Summary

SpaceX’s $60BN all-stock Cursor takeover closed looking cheap, not crazy: at ~$6BN year-end revenue it’s roughly 10x forward, and Jason’s verdict is “Elon was a shrewd buyer.” Rory’s structural point: the gross-margin critique of Cursor was true but was outweighed — “your gross margin problem is my revenue opportunity for my Colossus cluster” — because in a risk-on market with coding as AI’s biggest vertical, “pessimists sound smart, optimists die rich.”
The deal mechanics are the tell for who can do AI M&A: Cursor had a $2BN-at-$50BN round teed up from Andreessen, and Elon simply bid $60BN with a $10BN breakup fee and struck the deal in a week. Jason argues only Elon or Zuck can strike at that speed, and founder psychology matters — “I would much rather initially work for Elon than for Zuck” — while Rory notes SpaceX bought it “like a regular old corporation,” with quick antitrust clearance Meta might not have gotten.
The company for whom owning coding is existential isn’t Amazon or Meta — it’s Microsoft, with GitHub “now a trailing-edge product” and the Ballmer “developers, developers, developers” connection lost. Jason punctures the land-grab myth: losing number one doesn’t trigger a panic race for number two; acquirers who “thought I had more time” quietly move Cognition up their existing priority list.
Stripe’s $7BN OpenRouter buy (5x CapitalG’s Series B mark in four months, 12x for Menlo and Andreessen) is a bet on a “wonderful niche” that may not scale like payments. Jason’s product critique: serious B2B workflows down-spec to one or two models because “you can’t have model drift” — Rippling just standardized on two — and his five-year call is that the product itself won’t exist but there’s a “>51% chance it builds into a 20 or 30% revenue stream for Stripe,” like the Scale acquisition.
Anthropic’s first profit on $11.5BN of Q2 revenue was mechanically inevitable — “you can’t add expenses below the line fast enough to stop yourself making money” when revenue 12x’s against ~40% gross margins. For the IPO, Rory says the off-balance-sheet commits and jaw-dropping SBC (a $1M 2023 package that paid $51M) get a hyper-growth free pass: “the only thing that matters will be the growth rate and the ‘27 and ‘28 projected revenue… once revenue stop go up, all bets off.”
The path to Anthropic’s $200BN 2028 plan is stretched and $600BN may not exist: Rory calls a billion knowledge workers “absolute bollocks,” counting ~83M US knowledge workers and ~5M software-related workers grossing ~$600BN in pay. Both converge on ~$100K of tokens per engineer with dev teams cut 30-40%, which “terrifyingly” yields a ~$200BN US market — ever, not next year — versus Anthropic’s own estimate, with maybe $350BN worldwide before leaving software.
Silver Lake’s $43BN Workday take-private sets the high-water mark for SaaS LBOs: around 5x trailing revenue for a 13% grower versus Bending Spoons’ 2.7x for Airtable — “you pays your money, you takes your choice.” Jason’s key nuance: system of record is a moat, not a growth ticket (“just because your customers are prisoners does not mean they will spend one more dollar with you”), and Workday’s closed ecosystem is better protected from agentic abstraction than headless-friendly Salesforce.
The application layer is compounding moats fast: Lovable’s $13.3BN and Higgsfield’s $5.5BN on similar ~$600-700M ARR now look reasonable with Cursor at 10x forward as the comp, and Etched rose from $17BN to $21BN in roughly a month. Jason’s operating bar for everyone: “if you’re not into your 2027 roadmap, deep into it, by August of 2026 in the agentic world, your team is not good enough to survive today” — and on fundraising, “you only need one great month to raise today.”

Deep dive

1. Cursor’s $60BN exit was three companies in 70 weeks — and cheap by the close

Speaker 1’s opening reflection: over the show’s ~70 weeks, “it almost seemed like Cursor was dead” when Claude Code launched — “none of our portfolio companies were using it” — then the early multi-model pivot “changes everything and rockets to a 60 billion outcome.” From email-client Hacker News curiosity to this: “there’s probably been three different Cursors since we’ve started,” and the journey “was not actually 100% linear progress to 60 billion.”
His second point: by close it “ended up not even being that expensive” — 10x forward revenues on ~$6BN year-end, for the company VCs once mocked for “selling a dollar’s worth of tokens for 80 cents.”
Rory’s generalization from the gross-margin saga: the negatives cited along the way were all correct, but the market trajectory outweighed them. With coding underpinning “70, 80% of Anthropic’s trillion dollar market cap,” the number-two player growing hyper-fast gets looked through — “it’s the old cliché… pessimists sound smart, optimists die rich.” In a different capital market, the same facts produce a very different Cursor.

