Initiating coverage · Mobile advertising technology

APP

AppLovin Corporation · Nasdaq

Revenue grew 53% year over year. The shares are down 54% year to date. Held against a constant earnings estimate those two facts reconcile almost exactly: the whole drawdown is multiple compression and none of it is earnings. The market has stopped paying for the growth algorithm and started pricing the terminal question instead. Our band is $300–375 against $311.98.

What we foundThree headline conclusions
01

All of the fall is the multiple, and none of it is the earnings. Held against a constant $6.8bn estimate of this year’s adjusted EBITDA, the enterprise value has gone from 33.3x to 15.5x — a 53.5% de-rating against a 53.7% fall in the share price. Over the same period the estimate itself went up. The stock has not broken; the willingness to pay for it has.

02

The vertical the bull case rests on is not a reported line. Management discloses consumer advertiser spend — $1.28bn in the second quarter — and not consumer revenue. On a take rate near 10% that is plausibly under 8% of the group. When a company reports gross flow for its growth segment and net revenue for its mature one, the framing is a choice, and it is the largest information asymmetry in the name.

03

The price already pays for the base case, which is why this is not a mispricing. At $311.98 the market’s enterprise value of $105.2bn sits within 1% of our base-case enterprise value of $104.5bn. It is not pricing the bear case and it is not pricing management’s own 30% algorithm. The drawdown has done the work of pricing the risk, and there is no free lunch left in it.

ContentsEleven sections
01 · The central factThe de-rating, drawn to scale

Seventeen turns of multiple went missing, and the earnings never did

Start with the only arithmetic that matters on this name. AppLovin began 2026 at $673.78 and closed the last completed session at $311.98, a fall of 53.7%. Hold this year’s adjusted EBITDA constant at our $6.8bn estimate and the enterprise value went from 33.3x to 15.5x across the same period — a 53.5% de-rating. Those two numbers are the same number. Nothing was left over for the business to have caused.

And the estimate did not fall. Consensus adjusted EBITDA for 2026 rose over the period, which makes the comparison above conservative rather than generous: measured against the lower estimates that prevailed in January, the starting multiple was higher still and the de-rating larger. We use today’s figure on both dates so the two ends of the bar are like for like.

15.5x EV / FY2026E adjusted EBITDA today
17.8x of multiple removed since January
33.3x where it started the year, on the same estimate

That is the whole story of the year, and it is why this report spends most of its length on duration rather than on the quarter. A 15x multiple on a business growing 47% is only cheap if the growth persists for several years. The market is not making a mistake about arithmetic. It is making a judgement about how many years of it there are.

Where the shares sit

MeasureValueNote
Reference price, 17 Aug 2026 close$311.98The last completed session. Every multiple on this page scales with it
Market capitalisation~$104.5bnOn 335m shares outstanding at the second quarter, down about 1% sequentially
Enterprise value~$105.2bnAdding $3.7bn of total debt, less $3.05bn of cash. Net debt is $0.65bn
EV / FY2026E adjusted EBITDA15.5xOn our ~$6.8bn estimate. 18.4x on the trailing ~$5.7bn
EV / FY2026E revenue13.0x15.4x on trailing revenue of $6.83bn
Price / trailing earnings23.7xOn $4.41bn of trailing net income. Flattered by an effective tax rate near 9% — see Section 07
Free cash flow yield~4.9%On management’s ~75%-of-EBITDA conversion target for the year, not on delivered cash flow
Net leverage~0.1xAgainst a stated long-run comfort level near 1.0x. There is no solvency question here
52-week range$303.17–745.6158% below the high and 3% above the low. The stock is making lows, not basing
Year to date−53.7%From $673.78 at the 2025 close
Implied / realised volatility~54% / ~84%Implied sits well below what the stock has actually delivered. Position sizing matters more than entry price

Derived inputs. Enterprise value, net debt and every multiple in this table are ours, computed from the Q2 2026 release, the 30 June 2026 balance sheet and the 17 August closing price. Trailing revenue, adjusted EBITDA and net income include a Q3 2025 quarter that is backed out of reported annual and quarterly totals rather than separately reported, and is therefore an estimate. FY2026E revenue of ~$8.1bn assumes the fourth quarter grows about 8% sequentially, the conservative end of the seasonal path.

02 · The valuation, openAdjust any input

Underwrite it yourself

Assumptions

Result

$304.3
Modelled DCF value per shareAgainst $311.98 last
PV of 5yr FCF
PV of terminal
Enterprise value
Net debt
Equity value

Terminal value is 74% of enterprise value, and net debt is almost nothing. The duration assumption is the entire risk in this model.

This is an unlevered discounted cash flow on the enterprise, with net debt of $0.65bn deducted to reach the equity. The base year is our FY2026 revenue estimate of $8.1bn. Growth is expressed as a five-year compound rate to 2031, and cash tax is held at 18% throughout.

Move the growth rate first. On a business with almost no debt, no capital intensity and an 80%-plus cash margin, duration is doing nearly all of the work: the terminal value is half the enterprise value in the base case and 60% in the bull. Whether 2027 grows 12% or 32% is worth more than every margin and discount-rate assumption on this page combined.

Reading the inputs. Two departures from the standard form are worth stating. First, the margin slider is an EBITDA margin struck after stock-based compensation. AppLovin’s headline 84% adjusted EBITDA margin is defined to exclude SBC, which we estimate at 8–9% of revenue; we treat it as the cash cost it economically is, so a 70% setting here corresponds to roughly a 76% margin as the company would report it. This single choice is the largest methodological difference between this model and sell-side targets built off adjusted EBITDA multiples. Second, terminal growth is set at 3.25% against the 3.0% used in the full-path model below, because in a five-year form the perpetuity has to absorb both the 2032–36 fade and the terminal state. At those settings the widget prints $304 against the full path’s $310 — agreement within 2%. Where the two disagree, the full-path table in Section 03 governs.

What this number is

The number above is not a fact about AppLovin. It is what our assumptions produce, and on this name the most consequential assumption is the growth slider, because the thing driving growth is a research pipeline that nobody — including management, by their own admission — can schedule. That is why the growth input is exposed and why we say to move it first. Set it to what you believe about 2028 rather than what you observed in 2025, and use the number that falls out.

