Garrett Motion Inc.

Okay, so I know everyone's still getting settled, but in the interest of time, we're going to keep progressing. So it's my pleasure to host Susan Lee, the CFO of Meta Platforms. Susan, thanks so much for being part of the conference.

Thank you so much for having me.

Okay. So I do have to read the safe harbor first, so please, everyone, bear with me. Some of the statements made today by META may be considered forward-looking. These statements involve a number of risks and uncertainties that may cause actual results that differ materially. Any forward-looking statements made today by META are based on assumptions as of today. META undertakes no obligation to update them. Please refer to META's most recent Form 10-Q filed with the SEC for discussion of the various factors that may affect actual results. Okay.

The very safest of harbors.

Yes, we are fully safe harbored up now. Susan, the company's been on a significant journey over the last three years since you took on the role of CFO. Why don't you talk a little bit about, at a high level, the balancing of where the company wants to go in terms of investing for growth and achieving scale across multiple opportunities, but also driving efficiencies across the business at the same time?

Yes, great question. So I took on, I've been at the company for 17 years, which feels like a real lifetime, but took on this job in November 22, which some of you may remember as being maybe like a local minima in the financial trajectory of this company. So it's been a sort of kind of difficult, but I think also exciting in many ways, sort of climb from there. And looking forward, you know, we are excited, frankly, to have a portfolio of opportunities and investments that kind of span the range from your sort of Near term, very measurable. We've got very sort of robust instrumentation for how a lot of our core ranking and recommendations work pays off, both in terms of the sort of benefit to core engagement, the way we think about user engagement, and, of course, the sort of benefit to monetization. So there has been for a long time an ongoing pipeline of projects there. And actually about a year ago in the 2025 budgeting process, we're now beginning to kick off the 2026 budgeting process. In the budgeting process, you know, we had a pretty big portfolio of asks from those sort of family of apps and monetization teams. We funded a lot of those. And I think one of the things I had actually said about the time, you know, when you're looking at sort of this portfolio of asks, Each ask makes a lot of sense. You're like, okay, great, you need 25 engineers, they're going to build this thing, and this is what we think the return will be. And the thing that's a little bit unknown is, but like, where are you on the sort of marginal curve of returns right now? So sure, maybe we know sort of each 25 engineer unit of work is going to generate some amount of return, but what happens if you add, like, a thousand engineers? Like, how quickly does the curve drop off? And I'm happy to say, you know, I think we funded a lot of those investments, and a lot of them have, you know, been paying off for us. So we've been seeing that in our results. And so there's, you know, still a lot of sort of great sort of work to do there that's kind of an ongoing pipeline. But we also have, you know, in the medium and further term, a lot of really exciting work that is happening also. And that includes the sort of big bets that we're making in the AI, not the core AI, but sort of the generative AI landscape, the work that we are doing to build frontier models. I think there's been a little bit of reporting around some of the work that has coalesced over the summer. And then what we hope to use those models to do and to build and to kind of take advantage of the distribution platform and data flywheel that we have when we're able to put those experiences in front of a lot of people. And then in the even longer term, sort of building the, if that's sort of the foundational AI model, building what we hope will be the foundations of the next computing platform and How do you bring those AI experiences out with you into the world? Obviously, glasses are, in their current incarnation, are a very premature but exciting sort of form factor in terms of what you can take with you into the world and how does that evolve over time. So we've got things on all of those time horizons and with all of those levels of certainty and measurability. And, you know, there's no sort of... kind of magical sort of formula that links them all together in terms of how you think about like how you fund today's roadmap, how you, you know, how you allocate your resources across that portfolio of projects. But, you know, we have, we've been mindful, I believe Mark has publicly committed to delivering operating profit growth. And I realize that that is, you know, not standalone a benchmark that is extremely exciting and that in practice we in fact have to make sure that over the long run, you know, we are an attractive investment relative to any of the many other public equity investments that are available to all of you. So, you know, there will be obviously like lumps in the years. It would be a truly amazing thing if you could sort of just deliver nice linear compounding returns in a predictable way forever. But we're committed to making sure that we deliver attractive financial returns over time and across, you know, and we sort of think about managing the portfolio of investments in that way. I will add one more thing because you mentioned sort of AI driven productivity gains also. This is a place where we are really, I think, you know, we recognize that the tools and technologies are evolving very, very quickly. And we are really trying to push our own internal teams to become sort of first adopters of a lot of the tools. to figure out how to make sort of their teams substantially more productive in whatever the area of work is. And I think actually it's unclear to me how that impact will net out. You can imagine that there are teams like if each of your engineers can produce twice as much product impact because the AI tools have made them twice as productive as they previously were, then, you know, we should probably hire a lot more of those engineers. There are sort of, you know, other areas around the company where I think this will be more of an efficiency gain than like something that multiplies the volume of output that you're able to produce. So it will differ by group. But we are leaning very, I think we're leaning very hard into it. I suspect many of our performance reviews are increasingly written by AI tools and, you know, for the better, probably. So both in terms of not only saving people time, but probably more, frankly, comprehensively reading all the reviews that are written, looking at sort of, you know, the different like system metrics, diffs, committed, all those things, and probably producing kind of a better holistic view. So you've got a lot of, I think you've got a lot of opportunity, even in the really mundane stuff, to make processes better and more efficient. And then, of course, as a software and product development company, the number one thing we care about is just how do we make the process of building consumer products and experiences as efficient and productive as possible.

