Summary

Jason Lemkin’s provocation names the episode: when outcomes are $20-100BN, “seed is for suckers.” A late-stage fund just put almost nine figures into one of his unicorns and owns as much as he does — “they skip years of work and stress,” get a 1X worst case, and “achieve liquidity in a quarter of the time.” Rory O’Driscoll concedes the math but adds the catch: that game is only open to those handed large, forgiving capital — “no one offered me a billion dollars in 2009.”
Spreadsheet SaaS investing is dead. Rory’s Box position was “the exact same in 2010 as 2024”; for 20 years the direction was obvious and only the math needed analyzing. Now the installed market is saturated (“anyone who needed a DocuSign account got it in COVID”) while AI startups “acquire and lose product market fit two or three times in a two-year period.” Jason’s compression: “it used to take you five years to fall out of product market fit. Now it can be five weeks.” Every check today buys more risk per dollar of revenue — expect bimodal fund results.
The $3 trillion question: that’s the fair market value of private venture assets, and roughly $2 trillion of it is mature, slower-growth SaaS with no IPO trajectory. PE isn’t tire-kicking much — it wants “a boring-ass software company in a teeny-tiny vertical with 40% market share,” the opposite of venture’s sub-scale horizontal companies with no pricing power. “You can’t walk away from $2 trillion”: expect grim, case-by-case grinding — private-to-private mergers, profitability slogs, small IPOs.
Andreessen Horowitz’s $20BN rationale: Jason’s hypothetical spreadsheet of every S-tier deal they passed on (the whole Databricks round at $27BN in 2021 would already be 2.5X). Rory’s caveat: the game stops only “when the LPs’ bosses, the overall CIOs, stop allocating capital to venture” — the 2022 crash was merely “a pause for breath.” The buried risk is correlated multiple compression: a Nifty-50-style regime turns marquee-asset strategies into a 0.5-0.7X, versus 1.5X for conventional venture.
Founders Fund is idiosyncratic, not replicable. The leaked numbers show exactly what they advertised — 10-15-year holds (SpaceX from ~‘07-‘08) compounding 30-40% gross into 8-10X funds, plus the stones to take massive concentration. Rory now says “my Bayesian prior on financial matters in venture should be checking what Peter Thiel does.” But an LP funding ten funds to be “just like Founders Fund” makes no sense — “they’re not the same people with the same approach.”
Extended private markets screw US citizens — Rory’s public-policy indictment: the same Stripe compounding used to reach savers through Fidelity at 70bps; now it reaches them through Thrive at 2-and-20, cutting a 15% gross to ~10% net instead of 14.3%. “A monstrously stupid outcome” that ends only when late-stage privates underperform equivalent publics by the amount of the fees — then capital reallocates.
Cycle-top warning lights: term sheets with “every box checked to the maximum” in one-day hot deals, sell-5%-get-7% founder secondary refreshes, Galbraith’s “bezel” growing unseen in the boom, 93% of more than 2,000 SaaStr respondents admitting they lie to win deals, and ARR that is “neither A nor R, nor R” — Rory has shifted underwriting to GAAP because “ARR is a made-up number and GAAP is a fact.”
The buy-or-not-buy game splits the panel: OpenAI at $300BN — Rory no, Jason out of anything north of $100BN (“all my decisions are bad”), Harry “I would buy the shit out of this. Escape velocity reached.” Cursor at $10BN hangs entirely on durability: “if it’s SaaS, take my money” (140-200% NRR, competitors destroyed) — but users switch IDEs in a week.

