AppLovin (APP.US) Q3 Earnings Call: CEO Rebuts Competition Squeeze Narrative, Partial Launch of New Model Already Accelerating Advertising

AppLovin (APP.US) reported third-quarter 2026 results after the U.S. market close on Wednesday, with revenue of US$2.185 billion, up 55% year-over-year, diluted earnings per share of US$4.30, and adjusted EBITDA of US$1.842 billion at an 84% margin. All three hit record quarterly highs, coming in roughly 6% above consensus expectations of US$2.07 billion, US$4.04 and US$1.73 billion, respectively, and also exceeding the upper end of the company's own 46%48% guidance by about US$100 million. The report was released after the close; the stock closed the regular session up 10.69% at US$271.75, with after-hours gains widening from about 17% to about 23%. The focus of the call was whether the explanation for the second quarter's "late model improvements" could be validated by the data: management said the new-generation model was only partially rolled out, and gaming advertising revenue immediately grew nearly 50% year-over-year and by more than US$200 million quarter-over-quarter. CEO Adam Foroughi said, "A quarter ago we were here answering questions about missing expectations... everything we said is reflected in this quarter's numbers," and reiterated that roughly 30% annual compound growth is sustainable long term. Market Focus and Response In the second quarter, AppLovin fell as much as 28% after hours after revenue slightly missed consensus and guidance disappointed, with the market questioning whether its AI model iteration had stalled; a securities class action later emerged alleging the company overstated the durability of its technology. This quarter's revenue growth therefore became the focus: actual growth of 55% marked a rebound from the second quarter and was above the upper end of management's 46%48% guidance range. Foroughi attributed the beat to the model rather than demand, saying the new-generation model "performed strongly, advertisers spent more budget at their target ROAS, and budgets were then raised." He also denied that competitors' growth had squeezed the company's share, saying "that was never the casethe problem was on our side." But management also stressed that model gains would not be released evenly by quarter and did not commit to a timing for full rollout, saying "we won't commit to a cadence." All three core metrics hit record quarterly highs. Net income was US$1.439 billion, up 72% year-over-year, with a 66% net margin, up 7 percentage points from a year earlier; adjusted EBITDA margin was 84%, up about 190 basis points year-over-year. CFO Matt Stumpf said revenue exceeded the upper end of guidance by about US$100 million, mainly from the core mobile gaming business, with the consumer vertical also scaling. Data Center Costs as a Share of Revenue Actually Fell Despite increased compute investment to train new models, the infrastructure team optimized how models are served, reducing data center costs as a share of revenue. The conversion of incremental quarter-over-quarter revenue into adjusted EBITDA was 87%. Stumpf said the company manages the business by EBITDA dollars and free cash flow, not by margin percentage. New-Generation Model: Technical Complexity on Par with AXON 2.0, No Timing Commitment Foroughi said the new architecture's technical complexity is on the same order of magnitude as AXON 2.0, but the company does not name it by version number. Compared with the old architecture it replaces, the new model "gives us significantly more room to optimize," with the most pronounced improvements in data-sparse scenariosnew advertisers, small advertisers and new categories, which is exactly where the consumer vertical sits. Gaming Advertising Up Nearly 50% Year-over-Year; Consumer Vertical Up Triple Digits Year-over-Year Gaming revenue grew nearly 50% year-over-year, reaccelerating as the model rolled out. Consumer vertical advertiser spend grew double digits quarter-over-quarter and triple digits year-over-year, but the company did not disclose a specific split between the two verticals. Foroughi said that under a single auction mechanism the two compete with each other, making it difficult to judge one vertical's growth in isolation. Q4 Guidance: Growth Slowing to 43%44% The company expects fourth-quarter revenue of US$2.3652.395 billion, up 43%44% year-over-year and 8%10% quarter-over-quarter, below the third quarter's 14% sequential growth; adjusted EBITDA of US$1.9952.025 billion, at an 84%85% margin. Stumpf said the guidance philosophy is unchanged, incorporating only currently known information. E-commerce Weight Rising; Seasonality to Become "Bigger Q4, Harder Q1" Foroughi explained that the sequential slowdown is not a shift to conservative guidance but a structural change: