Revenue +53% but the Model Cadence Broke: First Guide Miss of the Coverage Arc, Margin and Buyback Both Step Down, Long-Term Growth Re-Anchored to "roughly 30%" — Downgrading to Hold

Results vs. Consensus

Q2 2026 Scorecard

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

Consensus is a moving target. The company's own guidance, issued on the Q1 call in May, is the commitment management is accountable to, and it is where this quarter actually broke.

In each of the three quarters of our prior coverage for which the company's guide range is on the record (Q3 2025 through Q1 2026), revenue landed at or above the guide high end. This is the first quarter in that arc where it did not, and the first where adjusted EBITDA fell outside the guided range entirely. Management confirmed both outcomes without hedging: revenue was "just below the midpoint of our guidance range," and adjusted EBITDA "was just below the range."

Year-over-Year Comparison

Note on comparability: Q2 2025 included $47.7M of income from discontinued operations (the Apps divestiture). The continuing-operations lines are the clean comparison, and on that basis EPS grew 66.4%.

Quarter-over-Quarter Comparison

Q1 2026 figures for revenue, adjusted EBITDA and the expense lines are derived by subtracting the reported Q2 quarter from the reported six-month totals in the same filing, so they are exact rather than estimated. Q1 free cash flow is approximate because the six-month capital expenditure split is not separately disclosed.

Revenue assessment. Growing 52.8% on a $1.26B base is not a bad outcome in isolation, and the print sits inside the guided range. The problem is the shape. Sequential growth of +4.4% follows +11% in Q1, and management confirmed on the call that this was the only single-digit sequential quarter in the last twelve. The entire premium multiple on this stock has rested on a specific mechanism: models improve, advertiser return on ad spend improves, budgets step up, and the step-ups arrive reliably enough to look like a growth algorithm rather than a research pipeline. This quarter demonstrated that the mechanism is a research pipeline. Management was candid about it, which we credit, but candor does not restore the predictability premium.

Margins assessment. The year-over-year margin story is still excellent: 83.9% adjusted EBITDA margin against 80.9%, up roughly 300bp. The sequential story is where the direction changed. Q1's 84.5% was the peak; Q2 came in 61bp lower, and the Q3 guide implies about 83.3%, another 56bp of compression. The distinction matters: the Q2 guide was approximately 84% to 85%, so this quarter was a miss against a flat guide, and the Q3 guide is the first time management has guided the margin down from the Q1 peak. Quarter-over-quarter EBITDA flow-through fell to 70% from 86% in Q1. Management's framing, that they manage to EBITDA dollars and free cash flow rather than a margin percentage and will deploy dollars whenever incremental revenue follows, is a defensible operating philosophy. It is also a materially different investment proposition from the one where margin ratcheted up every quarter.

EPS assessment. Diluted EPS of $3.76 was in line to modestly ahead of the $3.72 to $3.75 consensus, but the composition deserves scrutiny. Other income, net swung to a $62.4M gain from a $22.3M expense a year ago, an $84.7M year-over-year swing that sits entirely below the operating line and is excluded from adjusted EBITDA. Applying the quarter's 15.9% effective tax rate, that gain contributed roughly $0.16 per diluted share. Strip it out and EPS lands near $3.60, below the consensus range. The effective tax rate itself rose to 15.9% from 12.7%. In short, the operating measure missed and the GAAP measure was rescued by below-the-line items. Share count did real work as well, with diluted shares down 1.5% year over year, though the buyback that drives it slowed sharply this quarter.

Segment and Vertical Performance

AppLovin does not disclose revenue by vertical, so any gaming versus consumer split is an estimate. What the company did disclose this quarter is directional and worth separating from inference.

Mobile Gaming — the deceleration was here, and the ecosystem read is contested

Gaming remains the majority of revenue and is where the sequential shortfall originated. Management was careful to separate a demand problem from a product-cadence problem, and the disclosed evidence supports their version: publisher earnings across the MAX marketplace grew double digits quarter over quarter, and the company's share of publisher waterfalls was unchanged. If advertisers were pulling back or a competitor were taking share, neither of those would look the way they do.

"Importantly, nothing we saw suggested weakening advertiser demand or a change in the competitive environment. In fact, MAX publisher earnings grew double digits quarter-over-quarter, and our share of publisher waterfalls remained consistent." — Adam Foroughi, Co-Founder and CEO

The complication is that AppLovin is large enough to be an input to the ecosystem statistics it cites. Management said as much: when the company does not push model lifts, category-wide user acquisition slows with it. That is a candid statement of scale, and it also means the ecosystem health check is partially self-referential.

Assessment: We accept that this was not a demand or share event. But a gaming business that grows only when a model release lands is a business whose near-term revenue is a function of an R&D calendar, and the company has now told us that calendar is not reliably quarterly. That belongs in the multiple, not the estimates.