Deep dive

1. Raw IQ is not transferable — and neither are the old SaaS heuristics

The episode opens on bitter billionaires on Twitter, and Rory’s diagnosis doubles as the episode’s thesis: “just ‘cause you really understand one domain, investing or technology, it doesn’t automatically make you understand a totally different domain… raw IQ is not transferable, and you can’t walk into a different game where people have been playing it for 20, 30 years and think you’re good just ‘cause, hey, you’re smart.”
The provocation Harry imports from Victor Lazarte at Benchmark — spreadsheet SaaS investing is dead — gets ratified by the man whose yardstick Jason used. Jason recalls Rory handing him his yardstick in 2013: “one to 10 in five quarters or less is S tier,” which he “copied with attribution” for years. Then late 2020 broke it: every startup met the bar, and a top cloud VC offered two of Jason’s portfolio companies term sheets at high nine-figure valuations “without talking to the founders.”
Rory’s post-mortem: for 20 years the direction was obvious — “take X, move it to the cloud, compound” — so all that was left to analyze was relative growth math. “I invested in Box in 2010… the thing we invested in in 2010 was the exact same in 2024. It’s stunning.” Two things then hit at once: saturation (“anyone who needed a Zoom or DocuSign account got it in COVID — we’re done”) and AI, where “this shit changes every six months.”

2. Product-market fit now lasts five weeks, and the 100-hour week is back

Jason’s compression of the new regime: “it used to take you five years to fall out of product market fit. Now it can be five weeks.” Rory has “had companies acquire and lose product market fit two or three times in a two-year period. It’s terrifying.”
Rory’s two causes: model progress at a deep level, plus an unresolved exploratory phase — like ‘99-2003 before Salesforce defined what a SaaS company was. And it’s harder this time, because “instead of just automating some backup,” “you’re really trying to automate the head of the worker” — getting inside the sales rep or SDR you’re augmenting, a target that moves as the AI improves.
The labor-intensity subplot is real signal: Jason says all his best startups are in the office seven days a week, twelve hours a day — “in 2021, people were working 10 hours a week from home… now your competition’s working 100 hours a week for real, not for fake. If you haven’t evolved, you’re gonna die.”

3. You’re underwriting upside now, and knowing less at every check

Harry’s core question — if PMF and revenue are both transient, what are we underwriting? Rory’s answer, unhedged: “if you’re not understanding that you’re underwriting more risk, you’re missing the movie. What you’re underwriting is the upside… At every stage, on every check you’re writing today, you know less than you would’ve known 10 years ago at a similar stage SaaS company. You’re just taking on more risk for a dollar of revenue.”
Harry pushes on price: inflated entries mean you’re not paid for the risk. Rory concedes the trap exactly as framed — high PMF variability plus paying up so much “that even that upside has been competed away, then it’s a sucker bet.” His closing shrug: “turns out making a lot of money is hard.”
Asked whether venture returns go bimodal, Rory says yes for two stacked reasons: more risk per deal, and elongated holding periods — every extra dice roll distills the portfolio, best ones up, worst ones down. “You’re in a riskier game for a longer period of time… some of you are gonna make it, but a lot of you are gonna not.”

4. Triple-triple-double-double still raises — but 70-80% of SaaS investors have left the building

Rory’s standing bid: a SaaS company doing 3X-3X-2X-2X with no “AI magic pixie dust” is proof it solves a customer problem — “SaaS hasn’t been made illegal… I would do that deal all day, every day. If you’ve got one of those, call me.” He admits passing on exactly that deal two or three rounds ago: “I’m an idiot.”
Jason’s counter-observation from the field: roughly 70-80% of the SaaS investors he grew up with won’t take those meetings anymore — “they’re momentum investors, and they wanna put 200 million into the latest AI deal and triple it in eight months.” Rory’s charitable decode of the heuristic: “I just don’t believe I’m gonna kiss all those SaaS frogs and find my prince” — anything doable in old SaaS has probably been done.
Rory’s crucial reframe: the problem isn’t the triple-triple-double-doubles. It’s the myriads of companies “doing 50 million growing at 10% or 20%, or 100 million growing at 8 or 9%” — that’s where the real question of terminal value lives.