Got it. Okay. There was a lot in there, and I think we're going to touch on some of those topics as we go through the conversation as well, but maybe turning to the core advertising business. You've pretty consistently outgrown the digital advertising industry over the last 12, 18 months, despite your scale. When you think about the algorithm of growth going forward, what are some of the building blocks that could sustain advertising growth, and how do you think about them as either being impression-driven versus pricing-driven?

Okay, I'm going to do my best to You can call me out on time. My high school English teacher told me that brevity is the soul of wit, said Polonius, and I have never been able to internalize that. So you can cut me off when you need to. So kind of the core business, right, if you think about the building blocks, you've got on the supply side, you've got basically users of the platform, you've got the amount of engagement they spend on the platform, then our ability to monetize that, add load is sort of the traditional way, and then you have sort of pricing on the demand side. So on the supply side, you know, we still see, obviously there are many markets now, especially in more developed markets where user growth is hitting sort of saturation levels. But nonetheless, you know, we're growing users globally across Facebook and Instagram and WhatsApp. And there are still markets where we are not close to saturation yet, where there's a lot of growth to go. We have found that there just continue to be opportunities for us to, improve basically the performance of our core ranking and recommendations engine that powers sort of the, you know, what you see when you use our family of apps. And we've been really just happy with, frankly, the pipeline of those investments and how they pay off. We talked about this a little on the Q2 earnings call, but I think, you know, we sort of ranking optimizations helped drive another, I think, 5% lift in time spent and 6% on Facebook, 6% on Instagram. And there's a lot of work that we're doing to continue to really try to make your experience more personalized to you, to make it more adaptive to how you are engaging with whether it's Facebook or Instagram, whatever part of the product that is your wheelhouse, to make it most relevant to you as you're using it to adapt to your behavior on the personalization side. We're also doing a lot to try to make sure that we are surfacing really like the most timely and fresh, freshest content. And that's especially important for newer and younger creators who are like creating content, you know, something that is a, you know, maybe a meme about something that just happened in the world. That's going to be really interesting and funny for 24 hours, but it might not be for three weeks. So if you want to help that sort of creator be able to break through, you've got to do it immediately. You can't sort of hope it percolates through the system and gets to you like four weeks after the event happened. So really trying to make sure that we are helping sort of make sure content recommendations are very timely. That, again, is particularly good for creators also. And so there's just, I would say, a lot more work to be done on the core engagement side. And then with Adload, you know, it is also a story of personalization and of increasingly trying to infer when you are using our product, when you are in a session, Are you interested in buying something? Are you in a commercial state of mind? You know, if I just bought, you know, like binders and rulers, this is back to school is top of mind, now is a good time, you know, for me to buy like notebooks and protractors and calculators. And, you know, that's a great time to show me more ads. And then there will be times when I'm like clearly scrolling through friends and family content and probably not thinking about, you know, about shopping. And that's a good time to show me fewer ads. And that... enables us without any meaningful sort of engagement impact to really optimize the impressions that we show you and increase the value of those impressions. And so there's a lot of work that's being done, I think, to really make ad load. It's gone away from kind of like 12.5%, the one in every eight stories is an ad, to something that feels really, really sort of tailored to when you are most likely to want to have a commercial experience. So that's all on the supply side. On the demand side, you know, reported CPM is really an output of the work that we are actually doing to drive prices down, right? And so all of these sort of efforts that we have to sort of improve the performance of our ads, really what we're trying to do is make any individual ad convert more frequently and to drive higher value conversions, right? So those are the two things. that they're really trying to do. And if we're able to do that, then even as we are bringing the cost per business objective, the cost per acquisition, whatever it is that the advertiser is looking for, even as we bring that cost down, you should see reported CPMs go up because we are making each impression convert more frequently and be more valuable. So that is a place where it is a little bit confusing when you think about what does rising prices mean for us when we report rising CPMs. That's often not reflective of what the fundamental sort of cost per acquisition is for advertisers. But it's something we're able to measure on sort of our side and try to normalize across different conversion types, obviously like mobile app installs and like, you know, e-commerce purchase are very different value conversions. And when we normalize for those, we feel really good about sort of our ability to drive, again, the cost per business objective down and ROI up for advertisers.

