GLOBALFOUNDRIES Inc. (GFS) Earnings Call Transcript & Summary

Okay. Excellent. Good afternoon, everybody. Welcome to the Goldman Sachs Communacopia Technology Conference. My name is Jim Schneider. I'm the semiconductor analyst here at Goldman Sachs. And my pleasure to welcome GLOBALFOUNDRIES and CEO, Tim Breen to the stage today. Welcome, Tim.

Thank you, Jim. Great to be here.

Excellent. I want to start off with a couple of very high-level questions around your business. GLOBALFOUNDRIES kind of stand as a foundry providing specialized CMOS process disease, has manufacturing sites spanning the U.S., Asia and Europe. At the highest level, tell us a little bit about what kind of capabilities are you net to GLOBALFOUNDRIES, why customers come to you and why you win?

Yes. Thanks, Jim. Great to be here. Look, we see that in 3 broad categories. I think the first thing is there was once upon a time when in semiconductors that was leading edge and lagging edge. And that was really defined by the technology transitions that the industry went through. I think it's obvious today that we've moved beyond that paradigm. We've moved beyond it because where semiconductors find themselves in our daily lives, in the data centers that we're relying on in space, a broad set of set of use cases. And it means that when you think about what we do, you have to focus your strategy against that. And so we think of these 3 pillars that we focus on. Number 1 is our technology portfolio. So we have an incredibly rich diverse technology portfolio built over now 17 years since we've been a company called GLOBALFOUNDRIES in areas as diverse as analog mixed signal, in power, in ultralow-power CMOS and beyond. And so step number one, have for those diverse applications, a set of technologies at the process level and increasingly now beyond that process level with advanced packaging with IP, with software, with custom silicon as well. So pillar #1, kind of deep focus on those differentiated technologies that accomplish specific workloads. Number two, ultimately, you are a really high stakes decision for your customers. They don't get to choose you this year and not choose you next year. They bet on you for the long term. And so you need a track record of trust and delivery, and that takes time to build. And we've been doing this for some time now. I'm really confident to say now we're in a place where 10 or more customers have done a decade with us and made millions of wafers and hence, billions of chips with us. And so we have that foundation of trust with our customer base and with our ecosystem. And then the last point you mentioned, it's in the name, is that global footprint. And it's really hard to build. It takes a lot of time. It's actually quite painful to establish not just a set of fabs in different geographies, but ones that work together as 1 global network that have cross-qualified technology, so a customer can take out once in manufacturing 3 continents. That's a huge advantage in a world where supply chains are more fragile than ever and more -- and need more resilience than they've ever had before. And so again, when you put those 3 together, it's clear why customers are coming to us more and more for what we do.

You hosted an Investor Day back in May, we laid out some of the long-term growth drivers of business. I think you called them megatrends of AI data centers, physical AI and resilient global supply. So maybe unpack some of these and highlight the specific areas where you have to expect to have the greatest impact on your business over the next few years.