5. The $3 trillion question — and why PE isn’t riding to the rescue

Harry channels his LPs: where does liquidity come from for the Dataikus, your Calibras, your Algolias? Rory sizes it: ~$3 trillion of privately held venture assets, of which maybe $0.5-1T is high-growth new stuff and “the other 2 trillion is mature, slower growth SaaS” with no IPO trajectory but meaningful value. Unlike 2000-2002, when the bad deals were small enough to close down and “say whoopsie,” “you can’t walk away from $2 trillion.” The menu: grind to profitability, PE exits, private-to-private consolidation, small IPOs where “price does all markets” — “real case-specific, long and tiring work.”
Jason’s worry, which Rory calls shrewd: PE isn’t tire-kicking much in this cohort. At SaaStr annual, founders used to report 20 PE approaches each; now the mediocre-growth cohort gets none.
Rory’s explanation is a taxonomy worth keeping: PE loves “a boring-ass software company in a teeny-tiny vertical with 40% market share, where they can screw the customers for the next five years by raising prices.” Venture funds broad horizontal markets — which, when the billion-dollar outcome fails, leave sub-scale companies with no pricing power. Run the PE playbook on a 2016-vintage horizontal app and “your gross dollar retention will be 80%… your product will become irrelevant in two years. But other than that, have a great day.”

6. Andreessen Horowitz’s $20BN: the see-every-deal spreadsheet — but picking is still the job

Jason’s reconstruction of the logic, borrowed from an LP’s tweet that “opened my eyes”: fund size should be tied to the number of winners you can deploy into. If Andreessen sees every deal, they can literally spreadsheet the passes — “Databricks was 27 billion in 2021. What if we’d done the whole round? They would have already two and a half X’d their money” — and the analysis “maybe solves to 20 billion.”
Harry hard-pushes back on “they see every deal”; Rory dismisses that as not a top-three issue and lands the sharper point: seeing every deal means seeing every bad deal, and “there are 99 shit deals for every one good deal. The more deal flow you see, the more important picking is.” His mitigant: relative picking is easier than absolute picking.
The real question is whether privates can absorb the money — and Rory’s stopping condition is precise: it won’t stop because GPs mature, founders get careful, or LPs balk. “It will stop because the LPs’ bosses, the overall CIOs, will stop allocating capital to venture. Until that happens, this game goes on.” The 2022 crash, stunningly, “didn’t cause much more than a pause for breath.” And if allocation dies the day after Andreessen Horowitz closes — they win anyway: “they have 20 billion and no one else has any.”

7. The Thrive Monopoly strategy — and the one risk it can’t diversify

Rory’s explanation of why Thrive worked is the episode’s best set piece: “a real estate investor knows only one thing — buy the best damn house on every block. The FinTech block, Stripe, tick. The OpenAI block, tick. The infrastructure block, Databricks, tick. Then you just go home and wait for the checks to roll in. It’s genius.” Every fiber of his being would have said “that’s not venture” — “but I’m not living in three houses in Miami. He wins.” The only criterion left: “is it gonna make you money, and is it gonna make you money across the cycle?” First part yes; second part, “call me in a year or 10 years.”
Harry probes the correlation claim — the assets look uncorrelated. Rory: they’re uncorrelated on picking, but fundamentally correlated on equity values. If growth-stock PEs of 30-35 compress to 12-13 — “look what happened to the Nifty 50 between ‘68 and ‘82” — a conventional strategy’s 3X becomes 1.5X, but “a strategy entirely predicated on buying marquee assets at high prices could get a 0.5 or a 0.7X.”
The second buried assumption: continued compounding. Five or six trillion-dollar tech companies prove OpenAI can go from $300BN to a trillion — “what you forget is most tech companies don’t.” His nightmare: “be a bit of a bummer to discover you’d invested in BlackBerry, still private, and they just launched the iPhone, and you decide ‘screw it, I’ll put in another billion.’” The general law: “there’s no such thing as free money — when stuff looks like free money, it typically means the risk isn’t fully recognized.”