03 · Scenarios and the bandSelect to load into the model

Three ways this resolves, and the band that comes out

Same model, different inputs. Click any card to load it. The spread is wide — the bull is four and a half times the bear — and on a business whose growth rate is set by research outcomes rather than by a sales pipeline, that is the honest shape of the distribution rather than a failure of analysis.

Weighting those 25 / 50 / 25 gives $333, which is where the $335 central estimate comes from. That weighted case cannot be loaded as a preset — the model supports three — so it is stated here instead. Note that it sits above the base case: the bull adds more than the bear subtracts, because a business with no debt and no capital intensity has a floor under it that a levered one does not.

The assumptions, side by side

AssumptionBearBaseBull
2027 revenue growth+12%+24%+32%
Growth path to 2036fades to 2% by 203019 → 15 → 12 → 10, fading to 4%28 → 24 → 20, fading to 5%
2036 revenue$11.9bn$22.6bn$34.9bn
Adj. EBITDA margin78% → 66%82.5% → 76%83.5% → 80%
SBC, deducted as a cost$0.85 → 1.12bn$0.80 → 1.40bn$0.85 → 1.83bn
Capex, % of revenue~4.5%4.0–4.5%~4.3%
Cash tax rate18%18%18%
Working capital6% of revenue growth6% of revenue growth6% of revenue growth
Discount rate12%11%10%
Terminal growth2.0%3.0%3.5%
Enterprise value$43.9bn$104.5bn$196.3bn
Value per share$129$310$584
Terminal value as % of total37%50%60%

Why each assumption is where it is

2027 growth is the whole argument. The company exits 2026 growing 46–48%. Management’s stated long-run algorithm is roughly 30% a year. The Street sat at 30.9% before the second quarter; Benchmark has since cut to 21%, and BofA downgraded on the view that risks to the 30% trajectory have increased. Our 24% splits that difference and deliberately refuses management’s own number, because the quarter just gone demonstrated that the algorithm’s input — model releases — is not schedulable.

The bear margin fade to 66% is not a competitive-collapse assumption. It is what happens if mix shifts toward consumer — where AppLovin must eventually build a sales force it does not currently have — while compute intensity keeps rising. The base fade to 76% is gentler, and still below the company’s own “low 80s” long-run frame, because we deduct SBC separately and because ten years is a long time to hold 80%-plus margins in any advertising market.

Cash tax at 18% against a current effective rate near 9%. That gap alone is worth roughly a tenth of the value. Modelling 18% rather than the statutory 21% acknowledges genuine structural benefits without extrapolating a rate that depends on stock-compensation windfalls and a rising share price — the second of which has stopped happening.

An 11% base discount rate looks punitive on a credit view and generous on a volatility view. With 0.1x net leverage the balance sheet argues for less. A beta near 2.5, 84% realised volatility, single-product concentration and an unforecastable growth driver argue for the top of the range. The backdrop supports it too: the US 10-year sits at 4.71% and the 30-year at 5.29%, which is an unkind environment for long-duration cash flows and means a 2021-style 8% rate is simply not available. The sensitivity table below shows exactly what the choice is worth.

Sensitivity — discount rate against terminal growth

Base-case cash flows throughout, on the full ten-year path. Across the plausible core of this grid — 10% to 12% and 2.5% to 3.5% — the base case spans $266 to $374, which is very close to the published band. That is not a coincidence; it is where the band comes from.

WACC ↓ / Terminal g →2.0%2.5%3.0%3.5%4.0%
9%$383$401$422$446$475
10%$331$344$358$374$393
11%$291$300$310$321$334
12%$259$266$273$281$290
13%$233$238$244$250$257

One hundred basis points on the discount rate is worth roughly $50 a share here, or about 16%. That is far less violent than on a levered balance sheet, and it is why the growth slider rather than the rate slider is the one we tell you to move first.

The reverse test — what $311.98 is actually assuming

More useful than arguing about what the shares are worth is inverting the model, because the answer then depends on no view of the company at all. It states only what has to be true for a buyer at this price to earn a normal return.

The market is paying for the base case, almost exactly

At $311.98 the enterprise value is $105.2bn. Our base-case enterprise value is $104.5bn. The gap is under 1%, which means the price embeds roughly 24% growth in 2027, margins drifting from 83% toward 76%, an 11% discount rate and 3% terminal growth. It is not pricing the bear case, which would be $129. It is not pricing management’s 30% algorithm, which sits closer to the bull. That is a reasonable place for a stock to sit after a 54% drawdown, and it also means there is no free lunch here: to make money from this price you need growth to hold above roughly 25%, and you are no longer being paid to be wrong about it.

Where the peer tape actually points

The most-quoted bear argument this year is a read-across: the independent buy side is being repriced while the sell side re-rates, because walled gardens are absorbing performance budgets and supply owners get scarcer. The table below is every relevant listed comparable at the same close. It supports that story less cleanly than it is usually told.

CompanyWhere it sits17 Aug closeYear to date
AppLovin (APP)Demand side, mobile app$311.98−53.7%
The Trade Desk (TTD)Demand side, open web and CTV$13.40−64.7%
Unity (U)Demand side, mobile game UA$45.45+2.8%
Magnite (MGNI)Supply side$24.19+49.0%
PubMatic (PUBM)Supply side$17.27+94.7%
Meta (META)Walled garden$568.97−13.7%

Unity is the problem for the clean version of the argument. It is a demand-side platform competing directly with AppLovin for mobile-game user acquisition, and it is up on the year while AppLovin and The Trade Desk are down 54% and 65%. If walled gardens were structurally absorbing demand-side budgets, Unity should be on the losing side of that line, and it is not. What separates the three is not which side of the auction they sit on but what multiple they carried into the year: AppLovin and The Trade Desk both entered 2026 on premium growth multiples and both lost them, while Unity — growing around 24% off a much smaller base with GAAP losses — had no premium to lose. Read that way the divergence is a repricing of expensive growth, not a verdict on where value accrues in performance advertising. That is a materially weaker bear argument than the one usually made, and we say so even though it cuts against the more dramatic story.

This table carries closing prices and year-to-date moves only. We are not publishing peer market capitalisations, growth rates, margins or multiples because we cannot source them to primary company reporting, and a directional approximation in a valuation table is worse than an absent one. Year-to-date figures are computed from each name’s 2025 closing price to the 17 August 2026 close and exclude dividends. Meta is included as the competitive reference point in Section 06, not as a valuation comparable.

