I'll pull the data for AppLovin. Let me look up the ticker and load both the market cap history and the financial data needed for TTM PE.Now let me load the daily market cap and quarterly PE ratio data simultaneously with quarterly net income and shares outstanding so I can compute TTM PE.Let me check which metrics are available. Let me try the correct names.Good. Now let me compute the TTM PE and market cap. I need to: 1. Calculate TTM net income (sum of last 4 quarters at each point) 2. Use shares outstanding to get TTM EPS 3. Combine with daily price to get market cap and TTM PEThe early dates have nulls because TTM needs 4 quarters of data. Let me check when data becomes available and trim the null section for cleaner charts.Now let me chart both series.

Loading chart...Now let me also show a summary table of the latest values.Let me also grab the latest values to summarize.Here's the chart for AppLovin (APP) — two panels shown together:

AppLovin (APP) — TTM Market Cap & TTM P/E Ratio

As of 21 Jul 2026:

The chart tracks both metrics from late 2021 to present. Key takeaways:

• Market Cap has seen a dramatic ascent — from roughly $28B in late 2021/2022 to over $145B today, with the steepest climb occurring through 2025 into mid-2026.
• TTM P/E was deeply negative through much of 2022–2023 as the company was unprofitable on a trailing basis. It turned positive in late 2023 and has since compressed from elevated levels to ~37x as net income growth caught up with the surging share price.

Great question — let me dig into AppLovin's revenue drivers and performance segments from their SEC filings.Excellent. Now let me also get the latest 10-K for more detail on the AXON engine and the advertising model.Let me also get the 2025 10-K (if available) and look at the business description in the actual filing for more context.Now I have a complete picture. Here's what's really going on:

AppLovin: Where the Revenue Actually Comes From

AppLovin isn't a consumer-facing product — it's the invisible infrastructure layer underneath mobile gaming and app advertising. You never interact with AppLovin directly, but if you play mobile games or use ad-supported apps, you've almost certainly been exposed to their technology.

The Two-Segment Business

The company has two segments, and the shift between them explains everything:

1. Advertising Segment (was "Software Platform") — now the dominant business

This is where virtually all the growth is. Revenue composition:

2. Apps Segment — the legacy mobile gaming portfolio

These are owned-and-operated mobile games (think casual puzzle/match-3-type titles). Revenue comes from in-app purchases (IAP) from users buying virtual goods, plus in-app advertising (IAA). This segment is being wound down — AppLovin announced the sale of this business in Feb 2025 for $900M.

The Real Story: AXON Changed Everything

AppLovin launched an upgraded version of its AI bidding engine, AXON 2.0, in early 2023. The results were dramatic:

Notice 2023 revenue dropped but margins and profit exploded. That's because they pivoted from running capital-intensive owned games (low margin) toward the pure advertising platform model. By 2025, the Advertising segment was generating 88% gross margins and 76% operating margins — software economics, not gaming economics.

Scale That's Invisible

From the 2024 10-K: - ~1.6 billion daily active users move through apps with AppLovin's SDK embedded — that's the reach of their ad network - AppDiscovery installations grew 50% YoY in 2024, with NRPI (net revenue per install) up 22% — meaning AXON got better at both volume and pricing simultaneously - Just 1,563 employees (plus ~2,000 external contractors) — generating $3.5M+ in revenue per employee

Who Actually Pays AppLovin?

The customers are other mobile app developers — particularly mobile game studios. These are companies like Playrix, King, Supercell, and Zynga — the studios behind Candy Crush, Clash of Clans, etc. They pay AppLovin to acquire new users who will spend money on in-app purchases. AXON's machine learning predicts which users will have high lifetime value, making the ad spend ROI-positive for these studios.

In short: AppLovin is the picks-and-shovels AI advertising engine for the $100B+ mobile gaming ecosystem. You don't see it because you're the product being auctioned, not the customer.

Good question — let me pull the exact revenue recognition details from the filing.Let me also dig into the MD&A narrative to get more color on the actual pricing mechanics.This is the critical passage. Here's the actual answer:

It's Neither Simple CPI Nor Fixed Revenue Share — It's Dynamic

The key sentence from the 10-K (business description):

"Advertisers are charged for advertising dynamically based on the revenue they receive from acquired users, rather than a simple fixed price per impression or per action (click or installation)."

So AppLovin's pricing is a hybrid model — and it's what makes the business so powerful:

How It Actually Works

Advertisers don't pay a fixed cost-per-install. Instead, they tell AppLovin what return target they want (e.g., "I need to make back my ad spend within 7 days via in-app purchases from these users"), and AXON's machine learning dynamically bids on impressions accordingly.

