Summary

Rory O’Driscoll’s central call: the risk curve has shifted a full stage down — old wisdom said a bad Series B meant “paying Series B prices for Series A risk,” and “it’s exactly correct now at the A. You’re paying Series A prices for seed risk.” With 20 competitors instead of three, there’s no room in the price to bury errors: picking and win rate are the whole job now, and much of SF’s claimed product-market fit at $400K ARR is, per Harry, just “having good friends with some money.”
Jason Lemkin sees a gold rush bigger than 2021 — “this gold rush goes all the way down to high school kids” — where the April Trump/Nasdaq shock spooked growth investing “for at least three days” before a fresh unicorn got a higher top-up offer: “the shock doesn’t even last until the next 20VC comes out.” Almost every GP and market will deploy every dollar in a boom; simultaneously he insists 1X the fund is nothing: “I want 3X the fund. It’s just not worth it” otherwise.
The marks underpinning all of it may be fiction. Jason, as an LP, looks at underlying assets and says “these are not where they are priced”; Rory’s framing — with no IPOs or M&A there’s no feedback loop: “we’re grading each other’s exams, and we’re all saying we’re getting As, but teacher hasn’t graded the test yet.” A giant endowment told Harry that if illiquidity is structural rather than cyclical, “we cannot be in this asset class any longer” — Rory’s answer: only capital leaving the asset class will push companies public earlier; equilibrium does the rest.
On returns math: IRR is the lever managers control least, so the real algorithm is “maximize multiple subject to a constraint on IRR” (Rory, threshold ~20%). Fabrice Grinda’s cheat code for 30% IRR is the anti-VC strategy — selling winners via secondaries on the way up when he can no longer underwrite a 10X, which drove most of his exits for three years; Jason’s 2017 fund sits at 4.31X and a 32.56 IRR that no realistic upside scenario can improve, and his LPs “didn’t care.”
Rory’s number-one fear is now structural: “the technology life cycle to obsolescence is now shorter than the holding period of privately held software companies,” so every company faces an existential second-product crisis before it can exit. Corollary from Jason: AI strengthens the enterprise (ServiceNow popping 24%, its $3B Moveworks buy “shrewd”) and weakens SMB software, with Box the live case study — perfect AI use case, S-tier CEO, and if it still doesn’t re-accelerate, the AI-uplift thesis itself is in question.
On Benchmark leading Manus’s $75M round: Rory separates idiosyncratic deal risk (probably well paid) from firm risk — congressional blowback that made Sequoia walk away from billions — and passes: “the US is 25% of the world’s GDP, 50% of the world’s software market. If I can’t make money on 50% with the rule of law, adding another 10% with no rules whatsoever isn’t gonna help me.” Fabrice exited China completely “once Jack Ma was disappeared” and now funds Ukrainian defense startups on cost-per-kill efficiency, arguing “we would lose a war with China right now.”
The talent endgame: Jason claims “the value of a 10X engineer is 100X now,” whole teams are already ~2X’d (Salesforce commits 20% of code via AI), and “half of these sales and customer success teams will be gone in two years.” His “baguette culture” broadside against European pace drew Rory’s defense — enterprise R&D can sit anywhere, go-to-market must be in the US — while Fabrice posed the open question: does AI eventually shrink the gap between the best and average engineers, “because if that happens, that changes the game.”

Deep dive

1. The low-low quadrant: expensive to deploy, hard to exit

Fabrice’s opening read: we’re still mid-AI-bubble — roughly $100B flowed into the category, doubling from Q1 to Q4, while other categories “aren’t seeing much love” and LPs have gone without distributions in ‘22, ‘23, ‘24, “and frankly ‘25.” Venture is “an unloved asset class. I think it’s the best time to invest” — but contrarian, and via funds not chasing “all AI all the time.”
Rory’s two-by-two: normally either deploying is cheap or exits are open. “‘21 — easy to make money, easy to get money back, hard to put money out… Right now, we have the low low quadrant.” His standing caveat: nobody cares about VC problems — entrepreneurs hear “valuations are too high” as “you have to pay me more for my company. I like your problems. I want you to have more problems.”
Harry relayed one of the world’s largest endowments: is the liquidity drought “a temporary adjustment… or a permanent structural shift”? If the latter, “we cannot be in this asset class any longer.” Rory: “The sentence has the answer” — only when LPs pull capital does stay-private-longer end; “economics is about seeking equilibria.”

