way back in 2011. We came up with a standard molded package. We then went into this chip embedding because we recognize the reliability and the performance that chip embedding actually provides to our end customers. As you can see, we then went to 1 amp per millimeter squared with our first-generation module in 2024.
If you look now, already coming out in 2027, we're actually 3x that going to 3 amps per millimeter squared. These renderings that you see, they're not just renderings. These are modules actually being sampled into our customers for the next generation because those customers recognize the performance benefits of having these modules for the VRM solutions. Of course, moving forward, we very clearly state that the solutions have to be greater than 4 amps per millimeter squared. This really gives us a wonderful window of opportunity to demonstrate our capability of understanding the systems, taking the leading-edge technologies that we have within the company, and absolutely without compromise, focusing on reliability and system performance as well as quality.
Ultimately, we get the question, "Hey, Infineon, why are you seeming to be winning from grid to core?" I ask you, the audience, to really reflect on how many companies out there that can really understand and deliver solutions from the grid, what Peter very nicely presented, to the core. Here at Infineon, we pride ourselves on this from our system understanding. We've been in enterprise power for servers way over 20 years. We have a lot of understanding of the systems and how these systems have to come together. We also have got this broad portfolio. Honestly, what Peter and I have shown today is just the tip of the iceberg of the opportunities that we and our engineers are working in within our labs. It's incredibly exciting. Of course, customer first.
We, as a company, recognize that we need to accelerate our innovation to customer value. There is no better opportunity to prove to the industry that Infineon can accelerate our innovation to customer value than using and demonstrating this empowering AI. Of course, the deep D&A of Infineon. Quality. We honor our quality commitments. That translates to what we talked about earlier. Whilst we do not give the breakdown in all the customer specifics, as you can see here, we effectively tripled our revenue back there in FY 2024 to over EUR 700 million. We are at the moment committing the EUR 1.5 billion for FY 2026. That is SAM at the end of the decade, the EUR 8 billion-EUR 12 billion. Now, let me just take two more slides here to really explain the physical AI. There are lots of terms and Edge AI.
are terms out there in the industry on how the tokens are used. This is another really interesting area for Infineon Technologies. Here, I now, of course, take our hats off of our two divisions and really think about Infineon because Infineon's got an incredible play with enabling humanoids moving forward. This is some of the data analysis from Goldman Sachs on the base case plus what they call the blue sky. We have to see the adoption, but there is no doubt about it. We are seeing further adoption and a lot of money supporting these humanoids. Now, from an Infineon play, just look at this slide. Look at the building blocks, the functional building blocks that Infineon can service from the processor, from the computers, as it were, from the MCUs for the motor control ICs, making sure also for security ICs.
We also got the play within functional safety thanks to our automotive colleagues. On the connectivity side, of course, we've got the Wi-Fi and the Bluetooth where the new families are winning in the market. On top of that, we've got the Ethernet due to the recent acquisition of the portion of Marvell. This now is giving us a new entry into making sure that we're working with the humanoid companies to look at how Ethernet can win in those markets. Power is a given, but one of the areas that we see, especially in the hands, is the GaN for the motor control with our microcontroller. Of course, on the storage, on the memory, many abundances on the analog. In the sensing area, of course, you've got to sense the physical world.
This is a topic that we are now really looking at, focusing on for use cases with our end customers using the Infineon sensors. Here, we believe we have actually a SAM opportunity per humanoid of $450. My last slide before we open it up here for Q&A. We, of course, are proud of powering AI because of the full system understanding, the portfolios, the technologies, the manufacturing scalability, and of course, no compromise on the quality aspects. We also have the key functional blocks in the physical AI that I highlighted. Thank goodness, Infineon, we recognize that silicon, of course, will stay, but we need also mastering of the areas of wide bandgap in silicon carbide as well as gallium nitride.
Of course, at the end of the day, we believe that the revenue here will be EUR 1.5 billion just for the dedicated on the AI revenue. We will do this with the acceleration of the customer value through low cost of ownership as well as reliability. With that, now, I will hand back to Francois, and then we will open up to Q&A.
