Summary

Ellenbogen’s core empirical finding, done before the likely Bessembinder “4%” study circulated: over rolling 10-year periods only ~40 stocks (~1% of the market) compound at 20%/year, and ~80% of those “valedictorians” start as small caps. The formative scar: a likely T. Rowe Price New Horizon fund sold Walmart early — had it held, that one stake “would have been greater than the sum total” of the entire ~$8B fund. Durable Capital is purpose-built to own the 40.
The “man versus machine” market-structure call: the short-term alpha game “is probably going to be won by the machines paired with the humans,” and he estimates 80–90% of institutional flow is driven by one-to-three-month agency or quant models — producing the most volatile earnings season since the financial crisis last Q2. Durable’s response: “let’s go do less so we can do more,” leaning into stress only where they’ve known the people and culture for decades (likely Colliers) or where short-horizon players structurally can’t own the name (Duolingo, bought more in 2022 amid a 70%+ average fall).
AI is the new “China cost” — but for IP and white-collar work, not product. Just as every product business in the 2010s had to know its China cost or die, every company reliant on knowledge labor now must understand AI’s effect on human processes. Best specimen: Max (likely Levchin) says Affirm can keep growing “without adding headcount” — and Ellenbogen sent him to likely Mitch Rales, because Danaher’s DBS has been doing Kaizen in factories for 40 years and “we’re just getting started on processes that are done by humans.”
The tradeable pattern is good-to-great, not pure-play AI: already-advantaged physical-world incumbents (distribution, trucking) using AI to widen relative cost advantage and reinvest it into permanent moats — the Amazon playbook (3–5% cost deflation for 20 years, 30%+ incremental share) and the Domino’s playbook (best Russell 2000 Growth stock of the 2010s despite not averaging 10% growth). “That’s about to really advantage us in the other 70% of the economy.”
Robotics may be a steeper second Kaizen wave: Durable’s month-old, self-described “very early and probably deeply wrong” view is that robot costs, already at or below par with physical labor in some use cases, could deflate at 15–20%/year vs Amazon’s 3–5% because general-purpose models power the curve — leaving laggards “minimum two to three years behind” and creating power-law winners in five years.
Duolingo is a high-risk AI test case: the stock often falls when OpenAI demos translation or Apple shows AirPods live translation and “I don’t think the market’s wrong… probably wrong the magnitude” — but the chess product (built by ~6 people in 9 months, now 1M+ DAUs, would have taken 4–6x the people and 4x the time before) proves the company “was purpose-built for AI.” Discount rate up, opportunity up commensurately.
Memo discipline as portfolio construction: Durable can’t buy an early-stage growth name unless the memo says they’d want to buy more at higher prices in three years — “our thesis can’t be it gets bought” — while durable-growth compounders (likely Colliers) get averaged down into macro fear. Three-year look-backs (“we thought X, they did Y” — two slides) keep everyone honest.
The pitch for going public: the average 10-year 6x compounder endures a 50% drawdown during transition, and daily marks force discipline early — his Netflix pipe story (warned Reed the fixed-cost content transition would break the balance sheet; led half of a ~$4.5B recap) shows why. “You have to be in the and business, not the or business”: growth, innovation, and profitability simultaneously.

Deep dive

1. A biologist’s frame, a mentor’s philosophy, and the Walmart sale that changed everything

Ellenbogen came to investing from politics and organic chemistry, not finance, and imported a biology frame: organisms persist by staying “in balance with their ecosystem” — so why shouldn’t companies that balance customers, employees, shareholders, and communities be the ones that sustain? His likely T. Rowe Price mentor Jack Leaport believed exactly this: small companies run by people who think like owners, good cultures, good capital allocation.
Handed the New Horizon fund mid-career, he read 50 years of shareholder letters and found “it was really only 20 stocks over 50 years that drove the performance” of the country’s oldest and best-performing small-cap growth fund.
The scar tissue: a predecessor met Sam Walton on Walmart’s IPO roadshow — 50 stores at the time — bought it, then the firm sold it. The math: unsold, that single stake would have exceeded the entire ~$8B pool he was managing. “One bad decision… actually wiped out all these other good decisions mathematically.”

2. The 1% valedictorians: 40 stocks per decade, 80% born small-cap

His own study — done before what he calls the “Benheimer” study (likely Bessembinder’s “4%” work) circulated — found that over rolling 10-year periods about 40 stocks compound wealth at 20%/year (6x+), roughly 1% of ~4,000 public names. Durable’s entire philosophy is maximizing the probability of owning those 40.
Why small caps? Not sentiment: “80% of these great companies actually start as small caps.”
Pattern-recognition infrastructure: “If you’ve been one before, you have a higher probability of being one again.” Partner likely Anu Kote teaches Columbia’s value analysis program — about six case studies a year, building a library of the companies that actually did it — “our way of going back to school as an organization.”

