Adam Foroughi, Applovin CEO: Surviving a 92% Drawdown, Ads as ML 1.0 & the $50B Game Ad Market

Summary

Adam Foroughi frames AppLovin as a mobile gaming monetization company with a much larger hidden opportunity than most people realize: billions of daily players and roughly $50 billion in annual ad spend across the ecosystem. He argues advertising is effectively “ML 1.0,” with recommendation systems and LLMs sharing a common lineage, and says AppLovin moved from regression models to deep learning to improve performance. A major theme is the company’s public-market whiplash: a huge IPO, a collapse to about $3.8 billion market cap, and then a recovery driven in part by a large buyback program. Foroughi distinguishes discovery ads from search ads, says privacy changes made targeting harder but not impossible, and rejects the idea that AppLovin tracks location or microphones. He closes by arguing that AppLovin’s edge comes from being lean, algorithmically focused, and able to move faster than larger competitors.

Notes

• AppLovin is presented as a large ad platform embedded in mobile games, not just a game company.
• Foroughi says AppLovin’s core business is helping mobile game developers monetize their audience.
• He estimates over one billion people a day play mobile casual games.
• He says AppLovin disclosed $11 billion a year of ad spend on its own platform nearly two years earlier.
• He says that figure grew about 60% year over year afterward, implying roughly $20 billion on AppLovin’s own platform.
• He sizes the broader mobile gaming ad ecosystem at roughly $50 billion a year.
• He frames advertising as “ML 1.0,” the first big implementation path for modern AI-like systems.
• He says recommendation systems and LLMs share research lineage, so techniques can transfer both ways.
• He describes the product evolution as a move from regression models to deep learning, which he calls ML 2.0.
• The ad stack is described as performance based: better model quality improves advertiser ROI and revenue scaling.
• He says ads used to be spam, but better data plus better tech made them feel like content.
• He says users engage with mini-games and previews because recommendations are now highly relevant.
• He draws a hard line between bottom-of-funnel search ads and discovery ads.
• He says search ads mostly close transactions that were already intended.
• He argues discovery ads create economic expansion when they introduce an unfamiliar product.
• He says AppLovin’s model is closer to Meta-style discovery than Google-style search.
• Foroughi rejects the idea that AppLovin can track location.
• He says microphone-based ad targeting is not realistic.
• He says social-network relationships can still affect what ads people see.
• He says privacy concerns persist, but ad data collection is broadly constrained at scale.
• He says iOS privacy changes forced advertisers toward broader grouping and less precise targeting.
• He says consumers still dislike spammy ads even after privacy-driven changes.
• He says the company’s 2021 IPO started around a $28 billion market cap.
• He says the stock peaked near a $40 billion market cap before the collapse.
• He says 2022 was a brutal year, with the stock falling every day and the market cap dropping to about $3.8 billion.
• He says AppLovin generated $1 billion of EBITDA in the down year despite the collapse.
• He says investor quality and supply-demand imbalance drove much of the public-market rerating.
• He says the IPO was hurt by too much supply and too little demand from blue-chip investors.
• He says AppLovin responded by becoming its own best investor through aggressive buybacks.
• He says the company repurchased roughly $6 billion of stock and retired 20% to 25% of shares outstanding.
• He says the buyback position was worth over $50 billion at peak.
• He says the company’s stock later recovered from about $80 to $150 in one week after investor outreach resumed.
• He says the valuation moved from about $28 billion to $55 billion in that week.
• He says AppLovin bought game studios originally as a data-training play.
• He says owning game studios helped seed training data for the first deep learning model.
• He says AppLovin later sold the studios after third-party data became available.
• He says agents may help with repetitive subscriptions, but not broad shopping discovery.
• He says the average shopper values browsing and the dopamine hit of shopping more than agent-driven savings.
• He argues the average consumer is unlikely to outsource ordinary shopping to agents.
• He says AppLovin stays lean, focused, and able to move faster than giants.
• He says the company’s strategy is to remain specialized rather than broad-based.
• He says AppLovin’s EBITDA margins are about 84%.
• He describes AppLovin as an arbitrage layer between advertisers and consumers.
• He says the company has few leakage points because it is highly algorithmic and operationally lean.
• He argues complex models plus differentiated data can create durable moats.
• He says scale adoption itself can become a moat that is hard to dislodge.
• He says the China engineering team is humble, hardworking, and sharp.
• He says AppLovin’s engineers in China are among the brightest minds in the world.