Thank you very much. Very interesting. Maybe before we go, you can raise your hand, and then you can limit to one question and one quick follow-up. You can unmute yourself once we name you on the call. Let me start maybe with both of you. I will be very brief and have one set of questions for each. That is where we get the picture.
First, Adam, I mean, obviously, you talk about this AI is very interesting and this EUR 1.5 billion for 2026. I was wondering if there is any upside to this 2026 number and what is the driver? What would be the upside to that? Because you mentioned it's an ambitious target. I'm just wondering, is there any room for more?
Yeah. Let me really. We stick to the EUR 1.5 billion. I would say now, having the opportunity to work in this industry for, give or take, 30 years, I've never seen anything like this before in my history of the semiconductor market. On top of that, this is more of, at this time anyway, a supply issue, not necessarily a demand issue.
If I do not mind quoting our CEO here, that we have got recently a new COO, and it is a good problem for our new COO to address because, of course, this EUR 1.5 billion at the moment is also gated by supply.
Okay. Thank you. Maybe a bit more longer term, I mean, you showed all this product roadmap and also this 400 V, 800 V, and the content per rack increasing significantly. What is your ability on your share here? Because obviously, you know better than me that you have a lot of competitors out there, and everybody is trying to grab this very ambitious market. Do you have any visibility on that? Because the content is high, but we want to make sure you capture at least or more.
Yeah. Obviously, tracking share is obviously problematic and difficult because there are so many variables. Traditionally, in enterprise power, we had anywhere between 30%-40% share. In AI, we see at least 30%-40% going into the horizon.
Is it across all products area, or do you see significant differences depending on where you are, motherboard or PSU, etc.?
We do not break that down because, again, the complexity, and there will be variants, and there could be platform changes, mix changes, award changes, ramp-up challenges. It is very difficult. What we are saying is ultimately at least 30%-40% market share.
Makes sense. Thank you. Moving on to Peter. I mean, when I look at your divisions, there is one cycle and one long-term question I have for you. One is GIP as a whole is still like 26% below the peak of 2023. If you look at PSS, it is above for the obvious reasons.
Auto is 8% below. You seem to track a bit deeper in terms of turn cycle. Maybe can you assess to why? How can you explain this? Do you see any recovery from your perspective?
Yeah. Now, disregarding AI, definitely while we were enjoying, of course, the allocation and the good times heavily, now we are, of course, also we have to digest also the weak cycle of the market, and particularly by the weakness of the European market and also China not performing as we predicted also for 2025, we dropped further down than expected. The good news is that there's evidence that we have now reached the bottom. The other good news is that we kept our inventory very much under control. We do not have any over-inventory issues looking forward. There are first signs that now the demand comes back.
As typically for the seasonal pattern in the fiscal year, our first fiscal quarter, meaning the last calendar quarter, is typically our weakest. That will be also true this year. Looking forward into the coming quarters, I'm cautiously optimistic that worst is over and markets start to recover. That is now very interesting because the AI opportunities come on top, right? That was the question related to my core market, my core business. This is now an exciting additional opportunity.
Makes sense. The last one for me is your division is mainly power semis, if not power today, and mainly discrete, right, power. I was wondering, as you see the grid appearing, there is, of course, this fear that China is investing a lot in power discrete significantly.
How do you see the Chinese competition in your market here in terms of pricing behavior, market share, rewards? Do you see an impact from that?
Of course. I mean, just to be a bit precise, right, if we talk about discrete, that includes modules. And modules is the lion's share of my business. Here, luckily, we see still quite a lot of areas for differentiation. The differentiation goes in both ways, in performance and, of course, also in reliability. While definitely every competitor, also the Chinese, tries to eat the lunch of the other, meaning also ours, it is about price performance. Especially in demanding applications, we do excel in performance. Particularly true is it for the emerging wide bandgap topics.