Consumer — genuinely strong, and explicitly not yet big enough to matter

The consumer vertical had the quarter the bulls wanted. Advertiser spend set a record at 28% above Q4 2025 levels, which is the relevant comparison because Q4 is the seasonal peak for these advertisers and Q2 is structurally soft. Clearing the prior peak by 28% in the weak part of the year is a steep curve by any standard, and it is the second consecutive quarter of this pattern after April's record cited on the Q1 call.

"Advertiser spend set another record, finishing 28% above Q4 2025 levels. And remember, Q4 is the seasonal peak for these advertisers. Growing well past peak season levels in a seasonally slow quarter tells you how steep this curve is. Consumer isn't yet large enough to fully smooth a quarter like this, but that will change as we continue to ramp up our consumer business every quarter." — Adam Foroughi, Co-Founder and CEO

The second sentence of that quote is the one that matters for the thesis. The diversification argument for consumer was that it would reduce dependence on the gaming model cadence. Management is confirming it cannot do that yet.

Assessment: Consumer is working, and management's characterization of the trajectory is supported by the only metric they disclose. It remains roughly a fifth of revenue on our estimate, growing off a base too small to offset gaming volatility for at least several more quarters. This is a 2027 diversifier, not a 2026 one.

Cost Structure — Where the Money Went

Two things stand out. Research and development at 5.19% of revenue is up 170bp year over year and has roughly doubled in dollar terms, which is the visible cost of the architectural work management described. And sales and marketing was flat at 3.30% of revenue in the very quarter the company opened its platform to the public. The paid marketing scaling that the Q1 report flagged as an H2 signpost has not begun; management has substituted partnerships as the acquisition channel instead.

Key Topics & Management Commentary

Overall Management Tone: Direct and unusually plain about a shortfall, opening with the miss rather than burying it and declining to blame demand, competition or macro. The confidence was reserved for the mechanism (models improve, budgets follow) rather than for near-term numbers, and the one place the posture turned noticeably softer was around timing commitments, where the answers moved from dates to conditions. Compared with the assured, catalyst-dense posture of the Q1 call, this was a management team narrating a process rather than a plan.

1. The Miss, Owned Without Qualification

Management led the call with the shortfall rather than working toward it. There was no attempt to reframe a below-midpoint revenue print as a beat on some other basis, and no appeal to macro conditions or competitive dynamics. For a company that has spent the past several quarters defending itself against sentiment-driven narratives, choosing to open with the failure is a meaningful change in communication posture.

"This quarter, we delivered almost $2 billion in revenue, which was just below the midpoint of our guidance range, and our adjusted EBITDA was just below the range. We've always managed this business with the goal of outperforming our own expectations. And this quarter, we fell short of that standard." — Adam Foroughi, Co-Founder and CEO

The CFO closed the prepared remarks in the same register, describing it as "not the quarter we hold ourselves to."

Assessment: Credibility preserved. A management team that owns a small miss cleanly is more trustworthy on the next large one than a team that explains it away. This does not change the numbers, but it does mean we take the Q3 guide at close to face value.

2. What Actually Happened: A Research Pipeline, Not a Growth Algorithm

The substantive disclosure of the quarter is that the pace of meaningful model improvement ran lighter than normal, and the next step-up landed just after the quarter closed. This is a timing explanation, and it is unfalsifiable in the moment. It is also the most consequential thing said on the call, because it reframes what kind of business this is.

"This quarter came down to timing. Our pace of meaningful model improvement was lighter than normal during the quarter and the next step-up in model performance landed just after quarter end." — Adam Foroughi, Co-Founder and CEO

Pressed in Q&A on whether the absence of a lift could be diagnosed after the fact, management declined to manufacture a tidier answer, describing the process as a portfolio of A/B tests with uncertain outcomes and acknowledging that some periods simply will not produce material lifts.

Assessment: This is the downgrade. Investors were underwriting a compounding machine with quarterly cadence; management has now clarified it is a research organization whose output is lumpy and whose lumps are visible in the revenue line within one quarter. Both descriptions can be true of the same excellent business. Only one of them supports a 25x EBITDA multiple.

3. Q3 Reacceleration Is Already Banked, and Deliberately Conservative

The strongest element of the print is the Q3 guide, and specifically what management says is inside it. Revenue of $2,055M to $2,085M implies 46% to 48% year-over-year growth and 7% to 8% sequential, a clear reacceleration from Q2's +4.4%. Critically, the guide is built on model improvements that are already live and performing, and explicitly excludes anything not yet deployed.

"That outlook reflects the model improvements that are already live and performing, continued scaling in our consumer vertical, normal seasonality and the higher training and compute costs I mentioned. It does not assume additional model releases that have not yet been deployed." — Matt Stumpf, CFO

That last sentence is the single best reason to own the stock here. If the guide contains no speculative uplift, then the historical pattern of additional in-quarter releases represents upside rather than the assumption being tested.

Assessment: Genuinely constructive, and it is why this is a Hold rather than something worse. The reacceleration is not a hope; it is a shipped product already in the run rate. The offset is that the same guide sits about 0.6% below where consensus had been, so the reacceleration is not enough to restore the prior trajectory.