3. Good-to-great: Domino’s is the template for the AI trade in the physical economy

The cloud/mobile-era lesson: yes, own Amazon — but once Walmart and Costco understood the game and leveraged their scale, “62% of all retail has gone to those three.” All three were good.
Best Russell 2000 Growth stock of the 2010s? Domino’s Pizza, despite not averaging 10% growth. Facing a market split in thirds (great local shops, national chains, mediocre regional players), Patrick Doyle’s team picked the one value axis technology could transform — convenience — invested in the app and direct customer relationships, layered on a ROI-light franchise model, and the brand halo followed.
Durable’s internal “good to great thesis”: which already-good distribution and trucking companies can use AI to substantially lower relative cost, gain revenue scale, and reinvest “in a way where you create something that’s permanent” — because the best stocks come when “the technology advantage transitions into a physical mode advantage.”

4. Act-two entrepreneurs: solving the same problem with total clarity

Workday in 2012 at ~$100M revenue crystallized it: Duffield and likely Aneel Bhusri had built PeopleSoft, so they knew the “exception management” — the edge cases at super scale — that first-timers can’t. “If you don’t understand that exception management because you have done it before, you’re not going to properly be able to do it.”
Max (likely Levchin) is the archetype: one of Ellenbogen’s first private investments was Slide — “I used to joke, Max, I’m the only investor in Max who hasn’t made a lot of money” — and now Affirm. Act-two founders get a clean sheet: they align people, structure, and investors exactly how they want, and pick their backers.
The firm itself is the thesis: “I almost named Durable Act 2 Capital.” Of 5–10 new investments a year, many are act-two founders — often people Durable backed in the previous act.

5. A firm purpose-built so the same person owns Figma at $30M and at IPO

Structure follows mission: ~10–15% of capital private, the rest public, and continuity as the product — partner Katherine underwrote Figma privately, led the 2021 round, sat in the all-hands when Dylan announced the IPO, and still covers the public earnings. “That’s just different… investment firms aren’t structured that way.”
The memo rule that gates everything: a $20M Duolingo position isn’t “10 or 12 basis points” of a $15B vehicle — it’s “our future compounder.” “If we can’t write the memo that we want to buy more at higher prices, we can’t buy the shares — and our thesis can’t be it gets bought.”
LP alignment means promising volatility up front, because “increasingly the market has changed where capital is short cycle… so many people are on one month incentive models, not even yearly.”

6. Two sleeves: average up the early-stage names, average down the durable ones

Early-stage growth (Duolingo private or newly public): underwritten on three years; if it derisks and proves it can “financially balance growth, profitability, innovation,” buy more as it gets bigger. Track record of persistence: DoorDash, Affirm, Toast — Durable led the last private rounds and they remain among the largest public positions, out of 100+ private investments and 50+ IPOs.
Durable-growth names get bought into fear. Likely Colliers is his best specimen of a misunderstood asset: the brand screams cyclical CRE broker, but Jay (likely Hennick) — “a disciple of Peter Drucker” — has built “an incredible business” in real estate asset management (Harrison Street) and an emerging consulting platform. When rates fears sold the stock off last year, “we bought a lot more.”

7. Man versus machine: 80–90% of flow is short-cycle, so do less to do more

When the quants started winning, he studied them rather than dismissing them — humility is doctrine (“we never assume we’re good and the other person’s bad”) — and concluded “the short-term alpha game is probably going to be won by the machines paired with the humans.” Quants dominate repeat-actor problems on known data (the PMI-rerating trade is dead); humans stay advantaged at people and change. He taught this internally as “man versus machine” and doubled down on both.
On the pods: deep respect for Citadel/Millennium talent, but a firm that measures your risk daily and cuts capital after a bad month means “you can’t have a time horizon longer than your career horizon.” Durable estimates 80–90% of institutional flow is driven by one-to-three-month agency or quants — and last Q2’s earnings season was more volatile than any since the financial crisis, despite no systemic stress.
The response: “Let’s go do less so we can do more” — because the market is “probably right 90% of the time,” you only lean into stress where you truly know the business and people.
Case in point: the average loss-making Russell 2000 Growth name fell over 70% in 2022. Durable’s view — “the market’s probably right, but not all of these can’t adapt” — led to buying more Duolingo in ‘22. “If you’re rule-based or you have a one or three month time frame, you just can’t own Duolingo.”

