Alex Rampell: The Best Founders Materialise Capital, Customers & Labour | The Future of Venture

Summary

• Andre Horowitz’s $15B raise is a bet on the “death of the middle”: in venture you’re either a large generalist or a small specialist, and mid-sized generalists “are largely going to lose to the big generalists or the small specialists.” Rampell’s LP math: you’d rather have a 3x on $1B than a 5x on $50M — “the harder thing to do is to just return gross dollars, period.” His personal LP stakes in Ribbit’s 55x fund one ($85M) and AngelPad’s 120x-DPI $8M fund are, he concedes, probably not attainable at $2B scale.
• His people framework, from an internal memo: back founders who can materialize labor, capital, and customers — five people follow you tomorrow for a 50% pay cut, round n+1 gets easier, and first customers can be won despite “a week of cash and zero customers” (the Toast test) — plus deep history of the space and Count of Monte Cristo revenge motivation: “$100 million to an 18-year-old is transformative. You’d have to be an idiot to turn that down, or you have to want revenge.”
• “The best companies have hostages, not customers.” The “greenfield bingo” thesis: don’t try to rip out Workday’s hostages — sell the better product to new companies, so the play only works where the rate of new company creation is high (Stripe; Mercury, where Alex doesn’t think it stole a customer from SVB before SVB failed) and fails where it isn’t (“the rate of new hospital creation is too slow” for a better EHR).
• Early-stage venture is buying out-of-the-money call options — a $1M-revenue series A losing $10M a year “of course isn’t worth $100 million”; you buy 15-20% hoping the option expires in the money. The portfolio rule: “we either want to buy any percent of something that is absolutely working, or high ownership of something that could work” — and if you win 100% of deals at low ownership, “you’re probably not testing how far you can go.”
• Competitive compression is now brutal — VisiCalc took ~5 years to halve, Lotus ~15 years to die, but in 2025 “this can take weeks, which is bonkers” — and with few unicorns passing rule of 40, Rampell bets “maybe 5% of the unicorn class will ever be able to go public.” He “hates massive secondaries” for the moral hazard: a 2021 fund sought to double its stake, which could make a founder rich enough to stop caring about anyone’s liquidity.
• The AI apps playbook: hyperscaling “software that does the job of labor” (Eve lets plaintiff attorneys take the $1,000 contingency cases; $20K software replacing the $80K hire you couldn’t make) must “back into a system of record” — or hold a walled garden of proprietary data: vLex grew ~5x after adding AI to 25 years of Spanish legal records, and “I’d rather have GPT-3.5 plus infinite data of everything around medical science versus a sentient being that has no data whatsoever.”
• Selling a company is a years-long “cron job”: spend ~5% of CEO time getting to know three or four potential acquirers as genuine partners, never pitch corp dev (“they execute transactions”), because “in every M&A conversation, in every fundraising conversation, the first question is: what was your last round price?” — an insane price ends the conversation.
• His biggest miss: haggling $130M vs $135M with Plaid’s Zach at the series B (Goldman would have paid 200) — corrected by paying $2.4B at the series C. The meta-lesson: “the most valuable insight that you can have as an investor is the self-reflection to say, I’m an idiot”; on deals where a16z passed at round n−1 and paid up at round n: “I’d rather be rich than right.”

Deep dive

1. Death of the middle: go huge or go specialist — the middle dies

• Harry’s opener on the $15B raised that day — go really big or go boutique? Rampell’s answer: most asset classes suffer a “death of the middle” — “you either have to be a large generalist or a small specialist,” and the mid-size generalist “is largely going to lose to the big generalists or the small specialists.” Ribbit (fintech) and likely Kaszek (Latin America) can stay small because they’re specialists; a firm that does a little of everything, isn’t big, and can’t help much “is just going to lose.”
• Why scale works now: Series D didn’t exist in 1992 — “that was an IPO.” Amazon went public at roughly a $600M market cap; today the five biggest companies on earth are all technology companies (20 years ago all banks, 10 years before that oil, before that Japanese bubble names). Companies go public far later, so venture can deploy far more — almost $7B of the new raise is for the growth fund.
• The LP math against the “returns decay at scale” canon: a 3x on $1B beats a 5x on $50M — “the harder thing to do is to just return gross dollars, period.” Alex is personally an LP in a Ribbit fund led by likely Micky Malka (~55x on an ~$85M fund) and AngelPad (~120x DPI on an $8M fund): “can you get 120x on a $2 billion fund? Probably not.” Harry’s pushback — endowments can access small funds, so hasn’t a16z scaled out of the best risk-adjusted spot? Alex: “you can’t disprove an unknown future,” but he invests his own money in a16z’s funds — big generalist or small specialist is “where the best returns will be.”

