Initiated 2026-09-20 · Price $308.06 (as of 2026-09-18 close, stockanalysis.com) · Mkt cap $103.09bn · Technology / performance advertising · Model: lite, verified (generic, two years of continuing-operations history)

Rating: Buy — Conviction: Medium

AppLovin sits at a 52-week low, 59% below its high, at 17x forward earnings, while growing revenue 53% year over year at an 84% adjusted EBITDA margin and shrinking its share count. The de-rating is not about the results — Q2 2026 missed the guidance midpoint by a rounding error and the SEC closed its AXON data-practices investigation with no action — but about a genuine question: whether the sequential improvement cadence of one machine-learning model is a durable growth engine or a run of good quarters. That is the right question, and a securities class action now alleges the company overstated exactly that. The call is Buy anyway, because the price has moved much further than the evidence: the model's Base case, which already fades growth from 46% to 14% over five years and applies an exit multiple below today's, sits 18% above the last close, and the bull branch is worth roughly double. Conviction is Medium, not High, because the bear branch is −47% and the history is only two years deep.

Business overview

AppLovin is now a single business: a performance-advertising platform. The AXON engine matches advertisers — historically mobile game developers, increasingly e-commerce and direct-to-consumer brands — to in-app inventory through a real-time auction, and monetisation improves as the model improves. The mobile-gaming Apps segment was divested during 2025, so FY2023 and FY2024 were restated as discontinued operations; continuing-operations revenue was $3,224mm in FY2024 and $5,481mm in FY2025 (+70%), with H1 2026 at $3,766mm (+56%). Earnings turn on three variables: advertiser spend on the platform, the model's improvement cadence (Q2 net revenue per installation rose 58% on 2% fewer installations), and essentially nothing else — cost of revenue was 11.7% of Q2 revenue and capital expenditure was $1.4mm in the quarter.

Bull case

1. The unit economics are extraordinary and still improving. Q2 2026 revenue of $1,923.7mm produced $1,494.3mm of GAAP operating income — a 77.7% GAAP operating margin, 84% on adjusted EBITDA — on $1.4mm of capex. Almost every incremental dollar of revenue reaches cash. Model: ebitda_margin, capex_pct_rev
2. Growth is still near 50% at $8bn of run-rate revenue. H1 2026 revenue +56%, Q3 guided to $2,055–2,085mm (+46–48%), with net revenue per installation up 58% on 2% fewer installs. The engine is monetising better rather than simply buying more volume. Model: rev_growth
3. Cash conversion genuinely shrinks the share count. FY2025 free cash flow was $3,952mm on $5,481mm of revenue; diluted shares went 348mm → 342mm → 337mm with $551.3mm repurchased and withheld in Q2 2026 alone. SBC is only ~3.8% of revenue, so unlike most software the buyback nets to real shrinkage. Model: not in the value range — share_change, capital_return_pct move EPS, net cash or a cross-check only (see model note)
4. Two overhangs came off and the multiple did not recover. The SEC's AXON data-practices investigation closed with no action, the company kept guiding to mid-40s growth, and the stock still trades at its 52-week low on 17x forward earnings. Model: exit_ev_ebitda, wacc

Bear case

1. The thesis rests on one model's improvement cadence, and that cadence just missed. Q2 revenue came in below the guidance midpoint and the CEO attributed it to the next step-up in model performance landing just after quarter end. BTIG and Bank of America cut targets citing diminishing sequential returns from automated model updates and saturation in mobile-gaming user acquisition. A business whose growth is an engineering release schedule is harder to underwrite than one whose growth is a contract. Model: rev_growth
2. A securities class action attacks the improvement story itself. Filed over a 2026-02-12 to 2026-08-05 class period, alleging misstatements about AXON model stability, the commercial readiness of the generative-AI video creative tool and early e-commerce traction; lead plaintiff deadline 2026-11-16. The litigation cost is second-order for a company this size; the allegation is not, because it is about the exact variable the model turns on. Model: rev_growth, wacc
3. Margins this high are a target, not a moat. A 78% GAAP EBITDA margin in an auction business whose competitors for the same performance budgets are Meta, Google and TikTok. The Bear case does nothing exotic — it lets margin decay toward 69% while growth normalises, and that alone is worth −47%. Model: ebitda_margin
4. Concentration, and a history two years deep. One platform, one algorithm, an advertiser base still weighted to mobile gaming, and the e-commerce expansion the market was paying for is not separately quantified in the reported numbers. After the Apps divestiture the continuing-operations series runs only from FY2024, which is why this model is lite tier. Model: rev_growth, exit_ev_ebitda

