Summary

Meta’s $14.83B Scale AI deal is messaging, not M&A. Meta bought 49% non-voting, the investors immediately dividended the $14B back out, and Alexandr Wang left for Meta — leaving Meta owning 49% of “an $800 million declining revenue business with so-so gross margins” and none of the cash. Harry’s blunt read: Zuck was behind with Llama and needed the public markets to see a front-runner; at under 1% of a $1.7T market cap and one quarter’s $15B free cash flow, “he doesn’t give a [__]… Roll the dice.”
Scale is “a dead man walking instantly” — not because it lost its founder, Rory insists, but because labs may hesitate to keep giving frontier work to a supplier 49%-owned by a direct competitor. Hundreds of millions of data spend is already reallocating; Handshake’s demand tripled within the week, and Garrett names the sector’s only durable moat: “access to an audience.”
Founder-governance doctrine, sharpened on the Discord rumor: Rory’s data says 90% of B2B IPOs still had the founder as CEO, so “if the founder CEO is out, I’m out” — Jason would take a $10B IPO with Jason Citron over $20B with a hired gun. Replacing a CEO is “open heart surgery and you got a one in three chance of dying,” and if the fired founder is surprised, “you have massively failed as a board member.”
The Microsoft–OpenAI endgame turns on one undefined word. Rory thinks ambiguity favors Microsoft because the relationship isn’t existential for them; Jason counters they have no leverage — the profit-sharing ends at AGI and AGI is close. The line of the episode: “We’re going to hit AGI when Microsoft and OpenAI litigate what AGI is… Now half a trillion dollars of value depends on that word.”
The IPO window is open because issuers “bribed the buying public to like IPOs again” — buyers who bought the 2021 vintage and were looking back from 2022 after losing 30-40% are now up 70% on average, 250% on Chime and Circle. The real unlock is companies in the $200M–$500M range dusting off plans; Databricks is already “managing to Wall Street” after an analyst summit and “could file next week and no one should be shocked.”
Private markets still misprice revenue quality, Rory’s standing mystery: Ramp raised at $16B on roughly $700–800M of revenue while fintech trades 2–4x in public comps, and 80%-gross-margin ARR gets the same multiple as 20-30%. But Harry wins the exchange — a round every three months at 1% dilution is a momentum and relevance strategy: “You got to do it. They’re crushing.”
The old guard’s only right to win is its installed base. The incumbent pitch is “I’m the system of record, I can add an agent”; the new guard runs on any system of record — so Dropbox versus $7.2B Glean is “not a relevant question” because Dropbox has no enterprise base to defend. Rory: “I’m starting to lose confidence in the old guard… they’re still too slow.”
Positioning beats opinion in the quickfire: betting against a Chinese model hitting #1 in evals this year at sub-20% odds is “a Bay Area hubris” — “there’s no world where we’ve got all the smart people and they don’t.” On the S&P, Rory refuses the forecast but reveals the book: he’s ~75% equities and “mildly terrified”; Jason is 100% in, “back where I was in 2008.”

Deep dive

1. Meta paid $14.8B for Scale — and it was mostly messaging

The facts as the panel lays them out: Meta invested $14.83B into Scale AI for 49% non-voting control, investors took the money straight back out as a special dividend, and Scale’s CEO moved to Meta to run “seemingly a much broader portfolio.” Scale, founded ten years ago, had grown to $800–900M revenue selling data labeling and increasingly PhD-level post-training work to “five or six amazing model provider companies,” and had just raised at $14B.
Harry’s read, which the room adopts wholesale: “Zuck was behind with Llama. He needed to show the public markets that they were still a front runner… The price is less than 1% of the market cap. Honest, he doesn’t give a [__] and he gets good talent with some people that he likes. Roll the dice.”
Jason’s precedent math: Salesforce paid $750M for Quip in 2016 — “a word processor that barely worked, for one dude” (Brett Taylor) — and 20x-ing that for the 2026 exit environment lands almost exactly on this deal. Seven companies can now write stunning checks that cost one quarter of free cash flow: Meta’s cap is $1.7T, quarterly FCF $15B. “They’re literally thinking, ‘Tick, that’s Q2 done. What am I doing in Q3?’”
Rory pieces the logic to its absurd end: the $14B left the company, it wasn’t a share repurchase, so Meta owns 49% of a business that no longer has the cash — “an $800 million declining revenue business with so-so gross margins.” Jason’s summary: “It’s the worst purchase ever of all time for assets.” And it’s a clever DOJ getaround — the deal is already closed, and “it would be very hard for any kind of judicial review to unwind it in part because it’s so weird.”