8. “Seed is for suckers” — and the survivor’s rebuttal

Jason’s rant, delivered with relish: “Why struggle to pretend you can do 8X over 20 years on a seed fund when you can just write one big check into a winner and call it a day, and achieve liquidity in a quarter of the time? The multiple will be lower, but the absolute return will be higher… split with four partners on your tiny little fund after 20 years making nothing. It’s so stupid. When outcomes are north of 20, 100 billion, seed is for suckers.” His live example: a late-stage fund just put nearly nine figures into one of his new unicorns and owns as much as he does, underwriting a $10BN outcome with “your wonderful 1X” worst case. Rory, delighted: “I’m gonna steal that, and I’m not even gonna give you credit.”
Rory’s structural rebuttal: the game is gated by capital access. “If you have access to capital that’s large and forgiving, you should play the big-balls game… The reason you and I play a different game is I was wandering around in 2009 and no one offered me a billion dollars and said ‘have a go, and if it doesn’t work, we’ll give you another billion in 2030.’” With a smaller fund, higher entry prices mechanically lower your odds of being in the deal that transcends price.
His conservatism is scar tissue, stated without apology: “I am most proud of the fact that I made small amounts of money from 2000 to 2010 than I am about much larger returns from 2010 on” — 70% of the people he knew in ‘99 were out of the business four years later. The momentum-era ledger: “the two biggest momentum players of the last decade, Tiger and SoftBank, already out of the game”; Insight survived on savvy and earlier, lower-priced deals.
Even the best trade of the era illustrates the scale trap: Insight’s Wiz win returned a reported $2.6BN — a third of an $8.5BN fund. Jason: “I would probably quit venture if I did Wiz and it was only a third of the fund.” Rory, dry: “If you are pulling down the fees on an $8 billion fund, I don’t think you’d quit.”

9. Fund size is the strategy

On Emergence raising $1BN, Rory pulls out the deflator nobody uses: nominal GDP is 3X since ‘99 ($10T → $30T), so a $100M fund then needs to be $300-400M now “just to be the same thing” — and COVID-era nominal growth means “if you’re not up 50%, you’re falling behind.” On top of that sits a discretionary layer “driven by people playing to win”: Scale found that “you’d come into a deal with your little $20 million check and they’d laugh at you.”
His construction math, and his explicit disagreement with the Benchmark podcast’s portfolio-construction-doesn’t-matter line: with PMF variance this high, “I wanna make sure I have enough deals in every fund that the fund has a good probability of success” — ~25 deals, $20M initial / $30M total checks, “you’re at 7, 800 before you blink.” The maxim: “you’ve got to size the fund for the strategy, ‘cause fund size is the strategy.”
Harry’s corollary: $50M seed funds drive him nuts — $40M investable against $3-5M seed rounds can’t lead and get to 20 names. Jason’s blunt sorting: “you gotta take concentration risk. Or pretend.”

10. Founders Fund is a money-making machine you cannot photocopy

The leaked returns showed “exactly what they advertised”: 10-15-year holds (SpaceX bought around ‘07-‘08), strong-but-not-stellar IRRs that compound into monsters — “if you compound at 30, 40% gross not for 8 years but for 15, ‘cause you don’t give a damn about giving the LPs money back early, you end up with an 8 or 10X fund” — plus massive willingness to concentrate in winners (“if you don’t like the risk, take your money and go home”). Rory’s conversion moment: “my Bayesian prior on financial matters in venture should be checking what Peter Thiel does.”
Brian Singerman’s line, quoted by Harry — “the enemy of great venture returns is capital concentration limits” — gets Rory’s double-edged verdict: “it is the enemy of greatness, and it is the protector of massive wipeouts.” Exhibit: the stones to put $300M into Stemcentrx, take a 5X — and three years later the acquirer canceled the program. “Never forget the risk was there.”
On their no-B2B doctrine: it’s conviction about N-of-1 singularity deals — high technological bar, then untrammeled competitive free space, which very few B2B companies have. Jason’s inside color via Sam Blond: they told him they don’t do B2B and simply classified Ramp as a fintech. Rory’s refinement: “we didn’t back into it with a thematic focus… I only wanna do amazing greatness. If I run into a B2B guy who has it, I’ll do it. If I don’t, oh well.” The LP lesson Rory insists on: funding ten funds to be “just like Founders Fund” is incoherent — “they’re not the same people with the same approach.”