Maybe just one more on the core advertising business before we keep moving along. Probably the number one question I get from investors is just the role that AI plays today in driving outcomes in the advertising business and how that's going to evolve in the years ahead. You've launched a number of products that have AI at their core in the advertising solutions. Talk to us a little bit about what you're building to and how should we be thinking about that as a scaling effort in the years ahead?

Yeah. Okay. So this is like... This has the potential to be an epitome on its own, but we're going to restrain the Beowulf here and really try to keep it pithy. So, okay, there's a lot of work sort of on the back end today. We have sort of a very complicated ranking and recommendations back end that is sort of separated very broadly into, you know, we think of like the ads retrieval stage where there are, you know, tens of millions of possible ads for any individual person. And we basically have machine learning models that take that and retrieve several thousand to send then into the ranking stage. In the ranking stage, we figure out what is the sort of right order and time and sequence in which to show you those thousands of ads and then actually deliver the ad to you. And there are some other things that we factor into, you know, advertiser bid, the sort of estimated impact on sentiment, things like that. And the models, you know, so there are sort of very complicated machine learning models that power both of those. We've talked about them, I think, a bit on earnings calls. Andromeda is the name of the model that powers the ads retrieval. And then GEM is the name of the model that powers ads ranking. And in both of those cases, there's been a lot of work done to sort of basically refine the models, scale up their complexity, enabling us to basically retrieve more ads and rank more ads at a similar degree of efficiency as we have in the past that helps make sure that the ad is more relevant to the individual user and And in each case, these are things where, you know, in the case of Andromeda, I think we've rolled it out mainly across the Facebook surfaces and Q2 mobile feed and reels. And now the sort of forward-looking work is to roll it out across Instagram. And then we're scaling up the sort of complexity of GEM. And then there's what we call the sort of meta-lattice architecture that we use to then broadly scale our models from individual surfaces and objectives, which is how they all get developed, to sort of trying to make this basically a more like a global model that is doing ranking and recommendations work across all services and objectives at the same time instead of these sort of focused individual models. So there's a lot of work that's happening on the ads back end to sort of continue powering the sort of growing complexity of those models. And I think over time we're also going to find that we will use more sort of llm architecture also to think about how to power ranking and recommendations work that's relatively newer for us, but that is another sort of. that's another much earlier stage project that I think has the potential for a lot of a lot of upside in the sort of the recommendation landscape. And then on the front end, right, from the sort of advertiser experience, you know, advertisers right now come in, Advantage Plus is the name of the sort of front end tool that we make available for advertisers to create their ads. That is a tool where there is sort of a lot of AI-powered automation to basically try and streamline the ads creation and campaign process as much as possible. You know, who your audience is. If you think about the history of sort of targeting very specific demographics, now we basically try to serve your ad to the most likely and relevant audience base, and you don't have to tell us that much about who they are. We're going to be able to make those inferences for you. Setting budgets, how you allocate your budget across different campaigns and the campaign set, things like that. So we're trying to automate all of those things. And then finally, I think the sort of next frontier is is using Gen AI creative tools to make the creative. So right now in Advantage Plus, image expansion is sort of our most commonly used Gen AI tool. And it's basically if you upload a creative, you've probably made like one or two, we are going to figure out how to size this to all the different possible ad formats that we have so that you don't have to figure out how to upload like 12 different ads. You upload one and we expand the pixels as needed to fill space and so on and so forth. Text translation is another popular one. And I think kind of the next frontier for us is video generation, taking a still image and using it to generate a video, in particular because still ads don't feel native in Reels. So if you're in Reels and you hit a image ad, it's actually a little bit of a jarring experience. But a video ad feels very seamless, feels like this is just part of the Reels experience. Video generation, I think, is kind of the next frontier. That's the place where a lot of active work is happening right now, frankly. And then I think in the longer run, the two sort of things I'm most interested in are, excited about are, one is sort of ads becoming more interactive with you. I think that's not something we see at all today. I think that'll be super interesting and part of a general trend of content becoming more interactive with you. And then the second thing is just the idea that ads can be super, you know, super tailored to each person without the advertiser knowing that, you know, you and I could get the same, like a hotel in Hawaii could target you and I with the same ad. It's just trying to reach you and I and get each of us to go to this hotel. But we know that you should get an ad that's oriented around like big wave surfing and I should get an ad that's oriented around like hiking. And it just creates those for us because, you know, it knows that those are our interests and the hotel doesn't have to We do all that on behalf of the advertiser. That's the goal. So all this is to say the arc of the advertising universe is long, and I think we're still pretty early in what's possible.