2. Scale post-deal: “dead man walking”

Jason doesn’t hedge: “I love you, but it’s a dead man walking instantly. There’s no way that scale can recover from losing its founders… it can manage 200 of its 800 for 24 months but it’s a dead company.”
Rory’s precision — worth keeping: it’s not founder loss per se, “a company can survive the loss of one person.” It’s what the structure signals about “your ability as a customer of scale AI to continue doing business with them when 49% of the company is owned by one of your competitors in the LLM space” — the same dynamic as Windsurf being turned off by what was likely Anthropic.
Garrett’s over/under on Scale revenue in a year: over $100M — which Rory flags as savvy and probably factually correct, “because it takes a long time to go from 800 to 100.” Meanwhile Meta is “leaving a little pot of honey” — $200M in the bank that can last a long time once the cashed-out employees drift away. Jason proposes a future study of “the AI stubs… We got a big one called OpenAI, which was sort of bought by Microsoft. They’re trying to undo that one right now.”
Will rivals copy the move? Rory thinks not: labs may take work in-house — “I’m giving too much information to a third party” — or buy a small provider, but nobody else will “hire the CEO, buy 49%, give Harry the money, and let Harry dividend it out and pretend it’s still a company.”

3. Handshake’s windfall — and the meta-lesson for founders

Garrett’s numbers: demand tripled within the week of the announcement, he’s running on three and a half hours of sleep for ten days, and “hundreds and hundreds of millions of dollars of spend is trying to be reallocated” — the binding constraint is delivering volume. His moat claim: “the only durable moat in the entire human data business is access to an audience,” with frontier demand shifting toward audio, tool use, agentic trajectories, and domain experts in science, finance, law and medicine.
The $4B hypothetical: would Handshake sell to Microsoft tomorrow? “Absolutely not.” Jason deliberately plays the contrarian and tells him to take it, with the caveat that “no one’s opinion on what they would do at a hypothetical offer is worth a damn and what people actually do when the money comes into view tends to be very different on both sides.”
When Garrett reaches for the mission statement, Jason cuts him off: “No, that’s not the takeaway… the takeaway is stay in the [__] game long enough. Don’t burn too much cash and bluntly get out of business too soon… The business is about being ready to play when you’re called on the pitch.” The panel also tips its hat upstream: likely Accel made $2.5B on Scale, and Paige Craig’s pre-seed returned ~1,000x — the same investor famous for passing on Airbnb.

4. Founder firing doctrine, tested on the Discord rumor

The prompt is Delian’s tweet that Benchmark may have pushed the Discord founder out over a delayed IPO. Rory has fired founders — “mainly consensually,” some now serve as his references — and his rule: “If you walk in and fire a founder or any CEO and they’re surprised, you have massively failed as a board member.” Replacement is “open heart surgery and you got a one in three chance of dying.”
Rory’s data point: of the B2B IPOs he analyzed, 90% still had the founder CEO as CEO, and all but one departure was an elective step-down. Rory’s conclusion is categorical: “If the founder CEO is out, I’m out because there’s no hope in B2B.”
The seed-investor math, as Jason runs it: given a choice between a $20B+ IPO with a hired operator or a ~$10B IPO sticking with likely Jason Citron, “I’d stick with Jason in a heartbeat… I’d rather make less money.” Rory agrees, but for a different reason: it’s less money “with a higher certainty,” because one in three replacement hires is “an empty suit and a freaking disaster.”
The carve-out both concede: Uber. If a board member of a company that raised $10B genuinely believes change is a fiduciary obligation, Rory credits Gurley for doing “the hard thing, not the aesthetically pleasing soft thing.” He also notes likely Sequoia’s absence from Facebook traces partly to residual frustration with Moritz over a prior CEO change — which is why the pendulum swung so far. If you can only press one button forever, it’s never change: “it’s statistically the right outcome.”

