Summary

Jason’s Dreamforce takeaway is the episode’s tradeable core: for the first time ever, 100% of enterprise buyers are in market for AI tools simultaneously — versus the traditional 5% — and that’s warping every TAM and growth-rate estimate. “They’re not going to be in market every year for an AI tool. This will be a window that will disappear,” so today’s 5-10x growth rates could fall to 2-4x, and “if any kind of deceleration happens… everyone’s estimates on what’s going to happen here are wrong.”
Jason’s new vertical-AI filter: at Toast’s $22B in the biggest B2B vertical there is (restaurants), every vertical SaaS pitch must answer “why will AI make me much bigger than Toast?” — and deal sizes must be 10x, not marginally bigger, for venture math to pencil. His prediction: VCs will lose ~80% of their AI B2B investments because “we’re hyperfunding niches once again.”
On Revolut’s $3B raise at $75B (from $45B in 2024, on ~$3B revenue, making $1B, growing 60%): private markets have seized another IPO-ready asset, and Jason notes even at that price you’re making a TAM bet — Revolut is already as big as the biggest banks in England (one at ~$110B, Barclays ~$60B). Rory’s framing of the market vs founder debate: “the addressable market determines the size of the prize and the skill of the CEO determines who gets the prize.”
Poolside building its own 2GW data center (as prime developer, partnering with likely CoreWeave) is “the boiled frog of capital intensity” — you thought the business needed $500M to break even and suddenly it needs $5B. Rory’s read: they likely can’t buy capacity at scale because CoreWeave has committed $22B to OpenAI, $10B to Anthropic, $5B to Microsoft — a terrifying conclusion for SSI and Thinking Machines’ capital needs.
OpenAI’s corp-dev machine “has a ruthless instinct for weakness”: it has pawned off balance-sheet risk onto everyone else (including Oracle, whose debt-to-equity is now 4.6x and is “over their skis”), while Microsoft rationally stepped back — “shareholders should award medals to Satya, the CFO and their GC and hire someone else to do their technology.”
Rory’s crash framework: overinvestment is inevitable (“if it works at 10x, go 20; if it works at 20, go 30 — the only thing that stops you is when it hurts”), and “in a bull market, the most aggressive person will look the smartest just before the crash.” The unravel looks like the bandwidth bust: one marginal $2B data center sold for $1B and nobody builds again. But Harry’s counter is a B2B investment that wants 24/7 inference — three orders of magnitude more compute than affordable today — arguing demand absorbs the buildout.
Sharpest disagreement of the episode: Rory says investing today is “as tough as it’s ever been”; Jason says it’s “the easiest ever” — so many entrepreneurs, no gross-margin worries, LPs pushing go-go — even though Claude told him to model his current fund at only 2-3x. “When you’re most happy, you’re probably less likely to make money.”
Rapid-fire calls: Jason takes the bet Replit hits $1B ARR by end of next year (the TAM is every mediocre dev shop and WordPress agency dying); Harry disagrees — prosumer market, churn, “Lovable’s got a larger TAM because it is literally everyone.” On Deel vs Rippling, Rory (conflicted via Papaya) leans Deel’s TAM “despite the espionage thing”; Jason picks Rippling on installed-base defensibility.

Deep dive

1. Benchmark reloads with Everett Randall — and VC comp now loses to Meta engineers

Rory’s read on Everett Randall joining Benchmark from Kleiner (prior: Vista, Bond, Founders Fund — “five good ones in 8 years”): the two-months-ago “world is ending” narrative when a partner left was always wrong. The playbook is mechanical — “you draw a list… and you go and hire someone talented from one of the adjacent golden firms where the pitch is purely equal partner. Tick done.” Benchmark will be just fine too — likely Mamoon and Ilya are “wildly talented.”
Harry adds the sweetener: Benchmark is “extremely generous” with backdated carry, and being brought into a pool holding Fireworks, Mccor, Lora, Manis and others is “the Godfather offer I can’t refuse.”
Then the humbling pivot — Rory: VC used to be the best economic gig in tech, but there are people “vesting a billion dollars over four years at Meta because they wisely did computer science and AI at school 10 years ago.” Harry pushes back that a top-3 carry participant at a Thrive/GC/Lightspeed beats that over decades; Rory’s rebuttal: the Meta package is fully liquid stock in four years, while carry means owning “a ton of private stock” — and history shows 10-year periods of squat after crashes. “Venture is the get-rich-slow program.”
Jason’s supporting evidence: his 2017 fund should hit 5x on paper by year-end — “that’s a lot of years already… hopefully I’m not in a walker by the time I get my distributions.”