Okay. So continuing the theme of AI, you know, Mark has laid out. his vision for how the company's focus around superintelligence might evolve in the years ahead. Talk a little bit about how this company is uniquely positioned to execute on that vision, and then some of the challenges about delivering compute capacity and personnel to deliver on that vision.

Yes. This is sort of one of the most – I mean, frankly, this is just an extraordinarily exciting time to be working on this problem. So I feel really very lucky to get to be a part of this. Look, I think clearly the sort of rate of evolution of the way sort of like frontier models are evolving I think is very fast. I think the capabilities that we are going to sort of see applicable to our everyday lives, productivity, and other use cases I think are going to be tremendous. Obviously, we're very excited about, I think, you know, scientific and sort of economic advances that I'm very hopeful for. For Meta specifically, I think a really interesting angle is, you know, how do we make this experience very, how do we make these sort of technologies very applicable to you personally and your goals and your creative output and the way you sort of express yourself and the way you share with the world? It's, you know, we think of it as sort of a very personalized experience, and I think that is rooted in the fact that today we deliver to, you know, billions of people around the world an extremely personalized experience. Each of you using Facebook or Instagram has a totally different set of content that is, you know, that is being shared with you than any of the people around you, right? And so for us, the idea that, you know... the AI experiences we build should be an extension of that. So it should be sort of a sort of very personalized experience to you. How do we make the content that you see sort of more interesting, more interactive, more engaging? How do we give you the tools to create whatever it is that you want to create and, you know, put out into the world? How do we enable you to sort of more productively engage with the people you care about engaging with, you know, um, or undertake projects that you're excited about. So for us, I think the landscape here is we really think about kind of the AI efforts through the lens of building deeply personalized experiences. Now, in terms of, you know, what it takes, we've got, you know, I think it takes talent, it takes compute, data distribution. I think increasingly it takes, you know, a lot of capital Those are all things that I think we have. And it is a, you know, a very sort of exciting and competitive landscape, frankly, in terms of the sort of all of the folks who are engaged, you know, in doing this work right now. But specifically, I'll say, you know, on the infrastructure side, we are finding that the sort of amount of compute that we think it will take to do pre-training and distillation and post-training and enable us to build frontier models, you know, are sort of ability to see what we need there is that, you know, we need more compute and then we want to be prepared for what the inference sort of use cases, you know, will be. And so we are, we think we are sort of at the forefront of this. We've got some big, exciting infrastructure projects going on. Our sort of first gigawatt plus cluster is going to come online next year. We have called that Prometheus because we've got a lot of history nerds at the company. And we have a sort of 5-gigawatt project that is coming online that will have the ability at least to scale to 5 gigawatts if we need that. And then on the talent side, I'm sure, you know, probably a few of you have seen a few headlines here or there about our team-building efforts over the course of the last few months. But we're very excited about – what we call TBD labs. It's a placeholder name that sort of stuck and feels actually very appropriate in the sense that I think, you know, a lot of what the team is going to build is sort of yet to be sort of precisely shaped or determined. But it's a, you know, we conceive of it as sort of a pretty small, a few dozen people, very talent dense set of folks. They are kind of working on the next generation of, foundation models, and we hope that those will be sort of at the frontier over the course of the next year or two. So we're pretty excited about all this coming together.

