Issue 18 — April 27 – May 3, 2026

This Week in AI

Hosted by Rachel & Marcus · AI hosts

AppLovin's 84% EBITDA margins and $10M revenue-per-employee are the week's sharpest data point — a case study in what happens when a company bets on AI at its lowest moment and rebuilds from scratch. That story runs parallel to a broader structural argument taking shape across multiple conversations: AI has collapsed software production costs 100x, the SaaS moat is gone, and the 'AI company' valuation premium is a temporary label that will dissolve within two to three years. Threading through it all is a harder question about what elite performance actually costs — one that Morgan Housel and AppLovin's Adam Ferogi answer, independently and candidly, in the same week.

AppLovin's $10M EBITDA-per-employee is not a typo — it's the Axon 2 rebuild

AppLovin CEO · AppLovin CEO: Why Founders Shouldn't Angel Invest & Why the Best Don't Need Mentorship

AppLovin's core advertising business of ~400 people generates over $10M EBITDA per employee and 84% EBITDA margins — numbers so unusual that short sellers defaulted to fraud accusations rather than trying to explain them. The engine behind it: a 2022 decision to scrap their entire ML stack at the company's lowest point and rebuild from scratch.

  • $9 → $750/share in ~2.5 years; market cap from under $4B to ~$250B — Adam Ferogi believes no company has created that much value that fast
  • At the bottom in 2022, AppLovin threw out its older ML system and rebuilt around cutting-edge recommendation systems — internally called Axon 2
  • 84% EBITDA margins: "there's not another comp in the world that looks like it"
  • Short sellers "can't come up with anything other than we're cheating" — which Ferogi says became a source of internal pride

"We have 80, I think 4% Ebida margins. So, like it's the revenue growth since we launched Axon 2 model has been astounding... there's not another comp in the world that looks like it."

AppLovin's radical org design: no CRO, no COO, no CMO — and 40–50% staff cuts during triple-digit growth

AppLovin CEO · AppLovin CEO: Why Founders Shouldn't Angel Invest & Why the Best Don't Need Mentorship

Ferogi cut 40–50% of staff in most departments during a year of near-triple-digit revenue growth, driven by a belief that AI automation was coming and slow adopters should go now. The HR org went from 70–80 people to 15. The executive team has no CRO, COO, CMO, or Chief People Officer.

  • 80–90% of AppLovin's code is AI-generated — Ferogi notes the percentage metric is misleading without measuring value, but confirms it exceeds Databricks' reported 50%
  • New employees onboard via Claude queries on Slack history and call transcripts — no formal L&D program
  • "Token quotas and token budgets are no different than hiring quotas. Until they get efficient, they'll be inefficient and I think a lot of companies will just burn money."
  • On fixing a degraded culture: "The only way to fix a culture like that is to go and fire 99% of people and just rebuild it from the ground up."

"I think it was in 24 25 um but mostly in 24 we we had a year where we we probably grew near double digits um sorry near triple digits but we ended up cutting the team's staff by 40 50% in most departments and the reason I did that then is a belief that if the role was going to get automated or that AI was not being adopted fast enough in those departments it's time to let those people go."

The SaaS moat is gone — AI collapsed software production costs 100x

SaaS Challengers

The multi-decade competitive advantage of legacy SaaS — millions of lines of code built over years — has been neutralized by AI, and the next generation of great software companies will be built by replacing it with AI-native alternatives. The historical parallel: cloud replacing on-prem was the last great wave; AI-native replacing legacy SaaS is the next.

  • Attack vectors range from cloning and undercutting on price (1/10th the cost) to bundling 10 point solutions into one suite to open-source replacement monetized via services
  • Contrarian advice: skip the easy targets and go after "the products that seem invulnerable" — chip design software, ERPs, industrial control systems, supply chain management
  • Ferogi (AppLovin) independently echoes this: "I'm not sure it's actually done yet" on the SaaS sector's decline, citing AI disruption, high SBC burn, and collapsing terminal value

"The last generation of great software companies was built by replacing on-prem software with cloud. The next generation will be built by replacing legacy SAS with AI native software."

The trillion-dollar timing problem: being off by two years on data center ROI could be ruinous

The Trillion-Dollar Timing Problem in AI

Even if AI capabilities arrive on schedule — and the speaker believes a "country of geniuses in a data center" is 1–2 years away — the mismatch between capability arrival and revenue materialization could financially devastate companies that have already committed to massive infrastructure spend. The polio vaccine analogy is sobering: proven, life-saving technology still hasn't fully diffused after 50 years.

  • Core tension: "we know it's coming, but with the way you buy these data centers, if you're off by a couple years, that can be ruinous"
  • Technology diffusion is structurally slower than invention — the Gates Foundation is still fighting polio in remote Africa decades after the vaccine existed
  • Balanced conclusion: AI adoption "will be faster than anything we've seen in the world, but it still has its limits"

"One question is how many years after that do the trillions in revenue start rolling in... even if the technology goes as fast as I suspect that it will, we don't know exactly how fast it's going to drive revenue."

Exowatt's Xorrise: solar-powered multi-gigawatt data centers on America's empty 41%

Exowatt Founder · Exowatt Founder: "41% of the U.S. is empty—land isn't the constraint for solar + AI data centers"

Exowatt's founder argues that land is not a constraint for solar-powered data centers — 41% of the U.S. has zero residents per the Census Bureau — and has launched the "Xorrise" initiative to develop remote sites into multi-gigawatt campuses. The geopolitical framing is pointed: betting data center power on a gas backbone is a national security risk given current global conflicts.