8. AI is the new China cost — for IP-based businesses, not products

The framing: by the end of the 2010s every product business had to understand its China cost or die; even second-derivative spread businesses (HVAC distribution putting a spread on raw materials) could be hurt by input inflation versus deflation. “The same thing applies to AI — but it’s not product based this time, it’s IP based”: every company driven by white-collar and IP labor.
Affirm as exhibit A: regulated, hundreds of thousands of merchant contracts, heavy legal cost — and Max has said publicly the company can grow at current rates “for a reasonable period of time… without adding headcount,” via a team that leans out processes “not from a cost standpoint, but from a leaning-out standpoint.”
The punchline, as told: Ellenbogen’s reply to Max — “why don’t you come to DC and let’s go see likely Mitch Rales” — because Danaher’s DBS brought Kaizen back to the U.S. 40 years ago, and over a dozen Fortune 500 CEOs started their jobs at Danaher, including GE’s turnaround chief. “In many ways I feel like we’re just getting started on processes that are done by humans.”
The Amazon mental model, from twice-yearly lunches with Bezos when Amazon was a $10B company: the best businesses use technology to lower cost and drive revenue for 30%+ incremental market share, then reinvest the unit-economics advantage into something persistent “even if their competition were to wake up tomorrow and do the exact same thing” — fulfillment centers riding a 3–5% cost-deflation curve for 20 years.

9. Duolingo: the market’s right about the risk, wrong about the magnitude

Honest handicapping of his own book: Duolingo often drops when OpenAI demos translation or Apple shows AirPods live translation, and “actually I don’t think the market’s wrong — it’s probably wrong the magnitude, but what the market is saying is there’s a risk here.” He calls it one of the higher-risk names in the portfolio.
What he demands of CEOs facing discontinuous change: already operating well (“you can’t do a turnaround and do well”), already winning the first end market, proven resilience, and “a perspective… but also humble.” Durable rewrites its own AI views every six months — “probably we’re less wrong” each time.
The proof point: likely Luis von Ahn — whom Ellenbogen compared on first meeting to likely Tobi Lütke at private Shopify — built the chess product with two people for six months plus four more, shipped in nine months, now “well over a million DAUs” and growing at “astronomical rates”; pre-AI it would have taken 4–6x the people and 4x the time. “Yes, the discount rate has gone up, but this company was purpose-built for AI” — raising the odds it goes “from a point solution to the suite.”

10. Robotics: a second Kaizen wave, on a much steeper curve

Rare on-record humility: Durable only documented its robotics views “literally in the last month,” and “I know our views here are very early and probably deeply wrong” — but he shares them anyway.
The initial conclusions: in certain use cases robot cost is already below the equivalent physical-labor process, and “this is the earliest and worst robotics is going to be.” Because machines iterate with machines powered by general-purpose models, the curve is geometric — costs may deflate at 15–20% a year vs Amazon’s 3–5%, “or maybe it goes down even faster.”
The consequence: wake up in five years and companies on one cost curve versus the other “could be power law businesses.” Laggards who haven’t built the distribution or technological infrastructure are “at minimum probably two to three years behind, and every day they wait they’re getting further behind.” This is where Durable’s people-and-change edge extends into “the other 70% of the economy.”

11. Favorite moats: things you can’t spin up, and soft things that are incredibly hard

His stated favorite: “I love physical real estate” — Carvana’s reconditioning centers being the specimen. “You can’t spin those things up”: land acquisition, network placement, capex, systems, and then an operating culture on top. Misplace the real estate and your transport cost is permanently worse. It’s why robotics immediately took his mind to distribution — “where can robotics take already advantaged businesses and make them more advantaged?”
The other category: “soft things that are incredibly hard.” Danaher compounded at ~20% for nearly 40 years with no deep physical moat or data network effect — just sharpness on what talent really means (“not the sticker of talent”), operating excellence with a system behind it, and capital allocation that holds businesses accountable. Hennick’s FirstService/likely Colliers is the same species: decentralized incentives from condo management to roofing, selling anything without “a path to be great.”