4. Consumer Set Another Record, and Management Said It Cannot Yet Help

Consumer advertiser spend finished 28% above Q4 2025 levels, clearing the seasonal peak during the seasonally weak quarter. This continues the pattern established with April's record cited on the Q1 call and is the clearest evidence that the second vertical is compounding independently of the gaming model cycle.

The candid coda was that consumer "isn't yet large enough to fully smooth a quarter like this." That is management pre-empting the obvious question of why a fast-growing second engine did not offset a soft first engine.

Assessment: The operational signal is good and we do not want to understate it. The thesis implication is narrower than it looks. Consumer is not yet a shock absorber, and on management's own framing it will take several more quarters of ramp before it becomes one. Until then, quarterly results remain a levered bet on the gaming model calendar.

5. General Availability Happened, Then Was Deliberately Throttled

The June general availability launch was the binding catalyst in our Q1 thesis. It occurred, under the original AppLovin Ads Manager name rather than the AXON branding, and management immediately set expectations that it would not move near-term numbers. The sequencing is intentional: mid-market advertisers first, where the model performs best today, with the long tail deferred until the data compounds.

"During the quarter, we also opened up our platform to the public under its original name, AppLovin Ads Manager. As we said last quarter, we never expected the public launch to change the business overnight. We're sequencing deliberately mid-market advertisers first, where the platform performs best today, with the long tail unlocking as our data compounds, the same way gaming developed." — Adam Foroughi, Co-Founder and CEO

Asked directly to characterize onboarding volume six weeks in, management offered no advertiser count, no cohort metric and no comparison to plan beyond "as expected," and confirmed the base is few advertisers spending more rather than many spending little. Asked when the long tail opens, the answer was explicit that it is not predictable.

Assessment: This is the second pillar of the downgrade. The Q1 report treated June general availability as the event that would inflect advertiser count and re-rate the stock. The event has now happened, produced no disclosed operating metric, and been reframed as the beginning of a multi-year data-compounding process with no timetable. A catalyst that converts into an open-ended process is not a catalyst.

6. Compute Step-Up: The Ten-Cent Framework Holds, the Timing Does Not Flatter

The sequential cost increase was driven by compute for training existing models and developing new ones. Management reiterated the framework that roughly ten cents of every incremental revenue dollar goes to compute, and said the guide and the forthcoming data center disclosure both sit at that level.

"The primary driver of the sequential increase in costs was higher compute associated with training our existing models and with new model development. And that higher compute run rate is reflected in our outlook." — Matt Stumpf, CFO

The filed numbers support the framework on a year-over-year basis: incremental cost of revenue divided by incremental revenue is 10.6 cents. On a sequential basis it is 27 cents, which is precisely what happens when training spend lands in one quarter and the revenue it produces lands in the next. Management framed the architectural work as building models that benefit more from additional training compute, which is a deliberate bet on scaling.

Assessment: We do not read this as margin erosion from competitive pressure, and the ten-cent framework is intact on the annual math. We do read it as a structural change in working-capital-like timing: revenue now lags the spend that creates it, which introduces quarter-to-quarter noise the model previously did not have.

7. The Long-Term Algorithm Was Re-Anchored to "roughly 30%"

Management stated a forward compounding rate of roughly 30% annually over the longer term. Prior framing in our coverage referenced a 20% to 30% long-term framework that the business had exceeded for twelve consecutive quarters, with actual growth running 50% to 70%. Saying "roughly 30%" out loud on a miss quarter, rather than pointing again at the gap between framework and delivery, reads as a deliberate reset of expectations.

"Over the longer term, as we continue improving gaming and expanding consumer, we believe this business can compound at roughly 30% annually." — Adam Foroughi, Co-Founder and CEO

Assessment: Thirty percent compounding is an outstanding business. It is also roughly half the growth rate embedded in the valuation this stock carried in November. The distance between those two facts is most of the drawdown.

8. Free Cash Flow Conversion Fell to 53.5%

Free cash flow of $863.3M grew only 12.4% year over year against 58.5% adjusted EBITDA growth, converting at 53.5% of adjusted EBITDA versus 75.4% a year ago. Management attributed the gap to the timing of international cash tax and interest payments and had flagged the risk on the prior call. The filings corroborate the tax explanation: cash paid for income taxes in the first half was $639.8M against $100.6M in the prior-year period.

"As I previewed on the last quarter's call, conversion was below our normal cadence in the second quarter due to the timing of international cash tax and interest payments. This is a timing dynamic, not a change in the earnings power of our business." — Matt Stumpf, CFO

Management expects conversion to improve in Q3 and to normalize to roughly 75% of adjusted EBITDA for the full year. Reaching that requires a materially stronger second half, since first-half conversion ran near 68%.

Assessment: The tax-timing explanation is documented rather than asserted, so we take it. The item to watch is whether the full-year 75% target holds, because the buyback pace is now explicitly a function of quarterly free cash flow rather than of conviction.

