APP

We rate AppLovin a Buy, with low confidence, at $279.20. Our central fair value is $476 a share, 70% above the price; weighting in a 10% chance of a platform, regulatory or legal shock and a 25% chance that growth stalls gives $410, still 47% above it. We would buy below $452, the price at which the central case earns our required 12% a year.

How to read the cases. Each report values a company in several cases. The central case is our best estimate. The bear case is a severe but plausible bad outcome, here "saturation": AppLovin's share of game ad auctions stops rising and rivals bid harder. The bull case is the matching good outcome. This report adds a fourth, the shock case: a platform, regulatory or legal event cuts the data AppLovin's ad model uses. We weight shock/bear/central/bull 10/25/45/20; the weights are ours. The exit price is our estimate of the stock's value on 2031-12-31, the median of three methods: a multiple we judge fair (18x 2032 NOPAT, net operating profit after tax), the "long fade" (which lets returns on new capital drift down toward its peers' over time), and the "outside view" (what mature ad companies trade at). Cost of equity (10.80%) is the yearly return investors require; we discount future cash to today at that rate.

AppLovin runs the auction (MAX) that most large mobile games use to sell their ad space, and the ad network (AppDiscovery, driven by the AXON model) that buys most of it for advertisers. 2Q26 revenue was $1.92B, up 53% a year, at an 84% adjusted EBITDA margin; management guides 3Q26 to $2.06-2.09B. Our central case grows revenue from $8.2B in 2026 to $18.5B in 2031 as gaming slows to single digits and web advertisers (e-commerce, now open to self-serve) grow to a third of revenue, with the operating margin easing from 77% to 72%. That is worth $476 a share at 18x 2032 operating profit after tax, below the mature ad peers' 19.7x. Our revenue is within 2% of the Street's through 2028 and our EPS within 7% (at a lower margin than the Street's), and our value is near the Street's median target ($490). Weighting in a 10% chance of a platform, regulatory or legal shock ($114) and a 25% chance of saturation (the bear, $208) gives $410. The price, $279, is 14x our 2027 earnings: it pays for a business whose profit grows little after 2026, close to our bear, and is 34% above it. The market's reasons are recent and specific, though most are press or analyst claims rather than filings (section 6): a court's refusal on 2026-10-01 to stop Unity bidding in MAX while their data dispute goes to arbitration, channel checks that MAX's share has hit a ceiling, the first quarter in eight that missed its own guide midpoint, a county lawsuit on children's data and a new securities class action. None of it is yet in a filed number. We have no information edge; the call is that the price already charges the bear case at 73% weight, against a business that guided its next quarter up 7-8% sequentially. Buy below $452 with low confidence; the 4Q26 guide on 2026-11-04 is the first test (section 5).

1. AppLovin earns from an ad auction and an ad network, at an 84% adjusted EBITDA margin

• Who pays. Advertisers pay AppLovin per install or per purchase; AppLovin pays publishers for the impression and keeps the spread (revenue is reported net of what publishers receive for its own network). Mobile game studios are most of the buyers. Since late 2024 web advertisers (direct-to-consumer brands, "consumer" in management's word) buy through a pixel on their sites; the self-serve AppLovin Ads Manager opened to the public in 2Q26, sequenced to mid-market advertisers first with the long tail later (management claim, 2Q26 call).
• Why it wins. MAX mediation sees the bids of every network in the games that use it (the company's description; we have no count of its share), so AXON trains on a wide view of what an install and a purchase are worth. Reported proof (10-Q 2Q26 MD&A and cost note): net revenue per install +58% in 2Q26 with installs −2%; adjusted EBITDA margin 82% in 2025 and 84% in 2Q26; investing outflows of $7.7M in 1H26 (compute is rented: $181.0M of datacenter cost in 2Q26).
• What limits it. (a) Gaming ad budgets: installs fell in 2Q26, so growth came from price, which a ceiling on MAX share would stop (an outside view from Edgewater, a third-party research firm, not a filing). (b) Access to data that passes through publishers' apps: rivals' SDKs (software kits built into apps) can read it (the Unity dispute), and Apple's App Tracking Transparency (ATT) and Google's privacy changes set the rules (10-K risk factors). (c) Web advertisers can buy the same people on Meta and Google; AppLovin's pitch is incremental reach in games.
• Capital. $3.55B of notes (5.1-6.0%, 2029-2054), $3.05B cash (2026-06-30), $1.0B revolver undrawn; net leverage target about 1x (management). Buybacks of about $1.5B in 1H26 (10-Q) roughly offset stock pay; the model holds the count flat and expenses stock pay ($86M in 2Q26).
• History is split in the 10-K: Apps (first-party games, $1.4-2.1B of revenue a year in 2021-24) is discontinued; 2023-25 continuing lines come from the FY2025 10-K, 2021-22 from the advertising segment. Revenue grew 56%, 76%, 75% and 70% a year in 2022-25 (output.txt).