12. A writing culture, three-year look-backs, and the media memo

Why write everything down: “human beings are innately human” — you lose executive distance when you know and root for the founders. Memos must articulate why the company is (or will be) competitively advantaged, why the operating culture is excellent, and why the leader thinks like an owner; quarterly operating reviews cover every position with the whole 12-person team.
The process he wishes he’d invented decades earlier: every three-year holding gets a look-back — “three years ago we thought they would do X and now they did Y,” two slides. “If it deviates a little bit for 12 straight quarters, actually it’s kind of staring you in the face.”
The concision test — “if you really understand something, you can make it super concise” (robotics memos are “probably too long because there’s too much unknown”) — traces to his early media study: the whole industry’s economics rested on cable networks growing 20% with 20% ROEs for 20 years, John Malone at the center, all “predicated on a closed system” (the worst shows aired Sunday night, the best Thursday). The TMT bubble’s broadband overbuild laid the seeds of YouTube and Netflix. His memo’s summary line: “the riskiest thing is to own the durable asset and the safest thing to do is go buy the next standard.”

13. The 2022 tour: what only works when money is free

His confession about the free-money era: “every CEO I talked to, every investor I talked to — and even Durable — had made simplifying assumptions based on almost a decade of free money,” when 30% of all treasury bills carried negative yields. Rerunning the compounder study by regime: normally ~40 compounders; during free money, 120 — “it was three times easier.”
What’s perennial versus regime-dependent: growth-plus-profitability works always; small companies needn’t be GAAP-profitable “but you do have to show progress on your path towards it.” What he rejects: no progress toward returns, and leveraging cheap debt onto low-quality businesses.
The dinner with Luis and his CFO, as told: “It doesn’t mean you need to get to your long-term margin targets at 30% — but you have to show progress towards it.” Versions of that conversation happened with Aman at Toast and Max at Affirm. Today’s contrast: “we’re more back to learning and normal interaction than having a perspective we’re dying to explain.”

14. Why go public: the Netflix pipe and the “and business”

He concedes the indefinite-private view is “very thoughtful” — “maybe Elon is correct and SpaceX never has to go public… here’s the good news about life: we’re going to actually run an experiment.” But the public path to a generational company “is proven, and if you understand how to do it, actually very clear.”
The data plus the story: the average 10-year 6x compounder suffers a 50% drawdown — during transitions, not bear markets. Netflix: the likely Qwikster misstep, buying back stock at $280 as it fell to $70, and Ellenbogen’s Saturday call to Reed (likely Hastings) — the DVD-to-streaming shift meant going from variable-cost to fixed-cost content, and “you’re going to run out of cash, or at least the market’s going to think you will.” Hastings’ response: “I have not thought about this as much as I should. Let’s talk tomorrow.” Likely T. Rowe Price led half of the ~$4.5B pipe (TCV’s Jay Hoag the other half) — and a year later the stock was in the $50s. “Did it look like a great investment?”
The governance argument: daily marks signal transitions early and let leaders realign teams to the right incentives — better done “sooner rather than later,” before culture calcifies. His operating credo: “you have to be in the and business, not the or business” — growth measured by share, innovation/capital allocation, and profitability, with the CFO “not as a policeman but as someone who sets standards that forces you to make sharp decisions.”

15. Building a firm that outlives its founders — and roots for everyone

The founding goal: build a firm “that would be better the day I left… than when we were the best while we were there” — he did a listening tour of firms that had periods of greatness, and the lesson was “if you don’t architect the system on day one for success, you end up with conflicts that undermine what you could have accomplished.” No compromise hires: no one joins unless they could one day lead the firm.
The talent model is deliberately non-Wall Street: likely Anouk Day started at 26 from a nonprofit after an Oxford master’s; Corey at 21–22 out of William & Mary — and “the youngest person in the room can have the most valuable perspective” (her millennial-mindset work fed the DoorDash, Sweetgreen, Warby Parker privates). Making colleagues better is measured: 360 reviews demand specific investment examples, not “she’s a nice person.”
Process kept light but sharp: ideas gated together on Mondays so everyone learns what deserves time; Friday insight lunches (“I listened to OpenAI dev day and this is what I thought”); offsites that dropped team-building for look-backs and industry deep-dives — “we’re a group of people who likes learning from each other.”
The temperament, in his sports frame: not Jordan’s fierce-competitor greatness but Kerr’s Warriors and John Wooden — “we want to have fun and we actually root for everyone.” Durable sold likely Affirm when risk-reward demanded it (“it’s not our money, it’s our investors’ money — we got to be fiduciaries first”) but never stopped talking to Max. And the closing story: at 19, his mother let him drop out of Harvard to become chief of staff for Congressman Deutsch — “if you’re going to go do that, you need to be responsible for paying for your education… you’re making a real adult decision.”

Are Humanoid Robots Actually Coming to Your Home? | Nikolaus, Rerun Market Overview — December 16, 2025