While we now see also the design in pipeline year on year increasing by + 20% roughly, we see design win, especially for silicon carbide, now skyrocketing for the industrial space year on year by 80%. Compared to the same time last year, I do have silicon carbide design win pipeline, which is filled by more than 80% on top. That is also partly in line with the topics we just described, but that gives me high confidence that there's a future and that, of course, also prevents us, myself, our business for being cannibalized and commoditized.
Makes sense. Thank you. Maybe we can open to Q&A. If anybody has a question, Olga will state your name and you can unmute yourself.
Yes, please. The next question comes from Lee Simpson. Please unmute yourself, Lee, and ask your question.
Great. Thanks for squeezing me in.
Gentlemen, very, very interesting presentation. I really wanted to just sort of push on the supply constraint side, Adam, here. You've talked about these supply constraints and it's hitting numbers potentially. I just wanted to, if we look at the whole supply chain, are these supply constraints with your manufacturing or is it with the rack makers? Really, as the supply of parts broadens, and I'm thinking power sidecars and cooling systems, do we see an increasing risk of supply constraints in the future? Maybe as a quick follow-on related to that, you've utilized prepayments in the past, I think, with the fit out of Ku lim. Could we do the same thing here, given that you probably have to fund the supply capacity growth via these customers? Thanks.
Yeah. Hi, Lee. Good to see you.
Look, ultimately, we're not upstream from understanding how the installations of these data centers are going to be built out. Of course, contractors, for example, getting all of the energy, meaning the electricity or the power into the data centers. That is, of course, a topic that is something that we will be part of in the enablement, but ultimately, we're not the decision makers on that. Now, coming particularly to the Infineon view on the supply. Of course, these markets are moving quickly, right? The volatility, the chunkiness of the business. When you are winning the business, it comes, and it comes very quickly with very short lead times. Here, this is something that we recognize within Infineon.
We are taking now measures to really support the adoption of our equipment to the leading-edge technologies for powering AI, meaning even in the areas of some of our IGBT technologies, we can switch and convert some of the fungible equipment into supporting some of the AI on the front end. That is something that we are doing, and I believe that, of course, logically makes sense, and that gives us further capacity and opportunities to grow. On top of that, we're working with our ecosystem partners, especially in these power modules, for example. There's lots of components that go in, and that's something we've got dedicated groups on to support those areas there.
To really do the follow-up question on these capacity reservation agreements, look, at the moment, what we sense is a lot of the data centers have been, of course, focused on areas like the GPU supply or the TPU supply or the ASIC supply. They have been looking at the optics and making sure those areas are covered. What we are sensing now is the customer base are coming to Infineon to really understand our capacity corridors and now beginning to open up discussions about how they can secure capacity to support their ramps. A little bit premature, but the good news is we have got customers at the table recognizing that power management is a key topic because without power, there is no AI. Ultimately, they are now looking at securing those corridors.
Okay, thank you very much.
The next question from Joshua Buchalter. Joshua, please ask your question.
Hey, guys. Thank you for taking my question and for hosting such an informative presentation. I guess I think you gave a bunch of the pieces here, but maybe you could sort of simplify it for us. You guys have talked about this EUR 8 billion-EUR 12 billion TAM by the end of the decade. Could you maybe just help us? How much of that is grid, core, and on the rack? I think that would be helpful for investors. Thank you. I have a follow-up.
Okay. Thank you very much for the question. Yep. Look, rule of thumb on the EUR 8 billion-EUR 12 billion, right, on the addressable market, if you look at taking the first stage versus the second stage, and the second stage really is the core, right?
This is where we talk about the vertical power modules or the down solutions closer to the GPU. Roughly, give or take roughly, it's about 50% of the SAM. Yeah. Then the other 50% is on the first stage. Of course, as we get closer to the end of the decade, more and more of that SAM opportunities open up here with GIP as well. So roughly 50% second stage, roughly 50% first stage.
Since we don't know yet, of course, how fast the different opportunities will grow, our assumption is that inside the range you just mentioned, the EUR 8 billion-EUR 12 billion , a low double-digit percentage is GIP business.