AppLovin's revenue is then effectively a function of: - The predicted lifetime value of each user that AXON is bidding on - What the publisher needs to be paid for the ad slot - The spread between those two — which AppLovin keeps

The accounting confirms this: AppLovin reports advertising revenue on a net basis as an agent — meaning it only books the difference between what the advertiser pays and what it passes through to the publisher. The formula from the filing: (actions × price) − consideration paid to publishers.

Why NRPI (Net Revenue Per Installation) Matters So Much

The metric they report — Net Revenue Per Installation — is the spread AppLovin captures per install. In 2024, NRPI grew 22% while installations grew 50%. That means:

• AXON got better at predicting which users were valuable, so advertisers were willing to bid more
• AppLovin captured more of that value as its own net revenue per install

What This Means Competitively

This is fundamentally different from Meta or Google, which sell fixed-price impressions or clicks. AppLovin's model creates a positive feedback loop: the more installs AXON drives that result in valuable users, the more data it gets, the better its predictions become, the higher the bids from advertisers, and the more net revenue it captures per install. That's why operating margins jumped from negative in 2022 to 76% in 2025 — it's not just scale, it's a fundamentally better pricing model that compounds with data.

Let me pull the details on where AppLovin's ad inventory actually comes from.Here's the clear answer:

Both — But Third-Party Publishers Are the Dominant Source

AppLovin sources ad inventory from two channels:

1. Third-Party Publishers (via MAX SDK) — The Vast Majority

AppLovin's MAX SDK is embedded in thousands of third-party mobile apps. The filing states:

"Approximately 1.6 billion daily active users" are on apps that have AppLovin's SDK installed — and most of those are NOT AppLovin's own apps.

These are other developers' games and apps — the same studios that use AppDiscovery to acquire users also embed the MAX SDK to monetize their own ad inventory. The DAU figure explicitly counts "whether that mobile app is our own or a third party's."

The relationship works like this: - A publisher (e.g., a casual game studio) integrates MAX into their app - When an ad slot becomes available in that publisher's app, MAX runs a real-time auction where multiple demand sources bid simultaneously — including AppLovin's own AppDiscovery but also competitors like Google AdMob, Unity Ads, Meta Audience Network, etc. - MAX picks the highest bidder. If AppDiscovery wins, AppLovin captures the spread. If a competitor outbids, the publisher still gets paid and AppLovin earns a small take-rate on the transaction via MAX.

2. AppLovin's Own Apps Portfolio — Shrinking

Their owned games portfolio (~200+ titles) also carries ad inventory (IAA — In-App Advertising). But this is now the minority piece: in 2024, Apps revenue was $1.49B vs Advertising revenue of $3.22B. And within Apps, only ~32% was IAA — the rest was in-app purchases. On top of that, AppLovin announced the sale of the entire Apps business in Feb 2025 for $900M.

Does AppLovin Use Meta/Google as Supply?

No — the opposite is true. AppLovin doesn't buy inventory from Meta or Google to resell. Instead, MAX competes against them:

• MAX is a mediation platform that sits on the publisher side. When you see an ad in a mobile game, MAX's auction might pit Google AdMob against Meta Audience Network against AppLovin's own AppDiscovery — all bidding simultaneously in microseconds.
• AppLovin describes "most bidding platforms in the industry bidding simultaneously for developers' inventory" through MAX.

Meta and Google are listed as both customers and competitors — they advertise on AppLovin's platform (buying installs for their own apps like Facebook, Instagram, etc.) but they also compete as alternative ad networks that publishers can choose.

The Takeaway

AppLovin's installs come from ads shown across a massive network of third-party apps that have embedded the MAX SDK. AppLovin acts as the exchange: it doesn't own the inventory, it just runs the auction and takes a cut when its own demand (AppDiscovery advertisers) wins. That's why revenue is recognized net — they only book the spread, not the gross media spend. The 1.6 billion DAU reach is the aggregation of all these third-party apps, not AppLovin's own users.

You're asking the right question. Let me dig into what's actually technically different, not the marketing language.Here's the straight answer, cutting through the marketing:

What's Actually Different: Not the AI, It's the Training Data

Ad exchanges and programmatic bidding have existed for 15+ years. Google, Meta, Unity, ironSource — they all have machine learning optimizing bids. AppLovin's AXON engine isn't fundamentally different technology. What's different is what it was trained on.