2. 2Q26 was the first quarter in eight to land below the guide's midpoint

The guidance record (8-K releases, eight quarters, printed by model.py):

Seven quarters above the top of the range, then 2Q26 inside it and just below the midpoint: the beat shrank from about 10% to zero. Management's explanation (2Q26 call, a management claim): the quarter "came down to timing", the next model improvement "landed just after quarter end", MAX publisher earnings grew double digits sequentially and AppLovin's "share of publisher waterfalls remained consistent"; the 3Q guide, unusually, includes improvements already live and higher compute cost. Because that guide already counts the improvements, the revised 3Q26 forecast is the guide midpoint ($2,070M); the blind model added half the last-four median beat ($2,118M), which the audit reviewer (an outside reviewer who checked our numbers and logic) rightly questioned after a miss (−$6). The 4Q26 forecast is 3Q × 1.15 ($2,381M; 4Q over 3Q was 1.18-1.20 in 2024-25). Edgewater's channel checks (reported 2026-09-23) put 4Q at +8-9% sequential; that would take 2026 revenue to $8,082M and fair value to $468 (unblinded-output.txt). Ours, $8,217M for 2026, is 1% above the Street's $8,123M.

3. Our forecast: revenue of $18.5B in 2031 is worth $476 a share

model.py, output in output.txt; the blind record in assumptions.md. The company reports one segment and no revenue split, so the two lines are ours: gaming and consumer (web), with the 2026 consumer share of 14% inferred from management's "billion-dollar run rate" and "that sort of 10% range" in 2025 (management claims, 1Q25 and 2Q25 calls). Margin is one path, because no profit by line is reported.