as e-commerce weight rises, the same advertiser will have a "huge fourth quarter" and a "weak first quarter." Historically, gaming saw easing competition early in the year, creating a "Q5" buffer that made the company's first quarters relatively strong; the higher e-commerce's share, the weaker that buffer. Cash Flow Conversion Rebounds to 77%; Buyback Authorization Raised to US$3.6 Billion Third-quarter free cash flow was US$1.418 billion, or 77% of adjusted EBITDA, versus 53% in the second quarter (affected by the timing of cash tax payments); the company still expects a full-year conversion rate of 70%75%. In the quarter it repurchased and withheld 4.1 million shares for US$1.4 billion, and the board added US$3.2 billion to the authorization, bringing the total to US$3.6 billion. CEO Says Competition Is Not Zero-Sum; MAX Supply Maintaining Strong Double-Digit Growth Addressing competition concerns since the second quarter, Foroughi said "this industry has never been zero-sum," and that several companies performing well at the same time does not squeeze AppLovin. He said supply on the MAX ad exchange is still growing at a strong double-digit rate, while peers are mostly growing at low double digits to single digits. The biggest mid-term lever is conversion rate, which the company believes remains well below what it sees as the potential range. CTV Pilot and Data M&A: Not Counted in Results Near Term The company has begun pilot campaigns on connected TV (CTV); Foroughi said product progress is "quite promising" but explicitly cautioned that "you should not expect it to boost results near term." On M&A, citing the MoPub acquisition as an example, he said that if the company acts it would prioritize targets that quickly bring customer relationships and data relationships, but there is no urgency at present. Reiterates Long-Term ~30% Annual Compound Growth Foroughi reiterated that the company can sustain roughly 30% annual compound growth long term, saying "as long as we can, we'll run above that level." Priorities for 2027 are completing the model transition and continuing to optimize on top of it, expanding existing customer spend, accelerating new customer acquisition, improving creative tools and ad formats, and increasing investment in CTV. Full Conference Call Transcript David Hsiao, Head of Investor Relations Welcome to AppLovin's third-quarter earnings call for the period ended September 30, 2026. I'm David Hsiao, Head of Investor Relations. Joining me today to discuss results are co-founder and CEO Adam Foroughi and CFO Matt Stumpf. Please note that our filings with the SEC to date, as well as the financial data and press release discussing third-quarter results, are available at investors.applovin.com. On this call we will make forward-looking statements, including but not limited to the platform's future development and reach, expected growth opportunities, the company's expected future financial performance and other future events. These statements are based on our current assumptions and beliefs, and we undertake no obligation to update them except as required by law. Actual results may differ materially from projections. We encourage you to review the risk factors in our latest 10-Q for the fiscal quarter ended June 30, 2026. More information will also be available in our 10-Q for the fiscal quarter ended September 30, 2026, which we will file today. We will also discuss non-GAAP financial measures. These non-GAAP measures are not superior to or a substitute for GAAP results. Please be sure to review the GAAP results and reconciliations of GAAP to non-GAAP measures in the earnings announcement and financial update on our investor relations website. This call is being recorded, and a replay and transcript will be available on our investor relations website for a period of time. I'll now turn the call over to Adam and Matt for opening remarks, after which the operator will host the Q&A session. Adam Foroughi, Co-Founder, CEO and Director Thank you all for joining today. We delivered record quarterly revenue of US$2.185 billion, up 55% year-over-year, with the largest dollar increase quarter-over-quarter in our history, more than US$200 million above last quarter. Both revenue and EBITDA were well above the upper end of our guidance. That is the standard by which we have always run this companyexceeding our own expectations. This result belongs to our team. They are executing a model migration as significant as AXON 2.0 while running the business at full speed and doing both in the same quarter. I want to thank them. The drivers are as follows. This quarter, we partially rolled out the new-generation model based on the architecture changes described last quarter. It performed strongly, advertisers spent more budget at their target ROAS and were