The Quarter Management Promised, Delivered: Gaming Re-Accelerates to Nearly 50% on a Partial Model Rollout, Guide Cleared by $100M, Buyback Rises 2.5x. Upgrading to Outperform.

Results vs. Consensus

Q3 2026 Scorecard

The last two rows are the ones that distinguish this print from an ordinary beat. A quarter ago the Q3 guide arrived roughly 0.6% below where consensus sat, and that single fact did much of the work in converting a trivial dollar miss into a 19.7% one-day decline. This time the forward number beat as decisively as the reported one.

The More Important Comparison: Actual vs. the Company's Own Guide

AppLovin is accountable to the range it issued on 5 August, not to a consensus that drifts. That is where the Q2 quarter broke, by $1.2M on adjusted EBITDA, and it is where this quarter has to be judged.

The August guide was unusually informative because management narrowed its content on purpose: it was built only on model releases already deployed and performing, and explicitly excluded anything not yet shipped. That construction capped the size of any beat that did not come from a new release. A 4.8% revenue beat against a deliberately de-speculated guide therefore says something specific, which is that additional model work landed inside the quarter. Management confirmed as much, describing a partial rollout of the next generation as the driver.

Year-Over-Year Comparison

Two lines in that table deserve to be read together. Revenue grew 55.5% while cost of revenue grew 37.7%, which is why gross margin set a record at 88.98% in the very quarter management says it was training a new model generation. The year-ago quarter is a clean comparison: both periods are continuing operations, and the Apps divestiture is fully out of the base by Q3 2025.

Quarter-Over-Quarter Comparison

Q2 2026 figures come from the quarter column of the August filing rather than from subtraction, so they are reported rather than derived. The one line that looks ugly, general and administrative up 59.9%, is the opposite of what it appears: 9M 2026 general and administrative expense of $148.8M is down 10.0% against the prior-year nine months, and Q2's $40.3M was the anomalously low quarter. Separately, the transaction-related expense add-back inside the adjusted EBITDA reconciliation rose from $0.1M to $15.5M this quarter. Such costs customarily sit in general and administrative, though the filing does not say which line carries them, and either way the item is excluded from adjusted EBITDA.

The Sequential Step-Up in Historical Context

Management claimed on the call that this was the largest sequential dollar step-up in revenue in the company's history. We checked it against the full filed series rather than taking it on trust, because superlatives of this shape are the ones that quietly turn out to be wrong.

The claim holds. The prior maximum was Q4 2025 at +$252.9M, and $261.1M clears it. The series validates against the filed full-year totals exactly, so it is not an approximation. It also makes one point the company did not make: the margin column shows that 84.32% is still 18bp below the Q1 2026 record of 84.50%. This was a margin recovery, not a margin record, and the Q4 guide midpoint of 84.45% would do no better than match the peak.

Quality of Beat/Miss

• Revenue: organic, single-segment, no acquisitions and no currency effect of note. The composition is disclosed only directionally: gaming grew "nearly 50%" and consumer spend grew "triple digits" year over year with "double digits" sequentially. The $99.8M overshoot against a guide that excluded undeployed model releases implies the overshoot came from releases that landed in-quarter rather than from pricing or seasonality.
• Margins: sustainable on the evidence available. The incremental cost of revenue per incremental revenue dollar was 8.5 cents year over year and 5.7 cents sequentially, against the roughly 10 cents management has framed as the compute run-rate and against Q2's 10.6 cents and 27 cents on the same two bases. Research and development rose only 1.9% sequentially while revenue rose 13.6%, so the training step-up that compressed Q2 did not repeat. Stock-based compensation of $84.2M is 3.85% of revenue, up from 2.40% a year ago, which is a real dilution cost the adjusted measure excludes.
• EPS: operationally driven with a below-the-line assist. Operating income grew 58.4% against revenue growth of 55.5%, so the operating leverage is genuine. On top of that sits $0.21 per share of after-tax other income and a 1.82% reduction in the diluted share count. Strip the year-over-year other-income swing entirely and diluted EPS lands near $4.07, still above the $4.03 consensus.

Revenue assessment

Growing 55.5% on a $1.4B base is a strong absolute outcome, but the number that matters for the thesis is the sequential one. Sequential growth of 13.6% follows 4.4%, and the dollar step of $261.1M is the largest the company has produced. In August we wrote that the entire premium on this stock rested on a mechanism, that models improve, advertiser returns improve, budgets step up, and that Q2 had demonstrated the mechanism was a research pipeline rather than an algorithm. That characterisation was correct and remains correct. What this quarter adds is a second observation about the level of that pipeline's output: when a release lands, it is worth more than the market had assumed, and the architecture that produced it is only partially deployed.

The specific disclosure that changes our estimates is gaming at nearly 50% year-over-year growth. Through 2026 the framing had been a 20% to 30% long-run expectation for gaming with the law of large numbers pressing on it, and the bear case we articulated in August held that the core was approaching share saturation, leaving take rate or an immature consumer vertical as the only sources of growth. Gaming re-accelerating to nearly 50% in the quarter a new model generation partially rolled out is direct evidence against that reading.

Margins assessment

The direction reversed, which is the point. Adjusted EBITDA margin went 84.50%, then 83.89%, then 84.32%, and the Q4 guide implies 84.4% to 84.6% (guided 84% to 85%) against the 83% guided for this quarter. Our August note treated the guided step-down from the Q1 peak as the point at which the margin direction changed, grading that pillar neutral on the basis that the direction had turned while the level remained exceptional. The direction has now turned back. More useful than the level is the flow-through: 87.5% of the incremental revenue dollar reached adjusted EBITDA this quarter, against 70% in Q2 and 86% in Q1.

We would caution against reading this as a return to a ratcheting margin. Management restated plainly that it manages to EBITDA dollars and free cash flow rather than to a margin percentage, and will spend against identified incremental revenue whenever it appears. That philosophy produces exactly the pattern of the past three quarters, which is a margin that oscillates in a narrow band in the mid-80s rather than climbing. The band is the right expectation. The climb is not.

EPS assessment

Diluted EPS of $4.30 beat the $4.03 consensus by 6.7% and grew 75.5%, outpacing revenue growth by 20 points. Three things did that work, in descending order of quality. Operating leverage is the largest and best of them, with total costs rising 45.9% against revenue up 55.5%. The share count did real work for the second year running, down 1.82% on a weighted diluted basis and down 2.07% on shares outstanding since December. The weakest contributor is the $86.0M of other income, worth roughly $0.21 per share after tax, which sits below the operating line, is excluded from adjusted EBITDA, is not broken out in the filing and was not discussed. With $3.17B of cash on the balance sheet, interest income plausibly accounts for a meaningful part of it, but that is inference rather than disclosure, and the absence of an explanation on a line that grew to 3.9% of revenue is a gap.

Vertical Performance

AppLovin reports a single revenue line and discloses no revenue by vertical. Everything below separates what was disclosed from what has to be inferred, and every inferred figure is labelled.

Mobile gaming: the re-acceleration, and the first growth rate we have been given

Gaming is where the Q2 shortfall originated and where this quarter was won. Both the CEO and the CFO attributed the sequential beat to the core mobile gaming business responding to model improvements deployed during the quarter, and for the first time in our coverage the company put a growth rate on the vertical.

