Summary

Jason Lemkin’s core call is TAM exhaustion — “I see TAM exhaustion across my portfolio… I see TAM exhaustion everywhere,” something he didn’t think about even eighteen months ago. His new heuristic: invest only where a company at $100M ARR has 1% or less market share, because in public B2B “there’s almost no one except Palantir that’s having an easy time north of a billion” — even Klaviyo trades at 6X. His new sell rule, contra Paul Graham: if you get an M&A offer and your TAM isn’t accelerating faster than revenue, take it — “VCs are gonna lose, like, eighty percent of their investments in AI B2B.”
The episode’s sharpest mechanism: because of AI, 100% of buyers are in market at once instead of the traditional 5%, warping every growth signal. Lemkin: “It’s like 2020 all over again… they’re not gonna be in market every year for an AI tool. This will be a window that will disappear.” Rory O’Driscoll calls it “the COVID mistake” — Zoom’s growth went to 10% once everyone had an account — and warns “if any kind of deacceleration happens, everyone’s estimates on what’s gonna happen here are wrong”: expect 5-10X growth to fade to “a more prosaic 2, 3, 4X,” with the number three-to-five players “shit out of luck.”
Vertical SaaS fails the Toast test. Toast is the largest vertical in B2B — restaurants — and is worth $22B; “your vertical better be bigger than restaurants, and none of them are.” The partners-meeting question Lemkin says nobody can answer: “Why will AI make me much better than Toast?” The math only pencils if AI lifts deal sizes 10X — can 10,000 SMBs that paid $10K/year pay $100K? “If they do, it’s golden.” Rory’s counter: legal is “the most LLM-obvious market out there,” but “going in at pre-money equal to your total TAM” means “you’re never gonna make a dime.”
Rory’s read on where returns actually are: “the easiest way to make money in 2025 is to take the very biggest companies and double down one more time.” Revolut’s $3B raise at $75B (up from $45B, revenue ~$3B, making $1B, 60% growth) is the public market ceding another winner to privates — and Harry’s own three pillars (anointed winners, winners with great economics, truly early) are, per Rory, two-thirds “public style investing with a two and 20 comp structure.”
On the compute complex: OpenAI has “brilliantly palmed off all the risk on everyone else” — “we’ll pay you one fine day with money we don’t yet have” — while Microsoft rationally stepped back and Oracle (4.6X debt-to-equity, “over their skis”) took the bet. Poolside going prime on its own 2GW data center is “the boiled frog of capital intensity”: “you think you’re in a business that needs five hundred million to cash flow breakeven, and suddenly you’re in a business that needs five billion, and you own bulldozers digging a hole somewhere in Texas.”
The two veterans flatly disagree on the moment: Rory says “frankly, it feels tough today, as tough as it’s ever been” to write checks; Lemkin says “this is the easiest ever” — but only “to have a checkbook and to feel smart about yourself,” not to make returns. Nobody is doing temporal diversification (funds every 18-24 months, LPs “back to eighteen-month cycles”), venture has “massively” lagged public markets for five years, and Rory’s iron law: “in a bull market, the most aggressive person will look the smartest just before the crash.”
On the lightning round, Lemkin bets Replit hits $1B ARR by end of next year (“it’s only 4X, I’m all in”) because vibe coding erases “crappy WordPress agencies and terrible offshore dev shops” — and breaks early-stage diligence: when a 19-year-old’s product is “really, really good at 30 pre, the classic ways we could judge software go out the window.” Harry disagrees (Replit is prosumer; Lovable’s TAM “is literally everyone”). Lemkin picks Rippling over Deel; Rory, citing an adjacent Papaya investment, says Deel’s TAM and competitive matrix look better “despite the espionage thing.”