profitable, and budgets were then raised. Our gaming business grew nearly 50% year-over-year, reaccelerating in step with this rollout. I want you to appreciate this: this is only a partial rollout. The migration is still ahead, and the new-generation model gives us significantly more room to optimize than the architecture it replaces. It also improves most where data is sparsernew advertisers, small advertisers, new categories, exactly where our consumer vertical sits. As always, model gains won't fall on a perfect quarterly cadence, and we won't commit to a cadence, but the room to optimize is real and we are still very early. Just as important, we did this with discipline. While part of the team trained new models, our engineers optimized how models are served. So even with increased training investment, data center costs as a share of revenue still fell. Every dollar of that investment is predicated on a return. This quarter you saw both sides: the return on the revenue side, and discipline on the margin side. The consumer vertical had another strong quarter, and as the fourth quarter unfolds, these advertisers' seasonal peak is coming; the consumer vertical is in very good shape entering the biggest spending season of the year. Long term, nothing has changed. We run one auction across multiple advertiser categories. Gaming keeps improving. The consumer vertical keeps adding runway. We believe this business can sustain roughly 30% annual compound growth long term. And as this year has shown, as long as we can, we intend to run above that level. Priorities going into next year are consistent: complete the model migration and continue optimizing on top of it, expand existing customer spend while truly accelerating new customer acquisition, keep improving our creative tools and ad formats, and begin increasing investment in CTV. I'll close with this: a quarter ago we stood here answering questions about missing expectations. We told you what happened, that the problem was fixed, and that the business was reaccelerating. And everything we said is reflected in this quarter's numbers. That is how we will continue to operate: build for the long term, be transparent about challenges and opportunities, and compound through the noise. I'll now hand it to Matt for the financials. Matt Stumpf, Chief Financial Officer Thank you, Adam, and thank you all for joining today. AppLovin reported record third-quarter results. Net income grew 72% year-over-year to US$1.439 billion, revenue reached a record US$2.185 billion, up 55% year-over-year, with a 66% net margin and an 84% adjusted EBITDA margin, all industry-leading. Revenue grew 14% quarter-over-quarter, exceeding the upper end of our guidance by about US$100 million, driven mainly by the core mobile gaming business, with model improvements contributing this quarter and the consumer vertical continuing to scale. Consumer advertiser spend continued to grow healthily, with spend up double digits quarter-over-quarter and triple digits year-over-year. Adjusted EBITDA was US$1.842 billion, up 59% year-over-year, also above the upper end of our guidance range, at an 84% margin, up about 190 basis points year-over-year. The conversion of incremental quarter-over-quarter revenue into adjusted EBITDA was 87%. Operating leverage remains strong. Thanks to optimizations introduced by the infrastructure team, our data center costs as a share of revenue fell, even as we increased compute investment to train existing and new models. We manage this business by EBITDA dollars and free cash flow, not by margin percentage, and we will keep investing wherever we see opportunities that generate more revenue. Free cash flow this quarter was US$1.4 billion, or 77% of adjusted EBITDA, up from 53% in the second quarter. As I said on last quarter's call, second-quarter conversion was affected by the timing of cash tax payments. We still expect full-year free cash flow conversion of about 70% to 75% of adjusted EBITDA. Putting our results in context: this earnings season, the largest platforms in digital advertising reported third-quarter revenue growth of roughly 10% to 25% year-over-year. We grew 55%, several times the market's pace. And our margins are far above all peers in the advertising industry. The comparison at the cash level is even starker. This quarter, our asset-light model converted about 65% of revenue into free cash flow. We ended the quarter with US$3.17 billion in cash and US$3.5 billion in long-term debt, with net leverage below 0.1x adjusted EBITDA, well below the roughly 1x net leverage we expect to operate at long term. In the third quarter, we repurchased and withheld a total of 4.1 million shares for US$1.4 billion, up significantly from the second quarter, ending the quarter with 331 million shares outstanding. Given