"During the quarter, we partially rolled out the next generation of our models, built on the architectural changes I described last quarter. Performance was strong. Advertisers saw more spend deploying profitably at their target return on ad spend goals, and budgets stepped up. Our gaming business grew nearly 50% year-over-year, re-accelerating right alongside the rollout." — Adam Foroughi, Co-Founder and CEO

That sequence is the whole bull mechanism stated compactly: better prediction, better advertiser return, larger budgets. The supply side corroborates it. Management described the MAX marketplace as growing at a strong double-digit rate annually against an advertising industry whose supply growth it put at low double digits to single digits, and acknowledged the obvious consequence, which is that AppLovin must out-innovate the rising cost of that inventory rather than simply buy it.

"So cost effectively, companies getting better at monetizing the impressions, the value of the auction goes up, the publisher makes more. Now, on the other side of that, we are either going to have to pay out more for publisher inventory or improve our own technology at that same rate, or improve our own technology more than that rate to grow our business even faster than that market." — Adam Foroughi, Co-Founder and CEO

Assessment: This is the most important disclosure of the quarter. A gaming business compounding at nearly 50% is not one approaching saturation, and the mechanism that produced the acceleration is only partially deployed. The structural caveat survives: revenue in this vertical is still a function of an R&D calendar that management will not forecast, and the inventory cost curve means standing still is a slow decline. But the question we said in August we could not answer, whether Q2 marked a maturing curve or a missed release, now has evidence on one side of it.

Consumer: triple-digit growth, and still not a single dollar disclosed

The consumer vertical delivered again, and the disclosure shifted form. For the previous two quarters the only metric was advertiser spend measured against the Q4 2025 seasonal peak. This quarter the company gave growth rates instead, which is more useful and still not a number.

"Consumer advertisers' spend continues to have healthy growth, with spend up double digits quarter-over-quarter and growing triple digits from the same quarter last year." — Matt Stumpf, CFO

Those two rates plus the disclosed gaming rate and the reported total are enough to bound the vertical arithmetically, and the bound is uncomfortably wide. Solving the blend for a 55.5% total with gaming at "nearly 50%" and consumer at "triple digits" puts consumer revenue somewhere between roughly $150M and $540M for the quarter, or 7% to 25% of the total, with the midpoint of the feasible range in the mid-teens. The answer swings by a factor of three depending on whether "nearly 50%" means 45% or 50% and whether "triple digits" means 100% or 200%. That is the precision the current disclosure permits, and it is why no point estimate appears in this report.

On the structural question of who can actually use the product, management was franker than it has been. The self-service funnel still works mainly for sophisticated advertisers or those the sales team handles directly, and the true small-business case is explicitly years away.

"There is a roadmap on this thing where the more data you get into the system, the more predictive you can be in the absence of data. And we are not at the point yet where somebody can just plug in tiny business selling jewelry, we have never seen them before, very little data, very little GMV, and it just plugs in and they go." — Adam Foroughi, Co-Founder and CEO

Assessment: Consumer is the better-compounding half of the business and the worse-disclosed half. Triple-digit growth at a mid-teens revenue share means it is now large enough to matter to the consolidated rate, which is a change from August when management said it could not smooth a quarter. The new model generation is specifically described as performing best where data is sparse, which is where this vertical lives, so the next leg of the rollout should help here most. The gap remains that an investor cannot size the vertical, cannot count its advertisers, and cannot observe its retention.

Connected TV: from deferred to piloted

In August, connected TV was the third item in a supply-expansion queue with the note that the company was "just not at the point yet where we can go execute on it." This quarter it is one of four named 2027 priorities with live pilots and a clear statement of why it is attractive.

"The reason CTV is compelling to us is there's not a true top of funnel to bottom of funnel performance offering on connected TV. If we can figure that out and be the first to market with something that's scalable there's a lot of available inventory that's under monetized there. It's sort of what mobile looked like to us 10, 15 years ago when we started the business." — Adam Foroughi, Co-Founder and CEO

Management was careful to cap expectations in the same breath, calling it early, a whole new market, and something not to bank on for numbers anytime soon.

Assessment: A free option that has moved from theoretical to tested in one quarter, and the strategic logic is the strongest version of the argument: performance advertising on connected TV is genuinely unbuilt, and APP's claim to it rests on the same prediction stack that works in app. We carry it at zero in the model and will keep doing so until a pilot converts. The read-across worth noting is that it puts APP on a collision course with the open-web demand-side platforms in a category they regard as their growth engine.

Lead generation: silent

Lead generation was a disclosed vertical in testing two quarters ago and a named signpost in our August checklist. It was not mentioned in the prepared remarks, no analyst raised it, and no update was given.

Assessment: An omission rather than a negative, but a telling one. The verticals management chooses to talk about are the ones producing results, and the inference from a vertical that had nothing new to report in August and went unmentioned this quarter is that it is not yet one of them.

Cost Structure: Where the Money Went

Three readings. First, cost of revenue fell 142bp as a share of revenue year over year and 72bp sequentially, which is the arithmetic behind management's claim that data centre costs declined as a percentage of revenue even as training investment rose. Second, research and development remains 154bp higher than a year ago in ratio terms but has stopped growing sequentially, which is what a completed build phase followed by a deployment phase looks like. Third, sales and marketing fell again to 3.15% of revenue, the lowest of 2026 and down from 3.30% in each of the prior two quarters, four months after the platform opened to the public. The one quarter below it is Q4 2025 at 2.94%, which is the seasonal-peak revenue quarter and therefore the structurally lowest ratio of any year. The paid-acquisition ramp that earlier coverage expected still has not begun; partnerships remain the acquisition channel, and the cost of that choice is slower advertiser growth in exchange for a better margin.

Disclosed Operating Metrics

Key Topics & Management Commentary

Overall Management Tone: Confident and pointedly retrospective, opening by framing the quarter as the fulfilment of a specific claim made three months earlier rather than as a beat on its own terms. Where the Q2 call replaced dates with conditions, this one replaced conditions with results, and the single place the posture remained unchanged was disclosure, where management declined a direct request for predictive metrics in plain language rather than deflecting it. Tone was the most assured of the past three quarters without reaching for the catalyst-dense register that preceded the Q2 miss.

1. The Claim Management Made in August, and Kept

The most consequential thing on this call is structural rather than numerical. In August management gave a falsifiable account of the Q2 miss: that it was a timing problem, that the next model step-up had already landed in early Q3, and that the business was therefore re-accelerating. Three months later the sequential rate is 13.6%, the guide was cleared by $100M on both lines, and management chose to make the audit the opening of the call.

"A quarter ago, we stood here answering questions about a miss. We told you what had happened, that it was fixed, and that the business was re-accelerating, and everything we told you showed up in this quarter's numbers. That's how we'll keep operating, build for the long term, be transparent about challenges and opportunities, and compound through the noise." — Adam Foroughi, Co-Founder and CEO

The claim was independently corroborated later in Q&A, when management confirmed that A/B test lifts had landed in almost every one of the previous sixteen quarters with one exception, and that exception was Q2 2026. That is a much more specific statement than the one given in August, and it fits the data rather than merely being consistent with it.

Assessment: This is the strongest argument for upgrading, and it is about credibility rather than arithmetic. In August we wrote that a management team which owns a small miss cleanly is more trustworthy on the next large one. The sequel to that observation is that a management team which makes a specific forward prediction on a bad quarter and then delivers it earns the right to have its framework taken at face value. The 83% margin guide and the de-speculated revenue guide were both beaten; the "it was fixed" claim was the hardest of the three to fake, and it verified.