5. Ramp at $16B and the revenue-quality mystery

Ramp raised $200M at $16B — 1% dilution, a new round roughly every three months. Harry’s puzzle: Brex sits at “101” as quoted (likely $10.1B) and Mercury at $3.5B with $500M revenue while owning the banking relationship — though Rory thinks Mercury’s bank relationships “are worth zero.”
Rory’s structural gripe, unresolved in his own portfolio too: private markets hand out the same ARR multiple regardless of revenue quality — 80% gross-margin software priced like 20-40% fintech revenue that publicly trades at 2–4x. “That remains a mystery to me.” And the model is capital-consumptive: a billion of revenue implies roughly $4B of receivables to finance.
Harry’s pushback lands: constant fundraising is a relevance and momentum strategy — brand is the product when customers can’t tell the vendors apart, and nobody wants to sign with “the ones where the energy is seeping out of the company.” Rory concedes on air: “You did call me on it correct… You got to do it. They’re crushing.” Perplexity’s two-step raise — first tranche at 15, moved to 18 on demand — is filed under the same heading: “It’s a sign of deal heat… rational CEOs take advantage of it.”

6. OpenAI: sell to everybody, and the AGI clause worth half a trillion

The $200M Pentagon contract — the largest to a single provider — reads as good news to Rory: procurement is improving, Palantir and SpaceX opened the door, and “I want to make damn sure that we have the best stuff if we ever go to war and it looks like that’s not being made by the old fuddy-duddy companies.”
Jason’s frame: at OpenAI’s market share, “you got to be friends with everybody” — the pre-2020 era of principled CEO stands is over, being political is the biggest threat to the company, and “other than rogue states, OpenAI has got to sell to everybody.” Sam himself flagged the deal as immaterial.
On the Microsoft feud, Rory’s axiom: “In situations of ambiguity, the person who has the most leverage has the best chance to win. And I would argue Microsoft has quite a lot of leverage because it’s not existential for them.” Jason’s direct rebuttal: “I think they have no leverage” — the profit-sharing ends at AGI, and “whatever AGI is, we’re going to be there pretty soon.” Sam is already signaling the deal outline: Microsoft ends up ~33% of a new entity, Sam six or seven percent.
The exchange that should be the clip: “We’re going to hit AGI when Microsoft and OpenAI litigate what AGI is, because everyone else is using that term loosey-goosey. It could be now. It could be 2045… Now half a trillion dollars of value depends on that word.” Jason adds that the smartest AI people he knows say AGI is “something likely Sam Altman made up to simplify a lot of concepts” — nobody believes in a magic point. What Microsoft actually wants isn’t cash (“Buy Scale AI Prime?”) but the IP and the longest possible tail on the relationship.

7. The IPO window: pops that bribe the public, and a Databricks tell

Chime popped 50%; every 2025 IPO is up except likely SailPoint. Rory’s mechanism: windows only reopen when good companies go out at attractive prices, so issuers have “effectively bribed the buying public to like IPOs again.” Investors are Pavlovian — the 2022 memory was losing 30-40% on average, 90% in some cases; the new narrative is “I’m up 70% on average, up 250% in the case of Chime and Circle… I got to do me some more.”
On mispricing hand-wringing: the IPO price is “almost irrelevant for the VCs” distributing over the coming years — “getting out strong may be worth the incremental dilution.” That’s the one where Bill Gurley disagrees.
Databricks and Stripe “can go whenever they want” — Rory could “imagine an entire IPO cycle where once again neither of the two names chose to go out.” But Jason reads last week’s analyst summit as the tell: Databricks is “already managing to Wall Street like they’re already public… they literally could file next week and no one should be shocked.” The real unlock is companies in the $200M–$500M range now dusting off plans — Gusto’s $9.3B tender among them.
On LP liquidity from Scale’s $14.8B: Rory tempers the excitement — against “a couple of trillion of NAV,” it’s roughly 1%, and “that’s the terrifying thing about big numbers. They’re big.” LPs don’t get ahead of distributions; those IPO distributions come back over 36 months post-lockup, which is what makes Scale’s instant, no-lockup cash dividend so unusual.

