Financebotresearch desk研究台

APP › analyze

APP · Analyze from before

ACCUMULATE Technology

2026-09-10 · Price $314.49 (post-market $313.01, −57.8% off the $745.61 high) · Market cap $105.65B · 52wk range $297.50 – $745.61 · 4th consecutive watchlist scan with the name sitting in the $290-330 zone — this file exists because the watchlist's own rule says a repeated "still in zone" is not a decision.

This updates: Output/Stocks/Technology/APP/analyze-2026-08-05.md (WATCH, conviction 6.0, FV $300-450, entry $290-330, trim 25x fwd). No Q3 print has occurred since the baseline — the next print is early November, unchanged. What has changed is price (−10.4% further, from $351 to $314.49, now genuinely inside the zone the baseline itself set), analyst estimates (a sustained wave of target cuts, exactly as predicted), insider activity (the year-long one-way selling streak has gone quiet), and one new soft risk (AI-IPO capital rotation). The SEC investigation — the single item capping conviction — is objectively unchanged.

Event list since 2026-08-05: (1) continuing sell-side target cuts, Aug 6 – Sep 8, including one outright downgrade (BofA); (2) a 10-Q for the quarter ended June 30, 2026, filed in August, containing no new disclosure on the SEC matter; (3) zero new open-market insider transactions (only stock gifts and RSU grants); (4) a Sep 1 Evercore screen naming APP among ~40 stocks at risk from AI-lab-IPO capital rotation; (5) continued price decline to a new low close to the 52-week low ($297.50). No earnings, no guidance revision, no SEC development, no M&A, no leadership change.

Verdict — ACCUMULATE (small starter tranche) · Conviction [6.5] (was 6.0)

The explicit call: this is now a BUY for a small starter tranche, not a PASS and not a plain WATCH. Nothing about the business changed in the last five weeks — every Structural and Trend claim in the baseline survived re-test unchanged (see ledger). What changed is that the price did the baseline's work for it: spot fell from $351 to $314.49 and is now genuinely inside the $290-330 entry zone the baseline set for a reason — to make the SEC binary's discount explicit, not decorative. A zone that has now actually been reached, on unchanged fundamentals, is the "classic re-rating setup" this protocol looks for: the business held, the multiple compressed further.

Sizing is unchanged and still the load-bearing constraint: ≤2% of portfolio, even now. The SEC investigation is objectively no closer to resolved than it was on 8/05 — still "active and ongoing" per the last official confirmation (Feb 2026), no enforcement, no settlement, and the August 10-Q's Legal Proceedings section carries only generic boilerplate with no mention of it at all. Zero insider buying continues at every price down to the 52-week low — the single cheapest signal available to management, and it has never fired. That combination — real discount, unresolved binary, no insider conviction — is why this is a small starter, not a full position, and why conviction moves to 6.5, not 7.0+.

A portfolio-specific passage was removed from the public build.

The delta ledger

Legend: ✅ CARRIED · 🔁 REFRESHED · 📉 DRIFTED · 🔄 SUPERSEDED · ❌ RETRACTED · ⏳ UNTESTED · 🆕 NEW

Leading with the rows that moved something.

