AT & S Austria Technologie & Systemtechnik Aktiengesellschaft (ATS) Earnings Call Transcript & Summary

Ladies and gentlemen, thank you for standing by. I'm [indiscernible], your operator today. Welcome, and thank you for joining the AT&S conference call on the results for the first 3 quarters of 2025-'26. [Operator Instructions] I would now like to turn the conference over to Mr. Philipp Gebhardt.

Thank you, Mandy. Good morning or afternoon, ladies and gentlemen. Welcome to the AT&S 9 Months '25-'26 Conference Call. I'm delighted to welcome our new CFO, Gerrit Steen, to his quarterly -- first quarterly call for AT&S. As he has just officially joined us yesterday, our VP of Finance, Silvo Leitner, as in the previous call, will present the figures and answer any financial questions. However, in the recent months, we have received an increasing number of questions about current technical developments. We see that as a positive sign that the focus is shifting from our balance sheet back to our core business. Therefore, we would like to use this call to answer these technical questions in a more qualified manner than if they were addressed to me. I'm, therefore, particularly pleased to welcome our CTO, Peter Griehsnig, today. He will address the 2 current trends directly in the presentation and will also be available to answer your questions afterwards. So after this introduction, please allow me to begin on Page 2. As planned, Kulim and Hinterberg are now contributing to revenue. This marks an important milestone in our capacity expansion and supports our top line growth, bringing us up to EUR 1.3 billion in revenue. We also see positive momentum on profitability, product mix improvements and a more favorable pricing environment in Q3 have supported the margin recovery. In addition, our efficiency programs continue to contribute to earnings and make a significant impact on our EBIT of EUR 35 million. On the downside, currency fluctuations, in particular, the weak U.S. dollar, created notable FX headwinds, partially offsetting both top line and operational improvements. Nevertheless, supported by the restart of our factoring program, we achieved a positive operating free cash flow of EUR 223 million, demonstrating the strong cash discipline. And while global dynamics remain uncertain, we are seeing sentiment in our core markets becoming more positive. So based on the current performance and the outlook ahead, we have confirmed our full year guidance as well as the guidance '26-'27. Moving on to Slide #3. The PCB market grew by 15% year-over-year or 11% in euro terms from '24 to '25. This growth is driven primarily by increased demand in high-end computing, networking and advanced packaging applications. The IC substrate market also expanded last year, growing 18% or 14% in Euro. This continues to be supported by strong AI-related investments and ongoing data center build-out, which remain the growth engines. I'm continuing on Page #4. After adjusting the number of our Supervisory Board members last year, we now have, as of February 1, also reorganized the Management Board. Instead of 5 members, the company will be led now by 3: our CEO, Michael Mertin; our CTO, Peter Griehsnig; and our CFO, Gerrit Steen, to whom I would like now to give the floor.

Thank you, Philipp, and good morning, good afternoon, good evening, wherever you are, everybody. My name is Gerrit Steen, and I'm very happy and proud to have joined AT&S as CFO actually yesterday. So my immediate focus now is on gaining a detailed understanding of the business, the markets and the financial position of the company. But I can tell you already now that I have had very positive first impressions of the commitment and dedication I have seen across the organization. But for today, as Philipp already alluded to, I will primarily listen and leave the explanations about our last quarter in the very capable hands of my new colleagues. Certainly happy to support the discussion later where relevant. So thank you, and over to you, Silvo, to guide us through the numbers.