Deep dive

1. Benchmark hires Everett Randall — same playbook, and VC is no longer tech’s best gig

Rory’s read on the hire: two months after a partner left and everyone cried “the world is ending,” Benchmark did “exactly what they always do” — draw a list of top firms with good young people, call whoever they’ve overlapped with on deals, and pitch purely equal partnership. “Mission accomplished. On they go.” Kleiner will be fine too — “Mamoon and Ilya are wildly talented people.”
Jason’s gloss on the resume — Vista, Bond, Founders Fund, Kleiner, Benchmark in eight years: “if you’re ambitious today, you wanna go fast… no need to stop at anyone in the B tier.” Harry adds the sweetener: Benchmark’s back-dated carry pool (Fireworks, Manus, others — some names garbled in captions) is “the Godfather offer I can’t refuse,” per Rory.
The humbling context Rory insists on: the best AI engineer at “Facebook Meta” is vesting a billion dollars over four years in liquid stock. Harry argues a top-three carry participant at a Thrive or Andreessen beats that over decades; Rory: “you admitted the truth in the last sentence” — over 20-30 years yes, but “in terms of liquid stock cash payment, it appears to be the best gig on the planet right now.”
Venture remains “the get rich slow program”: Rory’s first carry check came fast, then “a 10 or 12 year period of squat” after the Nasdaq fell 80%. Jason’s 2017 fund should hit 5X on paper this year, but — riffing on Brian Halligan toasting Benchmark One’s 20-year vintage — “I’m not sure I wanna wait 20 years for my wine. I’d like a few sips tonight.”

2. Revolut at $75B: the public market cedes another one

The raise: $3B at $75B, up from $45B in 2024, massively oversubscribed. Rory: “the public markets have ceded that business to the private market” — roughly $3B revenue, making $1B, growing 60%, “could go public anytime it wants.”
The scale check that makes it a TAM bet: at $75B Revolut is already as big as England’s biggest bank — one at ~$110B, Barclays around $60B. “When you’re paying seventy bill, you’re buying the undisputed winner… you’re making some kind of the-TAM’s-even-bigger-than-you-think bet.” If you were writing the Revolut memo at $75B, you’d be asking the same question Jason asks at a $25M pre-round: how big can this thing get?

3. Jason’s conversion: TAM exhaustion is everywhere, and 1% share is the new filter

The new heuristic: “If I could, I would like to invest in startups that at 100 million ARR have 1% or less market share.” Evidence from publics: “there’s almost no one except Palantir that’s having an easy time north of a billion” — even Klaviyo, “crushing it,” trades at 6X.
The confession: “I see TAM exhaustion across my portfolio… Even eighteen months ago, I didn’t think about it. I see TAM exhaustion everywhere.” The old model hid it — do the A at 25 post, exit at a billion, and “TAM exhaustion is someone else’s problem because you’ve distributed twenty-four months after the IPO.” Staying private longer, where “a billion dollars doesn’t even count as an exit,” broke that.
Rory agrees on TAM but rejects the heuristic: a wide market where you “only need 1%” means “you’re probably undifferentiated.” The best pattern is Revolut’s own — a pointy niche (travelers with FX needs), good margins, then the market expands around you: “the best of all things is when, as you grow up as a company, your TAM grows up as well.”

4. “The addressable market determines the size of the prize” — the great-founder debate

Harry’s claim: the best founders unlock TAM — he cites Daniel at Spotify, Alex at Deel, Nick at Revolut, all of whom “expanded TAMs sequentially.” Rory, “to pile on and be direct, I think you’re wrong”: all three started in obviously huge markets, and “there are lots of thinly sliced SaaS markets that people invested in in 2017, ‘18, ‘19, ‘20, and you just ran out of space — no matter how amazing the founder is, there’s nothing they can do.” His formula, worth framing: “The addressable market determines the size of the prize, and the skill of the CEO determines who gets the prize.”
Rory’s arcane theory of why Spotify won: US music startups “got strangled at birth by lawyers” over IP, while “little Spotify got going in a bunch of European countries that your average big five record label didn’t really focus on” — better licensing, critical mass, rising leverage against labels who’d tormented Pandora and every US subscription player. “Great execution and a little bit of serendipity to keep them away from the fray.”