# Claim Status Old → New Force / source
26 Sizing: "do not add — not yet in zone" at $351 🔄 SUPERSEDED Spot $351 (below zone) → spot $314.49 (inside $280-320 zone) Price alone. The zone's own logic, not new bullish information about the company
2 FV $300-450 / entry $290-330 / trim 25x fwd 🔁 REFRESHED FV $300-430 (central ~$365, was ~$375) / entry $280-320 (was $290-330) / strong buy <$260 (was <$270) / trim unchanged 25x fwd FY2027 consensus EPS drifted $21.59→$20.98 as the sell-side cut estimates; see #21. Trim multiple untouched — it already tracks this ticker's mislabeled "fwd" field correctly per the baseline's own note
23 Reverse-DCF: $351 embeds ~13-14% 5yr FCF growth vs 46-48% guided 🔁 REFRESHED 13-14% required growth at $351 → ~10-11% required growth at $314.49 (same method, EV fell to $105.71B on unchanged FY26E guide) Recalculated directly, single self-consistent model — the margin of safety widened, not narrowed, as price fell on flat guidance
21 32 ratings/0 sells, mean target $656.20 (pre-print, flagged as stale) 🔁 REFRESHED Mean target $656.20 → $508.39 (−22.5%). One outright downgrade: BofA Buy→Neutral, 8/11 ($430→$400). Continuing cuts through Evercore $630→$510 (9/1) and BTIG $408→$396 (9/8). Still 0 sells, ~31 analysts Primary source (Yahoo recommendations feed, dated grades). The baseline's own prediction — "expect a wave of target cuts, don't anchor on $656" — materialized almost exactly as written. Even after five weeks of cuts, mean target is still ~62% above spot; treat the residual gap as lagging re-rating, not a signal
20 Insider selling $306M Mar-Jul 2026, zero buying 🔁 REFRESHED Continuous one-way selling since Aug 2025 → zero open-market sales Aug 5 – Sep 10 (only two stock gifts, no cash changes hands). Buying still zero Primary source (Form 4 feed). Weak positive, flagged not to over-read: Q3 quarter-end (Sep 30) blackout timing is a fully sufficient non-bullish explanation for a pause. This is a change in a headwind, not a new tailwind — it does not become a golden flag until an actual purchase prints
15 Net debt ~$755M, coverage 28.1x 🔁 REFRESHED Cash $2.76B → $3.05B; debt flat at ~$3.52B → net debt ~$462M, even more unlevered Q2'26 10-Q balance sheet, primary source
5 EBITDA margin step-down 85%→83%, Q3 guided ~83% ✅ CARRIED, flagged at the edge Confirmed: Q2'26 EBITDA/revenue = 82.6% Recomputed from raw Q2'26 income statement. This sits exactly on the break-trigger floor ("below 82% for two consecutive quarters" downgrades to 4.5) — one more soft quarter and it fires
18 Web/e-commerce run-rate ~$1B annualized 🔁 REFRESHED, low confidence ~$1B (within months of launch) → third-party model ~$1.45B for FY2026 Single source (industry blog modeling), not company-disclosed — same standard as baseline applied to the $1B figure. Directionally supportive, not verified

Everything else — CARRIED, no material change:

# Claim Status Check performed
3 SEC data-collection investigation active, unresolved, whistleblower-triggered ✅ CARRIED WebSearch confirms last official status is "still active and ongoing" (Bloomberg, Feb 2026); the Aug-filed 10-Q's Legal Proceedings section (Part II, Item 1) contains only generic boilerplate — no mention of the investigation, no subpoena, no settlement language, no accrual. Absence of specific disclosure is consistent with no material escalation, not with resolution
4 Sequential growth collapsed +11.1%→+4.4%, Q3 guided +7-8% ⏳ UNTESTED No Q3 print yet (early Nov). This remains the single most important open question and is unchanged from baseline
6 Vendor "forward P/E" is FY2027, not FY2026 ✅ CARRIED, confirmed a second time forwardEps $20.98, epsCurrentYear (true FY26) $16.76 — same trap, same direction, now observed on two consecutive passes. This raises confidence it is a durable characteristic of this ticker's feed, not a one-off. Real FY26 P/E is 18.8x at $314.49, not the vendor's headline 15.0x
7 Revenue CAGR corrupted by divested Apps segment ✅ CARRIED fin.py still prints "Revenue CAGR 24.8% (3y)" unchanged — no new annual print to update it. Continuing-ops math from baseline (+75.0%, +70.0%) stands
8 Graham IV void (BVPS hollowed by buybacks) ✅ CARRIED Graham IV now $53.33 vs $314.49 (using updated BVPS $9.43, EPS $13.40) — still a residual-of-buybacks artifact, still discarded per the same reasoning
9 Share count is Class A only; true diluted is higher ✅ CARRIED sharesOutstanding 305.73M vendor field vs impliedSharesOutstanding 335.94M vs true Q2'26 diluted 337.03M
10 ROIC 14.8%→42.0%→70.3%, rising ✅ CARRIED No new annual print; last confirmed point unchanged
11 Gross margin rising 80.8%→88.9% ✅ CARRIED Q2'26 recomputed: 88.26% — flat vs Q1'26, comfortably above the 85% break floor
12 SBC ~7.6% of FCF ✅ CARRIED (noisier by quarter) Q1'26: 6.5% of FCF; Q2'26: 9.9% of FCF (FCF itself lumpy quarter to quarter on working capital). Same order of magnitude, same conclusion vs peer comparisons in baseline
13 Buybacks ~50-56% of FCF, self-funded ✅ CARRIED Q1'26: 76% of FCF; Q2'26: 63% of FCF — still fully self-funded, zero new debt issued either quarter
14 Share count falling ~−2.7%/yr ✅ CARRIED Diluted shares 338.7M (Q1'26) → 337.0M (Q2'26), still declining
16 Moat rating 6.5/10 ✅ CARRIED No structural change to re-test against
17 Evergreen: No ✅ CARRIED Unchanged reasoning
19 R&D ~5.1% of revenue vs Meta/Google 15-30% ✅ CARRIED Q2'26: $99.9M / $1,923.7M = 5.2%
22 Short interest 4.3% of float, not crowded ✅ CARRIED Now 3.9-4.0% of float, 1.59 days to cover (down from 2.13) — the decline is not short covering
24 Net margin < operating margin (no unrealized-gain inflation) ✅ CARRIED Q2'26: net margin 65.9% vs operating margin 77.7% — still holds
25 Beta ~2.53 ✅ CARRIED Now 2.49 — unchanged within noise