2. Jason’s gold rush — and Fabrice’s refusal to chase it

Jason’s microlearning: after the April 7 Trump/NASDAQ drop (~15%), growth VC was freaking out “for at least three days” — then his just-minted unicorn got a top-up offer at a higher valuation. “All I see is gold rush… OpenAI says they’re gonna grow 1,000% by 2029… I’ve never seen anything like this gold rush. Never.” In 2020 you had to be “a crusty old B2B guy”; now “every 17-year-old kid is dropping out.”
Fabrice keeps 9% of investments in AI but demands differentiated data sets and thin competition, waiting for an emergent winner — “the traction-valuation metrics start aligning as you get a Series B, C, D,” and prices adjust with traction over time.
His cautionary tale: the AI profile-photo app whose MRR went $250K to $30M and back to $500K with “99% churn a month later” — raising at the peak. Lovable can go 0→18M ARR in three months, “but I think they can go back in the other direction.”
Platform risk, from his own behavior: he built Fabrice AI on LangChain and Pinecone, and when OpenAI shipped 4o, “I just ripped out my entire stack.” People are “underestimating the risk of zeros even though something can go from zero to 100 million ARR very quickly.”

3. Josh Kopelman’s speed tweet: IRR is the lever you control least

The prompt: “3X in 10 versus 3X in 17, obviously enormous, and people just talk about 3X funds.” Rory agrees but ranks the three return levers: picking is “100% in your control. If you can’t get that right, you should lose your job”; entry/exit valuations partially — his 2014 fund exited in 2021 and “one entire turn of that fund we don’t deserve. It was multiple expansion”; exit timing barely at all. Everyone is now getting in 2026 the return they expected in 2024.
Jason’s live math: his 2017 fund is at 4.31X with a 32.56 IRR, and no scenario — “5X fund, 6X fund, even an 8X fund” — moves the IRR above 32. He asked his LPs if they cared: “they didn’t care.” For early-stage managers, “we’re just looking at multiple.”

4. Secondaries on the way up — the anti-VC cheat code

Fabrice’s mechanism: owning only 1–3% per company, he sells winners into hot rounds when he can no longer underwrite a 10X — founders capping dilution take “30% primary, 15% secondary” while Sequoia, Andreessen, and Greylock fight to get in. Via Forge and SharesPost-type venues, “the vast majority of my exits in the last three years have come from secondaries.” In 2021 he sold at 100X ARR: “what I need to believe to underwrite this valuation is every star in the multiverse aligning.”
Rory’s counter-algorithm: “maximize multiple subject to a constraint on IRR” — above ~20% (7–800bps over small cap), keep holding. Selling the six-year board company you know cold to reinvest “at a slightly higher revenue multiple in a company I know nothing about… is such a risk escalation.”
Jason’s tax rider: with personal liquidity, “I don’t care about IRR as a GP. If I can get another X as a GP, tax-deferred or tax-free, whoa. I can’t beat that.”

5. “1X is not good enough for me anymore” — and the 1-1-1 fund

Jason’s confession: “1X is not good enough for me anymore at this point in life… that doesn’t even put me into real carry mode… I want 3X the fund.” Harry flagged the paradox with last week’s “if I can 5X a check I’ll do it” — Jason: “they’re paradoxical and they’re both true.” His proof the extra money matters: an extra deal in an outlying year of the 2017 fund is “almost a fund returner.”
Rory’s deadpan: “When Marie Antoinette took this attitude, she ended up with her head chopped off… I’m still willing to get out of bed for a million bucks.”
Fabrice’s answer is portfolio construction: 500 deals per fund, where 2% of deals return the fund 1X (46X average), 8% return another 1X (~8X average), and the long tail adds roughly another — “one, one, one” to 3X and 30% IRR via early DPI, “true for the last 28 years.”
Rory on why holding usually wins: compounding winners “covers up for your mistakes” — almost always, except 2021. Sequoia’s Evergreen structure was “a brilliant idea. Unfortunately, in the only year in the last 20 where it was the wrong timing” — but over 20–30 years the correct insight, dating to distributing Cisco’s $200M before it went public.