Valuation & balance sheet

Model-implied value range (from model-summary.json; generic module, DCF-Gordon and DCF-exit-multiple, midpoints): Bear $162.14 · Base $361.92 · Bull $606.00 per share, i.e. implied returns of −47% / +18% / +97% vs $308.06. These ranges show how the bull and bear drivers translate into value; they are not price targets. Nothing floors at zero, so the full range is informative. Two things about the shape matter. First, the Base case is not heroic: it fades revenue growth from 46% to 14% across five years, holds the GAAP EBITDA margin roughly flat rather than expanding it, and applies an 11x exit multiple against the 16.2x the market pays on FY2026E EBITDA today — a de-rating built into the base. Second, the Y1 output is a cross-check rather than a forecast: model Base EPS of $15.24 against the quote page's 17.1x forward multiple, which implies consensus nearer $18 on a non-GAAP basis, says the model is running slightly behind the street. The distribution is what makes this a Buy: at the current price the base branch pays 18%, the bull branch roughly doubles, and you are paid for taking the risk that the model cadence breaks — which is the bear branch, and it is a real −47%.

Balance sheet: gross leverage 0.83x FY2025 EBITDA, net leverage 0.12x, coverage 22.2x. Cash of $3,053.3mm at 2026-06-30 against $3,550mm of senior notes in four tranches — $1.0bn 5.125% due 2029, $1.0bn 5.375% due 2031, $1.0bn 5.500% due 2034 and $550mm 5.950% due 2054 — so nothing matures before December 2029 and the model's base path turns the company net-cash within a year. No agency rating was retrieved in this session.

Model note: Built at lite tier — the Apps divestiture in 2025 restated FY2023 and FY2024 as discontinued operations and no comparable continuing-operations D&A figure exists for FY2023, so the history runs only FY2024–FY2025. Treat base rates from two observations with appropriate suspicion; the Dashboard says so in red. 0 failing error checks, 0 failing warning checks, no scenario consistency CHECKs, no unverified inputs and no assumptions without a basis. verification.status is verified as of 2026-09-20 (LibreOffice's independent recalculation matches the Python values on every formula cell; it was not_run at initiation because the environment's LibreOffice install was missing its Calc component). Three conventions. (1) ebitda is GAAP operating income plus D&A, so SBC is expensed — this is not the company's adjusted EBITDA, which is why the FY2025 margin here reads 78.1% rather than 82%. (2) share_change therefore carries only the net count change, which is genuinely negative here. (3) cost_of_debt is the principal-weighted 5.43% blended coupon; because the module charges net debt times that rate, it books roughly $27mm of net interest against an actual net expense nearer $70mm — immaterial against ~$6bn of EBIT. Generator note (2026-09-24): in this module share_change, capital_return_pct and cost_of_debt move EPS and the net-cash path only, and peer_pe feeds a cross-check only; none changes the value range, so bull/bear tags no longer cite them. Scenario deltas were rebuilt after multiple_pct shocks were extended to the Gordon method (previously they moved only the exit-multiple half), and EPS now uses average diluted shares.

Scenario stress test

Reasoned from the bull/bear drivers above. The model column comes from the scenario overlays (Base case + shock).