2. Revolut at $75B — Jason’s new heuristic and the TAM-exhaustion alarm

Revolut’s $3B raise at $75B (up from $45B in 2024), on roughly $3B revenue, making $1B and growing 60% — “another round where the public markets have ceded that business to the private market.” It could have gone public years ago.
Jason’s new heuristic, born of fear: “I would like to invest in startups that at $100 million ARR have 1% or less market share” — because in public B2B “there’s almost no one except Palanteer having an easy time north of a billion,” and even Clavio is “crushing it and still trading at 6x.” His confession: “I see TAM exhaustion everywhere — I never used to, even 18 months ago… We used to have more time.”
Rory’s pushback: a wide-open market where you “only need 1%” usually means you’re undifferentiated. Revolut started as a pointy niche (travelers with FX needs), got traction and margins, then the addressable market expanded — “if from day one they’d gone after everyone in Europe for all consumer banking, they’d have got spread out.”
Jason’s kicker on the $75B price itself: at that multiple you’re buying the undisputed winner at a premium, so “all of these things are TAM bets” — Revolut is already as big as the biggest bank in England (one at ~$110B, Barclays at ~$60B) in the country it’s domiciled in.

3. Market vs founder — and why Spotify actually won

Harry’s case for founder-driven TAM: he’s close to Daniel at Spotify, Alex at Deal, Nick at Revolute — “all of them have expanded TAMs sequentially, opened up more new chapters.” Harry then says, “I think you’re wrong,” while Rory argues that all three started in obviously huge notional markets (music consumption, payroll, fintech) and merely threaded the needle into adjacent empty space. “There are lots of thinly sliced SaaS markets people invested in in 2017-20 and you just ran out of space, and no matter how amazing the founder, there’s nothing they can do.” He explicitly rejects “the great man theory.”
Jason’s arcane Spotify theory, worth the price of admission: “All the other music startups based in the US got strangled at birth by lawyers” — IP litigation and radio-type licenses (Pandora) crushed margins, while “little old Spotify got going in a bunch of European countries that your average big-five record label didn’t really focus on,” got a better licensing deal, built critical mass, then gradually flipped the leverage. “Combination of great execution and a little bit of serendipity.”

4. Everyone’s in market — AI’s rhyme with 2020

Jason’s big point, fresh from Dreamforce: “because of AI, everyone’s being yelled at and told ‘go find a tool’… The fact that everyone’s in market instead of 5% of the market — which is our traditional metric in B2B — is warping how we think about market size. It’s like 2020 all over again… They’re not going to be in market every year for an AI tool. This will be a window that will disappear.”
Rory amplifies it as the Covid extrapolation error: “when you looked at the growth rate for Zoom in 2021… there’s not a human being on the planet who didn’t have a Zoom account by late ‘22. So growth went to 10%. If any kind of deceleration happens, everyone’s estimates on what’s going to happen here are wrong.”
Harry’s pushback — kept because it’s the live disagreement: Covid was “a temporary moment that did not sustain,” whereas applying that here means expecting AI to stop improving productivity, “which we would all disagree with.” Jason’s concession-with-teeth: software was no better in 2021 than 2015, and today’s software is radically better — “that’s the only similarity” — but the exogenous everyone-in-market condition is identical, and “every CIO’s neck is on the line. It will not last.”
The practical implication all three converge on: what should have been 5-7 years of buying decisions is compressed into 1-2 years, so you must win now — “showing up two years from now when 90% of top-500 American law has made a decision is just too late” — but if you’ve leaned in on valuation as growth goes from 5-10x to “a more prosaic 2, 3, 4x… you might be over your skis. And if you’ve done the number three, four or five player, you might be st out of luck.”