Where the band comes from

MethodAssumptionPer share
DCF, full pathBase case at an 11% WACC and 3% terminal growth$310
Same at 10%$358
Same at 12%$273
DCF, constant-rate formThe Section 02 widget on its base preset$304
Probability-weighted25 / 50 / 25 across bear, base and bull$333
Bear caseGrowth fades to 2% by 2030, margin to 66%, 12% WACC$129
Bull caseConsumer scales, margins hold at 80%, 10% WACC$584
StreetConsensus target~$559
Fair value: $300 – $375 per share, central estimate $335

At $311.98 the shares sit inside that band, in its lower half, about 7% below the central estimate. The band is the model run across a reasonable input range, not a separate opinion, and it deliberately sits far below the $559 consensus target because that target is built off adjusted EBITDA multiples and ours deducts stock compensation as a real cost. That single choice accounts for most of the gap. The easy money from the de-rating has been made, and this is not yet a distressed price. We rate the shares fairly valued.

04 · The quarter, and the callQ2 2026, reported 5 August 2026

A $16m shortfall removed roughly $40bn of market capitalisation

AppLovin missed the midpoint of its own revenue guidance by roughly 30 basis points and its EBITDA guidance midpoint by roughly 100. Per Piper Sandler that is the first such miss since the company went public. In absolute terms the revenue shortfall was about $16m on a $1.92bn quarter, less than one percent, and the shares fell 16% after hours. The size of the reaction relative to the size of the miss is the fact worth sitting with: what broke was not the quarter but a three-year assumption that the beats would not stop.

Q2 2026ResultQ/QComment
Revenue$1,924m+4%Up 53% year over year. +4% sequential against +11% the prior quarter is the whole story
Adjusted EBITDA$1,614m+3%Below the guided range for the first time
Adj. EBITDA margin84%−100bpOff the record 85%. Compute is the stated reason
Sequential flow-through70%Incremental margin still extraordinary in absolute terms
Net income$1,267m+5%A 65.9% net margin. Flattered by a very low tax rate — see Section 07
Diluted EPS$3.76+6%Against $3.56 in Q1 and $2.39 a year ago
Free cash flow$863m−33%The one number that genuinely deteriorated. 53% of adjusted EBITDA against a ~75% full-year target
Consumer advertiser spend$1,280m+28% vs Q4’25Gross advertiser spend, not AppLovin revenue. Not disclosed as a revenue line
Cash / total debt$3.05 / $3.7bn0.1x net leverage. No balance-sheet risk at any reasonable stress level
Buyback and withholding$551m−45%Slowed deliberately on lower cash flow; $1.8bn of authorisation remains
Shares outstanding335m−1%Genuine net shrinkage, not just dilution offset

The quarterly path, and what changed in it

$m unless statedQ2’25Q3’25Q4’25Q1’26Q2’26Q3’26E
Revenue1,2591,405*1,6581,8421,9242,055–2,085
Growth, year over year+66%+59%+53%+46–48%
Growth, sequential+12%*+18%+11%+4%+7–8%
Adjusted EBITDA~1,021*~1,141*1,4001,5601,6141,710–1,740
Adj. EBITDA margin81%81%*84%85%84%~83%
Net income773~835*1,1021,2061,267

Starred figures are derived rather than reported. Q3 2025 revenue is backed out of the FY2025 total of $5,481m less the reported quarters, and Q3 2025 adjusted EBITDA out of the FY2025 figure of $4.5bn. Treat them as estimates. Full year 2025: revenue $5,481m up 70%, adjusted EBITDA $4.5bn up 87%, free cash flow $4.0bn up 91%, $2.2bn of stock repurchased. First half 2026: revenue $3.77bn up 56%, net income $2.47bn, adjusted EBITDA $3.17bn up 62%.

The explanation management gave, and why it is the right kind

Not demand, not competition — model release timing. Foroughi said the pace of meaningful model improvement was lighter than normal during the quarter and that the next step up landed just after quarter end. What matters is that this is a falsifiable claim supported by the evidence that would exist if it were true. MAX publisher earnings grew double digits sequentially and share of publisher waterfalls held — both of which would deteriorate if the business were losing share rather than waiting on a release. That is the right evidence to cite, and it largely would not be available to a company inventing an excuse.

The most useful sentence on the call

Third-quarter guidance is $2.055–2.085bn of revenue, up 46–48% year over year, with adjusted EBITDA of $1.71–1.74bn at roughly an 83% margin. Critically, Stumpf stated that the guide does not assume additional model releases that have not yet been deployed. That converts the guide from a hope into a floor built on shipped work, and it is why a third-quarter miss would be far more damaging than the second-quarter one: there would be nothing left to attribute it to.

Compute is now a real line item. The sequential cost increase was driven by higher training and inference spend, and the CFO offered a framework of roughly ten cents of compute per incremental dollar of revenue, with third-quarter margin guided down a point to about 83%. That number is checkable against the 10-Q, which is the reason it is worth having.

And the SEC inquiry closed. Stumpf disclosed that the request had been voluntary, was never deemed material, and that the SEC has advised the company it concluded with no recommended action. That removed the single largest tail risk in the name — the possibility of a forced change to the data practices that produce the margins — and the stock fell 16% anyway. That tells you what the marginal buyer is actually worried about, and it is not regulators.

What analysts pressed on, and what they did not get

Wells Fargo pressed on cost-per-install inflation and waning returns across mobile game publishers, citing third-party data showing downloads off 10–15% year on year. BofA pressed hardest on self-serve onboarding metrics and received no numbers, only that a few advertisers contribute more today. Loop Capital pushed on funnel leakage and the creative bottleneck. Multiple analysts asked whether compute spend is now structurally higher. Nobody got a consumer revenue figure.

The positive surprises were the SEC resolution, MAX publisher earnings up double digits sequentially, consumer spend 28% above the seasonal peak in a seasonally weak quarter, a falling share count, and a third-quarter guide built only on already-shipped work. The negative surprises were free cash flow conversion at 53%, the first guidance miss since the IPO, creative video generation still unsolved six months after it was first flagged, and Wurl and CTV explicitly deferred for lack of budget capacity. Lead generation remains in testing with nothing to report.