our continued confidence in the company's future, the board added US$3.2 billion to the share repurchase authorization, bringing the current total authorization to US$3.6 billion as of the end of October. Since the program began, we have repurchased and withheld 104 million shares for US$9.4 billion at an average price of US$90 per share, returning 85% of free cash flow to shareholders over that period and substantially reducing our share count. Our weighted average diluted share count has declined for 8 consecutive quarters, and is down about 2% from a year ago even as we continue investing in the business. That combination55% growth plus a shrinking share countis something few companies can achieve. Turning to our fourth-quarter 2026 outlook. We expect revenue of US$2.3652.395 billion, up about 43% to 44% year-over-year and 8% to 10% quarter-over-quarter. We expect adjusted EBITDA of US$1.9952.025 billion, up about 43% to 45% year-over-year. At the midpoint, this guidance implies full-year 2026 revenue and adjusted EBITDA growth both above 50%. The outlook reflects normal seasonality, further scaling of the consumer vertical, significant model improvements already rolled out and taking effect, and continued increases in training compute investment for new model development. In summary, we exceeded the upper end of guidance on both revenue and adjusted EBITDA, grew revenue 55% year-over-year, expanded margins, generated US$1.4 billion in free cash flow, and returned US$1.4 billion to shareholders. Our fourth-quarter outlook reflects the same trajectory. We'll now move to Q&A. Q&A Session Operator [Operator instructions] Our first question comes from James Heaney. James Heaney, Jefferies Research Great, I have two questions. First for Adam. Can you talk about the generations of new models you're rolling out? I'm curious what the biggest improvements you're seeing are versus AXON 2? And anything you can say on timing for full rollout? Then I'll ask Matt one. Adam Foroughi, Co-Founder, CEO and Director James, last quarter we talked about one of our goals being to figure out how to build larger models, use more compute, and train a model with better output. So I previewed that last quarter. Now we're beginning to roll out these models. We're seeing good results. We can't tell you or commit to when we'll see full rollout, nor can we predict the cadence of these models. But as I said in my remarks, these models have more room to optimize than the prior generation we're on. So we're very excited about the future. James Heaney, Jefferies Research Okay. And then a question for Matt. On the surface, the fourth-quarter revenue guidance of 8% to 10% quarter-over-quarter growth, versus 14% in the third quarter, seems a bit conservative given strong e-commerce this quarter? How would you explain that sequential deceleration? Matt Stumpf, Chief Financial Officer Yeah. Our guidance approach hasn't changed over the past several quarters. We guide to what we know at that point in time. So our guidance incorporates the points I mentioned in my prepared remarks: normal seasonality within the quarter, the model updates we've rolled out and that are taking effect, and then obviously increased data center costs. So there's nothing really changed there versus prior quarters. Operator Our next question comes from Jason Bazinet of Citi. Jason Bazinet, Citi Research Just one question. I think last quarter you talked about investing more in the referral program. I wanted to get an update on whether you're pleased with the progress working with these partners. Adam Foroughi, Co-Founder, CEO and Director Yeah, Jason, still in progress. Obviously, this will be a multi-quarter, multi-year trajectory for our business. We're early in doing partnerships. But as you can see, even from what I think you see in published Pixel account growth, we've been acquiring more and more customers, and at an accelerating pace. That only compounds over time, because not only will there be more partnership deals and we'll start running more campaigns, but as more customers start having good experiences and sticking around, there will also be more positive sentiment about the product in the market. Because the other thing you have to realize is that when customers come in, their campaigns have to scale and optimize on our platform, they have to do incrementality studies, they have to prove it works, and then they really scale long term. So we expect this to compound over time. Operator Our next question comes from Matthew Cost of Morgan Stanley. Matthew Cost, Morgan Stanley Research Great. I wanted to ask, does your experience with this new model you just rolled out change your expectations for how much room there is to improve gaming ad revenue? Because this year you've been relatively explicit about