2. The New Model Generation: Deliberately Unquantified, Explicitly Incomplete

The architecture described in August as "architectural changes that let us build more complex models" has begun to ship. Management put its technical significance at the level of Axon 2.0, the transition that produced the company's original re-rating, while refusing the branding and refusing a timetable.

"Here's the part I want you to sit with. This was a partial rollout. The transition is still in front of us, and the new generation gives us meaningfully more room to optimize than the architecture it replaces. It also improves performance most where data is sparser, newer advertisers, smaller advertisers, newer categories, exactly where our consumer vertical lives." — Adam Foroughi, Co-Founder and CEO

The second sentence of that quote is the part the market priced. If the generation that delivered a $261M sequential step is only partially deployed, then the remaining deployment is unpriced optionality rather than a hoped-for catalyst. The fourth sentence is the part the model should reflect: the architecture's comparative advantage is in sparse-data regimes, which is a direct statement that the consumer vertical is the structural beneficiary of the next leg.

Management then capped the enthusiasm in the same paragraph, which is the pattern of the entire call.

"As always, model gains won't land on a perfect quarterly schedule, and we won't promise a cadence, but the headroom is real, and we're still very early." — Adam Foroughi, Co-Founder and CEO

Assessment: Both halves are true and the market has to hold them together. The headroom is real and partly undeployed, which raises the expected value of the next several quarters. The cadence is still unforecastable, which keeps the variance high and is the reason this upgrade goes to Outperform rather than to a conviction-10 position. An investor is being paid to accept lumpiness, and after this quarter the payment looks adequate where in August it did not.

3. Margin Reversed Direction, and the Compute Framework Improved

The August thesis carried a bear point that compute spend was landing a quarter ahead of the revenue it funded, compressing margin on a lag the market could not see. This quarter is the clean test of that proposition, because the revenue arrived while the training continued.

"While part of the team trained the new models, our engineers optimized how we serve them. So even as training investment increased, data center costs declined as a percentage of revenue. Every dollar of this investment is gated on return. In this quarter, you saw both halves of that, the return in revenue and the discipline in margin." — Adam Foroughi, Co-Founder and CEO

The filed numbers support it more strongly than the narrative claims. Incremental cost of revenue per incremental revenue dollar came in at 8.5 cents year over year and 5.7 cents sequentially, against the roughly 10 cents management has repeatedly framed as the compute run-rate and against Q2's 10.6 cents and 27 cents. Research and development rose 1.9% sequentially on revenue up 13.6%. Flow-through to adjusted EBITDA was 87.5%.

The CFO paired this with the operating philosophy, restated verbatim from August, which is the thing that caps how good the margin can get.

"We manage this business to EBITDA dollars and free cash flow rather than to a margin percentage, and we'll continue to deploy dollars when we see an opportunity to produce more revenue." — Matt Stumpf, CFO

Assessment: The compute bear case does not survive this quarter in the form it was written. A one-quarter lag between training spend and the revenue it funds is now a documented pattern with one observation of each half, which is a working-capital-like timing effect rather than structural erosion. The honest residual is that both quarters since the Q1 2026 peak margin came in below it, and management has told investors twice not to expect a ratchet. Mid-80s with oscillation is the correct modelling assumption, and it is an outstanding one.

4. Capital Return Went From Halved to Largest Ever

The August report treated the halving of the buyback to $551M as a failed signpost, and specifically flagged Q3 deployment as the test of management's claim that the pause reflected cash flow rather than conviction. The test was passed emphatically: $1.4B of repurchases and withholdings on 4.1M shares, plus a $3.2B increase to the authorization taking it to $3.6B as of end-October. Management also aggregated the programme for the first time.

"Since we began the program, we have repurchased and withheld a total of 104 million shares for $9.4 billion at an average price of $90 per share, returning 85% of our free cash flow over that period and reducing our share count materially. Our weighted average diluted share count has now declined for eight consecutive quarters and is down approximately 2% from a year ago, even as we invest in the business." — Matt Stumpf, CFO

The year-over-year share-count reduction verifies at 1.82% on a weighted diluted basis, and shares outstanding are down 2.07% since December. The $90 average purchase price across the programme is a management disclosure that cannot be reconstructed from the filings, and the stated figure of 104M shares for $9.4B is arithmetically consistent with it.

Assessment: The behavioural signal investors thought they were buying in Q1 and did not get in Q2 appeared in Q3: 4.1M shares for $1.4B, an average near $341. Combined with net debt of $342M against TTM adjusted EBITDA of $6.41B, or 0.05 times, and a $3.6B authorization, this is now a genuine floor mechanism rather than a stated intention. It is also the clearest evidence of how management reads its own stock, which matters more than usual in a name where no predictive operating metric is disclosed.

5. Free Cash Flow Recovered, and the Full-Year Target Quietly Widened Downward

Conversion was the August worry and it recovered as promised, from 53.5% of adjusted EBITDA to 77.0%, with $1,418.4M of free cash flow in the quarter. The CFO's framing contained one change that was not flagged as a change.

"Free cash flow for the quarter was $1.4 billion, or 77% of adjusted EBITDA, up from 53% in the second quarter. As I mentioned on last quarter's call, second quarter conversion was depressed by the timing of cash tax payments. We still expect full-year free cash flow conversion of roughly 70%-75% of adjusted EBITDA." — Matt Stumpf, CFO

In August the commitment was to "normalize to roughly 75% of adjusted EBITDA for the full year." It is now "roughly 70%-75%," prefaced by the word "still." That is a reduced commitment delivered inside a beat, and it is the kind of revision that goes unremarked on a good day. Nine-month operating cash flow of $3,589.3M against nine-month adjusted EBITDA of $5,013.0M is 71.6% on a pre-capital-expenditure basis, which sits inside the new band and below the old target. It is also worth recording that 77.0% is 13.6 points below the 90.6% the company converted in Q3 2025.

Assessment: The quarter-level commitment was met and the tax-timing explanation given in August is now documented by outturn rather than asserted. The full-year band moving down by five points at the low end is a small negative that matters mainly because the buyback pace is explicitly a function of free cash flow. At the new band's low end, FY2026 free cash flow is roughly $4.9B against roughly $5.3B at the old target, which is still comfortably more than the authorization.

6. The "Roughly 30%" Algorithm, Restated With a Clause Added

The long-term growth rate that did most of the damage in August was restated in substance, with new sentences attached.

"On the long term, nothing has changed. We run one auction across multiple advertiser categories. Gaming keeps improving, consumer keeps adding runway, and we believe this business can compound at roughly 30% annually over the long term. As this year shows, we intend to run well ahead of that whenever we can." — Adam Foroughi, Co-Founder and CEO

Pressed in Q&A on whether the new architecture lifts the gaming framework specifically, management declined to raise the number, restated 30% as its long-term expectation, and said that reaching it does not require gaming to outperform.

What remains missing is the same thing that was missing in August: no base year, no bridge from the delivered 55.5% to the stated 30%, and no indication of whether it describes 2027 or a distant steady state. Three quarters of actuals now sit between 52.8% and 59.0%, so the gap between the framework and the delivery is not narrowing.

Assessment: The 30% figure continues to anchor the valuation debate and continues to be the least informative number management gives. The added clause is not nothing, since "we intend to run well ahead of that whenever we can" is a different posture from the bare statement given on a miss. We model FY2027 at 35% to 40% growth, above the stated rate and below the current run rate, and treat the 30% as a deliberately conservative promise rather than a forecast.