Yes. So we thought in the data center, I think, obviously, it's the area that everyone in the industry, everyone technology is very focused on. And if you leave aside for a second, just the shift scale of the build-out that's happening, we talk about the $800 billion of hyperscaler CapEx this year. Those numbers growing into '27 and beyond. But leave aside the scale of it and talking about what's happening within that data center today, that was different from 2 years ago or different than even a year ago. And it's that data center is increasingly heterogeneous in terms of the compute that's going into that. And so while before it was a GPU-centric kind of build out, now you're talking about GPUs, CPUs, LPU, NPUs, basically anything with the PU after it, working together in a much more complex compute environment because the AI workloads have changed. They were originally quite narrow in scope, now heavily skewed from training now to inference and even disaggregated inference workloads as well. So that introduces a new challenge called how do you connect all of that together. And so in that data center, we're very much focused on the optical connectivity piece of it. We'll talk more about that. I'm sure there's a lot of growth in that sector as we transition from copper to optical. But beyond that in data center, you're also seeing an increasing criticality of improving power conversion and power efficiency. So again, data center power being a very good growth driver for a business like us. They're specializing in the technologies for intelligent switching, dents power conversion and so on. So data center, very, very strong driver today where we're seeing very, very good demand growth where we're also investing CapEx for capacity for the short-term ramps. We think that is probably dwarfed by what the longer-term physical AI transition will be. And it's a bit like every transition we've seen in technology, you start with the boom and everyone focuses on the boom, but we forget that there's always an echo that comes later. And the reason we think that's the case is that when AI permeates the devices in our world around us, there are so many more use cases that can be deployed that we haven't thought of yet or we haven't taken full advantage of yet. We haven't seen that really materially hit the business yet. But of course, we see increasing interest in robotics, but also industrial automation, medical devices, agriculture, consumer devices, a broad spectrum of devices, all of which will need a lot more semiconductor content because operating in the real world, they all need to sense their environment, think with distributed compute, control a real-time system, actuation, motor control and so on. And they went in a ton of communication, largely wireless communication. So the second driver of physical AI, very much at the beginning and the early stage but we think that is a material business growth driver for us. If those 2 were all that was happening, that would always be pretty overwhelming for us and overwhelming for the industry, but they're also happening at this time where this reshoring trend is now truly accelerating. And for a long time, it was interesting to customers that we had a global footprint, but it was not a competitive differentiator. It wasn't something that said, okay, we need to do this with you because of the footprint you have. In the last couple of years, that has dramatically changed to the point that now we see procurement processes where NCT is the rule not a nice to have, right? Made in the U.S. is a critical feature of that. And so it's almost a multiplier effect on the first 2 that all of this demand growth is happening at a time when the industry is in a very committed way driving that reshoring of capacity into the U.S. but also into other parts of the footprint as well.

Great. I want to dive into those things, but maybe first ask about operational kind of question. what are 1 or 2 key objectives you're driving toward over the next, say, year or so? And if you think about investor expectations for the business heading into 2027, if you were to outperform, what would be the key areas of upside you see for the business.

So it's interesting how our industry moves through these periods of when you're chasing supply or chasing demand. We are clearly in a chasing supply moment today. And for us, what that mostly means is a deep focus on ramps of technologies that we are already producing today, but ramping to higher volumes, we'll be definitely increasing our CapEx. We already are increasing our CapEx. This year from last year, we'll be increasing further again next year, but it's all going to come down to how quickly we can get tools installed, qualified, ready to ramp for customers. That will gate or accelerate our growth going into next year. We don't see a demand-based reason that, that would be different. We definitely see a supply-based reason that, that will be different. And that puts us in a very much more executional phase of our business, right? We're delivering for customers working hand-in-hand finding ways to take risks together to go faster to meet their objectives for their markets. So priority #1, job #1 is really just build out that capacity and ramp as quickly as we can in those technologies that we are investing in, and really doing that in a very much partnership mode with our customers. By the way, not just in terms of commitment, but even in terms of like financially, right? They're also co-investing alongside us in some of those ramps as well. So that's priority number one. Priority #2 is in that broad set of technologies that we offer, every year, we're delivering critical road map intercepts, qualifications of new products, new technology launches, we just had our Global Technology Summit here in in Silicon Valley last week. We've never had as rich or diverse a road map as we do today in all these different end markets. But those are things that customers are counting on. They're counting on it being ready, first time right. And so job #2, very close second priority is delivering those road map intercepts for features and performance that they need to win in their end markets as well. So high execution focus.

Excellent. Now clearly, AI data center is 1 of the bigger drivers of the growth in the overall market. you have a differentiated position in silicon photonics specifically. Walk us through that competitive differentiation relative to some of your competitors like Tower Vanguard, et cetera, and just some of the actions that TSMC may be taking into the market as well.

