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

概览

AppLovin CEO Adam Foroughi explains how the company built a large advertising platform inside the mobile-gaming ecosystem while attracting relatively little mainstream attention. He estimates that roughly $50 billion is spent annually on advertising across mobile games, where more than a billion people play casual games each day.

The central business argument is that deep-learning recommendation systems can turn mobile-game ad inventory into a broader product-discovery channel. Unlike search advertising, which captures existing purchase intent, AppLovin aims to create new intent by showing consumers products they did not already know they wanted.

Foroughi also recounts AppLovin’s public-market collapse and recovery. After its valuation fell from approximately $40 billion to $3.8 billion, the company used its cash flow to repurchase shares, introduced performance-based equity incentives, and launched a deep-learning advertising model that accelerated growth.

The discussion closes with privacy regulation, AI agents, operating discipline, margins, technological moats, and AppLovin’s international engineering organization. Foroughi attributes the company’s position to focus, speed, differentiated data, automation, and a culture that assumes it must keep fighting rather than believing it has already won.

分段落总结

[00:49] What AppLovin Does and the Scale of Mobile Gaming

[事实] AppLovin is an advertising company that helps mobile-game developers monetize their games.

[事实] Foroughi says more than one billion people play mobile casual games each day, with many players being adults and heads of households.

[事实] AppLovin disclosed approximately $11 billion in annual ad spend on its platform nearly two years earlier and subsequently grew by roughly 60% year over year.

[推测] Using AppLovin’s growth and the presence of competing ad platforms, Foroughi estimates that annual advertising expenditure across the mobile-gaming ecosystem is about $50 billion.

[02:24] Turning Game Advertising into a Commerce Channel

[事实] Many mobile-game users watch ads in exchange for rewards, creating opportunities to influence their next action.

[事实] AppLovin historically used its recommendations to move users from one game to another.

[事实] The company now believes stronger deep-learning models can recommend shopping experiences to the same audience and connect mobile-game inventory with larger commercial markets.

[推测] AppLovin’s long-term opportunity depends on proving that game advertising can generate purchases outside gaming, not merely redistribute players among games.

[03:07] Advertising as “ML 1.0”

[事实] Foroughi describes advertising as an early commercial implementation of machine learning technologies that now underpin modern AI.

[事实] Recommendation systems and large language models are structured differently, but research and talent can transfer between the two fields.

[事实] Advertising models predict future actions, and the economic value of those predictions can be measured almost immediately.

[推测] Advertising is especially useful as a proving ground for machine learning because model improvements produce rapid, observable financial feedback.

[04:42] How Ads Became More Like Content

[事实] Foroughi says online ads around 2005 were largely spam because the available technology was not powerful enough to make them relevant.

[事实] Facebook demonstrated that extensive data combined with better technology could generate highly relevant advertising.

[事实] He argues that Instagram recommendations now influence much consumer shopping and that playable game ads can function like engaging content.

[推测] As recommendation quality improves, the distinction between an advertisement and a useful content recommendation becomes less obvious to users.

[05:29] Search Intent Versus Discovery Advertising

[事实] Foroughi divides advertising into bottom-of-funnel advertising, where consumers already know what they want, and discovery advertising, which creates new purchase intent.

[事实] He expects advertising inside large language model products to compete primarily with Google Search because both help users research and complete intended transactions.

[事实] AppLovin and Meta instead try to recommend products that consumers did not previously know about or intend to purchase.

[推测] Foroughi views discovery advertising as more economically additive than search advertising because it can create transactions that otherwise would not have occurred.

[07:39] Why Highly Relevant Ads Can Feel “Creepy”

[事实] Foroughi rejects the idea that advertising platforms need to keep microphones active and parse conversations to explain unexpectedly relevant ads.

[事实] He says users may forget earlier trackable actions—such as searches, browsing, or product research—that later influence the ads they receive.

[事实] AppLovin does not track precise location, according to Foroughi, although social platforms may use relationship data to shape advertising experiences.

[事实] He argues that more relevant digital advertising contributes economic value and that improved advertising technology can support GDP growth.

[10:06] The IPO, the 92% Drawdown, and Investor Supply-Demand

[事实] AppLovin went public in April 2021 at an approximately $28 billion valuation, briefly reached around $40 billion, and then fell to roughly $3.8 billion in 2022.

[事实] During the year in which its market capitalization fell to $3.8 billion, the company generated about $1 billion in EBITDA.

[事实] Foroughi attributes the decline partly to selling pressure from existing holders, limited demand from blue-chip investors, and the unusually large number of companies going public during the pandemic period.

[事实] The company’s valuation multiple fell from a high level to below four times the metric Foroughi referenced.

[11:56] Becoming the Company’s Best Investor

[事实] Foroughi stopped prioritizing investor outreach when investors were unwilling to buy AppLovin shares and redirected attention toward repurchasing the company’s stock.

[事实] AppLovin subsequently bought approximately $6 billion of its own shares and retired roughly 20%–25% of the outstanding share count.

[事实] At its peak value, the repurchased stock was worth more than $50 billion, according to Foroughi.

[推测] The buyback converted a crisis of market confidence into a major capital-allocation opportunity, provided management’s assessment of the underlying business was correct.

[12:46] Maintaining Culture During the Collapse

[事实] Foroughi says friends and family asked whether he was suicidal as the share price collapsed, illustrating the psychological intensity of the downturn.

[事实] He recognized that employees were receiving similarly alarming reactions from their own families without having his level of ownership or authority.

[事实] Management promoted an “us against the world” mentality and extended a performance stock plan to key employees, tying their upside to the company’s recovery.

[推测] The equity plan helped turn public-market volatility from a retention threat into a renewed incentive for employees to remain committed.