Okay. One more big-picture one, and then we'll go into sort of rapid-fire mode. But how central is meta-AI to your AI efforts, both within the family of apps, existing applications today, and how it informs what reality labs might produce as computing experiences longer term?

Yes. Meta AI is definitely an important sort of part of what we're building. It's, you know, pretty interesting. Meta AI is not today powered by a frontier model, and yet it is very widely used. We find that, like, there are a lot of really interesting use cases on it, and we find that the experience improves a lot every time we sort of improve the underlying model. So we're very sort of optimistic about the sort of the kind of trajectory ahead of us there, and especially, again, as we build frontier models. But also, I think the notable thing about it you were asking about Reality Labs is the sort of, you know, what is the form factor by which you are going to carry sort of AI technology with you into the world? And I think for us, the glasses form factor seems very intuitive. You know, it is the best way for AI to replicate the experience of what you are seeing and hearing and doing for it to be able to interact with you, whether it's talking to you, for example, as you are moving about in the world. And so this is a place where, I mean, obviously, I think there will be other interesting modalities that get developed over time, but glasses feels like a very intuitive one. There are like a billion glasses wearers in the world today, not counting sunglasses wearers. So But, you know, there's like already – it's like a very normalized experience, and it seems obvious that at least for, you know, for example, those people, it should make a lot of sense that they might choose to switch from, you know, regular to smart glasses. I may have to bring – I wear contacts because I'm like blind as a bat, but I may have to bring glasses back. But I would say – The sort of AI experience, you know, experience on glasses right now is pretty nascent, but there are a couple of things we are, and, you know, and as it is today, which is to say a pretty limited AI experience on glasses, RBMs are doing super well. We are having, you know, we're having trouble keeping them in stock. Frankly, we're trying to ramp up supply for the second half of the year. So, you know, growth in RBM sales accelerated in Q2, and we're happy about those. And we have sort of new models to announce. We'll be sharing some of them at Connect. The Oakley Meta Houston's, for example, those have ultra high def video, better battery life. It's like a more sports and performance oriented glasses. So, you know, if you really need to capture what the experience of hurtling downhill, you know, on skis looks like, those will be the glasses for you. So we are, you know, I would say we're pretty excited about that. about the portfolio of glasses as they exist today, and they're just going to get better because the AI experiences will get better. You know, the live translation that we just rolled out for English, Spanish, French, and Italian, I think, like, that's, in my mind, what the future was supposed to look like. Like, you wear glasses, and you talk to someone in one of those languages, and you hear what they're saying, and, you know, in my case, it would be English, you know, and they, if they have glasses, can have the inverse experience, or if they have a phone, then they can read what you're saying translated into their language on the Meta AI app. So I think this is a place where the sort of forward-looking developments are going to be really exciting.