So if you back up a little bit, the transition to optical, if we had this conference discussion a year ago, 2 years ago, we would be having an if discussion, not a when discussion. We're clearly now in a when moment for optical and it's because AI data center workloads are much, much more data-intensive than previous workloads that were ever considered in being. And so now we're firmly in that camp of transition, and the transition has kind of 2 dimensions to it. First of all, a transition from long distance kind of coherent optics that scale across to scale out and eventually scale up and can potentially scale in, right, all the scales, right, in the data center in terms of which links are being replaced by optical that were historically copper. But the other transition happening is form factor transition where today, you see the strong ramp of pluggable optical transceivers. But tomorrow, you're seeing even taking out enough has today, near package optics and co-packaged optics designs that are going into production in '27 and '28. And that's because you need to deliver that optical link closer and closer to the XPU, the GPU, the CPU, the switch ASIC and so on. So this is the backdrop of what's happening, and it matters because our differentiation isn't just about capacity today. It's very much about a road map that can deliver the performance that those transitions need over time. And so you have to have a core silicon photonics platform, which we've been building for more than a decade that can do high-speed modulation. So we're now talking about 100 gig per lane 200-gig per lane, 200 becoming 400, going beyond 400 gig per lane with new materials, new modulators and so on. So a core photonics platform that can accommodate the acceleration of bandwidth. But you also need to think about how that form factor is delivered that's where advanced packaging comes to play. As these solutions are delivered closer to the chip, hybrid bonding, ability to attach a fiber to the chip, these are all difficult and new techniques that were in the industry before. And again, part of our road map of the innovation we've been driving as well. So you need that technology platform. You need the capabilities to work with a diverse set of customers. We have more than 40 customers in photonics today, and that number keeps growing as players really invest in their optical road map. By the way, everyone who was touting copper 2 years ago has either launched an organic or inorganic effort to get into optical, right? And so suddenly, everyone is doing optical, but that means you have to support them in these new ramps, these new capabilities that they're in with not just your team, but your models, your PDKs and so. So the second piece of that story. But the third piece, maybe the most in a way difficult to replicate is, this is a fast-growth market. And so people are not betting on you to sell a few wafers today. They're looking to how do they scale for the next multiple years. We're ramping our photonics capability in existing scaled 300-millimeter fabs. These are big fabs. These are fabs that do today, hundreds of thousands of wafers of leading-edge, effectively leading-edge capabilities with lithography like DUV and so on, that are ready to scale for higher-performing applications in the future without building new fab space. And we've talked about being able to 10x our photonic capacity within our existing 4-wall infrastructure in the U.S. and in Singapore to really meet the needs of that market going through that high ramp period as well. You put those 3 things together, it's very clear why we're having a really good engagement with customers in Photonics today.

And to that point, you laid out growth targets of getting to a $1 billion growth run rate in Filcon Photonics in '28, $2 billion by 2030. How do you see that trajectory in the very short term? And then what would we have to see in either near package offer co-pay optics for you significantly both passive circuits out?

I think the industry adoption needed to reach those targets has already happened. So I think we now have very limited adoption risk baked into that. Now we're in the mode of again, chasing supply in the near term, how quickly can I qualify tools? Can I invest and ramp that capacity within my existing footprint? All indicators right now from a demand point of view are flashing kind of bright green relative to those targets. It's a question of how quickly I can build -- bring that supply on. And obviously, I'm going to do that in an intelligent way, understanding that when you lead with capacity, you've got to balance it from a long-term kind of profitability point of view as well. But right now, we see much more demand than we can supply out of our current installed base, and we'll definitely be accelerating that. That's still primarily today that path to that first milestone. Pluggables gets you pretty much all the way there. We're already seeing the NPO rollout picking up steam. And by the way, it's many of the same players who are imploggable today rolling out their NPA solution. So I think that transition is now very much inevitable. If you saw really broad adoption of co-packaged optics the latter part of the period, you could blow past those targets materially because co-package is the area which is critical for that scale up transition. Once you're moving into that, you're replacing 10x the links of scale out with those solutions. And so you're not going to see a linear growth story. You're going to see a much faster growth story on top.