7. Creative Was Reclassified From Company Hurdle to Market Condition

Out-of-the-box video creative is the single commitment in this coverage arc with a date attached to it. In May it was "a matter of weeks." In August it was "still work in progress" and "the biggest hurdle in our system." This quarter the tool has still not shipped, and the framing changed.

"Yeah, and look, on the latter, the large language model companies in the West are focused on something else, and so you have less R&D outside of us on this problem. On the former, I wouldn't call it a barrier, but it is an opportunity cost or opportunity lost for customers that don't invest in creative." — Adam Foroughi, Co-Founder and CEO

Two things moved in that answer. The absence of the tool is now attributed to external research priorities rather than to the company's own roadmap, and the constraint itself is downgraded from "the biggest hurdle in our system" to something that is explicitly "not a barrier." Management also repeated a useful datapoint from a year ago in the same exchange, which is that the platform's full-screen ads average over 35 seconds of watch time, and that creative volume and quality together determine how much an advertiser can scale.

Assessment: This is the weakest part of the quarter and the clearest instance of a good print being used to re-characterise an unmet commitment. The substance has not changed since May: an advertiser who cannot produce long-form video cannot scale on this platform, which is precisely why the self-service long tail remains gated and why the sales team still has to handhold new consumer accounts. Calling it an opportunity cost rather than a barrier is a reasonable description of the economics and an unreasonable description of the roadmap. We keep this bear point live.

8. Data Acquisition Enters the Story as Explicit M&A Appetite

A new element this quarter, and one that did not exist in prior calls in this form. Management stated plainly that the binding long-run constraint is data density, that organic data accumulation is slower than it would like, and that acquisition is a legitimate route to compress the timeline. The precedent cited is MoPub, which became MAX.

"We're going to be starved for data. The more data we get into the system quicker, the better the output of our predictions model is going to become. We're never going to stop looking to go get data, whether it's promoting ourselves through referrals and advertising, as we just touched on, for new customers, or potentially some sort of data acquisition." — Adam Foroughi, Co-Founder and CEO

Against that, management was equally clear that the current quarter required no new data at all.

"I mean, look, obviously this quarter, we had a fantastic quarter, best we've ever had with the data that we had. There was no incremental data into the system, right?" — Adam Foroughi, Co-Founder and CEO

Assessment: Read together, these two statements say that better processing of existing data produced a record quarter and that more data would raise the ceiling further. That is a constructive formulation. The risk it introduces is new: a nearly unlevered balance sheet, a $3.17B cash position and a stated appetite for buying customer and data relationships is a configuration that can end in a large, poorly-priced acquisition. The MoPub comparison is the strongest possible precedent and also a selected one. We would read a sizeable deal as a negative until proven otherwise, and note that it would compete directly with the buyback for capital.

9. Competition: the "Foot Fault" Framing, Now With Peer Results Behind It

The August quarter arrived amid genuine concern that the walled gardens and other performance channels were taking share. Management rejected that reading then and now has a cleaner factual basis for the rejection, since several large platforms reported strong quarters in the same season.

"We have said since we went public, this is not zero sum. Our prediction, the impression, what we price for a customer is going to be different than any other model. What makes this category great is that the sum total of all the models improving give these customers better businesses. So, it has never been the case that other companies growing should squeeze us. I said it on the last earnings call, that was not something that actually happened, and it was our own foot fault." — Adam Foroughi, Co-Founder and CEO

The CFO supplied the quantitative version, putting peer growth at roughly 10% to the mid-20s percent for the quarter against AppLovin's 55%. We have not independently verified each peer's figure for this report and present it as management's characterisation.

Assessment: The competitive bear point, which we have carried as not confirmed, now has a second quarter of contrary evidence. Peers growing well while APP grows 55% is the direct refutation of the share-loss thesis, and it also validates the "not zero sum" framing that was easy to dismiss as spin in August. What remains unresolved, and unmeasurable from outside, is whether the cost of winning impressions rises faster than the prediction improves, which is the version of competition that actually shows up in this business.

10. The Q1 2027 Seasonal Reset, Flagged in Advance

The most useful forward-looking negative on the call was volunteered rather than extracted. As the consumer vertical grows into a larger share of revenue, the company inherits e-commerce seasonality, and management said so directly.

"So, we've been more insulated than other advertising companies, but as we get more deep into e-commerce, we should see bigger Q4s followed by tougher Q1s." — Adam Foroughi, Co-Founder and CEO

The structural reason APP has been insulated is a gaming phenomenon management calls "Q5," in which post-holiday competition for impressions falls and gaming advertisers spend aggressively into Q1. That cushion does not disappear, but it gets diluted as consumer grows. Management also noted the offset, which is that the consumer vertical's own new-customer acquisition rate can carry growth through a seasonally weak quarter even when like-for-like advertiser spend falls.

Assessment: Credit for saying this in advance of the quarter rather than during it. The practical implication is that a soft Q1 2027 sequential print should not be read as a cadence break of the Q2 2026 kind, and investors who conflate the two will mis-price it. The modelling implication is that our FY2027 estimate should carry a weaker Q1 and a stronger Q4 than a smooth ramp would imply.

11. Advertiser Acquisition Still Runs Through Partnerships, and the Pixel Trail Is the Only Scoreboard

The partnership channel that replaced paid marketing in August remains the mechanism, and management's own progress metric is a third-party observable rather than a disclosure.

"Still a work in progress. Obviously, this is going to be multi-quarter, multi-year trajectory on this business. We're early in doing partnerships, but as you've seen, even in the rate of pixel counts that I think you're publishing, is that we've been getting more and more customers coming on, and the rate is accelerating." — Adam Foroughi, Co-Founder and CEO

Management went on to explain why a new advertiser takes time to become a large one: campaigns must be scaled and optimised, then validated through an incrementality study, before budgets commit for the long term. Asked why third-party pixel counts skew toward Asia-Pacific, management argued it is an attribution artifact rather than a penetration signal, since APAC advertisers frequently spin up single-purpose domains while North American enterprises run one.

Sales and marketing at 3.15% of revenue, the lowest of 2026, corroborates that no paid-acquisition ramp has begun.

Assessment: Pointing investors to externally-scraped pixel counts as evidence of customer growth is an unusual substitute for disclosure, and it cuts both ways: the company is implicitly endorsing a metric it does not control and cannot revise. The strategy itself is defensible and margin-friendly. Its cost is that advertiser growth proceeds partner by partner, and after four months of general availability the only quantitative evidence available to an investor comes from outside the company.

12. Supply Expansion: Conversion Rate, Not New Inventory

Asked whether the new models unlock materially more sellable inventory, management reframed the question. The constraint is not supply, it is the share of impressions that convert, and the consumer vertical is the lever.

"We're still very low on what we think is the possible conversion rate inside this gaming ecosystem. What consumer gives us is the chance to diversify the ads that someone's seeing. Instead of showing the same user 20 gaming ads in a week, we might show them 15 product ads and five gaming ads. That doesn't mean that the gaming customers lost 75%. It just means that the model can get much more targeted, so those same five impressions that would have driven the conversion are still going to sit there, and now all of a sudden you have all these product ads." — Adam Foroughi, Co-Founder and CEO

The scale of the headroom was quantified obliquely elsewhere on the call, when management noted that at a conversion rate "a little bit over 1%," roughly 987 of every 1,000 impressions purchased lose money. Management also sized the category, arguing that in-app performance advertising is plausibly the largest advertising opportunity outside social and search.