And the filings admit it — in the 2023 10-K, buried in the strategy section:

"Over the past several years, our Apps have been critical in providing first-party data and audiences for our Software Platform to enable us to test, design, and scale our technologies."

Translation: they used their own 200+ mobile games as a training ground.

The Real Moat Was Owning Both Sides

AppLovin spent 2018-2022 acquiring game studios. These weren't bought for their revenue — they were bought because they gave AppLovin something no other ad network had: closed-loop conversion data.

Here's why that matters:

• Google/Meta can see that someone clicked an ad and installed an app. They can't see what that user did inside the app — whether they spent $0 or $500 on in-app purchases. Post-install behavior is invisible to them.

• Unity/ironSource have some visibility through their game engine / SDK, but they don't own the games. They see aggregate data, not user-level revenue.

• AppLovin's owned games gave it the full picture: which ad impression led to which install, and exactly how much revenue that specific user generated over weeks and months. That's supervised training data at a level of fidelity nobody else had.

The model wasn't trained to optimize for "cheapest install." It was trained to maximize predicted IAP revenue per user. The filing confirms: "Advertisers are charged for advertising dynamically based on the revenue they receive from acquired users."

Apple's Privacy Changes: The Accidental Catalyst

When Apple launched ATT (App Tracking Transparency) in 2021, it killed IDFA-based tracking. Meta's ad performance on iOS collapsed — it famously lost ~$10B in revenue. Third-party attribution became unreliable.

But AppLovin didn't need cross-app tracking in the same way. Its own games provided first-party data. When AXON predicted a user would be valuable, it was relying on patterns learned from its own closed ecosystem, not on tracking pixels across apps.

The 10-K says Apple's changes "have had a relatively muted aggregate impact on our results of operations." That's not marketing — it's a structural advantage. While competitors scrambled to rebuild targeting, AppLovin's model was trained on data that ATT couldn't touch.

The Self-Cannibalization That Worked

Now the critical part: AppLovin is selling the games business for $900M. Why sell the thing that provided the training data?

Because the training is done. The 2023 10-K explicitly says: "Given the recent development of our technology, the current scale of our Software Platform, and the reach of our MAX solution, we believe we can reduce our reliance on the data from our Apps."

Once AXON was trained on enough first-party conversion data, the games became a drag — low margin (~19% Adjusted EBITDA), capital-intensive, and a distraction. By 2024, Advertising segment margins hit 76% while Apps margins were just 19%. Selling the games makes AppLovin a pure ad platform with software economics.

The Risks That Don't Show Up in the Marketing

There are real vulnerabilities:

1. The e-commerce pivot is unproven. They're pushing into non-gaming verticals, but those advertisers don't have the same conversion visibility. Without owned e-commerce apps generating first-party training data, AXON won't have the same edge.

2. Google Privacy Sandbox on Android could eventually do to Android what ATT did to iOS. AppLovin has been more resilient than peers so far, but Android is a big part of the mobile gaming market.

3. Meta and Google are rebuilding. Meta's AI-based Advantage+ campaigns have recovered much of their iOS performance. If their targeting catches up, AppLovin's edge diminishes.

4. Concentration risk. The filings note that Facebook and Google are both major advertising clients — they use AppDiscovery to promote their own apps while also competing directly. They could pull spend at any time.

Bottom Line

The "AI platform" story isn't fake, but it's not about having better algorithms than Google. It's that AppLovin's model was trained on a dataset — user-level install-to-revenue conversion data from owned games — that nobody else had access to. That data advantage, combined with the timing of Apple's privacy changes kneecapping conventional attribution, created a narrow window where AppLovin could outbid everyone profitably. Now they're cashing in by selling the low-margin games business and trying to prove the model works in verticals where they no longer have that training data advantage.

You're right to push on this. Let me find what specific data AppLovin had that an SDK alone wouldn't provide.Here's the real answer — stripped to boardroom format:

What AppLovin Had That No SDK Alone Could Provide

The data gap: what an SDK sees vs what an owner sees

• A third-party SDK (Unity Ads, AdMob, ironSource) embedded in someone else's app sees: impression served → click → install. That's it. Post-install, it's blind. The game developer owns the user purchase data and doesn't share it.

• AppLovin as the game owner saw the full chain: impression → click → install → every IAP that user made over 5-10 months → exactly when they churned. Complete closed loop.