8. Gusto at $9.3B — and the agentic payroll nobody has built

Gusto’s tender came at $9.3B on $900M ARR, and Harry admits he underestimated them. The market math: five bucks a month per US worker, half of American workers in SMBs — call it 60M people — with ADP and Paychex worth ~$100B and Paychex trading at 10x revenue, $55B, at an all-time high today; ADP at 15x. “With a $55 billion ancient public company trading at 10x, you can justify any of these deals.”
Rory on having passed: “I can occasionally be very stupid… I underestimated both the founder and the openness of the market to switch” — plus how fast investors paid up, pushing it out of his price range. “Being wrong sucks.”
Jason, a Gusto customer for a decade and “a case study on their website,” makes the sharper product point: he does more work than he did with Paychex — “I just called up my rep. Hey, process the payroll… My human did it for their commission. Now I got to log in, run the workflows.” Rory’s conclusion: “somewhere out there someone is building an AI-enabled agentic payroll” where the chatbot does all the work.

9. Old guard vs new guard: your only right to win is your installed base

The setup: Glean at $7.2B moving fast, Dropbox still working to ship Dash. Rory: “I’m starting to lose confidence in the old guard” — the LLMs are open, “it’s pretty simple to rag a bunch of data and stick it in,” so he assumed the big guys would catch up at the software level. “They’re still too slow. I just don’t see it. I’m worried.”
Rory reframes the question entirely: the old-guard play is always “I’m the system of record, I can add this stuff on top” — ServiceNow can add an agent, likely Zendesk can add an agent — while the new guard’s pitch is “we have this agent and we can run on any system of record.” Your right to win is your existing customer base, and in the enterprise Dropbox doesn’t have one: “they’re not in the same quadrant. It’s not a relevant question.” Even a perfect Dash only defends Dropbox’s smallish business installed base, then starts from ground zero against Glean.
On Salesforce throttling Slack’s data access: “pathetically lame, irritating to customers” — it probably doesn’t survive in that form; the real question is whether it becomes an API connectivity fee. And the panel’s resigned aside: “the three of us can complain, but if Allstate and State Farm don’t complain, ain’t going to change. We don’t matter at 30 billion ARR.”
The Slack postscript cuts both ways: sold at 27x revenue at peak excitement, so its best days are “by definition” behind it — yet Jason notes it’s doing ~$2.5B and growing in the teens while core Salesforce grows 7-8%, so “I don’t think Salesforce got a terrible deal in the end.” His deeper diagnosis of why it sold: likely Stewart is a generational founder, but “deep down in his heart he didn’t want to do multi-product… it had an existential ticking time bomb because it could only be so big as a single-product company.”

10. Quickfire: bet the odds, reveal the book

Will a Chinese AI model hit #1 in evals this year? Jason’s first answer: “yes, but not published” — “I’m right, but no one will ever know.” At sub-20% implied odds, he calls the under “a Bay Area hubris”: “if the government’s behind you in China… I don’t want to take the under.” Rory converts on air: “the bet at these odds has to be a yes — which is different than saying they’re going to displace OpenAI.” Harry closes it: “there’s no world where we’ve got all the smart people and they don’t. It’s silly the minute you say it.”
Will the S&P finish positive? Rory refuses the forecast — “if I knew that, Harry, I wouldn’t be talking to you” — but says at 70% implied odds “I would take the no,” purely on volatility, “a very ill-informed comment.” The panel’s better question: forget opinions, where is your position? Jason: “100% equity in my personal position. Back just like 2008.” Rory: about 75% equity, “mildly terrified,” almost nothing in bonds.
Will Apple announce US iPhone assembly this year? Harry says yes — Jason notes Trump announced a $499 US-built phone the day before, and “announcing it and beginning a process, you might have to do it.” Rory: politically smart to announce “even though it’s impossible,” but “Tim Cook is such a straight shooter that he just couldn’t pull it off with a straight face.” Harry’s benediction on the whole exercise: “Now you’re a true venture investor, Rory… You say things with little thought, but great confidence.”

Verification Notes

Brex’s raw caption says “101”; the likely $10.1B interpretation remains unresolved.

Vol.66 We’re Not Just AI Content Movers—A Conversation with 36氪, 特工宇宙, and Z Potentials Trump Is Selling a Phone + The Start-Up Trying to Automate Every Job + Allison Williams Talks ‘M3GAN 2.0’