2. Why Elon and not Zuck — currency, clearance, and stomach

Harry’s puzzle: Meta fits the same profile — lots of compute, no compute business on top — so why didn’t Zuck buy it? Rory’s answers: SpaceX “just bought it like a regular old corporation” with quick antitrust clearance a DOJ-shadowed Meta couldn’t promise; and SpaceX stock at ~40x revenues ($8BN last quarter) makes a 10x-forward asset “net accretive day one.” Also, bluntly: “no one else had the stomach for the bet in the way Elon does.”
Jason reconstructs the deal: Cursor was about to close $2BN at $50BN from “Andreessen and Friends,” and Elon asked what it takes — $60BN, a $10BN breakup fee if it falls apart, run the company your way. “It was three points, and they shook hands.” Zuck has done this (Instagram, WhatsApp “in an hour on the back of a napkin”), but here he’d have needed $70-80BN in a week — “only a handful of people can do this.”
Founder psychology as M&A edge, per Jason: “I would much rather initially work for Elon than for Zuck… this guy is fucking rockets, electric — he does everything, and Zuck’s firing everybody and going crazy ‘cause he doesn’t have an LLM.” He tells founders to ignore brand in M&A, “but it is incredibly emotionally important to founders to land in something they want to land in.”
Rory’s coda, citing Noah Smith — “only a fool denies that Elon Musk is wildly effective”: in just over a year SpaceX went from rockets and satellite-to-cell connectivity to, per its S-1’s future prospects, “89% an AI story,” building the cluster and then buying the product to fill it. “That’s just wildly effective management.”

3. The knock-on: Microsoft’s existential problem, and the myth of the land grab

Rory’s triage on who has to own coding: Amazon effectively already gets Claude Code’s inference revenue through AWS — “important, but not imperative” — and Meta’s $100BN+ ads machine means its AI push is really “I’m just fucking really interested in it.” The company for whom this market is “100% existential over the medium term” is Microsoft: Ballmer screamed “developers, developers, developers,” and “GitHub is now a trailing-edge product.”
Jason’s pushback on Harry’s assumption that rivals now rush to buy Cognition: big-company M&A doesn’t work that way. Losing number one “pushes it up my existing priority list” — “maybe Satya’s like, I thought I had more time with Cursor… I could do it at 100 billion next year” — so number two gets bought from disappointment, not panic. “That might be a VC partial myth.”

4. Stripe buys OpenRouter for $7BN — value to acquirer, not DCF

The facts as Harry lays them out: not the reported $10BN but $7BN, with some Stripe stock going to investors; CEO Alex Atallah (previously founded OpenSea); a 5x on the $1.3BN Series B from just four months ago for CapitalG, 12x for Menlo and Andreessen.
Rory’s mechanism: Stripe already takes a small cut of money flow to manage payment complexity; here it takes a cut of AI spend to manage model-picking complexity across “tens and maybe hundreds” of models. In a meta-market growing 10x year-on-year, “the value to the acquirer dictates a very healthy price” — Stripe will monetize OpenRouter faster than OpenRouter could standalone. “This is what happens in a crazy market.”
Jason on the weirdness of hypergrowth M&A: revenue is “the biggest input to price” yet economically irrelevant — OpenRouter does maybe $70-80M and “Stripe does not care for that money.” The grouchy version: if you’re this good at M&A, “maybe they should be better at building these themselves.”

5. The niche debate: model drift versus the routing dream

Jason’s product-level critique, which Rory concedes his own firm “agonizes” over: OpenRouter is superb for developer tools and chatbots where outputs needn’t be perfect, but high-reasoning B2B workflows down-spec to one or two models — “you can’t be routing from Kimi to Qwen to GPT-4.6 to Fable… even just going from one Opus model to another, you see drift.” His evidence: Rippling just published its stack — Opus 4.8 (“an N-minus-one model, but well-trained with their harness”) plus what Jason tentatively called GPT-5.5 Medium — and said “the rest isn’t worth it for Rippling today.”
Rory’s cloud analogy for the risk: everyone said “I wanna be multi-cloud” and it was really hard. If JPMorgan wants Anthropic plus one Plan B and won’t qualify ten models, “then you have niched and you don’t get that revenue” — though as long as frontier labs try to extract $100BN a year from enterprises, everyone wants something keeping them honest.
Harry demands a five-year verdict. Jason: “I think it’ll be like the Scale acquisition… I don’t even think this product will exist in five years,” but there’s a “>51% chance it builds into a 20 or 30% revenue stream for Stripe” — the product buried “11 layers down” inside a token-management platform. Rory adds the portfolio view: doing this TAM-expansion bet while simultaneously circling a $40-50BN PayPal consolidation play, issuing paper while private, “is pretty impressive.”