9. Buybacks Halved, With the Reason Stated Plainly

The company repurchased and withheld approximately 1.1 million shares for $551.3M, against roughly $1.0B in Q1. Shares outstanding ended at 335 million with about $1.8B remaining under the authorization. Management tied the moderation directly to the quarter's weaker cash generation and pre-emptively denied any signal about conviction.

"Our choice to moderate the pace of our buybacks this quarter relative to the roughly $1 billion we have deployed in the first quarter reflects consideration of our lower free cash flow during the quarter and does not indicate a change in conviction or in how we intend to use the authorization going forward." — Matt Stumpf, CFO

Assessment: The Q1 report treated a sustained $800M to $1B quarterly pace as a thesis signpost, on the reading that management would lean into weakness. The stock fell through the quarter and the buyback halved. The explanation is coherent, but the behavioral signal investors thought they were buying (aggressive repurchase into drawdown) did not appear. With roughly $1.8B left on the authorization and the stock now near 52-week lows, Q3 deployment is the test of the stated conviction.

10. Creative Tooling: A Dated Commitment That Did Not Advance

Delivering usable video creative automatically is the identified constraint on self-service conversion, and our Q1 report treated it as the binding pre-launch execution requirement. On the Q1 call in May, management put a timeline on it and tied it explicitly to the general availability launch.

"The main thing that we talked about last call, resolving, we're still in the midst of resolving. We will over the next couple of weeks before we go to general release, and that's delivering video out of the box… So we think we're pretty close to having that ready to go for anyone to just plug in and get video out of the box. And then we talked about opening up the platform in June. So in a matter of weeks, we'll think we're there." — Adam Foroughi, Co-Founder and CEO, Q1 2026 call (May 6, 2026)

The platform went to general release in June as planned. The creative tooling that was supposed to precede it did not.

"The creative is the biggest hurdle in our system. So nothing has really changed there. Lengthy video plus interactive end card, which is in a common ad unit, we can auto-generate the interactive end card with pretty high efficiency at this point. We're not at the point where we can get a high-quality video for 30 to 60 seconds in the hand of an advertiser out of the box. So that's still work in progress." — Adam Foroughi, Co-Founder and CEO, Q2 2026 call

Management connected this directly to the long-tail problem: smaller advertisers signing up directly typically lack the video assets the platform's ad formats require, which is why the mid-market sequencing exists.

Assessment: This is the third pillar of the downgrade and the one we weight most heavily, because it is the rare case where a specific, dated commitment can be checked against a later transcript. Three months after "in a matter of weeks, we'll think we're there," the answer is that nothing has really changed. The launch shipped without the tool that was supposed to make the launch work, which is a sufficient explanation on its own for why general availability produced no disclosed advertiser metrics.

11. Partnerships Replaced Paid Marketing as the Acquisition Channel

A genuinely new element this quarter: advertiser acquisition is now running through third-party partnerships, including with a large e-commerce analytics provider, rather than through the paid marketing campaign that prior calls had teed up for the second half. The logic is that buying ads brings in long-tail advertisers the model cannot yet serve well, while partners deliver targeted mid-market accounts.

"So rather than just start by buying ads and bringing in long tail, which is harder to make work with where the model evolution is today, we're going through these partnerships to try to get very targeted customers into the platform." — Adam Foroughi, Co-Founder and CEO

The financials corroborate the strategy shift. Sales and marketing was flat at 3.30% of revenue and grew only 35% year over year, well below revenue growth, in the quarter of the public launch.

Assessment: Strategically sensible and margin-friendly, and it explains why the expected H2 marketing ramp has not appeared. It is also slower and less scalable than the paid-marketing flywheel the Q1 framework assumed, and it makes advertiser count dependent on partner-by-partner negotiation. Management said the optimal structure is still being worked out.

12. SEC Inquiry Closed With No Recommended Action

Management used the prepared remarks to close out the reported SEC inquiry, characterizing it as a voluntary request the company never deemed material and confirming the agency concluded with no recommended action.

"It was a voluntary request, which we never deemed material. The SEC has recently advised us that it concluded its inquiry with no recommended action. We're pleased to have it resolved." — Matt Stumpf, CFO

Assessment: An unambiguous positive and the removal of a genuine tail risk that had been part of the bear file since the short-seller campaigns. It got almost no attention in the reaction, which tells you what the market was focused on instead.

Guidance & Outlook

Guide composition. The Q3 outlook is built on four disclosed components: model improvements already live and performing, continued consumer scaling, normal seasonality, and the higher training and compute run rate. Management stated explicitly that it does not assume any model release not yet deployed. That is a meaningful narrowing of the guidance's speculative content relative to prior quarters, and it should raise confidence in the midpoint while capping the size of any beat that does not come from a new release.

Implied quarter-over-quarter ramp. The $2,070M midpoint requires +7.6% sequential growth, which would restore the pace to roughly the Q1 level (+11%) minus the model-cadence gap. For context on the full year: first-half revenue of $3,766.1M plus the Q3 midpoint implies $5,836.1M through nine months, leaving a Q4 in the $2.2B to $2.3B range to reach our $8.05B to $8.15B full-year framework.

