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

Why does it matter? Because the ad business most investors can't name just explained why Google is vulnerable and Meta isn't.

AppLovin IPO'd at $28 billion, crashed to $3.8 billion while printing $1 billion in EBITDA, bought back $6 billion of its own stock at sub-4x, and watched that position peak at over $50 billion. CEO Adam Foroughi spent that entire stretch refusing to talk to investors — and the silence turned out to be the right move.

The framework he laid out redraws the competitive map for the entire advertising industry:

• Discovery ads create net-new GDP; search ads reroute transactions already in motion — which is why LLMs threaten Google but are structurally irrelevant to Meta and AppLovin • Mobile gaming is already a $50 billion annual ad market, social-scale, and most institutional investors still haven't modeled it • AppLovin's recovery started with a quiet model upgrade in April 2023; investors found out five months later, and market cap doubled in a week • Most shoppers aren't optimizing for efficiency — the dopamine hit of finding something is what they're actually buying

Google Search facilitates transactions. Meta and AppLovin create them — and that's why LLMs threaten one but not the others.

The transaction Google helps you complete was going to happen anyway. Foroughi is direct: "The transaction via search or LM was going to happen anyways. If the LM didn't exist and Google ads had never come to existence, but Google search existed, that transaction, the closed loop would have [happened]." Swap Google Ads for ChatGPT and the economic outcome barely changes.

Discovery is different. When AppLovin surfaces a game ad between levels, or Meta serves you a product you've never heard of, the purchase that follows is genuinely new economic activity. "When you show a consumer an ad for something that they had no idea existed, they didn't know they needed to buy... you create economic expansion." That's GDP creation, not GDP rerouting.

The pricing consequences are real. Google's ad model is a toll on existing intent — and a sufficiently good LLM can collect that toll just as efficiently, cheaper, with no ads at all. Meta and AppLovin are in the business of manufacturing intent from scratch, which requires a recommendation engine that knows you better than you know yourself. "This is what makes Meta so amazing in their ad business and what we aspire to do."

There's no LLM shortcut for that. The two categories deserve different multiples, and the market hasn't fully priced the distinction.

A 92% drawdown against $1B in EBITDA isn't a crisis — it's a $50B capital allocation gift, if you stop talking to investors and start buying your own stock.

The stock fell literally every single day in 2022. $28 billion at IPO, $40 billion at peak, then all the way down to $3.8 billion — sub-4x EBITDA — while the company printed $1 billion in cash. Foroughi's response: "I'm not going to talk to investors at all anymore. They're not buying our stock. It's a waste of time. But guess what? We generated a ton of cash. Let's start buying our own stock. Let's become our best investor."

They bought roughly $6 billion of stock, retiring 20 to 25% of shares outstanding. "At peak that six billion was worth over 50 billion."

The internal management challenge was just as real. Employees were fielding calls from family asking if everything was okay. Foroughi's answer was a performance stock plan spread across key people — not just himself — tying their upside directly to the recovery. "Us against the world." The psychology mattered as much as the capital allocation math.

Most CEOs in that position spend the year on investor roadshows, issuing reassuring press releases, trying to talk the stock up. The lesson Foroughi drew is less comfortable: silence plus performance compounds faster than communication.

AppLovin's stock doubled in a single week in September 2023 — not because anything changed, but because investors finally found out what had happened five months earlier.

April 2023: AppLovin upgrades from a regression model to a deep learning model. Revenue accelerates. The stock starts recovering. Nobody outside the company knows why.

September 2023: Foroughi goes to New York. First investor meetings in roughly two years. "In that week, the stock went from 80 to 150. And I think it was like 28 billion to 55 billion from you being in New York." He describes sitting in those meetings watching people quietly trade on their phones mid-presentation: "These people are literally calling their friends in the room going bye bye bye bye bye."

The business hadn't changed. The information availability had.

The pattern is worth isolating. The biggest single-week re-ratings in public markets are often driven by information gaps closing, not by fundamental shifts in the business. A company can compound quietly for months before the narrative catches up. AppLovin went from $9 to $750 a share in two and a half years — $3.8 billion to $250 billion in market cap — and a substantial portion of that move was just the market belatedly learning what had already happened. Finding companies with undiscovered model upgrades is a repeatable edge.

There's a $50 billion annual advertising market hiding inside mobile games — and it's already at the same scale social media was at its inflection point.

Over a billion people play mobile casual games every day. Adults, heads of households — and they watch ads, often voluntarily, to earn in-game rewards.