05 · The businessOne machine, two auctions

A recommendation engine sitting on top of an auction it also owns

AppLovin is now a pure-play advertising platform. It sold its mobile games business to Tripledot Studios, completing in June 2025 for a $106m pre-tax gain, and what remains is a single machine: AXON, a deep-learning recommendation engine that decides which ad to show which user at which price, sitting on top of MAX, the in-app bidding and mediation layer AppLovin acquired and now runs across a large share of mobile publisher ad waterfalls.

The economics are unusually clean. Advertisers — overwhelmingly mobile game developers — set a target return on ad spend. AXON buys impressions to hit it. When the model gets better the advertiser’s return improves, so the advertiser raises budget, and AppLovin’s revenue rises with no incremental sales effort. That is the entire growth algorithm, and it is why roughly a thousand employees generate eight billion dollars of revenue. It is also why a quarter without a model release produces 4% sequential growth instead of 11%.

Where the revenue comes from

StreamWhat it isDisclosed scale
Gaming (AppDiscovery)Performance user acquisition for mobile game developers, priced on return on ad spendStated as “the majority” of revenue. No precise split disclosed
Consumer / ecommerceWeb and app advertisers — direct-to-consumer brands, retailers — buying the same auction$1.28bn of advertiser spend in Q2’26, 28% above the Q4’25 peak. Revenue not disclosed
MAX mediationPublisher-side auction. AppLovin takes a fee even when a rival wins the impressionPublisher earnings grew double digits sequentially in Q2’26
Adjust / WurlAttribution analytics, and connected-TV distributionImmaterial today. Wurl explicitly deferred — see Section 09
Lead generationNew format, in customer testingDescribed as still work in progress. No revenue

The United States is roughly 52% of net sales, and the consumer vertical is more Western-concentrated than gaming. Note what the table above cannot tell you. The two largest streams are reported on different bases — net revenue for the mature one, gross flow for the growth one — and there is no line in any filing that sizes consumer as revenue. Every long thesis on this stock leans on that vertical, and nobody outside the company can measure it. It is the single most important thing about this business and it is Section 07.

What is meant to drive it from here

Five things, in the order management put them. Model improvement in gaming, still the dominant near-term driver and the one that failed to show up in the second quarter. Consumer scaling, mid-market first and long tail later. The AppLovin Ads Manager self-serve platform, opened publicly in late June 2026 after fourteen years as a closed system. Partnerships with analytics and attribution vendors — Triple Whale was named — as a targeted customer acquisition channel. And supply expansion into non-gaming apps, then the open web, then connected TV through Wurl, which is explicitly deferred until demand outruns inventory. That last point is worth reading twice: management is telling you the binding constraint today is demand, not supply.

The risk that dominates all the others

It is not competitive and it is not regulatory. It is that revenue growth is a function of a research pipeline with no external visibility. The second quarter demonstrated the mechanism precisely: one quarter without a meaningful model release took sequential growth from the 11–18% investors had been trained to expect down to 4%, and produced the first guidance-midpoint miss since the IPO. Nobody outside the building can forecast when the next lift lands, and by their own admission, nobody inside it can either. Secondary risks follow in Section 07.

06 · The moat, scoredAverage 6.5 / 10 — wide in gaming, absent elsewhere

The advantage is a compounding loop, and it does not travel

DimensionScoreAssessment
Network effects8/10Genuinely two-sided. More publishers on MAX means more inventory; more advertisers means higher clearing prices; more of both means more training data. The strongest single leg by a distance
Cost advantage8/10Roughly 1,000 employees producing about $8bn of revenue — close to the highest revenue per head in public software. It is also why the incremental margin is 70%
Scale8/10The largest mobile-game user-acquisition channel in the world by management’s account, with well over $11bn of annual advertiser spend across the network
Data7/10Deep and near-total in gaming, and explicitly thin in commerce — management’s own words are that data penetration in that market is a lot lower. The moat is vertical-specific and does not travel automatically
Distribution7/10MAX’s position inside publisher waterfalls is the structural asset. AppLovin earns a fee even when a competitor wins the auction. This is exactly what CloudX is attacking
Customer loyalty7/10Near-universal in gaming, where management argues no serious mobile advertiser can afford not to be on the platform. Entirely unproven in consumer
Intellectual property6/10AXON is real engineering, but it is architecture plus data plus talent, not patents. The CloudX founders built the two products AppLovin bought. Talent walks
Switching costs5/10Asymmetric. Publishers integrate an SDK and are sticky. Advertisers reallocate budget in an afternoon — loyalty is bought quarter by quarter with performance
Regulation5/10Neutral to slightly negative. The SEC has cleared the company, but platform terms of service and privacy regimes remain a live constraint on the signal supply that feeds the model
Brand4/10The weakest leg, and management says so — it concedes an awareness problem among ecommerce advertisers who do not already use the platform. They tried renaming the company to AXON and reverted

Who is actually competing

CompetitorOverlapThreat assessment
MetaDirect, on both game and commerce advertisersThe real one. Roughly $250bn of revenue against AppLovin’s $8bn, and it can subsidise indefinitely if it chooses to. The counter-argument is opportunity cost: the mobile-game user-acquisition niche is small relative to what else Meta could do with the same engineers
GoogleDirect on Android via Play, plus AdMobStructurally advantaged on its own distribution surface, and management concedes it — Foroughi allowed on the call that Android is the more competitive platform because of a very big competitor doing well there. Google also shipped Project Genie, an AI game-creation platform whose second-order effect on the mobile game ecosystem is unknown and potentially structural
Unity (Vector)Direct in mobile game user acquisitionHas runtime engine-level signals AppLovin does not natively hold. Foroughi frames it as non-zero-sum. Growing around 24% off a much smaller base with GAAP losses
CloudXAI-native supply-side platform for mobile publishersFounded by Jim Payne and Dan Sack — the people who built MoPub and MAX, the two assets AppLovin bought. $30m Series A led by Addition in November 2025, generally available on 4 February 2026, and it knocked 16% off the shares in a day. Attacks the mediation layer rather than the model
The Trade DeskAdjacent — open web and connected TVNot a competitive threat today. Relevant as a cautionary read-across on what a demand-side de-rating looks like, and as the counter-example to it in Section 03
Average 6.5 / 10 — and pointed in an uncomfortable direction

The strongest advantage is the compounding loop between MAX distribution, advertiser data and model quality. Nobody else in mobile owns both the auction and the buying engine at this scale with this cost base, and that combination is why 84% margins have persisted through three years of people predicting they would not. The biggest threat is not CloudX and probably not Unity. It is Meta. A $30m Series A attacking the mediation layer is an irritation; a company with $250bn of revenue deciding untracked iOS game inventory is worth pursuing is a different order of problem, and the opportunity-cost argument that makes it irrational today gets weaker every year AppLovin’s consumer vertical grows into Meta’s actual core market.