Assessment: This is the cleanest statement of the long-run bull case the company has given, and it is more credible than a total-addressable-market slide because it is mechanical. If a 1% conversion rate is the binding constraint, then every increment of model accuracy monetises the same inventory harder, and consumer demand raises the number of chances to find a match per impression. It also explains why management is indifferent to new supply sources other than connected TV, and why gaming and consumer cannot be analysed as separate businesses in a single auction.

Guidance & Outlook

Guide composition. The Q4 outlook is built on four disclosed components, stated in the prepared remarks: normal seasonality, further consumer scaling, model improvements already live and performing, and continued investment in training compute for new model development. That is the same de-speculated construction used for Q3, which the company then exceeded by 4.8%. Management explicitly declined to split the guide between gaming and consumer.

"That outlook reflects normal seasonality, further scaling in our consumer vertical, significant model improvements that are already live and performing, and continued investment in training compute for new model development." — Matt Stumpf, CFO

Implied quarter-over-quarter ramp. The $2,380M midpoint requires a $195M sequential step, smaller than the $261M just delivered, and 8.9% sequential growth against the 13.6% just reported. On the face of it the guide decelerates on both the sequential and the year-over-year measure. Two considerations qualify that. First, Q4 2025 is the hardest comparison of the year, having itself grown 18.0% sequentially and set the prior record dollar step. Second, if Q4 clears its high end by the 4.8% margin Q3 did, revenue lands near $2.51B, which is 14.9% sequential growth and 51% year over year. The guide is a floor built from shipped product, not a forecast.

Full-year implication. Nine-month revenue of $5,951.0M plus the Q4 midpoint gives FY2026 revenue of $8,331.0M, up 52.0%. Adjusted EBITDA of $5,013.0M plus $2,010M gives $7,023.0M, up 55.6%, at an 84.3% full-year margin. Management's statement that the midpoint implies more than 50% growth in both measures verifies.

Street at. This is where the guide differs from August's. Consensus for Q4 revenue sat near $2.25B and for adjusted EBITDA near $1.88B before the print, so the guide arrived 5.8% and 6.9% above those marks respectively. A quarter ago the Q3 guide landed 0.6% below consensus and that gap did much of the damage. A forward guide that beats by nearly 6% is a different event from a reported quarter that beats by 5%, because it resets the next four quarters of estimates rather than one.

Guidance style. Across the five quarters with a range on the record in our coverage, revenue has landed at or above the guide high end four times and inside the range once, with adjusted EBITDA falling below the range only once, in Q2 2026. Management's own characterisation, offered unprompted, is that it has beaten the high end in 14 of the last 16 quarters, which we report as its claim rather than as our count. The operative conclusion is unchanged from August and now points the other way: treat the midpoint as the floor, and treat the upside tail as a function of model releases the guide deliberately excludes.

Analyst Q&A Highlights

How the New Model Generation Compares to the Axon 2.0 Transition

The dominant topic in Q&A was how to size the new architecture, with several questioners reaching for the Axon 2.0 analogy that management itself had introduced in the prepared remarks. The answer accepted the comparison on technical complexity, rejected the version-numbering framing, and left the business impact deliberately open while confirming the rollout is incomplete.

Q: "Adam, in the talk track, you talked about or referenced this model step change is as big as Axon 2.0, I think you said. I am just curious, is it fair to think of this as Axon 3.0? Just really trying to frame this model update because when the model transitioned to Axon 2.0, you had a fairly significant inflection point. Obviously, the business is a lot larger now. But just if you could help conceptualize how impactful this new model will be once it is fully rolled out and the impact on the business would be great."
— Ralph Schackart, William Blair

A: "I would put it from a technology standpoint in terms of technical complexity on that level. In terms of business, look, this is the next generation of the model. We are not a direct-to-consumer AI company, and so we do not increment the model. But the reality is, this gives us a lot bigger path in the future in terms of processing more data and spinning up bigger clusters and hopefully getting better output. The team is really excited about what it means. We are still in partial rollout, so it has to roll through all of our models. There is still a lot of complex R&D work there, but it gives us a much longer roadmap than the old models would have given us, so that gets the team really excited."
— Adam Foroughi, Co-Founder and CEO

Assessment: The refusal to brand it is more informative than a brand would have been. Axon 2.0 was a step change that produced a step change in revenue; this is described as a step change in headroom, which is a different and more durable claim. "A much longer roadmap" is the operative phrase for valuation, because it converts the model-improvement pipeline from a depleting resource into a replenished one. The residual is that nobody, inside or outside the company, can say when the rollout completes.

Why the Q4 Sequential Guide Steps Down to 8% to 10%

A recurring line of questioning pressed on the apparent deceleration in the Q4 guide, noting that 8% to 10% sequential growth looks conservative against the 13.6% just delivered and against a seasonally stronger quarter for e-commerce advertisers. Management did not defend the number as a forecast; it defended the method, which is the more useful answer.

Q: "Then one for Matt. Just on the surface, it feels like the Q4 revenue guide of the 8%-10% sequential growth seems a bit more conservative versus the 14% in Q3 when you layer in the strength of e-commerce in the quarter for Q4. Just how would you frame up that sequential deceleration?"
— James Heaney, Jefferies

A: "Yeah, I mean, our guidance philosophy remains the same from what we've done over the past few quarters. We're guiding to what we know at that point in time. So, we're guiding to include the points that I said in my script, which are normal seasonality within the quarter, the model updates that we've already launched and are live and performing, and then obviously the increase in the data center cost. So, nothing there has really changed from previous quarters."
— Matt Stumpf, CFO

Assessment: The exchange confirms the guide is a floor rather than an expectation, which is the single most important thing to establish about a company whose shares fell 20% on a 0.07% miss. Taken at face value, the Q4 guide says management can see $2.38B of revenue from product already shipped. Everything the research organisation delivers between now and December is upside that the guide excludes by construction. The symmetry worth remembering is that this construction also removes any excuse if the quarter lands inside the range.

Whether the Gaming Growth Framework Has Been Raised

With gaming re-accelerating to nearly 50%, the obvious question was whether the 20% to 30% long-run framework for that vertical still applies. Management declined to raise the number, and the reason it gave matters more than the refusal: it regards vertical-level framing as analytically wrong in a single unified auction.

Q: "I guess, does the experience that you've had with this new model that just rolled out, does that change your expectations for how much runway there still is ahead to improve gaming ad revenue? Because you've been relatively specific over the course of this year about a 20%-30% growth expectation and sort of law of large numbers weighing on gaming. Would you reiterate all of that at this point, or is there potentially more headroom than you would have otherwise expected based on how it's gone so far?"
— Matthew Cost, Morgan Stanley

A: "That said, this model is for sure more accurate than the old model, and it has a lot more room to optimize going forward. If we're able to do that, of course, we would expect that gaming could continue to grow at higher rates for the long term. It's just not something that we need, though, in order to hit that 30% if we have both parts of the business growing pretty quickly."
— Adam Foroughi, Co-Founder and CEO

Assessment: The last sentence is the one to keep. It says the 30% consolidated floor does not depend on gaming outperforming, which means the floor is more robust than it looked in August when consumer was described as too small to smooth a quarter. The unified-auction argument is also a convenient defence against ever having to disclose vertical revenue, and it is used that way here. Both things are true at once: the economics genuinely are one auction, and that fact is doing useful work for the disclosure policy.

