The Quiet Giant's Comeback: How AppLovin Climbed From a 92% Crash to a $50 Billion Gaming Ad Empire

Founded in Palo Alto in 2011, AppLovin stayed unknown partly because its name invited mispronunciation and partly because it grew without early venture backing. In the All-In conversation Adam Foroughi ties that quiet to necessity rather than modesty: when capital is scarce, you build a cash engine before you build a story. A finance-trained founder chose to polish the engine in the workshop instead of unveiling it on a stage. Like a craftsman who sharpens a blade for years before anyone notices, the edge was earned away from spotlights. That long stretch without a need to be seen explains why the company could operate for years outside the headlines.

At its core AppLovin is an advertising company that helps mobile game makers turn attention into money. More than one billion adults open casual puzzle, match and simulation games every day, often as heads of households filling a short gap in a waiting room or on a train. They willingly watch a 30-second clip to earn an extra life or a handful of gems. In that moment the ad is not an interruption but a trade. Think of a free sample stand in a side alley: a small reward at the right moment makes the walk to the next shop feel effortless. For years that walk led from one game to another; the same corridor now leads to a retail shelf.

Scale becomes clear when numbers are stacked. Foroughi disclosed about $11 billion of annual spend running through his own platform roughly two years ago; with around 60% compound growth since then the figure now approaches $20 billion. That is AppLovin inventory alone. Other networks monetize the same in-game attention, so doubling the pool gives roughly a $50 billion yearly gaming ad economy. For comparison the entire social ad market was that size only a few years ago. The broader in-app advertising market exceeded $390 billion in 2024 and heads toward $950 billion by 2033; gaming is its fastest-growing pocket. In short, AppLovin swims in the deepest cove of a huge lake.

The Invisible Market Inside 100,000 Games

The shift is about creating intent that did not exist. In the early phase the system looked at behavior inside game A and suggested game B, getting paid when a player switched. The same signals are now used to surface a product the person never knew they wanted. Picture three stages: 1) collect behavioral traces from rewarded views, level passes and idle moments, 2) let a deep learning model — a multi-layer neural network that discovers patterns across millions of examples — translate those traces into purchase likelihood, 3) show the highest-likelihood product as a short playable preview. A concrete illustration: a cosmetics brand selling lipstick for $20 with $8 cost of goods can pay less than $12 to acquire a buyer through AppLovin and still profit; when the model finds the right person, the brand scales spend. It resembles stumbling upon a market stall and buying a scarf you did not know you needed.

Foroughi repeats a sharp line: advertising was ML 1.0. What Google built with AdWords and AdSense in the 2000s was the first profitable large-scale prediction exercise. Recommendation systems and large language models use different architectures yet follow a parallel learning curve. Many researchers now in language modelling started in ad systems, and techniques travel both ways. Both families predict what comes next — the next word or the next click. In advertising the value of a correct prediction turns into cash instantly. That immediate feedback shortens the experiment loop and sharpens the model every day.

Someone who started in 2005 has watched the tone of ads flip. Early web ads were mostly spam; data existed but the math to make them useful was weak. Facebook's early-2010s pairing of data and engineering turned ads into something that feels like content; today a large share of shopping ideas comes from Instagram feeds. AppLovin shows a similar jump: playable mini-demonstrations appear inside games and people choose to play them. Why? The model filters ten thousand candidates down to three that genuinely fit your taste. Like a good bookseller who puts only three titles on the table based on your reading history rather than the whole shelf, the selection turns 30 seconds from a burden into a tiny game break.

Why Advertising Was AI's First School

The rise of chatbots revived the debate over who gets the ad dollar. Foroughi splits the world in two. At the bottom of the funnel the shopper knows what she wants and researches to close the deal — Google's classic turf, now mirrored by chat windows where someone asks for dress-shoe advice. On the other side is discovery: showing something to a person whose intent is unknown and creating desire from scratch. Facebook and AppLovin live there. The first merely intermediates a transaction that would have happened anyway; the second creates new economic activity. One is hailing a cab you had already booked, the other is spotting a storefront while walking and changing your route. Even if chatbots must be free for 95% of users, growth will be written in discovery, because discovery sells the thrill of the arriving package and the dopamine of unboxing.