5. The episode’s big mechanism: everyone is in market, and it’s warping every signal

Jason, fresh from Dreamforce: because of AI, “everyone’s in market for the first time forever.” Every law firm is being yelled at to “go find a tool” and dropping $50-150K without blinking — versus the traditional B2B baseline where ~5% of a market is buying at any time. “It’s like 2020 all over again, when everyone was in market for a contact center or e-signature or a digital events tool like a Hopin. Everyone was in market and then they disappear the next year. They’re not gonna be in market every year for an AI tool. This will be a window that will disappear.”
Rory, crediting the point as “huge” and crisply put: “you’re making the COVID mistake” — Zoom’s 2021 growth looked structural until “there’s not a human being on the planet who didn’t have a Zoom account” and growth went to 10%. “If any kind of deacceleration happens because of any kind of saturation or slowdown, everyone’s estimates on what’s gonna happen here are wrong.”
Jason’s own caveat on the analogy: unlike 2020, “software is radically better today.” The exogenous piece is organizational panic — “every CMO is told, bring in an AI tool or you’re gonna get fired. That will not last.” And the hidden cost: Dreamforce CEOs say onboarding and business-process change costs are “the highest it’s ever been in their lifetimes” — they priced the vendor, not the change. “We may go back to 5% being in market in 24 months instead of 100%.”

6. The corollary: land the market now, then model 2-4X — not 10X

Harry’s turn of the argument: doesn’t this make the case for owning a market immediately, a la Harvey? Rory: exactly — “what should’ve been a steady progression of company-by-company decisions over five to seven years has been compressed into the next one to two years, and then roll it out and stick with it for the following five.” Show up two years from now, when 90% of big American law has decided, and “it’s just too late.”
The deceleration math to underwrite: growth rates go “from the unprecedented 5X, 10X that you’re seeing this year to a more prosaic 2, 3, 4X, which is still damn amazing. But if you’ve leaned in too much on valuation, you might be over your skis. And if you’ve done the number three, four, or five player, you might be shit out of luck.” Jason adds that switching is exhausting — buyers “settle on whoever they bought.”

7. Vertical SaaS fails the Toast test — and the new rule is sell

Jason has changed his mind on vertical SaaS: Toast is the largest vertical in B2B — restaurants — and is worth $22B — “your vertical better be bigger than restaurants, and you know what? None of them are. Why will AI make me much better than Toast? And that’s a tough question to answer at the partners meeting.” The excess: “we just don’t need that many legal apps or veterinarians that only treat cats.”
His new M&A learning, explicitly contra the classic Paul Graham never-sell advice: “if your TAM isn’t really accelerating, take it.” You can keep growing into TAM exhaustion “but your value doesn’t.” The prediction: “VCs are gonna lose, like, eighty percent of their investments in AI B2B… We’re hyper-funding niches we shouldn’t be.”
The exemplar of the euphoria: Sierra — “there’s no one better than Brett Taylor” — funded at $50M ARR at $10B, implicitly assuming $10B ARR in five years. “My gut tells me we’re over-romanticizing verticals in the age of AI… it’s gonna be worse because expectations are so high.”
Rory’s synthesis for the defense: these wedge products (document recognition, voice bots) genuinely build value — but “if you go in at pre-money equal to your total TAM, you’re never gonna make a dime… assuming that every market is as big as the biggest market is the fatal error.” And the brutal test: “Are you making a perfectly good product for a perfectly sensible world that no one gives a shit about? Because you’re not Revolut.”