Okay. Also welcome from my side. I will lead you through the figures, starting with the consolidated group. You can see very nice on a year-to-date basis, we increased in our revenues from EUR 1.1 billion to EUR 1.3 billion, so roughly plus 10% and also a very nice development from quarter-to-quarter, roughly plus 18 percentage, and we are landing now in the third quarter with a revenue of EUR 468 million coming from positive product mix volume and pricing effects, but also from the ramping. On the next slide, we go to the BU Electronics Solutions. There, you can see if we take Ansan out and compare apple-to-apple, I will say, we can see year-to-date revenue somewhere flat with EUR 670 million, so 0 increase. And over quarter-to-quarter, you see a declining of 4%, but you have to take in account that here, we have the seasonality of the mobile devices business. And even in the last -- Q2 quarter, we have really a nice revenue of EUR 256 million, so a little bit also on timing differences. Maybe next slide. Here, we can see the business unit microelectronics. It's a very, very nice development year-to-date, we see we are growing from EUR 463 million to EUR 621 million, sure coming from the ramping in Kulim and Leoben, supporting also with volume mix and price increases and developments. Also quarter-over-quarter, you can see a 65 percentage increase from EUR 154 million to EUR 254 million. And even quarter-to-quarter, Q2 to Q3, nice increase and also an increase in the margin. And I think that is for microelectronics. Coming to the maturity profile and our financial position. Also here, you see a very nice solid financial structure. We have EUR 886 million cash and cash equivalents, most of it, EUR 843 million in the cash and cash equivalent and some unused credit lines with EUR 43 million. If you look in our profile on the left side, you can see for this year, we have roughly EUR 140 million to repay next year, EUR 488 million. That is easily handled with our EUR 886 million on our pocket, I will say. Also, the financing costs are very nice with 3 percentage, everything from that point very well. On the next slide, you see our working capital that we always show. I think we have a healthy level now with 14.6% or 13.6 percentage in our Q3. If you know before this 19%, 18 percentage, what you see in the quarters before is coming from our factoring that we leveled out over the time. On the next slide, the cash flow. Cash flow from operating activities also very strong, coming from EBITDA and also a little bit helping from the factoring, EUR 332 million. Also the cash flow from investing activities, very nice, driven by the cash deposits that we have in the last year, but also lower CapEx, minus EUR 26, financing EUR 83 million, also okay. So we come to an operating free cash flow of EUR 223 million in this year. Last year, we had a minus EUR 357 million. So -- and the net CapEx is minus EUR 108 million. So everything I think also is very well. Balance sheet, total assets flat, EUR 4.6 billion. Equity, we lose a little bit out of the headwind of FX, mainly in the U.S. dollar. Equity ratio is with 20.8 percentage. And also in the net debt, we can reduce it from EUR 1.4 billion to EUR 1.3 billion and the net debt EBITDA ratio with 2 is also very nice, I would say. On the next slide, the cost saving program that we always show. Our target, as you know, is EUR 130 million. We achieved this in the third quarter, and our expectation is that we land at the year-end at EUR 160 million plus. Next slide, I can confirm, as you have heard also in the beginning, the guidance for '25, '26, we will stay with the EUR 1.7 billion. The margin, as [indiscernible] mentioned also in the last quarter, 23%. And the net CapEx we've reduced from EUR 250 million to EUR 200 million. And we will stay with a positive EBIT and operating free cash flow. And also the guidance for the year '26, '27 will stay as also mentioned in the last quarter between EUR 2.1 billion and EUR 2.4 billion. As I mentioned last time, have in mind that we have headwind from the FX and also a little bit maybe on glass shortage, itself shortage. So we see us at the lower level and other things are unchanged. I think that's it from my side.

Over to you, Peter.

