Under-the-radar AppLovin: mobile game ads, deep learning, and surviving the 92% drawdown

Chapters

1. 0:00 Deals and companies

   AppLovin's quiet rise in mobile gaming ads

   The hosts open by calling Adam Foroughi one of the least-known great founders: AppLovin is an ad platform hiding inside 100,000 mobile games and reportedly outperforming Facebook ads for e-commerce brands, with the potential to print about $6 billion in cash this year. Foroughi then explains that AppLovin was built without early VC funding, forcing it to operate quietly, and that its goofy name hurt recognition. The company is ultimately an advertising business helping mobile game developers monetize, a framing that leads into AppLovin's scale and the broader mobile gaming market.

2. 1:32 Deals and companies

   The $50B mobile gaming ad market

   Foroughi lays out AppLovin's market opportunity: more than a billion adults a day play mobile casual games, and on AppLovin's own platform annual ad spend was $11 billion nearly two years ago; after roughly 60% year-over-year growth, that is now about $20 billion. Since other ad companies also monetize the ecosystem, he estimates total mobile gaming advertising reach about $50 billion a year - matching the size social advertising hit not long ago. Users often watch rewarded ads, and that creates intent; AppLovin has historically used that intent to move players between games but now sees deep learning enabling it to turn the same adult audience into shopper behavior, opening much larger economies.

3. 2:34 AI and tech

   Ads as ML 1.0 and modern discovery

   Hosts ask whether the first wave of internet advertising, exemplified by Google's AdWords and AdSense, seeded technologies that later diffused across the internet. Foroughi responds that advertising was effectively ML 1.0 - the first implementation of deep-learning technologies and a very profitable one - and that the economic value of LLMs is now much larger, but ad models still share research trajectories with recommendation systems. Because ad businesses predict a future outcome and can immediately translate that prediction into revenue, they are a natural commercial home for new ML techniques. He also describes the evolution of ad quality from spam in 2005 to relevant, content-like recommendations on Instagram and inside mobile games, where users engage with playable mini-game previews.

4. 5:15 AI and tech

   Chatbot ads, search intent, and discovery

   The conversation turns to OpenAI's ad product and the future of advertising in a world where users run five or six chatbot queries. Foroughi splits advertising into bottom-funnel search: where a consumer already knows what they want and an LLM can close the loop - and discovery: showing someone something they had no idea existed. He argues search-based AI ads compete with Google but do not create much economic expansion, because the transaction would have happened anyway; discovery ads create new demand and are what make Meta's business powerful. Foroughi also pushes back on the idea that phones are listening to lunch conversations; ads are more likely triggered by trackable searches and browsing, though social graph connections can influence what friends see.

5. 9:58 Markets

   IPO, the 92% drawdown, and $6B buyback

   Hosts trace AppLovin's unusual public-market arc. Foroughi is based in LA with engineering offices in Palo Alto, Beijing, and Singapore. He says the company went public in April2021 at about $28 billion market cap on $600 million of EBITDA, peaked near $40 billion, then fell almost every day in2022 to around $3.8 billion - even though it earned $1 billion in EBITDA that year. Blaming the collapse on poor investor demand and too much selling supply at the IPO, he stopped talking to investors, launched aggressive buyback, retired roughly20-25% of shares with around $6 billion, and rolled a performance stock plan across key employees. He resumed investor meetings in September2023, and the stock re-rated dramatically, rising from $80 to $150 in a week and eventually to $750 per share.

6. 15:42 Policy and regulation

   Privacy rules, Apple, and relevant ads

   A host asks about privacy headwinds from Apple and the EU, conceding that regulators have tightened screws on user data and Zuckerberg had to deal with it. Foroughi says the key is clear regulation: once rules are clear, technology can adapt. He notes that before Apple's changes, advertisers could precisely target iOS users; after forced grouping into cohorts, users began complaining that they were seeing spam and demanded more relevant ads. He argues consumers actually want relevant advertising, especially when they watch a 30-second rewarded ad in a game in exchange for something of monetary value, and that deep learning networks have since adapted to the coarser privacy landscape.

7. 17:48 Deals and companies

   Bought game studios as a data play, then sold

   A host notes Bending Spoons' aggressive acquisition strategy and asks whether AppLovin might become a game studio, which would conflict with its developer partners. Foroughi says AppLovin already bought and sold all its games. The studios were originally a data play: when building its first deep learning model, AppLovin needed training data, but third-party game studios were unwilling to share user data with an ad company. Owning studios let it generate that data; once the model proved successful in market and third-party developers began coming in, AppLovin divested the game assets to avoid competing with partners, keeping itself focused on powering ads.

8. 18:49 AI and tech

   Agentic commerce versus the average shopper

   Hosts ask what advertising looks like in an agentic world, where software agents handle purchases and the interface shifts to glasses or other devices. Foroughi says agents will optimize certain recurring behaviors, like a supplement subscription delivered on time each month, but discovery platforms are not that. He describes AppLovin's audience as the New York Times audience, not the Twitterverse, noting that millions of people still use Yahoo properties every day. The typical shopper wants to window-shop, compare, track packages, and experience the transaction; even if an agent could save20% on a $50 purchase, the dopamine hit from the shopping experience is what they enjoy. Thus he expects agent adoption among advanced users but not mass discovery shopping.

9. 20:23 Deals and companies

   Beating tech giants with focus, China talent, and 84% margins

   Asked how AppLovin competes with Meta, Google, and other ad engines, Foroughi says the company never assumes it has won; it wakes up expecting to get screwed and relies on leanness, subject-matter experts, and speed to outmaneuver giants. On margins, he says AppLovin's EBITDA margin is about84%, attributing it to an algorithmically driven performance model where advertisers buy consumers at an arbitrage and the transaction covers customer acquisition cost immediately. A host pushes on the idea that high margins invite competition, but Foroughi argues complex technology and differentiated data create a moat, analogizing to Anthropic's position in LLMs. He closes by praising his China-based engineers as humble, hardworking, and sharp, saying he is often the dumbest person in the room.