Another key thing in them in the data center is high-voltage power, specifically for the 800-volt transition, we're expecting to happen sort of maybe by like middle or the end of 2027. Can you lay out for us your exposure in those areas, specifically for gallium nitride and BCD MOS?

Yes. So as you said, we're seeing a paradigm shift in how power converted while it's converted from 2. As you said, next year's data centers will be having 800-volt at the data center wall that needs to get delivered to 0.8 volts, let's say, at the chip, with as few conversion stages as possible in a very dense form factor, right? The rack is dense than it ever has been. The amount of current that's used by all of these processes is very, very high. And so you need technologies like BCD for intelligent switching. You need to do dense power conversion with GaN. The same reason your GaN fabric for your laptop is much smaller than your historical power Brick, is the same reason that the data center relies on GAM for that. We're shipping GaN products to the data center today. One of our partners, Navitas, talked about that last week. So that ramp is already underway today. And BCD is the same story. And our latest BCD technology, 55-nanometer, which is not from a digital point of view, very complex, but from a power point of view, it's kind of really very advanced. It is already being used in these new conversion stages that are going into production as of now. So that ramp is starting. We have yet to see the real pull-through for the business, but we'll see that going into '27 and even more into '28, which is when we're also putting capacity on for that technology. The latest in innovation in the data center is also the switch from lateral power delivery to vertical power delivery. There's not enough space in the rack. And so you need to find ways to deliver power vertically to the chip. We've been making investments in this area with process technology, integrated passives, but most recently with the acquihire of a small team focused on designing those voltage regulators with customers. Because again, when you make a rapid technology transition, it's not just about having a standard offering and use it if you want to use it, work as partners with your customers, be they a traditional power semiconductor player or a hyperscaler or anyone in between to really help them navigate that transition. And that's exactly what we're doing.

So very fair. So even as you're buying into these markets, we just talked about, you still have a significant amount of your business today tied to the smartphone market. I think a lot of people saw memory pricing driving down potential demand destruction, but I think it's maybe played out in a little bit different way than some people expected, some pull-ins happened in terms of builds rather than driving sort of immediate downdraft. And I guess, how do you think these dynamics now with the benefit of hindsight are going to play out in your business in the -- talk about what you saw in the first half and then how you expect that to kind of materialize in the second half?

So we've been very deliberate about our approach to mobile and mobile has taken this industry quite far and it's been an important driver of how many companies got to where they got to, right? TSMC, by the way, include it, right? We were more than 50% mobile going back a couple of years. That's down to the mid-30s today we've laid out a long-term model that takes it to the mid-20s. And so we think that over time, as a percentage of revenue, it decreases. In absolute terms, it remains relatively flat or, let's say, to achieve our plan, we do not rely on it growing to achieve our medium-term plan. And that's very deliberate because again, we want to see the growth coming from the other areas where we're driving the diversification even harder. But the humble cell phone still has a role to play. And by there, it creates also a platform for technology innovation. The BCD technology we're excited about for the data center, it was born in the mobile phone, right? And if you can deliver it in that form factor, you can deliver it in other 4 factors, too. Look, I think partially, when we started the year there was a view that it would be a tough year for mobile, it is indeed a tough year. It's not as tough for the higher end, right? I think it's very tough on the low end, more of our exposures to the higher-end handsets. We'll see what announcements including this week's announcement from the Cupertino company, what the yield do they drive appetite for a refresh cycle, what does pricing really due to demand. I'd say it's holding up better than perhaps some people thought because, again, there's still just a natural replenishment rate. I think the other dynamic that we're seeing is mobile companies were the first to the queue in COVID. They understood that by getting early, they would not be the 1 shortage in supply. Now they're worried about being late when the data center is the 1 pushing all the demand. And so you're seeing actually more visibility where they're actually still saying, I need my supply for '27. I need my supply for '28 because we're still going to sell plus or minus a large number of units of mobile, even if it's down low double digits or so on, it's still a material amount of content that needs to be secured.