All right. Good afternoon, good evening to everyone. This is Peter Griehsnig, CTO of AT&S. Allow me to take a few minutes to shine a little light on technology trends in the area of substrates, in particular, substrates for artificial intelligence because that's where the action currently is. AI has definitely changed the speed of progress. And with the traditional transistor scaling not being able to deliver what is needed as fast as it is needed, the industry turned to the package to support and with the package also the substrate. So what are actually the challenges of the industry where substrates can support, where substrates need to innovate to deliver their input. It's basically what's mostly 3 topics. One is the size of packages. The other one is the data rates and the third one is the power delivery for those processes. Now why are these challenges? And what can we do as substrate manufacturers? In order to understand that better, we need to look at the architecture of such a processor package. And you see that in 3 different variants. Take the middle one first, CoWoS. CoWoS is the workhorse of the industry. This is what TSMC does. This is how almost all the current AI processors are built up. It consists first on top of the active dies, the memory stack and the GPU, which is connected first by a silicon interposer. Then all the signals that go in and out of the package are routed and fan out by the substrate. The substrate also routes the power to every place in the package where it's needed. And these 3 top layers is what generally is called the package and the package is then connected to a much larger PCB. That architecture works quite well. so far, but it comes with its own challenges. When you want to increase the processing power, what you can do and what you must do is put more GPUs and more HBM, the high-bandwidth memories into that package, and that increases the size of the package significantly. Over the last couple of years, package sizes have quadrupled. And going forward, they will quadruple again. Very soon, you will see substrates the size of a tablet. That means, of course, massive more manufacturing demand. Not only the unit numbers increase, also the unit size increase -- will increase dramatically. Of course, also technology-wise, this is extremely challenging to deliver the very, very tight specifications that are necessary for a substrate on such large package form factors. What we can do, what we need to do is use alternative new, more exotic materials. They will be, of course, more expensive. They will be thicker in order to keep warpage under control. And ultimately, this may or may not lead to the introduction of glass-based substrates. What also is a challenge coming from the package size is how to scale up the silicon interposer on top of our substrate. That's a large piece of rectangular silicon cut out from a circular wafer. That's not good when you want to scale it up. There are natural limits to do so, and it is becoming hugely expensive. So what to do going forward? Multiple options are there. First of all, it would be an option if instead of cutting it from a round wafer, if we would cut the interposer out of a panel similar to what we do, a rectangular format, the size of maybe 0.5 meter x 0.5 meter, you could fill in many more, you would have much less waste, but it probably won't be silicon anymore. It should preferably be something organic, which then comes close to what we in substrates do. Of course, the challenge will be to achieve those fine feature sizes that the semiconductor industry can do with silicon. 1 micrometer or below 1 micrometer line space is something that is very, very difficult to achieve with organic material as of now. But if we look further into the future on the longer horizon, it is an option, what could be done. There is already today something that is a smart solution to scale up the size of packages. And that is when you look to the left side, it's called EMIB, basically Intel solution. And the thought is simple. You don't need an interposer full size. All you need is a connection from one edge of one die to the other edge of the other die. And so you can break down the interposer in multiple small pieces, which are called bridge dies. And when you scale up the size of the package, you need more of them, but not a bigger one. That makes scaling way more economic. And the larger the package size will be in the future, the more benefit this kind of architecture seem to have. So going forward, if Intel can get some design wins in the market with that kind of technology, their share in the market could probably increase based on the smartness of that package architecture. And currently and probably also going forward, there's less than a handful of suppliers that can manage and manufacture the extreme tight tolerances that are necessary to embed those silicon bridges slightly under the surface of the package substrate. So that solution has taken the advantage of merging interposer with a substrate, effectively eliminating the interposer at least for some applications. There's also another opportunity that you see on the right-hand side where the substrate is merged with the printed circuit board. effectively eliminating the substrate. That's what's called CoWoP. It is a new technology promoted by TSMC and tried -- NVIDIA is trying to build such kind of packages in the near future. However, the challenges are substantial. The challenges are -- the PCB already is not so easy to make. It's a high layer count board and a large board. Now you need to merge it with the additive processing technologies of a substrate, and you need to fulfill all the tight specifications that usually only were applicable to smaller -- much smaller substrates. Can that be done? Technically, probably, economically question mark. The future will show us how well that kind of architecture can work. Why would any customer want to have? What's the benefit out of it? There are lots of benefits. Probably the one that is reported quite often, the cost saving. I have some serious doubts that this will be much cheaper. But what definitely can be a benefit is addressing the second difficulty, which is data rates. Signals. In order to keep this GPU running, you need to bring into the package and out of the package massive amount of data. A GPU makes easily 100 trillion operations per second, and accordingly, you need to bring in data constantly. It's not a secret that GPUs are idle significant time because not enough data available. So solving that issue would be very, very beneficial. And signals, first of all, don't like to travel long distances. And secondly, they don't like to go through different metal while they travel along a conductor. So when I take out the substrate, I cut out 18 layers of substrates. Signals definitely like that. And also one of the layers of microballs, which is a complex sequence of copper, nickel, palladium, gold and tin. And signals also would like it if you take them out. Data rates could be increased by such a solution. Whether it's feasible to do so, we will see definitely. As a substrate manufacturer, you want to be prepared should this kind of architecture pick up in the future. Then the third topic, if -- before we talk about power, let's have a look how this -- such a package looks like in reality. You see the MD MIA350 accelerator bottom left. In the middle of that picture, the colorful thing are the GPUs surrounded by these squares, which are the memory stacks. They all sit on this darker gray, which is the silicon interpose and then you see slightly greenish with a lot of yellow dots, that's our substrate. And that's basically the package with a metal frame around it, and that sits on what's called the accelerator card, a large PCB that has, amongst other things, most of the power, the last stage power conversion on it. And that power conversion are all these black cubes that you see at the edge of the PCB. And that's the important topic that we need to talk about. Because power delivery has come from being a side show to the main stage of designing powerful processors. Such a processor today needs easily 1 kilowatt, and that's current generation, and you can expect to double that in a few years and to quadruple even further down the road. That's a massive amount of electrical energy that you need to feed into these processes through our substrate. And the art is to deliver that electrical power without losing half of it on the way to the processor. One noteworthy thing is that these processors run unfortunately for us at very, very low voltage levels, like 0.7 volt or 1 volt. That means we need to route about 1,000 amperes through the PCB first and then through the delicate structures of our substrate. And that's not that easy. 1,000 amperes is twice the amount of current that a Tesla supercharger would use to charge the battery of your e-vehicle. A lot of losses are happening at such gigantic currents. We can help. We cannot reduce the power that the processor needs, but we can help to reduce the losses significantly. Now these power modules here indicated as PMIC power management IC, also called voltage regulators. They reduce the voltage level from the incoming, let's say, 12 volts to the 0.7 or 0.8 or whatever voltage the processor needs. While it converts the voltage, it loses power. And the amount of power that gets lost is directly linked to the distance of the components inside this module. So the closer you can bring components, the less losses you will see. And it's perfect, a perfect application for our embedding technology. With embedding, we can bring components from millimeter distance to micrometer distance, and effectively cut down the losses that happen in all these black cubes that are there for power conversion. And in the next step, power loss also depends on the distance between this PMIC and the active dies. The longer the route is, the more power gets lost. So why not bring this power conversion stage closer to the active dies. We can do that by, again, embedding it, for example, in the PCB bringing it down from 5 centimeter to 1 centimeter distance or we can even go one step further and bring the power conversion into the substrate, effectively reducing the distance to the die to millimeter. And again, embedding technology is the crucial part, the crucial enabler of reducing the losses significantly. We have shown for these PMICs that power loss can be cut down in half. A first generation of such product is already out in the market, very successful. We're working on next generations and the demand is picking up significantly in the future. And with that, I hope I have given a little light on what's going on, what are the challenges and what are the trends on the physical layer of artificial intelligence. And with that, I believe we are open for questions.