11. LP money will overshoot, then flee at exactly the wrong time

Rory’s forecast on flows: the opportunity set genuinely expanded (private-for-longer consumes vast late-stage capital), but “in financial markets things tend to overshoot, especially when the indicators of success are lagging, and venture is the most lagging market.” LPs will steer on trailing 10-year returns, overshoot in, then withdraw “probably just at the point when they should be investing” — a significant availability change sometime in the next five years.
His timing indicator, from lived experience raising in the Lehman/AIG offices the week they went bankrupt: “it would not be a great time to invest in venture until people spit at you when you mention the word” — and 2009-10, when LPs said “get out of my office,” was precisely when you should have done nothing but venture. The inverse holds now.
Jason’s heuristic — over a ~7-year window, exits should roughly equal new venture inflows, and “it’s been a hot minute for IPOs.” Rory’s version is blunter: “people don’t stop doing stupid shit because they intellectually figure it out. They generally stop when there’s no more money to do stupid shit.” He cites what he calls a famous Ben Stein quote — “if something can’t go on forever, it will stop” — with his own corollary: “until idiocy has to stop, it will go on.” The tell to watch: whether a Stripe, Databricks, or OpenAI IPO brings the cavalry back fast enough.

12. Extended private markets screw US citizens — a public policy failure

Rory’s most systemic argument: it is now more attractive for great companies to stay private and take capital from GPs paid 2-and-20 than to go public and take it from Fidelity at 70bps. “We have defaulted to the higher-priced capital alternatives, which is absurd.” Since Stripe performs identically public or private, the only thing that changed is who eats the fees: a 15% gross compounds to 14.3% net through a mutual fund versus ~10% net through a venture vehicle. “The ordinary investors of America are either not getting the good assets or getting them at massively higher fees… a monstrously stupid outcome.”
The Collisons’ side, which Harry relays (“why do I need some analyst at a bank to tell me about my margins?”) and a top founder’s line — “there is no really significant reason for any great company to go public today” — Rory accepts as correct today, which is exactly why it ends: “at some point, the late-stage private investments will underperform equivalent public investments by the amount of the fees, and then it’ll switch.” Jason tags it: “that’d be the efficient market thesis.”

13. Superintelligence’s $32BN, the secret-recipe logic, and a pref stack under siege

On Superintelligence at $32BN with $2BN in and supposedly no product, both guests say “go team.” Rory’s underwriting logic: fast-forward the foundation-model field and everyone without the OpenAI pedigree has struggled (“with the exception of Groq, which is astonishing”), while Anthropic, “populated by people that came from OpenAI,” worked. “They snuck away from the Magic Kingdom with the secret recipe… you got the guy who invented it. Why not?” What the model is worth once cracked is “a totally separate discussion.” Harry goes further: with a late pref, “there is zero chance this does not get bought for at least the pref. It’s fucking Ilya — Microsoft will buy him for $10 billion tomorrow.”
Jason’s brutal question ruins the comfort: can you actually count on the liquidation preference? “Acqui-hire most of the team for $10 billion and you leave the liquidation preference over in a C corp — doesn’t that work?” Rory: “Yes. That works too.” Jason’s field report: acquirers in nine-figure deals are “super aggressive” about routing around VC preference stacks — “side deals, back deals, we just want nothing going to the VC.”
Rory’s confession is the honest core: “I can pretend I’m appalled, but when my late-stage companies are buying early-stage companies, I do exactly the same thing. I don’t give a shit about Jason and his bloody preference. I wanna hire those five great engineers.” The saving grace so far is friction, not virtue: “I’m just not paid enough as the VP corporate development to take the litigation risk… it’s easier to give them their 30 million bucks. When it’s $2 billion, who knows?”