Street at. Consensus for Q3 revenue sat near $2.08B before the print, so the $2,070M midpoint arrived roughly 0.6% light. The guide is a reacceleration in absolute terms and a small disappointment in relative terms, which is precisely the combination that produces a multiple reset rather than an estimate reset.

Guidance style. Across the three quarters of our prior coverage with a guide range on the record, this management team guided conservatively and beat, with revenue landing at or above the guide high end each time. That pattern broke this quarter. The Q3 guide's explicit exclusion of undeployed model releases suggests they are trying to restore the beat-and-raise setup by lowering what the guide promises. Investors should treat the midpoint as more reliable than in prior quarters and the upside tail as thinner.

Analyst Q&A Highlights

What Specifically Failed to Land in the Quarter

The first substantive question of the call went straight at the mechanism, asking what model breakthrough was expected and did not arrive. Management contextualized the quarter against the twelve preceding it, confirming that this was the only one with single-digit sequential growth, and framed model improvement as a continuous process that is not normally broken out because it normally works.

Q: "I just wanted to get under the hood of the gaming advertising business, and maybe it's difficult to explain it in a simple response, but just curious exactly what the sort of model breakthroughs that you were looking for that didn't happen? Just trying to understand like the timing of it and kind of what happened in the quarter, just to..."
— James Heaney, Jefferies

A: "If you go back the last 12 quarters, every quarter, we've had really good growth, except for Q2 in single digits. I think every other quarter since Axon 2 has been double digits. Now Q2 starts as the seasonally weakest quarter. But every quarter that we've had these outsized growth rates, we've had improvements to our model. So our team isn't just sitting there doing nothing for the last 12 quarters. It's constant improvement. We just don't really break it out."
— Adam Foroughi, Co-Founder and CEO

Assessment: The answer is honest and it is also the disclosure that reprices the stock. Twelve quarters of double-digit sequential growth had been read as a run rate. Management has now explained that it was twelve consecutive successful research outcomes, which is a different thing to extrapolate.

Whether a Missing Uplift Can Be Diagnosed After the Fact

A follow-on line of questioning pressed on whether the absence of a model lift was diagnosable, implicitly probing whether something structural had changed. Management declined to offer a tidier answer than the truth, describing the process as hypothesis testing with uncertain returns and explicitly accepting that some periods will produce no material lift.

Q: "Just, Adam, maybe to start on the model improvements. I'm guessing you've kind of seen this before, but just any sense of why you didn't get the model uplift that you would have expected in the quarter? I don't know if there's a way to sort of diagnose that?"
— Ralph Schackart, William Blair

A: "Look, it's R&D, right? Like there's no guarantee that we're always going to have lifts in every single period of 3 months. The team is always working on lifts. And usually, in every single quarter, it's not one lift that we get. We compound multiple small lifts. The impact was just smaller in Q2, followed by a pretty material uplift in Q3 in the early part of Q3."
— Adam Foroughi, Co-Founder and CEO

Assessment: Refusing to invent a diagnosis is the right call for long-term credibility and the wrong outcome for anyone modeling next quarter. There is no leading indicator here that an investor can track. That is a permanent feature, not a temporary disclosure gap.

Margin Trajectory as Compute Scales

Questioning turned to whether the compute step-up implies a structurally lower margin. Management accepted that short-term variability is now part of the profile and asked investors to read spending increases as a positive signal, on the grounds that dollars are only deployed against identified incremental revenue. A longer-term floor in the low 80% range was reiterated.

Q: "And then, Matt, as compute costs continue to increase in scale, how should we think about the margin profile of the business?"
— Ralph Schackart, William Blair

A: "To the margin question, over the longer term, we may see some variability, which is consistent with what we've told analysts and investors in the past as well that we may have some short-term fluctuation in the margin profile of the business, but it should be indicative and investors should look at that as a positive because when we're spending, we're very cautious about how we're spending and we're only spending if we see that there is incremental revenue behind that."
— Matt Stumpf, CFO

Assessment: Asking investors to read higher spending as a positive signal only holds if the incremental revenue shows up on a visible lag. This quarter it did not show up inside the quarter, which is exactly the case that makes the framing hard to verify in real time. The low 80s is now the operative margin anchor, not the 84.5% peak.

Six Weeks Into General Availability, No Metrics Offered

The most direct challenge on the launch asked for an explicit assessment of onboarding volume against expectations, and whether the eventual model is many small advertisers or few large ones. Management confirmed the latter shape for now, characterized the launch as in line with a deliberately unaggressive plan, and gave no count.