16 carried, 8 refreshed, 1 superseded, 1 untested, 0 drifted, 0 retracted.

How the close calls were decided

Does the price move alone justify acting, with zero new fundamental information? Yes, and this is the central judgment of the pass. Per the framework, a verdict requires two separate answers — is it a good business and has the market priced that in — and only the second changed. The baseline explicitly built its entry zone to require a discount for the SEC binary; spot reaching that zone on flat fundamentals is exactly the scenario the zone was designed to catch, not a coincidence to be waved off. Weighed against: acting purely on a price move without new information is also how a stale-zone or a "waiting for a price that no longer needs to come" mistake gets made in the other direction — but here the zone was re-derived this pass (FV recomputed from the current FY26/FY27 EPS, not inherited), so this is a tested entry, not an inherited one.

Does the paused insider selling deserve real weight? No — decided against over-crediting it. Five weeks without a sale, after fourteen consecutive months of selling, is a real change in state, but a Q3-quarter-end blackout window is a complete, boring explanation that requires no change in management's view of the stock. It is recorded as a refreshed claim, explicitly not folded into the conviction increase as an independent force. It would become a real signal only if a purchase actually prints, which has never happened on this name at any price.

Does the widened reverse-DCF margin of safety (13-14% → 10-11%) mean the market has become irrationally pessimistic? Weighed, not accepted outright — the model holds the FY26E FCF assumption (from the Aug 5 guide) flat and only lets EV fall, which mechanically lowers the required growth rate. That is a legitimate calculation, but it is also exactly what happens when a stock re-rates lower on unchanged near-term guidance while a market-wide capital-rotation narrative (Evercore's AI-IPO screen) adds unrelated selling pressure. The model result is real and countable; the "market is being irrational" story built on top of it is not. The ledger records the number, not the narrative.

Does the Evercore AI-IPO screen change anything about AppLovin's business? No — it is a market-flow observation (funds may rotate capital toward OpenAI/Anthropic IPOs, pressuring names with weak year-to-date performance and negative earnings revisions), not a finding about AppLovin's advertising business, data supply, or moat. Recorded as a new, soft, technical risk — worth naming because it could extend unrelated downward pressure into Q4 tax-loss-selling season — but given no weight in the conviction score, which is reserved for business and regulatory facts.

Thesis persistence and conviction delta

Thesis persistence: 14 of 15 testable Structural + Trend claims survived as CARRIED or REFRESHED-without-thesis-change (93%). The one exception (#4, sequential growth recovery) is UNTESTED, not broken — it simply has not had its resolving event yet. This is high persistence against a real price move (−10.4% since baseline, −57.8% off high), which is the pattern this protocol flags as a re-rating setup: the business held; the multiple did the moving.

Conviction: 6.0 → 6.5. Driven specifically by: - #26 — the zone was reached on unchanged fundamentals (the primary driver) - #23 — reverse-DCF cushion widened from ~13-14% to ~10-11% required growth against a 46-48% guide, a materially bigger margin of safety on the same method - #15 — balance sheet got safer still (net debt ~$462M, down from $755M) - #20 — a real, if weak and possibly mechanical, change in a standing headwind (selling paused)

Held below 7.0 (the baseline's own upgrade threshold) because: - #3 — the SEC investigation is objectively unchanged: still open, no closure, no enforcement - #20 (mirror) — zero insider buying persists even at the 52-week low, the cheapest possible price to have expressed conviction, and management has not done so - #4 — the sequential-growth-recovery question that would confirm Q2 was cadence, not trend, remains completely untested