2. Venture is a sales job — and the best deals are mostly consensus

• The job description: “find, pick, and win investments. If they’re good investments, the winning is very, very hard.” Unlike private equity — KKR and Blackstone bidding for RJR Nabisco just sell to the highest price per share — in venture the founder chooses you, so you must credibly say “I have this amazing specialty, and/or I’m connected to everybody on the planet.”
• Against the non-consensus fetish: “everybody wanted to invest in Uber, everybody wanted to invest in Facebook” — doubt only creeps in at price (“I don’t know if I want to invest at 87 million pre for the series A of Facebook, but everybody wanted to do it at 20 million pre”). Occasionally the deal nobody wants becomes a thousand-x for a lesser-known firm that was “sold that deal” — but “a lot of the best deals will go to the best firms.”
• Harry counters with 11 Labs — “the most non-consensus deal ever at seed”: competing with OpenAI, in London, pre-seed. Alex’s rebuttal: “the entrepreneur was pretty consensus” — a super-talented, high-agency team is consensus at seed, where there’s nothing yet to be non-consensus about; it flips only “when the price goes up high enough” or once numbers exist (“a series B with half a million in revenue shrinking every month — of course it’s not going to be consensus”).
• Agency, defined using Harry himself as the specimen: “they’re not going to be told what to do — they just take matters into their own hands… what you did is not normal” (emailing every famous VC to death at 17).

3. Out-of-the-money call options and the ownership frontier

• The financial frame for early stage: “we are buying out-of-the-money call options and we hope they expire in the money.” A series A with $1M of revenue losing $10M a year “of course isn’t worth $100 million” — you’re buying 15-20% and hoping; only later does valuation converge on discounted-cash-flow reality. (His kids’ Silicon Valley reference: “no, you can’t do revenue — you have to be pre-revenue, because then you’re a pure play.”)
• The hiring analogy for price discipline: if 100% of candidates accept your offers, either you’re the greatest hiring manager alive or you’re overpaying. “If you win 100% of the deals, that’s a very, very good sign — but if you’re winning them with very low ownership, you’re probably not testing this efficient frontier of how far you can go.”
• Harry pushes hard: with seven funds and follow-on firepower, why not take 10% at the A and run a higher win rate? Alex: that’s Zeno’s paradox — “why not 9%? why not 8? where do you draw the line?” — and big-fund math needs high entry ownership because dilution erodes it even taking every pro rata. He’s watching Standard Capital, the YC offshoot associated with a 10% model: “that’s very, very bad for big funds.” Harry’s fund-returner spreadsheet (10% entry, 5% at exit — can this be $15B?) gets swatted: “it’s garbage in, garbage out — you can always say that.”
• The rule he actually runs: “we either want to buy any percent of something that is absolutely working, or high ownership of something that could work.” Facebook’s ~$25M B round at ~$500M split by Greylock and Meritech is the template for the former — but the bar is high: “this is crushing, the fastest growing company we’ve ever seen — it probably comes around once every decade. Throw away the entire rule book.”

