Down 92% and Bleeding Out, He Bet Everything on a Total Rebuild

Source: 20VC with Harry Stebbings | Published: 2026-04-27T14:00:38Z

When AppLovin's market cap cratered from $40B to under $4B, Foroughi went all-in—buying back shares with every dollar and piling on debt while rebuilding the entire tech stack from scratch. Two and a half years later, the stock surged from $9 to $750, an 80x return.

He made a decision that seemed extraordinarily risky at the time—tear down the entire ad recommendation system the company had built its business on and rebuild it from scratch using a next-generation machine learning architecture.

How You Come Back From a 92% Drop

Every day he checked the market, it was down. A 92% drop means you need a 10x gain just to get back to where you started. Foroughi says he absolutely checks the stock price—"I don't really trust CEOs who say they don't look at their stock price."

The hardest part wasn't the numbers themselves—it was the erosion of confidence. Investors were dumping shares, the team was watching and waiting, his family was worried. "Every day, everyone around you is telling you your company is going down. It's easy to start doubting yourself: Did I do something wrong? Is the crowd smarter than me?"

He made several critical moves. First, he announced they were abandoning the old recommendation system and completely rebuilding the tech stack around a cutting-edge machine learning architecture—later named Axon 2.0. Second, he cut team members who were clinging to the old system. Third, he used all the company's cash plus debt financing to buy back shares that were being panic-sold.

The buyback strategy was deliberate. Rather than purchasing shares on the open market like most companies do, he went directly to shareholders who were inevitably going to sell—early investors from the private equity era, former co-founders—and negotiated one-on-one, eliminating selling pressure at the source. At today's valuation, that buyback accounts for roughly a third of the company's value—about $50 billion.

The Year of Triple-Digit Growth—When He Cut Half the Staff

After Axon 2.0 launched in April 2023, AppLovin's revenue began growing at triple-digit rates. But around 2024, Foroughi did something that baffled many: at the peak of the company's performance, he cut 40% to 50% of staff across most departments.

The logic: If you were building this company from zero today, knowing what technology is now available, what would the team look like? Then jump straight to that end state.

He gave specific examples. HR went from seventy or eighty people down to about fifteen—all individual contributors doing the actual work. The creative production team was drastically reduced because he judged AI would soon automate most design work. The engineering team was filtered down to top engineers who could achieve 10x or even 100x output using AI tools, while those who could only manage a 2x improvement were let go.

When asked whether cutting early was risky given the market wasn't ready, his answer was clear—rather than keeping people in roles heading toward a dead end, dragging down morale and slowing the pace, it's better to offer a generous severance package and let them find more suitable opportunities. A-players don't want to be distracted by unhappy colleagues.

"We Don't Have a Product Team"

AppLovin's core advertising business runs with about 400 people and generates the vast majority of the company's profit. The executive team has only four C-level roles: CEO, CTO, CFO, and General Counsel. No CRO, no COO, no CMO, no CHRO.

Foroughi doesn't do one-on-ones. Doesn't do performance reviews. If he's unhappy with someone's work, he says so in real time over chat. If he's satisfied, "they don't need to be told—they know I respect them."

Even more radical: the company has no product managers. Engineers are expected to also be product managers—understand business KPIs, know what to build, then use AI tools to accelerate execution. "You don't need to write code, but you must be able to review code. First, you need to know what the business needs and how to measure it."

This structure has an added benefit in the AI era: all communication happens on Slack or recorded video calls, so new hires can ask Claude to summarize "what Adam has cared about over the past quarter" and find top-performing sales reps' call recordings to learn from. "Way more effective than passively sitting through 'here's what you need to know' training."

80–90% of Code Is AI-Generated—But That Number Is Meaningless

Databricks recently announced that 50% of their code is AI-generated. Foroughi says AppLovin's ratio is 80% to 90%—but immediately added that the metric itself is misleading.

"If all you're chasing is the percentage of tokens consumed, you'll produce a mountain of garbage code. Your LLM bill will be high, but the business won't move an inch."

The right way to measure it, he argues: does the token cost invested get covered by the revenue the resulting code generates? AppLovin's business model is naturally suited for this calculation—every improvement to the ad recommendation model directly shows up in accuracy and revenue growth, and the team knows exactly which KPI they're optimizing.

This leads to his take on the hot topic of "token budgets": if you give teams a token usage quota and set up leaderboards for who uses the most, people will produce valueless junk to game the rankings. "Token budgets are no different from headcount metrics—until you figure out efficiency, you're just burning money."

The SaaS Death Spiral

Foroughi is bearish on the outlook for traditional enterprise SaaS companies. He believes the current valuation decline is not only justified—it may not be over yet.

The logic chain goes like this: frontier models' capabilities are advancing rapidly, making the long-term growth prospects for many traditional SaaS companies uncertain. Investors compress terminal value estimates, stock prices fall. Once stock prices fall, equity compensation that used to represent 3% of market cap suddenly becomes 10% dilution. High dilution drives talent attrition and further investor withdrawal, creating a vicious cycle.