5. The hidden cost nobody budgeted: business process change

Jason’s Dreamforce detail: CIOs said onboarding and business-process-change costs are “the highest it’s ever been in their lifetimes” — they got the vendor price right but not the soft costs. “They’re not going to do that every year. We may go back to 5% being in market in 24 months instead of 100%.”
Rory’s synthesis for investors: “don’t confuse ‘25 growth rates with long-term growth rates” — there will be plenty of $500-800M-revenue vertical winners, but in a world where you don’t go public until $1B+, the question is how you avoid overpaying for them versus doing “de facto public investing at scale” in a Revolut.
Jason’s Toast test for any vertical AI deal: restaurants are “the largest vertical there is in B2B” and Toast is $22B — “so you’re really going to be worth $220 billion? Your vertical better be bigger than restaurants. And none of them are. Why will AI make me much bigger than Toast? That’s a tough question at the partners meeting.”

6. Vertical AI’s bull case: 10,000 customers spending $100K instead of $10K

Jason’s old Emergence-slide math: vertical SMB SaaS as ERP — get 10,000 customers at $10K/year and you have a $100M business, “and that proved true again and again. It’s just $100 million isn’t enough today.” The AI question: “can those same 10,000 spend $100,000? If they do, you get a billion… Will a plaintiff’s law firm that used to spend $100-200K spend a million because they don’t need humans anymore? If they do, it’s golden.” If deal size is only “a little bit bigger,” “we’re going to get crushed.”
Harry’s counter-example: Solve Intelligence, selling to IP law firms, all contracts over $100K. Jason’s rejoinder: so are the incumbents you’re displacing — likely LexisNexis and others — “it’s not 10 times larger. I’m not saying it won’t return your fund, but it’s got to be 10x higher for the math to pencil out in venture today.”
Harry’s case for the defense: legal was a horrible market for years because you were “selling workflows to people who didn’t care,” but “LLMs manipulate words — that’s the core of what they do — and it’s the most LLM-obvious market out there… the past is not predictive in terms of dollars you can extract.” His resolution: if the exit bar is $100-200M you don’t hit TAM exhaustion; if the bar is $1B, “you could hit it in some of these markets pretty quickly” — so returns boil down to entry valuation and exit-market health.

7. Sell into strength — and Harry’s three-pillar allocation

Rory’s changed mind on M&A, stated as a new rule: “If you get an offer and your TAM isn’t really accelerating, take it.” The classic Paul Graham never-sell advice fails at TAM exhaustion — “you can still keep growing but your value doesn’t. It’s not all like Revolut where every year you go 40 to 70 to 140 to 280.” His categorical call: “VCs are going to lose like 80% of their investments in AI B2B… We’re hyperfunding niches once again. We just don’t need that many legal apps or veterinarians that only treat cats.”
Harry’s meta-worry about the whole vintage: “we’re funding Sierra at $50 million ARR at $10 billion — assuming what, that it hits $10 billion ARR in 5 years? My gut tells me we’re overromanticizing verticals in the age of AI. We’re going to hit the same TAM exhaustion and it’s going to be worse because expectations are so high.”
Harry’s three pillars if he had infinite capital: the anointed winners (OpenAI, Anthropic), winners with great economics (Revolut, Deel), and really early — explicitly not “likely Mira Murati’s $2 billion at $10 billion” or “$300 million into Periodic Labs… a huge amount of money into a very still questionable early asset.” Rory’s needle: two of Harry’s three pillars are post-public-eligible anointed winners — “with two-thirds of your money you’d like to do public-style investing with a 2-and-20 structure. And the market seems to agree with you.” Rory notes the easiest way to make money in 2025 “is to take the very biggest companies and double down one more time.”

8. OpenAI > Oracle > Microsoft: who’s holding the risk

On OpenAI spending more with Oracle than Microsoft, Rory’s verdict: “Microsoft didn’t want to spend money irrationally, Oracle wanted to be in the game, and OpenAI seems extraordinary at reading other people’s needs and taking advantage of them… Shareholders should award medals to Satya, the CFO and their GC — and hire someone else to do their technology, because they haven’t shipped.”
Jason’s structural read: OpenAI needed “probably two orders of magnitude more” capital than Microsoft’s high-end model assumed; by effectively spinning it out at ~30% ownership, Microsoft escapes “an awkward situation… funding their subsidiary for eternity,” and Oracle — willing to tolerate much lower margins — replaces them.
Rory dismisses the “OpenAI is going to go bust” takes as dumb: “They have brilliantly pawned off all the risk on everyone else… ‘We need gigawatts of data centers, gazillions of chips — you all should do it. Go team. We’ve signed commitments and if we need them we’ll pay you one fine day with money we don’t yet have.’” The counterparties’ best case is being “the commodity compute provider to someone very rational who will grind you down at scale”; worst case is billions in fixed assets earning no return. On Oracle’s 4.6x debt-to-equity: “we said two weeks ago they’re over their skis and since then the stock’s down — we can claim an attaboy.”