Why that matters for training

AppLovin ran the same user acquisition campaigns any advertiser would on its own platform — spending real money to acquire users for its own games. Then it could observe exactly which users generated what revenue inside games it owned. That created a labeled training dataset:

The specific technical advantage

• EAUL tracking: AppLovin calculated Estimated Average User Life on a per-game, per-user basis — knowing exactly when each paying user went inactive. The filing states EAUL is 5-10 months. An SDK in a third-party app never sees this.

• Consumable vs durable goods tracking: AppLovin categorized every virtual item as consumable or durable per GAAP accounting — meaning it tracked per-user, per-item purchase behavior at granularity no external SDK could access.

• Virtual economy management: AppLovin's studios ran live ops — discounts, promotions, loot box tuning. They could A/B test which game economy changes drove LTV, then feed that signal back into AXON's bid optimization. No external ad network can do this.

• Attribution ground truth: When the filing says "return on advertising spend is measured based on third-party attribution," that's for external advertisers. For its own games, AppLovin didn't need attribution — it had server-side truth on exactly what each user spent. This let AXON validate its predictions against reality, continuously.

The competitive moat in one sentence

Every other ad network trains on "did they install?" AppLovin trained on "did they install AND spend $X over Y months in MY game?" — and it had 200+ games, 1.6M monthly payers, and $51 ARPMAP of ground-truth revenue data to train on.

Why they're selling the games now

The model is trained. The 2023 10-K says it explicitly: "we believe we can reduce our reliance on the data from our Apps." The games served their purpose — they were a $4.1B data acquisition engine disguised as a gaming portfolio. Now they're being sold for $900M because the data is already embedded in AXON's weights, and the 19% margin Apps business drags down the 76% margin Ads business. Selling shrinks revenue but expands margins and multiple.

Let me look at what AppLovin's filings say about advertiser retention, switching costs, and whether there's actually lock-in or just performance.Here's the unvarnished answer:

Why Advertisers Don't Just Shift to Meta/Google

They already use both — simultaneously

The filing states plainly: "Advertisers typically engage with several advertising platforms and networks." No mobile game developer spends 100% of UA budget on one network. They run Meta, Google, AppLovin, Unity, ironSource, TikTok — all at once — and shift budget daily based on which delivers the best ROAS. AppLovin doesn't need exclusivity; it just needs to win its share of the real-time budget allocation.

The product is fundamentally different from what Meta/Google sell

Meta/Google: "We'll show your ad to people who might be interested based on their social graph / search intent / browsing behavior."

AppLovin: "Set your ROAS target. We'll bid on impressions across 1.6B DAU of in-app inventory, and we only get paid when we hit your target."

Meta and Google ads optimize toward clicks or installs. AppLovin optimizes toward post-install revenue. For a game developer who lives and dies by LTV > CPI, that's a fundamentally different value proposition. Meta can't offer it because Meta doesn't see what happens inside the app after install.

Zero switching costs — and that cuts both ways

From the 10-K: "Typically our advertising agreements can be executed in as little as one day and can be terminated for convenience on two days' notice. Substantially all contracts are fully cancellable at any time."

This is the brutal reality: there is no lock-in whatsoever. AppLovin lives or dies daily by algorithm performance. The fact that revenue grew 75% in 2024 and 91% YoY in Q1 2024 with contracts that can be killed in 48 hours is actually the strongest signal that AXON is delivering real, measurable ROAS.

Meta and Google ARE customers, not just competitors

The filing lists "Facebook and Google" as "some of the largest global internet platforms" who are AppLovin advertising clients. Meta and Google run AppDiscovery campaigns to acquire users for their own apps (Facebook, Instagram, YouTube, etc.). This is the "coopetition" dynamic — they compete on one axis but spend money on AppLovin when it delivers.

The catch: no moat, just performance

There is nothing preventing an advertiser from shifting spend tomorrow. The only thing keeping them there is that AppLovin's ROAS is better than the alternatives for a given budget allocation. If Meta's Advantage+ catches up, or if Google's Privacy Sandbox on Android levels the playing field, or if Unity's ironSource integration starts matching AXON's performance — advertisers shift instantly.

The 76% operating margin in 2025 is remarkable, but it also means AppLovin has enormous room for competitors to undercut on price. A rival willing to take 40% margins instead of 76% could offer advertisers better effective rates. The only defense is sustained algorithmic superiority — and the sale of the games business removes the training data advantage that created that superiority in the first place.

You've caught the critical point. Let me re-examine this precisely.You're right. Let me correct the record:

Correction: AppLovin Does NOT See Inside Advertisers' Apps Either

I was wrong earlier. The filing is explicit:

"Return on advertising spend is measured based on third-party attribution."