a 20% to 30% growth expectation and mentioned the law of large numbers dragging on the gaming business. Do you reiterate that now? Or based on progress so far, could the potential be larger than you originally expected? Adam Foroughi, Co-Founder, CEO and Director I'd say what we've explicitly committed to over the past two quarters is that we believe this business can grow 30% year-over-year long term. We're a platform with a unified auction. Suppose in some world our consumer vertical grew 1000% next quarter. That would create more competition, right? So you can't look at this business vertically and think the consumer business is now growing faster and faster. It's hard to judge any one vertical in isolation. That said, this model is definitely more accurate than the old one, and there's more room to optimize going forward. So if we can do that, we would certainly expect gaming to grow at a higher rate long term. But that's not required for us to achieve that 30%, as long as both parts of the business grow reasonably fast. As we mentioned in our remarks, the other side of the business, the consumer side, is growing triple digits. So when you have a business like that, which is no longer immaterial to our business, growing at that rate, and with a lot of growth opportunity ahead as we penetrate more customers, that gives us a lot of room for future growth. Matthew Cost, Morgan Stanley Research Understood. And then I noticed you mentioned CTV in your prepared remarks. It's been a while since you last highlighted that. How important is it in the product roadmap now versus six months ago? Adam Foroughi, Co-Founder, CEO and Director Yeah, we've made more progress there. I mean, if people watch any content on connected TV, you'll start to see some of our ads; we have pilot campaigns running now. The product actually looks very promising. It's still very early. It's a brand-new market. So I would completely not recommend expecting it to boost results near term. But the team has really built a lot of confidence in where they are on the product roadmap. So if we get to the point where it's really starting to show very good results for customers, we'll talk about it more. Internally, we're very excited about it now. Operator Our next question comes from Benjamin Black of Deutsche Bank Aktiengesellschaft. Benjamin Black, Deutsche Bank Aktiengesellschaft Research Adam, you've talked in the past about needing to collect more data on the consumer side to optimize the model, improve performance, and the flywheel around it. With the new architecture, do you see a path for that flywheel to reach some kind of tipping point, like gaming did a few years ago? If not, what incremental investment do you think is needed? Adam Foroughi, Co-Founder, CEO and Director No, look, these two things go hand in hand, right? Every day we're live, customers come onto the platform, they share data with us, and so we get higher data density. So as quarters and years go by, you need to pair that with a model architecture that can actually handle more data and generate better output. So you write a bigger model, a more complex model, train it on a larger compute cluster, and ultimately get better output. What this new architecture allows us to do on both the gaming side and the consumer side is train larger models. So we will always be data-hungry. The faster we bring more data into the system, the better our predictive model output will be. So we'll never stop looking for ways to acquire data, whether by promoting ourselves through referrals and ads, as we just mentioned, to acquire new customers, or some form of data acquisitionwe'll look for all kinds of ways to bring data into this model, because we've seen time and again that whenever we add more data, we can write more complex models, train larger models, and get better output on the other end. Benjamin Black, Deutsche Bank Aktiengesellschaft Research Okay. I also wanted to ask, how are consumer advertisers' budgets for the holiday shopping season compared with last year? And if advertisers saw strong returns this quarter, how much of that spend do you think ultimately sticks into 2027? Or should we expect the usual seasonal pullback in the first quarter? Adam Foroughi, Co-Founder, CEO and Director I think it just follows the normal pattern. Look, we're still early on the curve. We're not even into the holiday season yet. So this will take shape over the next few weeks. But that seasonal pullback is consistent in e-commerce. We do have one advantage, though: as long as we grow this business at this rate, and we're still early in its evolution, we still have a rate of customer acquisition that lets us grow the business even in traditionally seasonally harder quarterslike the first quarter versus the fourth quarter. That said, the same