Direction of travel: expanding in absolute terms, narrowing in relative terms. The gaming moat is as wide as it has ever been. But the company is voluntarily walking into web commerce advertising, where it has admitted its data is thin, its model is less sophisticated, its brand is unknown and its creative tooling does not work yet. Every one of those is a moat the incumbents already have. AppLovin is trading a category it dominates for optionality on a category it does not. That is very probably the right strategic call, and it is also a large part of why the multiple has halved.

07 · Red flagsOverall 5 / 10 — moderate

Not a fraud profile. An earnings-quality and disclosure profile

We ran a deliberately hostile read of the release, the balance sheet and the insider filings. Severity is scored one to ten, where one is immaterial and ten is severe. This is not a fraud pattern and we are not implying one — the reported numbers are large, audited, cash-backed and corroborated by a balance sheet that has swung from leveraged to roughly net cash neutral. What the flags describe instead is earnings quality, where adjusted metrics exclude a real cost and a tax rate will not persist, and disclosure opacity, where the growth vertical everyone is underwriting is not a reported line. The score would have been 7 before the SEC closed its inquiry.

ConcernSev.Evidence, and why it matters
Consumer is not disclosed as revenue7/10 Management reports consumer advertiser spend of $1.28bn rather than consumer revenue. If the take rate is near 10%, consumer is plausibly $130–150m a quarter, or under 8% of the group. The entire long-duration bull case is unmeasurable from outside. When a company reports gross flow for its growth segment and net revenue for its mature one, assume the framing is deliberate
Growth depends on an unforecastable pipeline7/10 The second quarter is the proof rather than the theory. Foroughi’s own description is that there is no guarantee of lifts in every three-month period and that the work is hypothesis testing. That is an honest description of research and a terrible description of a forecastable revenue line. A business whose sequential growth swings from 18% to 4% on internal test outcomes cannot be valued on a smooth curve
Effective tax rate looks unsustainably low6/10 Derived, not disclosed. Q2 adjusted EBITDA of $1,614m converting to $1,267m of net income leaves roughly $350m for stock compensation, D&A, net interest and tax combined. Backing out plausible values for the first three implies an effective rate near 9%. Normalising to 21% would cut EPS by roughly 13%. Verify against the 10-Q before relying on the headline multiple
The 84% margin excludes stock compensation5/10 Adjusted EBITDA is defined to exclude SBC, which on our estimate runs 8–9% of revenue. The 84% figure is a real operating statistic but not an economic margin. Our DCF deducts SBC as a cash-equivalent cost, which is the main reason our fair value sits well below sell-side targets built off adjusted EBITDA multiples
Free cash flow diverged from net income5/10 Q2 free cash flow of $863m against $1,267m of net income, a 0.68x conversion. Management attributes it to the timing of international cash tax and interest payments and, to their credit, pre-flagged it on the Q1 call. Q1 conversion was 1.07x, which supports the timing explanation. Downgrade to 2/10 if Q3 recovers; upgrade to 7/10 if it does not
Insider selling is one-directional5/10 Zero insider buys and 55 sells over the trailing twelve months, roughly $900m in value. Foroughi sold 477,131 shares over the period, including 52,165 in June 2026 at $472–517. Mitigants: he retains about 2.35m shares directly plus material trust holdings, sales appear to run through scheduled plans, and founders of a stock that rose 700% then 100% selling some of it is not by itself a signal. But there has been no insider buying into a 54% drawdown, which is the datapoint a bull would most want to see
Customer and vertical concentration4/10 Gaming remains “the majority” of revenue by management’s own description, in a category where third-party data suggests downloads are down 10–15% year on year. Foroughi disputes the measurement, arguing analytics providers miss off-platform in-app purchasing. He may well be right. He is also the interested party
Buyback and tax withholding reported together3/10 The disclosure blends open-market repurchases with shares withheld to settle restricted-stock tax — about 1.14 million shares for $551m. These are economically different. Net share count did fall to 335m so the reduction is real, but the presentation makes the buyback look larger than it is
Unadjudicated short-seller allegations3/10 Fuzzy Panda, Culper Research and Muddy Waters alleged in 2025 that AXON extracted user identifiers in breach of platform terms of service. The SEC has now closed its inquiry with no action, which is a genuine and material clearing event. Separately, CapitalWatch retracted and apologised for claims about a shareholder in February 2026. The underlying platform-retaliation risk was never formally adjudicated, but the regulatory tail is materially thinner than it was
Leverage1/10 $3.7bn of debt against $3.05bn of cash and roughly $5.7bn of trailing adjusted EBITDA. Listed for completeness. Not a concern at any reasonable stress level
Overall 5 / 10 — moderate, and the composition matters more than the score

Two of these are load-bearing and the rest are housekeeping. The consumer disclosure gap and the unforecastable pipeline are the same risk seen from two angles: you cannot size the segment that is supposed to extend the duration, and you cannot forecast the driver of the segment that currently produces the cash. Everything else on this list — the tax rate, the SBC treatment, the buyback presentation — is knowable from the filings and already inside our numbers. If you own this, the two things to watch are the Q3 cash conversion and whether consumer ever becomes a reported line.