AppLovin does not have server-side access to an advertiser's IAP data. It relies on MMPs (Mobile Measurement Partners) like its own Adjust — or competitors like AppsFlyer, Branch, Singular — to tell it whether an install generated revenue.

And critically, the filing contains this firewall:

"Adjust's marketing platform is operated by our wholly-owned subsidiary and data generated by Adjust's services is not shared with AppLovin or incorporated into or used to optimize its recommendation engine or other technologies."

So even when an advertiser uses Adjust (which AppLovin owns), that attribution data is walled off from AXON. Adjust operates as an independent SaaS business.

So What Was the Real Advantage of Owning Games?

It wasn't that AppLovin could see into other advertisers' apps. It can't. The advantage was narrower but still real:

• AXON's training data came from AppLovin's own games. AppLovin ran UA campaigns through its own platform → acquired users for its own games → observed exactly how much those users spent inside games it owned. It used this first-party data to train AXON to predict which types of impressions/users correlate with high LTV.

• But after training, AXON doesn't need to see inside advertisers' apps. It bids based on the signals it can see — device type, OS, time of day, which publisher app the impression is in, historical patterns — to predict which impressions will produce high-LTV users. The advertiser's MMP reports back whether AXON's predictions were right, and AXON adjusts.

• This means Meta and Google could theoretically do the same. The difference is they don't have a portfolio of owned games generating labeled training data (install → LTV pairs) at scale. They get post-install revenue signals through MMPs, but that's noisier and delayed compared to AppLovin having server-side ground truth from 200+ games, 1.6M monthly payers, $51 ARPMAP, over years.

The Truth Is Less Magical, More Pragmatic

AppLovin's edge isn't a secret data pipe into advertisers' apps. It's that:

1. They spent ~$4.1B acquiring game studios that generated unique first-party training data
2. That data let AXON predict post-install LTV better than competitors who only had MMP-reported data
3. Apple's ATT crippled cross-app tracking, which hurt competitors more than it hurt AppLovin (because AXON was already trained on patterns that didn't depend on IDFA)
4. The games served their purpose and are now being discarded

Once the model is trained, no competitor can replicate it without equivalent training data — and the window to get that data (pre-ATT, when mobile gaming UA was booming) has closed.

That's exactly right — and it exposes the real weak spot in this whole story. Let me think through what's actually defensible here versus what isn't.

The Brutal Assessment

What's Not Defensible

• Owning games for training data. If this were a durable moat, Meta (which has 3B+ users, unlimited capital, and existing gaming integrations through Facebook Gaming/Instant Games) or Google (which owns Google Play itself, has cloud infrastructure, and could buy any game studio portfolio) would have done it already. The fact that AppLovin is selling its games for $900M — essentially admitting they're worth more as training data embedded in AXON than as operating assets — tells you the market doesn't value game studios as ad-tech training infrastructure.

• The "AI" itself. Every major ad platform has machine learning optimizing bids. Meta's Advantage+, Google's Performance Max, TikTok's algorithm — they all do automated bid optimization toward conversion goals. The claim that AXON's architecture is uniquely superior is unverifiable from public filings. The 10-K discloses 7 patents for advertising solutions and 58 trademarks. Seven patents is not a deep AI moat.

So What Actually Explains the Numbers?

If it's not proprietary data, not defensible AI, and anyone with money could replicate the approach — why are the numbers what they are? A few theories:

1. First-mover + ATT timing luck. AXON 2.0 launched right as Apple's ATT destroyed competitor targeting. For a 12-24 month window, AppLovin was one of the few platforms that could deliver performance on iOS while Meta and others rebuilt. The revenue numbers from 2023-2025 partially reflect competitors being temporarily impaired, not permanent superiority.

2. Advertiser inertia + measurement lag. Mobile UA is an opaque market. Advertisers see ROAS numbers from their MMP and reallocate budget. If AXON works "well enough" and advertisers are busy, they don't constantly re-test every alternative. But this is fragile — any sustained underperformance and budget shifts in days (contracts are 48-hour cancellable).

3. The e-commerce pivot is the real test. AppLovin has no owned e-commerce apps generating training data. The Flip (Humans, Inc.) $50M investment is an attempt to replicate the games data strategy in commerce. If AXON performs as well in e-commerce without first-party training data, the "trained on owned games" narrative was overstated. If it doesn't, the entire post-games growth story collapses.