• Central. Gaming +20% in 2027 slowing to +6% by 2032, below management's 20-30% long-term claim, because 2Q26 installs fell and the growth was price. Consumer +70% slowing to +14%. Operating margin after stock pay 76% easing to 72% (compute grows with revenue; web needs sales spend gaming did not). Tax 16-17%. 2031 revenue is $18.5B, against management's 25-30% compounding ($25B+).
• Bear: saturation. Gaming +10%, +5%, then +3%; consumer +35% to +5%; margin to 63%. MAX's share stops rising, Unity and Google bid harder in its auctions, the take rate and model edge erode. NOPAT is flat from 2027 to 2031.
• Shock (added after review). A platform, regulatory or legal event cuts the data AXON uses (Apple or Google rules, children's-data enforcement, rivals reading MAX): gaming revenue falls 10% in 2027 and 5% in 2028, consumer grows 15% then low single digits, margin falls to 55%, exit at 10x. The blind bear carried this label but its numbers are a slowdown; the shock was printed only as an unweighted stress. Weight 10% (our assumption: the pending children's-data suit, the Unity arbitration and the platform rules are live, none has yet cut a reported number). The weights moved from 25/50/25 to 10/25/45/20: the shock takes 5 points from central and 5 from bull, because a world with that event has no bull path.
• Bull. Just under management's pace (22.4% a year 2026-31 against their 25-30%): gaming +25% to +8%, consumer +100% to +15%, margin 77-78%.
• Exit multiples are judged from the mature ad peers' forward P/E (median 19.7x; Meta 21.6x, Alphabet 22.9x) and our own long fade (15.2x on central 2032 NOPAT): central 18x sits between them; bear 13x; bull 22x. The blind record also gave "a single product line" as a reason for 18x over 19.7x; our report checklist (research-eval) rightly read that as concentration risk, which the shock and bear outcomes charge. At 19.7x the central value is $509 (+$33). We keep the blind 18x as the headline and price the alternative here. The bear-side reviewer (an outside reviewer asked to argue the bear side) argues 16x (our fade 15.2x; peers 17.5x on FY2028 EPS): −$42 alone; rejected, because the fade already sets the long-fade method's value and taking 16x would put two of the three methods on the fade's assumptions. The printed cross-check says 18x implies 5.6% perpetual growth at a 60% return on new capital, below the path's 8.5% in 2032 and 7.7% in 2033: the multiple is conservative against the path, as the fade's 3.5% end point implies. The value-driver formula gives 19.5x / 12.1x / 29.1x, a cross-check only. In the bear the outside view ($365) sits 48% above the median: the peer multiple is for growing companies and the bear has stopped growing, so the judged multiple and the fade are believed there.
• Peer multiples. The FY2027 EPS some peers carry is odd (Alphabet $20.58 in 2026, $15.25 in 2027, $18.14 in 2028; The Trade Desk similar). On FY2028 EPS the peer median is 17.5x, the outside view $624 becomes the median, and fair value is $463 (−$13).
• APP's own multiple history was not used blind (it needs the price); after the blind commit it reads 31x at the end of 2024 and 43x at the end of 2025 on the next year's NOPAT (market data enterprise value, hindsight NOPAT). It does not argue for a lower multiple; it shows how far the market's view has swung (14x now).
• Tax. 16-17% book. 1H26 cash taxes were 21.8% of pre-tax income (timing of international payments, per the CFO). At 22% for 2027-32 the value is $447 (−$29).
• Singapore incentive (added after review). The 10-K (FY2025 Note 13) says the Singapore tax incentives cut 2025 tax by $272.1M (6.9 points, before US tax effects) and "are expected to expire in June 2028", which the company "can affirmatively elect to renew". Every case assumes renewal. The net effect visible in the 2025 rate reconciliation is about 2.5 points (the Singapore rate difference and local rate, after the $82.7M Pillar 2 top-up). Not renewed from mid-2028: central $462 (−$14) at +2.5 points, $439 (−$37) at +6.9; case-weighted $399 and $379.
• Consumer share. The "billion-dollar run rate" in 1Q25 was about 22% of that quarter's annualised revenue, against the "10% range" claimed for 2025, so the two quotes disagree (perhaps gross spend against net revenue; not stated). At 25% the value is $552; at 8% $434. The bear-side reviewer's 12% with slower growth (55% to 12%) is −$46 alone; rejected as a central case because management's latest figure (consumer spend 28% above 4Q25 in 2Q26) points the other way, though both are claims, not reported lines.
• Cost of equity 10.80% = US 10-year 5.089% (market data, 2026-10-15) + beta 1.36 × 4.20% (Damodaran implied premium, 2026-10-01). The beta is the median of six ad peers' five-year weekly regressions; APP's own is 2.55, which gives 15.80% and $387 (printed). We use the peer median because a stock's own beta after a rise of about 10x in 2024-25 (training memory) and a fall of more than half in 2026 measures its price swings more than its business risk; the reader who prefers 2.55 still gets $387, 39% above the price. The audit reviewer's ad-tech-only median (TTD, PINS, CRTO, MGNI) is 1.46, $467 (−$9); rejected, because the set was fixed blind and Meta and Alphabet are advertising businesses at APP's 2031 scale.
• Where the platform, regulatory and short-seller risk is charged: once, in the shock outcome (10% weight; the blind report said the bear, whose numbers did not carry it). Not in the discount rate, not in the central multiple, not as a legal deduction in the bridge (no loss is accrued, 10-Q Note 5). Evidence counted: the 10-K risk factors, the pending securities case, and management's statement that the SEC's inquiry ended without action (CFO, 2Q26 call; a management claim, no SEC document). The 2025 short-seller reports and the new lawsuits are allegations, not evidence of a loss.
• Bridge at exit. Enterprise value + 2026 year-end cash ($6.1B, after the guided 2H FCF) − operating cash (5% of 2031 revenue) + Tripledot and fund stakes at book ($483M) − notes $3,550M − leases $154M. FCF to equity in 2027-31 ($131 a share, central) is treated as paid out in buybacks at intrinsic value.