6. The ungraded exam: nobody’s marks are real

Jason, wearing his LP hat: “I look at the underlying assets and I go, I know that company… These are not where they are priced. Do not hang your hat on the IRRs of the managers.”
Rory’s extended metaphor — worth quoting whole: with no IPOs or M&A “there’s no feedback loop. We’re grading each other’s exams, and we’re all saying we’re getting As, but teacher hasn’t graded the test yet, and teacher appears to be on strike.” The reckoning is the S1: “it’s like you’re gonna take your clothes off and everyone’s gonna see what you really got.”

7. Going public: liquid money must cost less than illiquid — eventually

Rory splits Gurley’s position: pro going-public-early, but companies “stay private because they can… People don’t do what they should. People do what they must.” His first-principles absurdity: “how in God’s green earth, if the companies are the same,” does daily-tradeable capital cost more than five-year-locked capital? “It’s a point in time absurdity, and it will change.”
Fabrice, twice a public-company founder: “The last thing I would ever want is to be public ever again” — the quarterly budget treadmill “took all the fun out of being a founder.”
Jason on subscale IPOs: a $600–800M market cap with no analyst coverage, a 3X multiple, and employees who know their equity’s worth “to the nano cent” — “they’re all miserable.” HubSpot IPO’d around $100M growing ~60%; Rory’s bar: $100–150M revenue at 50–70% growth should be able to go public, whereas 1999’s 350 IPOs at median trailing revenue of $18M was too early — though it minted 6X-net, 100%+-IRR 1996 vintages.

8. The marathon problem: founder CEOs tap out before the finish

On Discord’s Jason Citron leaving, Jason’s numbers: 90% of B2B companies that IPO have a founder CEO — will that survive 15-year timelines? “You really do have to reinvent yourself as a CEO every five years and sign up for another tour of duty.” If you don’t want Wall Street, “leaving 12 months before the IPO is the right time”; otherwise companies drift into “terminal decay.”
Rory keeps the bar for replacement extraordinarily high: “a founder CEO with some managerial limitations usually performs a lot better than a reasonably good manager with no founding DNA.” But the race changed: 1999 was a three-year sprint, then a 10K, “now we’ve converted this into a marathon… it’s not surprising that lots more people tap out.”
Harry finds the length clarifying for picking: “you have to be a fucking psychopath… I look for nothing normal in the founders.” Jason’s companion warning — the “20 percenter club,” nine-figure-revenue CEOs in a WhatsApp group convincing each other 20% growth is great: “you need people who log out of that WhatsApp group.”

9. Rory’s number-one fear: obsolescence now arrives before the exit

The structural claim: “the technology life cycle to obsolescence is now shorter than the holding period of privately held software companies, which means every company at least one time before it gets to go public will have an existential reinvent-itself second-product crisis.” Jason sharpens it: at 15 years to IPO “you’re so far architected before the AI age, no matter what agent you add on top of it, you’re having an existential crisis.”
What separates survivors is the reinvention act — Zuckerberg’s mobile pivot as the archetype; Rory’s own case is Lattice: a 20% growth year, acquisitions, doubled engineering in the downturn instead of cutting, “now it’s back up to 40% plus.”
Fabrice’s carve-out: in marketplaces, AI benefits startup incumbents with data moats. Not eBay — its horizontal stack “is not built such that you’re best in class in every single vertical” (a Pokémon marketplace beats it; his portfolio company Rebag photographs a handbag and instantly returns model, authenticity, quality, price). But at the LLM layer, horizontal wins: DALL-E killed his Midjourney use — “the same way that Google won search reasonably writ large except maybe Kayak for travel.”