8. The 10X deal-size question, with legal as the test case

Jason’s old vertical-ERP math, stolen from an Emergence slide: get 10,000 SMBs to $10K/year and you have a $100M business — “it’s just 100 million isn’t enough today.” The AI-era version: “will people really spend $100,000, small businesses, on your same vertical agent software?” A plaintiff firm that spent $100-200K paying $1M “because they don’t need humans anymore — if they do, it’s golden.” But deal size must be “10X what it was 24 months ago,” and confusing everyone-in-market with bigger deals is the trap: “if it’s just a little bit bigger deal size, we’re gonna get crushed.”
Harry’s counter from his own book — Solve Intelligence, selling to IP law firms, “all of their contracts are over 100 grand, several hundred grand in a lot of cases.” Jason: so are the LexisNexis contracts you’re displacing — “it’s not 10 times larger.”
Rory’s case for the defense: legal was historically horrible — “selling workflows to people who didn’t care” — but “LLMs manipulate words… lawyers, it’s the most LLM-obvious market out there.” The past isn’t predictive; it could be 10X. The slicing risk survives: “if the bar for an exit is a billion, then you could hit TAM exhaustion in some of these markets pretty quickly. The ultimate return boils down to entry valuation and the healthiness of the exit market.”

9. Where the money actually is: double down on the anointed

Harry’s three pillars with infinite capital: the anointed winners (OpenAI, Anthropic), the winners with great economics (Revolut, Deel), and the truly early. What he’d skip: “your Mira Murati’s 2 billion at 10 billion, your 300 million into Periodic Labs — a huge amount of money into a very still questionable early asset.”
Rory holds up the mirror: two of Harry’s three pillars are “effectively post-public eligible anointed winners” — meaning two-thirds of the money is “public style investing with a two and 20 comp structure.” His summary of the whole market: “It looks like the easiest way to make money in 2025 is to take the very biggest companies and double down one more time” — and most of the dollars agree.

10. OpenAI plays everyone; Oracle takes the bet Microsoft declined

Rory on OpenAI spending more with Oracle than Microsoft: “Microsoft didn’t want to spend money economically irrationally, and Oracle wanted to be in the game. OpenAI seems to be extraordinarily good at divining other people’s needs and taking advantage of them… a ruthless instinct for weakness.” Microsoft’s shareholders “should award medals to Satya, their CFO, and their GC — and hire someone else to do their technology, ‘cause they haven’t shipped.”
Jason’s structural read: OpenAI needed “probably two orders of magnitude more than Microsoft’s high-end model” of required capital; by “de-acquiring” it for ~30%, Microsoft escapes “funding their subsidiary for eternity” and hands the low-margin hosting to Oracle.
Is Oracle over its skis at 4.6X debt-to-equity? Rory called it two weeks ago and, with the stock down since, claims “an attaboy.” The deeper point: OpenAI has “brilliantly palmed off all the risk on everyone else” — “we’ll sign commitments, and if we need them, we’ll actually pay you one fine day with money we don’t yet have.” The provider’s best case is “commodity compute provider to someone very rational who’s gonna grind you down at scale”; the worst case is billions in fixed assets that don’t return.

11. Poolside’s 2GW data center — the boiled frog of capital intensity

Poolside — building an enterprise coding LLM, still no publicly launched product — announced its own two-gigawatt AI data center, and not as build-to-lease: it’s going prime, with CoreWeave as partner. Rory: “if they’re right or if they’re wrong, either way it’s terrifying” — the conclusion smart people have reached is that “this game that you thought was a software game is now a fixed asset at scale game,” raising questions about the capital intensity of superintelligence and thinking machines.
The likely trigger is that capacity simply can’t be bought: “you’d ring CoreWeave and they’d say, look, I promised twenty-two billion to OpenAI, ten billion to Anthropic, five billion to Microsoft — I got nothing for you.” Rory’s pointy distinction: “rationale is why you think you’re doing this; rationality is whether you’re right — we’ll know in five years.”
The VC nightmare framing: “It is the boiled frog of capital intensity… You think you’re in a business that needs five hundred million to cash flow breakeven, and suddenly you’re in a business that needs five billion, and you own bulldozers digging a hole somewhere in Texas. Oh my God, what the fuck just happened?” Jason’s twist: Poolside likely did not model coding tools reaching $1B in revenue when it started — competition exploded, but so did the category, which is exactly why “it’s probably fundable today” for $5B-$10B raises when it wasn’t at inception. Harry, an early investor via the pre-pivot company (“Thank you, Iso” — a ~50x), notes the build is also a bet on their own future ability to fundraise.