08 · Management qualityAverage 3.5 / 5

Exceptional operators who have chosen to tell you as little as possible

AreaScoreAssessment
CEO track recordAdam Foroughi5/5Turned a mobile game studio into an $8bn-revenue advertising platform with 84% adjusted EBITDA margins and roughly 1,000 employees. Returns above 700% in 2024 and above 100% in 2025. Whatever happens next, the record is exceptional
Capital allocation4/5Sold the Apps business to Tripledot in 2025 at what now looks like an excellent moment, repurchased $2.2bn of stock in 2025 and $1.55bn in the first half of 2026, and runs the balance sheet at 0.1x net leverage against a 1.0x stated tolerance. Deliberately slowed second-quarter buybacks on lower cash flow rather than stretching — disciplined, if arguably the wrong call at $400
Communication style4/5Opened the worst call in the company’s public history by conceding the miss in the first thirty seconds without reframing it, and without over-apologising either. Answers analyst questions in plain language and volunteers unflattering detail — the creative bottleneck, the branding problem, Google’s Android advantage
CFO credibilityMatt Stumpf4/5Pre-flagged the Q2 cash conversion issue on the Q1 call, gave a specific compute framework that can be checked against the 10-Q, and closed the SEC question directly rather than leaving it hanging
Acquisitions4/5MoPub, MAX and Adjust were all integrated into the core product and are visibly load-bearing. Wurl is the exception — acquired and effectively parked. Divestiture execution has been better than acquisition execution
Insider ownership3/5Foroughi retains roughly 2.35m shares directly plus substantial trust holdings, so alignment is real. But 55 insider sells and zero buys over twelve months, with selling continuing at $472–517 in June, is a one-way pattern
Guidance accuracy3/5Beat its own guidance midpoint every quarter from the IPO until Q2 2026. That is an outstanding record and it just ended. Scored on the break, not the streak — the streak is precisely why expectations were where they were
Dilution3/5SBC is material at an estimated 8–9% of revenue and sits outside the headline margin. Net share count is nonetheless falling, at 335m, so shareholders are ahead on the arithmetic even where the presentation flatters
Compensation3/5Modest cash base with the value concentrated in equity, which is structurally correct. The absolute size of historical grants is large enough that the SBC line is a real drag on economic returns
Board quality3/5Adequate but not a differentiator. Worth verifying independently: call participant titles suggest Foroughi may no longer be listed as Chairperson as he was in 2025. If accurate that is a governance improvement. Treat it as an observation to check against the proxy, not a finding
Transparency2/5The weak spot, and it is deliberate. No consumer revenue disclosure. Gross advertiser spend for the growth vertical, net revenue for the mature one. No advertiser counts for the self-serve launch. No quantification of model uplift, ever. The AXON black box extends to the income statement
Do they act like long-term owners? Operationally yes, informationally no

Operationally it is not a close call. They shed a good business to focus on a better one, they run the leanest cost base in large-cap software, they buy back stock consistently, they carry almost no net debt, and they spend on compute only where it demonstrably returns. When they missed, they said so plainly in the first thirty seconds. Informationally the answer is no, and it is now expensive. This is a team that has systematically given investors the minimum disclosure the growth story requires and not one line more. That was costless while the stock compounded. With no segment data, investors cannot distinguish “consumer is scaling and gaming had a timing blip” from “gaming is maturing and consumer is too small to matter” — and faced with that ambiguity the market marked the stock down 54%. The disclosure choice is a direct contributor to the de-rating, and it is entirely within management’s control to fix.

09 · What moves it nextTwo of these are thesis-deciding

The calendar, and what each item actually tests

CatalystTimingWhat it tests
Q3 2026 resultsEarly Nov 2026The whole thesis in one print. The guide of $2.055–2.085bn is built only on already-shipped model work, so a miss would be very damaging — there would be nothing left to attribute it to. A beat plus reacceleration confirms Q2 was timing and the multiple re-rates hard from a depressed base. A second consecutive miss converts the timing narrative into a maturity narrative and opens the bear case. Confidence high; date approximate, based on reporting cadence
Q3 free cash flow conversionEarly Nov 2026Management guided conversion to improve in Q3 and normalise near 75% of adjusted EBITDA for the year. Q2 was 53%. A second weak quarter makes the timing explanation untenable and puts earnings quality in play. Confidence high — this is explicit guidance from the Q2 call
Holiday ecommerce seasonNov–Dec 2026The seasonal peak for consumer advertisers, and the first one with Ads Manager generally available. Consumer spend ran 28% above the Q4’25 peak in a seasonally weak Q2. A large Q4 print makes the consumer runway credible without needing a segment disclosure. Confidence medium-high
Q4 and FY2027 guidanceFeb 2027The first formal framing of 2027 against the “roughly 30% annually” algorithm, and where the Benchmark 21% against Street 30.9% argument gets settled. This is the input our base case is most sensitive to. Confidence high
Buyback execution or a new authorisationWatch Oct 2026$1.8bn remains; the board added $3.2bn in October 2025. A large increase at these prices is the clearest capital-allocation signal available, and management has earned credibility here. Confidence medium
Consumer revenue disclosureUnscheduledThe single largest information asymmetry in the name. If AppLovin ever breaks consumer out as a revenue line the stock re-rates — direction unknown, which is exactly why it matters. Low confidence that it happens; high confidence that it would move the shares if it did
Creative and video generation releaseH1 2027?Foroughi named this the biggest hurdle in the system: interactive end cards auto-generate well, 30–60 second video does not. Solving it is the gate to the long tail. Low confidence on timing — the same bottleneck was flagged six months ago
Supply expansion — non-gaming, open web, CTV2027+Explicitly sequenced and explicitly deferred, on the stated grounds that the company is budgetarily constrained and expanding supply would not expand revenue. A genuine future lever, and a signal that current demand is the binding constraint. Confidence medium
Competitive moves — Meta iOS, Google Play, CloudXAny timeCloudX has signalled demand-side expansion. Meta’s untracked-iOS push is the one that would genuinely change the model. Historically these headlines have moved the shares 13–16% in a day on announcement alone. Low confidence on timing, high on magnitude
Rates and high-multiple positioningOngoingThe 10-year at 4.71% with hike expectations fading. A beta near 2.5 means the shares amplify every move in growth-equity risk appetite. High confidence on the mechanism, none on direction
Residual securities litigationOngoingClass-action investigations were opened following the 2025 short reports. The SEC’s no-action conclusion substantially weakens them. Confidence medium
A positioning note, because it cuts against the obvious instinct

Implied volatility on the shares is roughly 54% annualised while 30-day realised volatility is roughly 84%. Implied is trading well below what the stock has actually delivered, which is the wrong way round for anyone drawn to selling premium into a beaten-down name. You would be paid less than the recent distribution of outcomes, on a stock that has printed 16–20% single-day moves on both earnings and competitor headlines, with the next catalyst roughly eleven weeks out and the price making 52-week lows rather than basing. Options here are not obviously cheap or expensive in absolute terms — but the implied-to-realised relationship does not favour a short-premium structure.

