Context

AppLovin CEO Adam Foroughi talks with the All-In hosts about building a $50 billion mobile-gaming ad business almost entirely outside the public eye, surviving a 92% stock collapse after its 2021 IPO, and why he thinks discovery-based advertising (showing someone something they didn't know they wanted) creates genuine economic expansion in a way that closed-loop search advertising doesn't. The episode matters to PMs and operators because Foroughi gives an unusually specific, mechanical account of two things companies rarely explain in public: exactly what he did operationally during a near-total loss of investor confidence, and a clear technical distinction between two categories of advertising value that's directly relevant to anyone building a recommendation or ranking system.

The Big Idea

When AppLovin's stock fell 92% after its IPO because it had the wrong investor base rather than a broken business, Foroughi's response wasn't to keep pitching investors who'd stopped listening, it was to redirect that same effort entirely inward: stop talking to the market, use the company's own cash to aggressively buy back stock, and let a genuinely improved product (a shift from a regression model to deep learning) be the only thing that eventually brought investors back.

The mechanism and result: AppLovin bought back roughly $6 billion of its own stock at depressed prices, retiring 20-25% of shares outstanding, while simultaneously shipping a new advertising model (moving from what Foroughi calls "ML 1.0," a regression model, to "ML 2.0," deep learning) that directly improved advertiser returns. When Foroughi finally re-engaged with investors roughly 18 months later, once the model's performance had already proven out and the market cap had room to grow again, the stock moved from $80 to $150 in a single week purely on the reintroduction of investor attention to a business that had quietly kept improving the entire time.

Key Insights

Stock price, in Foroughi's framing, is a function of investor quality, not just business quality

His specific diagnosis of the 92% collapse: "your price in the markets is determined by the quality of your investors," and AppLovin's IPO timing during the COVID IPO wave meant it went public without attracting the kind of "blue chip" institutional investors who do deep fundamental research, while early private investors and insiders were net sellers. The result was a supply-demand imbalance (heavy insider selling, no informed buying) independent of the business's actual performance that year, AppLovin generated $1 billion in EBITDA in 2022, the same year its market cap fell from a peak near $40 billion to roughly $3.8 billion. The lesson: a stock price collapse can be a liquidity and ownership-composition problem rather than a verdict on the underlying business, and distinguishing the two determines whether the right response is fixing the business or fixing who owns it.

Buying back your own stock is framed explicitly as "becoming your own best investor"

Foroughi's specific reframe of the aggressive buyback decision: rather than treating a depressed stock price purely as a crisis, he treated it as a rare opportunity, since the company was generating substantial free cash flow and no external investor was willing to buy at that valuation. His explicit internal message to the team: stop wasting time pitching investors who "weren't buying our stock" and redirect that same energy toward the company acting as the highest-conviction buyer of its own equity, on the theory that if you have genuine conviction in the business and the market doesn't, that gap is where more of the future upside concentrates for existing shareholders.

Managing internal morale during a public confidence collapse required proactively naming the asymmetry between leadership and staff

Foroughi describes fielding calls from family and friends asking if he was suicidal during the drawdown, and explicitly recognizes that his team faced the same pressure without his ownership stake or context: "they don't have the gravitas that you do, nor the ownership." His specific response wasn't just reassurance, it was a concrete incentive redesign: implementing a performance-based stock plan (a structure typically reserved for CEOs) across key people company-wide, explicitly acknowledging the team's losses while giving them direct, structural upside if the company recovered. This distinguishes his approach from generic "stay strong" messaging: he changed the actual compensation structure to make the recovery scenario personally, financially concrete for the people being asked to endure the crisis.

AppLovin's core advertising thesis draws a specific line between "closed-loop" and "discovery" advertising value

Foroughi's clearest technical distinction: search-style advertising (his example: someone researching dress shoes on Google, or increasingly through an LLM) monetizes a transaction that "would have happened anyway" even without the ad, meaning it captures value rather than creating new economic activity. Discovery advertising, his company's actual business, shows a user something they had no prior intent to buy and didn't know existed, creating a genuinely new transaction rather than redirecting an existing one. His explicit claim: this is why Meta's ad business (and AppLovin's) drives real GDP expansion in a way pure search-intent-capture advertising doesn't, a distinction directly relevant to anyone evaluating an ad or recommendation product's actual economic contribution versus its capture of pre-existing demand.