Whether the Enlarged Authorization Means Opportunistic Buying

A question framed around the stock's year-to-date decline asked directly whether the buyback is a programmatic exercise or a valuation judgment. The answer was the most unguarded statement of the call and the first time in our coverage that management has characterised its own shares as dislocated from intrinsic value.

Q: "For Matt, I guess just sort of net of what I think we were seeing just a moment ago in terms of after-hours share price reaction, your stock is still down ballpark 40% year-to-date, and it's only retraced most of the sort of post 2Q earnings step-down. Does the buyback give you flexibility, or the added authorization that you guys just layered on give you flexibility to be more aggressive, I guess, and opportunistic? If you think that the share price is meaningfully disconnected from intrinsic value, or is this mostly going to be programmatic going forward?"
— Clark Lampen, BTIG

A: "No, I mean, you saw within the quarter as well, Clark, that we did a meaningful step-up, and so that was directly as a result of what you're referring to, right? That was us opportunistically ramping up the buyback because we saw this dislocation in the stock price versus the actual intrinsic value of the company. So, we have the ability to do that going forward as well."
— Matt Stumpf, CFO

Assessment: In August the buyback moderation was explained by cash flow and accompanied by an explicit denial that it signalled anything about conviction. This quarter the acceleration is explained by conviction. Management cannot have it both ways over time, and investors should hold it to the symmetry. For now the disclosure is valuable precisely because this company gives no forward operating metric: the repurchase pace is the one quantitative read available on how management values its own equity, and in Q3 it bought $1.4B worth.

Whether the Self-Learning Contribution Can Still Be Sized

The most technically precise question of the call asked whether the 3% to 5% growth contribution historically attributed to continuous retraining on live impression data still exists, whether it appeared in the quarter, and whether it is inside the guide. Management declined to quantify it and in doing so made a disclosure that retroactively settles the Q2 debate.

Q: "In your old models, you used to continuously retrain on live impression data, and that resulted in a steady 3%-5% self-learning growth, and that is distinct from engineer-directed model uplifts. You did not address it specifically on the last conference call, so can you tell us if you see that same 3%-5% tailwind from self-learning and intra-quarter model retraining off of live impression data? Did you see it in Q3? Is it assumed in your Q4 guidance? Looking forward, how much visibility do you have as to that component of growth? Are you just wholly reliant on engineer-directed model enhancements going forward?"
— Omar Dessouky, Bank of America

A: "We just wouldn't split it up into, this is a growth driver, let's plug this number in here, and then on top of it, let's put a model enhancement, because we talked about back in the day also, we'd have one big one a year. Well, probably over the last 16 quarters, we've had multiple A/B test lifts pretty much every single one of those quarters, except the last one, that contributed to very outsized growth for the business. This stuff's not all that predictable, nor is it able to be dissected down to a specific percentage, but it all compounds into the business that we have today."
— Adam Foroughi, Co-Founder and CEO

Assessment: Two findings in one answer. The first is a withdrawal: a growth component the company once sized at 3% to 5% is now declared unmeasurable, which removes a modelling input investors had been using. The second is a confirmation that matters more: sixteen quarters with multiple test lifts in almost all of them, and exactly one exception, which was Q2 2026. That is the strongest available corroboration that the August quarter was a single missed draw rather than the first point on a declining curve, and it is more precise than anything management said in August.

What the Company Would Buy, and Why

A question on the possibility of tuck-in acquisitions to accelerate e-commerce capability drew the most strategically significant new disclosure of the call. Management described the target profile as data and customer relationships rather than technology, and anchored it to the MoPub precedent.

Q: "I think first, you mentioned in response to a question about e-commerce or web advertising, the idea of a data acquisition, and I think that there is this kind of idea percolating that there could be tuck-in M&A leverage to kind of accelerate the capabilities in e-commerce. So, perhaps if you could talk about what the ideal acquisition candidate or what the ideal data you are looking for is, that would help people kind of frame that opportunity."
— Alec Brondolo, Wells Fargo

A: "What we would look for is, can we buy data relationships, customer relationships that are valuable for us? We're going to get all these customers over the next decade anyways, but if we could buy it and get the shot in the arm to bring all these customers onto our platform quickly and get all that data into our platform quickly, it'll accelerate this business to get through what could be a decade's worth of work in much shorter amount of time. Same sort of structure as that kind of deal is what we would look for."
— Adam Foroughi, Co-Founder and CEO

Assessment: A real change in the capital-allocation story, and one the market has not priced in either direction. The phrase "a decade's worth of work in much shorter amount of time" is the clearest admission yet that organic customer acquisition in the consumer vertical is slower than the company would like, which is consistent with four months of general availability producing no disclosed advertiser count. The MoPub analogy is apt and self-serving: that deal worked, and management itself called it maybe its best. A balance sheet with $3.17B of cash, 0.05 times net leverage and a stated ambition to buy customer relationships is a configuration where discipline has to be demonstrated rather than assumed.

Whether Any Predictive Metric Will Ever Be Disclosed

The central disclosure question of this coverage arc was finally asked in the most direct form available, as an open invitation to volunteer whatever would improve investor confidence in the durability of growth. Management declined on the record, and explained why rather than deflecting.

Q: "I have a high-level question. Some of my questions were asked. As an ad tech company, for investors who are questioning the visibility of your business model and how much confidence you have in the durability of your top-line growth, is there something else incremental that you would consider sharing that would give more confidence in terms of the way your product works, the visibility you can offer, and the moat that you're building?"
— Shweta Khajuria, Wolfe Research

A: "We cannot give those KPIs that you are looking for because we are just not at that point of stability, but we have something that is a catalyst that is also very valuable. Our engineers improving our models and more data into the system is not predictable, but it can create very high upside."
— Adam Foroughi, Co-Founder and CEO

Assessment: This is the answer that keeps a multiple capped, and it should be read as policy rather than as a stage of development. Our August note set the first disclosure of a hard advertiser-count metric as one of two upgrade triggers; this exchange is that trigger being explicitly declined rather than merely unmet. An investor in this name is accepting, permanently and by design, that the primary growth driver is unobservable until it appears in revenue. The honest conclusion is that the correct response is a lower multiple than the business would otherwise command, which is different from the conclusion that the correct response is to not own it.

Whether Incremental E-Commerce Bidders Compress Incumbent Returns

A question built on advertiser fieldwork reported that some incumbent gaming advertisers were pre-emptively shifting budget away in anticipation of auction inflation from new e-commerce demand. Management rejected both the mechanism and the methodology, and in doing so gave the clearest explanation yet of why rising prices are not the same as falling returns.

Q: "They were anticipating that the pricing on the auction is going to start getting inflated because you are going to have incremental bidders from e-commerce coming in and taking up the prices. They were preemptively taking some of the budgets to some of your competitors in anticipation of what might be compression of ROI. Which kind of seemed like a strange thing to do, but it did make me worry about, I guess, the amount of inventory, which I think you have addressed in prior periods of this being, I guess, a demand-constrained industry as opposed to a supply-constrained industry."
— Stephen Ju, UBS

A: "I think these industry checks tend to end up being flawed because you're looking at a small sub-segment of the ecosystem, and the customers that are taking time talking to you all probably aren't working that hard. The ones that are working hard and succeeding probably aren't talking to you all. I would say we're a pretty good indicator on the market, and our technology has proven to be great at getting better pricing for customers. That doesn't mean that just looking at a CPM, the price is not going up. In fact, if we get better targeted so that our conversion rate goes up, the CPM will go up. But the down funnel effect of a better prediction is that the return on ad spend is going up, and that's why we're able to grow so quickly."
— Adam Foroughi, Co-Founder and CEO

Assessment: The substantive argument is correct and important: a rising cost per impression is the expected consequence of better targeting, not evidence of margin pressure on the advertiser, and the metric that matters is return on ad spend. The dismissal of channel checks is less satisfying. The company discloses no advertiser count, no retention, no spend-per-advertiser and no take rate, and then instructs investors to disregard the third-party research that exists because the willing respondents are self-selected. That may well be true, and it leaves an investor with management's assertion and the revenue line as the only evidence.