Q: "So you opened your platform, I think, to everybody in late June. It's been about, I think, 6 weeks now since it was open. Can you tell us a little bit about how you would evaluate the performance of your onboarding in terms of the number of advertisers onboarded, whether it kind of met your expectations, exceeded or was below?"
— Omar Dessouky, Bank of America

A: "It'd be the latter on your own words, Omar, that few advertisers contributing more today. So we look at the launches as expected. We said it's not going to be a big marketing push behind the launch. We're targeting the marketing relationships and the dollars that we spend to where we can go get those mid-market brands."
— Adam Foroughi, Co-Founder and CEO

Assessment: Describing the launch as "as expected" is not a metric, and after an event positioned for four consecutive quarters as the company's defining catalyst, the absence of any onboarding disclosure is conspicuous. On the same exchange management said the timing of the broader unlock is not predictable. The catalyst has been converted into a process.

The Creative Constraint on Self-Service Conversion

A line of questioning on the self-service funnel asked what is working post-launch and what still needs improvement. The answer identified creative as the unchanged bottleneck, distinguishing between interactive end cards, which can now be auto-generated efficiently, and long-form video, which cannot yet be produced to a usable standard automatically.

Q: "And then a little bit on feedback from commerce advertisers. What's been working well with the self-service platform now that you're GA, what are some things you still need to improve upon or that you'd like to get better?"
— Robert Sanderson, Loop Capital

A: "So both answers are sort of related. The creative is the biggest hurdle in our system. So nothing has really changed there. Lengthy video plus interactive end card, which is in a common ad unit, we can auto-generate the interactive end card with pretty high efficiency at this point. We're not at the point where we can get a high-quality video for 30 to 60 seconds in the hand of an advertiser out of the box. So that's still work in progress."
— Adam Foroughi, Co-Founder and CEO

Assessment: Measured against the prior call's "matter of weeks," this is a full quarter of no progress on the single item management itself identifies as the binding constraint on long-tail conversion. Until it resolves, the addressable advertiser base is structurally capped at those who already own broadcast-quality video.

Mobile Gaming Ecosystem Health

A skeptical question cited third-party evidence of cost-per-install inflation and mobile game downloads falling 10% to 15% year over year, asking whether the category itself is deteriorating. Management rejected the download metric as a mix artifact of the shift away from hyper-casual titles, argued in-app purchase data is systematically undercounted as transactions move off-platform, and made a notable admission about the company's own influence on category statistics.

Q: "I'd love to ask about the health of the mobile game ecosystem. I think a lot of us are having conversations with mobile game publishers and developers. They're speaking to CPI inflation, maybe waning ROAS, there's some market data that suggests that mobile game app downloads are down something like 10% to 15% year-over-year over the last several months. And just how do you guys feel about the category?"
— Alec Brondolo, Wells Fargo

A: "And so when we have a quarter where we don't push lifts, that's not great for the category as well. We're driving growth in the category given the scale that we operate at. Then when we go follow it up and we push a model release early in Q3, if you talk to a lot of those same advertisers, they'll say, install rates went up, CPIs went down, performance has improved."
— Adam Foroughi, Co-Founder and CEO

Assessment: The rebuttal on downloads is reasonable and the deeper-funnel mix argument is sound. But the admission that the company's own release cadence moves category-level statistics cuts both ways: it defends this quarter while confirming that the ecosystem indicators cited as evidence of health are partly a reflection of the company's own activity.

Compute Investment Beyond Q3

A question on the forward investment runway sought visibility into compute costs past the guided quarter. The response reaffirmed a specific unit-economic framework, tied it to a forthcoming disclosure in the quarterly filing, and was immediately qualified by the CEO, who made clear that a material model breakthrough would trigger unbudgeted spend without waiting for a quarter boundary.

Q: "How should we be thinking about the incremental investment in tech and compute going forward? Obviously, we have the 3Q guide, so we can kind of back into that. But just thinking beyond 3Q, should we expect an elevated runway for investment? And just how much visibility into those training and compute costs going forward do you have?"
— Aaron Lee, Macquarie

A: "I mean, we don't expect any departure from the higher-level guidance that we've given to analysts and investors in the past that of the incremental dollar in revenue, we're spending about $0.10 on compute. We're at that level within the guide. And we're at that level if you look at the data center disclosure, you'll get in the 10-Q as well."
— Matt Stumpf, CFO

Assessment: The framework is verifiable and it checks out on the year-over-year math at 10.6 cents. The CEO's follow-on qualification, that a proven uplift would be funded immediately regardless of the guide, is the honest position for a research-driven business and simultaneously removes the ceiling the framework appeared to provide.