advertiser, year-over-year, will have a huge fourth quarter followed by a weak first quarter. That's the reality of the e-commerce business. On that point, traditionally if you look at our business's quarter-over-quarter growth, our first quarters have been quite strong because gaming has a phenomenon called "Q5." Going into the first quarter, they know competition is less intense. So the modelwhether it's choosing gaming more frequently because ROAS is attractive, or customers being willing to take more risk because they're coming out of a very expensive fourth quarter into this post-fourth-quarter periodultimately spends very competitively. So we're less affected than other ad companies. But as we go deeper into e-commerce, we should see a bigger fourth quarter followed by a harder first quarter. Operator Our next question comes from Stephen Ju of UBS Group AG. Stephen Ju, UBS Group AG Research Okay. I think I spoke with some advertisers on your platform and they said something quite strange. They expect auction pricing to start getting pushed up because incremental bidders from e-commerce will come in and bid up prices. So they preemptively expect ROI to potentially compress and moved some budget to your competitors. It seems like a strange thing to do, but it does make me worry, and I think you've addressed this in prior periods, that this is a demand-constrained rather than supply-constrained industry. So can you talk about whether your thinking has evolved, or whether anything has changed, regarding AppLovin's ability to absorb all this incremental ad spend without compressing ROI for some existing customers? Adam Foroughi, Co-Founder, CEO and Director Yeah. If you look at our business over the past 3 years, we've become much more competitive. The business is much larger. Customers' ROI or ROAS is stronger than it was 3 years ago, because spend is substantially larger. So the better these models get at taking an impression and predicting the outcome, the more accurately they price. The incremental net revenue dollars this quarter versus last quarterI don't know how many of the companies we compete with that are publicly disclosed are directly comparable in size. But I think that's equivalent to the full-year revenue many of them report, and we did it in just one quarter's increment. So I think these industry surveys tend to be skewed, because you're looking at a very small subset of the ecosystem, and the customers who take the time to talk to you are probably not the ones doing particularly well. The ones really trying and succeeding probably aren't talking to you. So I'd say we're a pretty good indicator of this market, and our technology has proven very good at getting better pricing for customers. That doesn't mean CPM prices won't rise if you just look at them. In fact, if we target more accurately and conversion rates rise, CPMs will rise. But the downstream effect of better prediction is higher ROAS, which is exactly why we've been able to grow so fast. Operator Our next question comes from Robert Coolbrith of Evercore. Robert Coolbrith, Evercore ISI Institutional Equities Research I wanted to ask a bit more about the architecture. As models get larger and can handle more data, does that also make them more general-purpose? I know you've talked in the past about separating consumer and gaming models, but could models become more general-purpose across media types and customer types over time? Adam Foroughi, Co-Founder, CEO and Director Yeah, potentiallysorry to cut off your second question. But you're right, it's possible we eventually push this architecture in that direction. Robert Coolbrith, Evercore ISI Institutional Equities Research Okay, got it. And then I wanted to ask about brand advertising performance. We hear that because of mobile web issues, whether in terms of inventory supply or iOS 26 focusing more on mobile app inventory, more brands may be coming to you. Do you see that as an attractive, interesting opportunity for AppLovin going forward? Adam Foroughi, Co-Founder, CEO and Director Sure, if it's a performance marketerso I'd split performance brands into those truly doing performance, meaning they optimize incremental ROAS or downstream purchase transactions, and then performance brands still seeking audience reach. A lot of mobile web advertising, and a lot of what's traditionally called the open web, falls more into the latter. They just want to buy audience and reach it as efficiently as possible at a lower price. We try to price from discovery all the way to transaction. So if that's what they care about, and it's trackable end-to-end digitally, then yes, we think a lot of that budget will shift to the app space. It clearly performs very well, and it's much larger. If you lookI said this on the All-In podcast at the All-In