What They're NOT Saying

1. Any advertiser count, and now an explicit refusal: four months after general availability there is still no count, no cohort curve and no comparison to plan. The difference this quarter is that management addressed it directly, saying it cannot give the requested metrics because the business is not at that point of stability. We read that as a position rather than a passing absence.
2. Whether the $70,000 first-year revenue per new advertiser framework still holds: not mentioned, not asked, not updated. The figure that anchored the prior valuation case for the consumer vertical has now gone two consecutive quarters without being quantified, after management said in August it would "likely go up."
3. Lead generation, at all: a disclosed vertical in testing two quarters ago received no mention in prepared remarks and drew no question, one quarter after management said it had nothing new to report. Two calls without progress is the answer.
4. What the $86.0M of other income is: the line swung $92.7M year over year and contributed roughly $0.23 of the $1.85 EPS increase. The 8-K does not disaggregate it and no analyst raised it. With $3.17B of cash, interest income plausibly explains part of it, but that is an inference and the company has not offered one.
5. What the $15.5M of transaction-related expense relates to: the adjusted EBITDA add-back for transaction-related expense and other rose from $0.1M to $15.5M. On a call that introduced an explicit appetite for data and customer-relationship acquisitions, a sudden transaction expense is a conspicuous coincidence that nobody connected out loud.
6. When the model rollout completes: stated as partial, with the transition "still in front of us," and management said it cannot tell or commit to when a full rollout happens or how the progression goes. The single most valuable piece of forward information in the business remains undated by design.
7. Any dollar figure for consumer: two growth rates were given this quarter, a comparison against the Q4 2025 peak was given last quarter, and a revenue level has never been given at all. The disclosed constraints bound the vertical only between roughly 7% and 25% of revenue, a three-fold range.
8. That the full-year cash conversion target moved: "roughly 75%" in August became "roughly 70%-75%" this quarter, introduced with the word "still." The change was not flagged as a change, and it was delivered inside a quarter that beat the metric.
9. What "roughly 30%" means operationally: no base year, no bridge from 55.5%, no indication whether it describes 2027 or a distant steady state. Unchanged from August, and the gap between framework and delivery has not narrowed in three quarters.
10. Full-year guidance, for the third consecutive quarter: management will state that the Q4 midpoint implies more than 50% full-year growth but will not issue an annual framework. Easier to accept now than in August, and still a gap for a company of this scale.

Market Reaction

• Pre-print setup: the stock closed at $271.75 on 4 November, entering the print down 59.7% year to date against the S&P 500's +12.3%, down 55.4% over the trailing twelve months and down 3.6% over the trailing 30 days. The 52-week closing range was $228.50 to $733.60, and the low end of it had been set only three sessions earlier, on 30 October.
• The three sessions before the print: from that $228.50 closing low the shares rallied 18.9% into the release, including +10.7% on the print day itself on 14.5M shares against a 4.4M 30-day average. No company news accompanied the move; it was positioning into a quarter the market had been told would re-accelerate. On the same day APP was the worst-performing stock in the S&P 500 for the year.
• After-hours move: shares rose roughly 18% to 23% in the extended session on 4 November as the beat on both lines and the above-consensus Q4 guide landed, with the gain extending through the pre-market.
• Reaction session (5 November): opened at $349.94, a 28.8% upside gap, traded a $338.88 to $365.33 range and closed at $356.10, up 31.0% or $84.35. Volume of 22.0M shares was 5.0 times the 30-day average.
• The tape it happened in: the S&P 500 fell more than 4% on the same session and the Nasdaq Composite fell 5.5% in an artificial-intelligence and data-centre selloff, the market's worst day since April 2025. APP therefore outperformed the Nasdaq Composite by roughly 36 points in a single session.
• Peer read-across: the split was unusually clean. In-app and mobile-game advertising names rallied with APP, with one supply-side peer up 20.8% and a mobile-game ad platform up 14.9%. The open-web and connected-television demand-side peer fell 3.2%, and the two large walled gardens fell 4.0% and 2.2% with the tape.

Why a 5% revenue beat produced a 31% move. The arithmetic of the beat does not explain the reaction any better than the arithmetic of the Q2 miss explained a 19.7% decline. In both cases the market was repricing the algorithm rather than the quarter. In August it learned that the growth it had treated as a run rate was the output of a research pipeline that can miss a quarter. This week it learned three further things: that the pipeline's output when it lands is larger than assumed, that the architecture producing it is only partially deployed, and that the company's forward guide, built only from shipped product, is 5.8% above where the Street had modelled the next quarter. The last of those is what moves estimates across years rather than quarters.

Positioning did the rest. A stock that is the worst performer in its index, trading three sessions off a fresh 52-week low at roughly 12.7 times forward EBITDA, carries no expectations. Clearing the guide high end on both lines against that setup is the configuration that produces a 28.8% gap at the open. The important detail for anyone looking at the chart is that the +31.0% close is measured from a base that had already risen 18.9% in three sessions; from the 30 October closing low of $228.50 the move is +55.8%.

The relative move is the real artifact. Rising 31% on the day the Nasdaq Composite falls 5.5% is rare, and the specific reason matters here. The selloff was driven by concern that artificial-intelligence infrastructure buildout will slow and get more expensive. AppLovin is a beneficiary of applied machine learning whose incremental cost of revenue ran at 8.5 cents per incremental revenue dollar this quarter and which converts 65% of revenue into free cash flow. It is on the opposite side of that trade from the infrastructure names, and this session priced it that way.

Street Perspective

Debate: Does One Quarter Settle the Cadence Question?

Bull view: The bull case on the Street is that the question is now settled on the evidence that exists. Management diagnosed the Q2 problem as timing, said the fix was already live, and then cleared the high end of the guide by 4.8% with gaming re-accelerating to nearly 50%. The call added a specific corroboration, that sixteen quarters produced test lifts in all but one, and that one was Q2. A research pipeline that misses one quarter in sixteen is a reliable pipeline with a variance estimate attached, and the architecture that produced this quarter is only partly deployed.

Bear view: The bear camp contends that nothing structural has changed and that a single favourable draw from an unpredictable distribution is not evidence about the distribution. The company still discloses no leading indicator, has now explicitly refused to create one, and has withdrawn the one quantified growth component it used to give. The company's own Q4 guide decelerates to 8% to 10% sequentially, so by management's own arithmetic the double-digit cadence does not persist. On this reading the appropriate response to a good quarter is to keep the multiple where a lumpy grower belongs.

Our take: The bull has the better of it, narrowly, and for a reason neither side leads with. The valuable thing about this quarter is not the beat but that it resolved a conditional claim management made when it had every incentive to say something vaguer. In August we wrote that investors had no way to distinguish a 30% grower from a 55% grower in advance, and that this justified a lower multiple. That remains true as a general matter. What has changed is the estimated level of the distribution: gaming at nearly 50% on a partial rollout is a different mean than the 20% to 30% framework implied. A wide distribution around a higher mean is worth more than a wide distribution around a lower one, and the multiple has not moved to reflect it.