12. How the unravel would look — bust mechanics vs bottomless inference demand

Rory’s bust anatomy, via the bandwidth bust of ‘96-2000: growth slows more than modeled, the marginal player cuts purchases, “instead of having a shortage of data center capacity, you have a mild glut” — and once a $2B data center has to sell for $1B, “no one’s gonna build another data center for two billion.” Crucially, not a technology bust: “it’s still a dominant long-term trend, but the diffusion is gonna take 10 years, not two, and we’ve over-invested in capacity.” On Harry’s bubble-definition musings: that’s the bust definition, and “of course” he can’t time it — “if I had certainty on that, you think I’d be wasting my time talking to you, Harry?”
Jason’s live counter-datapoint: he’s making a B2B AI investment this week that wants 24/7 inference — twenty passes through the Claude API running round the clock, “three orders of magnitude more inference than you really wanna use today. If it was available cost effectively, they would consume all of it.” So should we all just plow into Nvidia? “We’re already deep. All of our QQQ and 401k are already long Nvidia.”
Rory’s discipline: overinvestment is inevitable — “if it works at 10X growth, go 20. If it works at 20, go 30. The only thing that stops you doing it is when it hurts.” The reminder he repeats twice: “in a bull market, the most aggressive person will look the smartest just before the crash, because the more risk you’ve taken, the more money you’ve made.” The correct algorithm: “how aggressive can I be to be one step below the level of aggression that blows up in my face in the crash.”

13. Nobody’s diversifying, and venture is losing to the S&P

Jason’s jab: “no one’s doing temporal diversification now, are they? Everyone’s just raising a fund every eighteen to twenty-four months… You can go from ChatGPT four to five in one fund. We’re getting an LLM or two.” Harry confirms from an LP: “we’re back to eighteen-month cycles… you love that you’re three years, but you’re the only one, dude.” Salesforce has already invested $850M of its $1B AI fund.
The number that gnaws at Rory: “the venture return over the last five years has been massively lower than the public market returns” — and somewhere an LP with a spreadsheet needs 300-400bps over the S&P to justify illiquidity. “If this AI boom doesn’t come good, the infinite spigot venture capital faucet is gonna get impacted.”
The long view: the Cambridge 30-year pooled return runs ~600bps above small cap — “venture is worth doing on aggregate over time” — but it’s massively cyclical, and “riding those is brain dead hard.” His underfunded eras: roughly ‘87-‘95 (when the internet gestated) and 2000-2010. Since 2010 there’s been no correction longer than a year — only ‘22-‘23, “and God bless ChatGPT, it ended that.”

14. “As tough as it’s ever been” vs “the easiest ever”

Asked when he enjoyed his 30 years most, Rory answers “Tomorrow” — then reframes: it was clearly a great time to invest in 2010-2015, tough in ‘21, and “frankly, it feels tough today, as tough as it’s ever been… when you’re writing checks, you’re like, wow, it’s sobering, the risk you have to take here to play.”
Jason’s flat disagreement: “No, I think this is the easiest ever” — change is when you make money in venture, gross margins no longer worry B2B investors, and LPs are “still pressuring you to go, go, go.” The distinction that saves both claims: “it may not be the easiest time to make returns, but it’s the easiest time to have a checkbook and to feel smart about yourself.”
The confessional detail: Jason ran his own fund analysis through Claude, which told him “you should assume 40 to 50% lower fund… you might end up with only a 2X to 3X fund, and it’s okay.” Rory’s coda: “in investing, when you’re most happy, you’re probably less likely to make money.”