AppLovin's own platform ran $11 billion in annual ad spend as of early 2024. Since then, the company has grown roughly 60% year-over-year, putting its platform alone at approximately $20 billion. Double that for the rest of the ecosystem and you arrive at "probably about $50 billion of advertising being spent every single year in this mobile gaming ecosystem."

Foroughi's reference point is deliberate: "It was not very long ago that social was a $50 billion opportunity. Space is growing really quickly."

Most institutional investors are still benchmarking AppLovin against niche ad-tech players — The Trade Desk comps, programmatic display conversations. The actual market is social-scale, growing fast, and serviced by a dominant player running 84% EBITDA margins. That's a different analysis entirely, and the gap between the analysis investors are running and the one that fits the reality is where the opportunity lives.

AI agents won't replace discovery shopping — because for most people, finding something they didn't know they wanted is the whole point.

The bear case on discovery-ad platforms assumes that AI agents will eventually optimize all consumer purchasing — set your preferences, let the agent find the best option, eliminate the messy browsing-and-discovering process entirely. Foroughi finds this argument unconvincing, and not on technical grounds.

"The typical shopper is not the person who's deep into agents and sitting on Twitter and adopting the latest technology." His actual audience looks more like the New York Times readership than a tech conference crowd. "There's still a ton of people using Yahoo properties every single day."

More pointedly: even if an agent could optimize a $50 purchase and save 20%, that misses what shopping is actually for. "I don't think that matters on a $50 transaction because the dopamine hit from going through it is what they enjoy." The window shopping, the comparison, the tracking of a package — that's not friction to be optimized away. It's the product.

"I think we really overindex on the Twitter verse and forget that the average shopper is not that."

The behavioral moat isn't demographic inertia — it's that the experiential value of discovery shopping is structurally irreducible. No agent arbitrages away the pleasure of finding something you didn't know you needed.

AppLovin bought game studios to bootstrap ML training data — the moment it worked, they sold everything.

The market read AppLovin's studio acquisitions as vertical integration — a company hedging into content. The actual logic was narrower: "We bought them originally as a data play." Building a deep learning model requires training data, and game developers don't share data with third-party ad networks. So AppLovin bought its own studios, seeded the training data internally, built a model that performed in market, and once outside developers started coming in on their own, divested the studios entirely.

The acquisition was a temporary scaffold to bootstrap ML without partner cooperation. The exit was the tell — not a strategic pivot toward gaming, but confirmation that the data problem was solved and the scaffold was no longer load-bearing.

This pattern matters beyond AppLovin. When an AI-driven company acquires content or distribution assets, the first question should be whether the play is data bootstrapping or genuine strategic expansion. The two have very different valuation implications. AppLovin answered that question explicitly by selling.

Advertising was the original deep learning lab — and the research lineage runs straight to today's frontier models.

"Advertising is like ML 1.0 — really was the first implementation of all these technologies that now are driving AI today." The genealogy runs deep: many researchers now building large language models started their careers studying advertising recommendation systems. Techniques port in both directions, and the two fields share more architectural DNA than most AI investors appreciate.

The practical edge for ad businesses is the feedback loop. "When you build a model, you're predicting a future outcome... you can translate the value of that prediction immediately." An advertiser knows within days whether a model change worked. That tight reward signal has been compounding ad-system ML for twenty years — long before "deep learning" was a mainstream term.

The moat, as Foroughi frames it, is academic talent and differentiated training data, not distribution scale. Which explains how a lean company competes against Google and Meta and wins in a specific domain: "If you're very focused, you remain lean and you can just move faster than them." Ad-tech ML talent is a leading indicator for frontier AI capability — the lineage runs deeper than the category label suggests.

The reclassification is already underway — and most institutional investors are still running the wrong analysis.

Internet advertising isn't one category. It's two. Platforms that reroute existing demand will face direct LLM substitution. Platforms that manufacture new demand have a structurally different moat — and no chatbot threatens it.

Foroughi has built his career around finding the gap between what the market believes and what's actually true. In 2022, that gap was a $50 billion buyback position hiding inside a $3.8 billion market cap. Right now, the gap might be the category itself — a social-scale advertising market that most institutional investors haven't yet put on the map.

The companies that create demand don't just facilitate GDP. They expand it.

---

Topics: advertising, mobile gaming, deep learning, AppLovin, capital allocation, buybacks, ad-tech, discovery advertising, agentic commerce, ML, IPO, public markets, e-commerce, Meta, Google