10 · Investment committeeBoth sides, argued properly

The debate

BULLYou are looking at a business growing revenue 53% with an 84% EBITDA margin, trading at 15.5x forward EBITDA on a 4.9% free cash flow yield, and calling it risky. Take the quarter apart. Revenue missed by $16m — less than one percent — and the stock lost roughly $40bn. The explanation was specific and testable: model releases are lumpy, this one landed just after quarter end, and Q3 is guided at 46–48% growth built only on work already shipped. Management deliberately excluded unreleased models from the guide. That is a floor, not a hope.

BEAREvery number you just cited is backward-looking, and the one forward-looking disclosure is missing on purpose. Where is consumer revenue? You quoted $1.28bn of advertiser spend. On a plausible take rate that is maybe $130m of revenue, under 8% of the group. The entire duration argument — the thing that turns 15x into cheap rather than fair — rests on a segment the company will not size.

BULLThen look at what did not deteriorate. MAX publisher earnings grew double digits sequentially. Waterfall share held. Consumer advertiser spend hit $1.28bn, 28% above the Q4 seasonal peak, in the weakest quarter of the year. And the SEC — the one genuine tail risk, the thing that could have forced a change to the data model that produces the margins — closed its inquiry with no action. The market ignored it.

BEARThe inquiry closing removes a fine. It does not remove the underlying question. The short sellers alleged terms-of-service violations against Meta, Google, Apple, TikTok and Snap. A securities regulator declining to act says nothing about whether those platforms can restrict the signal supply. That risk sits with the platforms, not the SEC, and it was never adjudicated.

BULLBalance sheet: net leverage of 0.1x. $1.8bn of buyback authorisation left. Share count falling. A thousand people producing eight billion dollars of revenue. And the de-rating from 33x to 15.5x happened while estimates went up. That is sentiment, not fundamentals.

BEAROr it is the market discounting forward rather than reacting. You have described the growth engine as unforecastable and called it a strength. Foroughi’s own framing is that you are testing hypotheses hoping for a good result. One quarter without a lift took sequential growth from 11% to 4%. That is not a lumpy cash flow. That is a business whose primary revenue driver is a research outcome, and you are capitalising it at a fifteen-year duration.

BULLGaming took fourteen years to reach full penetration. Consumer has had eighteen months and is compounding faster. Growing 28% above the seasonal peak during the seasonal trough is a hard number to fake, and advertisers are telling analysts they are nowhere near their return ceiling.

BEARThen look at the tape rather than the transcript. The Trade Desk is down 65% and grew 3%. AppLovin is down 54%. Magnite is up 49% and PubMatic is up 95%. The market is not selling ad tech — it is selling the independent buy side specifically, because walled gardens are absorbing performance budgets.

BULLThat argument does not survive contact with Unity. Unity is a demand-side platform competing with us directly for mobile game budgets and it is up on the year. If the walled gardens were structurally eating the buy side, Unity would be down with us. What actually happened is that the two names carrying premium growth multiples lost them and the one that never had a premium did not. That is a repricing of expensive growth, not a structural verdict.

BEARFine — take the read-across off the table. I do not need it. The earnings you are paying 23x for are flattered anyway. Adjusted EBITDA excludes stock compensation running 8–9% of revenue. The effective tax rate looks like nine percent. Free cash flow came in at 53% of EBITDA. Fifty-five insider sells and zero buys in twelve months, including the chief executive selling into the decline at $500. And your best defence of the quarter is management’s promise that it is already fixed.

BULLWhich is checkable in eleven weeks. That is the point of a guide built on shipped work.

CHAIRThe bear wins the disclosure argument and loses the read-across. The bull wins the quarter and has already been paid for it. On evidence, the consumer gap is real and load-bearing: you cannot underwrite what you cannot size, and the absence is a choice rather than an oversight. But the sector argument, which is the one most often deployed against this stock, does not hold once Unity is in the table — and we should say so plainly even though it makes the bear case less tidy. On the quarter, the evidence offered for the timing explanation is exactly the evidence that would exist if the explanation were true and largely would not exist if the business were losing share, and the SEC resolution is a genuine, underweighted positive. The reason this is fairly valued rather than a mispricing in either direction is arithmetic, not judgement: at $311.98 the market’s enterprise value sits within 1% of our base case. The 54% drawdown has already done the work of pricing the risk the bear describes. You are neither being paid to take it nor overpaying to avoid it.

What is genuinely uncertain

One thing, and nobody in the debate above knows it: whether the model-release cadence is stochastic around a stable mean or has structurally slowed as the gaming category matures. That single question separates the bear case from the bull case, it is unanswerable from outside AppLovin, and the third-quarter print is a single noisy observation on it. Anyone quoting a point target on this name — including the sell side at $559 — is expressing a view on that question and calling it a valuation.

What to verify next, in priority order

One. The Q2 Form 10-Q — actual stock-based compensation, the actual effective tax rate, and the data-centre disclosure Stumpf referenced, which together test the ten-cents-per-incremental-dollar compute claim. Two. Third-quarter free cash flow conversion against the ~75% full-year target. Three. Third-party mobile-game download and cost-per-install data through Q3, to test whether Foroughi’s rebuttal on category health holds. Four. The proxy, on board composition and on the size of the stock-compensation pool.

11 · In plain languageNo jargon

If you’re newer to this

What the company does

Think about the last free mobile game you played. Every so often an ad appears for a different game. Some company had to decide, in the fraction of a second before that ad loaded, which ad to show you and how much to pay for the privilege. AppLovin is very often the company making that decision. It built an AI system called AXON that became extremely good at guessing which ad you personally will respond to, and it owns a big chunk of the plumbing that runs the auction deciding who wins the slot. It used to make games itself and sold that business in 2025 to focus entirely on the advertising.

How it makes money, and why it is so profitable

A game developer says: I will pay for new players as long as each one earns me more than they cost. AppLovin’s system finds those players. When the AI gets better at finding good ones the developer’s returns improve, so the developer hands over more budget. No salesperson required. That is why about 84 cents of every revenue dollar becomes operating profit before certain costs, and why roughly a thousand employees generate eight billion dollars of revenue. Those are unusual numbers even by software standards.