What the table shows is that macro is not the main risk here. Even the worst row, S2, leaves the implied value only 6% below the current price, and most rows stay above it — because the Base case starts 18% above the price. The variance that matters is between the cases, not across the scenarios: Bear to Bull spans $444 per share, against a worst-case macro delta of $72. Whether the AXON improvement cadence continues dominates everything the playbook can throw at it. Currently active/on watch per the playbook: S3 partially active; S8, S10, S11 on watch. S8 is the second-heaviest row here and S10 the one this name would lead.

What would change the call

Upgrades if: Q3 2026 lands at or above the $2,055–2,085mm guide with Q4 guided above 40% growth — evidence the Q2 shortfall really was a timing artefact; or e-commerce advertising revenue is disclosed separately at scale. Downgrades if: Q3 misses the guidance range, or Q4 is guided below roughly 30% growth; or the GAAP operating margin falls two or more points below the 77–80% band without a stated mix reason; or the class action produces a restatement or a disclosure that changes how AXON's reported performance gains were measured.

Watch items

• W1: Q3 2026 revenue against the $2,055–2,085mm guide, and the Q4 guide — the single test of whether the improvement cadence resumed — expected early November 2026 — Model: rev_growth
• W2: GAAP operating margin — holds in the 77–80% band set by FY2025 (75.8%) and H1 2026 (77.9%) — quarterly release — Model: ebitda_margin
• W3: Diluted share count and repurchase pace — 337mm in Q2 2026 after $551.3mm of buyback; the base case assumes 3% annual shrinkage — quarterly release — Model: share_change, capital_return_pct
• W4: Class action docket, lead plaintiff appointed after 2026-11-16 — watch for any amended complaint that cites internal metrics on AXON performance rather than public statements — Model: none

Sources

• AppLovin Announces Second Quarter 2026 Financial Results (Form 8-K Ex. 99.1, filed 2026-08-05) — https://www.sec.gov/Archives/edgar/data/1751008/000175100826000057/exhibit991-2q26earningspre.htm — accessed 2026-09-20
• AppLovin Announces Fourth Quarter and Fiscal Year 2025 Financial Results (Form 8-K Ex. 99.1) — https://www.sec.gov/Archives/edgar/data/1751008/000175100826000005/exhibit991-4q25earningspre.htm — accessed 2026-09-20
• SEC XBRL companyconcept — APP RevenueFromContractWithCustomerExcludingAssessedTax, continuing operations restated in the FY2025 10-K (accession 0001751008-26-000010, filed 2026-02-19) — https://data.sec.gov/api/xbrl/companyconcept/CIK0001751008/us-gaap/RevenueFromContractWithCustomerExcludingAssessedTax.json — accessed 2026-09-20
• SEC XBRL companyconcept — APP OperatingIncomeLoss (same accession) — https://data.sec.gov/api/xbrl/companyconcept/CIK0001751008/us-gaap/OperatingIncomeLoss.json — accessed 2026-09-20
• AppLovin Form 8-K, 2024-12-05 — $3.55bn of senior notes in four tranches; proceeds repaid the 2018 term loans — https://www.sec.gov/Archives/edgar/data/1751008/000119312524271603/d896321d8k.htm — accessed 2026-09-20
• AppLovin (NASDAQ: APP) faces securities class action amid concerns over model improvements (2026-09-19) — https://www.globenewswire.com/news-release/2026/09/19/3365039/32716/en/applovin-nasdaq-app-faces-securities-class-action-amid-concerns-over-model-improvements-hbss.html — accessed 2026-09-20
• AppLovin stock lost $40bn on the Q2 miss while the SEC closed its AXON data probe — https://www.techtimes.com/articles/323336/20260806/applovin-stock-lost-40b-16m-miss-while-sec-cleared-axon-data-probe.htm — accessed 2026-09-20
• AppLovin (APP) quote and key statistics — https://stockanalysis.com/stocks/app/ — accessed 2026-09-20