15. Erotica is the wedge; content moderation is the real hot seat

OpenAI will allow erotica — already the largest use case on Grok’s image and video generation. Rory’s story from ‘22, pre-ChatGPT: an online role-playing-game startup told him it had to switch off OpenAI because user demand was for conversations OpenAI wouldn’t support — “another LLM provider, who shall remain nameless, was very happy to.” His shrug: “Human beings like to talk about sex. Shock horror. The question is which businesses meet that demand, and how.”
Jason’s worry is the pattern, not the product: “just like in the beginning, to get these off the ground, we had to trample copyrights… All of my IP is stolen. Everything I’ve written, all my videos were taken without my consent.” Erotica “is just the wedge… I worry it’s much more than that, just like everything Sam says” — noting Sam Altman already walked the announcement back on Twitter.
Rory’s bigger claim: social platforms hid behind “we didn’t write the content — we’re just a connection mechanism,” but “what’s super clear on ChatGPT is you are writing the content” — so on bad medical advice and politics, OpenAI sits in the crossfire. “The content moderation job at ChatGPT is gonna be a hot seat for the next five years, and I don’t think erotica is going to be the hardest problem they face.”
The kicker question — would you share your ChatGPT history? Jason: “I would not be remotely comfortable.” Rory: fine, “except if it were another venture firm” — he was more terrified of Spotify’s old music-sharing feature exposing his “sad ass songs.”

16. The game: Replit to $1B, Rippling over Deel, and the $1M-ARR paradox

Replit at $1B ARR by end of next year: Jason takes the bet (“it’s only 4X, I’m all in”); Rory disagrees on TAM clarity; Harry disagrees too — Replit skews prosumer while “Lovable’s got a larger TAM because it is literally everyone,” and cohort maturation and real churn are coming. Jason’s reasoning is bigger than the bet: Replit is “so much better” than when he started 110 days ago, maybe 20-30% of recent YC demo-day sites look vibe-coded, and “when a 19-year-old founder walks into 20VC and the product is really, really good at 30 pre, the classic ways we could judge software at that stage go out the window.” Rory’s clean summary: as a tools market it flattens; as compressed labor spend replacing “crappy WordPress agencies and terrible offshore dev shops that never finish a project,” the TAM clearly supports a billion — the only question is Replit, Lovable, or both. Harry dares Jason to invest as a top-0.1% power user; Jason: “gotta be 10 [million] to make enough money, but I’m with you.”
Deel vs Rippling: Rory half-punts, citing an adjacent Papaya investment, but gives the structure — the US is a served market where payroll supports ADP at $100B+, Workday ~$70B, Paychex ~$50B, so Rippling is grinding through replacement; internationally “it’s much more the Wild West” with no ADP-equivalent, which is Deel’s opening. His margin call: “the TAM and the competitive matrix is more attractive for Deel — despite a little distaste for the espionage thing.” Jason picks Rippling anyway: “I’m not gonna say in the age of AI that having these massive installed bases isn’t a huge asset,” and Deel proved agile enough that “it can build everything Rippling has.”
Rory’s closing trap, sprung on Jason: “If a billion is still early, why on God’s green earth are you writing checks to people doing a million dollars in ARR?” Jason’s honest answer: sweet spot — “all my losses are when I strayed out of my sweet spot… take more risks was the worst advice I ever got”; he has no unique value to add to a Deel or Rippling. Rory’s parting observation on the industry: the same “asset class” now contains $5M checks into $1M-ARR companies and half-billion-dollar checks into companies doing $5-6B in revenue — “those two things are so not like each other that it’s absurd. But that’s the world we live in now.”

OPPO’s Wan Yulong on Mobile AI That Understands You | Agent #7 When Alibaba Entered Global AI Coding: 60 Days on the Battlefield | A Conversation with 叔同, Founder of Qoder