"These companies won't disappear entirely, because once an enterprise has embedded a piece of software, they typically don't switch. But the growth opportunity may already be gone."

The Limits of Founder Mode

Paul Graham's "founder mode" concept is popular in Silicon Valley, but Foroughi sees it as an overcorrection to organizational bloat—when companies have too many layers and too much process, founders have to seize back control.

But once you've streamlined down to a high-caliber team of executors, continuing to micromanage is a different problem entirely. "Everyone around you is more expert in their domain than you are. If you don't let go at that point, what are you even doing?"

His own journey to letting go wasn't easy. The turning point was—again—2022. When new CTO Giovanni Ge joined, he started relentlessly asking "why"—why does this process exist? Why does this role exist? Why does this management layer exist? Foroughi was reminded of something he instinctively knew as a first-time founder but had forgotten over a decade of scaling. Giovanni Ge straight-up took over the product function Foroughi had managed for ten years. "It was great. I could sit in the passenger seat and watch the team execute without getting buried in the details."

The Bet Behind an $83 Million Annual Salary

In 2023, Foroughi's total compensation was $83 million—the eighth-highest among U.S. CEOs. But to understand that number, he says, you have to go back to 2022.

Since the company went public, he had never taken any compensation—only the minimum required to qualify for benefits. When the stock hit rock bottom in 2022, he requested compensation for the first time—but with a condition: he wouldn't receive a single dollar unless the stock price climbed from $9 back above $38–40. After that, there were five or six more tiers, all the way up to the $80 IPO price. It was a pay structure entirely aligned with investor interests.

"If you believe founders shouldn't receive further incentives because they took enormous risks in the early years, that's flawed logic. You have to give people who pursue big returns the ongoing possibility of big returns—otherwise they'll go do something else."

Short Sellers and the Cost of Having No Brand

From $9 to $750—an 80x-plus gain in two and a half years, from under $4 billion to roughly $250 billion in market cap. That kind of rally naturally invites aggressive attacks from short sellers.

Foroughi believes the short-selling mechanism has a structural asymmetry: short sellers can build their positions first, then publish sensational reports that open with "by the time you read this, we've likely already covered our position." They face no accountability for the accuracy of their reports. Meanwhile, a public company CEO is regulated by the SEC—every word must be precise.

But he also acknowledges the attacks exposed a real problem: the company had been heads-down executing for years and never seriously invested in brand communication. "When people don't understand what you do, you become a target." The short-seller episode became a catalyst, forcing the company to start telling its own story.

"Many things we're able to do just seem unreasonable to others. In a world where things look unreasonable, people assume you're cheating."

You Don't Need a General AI Platform to Reach a Trillion

Someone asked whether AppLovin needs to build a social network to become a trillion-dollar company. Foroughi's answer: no. He did the math—if the company's annual free cash flow (minus stock-based compensation) reaches $30–35 billion, that roughly gets you to a trillion-dollar market cap.

Three paths to get there. First, keep deepening the existing mobile gaming ad market—over 1 billion daily active users, a massive pool of adult and household decision-makers. Second, bring the mobile performance advertising model to connected TV (CTV), enabling small and medium businesses to run performance ads on television. Third, explore new applications for recommendation system technology—not because it has to succeed, but also to attract top-tier research talent.

On whether they need to build their own large language model, his position is clear: recommendation systems are a distinct model domain—you can't simply feed user data into a general-purpose LLM and have it recommend the next ad. This is fundamentally different from building an application layer on top of an LLM. "If you're a company building an interface layer on top of large language models, you'd better build a moat fast."

The Person Changed by Being CEO

Foroughi admits that during the company's high-growth years, he was virtually absent from his family's life. "I was physically there, but my mind was always on the business. My memories of my kids growing up are a blur."

The 2022 low point actually pushed him to change. He took up surfing—because when you surf, you have to put down your phone and completely disconnect. He started giving himself 10 minutes of full presence, spending time with just one or two kids at a time, but being 100% focused for those 10 minutes. He cut out his eight cups of coffee a day and started paying attention to his health.

"If I can't even maintain my own stability and health, how can I possibly have the mental clarity to run a public company for the next ten or twenty years?"

When asked what uncomfortable truth comes with trying to be both the best founder and the best parent, his answer was: The moment you decide to be exceptional at one thing, you're deprioritizing everything else. It requires understanding from your family and a balance that's extraordinarily hard to achieve. He says he's still figuring it out.

Every Morning I Wake Up Thinking: Is Today the Day It All Blows Up?

Despite AppLovin's current market cap of roughly $150 billion, an 84% EBITDA margin, 70% revenue growth, and a Rule of 40 score above 150—numbers that are virtually unmatched globally—Foroughi says he has never felt like "we've made it."

"Almost every morning when I wake up, the first thing I do is check the data, confirm we're still running normally, confirm we're not going bankrupt today. This fear of blowing up is one of my biggest drivers."

He says the day investors or the team hear him express a lack of conviction about the future—that's when they should really worry. So far, he hasn't had that feeling.