What is genuinely new

  1. Evercore's Sep 1 screen names APP among ~40 Russell 3000 stocks at risk from AI-lab-IPO capital rotation (OpenAI/Anthropic IPOs pulling investor dollars from names with weak YTD performance and negative earnings revisions, compounded by October tax-loss-selling season). A market-flow risk, not a business risk — noted, not weighted.
  2. The one-year insider-selling streak has gone quiet for the first time since it began (Aug 2025) — five weeks with no open-market sale, only gifts and grants. See caveats above; this is refreshed evidence on claim #20, not a new independent claim.
  3. A second consecutive confirmation of the forward-EPS vendor trap on this exact ticker (claim #6) — this is now a pattern observed twice on the same name, which is stronger evidence than a single catch that the "fwd" field is durably mislabeled here, not a transient pre-print artifact.

Updated verdict — full detail, written to stand alone

ACCUMULATE — small starter tranche within $280-320, capped at ≤2% of portfolio in aggregate. Not a full position, not a pass. The company remains a genuinely elite business (ROIC 70%+, gross margin 88%+, self-funded buybacks, falling share count, net debt near zero) trading at a price that has moved into a zone the framework itself set with an explicit discount for one named, unresolved risk. That risk has not gotten worse — it also has not gotten better, and it is the reason this is a 6.5 and a starter, not a 8+ and a full position.

Fair value range (updated)

Case Basis Value
Bear Growth → ~15%, 15x FY26E $16.76 ~$251
Severe bear SEC action / platform restriction, 12x FY26E ~$201
Base 22-25x FY26E $16.76 $369-419
Base (FY27 cross-check) 18-20x FY27E $20.98 $378-420
Bull (tail, excluded from central range) Growth holds ~40%, 25x FY27E $20.98 ~$524

Fair value: $300 – $430 (central ~$365) — spot $314.49 sits inside the range, modestly below centre, and inside the entry zone for the first time since the baseline.

Entry $280-320 · Strong buy <$260 · Trim 25x fwd (≈$524 at current FY2027 consensus — unchanged multiple; this ticker's "fwd" site field is FY2027, confirmed a second time)

Break triggers — unchanged from baseline, none fired, one at the edge

  • Sequential revenue growth below +5% again in Q3 (guide is +7-8%) — untested, the whole question this file could not resolve
  • SEC enforcement action, or any Apple/Google policy change restricting data access — no
  • Adjusted EBITDA margin below 82% for two consecutive quarters — 🚩 Q2 printed 82.6%, effectively at the floor already; Q3 guided ~83%. One soft print away from firing
  • Web/e-commerce run-rate flat or declining — no, modeled growing to ~$1.45B (low-confidence)
  • Buyback paused, or funded with new debt — no, still self-funded both quarters
  • Gross margin below 85% — no, 88.3% in Q2

Upgrade path to [7.0]+ — unchanged, none met

  • SEC matter closed without enforcement (the single largest re-rating event available) — no
  • Q3 delivers ≥+8% sequential with EBITDA margin back to ≥84% — unresolved until November
  • Any open-market insider purchase at any price — still none, ever, on this name
  • Web run-rate above $2B with disclosed advertiser counts — not yet disclosed at that scale

What this pass did NOT test

  • Claim #4 (sequential growth recovery) is UNTESTED for a second time in the sense that it was always going to require the Q3 print — this is not a system failure, it is the correct deferral; flagging it explicitly per protocol so the next pass does not silently inherit it as settled.
  • Claim #18 (e-commerce run-rate ~$1.45B) rests on a single third-party model, not a company disclosure — treat as directional only until the Q3 print or an investor-day disclosure.
  • Claim #20's "paused selling" was corroborated only against the Form 4 feed's most recent entries; it was not cross-checked against 10b5-1 plan filing dates, so the blackout-window explanation is a plausible alternative, not a confirmed one either way.
  • The SEC investigation's actual internal status (Wells notice, staff recommendation, scope) is not observable from any source available to this pass — the finding is limited to "no public escalation has occurred," which is a weaker claim than "no escalation exists."
  • Did not re-verify 10b5-1 plan structure/adoption dates for any insider, per baseline's own caveat, still unverifiable from this data.

Sources

Quantitative data: Yahoo Finance via .mcp/fin.py and the yahoo-finance MCP, 2026-09-10. Statements are GAAP as reported; adjusted figures are labelled as such.