Okay. Yes. Thank you, gentlemen. We will now start the Q&A. [Operator Instructions] Now I would like to hand over to Mandy to handle the session.

[Operator Instructions] And I can see the first question from George Brown from DB.

I have 2, if I may. I can take them one at a time. Just firstly, we've had a stronger Q3 report this morning, but you've not changed the full year sales outlook. So I'm just looking at EUR 1.7 billion, then that implies basically quite a big sequential step down in Q4. However, we see news around pricing in ABF substrates improving from the beginning of this year and stronger demand for server applications from Intel, AMD, et cetera. Apple, of course, just had a great quarter and is sort of now in supply chase mode as they sort of stated on their earnings call. So I'm just wondering how we should interpret this implied Q4 guide, it feels quite conservative. And then I have a follow-up.

I will answer this question. Yes, I would say I agree with you. It's a conservative guidance for this year with this EUR 1.7 billion. We are very confident that we will achieve, and it will be surely EUR 1.7 billion plus.

Okay. brilliant. And secondly, just on the FY '27 guide, actually, whilst I have you. So again, that's been reiterated. I remember you stated in your report, I think, in Q1 or Q2 that you have shifted your expectations from the high end to the lower end of the guide due to FX headwinds. But I think more recently, again, like I said, we've seen positive pricing developments in the last few weeks or months in ABF substrates. People are talking about ABF substrate pricing being up potentially 15%, 20% in 2026. And you also added in your report interestingly that customer forecasts are being revised upwards for the second half of 2027. So could that -- could the high end of the '27 guide be within reach? Or are you still sticking to the low end basically?