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

Poolside — enterprise-focused coding LLM, no public product yet — announced building its own 2-gigawatt AI data center, and per documents Rory read, Poolside is the prime developer (partnering with likely CoreWeave), not doing a build-to-lease. His framing: “this game you thought was a software game is now a fixed-asset-at-scale game… If they’re 100% right, do all the other model builders have to do the same? Is there a conclusion here for the capital intensity of Safe Superintelligence, of Thinking Machines?”
Rory’s English lesson, delivered to Harry: “rationale is not the same as rationality — rationale is why you think you’re doing this; rationality is whether you’re right. We’ll know in 5 years.” The likely rationale: they rang CoreWeave and heard “I promised $22 billion to OpenAI, $10 billion to Anthropic, $5 billion to Microsoft — I got nothing for you.” So it’s build it, give up the dream, or pause until ‘27.
The VC horror framing: “It is the boiled frog of capital intensity. You think you’re in a business that needs $500 million to cash-flow break-even and suddenly you’re in a business that needs $5 billion, and you want bulldozers digging a hole somewhere in Texas. What the fk just happened?” Jason adds the competitive driver — Claude Code, GPT-5 Codex — plus disclosure that nobody expected Claude Code at a billion, Cursor at a billion, Replit approaching, so a $5-10B raise is “probably fundable today” when it wasn’t at inception. (Harry’s aside: he was rolled into Poolside’s first round via a pivot — “it’s like a 50x. Thank you, Eiso.”)

10. Bubble math: the bust definition, the bandwidth analogy, and the 1000x-inference counter

Harry’s homework: a bust requires a >20% asset-value drop plus productive capital leaving a market for 3+ years — “we will not have productive capital leave AI and data centers for more than 3 years,” so the bubble-callers are wrong. Rory accepts the definition, not the conclusion: if it goes wrong, it won’t be because the tech fails — “we over-extrapolated one year’s adoption… the diffusion is going to take 10 years, not two, and we’ve overinvested in capacity.”
Rory’s mechanism, via the bandwidth bust (boom ~‘96-‘99, then 5 years of nothing): “once there are existing assets at less than the price to build new ones, no one rationally builds new. If people don’t need the marginal data center built for $2 billion and the only offer is a billion, that’s what they take.” Will it happen? “If I had certainty on that, you think I’d be wasting my time talking to you, Harry? I’d be trading as we speak.”
Harry’s bull-side evidence: a B2B AI investment he’s making this week wants 24/7 inference, 20 different passes, 365 days a year — “three orders of magnitude more inference than you’d want to use today. If it was available cost-effectively, they would consume all of it.” Harry’s caveat lands anyway: “that was the key sentence — cost-effectively. The bet you’re taking is as price comes down, that gets used up.” (On the obvious trade — “should we not just be plowing money into Nvidia?” — Jason: “we’re already all long Nvidia in our QQQ and 401ks. Just depends how much more you want to concentrate.”)

11. Temporal diversification is dead — and venture’s cycle math

Rory’s core risk algorithm: “in a bull market the most aggressive person will look the smartest just before the crash… the correct algorithm is how aggressive can I be to be one step below the level that blows up in my face, such that I can power through.” The most aggressive 2021 funds hit acute problems; the merely aggressive “took a little licking, got a lot of the upside, and kept rolling.”
Harry’s ground truth from a large foundation: “we’re back to 18-month cycles, Harry. We love that you’re three years, but you’re the only one, dude.” A guest’s dry translation: the only diversification anyone’s getting is “an LLM or two” per fund — “you can go from ChatGPT-4 to 5 in one fund.”
A guest’s reckoning-in-waiting: venture returns over the last 5 years have been “massively lower than public market returns,” and someone at every endowment has a spreadsheet saying “I need 3-400 basis points more than the liquid S&P and we’re not getting it.” The long-run facts still favor the asset class — Cambridge 30-year pooled returns run ~600bps over small cap — “venture is worth doing on aggregate,” but it’s massively cyclical: underfunded ~‘87-‘95, money roaring in by ‘96, unwinding 2000-2010, massively underfunded by 2010, and now an unusual 15-year cycle because equity markets have been so forgiving. “Since 2010 there has never been a period longer than a year of substantive correction — the only one was ‘22-‘23ish, and God bless ChatGPT, it ended that.”