What They're NOT Saying

1. Any advertiser count, ever. Six weeks past the launch that four consecutive calls positioned as transformational, there is no number of advertisers onboarded, no cohort curve, no comparison against an internal plan. Management said only that it looked "as expected," and that was the entire disclosure. A company confident in the launch metrics would give at least one.
2. What "roughly 30%" applies to, and from when. The long-term compounding rate was stated without a starting year, a bridge from the current 46% to 48% growth, or an indication of whether it describes 2027 or a further-out steady state. That ambiguity is doing a lot of work.
3. Any quantification of the Q3 uplift. The model release that landed after quarter end is described as material and as already live and performing, and it underpins the entire reacceleration. Its magnitude is never sized, so investors cannot distinguish a lift that recovers the trend from one that merely stabilizes it.
4. Whether the $70,000 per-advertiser figure still holds. Asked directly, management said it will "likely go up" because of the mid-market targeting, but declined to update the figure or provide any post-launch cohort data against it. The framework that anchored the prior valuation case is now qualitative.
5. A revised timeline for anything. Generative video creative, long-tail unlock, lead generation, connected TV: each was addressed and each answer replaced a date with a condition. The most explicit statement on timing was that it is not predictable.
6. Full-year guidance, again. Two quarters into 2026 with a materially changed trajectory, there is still no annual framework. The absence was more tolerable when every quarter beat the high end.
7. Why the buyback was not larger given the price. The cash-flow explanation covers the arithmetic. It does not address why a company holding $3.05B of cash, with net leverage near 0.1x and the stock approaching 52-week lows, halved its repurchase rather than borrowing against a nearly unlevered balance sheet.

Market Reaction

• Pre-print setup: the stock closed at $417.80 on August 5, entering the print down 38.0% year to date against the S&P 500's +12.8%, down 23.2% over the prior 30 days, and up 10.5% over the trailing twelve months. The 52-week closing range was $366.91 to $733.60. This was already a badly de-rated stock heading into the release.
• After-hours move: shares fell roughly 16% in the after-hours session on August 5, to about $350, as the top-line shortfall and the model-cadence commentary landed.
• Reaction session (August 6): opened at $338.45, a 19.0% downside gap, traded a $332.19 to $352.00 range, and closed at $335.67, down 19.7% or $82.13. Volume of 15.2M shares was 2.6x the 30-day average.
• Follow-through (August 7): closed at $346.80, up 3.3%, leaving the cumulative two-session move at −17.0% from the pre-print close. Both closes are below the prior 52-week closing low, so the stock set a fresh 52-week low on the reaction day.
• Peer context: the move was idiosyncratic rather than sectoral. On the same session one ad-tech peer rose 18% while another fell 6%, indicating traders were separating names rather than de-rating the group.

Why a $1.2M miss cost roughly $27.5B of market value in a single session. The proximate figures are almost comically small: adjusted EBITDA missed the low end of the guide by 0.07%, revenue missed the midpoint by 0.33%, and EPS was in line. Nothing in that arithmetic supports a 19.7% move. The market was not repricing the quarter. It was repricing the algorithm.

Three disclosures did the damage, and none of them is an estimate revision. First, sequential growth decelerated to single digits for the first time in twelve quarters, and the explanation was that the research pipeline did not deliver on schedule. That converts a perceived growth rate into a probability distribution. Second, management stated a long-term rate of roughly 30%, materially below both the delivered trajectory and what the multiple implied. Third, the general availability launch, the event that was supposed to inflect the second growth vector, arrived with no metrics and an explicitly undated path to the long tail. A stock trading at a premium for predictability lost the predictability, not the growth.

The de-rating in context. At $346.80 the stock sits 52.7% below its 52-week closing high of $733.60 and 48.8% below the post-Q3 peak that our November coverage marked. The prior drawdown, from November through January, was a sentiment event that the Q4 print refuted with operating results. This one is different in kind: the operating results are the source. That distinction matters for anyone treating the current level as a repeat of the setup that worked eight months ago.

Street Perspective

Debate: Is the Model Cadence a Timing Artifact or a Maturing Curve?

Bull view: The mechanism is intact and this was a scheduling accident. The uplift landed weeks later, it is already live and performing, and the Q3 guide of 46% to 48% growth is built on it without assuming anything further. Demand indicators held: publisher earnings grew double digits sequentially and waterfall share was unchanged. A business that missed by 0.07% on EBITDA and reaccelerates the following quarter has not broken.

Bear view: Twelve quarters of uplifts is exactly what the early, easy part of a model-improvement curve looks like. As the models mature, the marginal gain per experiment falls, and this is the first observable evidence of that. On this reading the core gaming business is approaching share saturation, so incremental growth must come from take rate or from a consumer vertical that management concedes is too small to matter yet. That is a lower-quality growth mix and warrants a lower multiple regardless of any single quarter.

Our take: The bear framing is unfalsifiable in the near term but directionally more useful. One missed cadence does not establish diminishing returns. What it does establish is that investors have no way to distinguish the two hypotheses in advance, because management explicitly cannot forecast which quarters produce lifts. When the difference between a 30% and a 55% grower is unobservable until after the fact, the correct response is a lower multiple, which is what happened. We would need two or three consecutive quarters of double-digit sequential growth to accept the timing-artifact reading.

Debate: Is the Consumer Vertical a Second Engine or an Expensive Option?

Bull view: Consumer spend cleared the Q4 seasonal peak by 28% in a seasonally weak quarter, the second consecutive quarter of that pattern. The vertical is roughly where gaming was in its early compounding phase, with the same flywheel of more advertisers, more data, better models. Mid-market sequencing is disciplined rather than defensive, and the partnership channel produces higher-quality accounts than paid acquisition would.