4. Back founders who materialize labor, capital, and customers — and want revenge

• From his internal memo on how to invest in people: the rare founder can materialize labor, capital, and customers. Labor: when people are paid fortunes to stay at OpenAI, Anthropic, or Meta, “if you can snap your fingers and five people follow you tomorrow for a 50% pay cut, that’s pretty magical.” Capital: a great fundraising story makes rounds n+1, n+2, n+3 easier. Customers, hardest of all: imagine Chris at Toast pitching a restaurant — “How much cash do you have left?” “A week.” “How many other customers?” “Zero.” Pulling that off is the tell.
• Sub-trait one: study the history of the space. Patrick Collison — then /dev/payments — knew payments history cold; Alex thinks he met likely Dee Hock, Visa’s founder. John Collison gave Alex an academic textbook on the origins of payment systems. Same pattern for likely Vlad Tenev at Robinhood, likely Apoorva Mehta at Instacart (met Webvan’s founders), and likely Brian Chesky (studied 1800s bed-and-breakfasts). Alex still passed on Stripe’s seed — Patrick’s “my customers don’t exist yet” was “the stupidest answer I’ve ever heard, but obviously it was genius.”
• Sub-trait two, via his favorite book: the Count of Monte Cristo. Likely Edmond Dantès becomes the richest man in the world and “doesn’t give a shit — he wants revenge.” That matters because “$100 million to an 18-year-old is transformative. You’d have to be an idiot to turn that down, or you have to want revenge or redemption.” Dave Duffield, after PeopleSoft’s hostile takeover, starts Workday — “Fuck you, likely Larry Ellison.” Likely Renaud Laplanche, fired from Lending Club, starts Upgrade — “no accident that the company is called Upgrade” — now worth maybe 10 times Lending Club.
• Harry’s trap: doesn’t expertise curdle into the dismissiveness that cost him Stripe (as Harry now feels about lending)? Alex’s two fixes: force a beginner’s-mind sparring partner into every pitch in his own domains (“you always have to be like, what if it works?”), and ask the founder what’s genuinely different this time.

5. Hostages, not customers — greenfield bingo and the speed problem

• The signature line: “the best companies have hostages, not customers.” Something marginally better than Workday can’t sell GE — “Workday has hostages… never going to happen.” But if the rate of new company creation is high enough, new companies pick the best product: “greenfield bingo” — “you pick every software category, you build a better version of that, and you’ve got a shot.” That’s why Stripe worked; it’s why a better EHR fails (“the rate of new hospital creation is too slow”). Mercury, where he was first investor, “never stole a customer from SVB” — Alex doesn’t think that happened before SVB failed.
• Don’t mourn the hostages who won’t switch: “hopefully they die because they’re using shitty software. Let’s just sell into the future.” Fast-ramping markets also keep your hotshot Meta and Google recruits from quitting out of boredom after eighteen months and one sale.
• Harry: what about foundation models and Cursor, where “the promiscuity of customers has never been higher”? Alex: every revolution splits into an infrastructure layer and an application layer; the app layer tends to be stickier but can face 9,000 competitors, and infra “is pretty hotly competed as well.” Promiscuous customers are exactly why infra players specialize — “that’s why Anthropic, I imagine, has gotten very good at coding.”
• The compression stat, from his own chart: VisiCalc took ~5 years to fall from 100% to 50% share; Lotus 1-2-3, at ~70% share in 1986, took ~15 years to die to Microsoft. “In 2025 this can take weeks, which is bonkers.” When a two-year build takes two weeks, app-layer pressure explodes — so go boring (Vlad’s Ask Leo in procurement “is not attracting 9,000 competitors”), get all the customer’s data inside, and win “the battle of every startup versus incumbent: whether the startup gets the distribution before the incumbent gets the innovation.”

6. The three theses of the $1.7B apps fund

• Thesis one, greenfield systems of record: exactly what he pitched LPs. His board company Rillet (likely — spoken as “Realet”) “is never going to grow zero to 100 in a month, but it is very, very sticky revenue” — NetSuite has hostages, and if you sell into every new company you win slowly and permanently, with “infinite option value” stacking things like a collections AI agent on top of overdue invoices.
• Thesis two, “software that does the job of labor” — where the insane growth lives. Eve sells to plaintiff attorneys, whose dominant software “is called Microsoft Office.” On contingency, a guaranteed-win $1,000 case isn’t worth a lawyer’s time — software that does the work makes every small-ticket case worth taking. “Instead of hiring somebody for $80,000 a year that I cannot hire, I can hire this software for $20,000 — and before, I was paying $0 a year for software.” The catch: a “thin wrapper on OpenAI plus 11 Labs plus something else” won’t be sticky, so his question to every such founder is “how are you going to get the hostages?” Saliant’s Ari, asked what happens when make-believe rival “Taliant” undercuts by 50%: outbound calls are the wedge, “we are going to back into a software product” — “I love that answer, and it’s true — that’s what they’ve done.”
• Thesis three, the walled garden: vLex bought and digitized every Spanish legal record over ~25 years to roughly $20M ARR, added AI, and grew ~5x — because even if “GPT 5.5 is here” and OpenAI is a sentient being, it doesn’t have the data to draft the Spanish court response. OpenEvidence did the same for health data: “I’d rather have GPT-3.5 plus infinite data of everything around medical science, versus a sentient being that has no data whatsoever.”
• Harry’s confession as stress test: his German POS company called Aloe (“a Toast for Europe, but a little bit better”) 5x’d from ~$500K to $2.5M, yet raising “8 or 10 on 50” was “horrible” — is triple-triple-double-double dead? Alex: no — “what really matters is growth and stickiness,” and he’d take “a slower-growing permanent system of record that will never get ripped out” over “the fastest growing thing on the planet that has 9,000 competitors that are all built in Lovable by 17-year-olds… I’m surprised it was as challenging as you portray it.”