14. Every box checked: term sheets, secondaries, and the growth-stuffing trap

Jason on hot rounds now: later-stage investors “put everything into the term sheet possible to win — maximum secondary, maximum refresh, maximum even cram down the prior investors… every box checked to the maximum.” He’s seen two hot deals done in one day; check enough boxes and “there’s even an argument the valuation doesn’t even matter.”
The signature structure: “sell five, we’ll give you seven” — founder sells 5% of their position, gets a pre-approved 7% equity refresh, coming out ahead of a dividend. Rory finds it “nauseating ‘cause you’re effectively replacing the comp committee” — but he’s lost a deal by refusing, and concedes the logic: “bad money drives out good, and bad habits drive out good habits. If you gotta win the deal, maybe you do it.”
Harry’s worry deserves its own line: growth funds assume outcomes are equiprobable in size regardless of what they do — but “if I stuff Rory with $200 million before Rory’s ready for 200 million, that $10 billion outcome will be a $4 billion outcome.”

15. Rippling vs Deel, the bezel, and ARR that’s neither A nor R nor R

On the alleged Deel spy inside Rippling ($5,000/month payments), Rory draws the line at criminality: “you can be pretty driven without actually planting spies… if it trends into criminal liability, you probably have to find a new payroll provider.” He salutes the counter-intelligence: “very clever of the Rippling team to trap the person involved — you win, dude. They probably could have run him as a double agent, feeding false information” — straight Le Carré. Jason’s read on why Parker went public: “I guarantee you, it’s got to be worse. He would not do this otherwise — this stuff is so distracting.” Both doubt customers churn (“you know how much work it is to change payroll providers? I’m outraged, but not that outraged to do any work”), though it arms competing sales teams at the margin — and Rippling raising at $18BN right after was, in Rory’s words, “pretty damn clever.”
Jason’s number that reframes the scandal: a SaaStr survey of more than 2,000 B2B folks found 93% admit lying to win deals. “If 93% are lying about features, and you’ve just been handed billions — you really think none of them would plant someone at a competitor?” Harry protests the difference between fudging a roadmap and orchestrating espionage; Jason: “I’m not sure the line is as black and white as you think… there’s gonna be 100 of these in this environment.”
Rory reaches for Galbraith’s The Great Crash and the concept of the bezel — the standing stock of undiscovered embezzlement that grows in booms “‘cause nobody knows” and surfaces when the tide goes out. His practical shift: “we started really focusing on GAAP revenue now ‘cause ARR is a made-up number and GAAP number’s a fact.” The riff that closes it: today’s experimental ARR “doesn’t really recur, no way it’s annual if everyone can get out after a month, and it may or may not be revenue — neither A nor R, nor R.”

16. Buy or not buy: OpenAI at 300, Cursor at 10

Harry’s closing game splits the desk on OpenAI at $300BN: Rory — “not buy.” Jason — out of anything north of $100BN: “I just can’t make any decision well north of 100… all my decisions are bad.” Harry — “I’m the opposite of you two. I would buy the shit out of this. Escape velocity reached.”
Cursor at $10BN becomes the episode’s thesis in miniature. Jason: “if it’s a SaaS company with highly durable revenue, then Cursor at $10BN is a good deal” — probably 140-200% NRR on paper, “a massive moat that has destroyed its competitors… there’s nothing better than those metrics. I just wish I had 500 million.” But: his portfolio companies switch IDEs back and forth, his son is switching — “is this revenue durable? This is the question of the ages for us.” Rory’s back-of-envelope ceiling: if Cursor replaces all of Google, that’s a bit over a trillion — “3 or 4X from here… is that the best 3 or 4X you can do? I don’t know.”
Rory’s last word on why anyone plays at these prices at all: “the whole reason this business is awesome is there are singly amazing companies in every generation, and maybe these are they. When you do those companies, everything works, and you’re just so glad you bought them at any price.” And the sign-off is self-aware comedy: Jason prescribes Harry a $4.5BN fund with “a hard cap around five or six”; Harry recites “we like to stay small”; Rory, wise to him: “suddenly everyone remembers right at the end that we gotta stay on message… no wonder you’re a fundraising genius, Harry.”

Predicting the Earth with Josh Goldman: How KoBold Uses AI to Find Critical Minerals Helen Toner: OpenAI Reflections, Adaptation Buffers, and AI in Warfare