Summitthis space is getting very, very large. We haveyou can back out our gross spend, and it's quite substantial. And if you scale up the whole space, it's one of the largest advertising opportunities outside of social and search, maybe the largest outside of social and search. So we fully expect more budget to follow, because performance is there. Operator Our next question comes from Clark Lampen of BTIG. Clark Lampen, BTIG Research I have two questions if I can. Adam, I want to pull on this model thread. Based on my understanding of where you're headed, one of the bigger derivative benefits of success, I think, is opening up more inventory or available inventory on top of your existing gaming and supply, which would presumably get filled by non-gaming advertisers because of auction dynamics. Is that the right way to think about itthat's the first question, the second-order benefit of turning into revenue-generating customers for e-commerce or gaming developers beyond just more ads per 100 impressions currently bought. And if so, can you help frame how significant that expansion of available or effective supply could be? Adam Foroughi, Co-Founder, CEO and Director Yeah. I think the challenge with supply expansion is that the MAX market, or the gaming advertising market, is much bigger than people realize. So when you're looking at a market in the tens of billions of dollars, we can go acquire new supply. In fact, the model we're on now and the growth of the consumer vertical give us a lot of opportunity to acquire more supply. But there aren't that many sources of supply open to third parties that can provide a huge boost all at once. So you need to compound over time. The reason CTV is attractive to us is that there's no real top-of-funnel-to-bottom-of-funnel performance advertising product on connected TV. If we can solve that and be the first to launch a scalable product there, there's a lot of under-monetized available inventory. It's a bit like mobile 10 to 15 years ago when we started. So something like that could have an impact. Not much else. But that's why we keep talking about conversion rate. Within the gaming ecosystem, our conversion rate is still well below what we think is possible. What the consumer vertical gives us is the opportunity to diversify the ads a user sees. If we show the same user 20 gaming ads a week, we might show them 15 product ads and 5 gaming ads. That doesn't mean gaming customers lose 75%. It just means the model can target more precisely. So the same 5 impressions that would have converted are still there. And now suddenly you have all these product ads. So that's the real area of focus. That's the biggest driver in the short to medium term, but long term, supply expansion is definitely something we'll look at. Clark Lampen, BTIG Research Okay, that makes sense. And then for Matt, setting aside the after-hours stock reaction I just saw, your stock is still down about 40% year-to-date and has only recovered most of the post-Q2 decline. Does the buyback give you flexibility, or does the authorization you just added give you more aggressive, opportunistic flexibilityif you think the stock is clearly disconnected from intrinsic value? Or is it mainly about executing on plan going forward? Matt Stumpf, Chief Financial Officer No. You saw within this quarter, Clark, that we did step up materially. So that's exactly the result of the situation you described, right? That's us opportunistically stepping up buybacks when we see that disconnect between the stock price and the company's actual intrinsic value. So we'll have the ability to do that going forward as well. Operator Our next question comes from Martin Yang of Oppenheimer. Martin Yang, Oppenheimer Research A question for Adam. At industry conferences, you mentioned dynamic product ads for e-commerce, which sometimes perform better than video. I'd like to hear your long-term view: do you think more diverse ad formats are the ultimate goal for making e-commerce more effective? Or are you still waiting for generative AI video creation to improve the volume and quality of video ads, with dynamic product ads just a transitional step for advertisers to get impressions? Adam Foroughi, Co-Founder, CEO and Director I'd say both. As language models applied to creative keep improving, and many companies in this ecosystem offer out-of-the-box ads, you'll see advertisers perform better. We'll also create more ad formats, hopefully serving advertisers in better ways. What we want is diversity in what consumers see. When you're running as many ads as we do every day, diversity of content can drive improvement. That's why it was a point in my remarks. It's an area we're focused on. The product is still very new. I know we've been at it for a few years, but it really is a very new