4. The MAX mediation position. This might be the closest thing to a real moat. MAX is embedded in thousands of apps, running real-time auctions where AppLovin's own demand competes against others. Every auction gives AppLovin data about what competitors are bidding. That visibility into competitor pricing across 1.6B DAU of inventory is hard to replicate — not because of technology, but because publisher SDK integrations are sticky. MAX is like being the exchange that sees every order before it executes.

The Honest Answer to Your Original Question

You were right to push back. There is no structural reason publishers can't shift budget to Meta/Google. They already use both. The only thing keeping money on AppLovin is that AXON currently delivers better ROAS for certain campaign types — and that advantage is contestable.

The stock is pricing in the idea that AXON's superiority is sustainable and transferable to new verticals. The 10-K itself warns: "Our clients do not have long-term advertising commitments with us... contracts are fully cancellable at any time or upon short notice." That's not boilerplate — it's the central risk.

Yes. One hypothesis: LLMs changed creative production costs.

• Pre-LLM: A/B testing 100 ad creatives meant hiring designers, copywriters, video editors. Expensive and slow. Most advertisers ran 5-10 variants.

• Post-LLM (2023+): Generate 1,000 ad variants in hours. Text, images, video scripts — near-zero marginal cost.

• Why this helps AppLovin specifically: AXON's core loop is: serve impression → observe which creative/placement/user combination produces high-LTV installers → optimize bids. More creatives = more signal per campaign = better optimization. The platform that can ingest and test 1,000 variants wins over the one testing 10.

• Why Meta/Google don't benefit equally: Their platforms were already saturated with creative volume. Going from 100 to 1,000 variants doesn't improve their models as much. AppLovin went from limited creative diversity to near-infinite — a step-change in training signal density.

• Timing matches: AXON 2.0 launched early 2023. GPT-4 launched March 2023. AppLovin's revenue accelerated Q2 2023 onward.

Bottom line: If true, this isn't about AI bidding. It's about LLMs making creative production free, which disproportionately benefits platforms with smaller existing creative libraries that optimize toward downstream revenue rather than clicks.

Boardroom Summary: Bull vs Bear

Bull Case

• AXON 2.0 delivered a genuine breakout. Revenue went from flat ($2.8-3.3B, 2021-2023) to $4.7B (2024) to $5.5B (2025) — 70% growth. Q1 2026: $1.84B, +59% YoY. 12 consecutive growth quarters.

• AI pricing, not volume. FY25: install volume +3%, revenue per install +72%. AXON is extracting dramatically more value per impression. 99.3% incremental operating margins.

• E-commerce gaining traction. From zero to >$1B annual run rate in ~18 months. March 2026 consumer spend +25% vs January. April exceeded any Q4 month. Self-serve Axon Ads platform launching publicly June 2026.

• LLM tailwind is real. AI-generated creative (interactive pages, video) removes the #1 barrier to advertiser onboarding. CEO: video output "indistinguishable from human-produced creative at a fraction of the cost." Gaming clients test thousands of variants; e-commerce clients couldn't — until now.

• Financials are extraordinary. 85% EBITDA margins. $4B FCF in 2025. Net cash position. Aggressive buybacks ($2.6B in 2025). 13% effective tax rate via Singapore structure (expires 2028 — but renewable).

Bear Case

• Growth shifted from volume to pricing in one year. Install growth collapsed from +50% to +3%. All growth is now extracting more per install. Diminishing returns ceiling exists. If pricing growth slows below 30%, the $228B valuation is unsupported.

• Muddy Waters: "just another scammy ad-tech company." Three short seller reports (Fuzzy Panda, Culper, Muddy Waters). Core allegations:

— 52% of e-commerce conversions are retargeting, not new customer acquisition. Incrementality only ~25-35%, vs CEO claim of "nearly 100%."

— AppLovin's JavaScript extracts proprietary IDs from Meta, Google, Snap, TikTok, Reddit and stitches them into "Persistent Identity Graphs" (PIGs) — cross-app user fingerprinting that violates platform TOS.

— Shopify event data collected (add-to-cart, checkout initiated) gives AppLovin a "black edge" in ad auctions — it knows which users are about to buy and retargets them.

— ~23% e-commerce advertiser churn in Q1 2025, vs CEO claim of "almost no churn."

— CEO and CTO "lied" about persistent identifiers in their rebuttal blogs, per Muddy Waters follow-up with technical video evidence.

• SEC probe confirmed active and ongoing (Feb 2026). Investigating data-collection practices. Whistleblower complaint filed. Risk of deplatforming by Apple/Google — the 2023 10-K added specific disclosure warning about this exact scenario.