12. Tough vs easy: the episode’s cleanest disagreement

Rory on today: “Frankly, it feels tough today. As tough as it’s ever been.” Stuff is working but there’s a lot of variance, a huge amount of capital, and you’re way out on the risk curve… you got to enjoy the process, not the outcomes." Great times to invest were 2010 and 2015; tough in ‘21 and now.
Jason’s flat contradiction: “No, I think this is the easiest ever” — so many entrepreneurs, “change is when you make money in venture,” no gross-margin worries in B2B, and LPs “still pressuring you to go go.” Then the honest asterisk: he uploaded his fund analysis to Claude, which told him to assume 40-50% lower returns — “you might end up with only a 2x to 3x fund and it’s okay.” His summary of what’s wrong: “just those three small things — entry point, ownership and margins. The rest is great.” Rory’s closing needle: “when you’re most happy, you’re probably less likely to make money.”

13. Erotica, content moderation, and the rapid-fire: Replit to $1B, Deel vs Rippling

On OpenAI allowing erotica (erotic creation is reportedly likely Grok’s largest image/video use case), a guest recalls a 2022 roleplaying-game company that had to switch off OpenAI because demand was for conversations OpenAI wouldn’t support — “human beings like to talk about sex, shock horror.” The real warning: unlike social platforms, “what’s super clear on ChatGPT is you are writing the content” — no we’re-just-a-pipe defense — so “the content moderation job at ChatGPT is going to be a hot seat for the next 5 years, and erotica is not going to be the hardest problem they face.” Jason’s unease: “just like we had to trample copyrights to get these off the ground… I think it’s just the wedge, just like everything Sam says.” (Telling exchange: would you share your ChatGPT history? Jason: “Absolutely — except for another venture firm.” A guest: “No way. I would not be remotely comfortable.”)
Replit to $1B ARR by end of next year — Jason: “It’s only 4x. I’m all in,” because vibe coding is “so much better than when I started 110 days ago,” 20-30% of the last YC class’s sites looked vibe-coded, and the TAM is “mediocre outsourced dev shops and WordPress agencies — they’re all going to be gone.” His deeper worry: when any 19-year-old can ship a really good product, “the classic ways we could judge software at that stage go out the window — this is super disruptive for early-stage investing.” Harry disagrees: Replit is prosumer, “Lovable’s got a larger TAM because it is literally everyone,” and cohort maturation and real churn are coming. Rory’s synthesis: as a tools market it flattens; as labor-spend compression the TAM clearly supports it — “the only remaining question is Replit or Lovable or both.”
Deel vs Rippling — Rory half-punts (Papaya is an adjacent investment) but gives the structure: payroll is a giant horizontal (“ADP $100B+, likely Workday ~$70B, likely Paychex ~$50B… if you don’t pay people on Friday, you don’t have workers Monday”), the US is a grinding replacement market, while internationally “it’s much more the wild west” with no ADP-scale vendor — so “the TAM and competitive matrix is more attractive for Deel, despite a little distaste for the espionage thing.” Jason, despite calling Deel’s pain point more acute and Deel “a much more agile company than I realized,” picks Rippling: “I’m not going to say in the age of AI that having these massive installed bases isn’t a huge asset.”
The closing self-examination, prompted by Rory: if $1B ARR is “still early,” why write checks at $1M ARR? Jason: “All my losses are when I strayed out of my sweet spot… The worst advice I ever got was to take more risk in venture.” Rory’s parting observation on the asset class itself: “you have Jason doing $5 million into companies doing a million, and people doing half-billion-dollar investments into companies doing $5-6 billion in revenue, and we think of them the same — it’s obvious those two things are so not like each other that it’s absurd. But that’s the world we live in now.”

Verification Notes

Several mid-episode guest turns remain attribution-ambiguous in the raw captions and are marked [Speaker?] in the transcript.
Garbled entity names are retained or softened rather than silently resolved.
“Making $1B” is retained instead of asserting that the captions said $1B profit.

Celebrities Fight Sora + Amazon’s Secret Automation Plans + ChatGPT Gets a Browser (Preview) Understanding the AWS Outage, Resiliency Realities and Innovation Urgency, F1 Heads to Apple TV