4. The price needs profit that barely grows after 2026

unblinded-output.txt. Market value $94.0B; net of cash, stakes, notes and leases the enterprise value is $94.1B: 17.6x our 2026 NOPAT and 14.1x 2027. On our central earnings the price needs an exit multiple of 8.6x 2032 NOPAT. At our 18x, with the distributions following the same lower path, it needs growth about 12 points a year below central (+15%, +10%, +5%, +2%, −1%, −3% in 2027-32): 2032 NOPAT of $6.44B, 20% above 2026's $5.35B and close to our bear ($6.12B). With profit held flat at 2026's level it needs a 23x exit. Or our central case at a 24.3% cost of equity. (The blind draft said "2032 profit flat on 2026" from a calculation that cut the exit but kept central distributions; the audit reviewer caught it.) The Street's earnings are close to ours (2027 EPS $20.06 vs $19.85; 2030 $29.78 vs $30.21), and its median target ($490) is near our fair value: the gap is between the price and both of us. The vendor's end-2025 market value was $228B, against $94B today. The stock rose 6.5% today ($262.19 at the 2026-10-14 close to $279.20) with no company news we found.

5. Why we disagree with the market, and what settles it

• The market's thesis. AppLovin's gaming engine has peaked: MAX's share has hit a ceiling, Unity now bids inside MAX and narrows the spread, and the web business started slower than claimed (a Wells Fargo note called the pixel's install spike "a false start"). Add legal overhang. On that view 2026 profit is roughly the peak.
• Ours. Growth slows hard (27% in 2027, 8.5% by 2032) but does not stop, because the 3Q26 guide is +7-8% sequential after +4% in 2Q26, consumer spend rose 28% in two quarters (management claim), and the cost base is almost all variable. Even the market's thesis, written as our bear, is worth $208, 25% below the price; the price only fails if profit falls, which our shock outcome (gaming −10% in 2027, margin to 55%, 10x) puts at $114, weighted at 10%.
• The bear-side reviewer's case (Hold to narrow Buy). Its five levers together give a central of $329 and a weighted $325, still above the price; each alone is in section 9. We reject the gaming, web, multiple and margin levers as a central case because their evidence is third-party channel checks (Edgewater, Wells Fargo on Similarweb data) set against reported 2Q26 revenue +53% and a company guide for 3Q +7-8% sequential, and because the bear-side reviewer's own search found Tenjin data showing share gains on iOS and Android in 2Q26. They are the bear's world, which we weight.
• What settles it. The 4Q26 revenue guide on 2026-11-04. At or above about $2,380M (+15% on a $2,070M 3Q) our central stands; below about $2,250M (+8.5%) the bear-side reviewer's $330 central becomes live and we would revise toward Hold. 3Q revenue against the $2,055-2,085M guide comes the same day; a 3Q below the range moves us toward the bear.
• Our edge. None on information. The call rests on price against a case already charged with the market's fears.