Regulatory constraints that remove precision targeting can paradoxically increase relevance complaints, not decrease them

Discussing Apple's privacy changes (which limited precise user-level ad targeting), Foroughi describes a specific, counterintuitive outcome: once targeting became coarser, users began complaining about receiving less relevant ads, essentially the opposite of privacy advocates' stated goal. His broader point: "you want the regulations to be clear... so technology can deal with them," arguing that once a rule is unambiguous, well-resourced technology companies adapt their models to it quickly, but the resulting user experience trade-off (less precision, less perceived relevance) isn't always the one regulators or privacy advocates anticipated.

A "we could get replaced tomorrow" posture, not scale, is what he credits for outcompeting Meta and Google in a narrow domain

Asked directly how a company far smaller than Meta or Google competes and wins in advertising, a category those companies have invested in for one to two decades, Foroughi's answer isn't a technical moat claim, it's an operating posture: "we never think we won. We think every day we wake up and we're probably going to get screwed right now and we better work hard." His structural explanation for why this works: a small, deeply specialized team focused narrowly on one specific domain (mobile gaming ad monetization) can move faster than a much larger organization managing many product lines simultaneously, even against vastly larger R&D budgets.

Mental Models & Frameworks

Separate a stock-price collapse into "business problem" versus "ownership composition problem" before reacting

When a company's valuation drops sharply, explicitly diagnose whether the underlying fundamentals (revenue, margin, product performance) have actually deteriorated, or whether the collapse reflects a temporary mismatch between share supply and the quality or conviction of the current buyer base. The correct response differs completely depending on which diagnosis is true, fixing the business itself versus using available capital to bridge the gap until the right investor base returns.

The discovery-versus-closed-loop test for evaluating any recommendation or advertising system's real economic value

Before assuming a recommendation, search, or advertising system creates genuine economic value, ask whether the transaction it enables would have happened anyway through some other channel (closed-loop, capturing existing intent) or whether it surfaces something the user had no prior awareness of or intent to buy (discovery, creating new demand). Systems in the second category contribute more directly to real economic expansion, a distinction worth making explicit when evaluating or pitching any product built around matching users to products, content, or opportunities.

Trade-offs & Nuance

High margins invite a "they must be over-earning" assumption that hasn't materialized as vulnerability

Foroughi acknowledges directly that AppLovin's roughly 84% EBITDA margins invite the classic competitive assumption that a rival could undercut on price and compete the margin away, and he's candid that this hasn't happened despite being a reasonable expectation. His explanation ties back to model and data complexity functioning as a real moat once a system reaches scale and adoption, comparable to how frontier AI labs maintain leads despite intense competition, but he doesn't claim the margin is permanently safe, only that the moat has held longer than the "someone will just undercut you" logic would predict.

Divesting owned game studios shows a deliberate, time-limited use of vertical integration

AppLovin previously owned game studios specifically to solve a data bootstrapping problem (game studios wouldn't share behavioral data with a third-party ad platform, so AppLovin acquired its own studios to generate training data for its first models), then divested them once the resulting model succeeded and external data-sharing partnerships became available. This is a specific, worth-noting pattern: vertical integration used deliberately as a temporary means to solve a specific technical bottleneck (data access), not as a permanent strategic commitment to owning the adjacent business.

Practical Application

Before responding to a market or stakeholder confidence collapse, diagnose which problem you actually have

When facing a sharp loss of external confidence (investor, customer, or partner), explicitly separate whether the underlying product or business has actually degraded from whether the current audience or stakeholder base simply lacks the context or incentive to evaluate it fairly. Redirect effort toward fixing whichever one is actually broken rather than defaulting to more communication effort aimed at an audience that may not be the right one to convince.

Redesign incentive structures, not just messaging, when asking a team to endure a prolonged setback

If your team is absorbing real financial or reputational damage from a setback outside their control (following Foroughi's example of a stock collapse), consider whether a structural incentive change (extending upside-sharing mechanisms typically reserved for leadership) is more credible and more effective than reassurance alone, since it gives people a concrete, personal stake in the recovery scenario you're asking them to believe in.

Classify your own product's value creation as discovery or closed-loop before making an economic-impact claim

When describing the value your product or feature creates, apply Foroughi's distinction directly: does it primarily surface something a user had no prior intent to seek out, or does it capture and redirect a transaction that was going to happen regardless of your product's existence? Claims about broader economic contribution are more defensible when the product genuinely falls into the first category.

Bottom Line

Adam Foroughi's account of AppLovin's 92% drawdown and subsequent run to a $250 billion valuation is a concrete case study in correctly diagnosing whether a confidence collapse is a business problem or an audience problem, then redirecting effort (capital into buybacks, incentive structures into retention, and continued product investment into a genuinely better model) toward fixing the actual constraint rather than continuing to talk to stakeholders who had already stopped listening.