7. Moral hazard: secondaries, foie gras, and founder-capital fit

• The liquidity backdrop: few unicorns conform to rule of 40 and many are shrinking, so “probably of the unicorn class, maybe 5% will ever be able to go public.” Harry’s example — a cybersecurity company he called “sneak” (likely Snyk) getting eaten by newcomers before ever returning shareholder money — draws full agreement: “this is a big challenge.”
• On the record: “I hate massive secondaries.” A $50M founder secondary is fine if they just turned down a $10B Google acquisition — that’s the Count going for it. But a fund in 2021 did a massive secondary into one of his companies (“we own 4%, we want to own 8%, because 8% is more than 4%”) and “you have now introduced moral hazard”: a founder with generational wealth may stop caring about liquidity for employees and investors. Harry’s sharper cut — the “foie gras-ing of startups”: over-capitalized companies “do 10 things, not two things, none of them work, the team is disincentivized, culture sucks.” Alex: “moral hazard — on both primary and secondary. Necessity is the mother of invention.”
• He generalizes to the input fallacy behind big-government thinking: people assume more input means better output, but “sometimes there’s addition by subtraction — I would rather have two people at the IRS than 80,000, but have those two people be likely Noam Shazeer and Jeff Dean.”
• Hence “founder-capital fit — nobody ever talks about that.” He led likely Rillet’s B just 60 days after its A — “that’s unfortunate; I would have rather done the seed, but if you find the winner, it’s also very expensive not to do that deal” — comfortable only because CEO Nick “has a bit of the Count of Monte Cristo in him.” Infinite capital lets founders dodge the fork-in-the-road decisions (Yogi Berra: “when you come to a fork in the road, take it”), and “making no choice at all — that’s the worst option.”

8. Price psychology: the first question is always your last round price

• The speech that “works maybe one time out of a hundred”: with great capital comes great responsibility. He lived the downside — TrialPay’s high-priced series C tanked a same-price Google acquisition and poisoned the next raise. “In every M&A conversation, in every fundraising conversation, the number one question, the first question is: what was your last round price?” Raise at a billion-plus with under $1M of revenue (a real deal he turned down at ~$200M post that later did exactly that) and “you have ended the conversation — the psychology of that round is all wrong.” The speech fails because founders are definitionally irrationally exuberant: “if they thought they had a 0% chance of raising a series B, they wouldn’t start the fucking company.”
• Harry’s tweet that got him “in trouble with my team”: series A is the worst place to be investing — minimal progression, 4-5x the seed price, 150-200x ARR, little sign of PMF. Alex half-agrees but blames nomenclature: his 2006 TrialPay A was $3.1M on $9.5M pre (then “the most expensive deal” his Battery partner had done); today one series A is five OpenAI superstars needing GPUs (“no moral hazard — you’re not going to go spend money on people”), another has $10M of ARR. The real trap was the old series B: “the only difference between series A and series B is that you increased your burn and built scaffolding… why would I invest at half the ownership when nothing has changed?”
• Harry’s own scar tissue: he lost Ask Leo to likely Seema on Alex’s team — “you did not pay more than me, you just beat me fair and square” — and now berates himself: mapping revenue 18 months forward, “you idiot, you should have paid 300 and doubled them” and still cleared a 3x. Alex on his own losses: “a lot of times the difference is just on price or ownership… eventually it’s like, I don’t want to be a fucking idiot — this is the winner, I want to be in the B at lower ownership. Most humans do not have the capability to admit that they were wrong — if you’re an investor like that, you’re just going to lose money all the time.”