• E-commerce is unproven at scale. Still ~85%+ gaming-dependent. Conversion rate per 1,000 impressions for e-commerce is dramatically lower than gaming (gaming: ~50/1000; e-commerce: undisclosed but "moving toward mid-single digits"). AppLovin admits model is where gaming was "10 quarters ago."

• Moat is contestable. Muddy Waters: if platforms don't deplatform AppLovin, "numerous competitors will start copying APP's techniques because there is little technology involved." No patents protect the fingerprinting approach. Meta's Advantage+ and Google's Performance Max are improving rapidly.

• CEO and insiders selling. Foroughi's blog posts defending the company are aggressive but not reassuring when the SEC is investigating the exact practices he's denying.

The One-Slide Verdict

The LLM angle holds. AI-generated creative lowers advertiser onboarding friction — real and verifiable. But Muddy Waters' technical evidence on fingerprinting suggests the "AI" advantage may be partly old-fashioned user tracking dressed in new language. The SEC investigation is the binary catalyst: if deplatforming happens, the bull case breaks. If it doesn't, the question becomes whether the retargeting-heavy, low-incrementality model retains advertisers once they measure properly.

In plain English — 20 words max each:

What Muddy Waters says AppLovin does:

1. Steals user IDs from Meta, Google, TikTok, Snap, Reddit pixels on advertiser websites.

2. Stitches those stolen IDs together into a single persistent profile per user.

3. Grabs Shopify event data — what's in your cart, whether you started checkout.

4. Uses that data to know exactly which users are about to buy something.

5. Outbids everyone in ad auctions only for those high-intent users.

6. Claims credit for sales that were going to happen anyway.

7. Calls this "AI" — but it's old-school cross-site fingerprinting, illegal under platform rules.

8. Runs the fingerprinting on its own servers, outside Apple/Google's view, to avoid detection.

Why it matters:

1. Apple and Google explicitly ban this. Could get AppLovin deplatformed — apps pulled, SDKs revoked.

2. If not banned, anyone can copy it. Zero tech moat. Just a willingness to break rules.

3. SEC has an active, ongoing investigation. Whistleblower involved. Outcome uncertain.

AppLovin's defense:

1. CEO: "We don't create persistent IDs." CTO: "We handle data properly."

2. Muddy Waters follow-up: video proof showing persistent IDs crossing apps and websites. "They lied."

Sharp question. Here's how — 20 words max each:

The setup:

1. AppLovin's JavaScript pixel sits on advertisers' websites (e.g., a Shopify store), not publishers' apps.

2. That advertiser website also has Meta, Google, TikTok, Reddit pixels loaded — standard marketing stack.

3. Browsers make all pixels on a page visible to each other. AppLovin's pixel can read other pixels' cookies.

4. AppLovin's code literally scrapes Meta's _fbp, Google's _ga, Reddit's user tokens from the page.

5. Muddy Waters found the code doing this — labels each stolen ID as a "key," sends to AppLovin servers.

The Shopify piece:

1. Shopify stores fire events into the browser: "addedtocart," "checkout_started." Standard e-commerce tracking.

2. AppLovin's pixel listens for these events. No hack needed — Shopify pushes them to all pixels on page.

The connection:

1. AppLovin server now has: Meta ID + Google ID + TikTok ID + Reddit ID + "this person just started checkout."

2. Cross-references against its own device graph from MAX SDK (in-app gaming behavior).

3. Same user playing Candy Crush → sees ad for the product they just abandoned in cart → buys → AppLovin claims credit.

It wouldn't. Let me correct the flow:

The two separate moments:

1. User visits Shopify store → AppLovin pixel scrapes Meta/Google IDs + sees "addedtocart" event → sends to AppLovin servers.

2. Later, same user plays a mobile game with MAX SDK → AppLovin recognizes the device → matches to the earlier Shopify profile.

3. AppLovin knows: this device = person who abandoned a cart. Bids aggressively to show them that product's ad inside the game.

The NYT example:

1. NYTimes.com is irrelevant. The ad isn't shown there. It's shown inside a mobile game where MAX runs the auction.

2. The "Shopify data" was captured earlier, on the Shopify store, not served alongside NYT content.

The cross-domain link:

1. The key: AppLovin sees you on the Shopify store (via pixel) AND in the game (via MAX SDK). Same device. Persistent ID links the two moments.

Exactly right. And that's the point:

Shopify doesn't integrate AppLovin. The advertiser does.