Okay. Thank you. I guess that's a natural sort of one for my follow-up. Within that, should we think about silicon carbide as only playing on the grid side, or do you see opportunities even within the rack moving forward? I guess same question for GaN as well. Thank you both.
Yeah. Very clear. All the things which relate to the data center start with the solid-state transformer, go forward to the solid-state circuit breaker. These are all silicon carbide-based solutions in different variants, but that is silicon carbide pure play, so to say. Therefore, also to a certain extent, that explains the high CAGR in my design win funnel. While, of course, on the grid side, it is a bit more complicated. The high-voltage transmission lines, which will also need significant investments and build-out, they are typically IGBT play for the foreseeable future far beyond 2030.
In the rack for the PSUs, as we presented earlier, from the 3.3 kW, especially move up now into the 8 kW, 12 kW up to the 30 kW three-phase. This will have all three technologies of silicon carbide, gallium nitride, and silicon. Openly, the customers have told us they have to use the wide bandgap to get the efficiency as well as the density. The other area for GaN, where you asked where the adoption of GaN will come in, most definitely in the IBC, the intermediate bus converters on the high voltage as well as on the mid-voltage, for example, on the interleaved design, as well as, as mentioned, on the PSU. Wide bandgap will be used in powering AI in the next generations and, believe it or not, even today.
Thank you both. Yeah, congrats on the high order book.
Thank you.
Thank you.
Next question from Jakob Bluestone. Jakob, please.
Hi. Thanks for taking my question and for the interesting presentation. Just on the EUR 8 billion-EUR 12 billion , is there any way you can break down how much is GaN, how much is silicon carbide? Maybe just sort of break it into some of the different materials. I had a follow-up as well.
Yeah. Jakob, I do not know if we can go back to the material. I think your question is on the intermediate bus converters, if I am not mistaken, Jakob, right? This is on this slide here. Is that correct?
Yes.
Yeah. We do not break it out, but if you see here, the darker green color really shows the percentage of the dominance of the technology. Where you see GaN darker, that really shows that there is more of a GaN bias within the solution.
That, of course, happens most definitely in the two-stage approach on the high-voltage IPC. I will quote one of the engineers at the customers, "No brainer." They literally told us, "No brainer. They will use a GaN-based solution." On the mid-voltage, as we go through, of course, there is more of a mix between the silicon as well as the gallium nitride. It really depends upon ultimately, are they going to be looking for more efficiency on the GaN, but on top of that, the size as well for the density. I cannot break it out really any more than that because we have many, many solutions all coming through the pipeline, and there really is a fusion, a mixture of those solutions. Sorry, I cannot be more precise.
Understood. Just a quick follow-up. From an ASP point of view, does it make a big difference whether it's GPU, TPU, ASIC, or is it all pretty similar from your point of view?
Yeah. It's a great yes. I like it, Jakob. So what we will say is the GPUs, of course, are growing significantly in the power. The TPU and the ASIC vendors, it's really been surprising us. They've even got more aggressive power ratings coming up. Remember, these GPUs were starting off, whatever, 800 W, 1,000 W. They're going up to 2,000 W and beyond. We now, of course, are seeing some roadmaps well and above 4,000 W now on these GPU, TPUs, and ASICs. What makes, of course, that brings in more content. Openly, that brings in more power requirements for Infineon as well as our competitors.
Now, in saying that, where we see the major difference is when the industry fully adopts into the vertical power modules, then, of course, the ASPs is a significant factor. That can be anywhere between three or greater than three from an ASP perspective as we go into the module solution versus discrete down solutions.
Great. Thank you.
Next question from Jonathan Mennon. Jonathan, please ask your question. Jonathan, I think.
Yeah. Yeah. Hi. Thanks for taking the question. Just following on for that last answer, am I correct in understanding that the GPU is already a vertical solution and therefore gives you a higher ASP? What is the rate of conversion of broad ASICs GPU to vertical power in the next will it happen in the next couple of years? This is going to take time. Similarly, you said 3x for vertical power.