6. The risks that could make the market right

• The Unity ruling may be structural, not a skirmish. On 2026-10-01 a San Francisco Superior Court judge denied AppLovin's request for a temporary restraining order and preliminary injunction against Unity (Seeking Alpha, 2026-10-02; we did not read the ruling). AppLovin alleges Unity's Ad Quality SDK takes MAX data while Unity bids in MAX; Unity calls it a publisher-authorized tool; the merits go to private arbitration. If a rival can read MAX's auctions through a publisher-installed SDK, AXON's data edge narrows and the take rate with it (Wedbush's view). This is the bear's mechanism; it is not yet in a reported number.
• Price per install cannot carry growth forever. 2Q26 installs fell 2% while revenue per install rose 58%. Growth by price alone ends when advertisers' returns fall; the first guide miss in eight quarters may be that point.
• Children's data. San Diego County sued AppLovin (reported 2026-10-05) alleging it collected children's data and showed adult ads on children's devices. COPPA and state laws carry per-violation penalties and, worse, could limit the data AXON uses. Allegations only; the 3Q26 10-Q will say how the company accounts for it.
• A new securities class action (law-firm notices, class period 2026-02-12 to 2026-08-05; the bear-side reviewer cites Talbot v. AppLovin, 3:26-cv-10584, filed 2026-09-16) adds to the 2025 case. Settlements are usually small next to a $94B company; the cost is management time and the disclosures that discovery brings.
• Our own exposure. We saw a 1H26 average repurchase near $460 before the model was built, and the saved 2Q26 release (in the blind data) says 1.1M shares cost $551.3M in 2Q26, about $480-500; our blind central value was $482. We cannot show that line was not read. The choices are sourced and the Street's earnings agree with ours, but the reader should discount the central value for anchoring (a first number seen pulls later ones toward it). The bear, which ignores it, is still close to the price.
• Weights. At shock/bear/central/bull 15/30/40/15 the value is $370 (+32%); at the bear-side reviewer's bear/central/bull 40/45/15 it is $402 on the blind model; the price needs 73% on the bear and none on the bull or shock (unblinded-output.txt).

7. What to watch: dates and results that would change our view

• 2026-11-04: 3Q26 results and 4Q26 guide (company-announced date). Our 3Q: revenue $2,070M (range $2,030-2,130: 2% below to 3% above the guide midpoint), adjusted EBITDA $1,718M (83% of our revenue; range $1,680-1,770). These ranges are in record.json. Our 4Q26 revenue is $2,381M.
• About 2026-11-05: 3Q26 10-Q legal note: San Diego County suit, the new class action, any regulator inquiry.
• Unity arbitration outcome (date not public).
• About 2027-02-11: 4Q26 results and the first 2027 outlook (date not confirmed; February in past years).

Falsifiers (results that would show us wrong):

• 4Q26 revenue guide midpoint below $2,250M: the bear-side reviewer's central (about $330) is live; revise toward Hold. 3Q26 revenue below $2,055M, or a 4Q26 guide implying under +5% sequential: move toward the bear ($208).
• AppLovin does not elect to renew the Singapore tax incentive by June 2028 (10-K tax note): $462 central at +2.5 points of tax, $439 at +6.9.
• Net revenue per install flat or down year on year in any 2027 quarter with installs still falling: the gaming engine has peaked; bear.
• A regulator (FTC, a state attorney general, the SEC) files an action on data use or disclosure, or the company accrues a loss: re-underwrite the shock outcome's weight.
• Consumer (web) revenue disclosed below 10% of 2026 revenue: the line is smaller than our 14% inference; $434 at 8%.

8. Data and tooling notes

• Market data quarterly statements mix total and continuing operations in 2024; the 10-K's recast lines were used instead.
• revenue-product-segmentation gives one reportable segment for 2025; no gaming/web split exists in any source we could reach.
• The filter that hid price-bearing filing lines (safe_cat.py) missed shares-withheld and repurchase lines on first use; the leak is disclosed above and in assumptions.md.