10. AI helps enterprise, hurts SMB — ServiceNow and Box as the tests

ServiceNow popped 24% at ~20% growth on $12B ARR. Jason’s caution: everyone exaggerates AI’s revenue impact — Benioff’s 500K Agentforce transactions against SaaStr AI’s own 100K+ “sounds great at first blush,” but it’s early. His tentative call: “the enterprise overall get stronger with AI and the SMB leaders get weaker and weaker because they’re disrupted faster.”
Rory’s three-player map for these categories: the pre-AI behemoth (ServiceNow), the AI teenager (built pre-LLM from ~2018, like Moveworks), and the post-LLM YC generation. ServiceNow buying Moveworks — $3B, roughly 1% of market cap — was “a shrewd move” to get relevant.
Box is Jason’s live case study: “there’s fewer spaces that you could disrupt more with good AI than documents,” an S-tier CEO all over it, trillions of documents — “if we’re shooting from the hip as investors, it should re-accelerate to 20 or 30%. But if it doesn’t, then I’m trying to learn how AI will change it.” Rory’s context: Box built a cash-flow-positive business “competing against the two largest companies on the planet who give the damn thing away for free,” and Adobe was stuck at $1B for years before its unlock.

11. Windsurf’s price cut, the eight-team problem, and the OpenAI SPV

Harry’s worry: Cursor and Windsurf generate “a billion lines of code a day” while cutting prices ($30 down to $15–20) — value creation without extraction? Rory: it’s shrewd tiered pricing, not charity — hook the base tier, escalate with value, “in much the same way as OpenAI has zero, 20, 200, and 2,000.” Jason’s gloss: a HubSpot-style barbell — cheap entry (HubSpot’s Essentials is ~45% of new customers) plus 200–300 enterprise sellers at $60–100/seat. “A quietly better model than it looks.”
Fabrice’s ‘21 flashback: eight great, well-funded teams per category “actually killed the economics” until a winner emerged. Winner-take-most dynamics mean “there may be a lot of investor value destruction on the way up” — so he stays sidelined until dominance, when “price and traction will be more aligned.”
Harry’s counter-exhibit: a friend led a model-company round at $4B, now marked $60B — but the multiple is 3.1X, “because they’re shedding 9% a year in employee stock comp… it’s not really a venture fundable asset.” Rory shrugs: they’re still up 3X with upside, and “walking away from the trend entirely is just too hard… It’s not the only game in town per se, but it is the biggest game in town.”
The other exhibit: a quarter-billion-dollar SPV into OpenAI at $300B. Harry: “there’s a non-zero chance you’re gonna three to four X that… I’d do it.” Jason’s cynical rider — a no-GP-commit SPV has no downside: “you just don’t talk about it if it doesn’t work out” — prompting Rory: “when you get a little bigger, you’ll hire a GC, and this will be the moment your GC has a heart attack.”

12. Series A now means paying A prices for seed risk

Rory turned Harry’s own tweet on him: awesome founders, directionally correct market, economics that make sense — but by an in-revenue A “you have to have product market fit. That’s why we use the word… if you pay up for a company that has product market fit and then you lose it and have to reacquire it, by definition you’ve overpaid.”
Harry’s jab landed: SF deals at $400K ARR aren’t PMF — “you’ve got all your YC mates around you… That’s not early product market fit. It’s having good friends with some money.” Rory: “Okay, you caught me. That’s the big problem, dude.” His resolution is the episode’s thesis: the old B-stage trap — B prices for A risk — “is exactly correct now at the A. You’re paying series A prices for seed risk. But the whole point of being good at this job is figuring out which is which… You don’t have enough room in the price to bury a lot of errors.”
Jason’s complication: Windsurf and Cursor “were radically different companies when they were seed funded.” Green Oaks doubled down when Windsurf was still Codium — later abandoned, every engineer repurposed. That’s betting on an S-tier founder like Varun, “and those bets don’t work out most of the time.” (Harry credits the seed lead as Neil Masak as heard — likely Green Oaks’ Neil Mehta.)