product. Ad formats haven't evolved that much yet. I fully expect them to evolve much more over the next few years. Martin Yang, Oppenheimer Research Understood. If I can, a follow-up on e-commerce. Comparing this year entering the holiday season with last year, how would you characterize e-commerce customers' learning budgets or learning time? Has it improved? Adam Foroughi, Co-Founder, CEO and Director Yeah. Look, learning goes both ways. Our models have gotten better. And this new model architecture, as I said in my remarks, is also faster when there's less data. On the other hand, customers are risk-averse when entering a new relationship. So when they spend with us, even if results look better, they still need to validate. So in this category, if you talk to customers, they not only want to see spend and ROAS, they also need to do incrementality studies. So there's a slow ramp. That's the reality of the industry, but our product is getting better. Every quarter and every new release, part of it is improving things for existing customers. The other part of the improvement is how quickly you can show a good result to a customer who just signed up and doesn't have much data in our system, because that's how you reduce churn. Operator Our next question comes from Rob Sanderson of Loop Capital. Rob Sanderson, Loop Capital Research A couple of questions on the consumer vertical as well, with triple-digit year-over-year growth, clearly a good trajectory. Can you comment on conversion of new leads coming through self-service or referral partners, or specifics on customer outcomes. I'm just wondering, there's still a bit of a mixed read on advertiser sophistication. Are we seeing similar trends, or is there anything you can comment on regarding go-to-market progress? Adam Foroughi, Co-Founder, CEO and Director Yeah. Look, we're still early on the product. So if a customer goes through our team, the likelihood of success is much higher than going through self-service and figuring it out themselves; but if they go through self-service and are sophisticated enough, they can absolutely succeed. If they're a very small business, what you'd classify as fully small and medium-sized (SMB), that's still further out for us. So there's a roadmap here: the more data you bring into the system, the better you can predict without data, and we're not yet at the point where a very small business selling jewelry, whom we've never seen, with little data and little GMV, can plug in and run. But 5 years from now, if we have massive data scale and the model has evolved to where it will be in 5 years, you'll have better results at every level. So it's a roadmap. We need to build the product over time to handle every scenario. For customers we either hand-hold or who are sophisticated, it works very well today. Rob Sanderson, Loop Capital Research Is ad creative still the main obstacle? That used to be a big failure point, with a lot of things that didn't work getting stuck on ad creative. Is that still the main sticking point? If we'd asked this a year ago, would you have expected the industry to be further along on generative AI tools? At least to me, things may be slower than we thought? Adam Foroughi, Co-Founder, CEO and Director Yeah. On the latter, the big Western large language model companies are all focused on other things. So there's less R&D on this problem outside of us. On the formerI wouldn't call it an obstacle, but for customers who don't invest in creative, it's an opportunity cost or lost opportunity. If you talk to customers who advertise on us, Meta and TikTok, they'll consistently emphasize one thing: you need to invest more creative in AppLovin to succeed on our platform. Those who know this are more successful than those who don't. Those who don't may be smallerit's probably not an obstacle to running, but they're missing a scale point. The challenge is that our ads are longer and they get seen. When you load 10 ads on another platform, they aren't seen for as long, and many aren't seen at all. Load 10 ads on our platform and on average you get over 35 seconds of view time, with nothing else you can do on the screen. It's a full-screen ad. So they are seen with 100% certainty. So the combination of ad volume and ad quality is what creates scale on our platform. Some customers know this and can become big spenders with us. Others need to learn. Rob Sanderson, Loop Capital Research I want to squeeze in one more. You talked about significantly larger models, a step change in scale. Are you organically generating enough data to feed the models? Adam, you also talked about possibly acquiring data. I'd guess you'd look for customer data, transaction data. What types of data are most helpful? Does it need to be long-lived? Or is what you generate organically enough?