9. Two reviews moved the central value from $482 to $476 and the weighted value from $462 to $410

Second review: an audit reviewer and a bear-side reviewer (2026-10-15)

After the blind report, two outside reviewers read it. The audit reviewer checked its numbers and logic. The bear-side reviewer argued the case for Hold.

Each finding measured alone from the blind model (unblinded-output.txt, "Review findings, each measured alone"):

Nothing moved toward the price without a stated cause. The weights finding corrects a risk the blind report claimed to charge and did not; the 3Q beat finding applies a fact from the blind reading (the guide already counts the improvements) that the blind model ignored. The rejected levers are all priced above.

First review: our report checklist (2026-10-15)

One pass of our report checklist (research-eval), run by a Claude Sonnet agent on draft 7f9ccd1, 2026-10-15. No hard FAIL changed a headline figure; the model and its blind record were unchanged. Every finding and its answer; the dollar effects were measured on the blind model (the revised model's are in unblinded-output.txt, "Research-eval alternatives").

The two-reviewer adversarial review (the audit reviewer and the bear-side reviewer) was run by the coordinating agent that briefed the analyst; the second review above answers it.

10. What slowed the work (friction log)

• Bigdata (bigdata_search), three calls, the trial limit:
  1. Smart search on the e-commerce run rate: found the June 2026 Nasdaq conference passage (2026 at about $8B; "25% compounding for 10 years"). Market data also lists that conference and we did not try its transcript there, so this may not have been new.
  2. Fast search: returned nothing.
  3. Smart search on the "$1 billion run rate": found the 1Q25 and 2Q25 call passages. Market data holds those transcripts; Bigdata located the quotes faster but found nothing new.
• The price filter leaked (section 8); a tighter pattern list was added mid-task, after the 10-K equity statement and the 2Q26 10-Q equity note had been seen.
• Damodaran's home page is a frameset; the premium is on New_Home_Page/home.htm.
• The 1Q26 market data transcript was too large for one tool result and had to be converted from its saved JSON.
• The worktree guard refused find -exec and a write to the shared checkout path; plain commands worked.
• The full 10-K and 10-Q texts (1.9 MB) were left out of data/ and are listed by accession number in assumptions.md.

11. Sources (all read 2026-10-15)

• SEC EDGAR, CIK 1751008: 10-K FY2025 (0001751008-26-000010), FY2024, FY2023; 10-Q 2Q26 (0001751008-26-000059), 1Q26 (0001751008-26-000044); 8-K earnings releases and shareholder letters 2Q24-2Q26; 8-Ks 2025-06-30 (Apps sale), 2026-04-02 (management changes). Saved in data/ except the 10-K and 10-Q texts.
• 1Q26 and 2Q26 call transcripts (market data documents), data/.
• Market data: income, balance sheet and cash-flow statements; revenue segmentation; peer quotes and FY2027 estimates (13:01 PDT); weekly returns for betas. After the blind commit: APP analyst estimates, price-target consensus, key metrics.
• Market data TVC:US10Y (5.089%, 20:05 UTC). Damodaran implied ERP (2026-10-01).
• Market quote, read about 20:10 UTC (data/quote-2026-10-15.json).
• News feed (after the blind commit, untrusted, press and analyst claims): Seeking Alpha on the Unity ruling (2026-10-02); market data weekly recaps (2026-10-05, 2026-10-12); Stocktwits on Edgewater (2026-09-23); Business Wire, 3Q26 date (2026-10-07); Dow Jones target-change headlines (2026-10-13/14); Kirby McInerney class-action notice headline (2026-10-15).
• Bigdata bigdata_search: three calls (section 10).
• Alpha: alpha rank APP --raw --json --allow-stale-model-cache, as of 2026-10-14.