Debate: Is the Absence of Disclosure a Stage or a Policy?

Bull view: The argument on the Street is that the consumer vertical is genuinely too early to produce meaningful metrics, that premature disclosure of a noisy cohort curve would be worse than none, and that the revenue line is the only metric that matters anyway. Triple-digit growth in a vertical compounding toward a mid-teens revenue share is observable in the consolidated number, and the new architecture is explicitly best in exactly the sparse-data conditions that vertical presents.

Bear view: The skeptics' position is that four months after a public launch that four consecutive calls had positioned as the company's defining catalyst, and with the $70,000 per-advertiser framework now lapsed, the absence is a choice. Management has stated outright that it cannot provide the metrics requested at this stage, has pointed investors toward externally-scraped pixel counts as a substitute, and has simultaneously dismissed third-party channel research as methodologically flawed. An investor is being asked to accept the revenue line and nothing else.

Our take: The bear is right on the facts and the bull is right about what follows. This is policy, not a stage, and it should be priced permanently rather than treated as a disclosure gap that closes. The consequence is a structural discount, which is the correct outcome, and it is already in a 16.7 times forward EBITDA multiple for a business compounding above 50% with 84% margins and a shrinking share count. The error in August was treating the first disclosure of an advertiser metric as an upgrade trigger. It was never likely to arrive, and conditioning a rating on it makes the rating hostage to something the company has now said it cannot do at this stage.

Debate: Does 16.7x Forward EBITDA Still Represent Value After a 31% Day?

Bull view: The bull case being made is that the re-rating has barely begun. The stock closed at $356.10 against a 52-week closing high of $733.60, remains down roughly 47% year to date, and trades at 16.7 times our FY2026 EBITDA estimate and 12.1 times FY2027, against the roughly 25 times it carried in May. The one post-print target we could confirm, a Hold-rated raise to $445, sits 25% above here, and reported targets from the bullish desks run higher. A company buying back $1.4B of stock per quarter against a $3.6B authorization and 0.05 times net leverage is its own marginal buyer.

Bear view: The bear camp's objection is that a 31% single-session move has already captured the re-rating, that the multiple expanded from 12.7 times to 16.7 times in one day on a quarter that was pre-positioned for, and that the stock has now returned to roughly the level at which the previous quarter's disappointment left it. Nothing in the disclosure regime has improved, the creative tool has not shipped, management has flagged seasonally tougher Q1s, and the full-year cash conversion target moved down. Buying after the gap means paying for the resolution of a doubt that has only one quarter of evidence behind it.

Our take: The comparison that resolves this is the one the bear raises and then draws the wrong conclusion from. We downgraded at $346.80 on 7 August, computing roughly 17.3 times our then-FY2026 EBITDA estimate of $6.75B. The shares now sit at $356.10, 2.7% higher, at 16.7 times our raised estimate of $7.10B. That is the same price at a lower multiple, with the central doubt of the downgrade now answered and the forward guide 5.8% above where the Street stood. An investor who sold on our August call is being offered the chance to buy back in at the same price with materially more information, and the information favours ownership. The bear's individual points are all correct and all second-order against that arithmetic.

Model Update & Valuation Framework

Valuation framework. At the 5 November close of $356.10 and against 331.322M Class A and Class B shares outstanding at quarter end, market capitalisation is approximately $118.0B. Net debt is negligible: $3,516.1M of long-term debt against $3,173.7M of cash, or $342.4M, giving an enterprise value of roughly $118.3B. That is 16.7 times our FY2026 adjusted EBITDA estimate of $7.10B and 12.1 times our FY2027 estimate of $9.75B. For reference, the same calculation at the pre-print close of $271.75 implied 12.7 times FY2026, and our August report marked the downgrade at roughly 17.3 times against a lower EBITDA base.

Risk and reward. From $356.10, the base case of $405 to $470 spans +13.7% to +32.0%, with a midpoint near +22.9%. The bull case of $535 to $600 implies +50.2% to +68.5%, and the bear case of $275 to $320 implies −22.8% to −10.1%. Measured midpoint to midpoint, that is roughly +23% of upside against −17% of downside, a favourable skew of about 1.4 to 1, with the bull case as a genuine tail rather than a decoration. Base-case upside above 20% with that skew clears our Outperform threshold.

What would break the bull case. Three things, in order of likelihood. A Q4 print inside the guided range rather than above it would re-establish the Q4 guide as the real trajectory and make the Q3 beat the outlier rather than the rule. A Q1 2027 sequential decline larger than e-commerce seasonality explains would reopen the cadence question with a second data point on the bear's side. And a large acquisition funded from the balance sheet, in place of the buyback, would change the capital-allocation story that underwrites the floor. We would also re-examine the margin assumption if cost of revenue returns above 12% of revenue for two consecutive quarters.

On overriding our own upgrade trigger. Our August report conditioned any move back toward Outperform on one of two developments: two consecutive quarters of restored double-digit sequential growth, or the first disclosure of a hard advertiser-count metric. Neither has occurred. The first is one quarter of two, and the company's own guide would not complete it. The second was explicitly declined on this call. We are upgrading anyway, and the reason should be stated rather than glossed. Those triggers were constructed as tests of a single question, whether Q2 reflected a maturing improvement curve or a missed release. This quarter answered that question with evidence the triggers did not anticipate: a new architecture, partially deployed, driving gaming back to nearly 50% growth, with sixteen-quarter test-lift history confirming Q2 as the sole exception. A decision rule that cannot be updated by better evidence than it was written to accommodate is not discipline. The separate test we set for the bear case was a Q3 print at or above the guide high end accompanied by a Q4 guide that re-accelerates rather than merely holds, plus a return to the $1B quarterly buyback pace. The first and third were met outright, the first by $99.8M and the third at $1.4B. The second was met only on the Street's terms and not on ours: the guide beat consensus by 5.8% while still decelerating to 8% to 10% sequentially from 13.6%. We are not claiming otherwise, and that single unmet half is why the base case here is +23% rather than the +50% the bull tail implies.

Thesis Scorecard Post-Earnings

Graded against the standing thesis pillars carried in our thesis of record following the Q2 2026 downgrade, not against a fresh set invented this quarter.

Overall: Thesis strengthened materially. Of the five bull pillars carried into this quarter, four were confirmed and three of those recovered a status tag lost in August. Two of three bear points eased and the third stayed contained. Post-launch execution eased but survives on the creative tool rather than on the model pipeline that was its original substance. Two new negatives were added, and both are about disclosure and capital allocation rather than about the business: a permanent absence of predictive metrics, and an untested appetite for acquisitions. The asymmetry of what moved is the point. The pillars that recovered are the ones that drive the earnings; the risks that were added are the ones that cap the multiple.

Action: Upgrade to Outperform. Add on the print rather than waiting for the Q4 confirmation, because the information that mattered was disclosed this week and the price sits only 2.7% above where we downgraded. New money can buy here with a base case of $405 to $470 and should size for the variance management has been explicit about, since a quarter without a model release remains possible and would be worth roughly 20% of the share price on the Q2 precedent. Existing holders who rode the drawdown should not trim into strength that has only returned the stock to its August level on a better earnings base. We would revisit toward Hold on a Q4 print inside the guided range, on cost of revenue rising above 12% of revenue for two consecutive quarters, or on a balance-sheet-funded acquisition that displaces the buyback.