  • Solar scale claim: a tiny fraction of U.S. land in panels could power the country 100,000 times over (echoing Elon Musk)
  • Remote siting also protects urban communities from health impacts and energy price spikes
  • Gas dependency framed not just as environmental risk but as geopolitical vulnerability

"if you think about the context of like the war we're in right now, it's not really a good idea to bet your data center power on on a gas backbone"

AI will raise the average IQ to 150 — and destroy every business model buried in fine print

Max Levchin · Max Levchin says AI will raise the average IQ to 150

Max Levchin argues that AI as a constant personal advisor will effectively supercharge consumer financial intelligence, threatening the entire business model of lenders who profit from confusion, inattention, and buried terms. Affirm was founded explicitly to fight this — the company is "fastidiously precise about the cost of credit."

  • US lending industry's dirty secret per Levchin: two core practices — betting customers will screw up and burying terms in fine print — underpin a huge percentage of lender revenue
  • Affirm on track for $47–48 billion in loans this fiscal year — and Levchin calls it "a drop in a bucket" relative to total addressable market
  • The AI-empowered consumer is an existential threat to opacity-dependent business models across finance

"I think the average IQ with AI in your ear at all times is about to go up to 150, which is like north of the genius definition and companies that have business models that are buried in a fine print of some kind are all in for a very rude awakening."

The 'AI company' label carries a 30.9% valuation premium — and will be worthless in 2–3 years

Alfred Lin · Alfred Lin on why "AI company" will mean nothing in 2–3 years

Sequoia's Alfred Lin warns that the AI company label is a temporary premium that will dissolve as every company becomes an AI company — the same way every company eventually became an internet company or a mobile company. The data point on current froth is concrete.

  • AI software companies raising in early 2025 commanded a 30.9% valuation premium: ~$55M vs. ~$42M median
  • Historical parallel: "internet company" and "mobile company" were once premium labels; both became table stakes
  • Sequoia's filter: they see 1,000 companies per investment, looking for founders four standard deviations above the mean

"In probably two or three years, we will now call them AI companies. Just the same way that uh internet companies over time... every company became an internet company, every company became a mobile company."

Anduril's thesis: DoD R&D share collapsed from 36% to sub-1% — the Pentagon can't define the solution anymore

How Anduril flipped 100 years of Pentagon procurement on its head

The core justification for Anduril's entire model is a single statistic: the DoD went from controlling 36% of global R&D in the mid-20th century to sub-1% today. When the Pentagon was funding nearly half of all cutting-edge technology, it made sense for them to define solutions. That world is gone.

  • Anduril's philosophy per Pomerle: "listen to customers because they understand their problem" but don't let them define the solution — they lack the framework for what's now possible
  • Built to mission vs. built to spec; fixed-cost contracting vs. the entrenched cost-plus model
  • Silicon Valley was literally born out of defense industry R&D — the relationship has now fully inverted

"Today, it is sub 1% of global R&D."

The tortured founder problem: admiring success without accounting for the cost

Morgan Housel · Morgan Housel on why the most successful founders are actually tortured

Morgan Housel argues that the world's most successful founders are often "tortured" people whose achievements came at the direct expense of family, health, and personal happiness — and that admiring their outcomes without acknowledging that tradeoff leads people to set incoherent life goals. AppLovin's Ferogi independently confirms this pattern in the same week.

  • Ferogi: "Almost in every relationship of my life, I was never really present" and "I think back at like moments where the kids were growing up and sort of a blur"
  • Housel's framing: "I the world is so much better because of you, and I would never want to be you"
  • The specific trap: believing you can simultaneously be at the absolute top of your field AND maintain a rich personal life — Housel says conflating these goals leads people astray
  • Ferogi's motivator: waking up every morning checking stats, fearing bankruptcy — "that fear of blowup is one of my big motivators"

"a lot of these people, when you dig into their life, you might really admire them and say, 'Oh, that person is so great. I would love to be that person.' But, their success came at the expense of their family life, their marriage, their kids, their own mental health, their own physical health."

Key Takeaways

  • The Axon 2 lesson: rebuild at the bottom. AppLovin's $250B market cap and 84% EBITDA margins trace directly to a 2022 decision to scrap their entire ML stack during a 92% stock collapse — the most counterintuitive moment to make a massive technology bet.
  • AI is collapsing software production costs 100x, and the SaaS moat is gone. The next generation of great software companies will be built by replacing legacy SaaS with AI-native alternatives; the contrarian move is targeting the most "invulnerable" incumbents.
  • The data center timing problem is the defining infrastructure risk of this AI cycle. Even if capabilities arrive on schedule, a 2-year mismatch between deployment spend and revenue materialization could be ruinous for companies that have already committed.
  • The 'AI company' valuation premium (currently ~31%) is temporary. Alfred Lin's historical parallel is clear: every company will be an AI company within 2–3 years, just as every company became an internet company — the label will stop commanding a premium.
  • AI-empowered consumers are an existential threat to opacity-dependent business models. Levchin's "average IQ to 150" framing captures a structural shift: any business model that depends on customer confusion or fine print is facing a reckoning.
  • The tortured founder tradeoff is real and underreported. Both Morgan Housel and AppLovin's Ferogi independently confirm this week that elite founder success is frequently purchased at the cost of family, presence, and mental health — a tradeoff that gets systematically edited out of the success narrative.

Sources

Source episodes

Sourced from 106 episodes across 10 podcasts this week