I take also this one. I told it, I think in the last call, our guidance was calculated with U.S. dollar from 10.7, and we are now at 11.7, 11.8, even we have last week 12.0. So there's a lot of headwind in the FX. And the second point that we see is also glass shortage. Even if there are demand, we have to see if we can then fulfill everything so that we stay with the lower end.

And the next question is from Daniel Lion from Erste Group.

I would like to follow up on next year's guidance. Can you give us maybe a little bridge in terms of where you stand now in terms of capacity utilization and capacity itself? And to what extent you will need further capacity ramping up in order to meet the guidance? Just to give you some insight where I come from, the delta is EUR 350 million at least, it should be EUR 350 million to EUR 400 million if you exceed the EUR 1.7 billion for the full year. So where should it come from? And how you think about additional capacities here?

Yes. Sorry, Daniel, as Silvo already answered, I will take this one. You will get the IR answer. Well, that's an additional capacity coming out of Kulim. So we are in the middle of the ramp. And in not technical terms, having the CTO next to me, I have to be careful, but just add another line, and that brings you the revenue into the EUR 2.1 billion up to EUR 2.4 billion. Don't forget also Hinterberg contributing to this, but really the big change in the setup is more capacity out of Kulim.

Okay. And maybe a second one also somehow related to this. When you look at peers announcing already additional capacities, substrates in order to live up to increased market expectations or market demand. How would you think now of going forward in terms of CapEx? You level down to EUR 200 million for this year, obviously, cash flow due to free cash flow reasons. But how should we think of next 1 or 2 years? Would this be a level that is reasonable? Or would we have to add maybe the lower CapEx this year to CapEx in the coming 2 years maybe?

Yes. I think the lower CapEx that you see in the Q4 is -- or will see in the Q4 compared to what we planned for full year is, of course, again, at every year a bit of a timing difference. So we are shifting it a bit into the next year, whatever is possible, so you can add it to whatever you expected for the maintenance next year and the additional line might be a bit higher than what we said a year ago. When it really comes to the big expansions that we heard from competitors in the recent hours, I think that is something that we have to look in very carefully. And we will announce whenever there is something happening there.

Next question comes from George Chang from Aletheia Capital.

Just a couple of questions. First of all, what are the considerations for the ASIC clients when they choose between CoWoS and EMIB-T?

Well, it's technology-wise, difficult to say what really drives one in this direction, what drives one in the other direction because both of these architectures have their own benefits. It is also traditionally when you have bought from TSMC, you only get CoWoS. If you have bought from Intel, which very few had in the past, then you get EMIB. Only recently, Intel opened up for EMIB. When they decide now, what you probably will assess is the overall the total cost for your package. There could be going forward, in particular, when the size becomes bigger, there could be a significant difference. Other than that, that's a question that's better answered by the package designers than by the substrate manufacturers.

Okay. Great. The thing I just want to just verify that my understanding is that the capital intensity for EMIB-T is probably double of that regular substrate. And it sounds like it's a lot more than traditional EMIB. So I'm wondering, can you explain sort of why EMIB-T investment is so expensive?

Yes. It is -- well, is it expensive or not that is in the -- everyone might have a different judgment on that one. But it is more expensive capital-wise, definitely to build an EMIB compared to a standard substrate because one of the reasons is the precision that you need to exhibit when you assemble the die into the cavity of the substrate. That precision is semiconductor level precision talking about micron 1, micron, 2, 3 microns. And when you go from EMIB where you have a single-site connection to EMIB-T, where you also connect from the bottom side, all of that becomes even more tight. The tolerances are going down and the process steps are going up because connecting from the bottom side in an architecture where you have the die not in the center, in the core of a substrate, that is a complete unique process that requires a lot of additional process steps that make the whole process more expensive, the investment bigger. But whether it's twice as much as a normal substrate, that remains to be seen. I would challenge that.

The next question comes from Lucas [indiscernible] Bank.

I had the same question as my colleague, so...