13. Manus and China: idiosyncratic risk versus betting the firm

Benchmark led Manus’s $75M round at what Harry describes as 4x the last valuation, into a Chinese AI company. Jason’s layered risk list: liquidity, “the government will take away your shares,” repatriation — plus the meta-observation that nobody cares anymore: “people love defense tech that’s off killing people… it was just a couple months ago we were talking about AI might kill us all, and I haven’t heard a peep out of that since Anthropic was formed.”
Rory’s three-part test, deliberately non-moralizing: deal-level, the risk is idiosyncratic and “probably massively getting paid for”; firm-level is where he balks — congressional pressure, “Sequoia… got out of billions of dollars of value ‘cause they just wouldn’t wanna be there. Having got out of something like that, I wouldn’t wanna go back”; the moral question he punts. His clincher: “the US is 25% of the world’s GDP, 50% of the world’s software and enterprise technology market. If I can’t make money on 50% with the rule of law, adding another 10% with no rules whatsoever isn’t gonna help me. Good luck to the Benchmark boys.”
Fabrice’s scar tissue: an early Alibaba investor who “pulled out of China completely once Jack Ma was disappeared” — his remaining holding, Ant Financial, saw its IPO personally shut down by Xi Jinping — and whose Russian unicorns were orphaned after Crimea in 2014, fundable only by “local, well-connected oligarchs.” He’s even stepped back from Turkey under Erdogan. “Would I take that geopolitical risk today? No, absolutely not.”

14. Cost per kill: Ukraine as the West’s defense manufacturing hub

Fabrice’s contrarian defense thesis: he’s an Endural investor but “their cost is extraordinarily high, and they’re not battle-tested” — the metric that matters is cost per kill, and Ukrainian startups minimize it. “We would lose a war with China right now… The reason we won World War II is we out-manufactured the Axis. Today, we don’t manufacture.” His verdict: “it’s distasteful… but it’s existential and essential.”
The table ran with it — Jason: “CAC, CLTV, and CPK are the metrics I really run the fund based on”; Rory: “It’s five to one.” On the defense-tourist seed funds popping down to El Segundo, Fabrice is blunt: mostly lemmings latching onto a mega trend (a fund apparently named Shield AI excepted — “they know exactly what they’re doing”). Rory’s kicker: “they’re probably just as thoughtful as the dude who piled in 250 million on the SPV… momentum investors may actually be impacted negatively by overthinking. Please stop thinking — that sentence is all you need.”

15. Baguette culture vs the 100X engineer

Fabrice to his French founders: come to the US — “300 million rich people that are early adopters… it’s easier to go from 100 to 200 million in the US than 0 to 100 anywhere else. Just buy the Europeans.” Harry’s pushback: lower salaries, higher retention, real capital in Europe, and Pigment proving you can sell into the US from Europe.
Jason concedes the arbitrage — Paris S-tier engineers at “half the price and people stay past their cliff. Everyone at OpenAI, they leave at their cliff” — but says his founders report the real blocker: “we cannot get our team in Nice or Barcelona to work at the pace of the US.” His anecdote: a hot SVP-of-engineering candidate back from a month’s vacation who “wants to spend some time thinking about AI” — he fed the chat to SaaStr AI, which returned “whatever you do, pass on this guy.” His coinage: “baguette culture, I think, is worrisome.”
Rory, defending his continent: “there’s a visible difference between the Europe of the people who get it and the Europe of the people who don’t” — Dublin cranks “a la American.” The clean version of the claim: in enterprise software “your R&D can be anywhere. Your go-to-market will have to be in the US” — Willie Sutton, “that’s where the money is.” Plus the media warning: Harry’s social team will clip Jason into “No Europeans worth a damn,” “and we’ll never go home again.”
The closing debate: Jason — “the value of a 10X engineer is 100X now”; whole teams already ~2X’d (“even Salesforce commits 20% of their code through AI… it’s 50% across Windsurf”), and “half of these sales and customer success teams will be gone in two years because they’re mediocre.” Fabrice’s open question, flagged as a question: could AI eventually shrink the gap — average engineers reaching “70X and that’s good enough”? “Because if that happens, that changes the game.” Jason’s bet: for all of history 10X engineers have gotten better, the Cursor-Windsurf arms race breeds “six and a half day a week teams,” and B2B companies “are all gonna die if you’re not number one in AI.”

Ecolab: Clean Machine - [Business Breakdowns, EP.214]Tim Sweeney: Fortnite, Unreal Engine, and the Future of Gaming | Lex Fridman Podcast #467