1. The advertiser (e.g., a DTC brand) installs AppLovin's pixel on their own Shopify store — just like they install Meta pixel, Google Analytics, etc.

2. AppLovin tells advertisers: "Install our pixel to track conversions from our ads." Standard pitch. Advertiser complies.

3. Once installed, the pixel does what it's told — but also scrapes every other pixel's IDs on the page.

4. The advertiser likely doesn't know AppLovin's pixel is harvesting Meta/Google/TikTok tokens. They just see ROAS reports.

So the chain is:

1. Advertiser voluntarily installs AppLovin pixel → pixel scrapes neighboring pixels → sends data to AppLovin → matched to device graph from MAX → retargeted in mobile games.

The alleged violation:

1. Nothing illegal about the advertiser installing a pixel. The issue is the pixel secretly harvesting other platforms' user IDs — which violates those platforms' TOS.

Because a GA ID isn't just a Nike ID. Here's the key:

A GA ID (Google Analytics cookie) is per-user, not per-site.

1. Same GA cookie value follows you across every site that runs Google Analytics — Nike, Adidas, Allbirds, your local bakery.

2. AppLovin scrapes it once from Nike.com. Now it has your GA ID.

How that becomes useful:

1. Later, you visit another site with an AppLovin pixel (or AppLovin sees you via MAX in a game) → your GA ID is the same → confirmed identity match.

2. AppLovin also has Meta's _fbp from Nike. Meta uses that same cookie across every site with Meta pixel — i.e., most of the internet.

3. AppLovin doesn't need Nike's purchase data. It just needs to know: this human = this GA ID = this Meta ID = this device.

The real play:

1. It's an identity resolution hack. Stitching external IDs creates a cross-platform user graph. Then retargeting becomes trivial: you browse Nike → AppLovin now knows who you are everywhere → shows you Nike ads in games until you buy.

You've landed on the core debate. Here's the counter:

Why it might not be defensible:

1. Advertiser installs pixel for attribution. Pixel does attribution and silently harvests neighboring platform IDs. That dual use isn't disclosed.

2. Meta, Google, Snap TOS explicitly prohibit scraping their pixel cookies. Not "please don't." Contractually banned. Grounds for revoking API/business access.

3. Apple's ATT framework bans fingerprinting — creating persistent user IDs without consent — regardless of where the code runs. Letter of policy, not just spirit.

Why they might get away with it:

1. The harvesting happens on advertiser-owned websites via advertiser-installed pixels. Apple/Google don't control what runs there.

2. Platforms would need to audit third-party pixels on millions of merchant sites. Hard to enforce at scale.

3. As you note, GA4 already supports property-scoped IDs. If Google tightens this, AppLovin's stitch breaks — but that's Google's move to make, not a legal ruling.

The real risk isn't "illegal." It's commercial:

1. Meta/Google can cut off AppLovin's access to their platforms — ad accounts, APIs, SDK integrations — without any court. Cheetah Mobile precedent exists.

2. The SEC probe isn't about criminality. It's about whether AppLovin misled investors about how AXON works. "AI" vs "fingerprinting" is a disclosure question.

Bottom line: Not clearly illegal. But depends entirely on platforms choosing not to enforce their own rules.

You've cut to it. The "deplatforming risk" is overstated. Let me re-frame honestly:

What Meta/Google could actually do:

1. Revoke AppLovin's access to their APIs, SDKs, business tools. But AppLovin doesn't depend on those for its core ad delivery.

2. Ban advertisers who use AppLovin from also using Meta/Google. Commercially absurd — would hurt Meta/Google more than AppLovin.

3. Technically block their pixel cookies from being readable by other pixels. Possible — but breaks legitimate multi-pixel setups across the industry.

What they can't do:

1. Stop AppLovin from running ads inside mobile games. That inventory has nothing to do with Meta/Google.

2. Stop advertisers from installing whatever pixels they want on their own websites.

The real risk isn't deplatforming. It's two things:

1. Advertiser disillusionment. If incrementality is ~25-35% as Muddy Waters claims, not ~100% as CEO claims — advertisers eventually notice. Budget shifts. Already 23% churn observed.

2. Copycats. If no one enforces rules, every ad network does the same thing. AppLovin's "AI advantage" becomes commodity fingerprinting. Margins compress from 85% toward industry norms.

The honest summary: The "existential TOS violation" narrative sells short reports. The actual risk is boring: advertisers getting smarter about measurement, and competitors replicating a technique that has no patent protection.