Why the stock halved while the business grew

For three years AppLovin beat its own forecasts every single quarter. In August 2026 it missed — by less than one percent — and the shares fell 16% in a night. Investors had been paying a very high price for the assumption that the beats would never stop. When one stopped, they stopped paying that price. Nothing happened to the earnings. What changed was how many years of future growth people were willing to pay for today.

What could go right

The company is pointing the same AI that dominates game advertising at online shopping ads, a far bigger market. Early signs are good: ecommerce advertisers put $1.28bn through the platform last quarter and are apparently still spending less than they profitably could. If that works, today’s price will look cheap in hindsight. Also worth knowing: a US regulator was investigating how AppLovin collects data and closed that investigation in August 2026 without taking any action, which removed the scariest thing that could have happened.

What could go wrong

Three things, in rough order of how much they should worry you. Meta and Google are enormously bigger and sell ads too; if either decides to compete hard for this exact niche, AppLovin has a real problem, and so far neither has fully committed. The growth is lumpy in a way nobody can predict, because it comes from the AI getting better and AI research does not run on a schedule — that is exactly what happened last quarter and it will happen again. And you cannot see the important number: AppLovin does not tell investors how much revenue the shopping-ads business actually makes, only how much advertisers spent, which is a different and much bigger figure. The part of the story you most need to check is the part you cannot check.

The numbers, plainly

  • Is it profitable? Extremely — about $4.4bn of profit over the last twelve months.
  • Is it growing? Yes, 53% last quarter. But slowing: it was 66% two quarters ago.
  • Does it have too much debt? No. $3.7bn of debt against $3.05bn of cash. That is comfortable by any standard.
  • Is it cheap? About 20x next year’s earnings. Cheap for this growth rate, fair if the growth slows sharply.
  • Do insiders own it? Yes, the founder owns a lot. He has also been steadily selling, and has not bought any during the fall.

Your checklist

  • Easy to understand? Partly. The business model is simple. Why the AI works better than everyone else’s is a black box, and that matters.
  • Financially strong? Yes. Huge margins, real cash generation, almost no net debt.
  • Growing? Yes. 53% last quarter, decelerating but from a very high base.
  • Reasonably valued? Yes. It sits inside our $300–375 band, in the lower half.
  • Risks understood? Partly. The competitive and lumpiness risks are visible. The size of the ecommerce business is not disclosed, so one key risk cannot be measured.
  • Needs more research? Yes. Read the Q2 10-Q for the real tax rate and stock compensation figures, and wait for the November quarter before drawing conclusions about whether last quarter was a blip.

One closing thought, and it is the most useful habit in this whole report. A stock falling 54% while its revenue grows 53% is not automatically a bargain, and it is not automatically broken either. Both can be rational at the same time, because what changed was not the business but how many years of future growth investors will pay for today. Decide what you believe about 2028, not about last quarter. That is the actual question. And on a stock that moves 84% a year in volatility terms and has printed 16% single-day moves, position size matters more than entry price — if a further halving would be intolerable, the answer is not to wait for a better price. It is to own less of it, or none.

AppendixSources & method

What is verified, what is ours, and what is still open

StatusWhat it covers
Verified from AppLovin All Q1 2026, Q2 2026 and Q2 2025 revenue, adjusted EBITDA, net income, diluted earnings per share and free cash flow; the FY2025 revenue, adjusted EBITDA, free cash flow and repurchase figures; the 30 June 2026 balance sheet including cash, total debt and share count; third-quarter revenue and adjusted EBITDA guidance including the statement that it excludes undeployed model releases; consumer advertiser spend of $1.28bn and its relationship to the Q4 2025 peak; the sale of the Apps business to Tripledot and the $106m pre-tax gain; and management and analyst commentary from the 5 August 2026 call, including the model-timing explanation, the compute framework, the MAX publisher-earnings and waterfall-share evidence, and the SEC no-action disclosure.
Verified from market data The $311.98 closing price of 17 August 2026 and every closing price and year-to-date figure in the peer table in Section 03; the $303.17–745.61 52-week range; the $673.78 2025 closing price from which every year-to-date figure is computed; and the implied and realised volatility readings of roughly 54% and 84%.
Our estimate or judgement Enterprise value, net debt and every multiple in Section 01, including the 33.3x January and 15.5x current EV/EBITDA figures; FY2026E revenue of ~$8.1bn and adjusted EBITDA of ~$6.8bn; the entire ten-year cash flow path in Section 03, the sensitivity grid, the reverse test and the $300–375 band with its $335 central estimate; the calibration of the Section 02 widget and its after-SBC margin convention; the estimate that stock compensation runs 8–9% of revenue and that the effective tax rate is near 9%; the take-rate assumption behind the under-8%-of-group consumer estimate; all scores in Sections 06, 07 and 08; and the reading of the peer tape as a repricing of premium multiples rather than a structural verdict on demand-side platforms.
Still open Q3 2025 revenue and adjusted EBITDA are backed out of reported annual and quarterly totals rather than separately reported, and every trailing-twelve-month figure on this page inherits that estimate. Stock-based compensation and the effective tax rate are derived by subtraction and will be reconciled against the Q2 Form 10-Q when we next review. Board composition is flagged in Section 08 as an observation to check against the proxy rather than a finding. We are not publishing peer market capitalisations, growth rates, margins or multiples, because we could not source them to primary company reporting; the peer table carries verified prices and year-to-date moves only.

Method. Unlevered discounted cash flow, methodology version 1.2. Ten explicit years from 2027 with terminal value by perpetuity growth, discounted from the 30 June 2026 balance-sheet date against net debt of $0.65bn measured at that date. Stock-based compensation is deducted as a real cost, which is the single largest methodological difference between this model and sell-side targets built off adjusted EBITDA multiples, and it accounts for most of the gap to the $559 consensus target. Share counts are 335m outstanding today, with terminal counts of 300m to 340m across the scenarios. The reference price used throughout is the $311.98 close of 17 August 2026, the last completed session before publication, and every multiple in this report scales directly with it.

Access

Get the next report when it publishes.

Every report is published in full and free, permanently — model included. Get new coverage by email as it publishes, or see the founding-member tier for reports 72 hours early and a say in what gets covered.

No card, no paywall. Privacy policy.

Open the model