[13:43] From Regression Models to Deep Learning

[事实] AppLovin moved from what Foroughi calls “ML 1.0,” based on regression, to an “ML 2.0” deep-learning model.

[事实] Because advertisers buy measurable performance, better recommendations improve advertiser returns and encourage customers to spend more on the platform.

[事实] The new model launched in April 2023, after which the business began growing rapidly.

[事实] When Foroughi resumed investor meetings in New York, the share price rose from roughly $80 to $150 in one week, while market capitalization increased from about $28 billion to $55 billion.

[事实] He says the stock ultimately moved from approximately $9 to $750 within two and a half years, and the company briefly reached a valuation near $250 billion.

[推测] AppLovin’s experience demonstrates how operational improvements and renewed investor awareness can amplify one another in public markets.

[15:59] Privacy Rules and Advertising Relevance

[事实] Foroughi argues that clear privacy rules are important because technology companies can adapt once the permitted boundaries are defined.

[事实] When precise targeting is unavailable, advertisers group users more broadly and may serve less relevant ads.

[事实] He says AppLovin received complaints from users asking for more relevant advertising after Apple’s privacy changes reduced targeting precision.

[事实] Users watching a 30-second ad to receive an in-game reward are exchanging attention for something with monetary value and may prefer relevant content during that exchange.

[推测] The discussion frames privacy and relevance as a tradeoff: limiting personal data can protect users while also reducing the quality of the advertisements that subsidize free products.

[17:32] Why AppLovin Bought and Then Sold Game Studios

[事实] AppLovin acquired game studios because it needed proprietary data to train its first deep-learning model, while independent studios were reluctant to share their data with a third party.

[事实] The acquired studios supplied the initial training data that helped AppLovin build a successful model.

[事实] Once the model gained traction and third-party customers joined the platform, AppLovin divested the game businesses.

[推测] The studio acquisitions functioned as temporary infrastructure for solving a cold-start data problem rather than as a permanent strategy of vertical integration.

[18:25] AI Agents Will Not Eliminate Shopping Behavior

[事实] Foroughi expects consumers to use agents for repetitive purchases, such as optimizing and delivering recurring supplement subscriptions.

[事实] He distinguishes such automation from discovery-oriented shopping, in which consumers enjoy browsing, comparing products, completing transactions, and tracking deliveries.

[事实] He argues that technology-focused social-media communities overstate how quickly the average shopper will adopt agentic commerce.

[推测] Even if agents automate routine purchasing, advertising may retain its role in product discovery because shopping itself provides entertainment and emotional reward.

[19:59] How a Focused Company Competes with Meta and Google

[事实] Foroughi says AppLovin never assumes it has won; employees approach each day as though the company remains vulnerable and must continue working hard.

[事实] He describes the organization as lean and staffed with specialists focused on converting mobile-game engagement into transactional behavior.

[事实] He believes a smaller company can challenge much larger competitors by staying narrowly focused and moving faster.

[推测] AppLovin’s advantage is presented less as a single technical breakthrough than as the combination of specialization, urgency, organizational speed, and accumulated expertise.

[21:04] An Automated Business with 84% EBITDA Margins

[事实] Foroughi states that AppLovin’s EBITDA margin is approximately 84%.

[事实] Advertisers acquire customers through the platform and scale their spending when customer transactions immediately cover acquisition costs after accounting for cost of goods sold.

[事实] AppLovin does not seek to become the advertiser itself; it aims to power the connection between advertisers and consumers.

[事实] Foroughi attributes the high margins to a lean organization, algorithmic decision-making, and extensive automation.

[推测] The high margin invites concerns that AppLovin may be over-earning, but it also indicates that software and models—not labor-intensive services—perform much of the platform’s work.

[22:20] Why the Margin and Technology Moat Have Persisted

[事实] The hosts note that competitors have not yet competed away AppLovin’s margins despite the apparent incentive to accept lower profitability.

[事实] Foroughi says the underlying technologies are complex and that innovation combined with differentiated data can create a durable advantage.

[事实] He argues that once a strong model reaches scale and gains adoption across a large community, it becomes difficult for competitors to overcome.

[推测] AppLovin’s moat appears to rely on a feedback loop in which scale supplies data, data improves the model, and better performance attracts more advertisers and publishers.

[23:02] Global Engineering Talent and Leadership Humility

[事实] AppLovin’s engineering offices are located in Palo Alto, Beijing, and Singapore.

[事实] Foroughi praises Chinese colleagues as humble, hardworking, and highly capable.

[事实] He says one of his founding goals was to work with exceptional people and solve problems together.

[事实] He feels energized when he enters a room and believes he may be the least intelligent person there.

[推测] His closing remarks suggest that recruiting stronger specialists—and being comfortable with their expertise—is central to AppLovin’s operating culture.

播客点评/总结

The episode is most valuable as a compact case study in the intersection of machine learning, digital advertising, and public-market capital allocation. Foroughi connects technical model improvements to advertiser returns, company growth, margins, and valuation without turning the conversation into a purely technical discussion.

Its strongest section is the account of AppLovin’s 92% drawdown. The combination of aggressive repurchases, broad performance incentives, continued operating growth, and eventual investor re-engagement offers a concrete example of managing both financial and organizational stress.

[推测] The episode’s principal limitation is that most claims about targeting, privacy, competitive advantage, and market size come from AppLovin’s CEO and are not independently challenged or verified within the discussion. The extraordinary valuation recovery and margin profile receive admiration, but comparatively little examination of downside risks, attribution, or competitive responses.

[推测] The conversation is particularly suitable for founders, technology investors, advertising professionals, and product leaders interested in recommendation systems, marketplace economics, corporate resilience, and the strategic difference between capturing existing demand and creating new consumer intent.