If you go to the substrate integrated one, what would be the step up there? And is that something that's going to happen within the next five years, or is that sort of beyond the next five years?
Okay. Yep. Good. I would say the majority of the industry is already shipping or working on the vertical power delivery modules. We are working a lot with those companies. There are, of course, one or two, maybe they're still on the down solution. The next question is, will they stay on down, or will they also see the benefits as the power of the GPU increases, whether they will need to go to a different architecture? There, this luxury with Infineon, we can support both, right? We will listen to the customer.
We will focus on the customer voice, and we will then support what the customer needs. Now, why then are the others in the industry already on the vertical power module? Because, honestly, it's again, the topics around the efficiency levels. It's about the total cost of ownership levels. It's also down to space saving and simplicity of these motherboard layouts. There are many reasons why the industry has gone that way. We are at least developing heavily on the vertical power modules, which I think should give you enough hints. What I showed earlier on, the greater than 4 amp per millimeter squared, that's where we believe the industry will further adopt to. On the question on SiVR, yeah, if you ask me for a five-year window, I believe yes.
I know certainly Infineon will be developing and be working with customers for that type of horizon for SiVR.
Just you said that sort of stage two, stage one, roughly 50/50. I'm assuming that the competitive dynamics on both sides will be different. I mean, in stage two, we have sort of established competitors, what we hear about like Monolithic and Renesas and people like that. Whereas on the stage one or intermediate, whatever, all of that, we want to put together. NVIDIA is signing agreements left, right, and center with everyone they can catch on the roads. Would that stage one in the HVDC be a more competitive environment over the next five years? Therefore, is there a potential that your market share there could reduce?
Similarly, as you go more to vertical power and SiVR, given that that's where Infineon's capabilities and technologies are extremely differentiated, maybe that your market share there can increase. Would that be a way of looking at it?
Yes. Look, I would like to answer it really on it's, of course, a technology-led discussion, especially on the second stage, as you highlighted. Yeah. There's, of course, an opportunity for the industry to really demonstrate the leadership and the capabilities there. Again, it's not just having the latest technology. You need to make sure it works at the system level because the closer that you are to the GPU, TPU, ASIC, then you've got to get it right because you've got to protect that very expensive component. On the first stage, yes, there's more overarching competitive landscape.
Again, one thing that Peter and I are incredibly proud of, we're working for a company that is leading in multiple dimensions on figure of merits, meaning that we are really coming out with leading-edge technologies in the areas of silicon, silicon carbide, gallium nitride. In those areas, we're not just thinking about today. Look at gallium nitride. We announced a 12-in GaN capability within the Infineon offerings. On top of that, Peter, you may want to talk about silicon carbide, the opportunities for capacity.
Yeah, of course. I mean, we have our fab in Kulim, and we're now, as we speak, converting from 6-in to 8-in. We have mastered the 6-in RAM with high yields, and we have invested for Kulim only into bridge tools, meaning that we have the full capacity available once the material and the business case is there.
That is what is now the case, right? Of course, per square millimeter, 8-in wafer being silicon carbide I am talking now about, right? It is being a bit more expensive. Due to productivity coming from 200-mm tools, we now already see a cost benefit for moving to 8-in. I think that is not big news. All the industry, including our competitors, do see it, but we are well prepared. The whole capacity in Kulim we have invested is 8-in ready, and that will give us an edge in ramp-up speed for sure, right? If there are significant volumes required, we are able to supply and, of course, down the road, cost advantage.
My final statement here is even on the silicon, we are, of course, developing the next generations of silicon to support these markets.
On top of that, with our microcontroller families, embedded security in the microcontroller, we're also coming up now and working and offering system play within, for example, the PSUs, and there's a lot of interest there. Of course, that's the reason why on the first stage, yes, there's more competitors. Again, Infineon, we're very proud of the fact at a discrete level, including modules, we're a clear number one in the industry, and we don't stop there.
Understood. Thank you very much.
Welcome.
The next question comes from Sandeep Deshpande. Sandeep, please.