Okay. Quick two-parter. You gave a framework around investments on the last earnings call for the next 12 to 18 months. Then Mark made some comments at the White House about spending $600 billion in the U.S. by 2028. How should investors think about that broad framework of what you want to invest in And the second part would be how should investors think about earning a return on that investment cadence in the years ahead?

Yes. Well, the way that we talk about these things reflects accurately that one of us is a CFO and one of us is a builder and tech visionary who runs one of the largest companies in the world. But there has been a lot of excitement or questions, too, about Mark's comments. So just to clarify, Mark's comments are referring to sort of the total envelope of our planned U.S. investment from this year, so including 2025 through 2028. So that includes, obviously, all the data center infrastructure that we are building in the United States, but it also just includes, like, all the investments that go towards supporting our U.S. business operations, all the people we hire, you know, in the U.S., you know, where our biggest offices are. So that's What that is referring to, and obviously we don't have a perfect crystal ball, but that's kind of the best line of sight we have today in terms of what we think we're going to be spending in the U.S.

Okay. So just the second part of the question, when you think broadly around some elements of the return profile, just curious because you referenced earlier operating profit growth. How should we be thinking about return profile if there's a prism that you want to share?

Yeah. I mean... You know, we don't have, you know, anything I think meaningfully more specific to share today. I think the most important parts of the framework here for us, again, are that we have this sort of balance of newer term, higher certainty, very measurable projects. We have these sort of medium and longer term sort of portfolio of things that are less sort of high fidelity in our ability to build a revenue forecast and timeline today. But at the same time, you know, we think of those in a little bit more of a VC style, like what is the set of opportunities we could unlock? If you have to unlock all of them to make the investment work out, then that's probably not a great investment. But if it's a place where a probability-weighted, you know, set of returns that seems achievable is going to justify the investment you're making, then that seems like a sort of, you know, a reasonable path to embark upon. And then we really weigh all of those things sort of in kind of a cohesive, like, can we do this? And, you know, do the near-term sort of high ROI investments today that we have give us the right to continue investing in these longer-term sort of more uncertain projects? And how can we navigate that over the, you know, the upcoming years and do that in a way that, you know, continues to deliver sort of solid financial returns and That's really the framework, not like a margin target in part because if we evaluated new projects relative to our existing business, a lot of projects would frankly look not as attractive. And so we're really focused on growing like operating profit dollars more than sort of any other metric and doing it in a way also making sure that we return capital to shareholders in a thoughtful way, offset the sort of equity dilution that comes from compensating our employees and continuing and growing our dividend program over time.

Maybe I'll squeeze just one more in on CapEx because there have been press reports around the company possibly partnering with external parties to look at elements of funding some of the capital needs of the company in the years ahead. What's the framework we should be thinking about about how much you need to build this yourself rather than look across an array of partners to possibly deploy capital and build capacity against the longer-term vision Mark's trying to build to?

Yeah. So, you know, for most of our history, we basically, you know, built O&O data centers. And now as the sort of, frankly, the ambition of our infrastructure capacity unfolds ahead of us, it kind of dwarfs what we've built before and we need to be sort of more expansive in the way that we are thinking about this. So we haven't announced any particular transactions yet, but we are looking at sort of partnership opportunities and kind of financing structures that will enable us to sort of achieve some of the flexibility that we are looking for in kind of a longer run timeline of these data centers. The fact that there's a lot of unknowns over these sort of 20 year life cycle of a data center and how it might, you know, be used over that 20 years, while also sort of being an attractive, you know, project, obviously, for investors. So we're looking at some structures there. We're also looking at, you know, more traditional, frankly, like cloud leases. I think the next speaker, you know, may be able to opine more on that. That's a good business to be in right now. But there's a whole range, I would say, of projects financing options from our own balance sheet on one end to fully leased on the other end, and we're kind of looking at everything.

Susan, thanks so much for making yourself available. I really appreciate the opportunity to have the conversation. Please join me in thanking Susan and Metta for being part of the conference. Thank you.