Bear view: After roughly eighteen months, the vertical is still around a fifth of revenue, cannot absorb a soft gaming quarter, has no disclosed advertiser count, and remains gated by a creative tool that has not shipped. Substituting partnerships for paid marketing is a slower path dressed up as a strategic choice, and the long-tail unlock has no date.

Our take: Both are right about different time horizons, and the bear owns the one that matters for the next several quarters. The 28% figure is real evidence of product-market fit, and we do not doubt consumer becomes a genuine second engine in time. But that horizon now has no timetable attached to it, the creative constraint that gates the long tail is unresolved, and management has told us plainly that consumer cannot smooth a quarter like this one. An option with an indefinite expiry is worth less than a catalyst with a date, and the market has correctly re-priced from the second to the first.

Debate: Does 17x EBITDA Make This a Value Opportunity?

Bull view: A business growing revenue 46% to 48% with 83% EBITDA margins, a nearly unlevered balance sheet, and $1.8B of buyback authorization trading at roughly 17x forward EBITDA is mispriced on any historical comparison. The multiple has halved from the level of eight months ago while the business still compounds far above the market. The last time the stock de-rated this hard, the next print refuted the fear and it rallied 17%.

Bear view: The multiple compressed because the earnings stream it was applied to became less predictable and its terminal growth rate was revised down by management. Seventeen times is not obviously cheap against a business the CEO says compounds at roughly 30%, particularly with margin guided down two quarters running, free cash flow conversion at 53.5%, and the buyback slowing precisely when conviction would be most visible.

Our take: The valuation is genuinely more attractive than at any point in our coverage, and that is why this is a Hold and not an Underperform. But the bull's comparison to the November drawdown is the wrong analogy. That episode was a sentiment de-rating that operating results disproved. This one was caused by operating results and by management's own revision of the long-term algorithm, and there is no equivalent refutation available in Q3 because the Q3 guide is already known and already below where consensus sat. At roughly 17x our FY2026 EBITDA estimate against a business now guided to a lower and lumpier growth path, the stock is approximately fairly valued. We want either two quarters of restored cadence or a disclosed advertiser metric before paying up again.

Model Update & Valuation Framework

Valuation framework. At the August 7 close of $346.80, and against 335.291M shares outstanding at quarter end, market capitalization is approximately $116.3B. Net debt is small: $3,515.1M of long-term debt on the balance sheet against $3,053.3M of cash, or roughly $462M, giving an enterprise value near $116.7B. That is about 17.3x our FY2026 adjusted EBITDA estimate of $6.75B and roughly 13.3x our FY2027 estimate of $8.8B. For comparison, the same calculation at the pre-print close of $417.80 implied a market capitalization of $140.1B, and our Q1 report marked the stock at roughly 25x forward EBITDA at $530.

Risk and reward. From $346.80, our base case of $340 to $400 spans −2% to +15%, with a midpoint near +7%. The bull case of $470 to $540 implies +36% to +56%, and the bear case of $250 to $290 implies −28% to −16%. Measured midpoint to midpoint, that is roughly +7% of upside against −22% of downside, a skew of about one to three against us; the bull case is the offsetting tail rather than the central expectation. Single-digit base-case upside with that skew is the arithmetic definition of a Hold under our framework. We would revisit toward Outperform on either of two developments: two consecutive quarters of restored double-digit sequential growth, or the first disclosure of a hard advertiser-count metric for the post-launch cohort.

What would break the bear case. A Q3 print at or above the high end of the guide, accompanied by a Q4 guide that reaccelerates rather than merely holds, would suggest the Q2 cadence gap was a genuine one-off. Resumption of the $1B quarterly buyback pace at these prices would also carry real signaling value given management explicitly tied the pause to cash flow rather than conviction.

Thesis Scorecard Post-Earnings

Graded against the standing thesis pillars carried forward from our Q1 2026 report and its signpost list, not against a fresh set invented this quarter.

Overall: Thesis weakened. Of the five bull pillars carried into this quarter, one was confirmed, one is neutral, and three were challenged. One bear point moved to materializing, one was partially confirmed, and one remains unsupported. The offsetting positives (the SEC resolution and a Q3 guide built on already-shipped product) are real but do not restore the two pillars that carried the Outperform: reliable model cadence and a dated general-availability inflection.

Action: Downgrade to Hold. Existing holders should not sell into a 19.7% gap at a fresh 52-week low with a reacceleration already guided and banked; the asymmetry no longer favors adding, but it does not favor capitulating either. New money should wait for the Q3 print rather than anticipate it, since the guide is known, already sits below where consensus was, and the upside tail depends on model releases management has explicitly excluded from the outlook. We revisit toward Outperform on two consecutive quarters of restored double-digit sequential growth, or on the first hard advertiser-count disclosure for the post-launch cohort.