Hi. My question is regarding your exposure in the AI market. You talked about, I think, 10% or 12% of the exposure by the end of the decade will come from the GPU side.
Is your competitive position in the GIP side much better than your position on the PSS side, given that many of these multiple competitors who are coming into the AI space do not really have that high-power capability that you guys have in GIP? I have one quick follow-up.
I would say, I mean, I do see certain areas of the market where we have very high differentiated solutions. But since being quite a while in this industry, namely also much more than a decade, whatever kind of solutions we offer, right, there will be competitors around, for no doubt. First of all, our customers hate single sources, so they will do everything to enable competition there. Of course, it is also healthy for us. While still on the topics we already mentioned, in certain areas, we face strong competition, commoditization. In other areas, we have unique technology.
I would say it's very similar for GIP and PSS. Those kind of differentiated pockets, I would call them, are available on both sides. That would be my answer, my first answer to your question. Adam?
No, no. It's good. GIP.
I mean, my question now, I mean, to Adam, to you on your side is that, I mean, when you look at the solution, it's not a single-point solution. There are a lot of components to the solution. When Infineon is on a particular board, is it that all those parts associated with power are Infineon, or are you sharing various maybe there is one stage where some parts are Infineon and some other ones of those competitors also on that same board? How does this play out overall?
Associated with that, I mean, particularly associated with PSS, is this AI part of your business accretive to your overall margin?
Yeah. I'll take the second question first. Yes, for Infineon now, overall AI will be accretive and, of course, higher than the average margin for Infineon moving forward. Second of all, some of the customers are very particular of having a supplier that, for example, with the controller as well as the power stages, that there's no mixing and matching because at the end of the day, they want system performance. They want to lead with a supplier that can offer all of the technical competency at a system level. Now, saying that, of course, if we stand back and look at the compute tray and look at the motherboard, it's amazing.
The auxiliary rails, for example, the point of load rails, for example, the protections, the EFUs, the hot swaps, the IBCs, this, of course, can really be a little bit more mix and match because those will be qualified under their own sort of conditions. Ultimately, what we're finding at the moment, and very transparently, that some designs and some of the orders that we're picking up now may have just been macro designs because we're the best performing, and they're dropping us on the boards. Some people call it the sort of the sprinkling and the shrubbery, but it's amazing how all of that adds up to a very healthy opportunity for Infineon as well. That's because customers really believe in our technology, leading edge, and, of course, on the quality and the reliability.
Thank you.
Yeah.
Now the last questions in the sake of time. Sebastian Sztabowicz, please ask your question.
Yeah. Hello, everyone, and thanks for taking my question. On NVIDIA and the 800-V DC architecture, what kind of market share do you expect there? There are a lot of competitors in this project. Do you expect the same kind of market share you have on AI data center today, or lower or higher?
Yeah. We do not talk particularly about customers because why we're not allowed to. What I will say here is that, again, we are targeting at least a 30%-40% share. If you go back on the public domain information with NVIDIA, we did the collaboration announcement very early on with NVIDIA because, of course, that is a revolutionary opportunity for the industry.
We were one of those suppliers that could, of course, give them confidence that we could head towards the 800-V DC opportunity. To answer your question, we do not break it out by customers. Again, market share at least 30%-40% to the end of the horizon.
Okay. Thanks a lot. The last question is on the ASIC market and the acceleration of the investment for the large IPS scalers around ASIC. Do you have the same kind of exposure between IPS scalers or between IPS scalers and GPU vendors, or are you overexposed to one of those kind of customers in terms of AI-powered semiconductor business? Thank you.
Yeah. It is a great question. I would say at the moment, Infineon, we are heavily engaged with the majority, if not all, of the GPU, ASIC, and TPU vendors.
There is a rich opportunity there, meaning that most definitely on the EUR 1.5 billion here that is not coming disproportionately from one customer, most definitely not.
Okay. Thank you.
Yes. I am very sorry. We have now to end the call here. Thank you very much for your participation and your questions.
Thanks a lot.
Thank you. Thank you very much, everyone.