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

Summary

This All-In interview with Adam Foroughi presents AppLovin as a machine-learning advertising platform built first around mobile-game inventory and now aimed at broader product discovery. Foroughi contrasts intent-capturing search ads with Discovery Advertising, describes advertising as a fast-feedback proving ground for prediction models, and says AppLovin used acquired game studios to solve its initial data cold start before divesting them.

The episode also connects AppLovin’s 2022 public-market collapse to Drawdown Capital Allocation and Alignment: management repurchased shares, extended performance equity to key employees, launched a deep-learning model, and later re-engaged investors as growth accelerated. Its claims about a roughly $50 billion mobile-game ad market, 84% EBITDA margins, targeting, privacy, repurchase value, and valuation recovery are management statements in a friendly interview rather than independently verified findings.

Key Claims

• Foroughi estimates that more than one billion people play casual mobile games daily and that the mobile-game advertising market is roughly $50 billion annually.
• AppLovin aims to move mobile-game advertising from game-to-game acquisition into broader commerce by recommending products users did not already intend to buy.
• Advertising supplies a rapid Performance Advertising Learning Loop because a better prediction can quickly improve advertiser returns, spending, and platform revenue.
• AppLovin acquired game studios to obtain proprietary training data, then sold them after third-party adoption supplied a broader data stream, illustrating Vertical Integration for Data Cold Start.
• Foroughi says AppLovin fell from roughly $40 billion to $3.8 billion in market value, bought about $6 billion of its own stock, retired roughly 20%–25% of shares, and paired the recovery effort with performance equity for key employees.
• The April 2023 deep-learning model is presented as the operating inflection behind faster growth, while focus, differentiated data, automation, and organizational speed are presented as the durable moat.
• Foroughi expects agents to automate repetitive purchases but argues that browsing and product discovery will remain valuable consumer activities.

Key Quotes

“ML 1.0” - Foroughi’s label for the regression-based advertising systems that preceded AppLovin’s deep-learning model.

“ML 2.0” - his label for the deep-learning recommendation system launched in April 2023.

“us against the world” - the internal posture he says management used during the share-price collapse.

Connections

• AppLovin and Adam Foroughi - company and chief executive at the center of the interview.
• All-In, Chamath Palihapitiya, Jason Calacanis, David Sacks, and David Friedberg - show and host context.
• Discovery Advertising and AI Consumer Decision Shaping - demand-creation and recommendation branch.
• Performance Advertising Learning Loop, AI Advertising Targeting, and Recommendation System Productization - model, measurement, data, and productization branch.
• Vertical Integration for Data Cold Start - temporary studio ownership used to obtain initial training data.
• Drawdown Capital Allocation and Alignment and Public Market Communication - buyback, employee-incentive, and investor-reengagement branch.
• Agentic Commerce - qualified by the claim that agents may automate replenishment without eliminating discovery shopping.
• Apple, Meta, and Google - privacy, discovery-advertising, and search-advertising comparison set.

Contradictions

• No settled contradiction found.
• The source’s market-size, user-scale, margin, buyback-value, valuation, targeting, and competitive-moat claims come primarily from AppLovin’s CEO and remain source-scoped rather than independently verified.
• The privacy discussion qualifies rather than overturns privacy-centered pages: the interview emphasizes relevance lost under weaker targeting, while the broader wiki also treats consent, collection, and surveillance incentives as substantive user protections.
• The agentic-commerce discussion narrows strong automation forecasts by separating repetitive replenishment from discovery-oriented shopping; it does not show how quickly either behavior will change.