9. Selling a company is a cron job you start two years early

• Prompted by David George’s tip that this is Alex’s never-discussed specialty: M&A is “a very highly choreographed dance,” nothing like raising off hot metrics. If you see the wall coming in 18 months, start a background process — “like a cron job in Unix terms: 5% of your time as CEO should just be getting to know people at the three or four companies that might buy you.” Never say “please buy my company” — “that’s DOA.”
• Skip corp dev: “corp dev doesn’t buy companies — they execute transactions.” A $500M-to-$1B acquisition can involve “this SVP who has some hole in their personnel or needs revenue growth to get their bonus.” At TrialPay he spent years genuinely partnering with Visa and PayPal — “if I got that deal, I didn’t give a shit if they bought us or not” — until a partner concluded “this is so valuable for us, we have to buy that company.” His frame is his favorite movie, Inception: “how do you incept this idea?” — except it takes 18-24 months, not one flight. (He once sat Zack at Plaid next to likely Alec Kelly at Visa at a dinner: “that worked until it didn’t — because of the Justice Department or something.”)
• Timing is cruel: the best moment to sell is mid-rocket-ship, “but rarely does that intersect… nobody wants to buy this falling knife.” The same background process applies to fundraising — 5-10% of CEO time casually meeting investors; he raised TrialPay’s series D from Greylock only because likely Reid already knew him after ~20 meetings: “they never would have done the deal otherwise.”

10. AI and labor, the Plaid miss, and venture eating more of the world

• On likely Jason Lemkin’s call that labor displacement shows up materially this year: “I’m not sure about that — in certain areas, for sure.” His taxonomy: three types of SaaS — impervious hostage-holders where AI is a tailwind (Workday, NetSuite); Zendesk, where “revenue could go down 100%” if every ticket auto-answers; Adobe in the middle. Harry cites likely Decagon in customer support and Harvey; Alex concedes the Zendesk-user labor market “might get decimated, 100% agreed,” but sees possible reallocation as well as elimination: United Airlines giving its best travelers “a personal human that remembers their birthday,” likely Tony Hsieh’s Zappos treating support as a revenue center (flowers for a bereaved customer), and his talk to JP Morgan’s exec team naming wealth management least-touched — “if you have a high EQ and you’re good at playing golf with people, you’re going to start hiring more people like that.” The upskilling “is not ’everybody should learn how to code.'”
• Quick-fire changes of mind: he was “probably the first,” in 2023, to describe private-equitizing venture — “barbarians at the gate with an AI” — which General Catalyst and others now run; “I’ve probably become more bearish on that just because it feels like a founder-market mismatch.” The product Andre Horowitz lacks and might want: credit (GC has a credit fund) — but there’s a solid reason not to: a lender must “beat up the companies that are struggling,” and a venture firm earning 1000x on winners can’t be at odds with its entrepreneurs.
• Biggest miss: Plaid’s series B, haggled over $130M vs $135M with Zach while Goldman was willing to pay 200 — “no, no, five mill… that was just so stupid.” He was “burdened by what has been” (his likely Kamala Harris joke): Yodlee’s $600M terminal valuation anchored him. He “corrected myself for it by doing the series C of Plaid” at $2.4B — the twin lesson being correct yourself without pride, and don’t let precedent cap your imagination.
• His distilled advice — set against Josh Kushner’s line Harry carries (“if you’re willing to take less, don’t do the deal”): find high-agency people who know the history of the space, can materialize labor, capital, and customers, and have the Count of Monte Cristo in them — “don’t second-guess anything… I’ve become 100% convinced this is entirely about people. 100%. And every round — it could be a D round, it could be a seed.” (Martin Casado’s midwit-meme mockery of Harry notwithstanding; Alex’s caveat: “it converges on reality at some point — you can’t tell an undersubscribed IPO order book, no, no, the founder is really good.”) Venture in five years: “it ends up eating even more of the world” — software-does-labor mints new markets the way Toast can be worth $20B, “and this is before even things like robotics — if robots actually work, you’ve expanded the market like another 100x. My guess and my hope is that it’s going to go up and to the right.”