The liveliest part of the talk concerned privacy and the feeling of being listened to. Most listeners have had the experience of mentioning a product at lunch and seeing its ad that evening. Foroughi calls the always-on microphone story unrealistic: streaming that much audio, understanding it and rendering a targeted ad for a single impression would be both heavy and expensive. A simpler explanation is overlooked traces — an earlier search, a product page view, then a related conversation, then a relevant ad. After Apple introduced App Tracking Transparency in 2021, precise targeting shrank and cohort grouping grew; when a user opts out of precise targeting, she is placed in a broader bucket and sees less relevant advertising. Paradoxically many users then complained, 'show me more relevant ads, this is junk.' Once rules became clear, technology adapted, and deep networks learned to extract signal from less data. Privacy versus relevance still swings like a pendulum in 2026.

A 92% Fall and a $6 Billion Countermove

The stock market story alone could fill a textbook. AppLovin listed in April 2021 at about $28 billion, briefly touched $40 billion, then slid throughout 2022 to roughly $3.8 billion. Strikingly the business generated $1 billion in EBITDA that same year, so the operation grew while the price shrank. Foroughi links the gap to investor quality: private equity holders and founders were selling, while more than a thousand IPOs flooded the market and blue-chip funds had no bandwidth to research a strangely named company. Supply overwhelmed demand and the valuation multiple compressed from around 50 times EBITDA to below four. As the share price fell from about $115 to $9, cash generation continued independently of the noise outside.

That is when a finance reflex took over. Foroughi describes stopping investor roadshows and telling the team, 'we will become our own best investor.' An aggressive share repurchase programme began, totaling about $6 billion and retiring 20 to 25% of shares outstanding. At the peak the market value of those retired shares exceeded $50 billion. Managing morale mattered as much as managing the balance sheet; phone calls from family members asking if you were okay echoed across the team. The answer was to broaden a performance share plan that usually covers only chief executives to a wider group of key people, with a clear 'us against the world, if we dig in the recovery pays manifold' message. The move was not a gamble but parking capital inside at a deep discount — like buying back your own house at half price during a panic; when prices normalize the leverage multiplies.

The technical engine of the rebound was a model switch. In April 2023 the company moved from a regression-based system to Axon 2.0, a deep learning recommendation engine with larger embedding tables and sequential behavior modelling, driving rapid gains in advertiser returns. When Foroughi returned to investors in New York in September 2023 the stock stood near $80; that week it climbed to $150 and the market value jumped from $28 billion to $55 billion. The next two and a half years stretched the run from $9 to $750, reaching a peak near $250 billion. Financial reports matched the narrative: in the second quarter of 2025 revenue rose 77% to $1.26 billion and net revenue per installation grew 70%. The promise of the model was thus confirmed on the income statement, not merely in a slide deck.

Some criticism focuses on appetite for data. Foroughi notes that when the first deep model needed training data, studios refused to share with a third party, so AppLovin bought studios to seed the model, then sold them once third-party traffic arrived; today it is a pure marketing platform. Questions linger: some analysts call its software kit the most aggressive third-party data collector, and researchers say its mediation encryption still carries enough device signals to re-identify phones across apps. European and U.S. scrutiny has surfaced, with reports of regulatory inquiries. The company's stance is that once rules are clear, technology adapts and deep networks can produce relevant advertising with less personal data. That tension remains the liveliest debate around advertising technology in 2026.

Anatomy and Future of a Profit Machine

The most forward-looking question is whether agents — software helpers acting on your behalf — will take over shopping. Foroughi draws a line: for repeatable subscriptions an agent makes sense and can optimize supplement delivery each month. For a $50 impulse purchase, people enjoy window shopping, comparing and the anticipation of the parcel; telling them an agent would have saved 20% does not replace that dopamine. The average buyer does not behave like the narrow crowd on technology forums; a broad audience still uses Yahoo properties daily and reads general news. How to compete with giants such as Meta and Alphabet? The answer is focus and leanness. Small expert teams spread across Palo Alto, Beijing and Singapore operate with a daily sense that something could break if they relax. That culture joins an 84% EBITDA margin. The formula stays simple: an advertiser arrives, buys a consumer through AppLovin, the consumer pays $20 for lipstick, the brand pays AppLovin less than purchase price minus cost of goods, and scale follows. Because the technology is intricate, differentiated data and continuous innovation form a moat that keeps margins resilient even when rivals offer to work at 60%. When a leader can say he is the least clever person in the room, that moat deepens a little every day.