6. Anthropic’s first profit was unsurprising given the growth

Rory refuses to be surprised by profitability on $11.5BN Q2 revenue: gross margins went from negative to ~30% to roughly 40% on track, and when you go from $4.5BN in a year to ~$10BN in a quarter, “you can’t add expenses below the line fast enough to stop yourself making money.” You’re not 14x-ing headcount or below-the-line training costs in six months. His hedge: once they’re buying Elon’s expensive compute — which had a big price increase two months in — he doubts they’ll forecast continuing profitability as the IPO base case.
Jason’s IPO worry: off-balance-sheet compute commits and “probably stock-based compensation like we’ve never seen in the history of mankind” — if fully accounted, “they will be jaw-dropping.” He thinks the market must ignore it, and “it’s important for Anthropic it get ignored.”
Rory’s defense of the free pass: the off-balance-sheet commits are self-hedging — if revenue 10x’s you need and can afford the compute; if it slows, “it all gets hard.” And the excess in the SBC example is luck, not run-rate: the person “hired with a million-dollar package in ‘23 ended up making $51 million four years later,” but you don’t have to pay the next hire $51M. The unfairness is real and symmetrical — in mature Workday/Salesforce, SBC “is real money” — but here: “you get a free pass on margin, off-balance sheet, SBC, provided revenue go up. Once revenue stop go up, all bets off.”

7. The $600BN math: “a billion knowledge workers — absolute bollocks”

Rory’s hard-nosed TAM teardown of Anthropic’s $200BN-by-2028, $600BN-after plan: the US is ~50% of world software spend because it’s half the world’s high-end knowledge workers; “that’s why we have crappy internet when we go to Europe.” There are 83M US knowledge workers (including nurses and teachers you won’t replace), and the “uber-sweet spot” is ~1.8M coders, ~5M software-related workers grossing ~$600BN in pay. $200BN “means you’re replacing a third of them. That’s a lot.” The single most important ratio: salary dollars to AI dollars in steady state.
Jason’s live data: in the last 60 days “every single scale-up is capping their AI budget for real” at $6-8M a year, and he thinks it lands at “$100,000 of tokens” per best engineer, in exchange for cutting dev teams 30-40%. Rory agrees, extends it to sysadmins and QA — and that math produces “terrifyingly about a 200 billion plus-or-minus market in the US.” Jason says “Next year”; Rory says “No, ever” and calls Anthropic’s estimate too optimistic. It probably means struggling past $350BN worldwide before leaving software, where nothing is “such a sweet spot.”
The corroborating dispersion: Ramp’s top 1% of a tech-forward sample spends ~$7K — only $84K a year at the cited monthly run rate, so even the pointy edge is ~50 cents of AI per salary dollar. And Jason’s board-room evidence that the spend is no longer performative: “I had two board meetings in the last week where they finished the roadmap for the year. They’re into 2027.” His gauntlet: “If you’re not into your 2027 roadmap, deep into it, by August of 2026 in the agentic world, your team is not good enough to survive today.”

8. IPO sequencing: Anthropic first, OpenAI capitulates

Rory wants to be “first out rather than second out,” especially with a bouncing-around-profitability story: going out in the fall as the enterprise winner beats going out next year into possibly slowing growth. Jason mostly agrees but thinks OpenAI has “just capitulated” — after executive turmoil and a first half “slower than its previously junior competitor,” they’ve accepted trading against a live comp: “the world will not end if we trade at 1.3 trillion.”
The one place Rory pushes hard: “there are no two companies on the planet that need more capital than these guys,” and when you need hundreds of billions, “being second sucks” — would you rather trade at $1.5T needing $100BN, or at $1T needing $300BN? Jason’s rebuttal: at an implied $2T, OpenAI can sell at $1.8T — “it’s okay to sell at a small discount” — and Sam’s team has decided “the cards are the way they are.”
On the executive churn Harry flags — Denise Dresser out as CRO, Wiz’s Dali Rajic in (“the best of the best,” per Chad Peets) — Jason reads it as Greg Brockman having “had enough of the Salesforce crap,” while defending the broader pattern: at $45-50BN run rates “you don’t wanna hire kids,” and a Wiz operator from a hypercompetitive technical market beats someone “asking how many seats of Slack you want.”

9. Silver Lake’s $43BN Workday bid is precision engineering, not a SaaS revival

Rory rejects Harry’s “SaaS isn’t dead” framing: this is a savvy financial buyer betting sticky system-of-record revenue at 13% growth, bought at roughly 5x trailing revenue / 16x trailing EBITDA, levered, debt paid down from ~$3BN annual cash on ~$10BN revenue, for a ~20% IRR and almost 2x over four-five years. The sensitivity is the tell: overpay by 20-30% and IRR “dips from 20 to low teens” — “it’s almost the exact opposite of venture,” where “if you’re in the right thing, it almost doesn’t matter what you paid. See Cursor for details.”
Jason’s distinction “everyone on X gets wrong”: system of record is a moat, not a growth ticket. Workday’s five years are more predictable than 95% of companies, including poor Monday — but stickiness “sure as hell doesn’t mean I want to spend more money with that vendor.” The unpriced upside: co-founder Aneel came back as CEO, and if he builds “the agentic version of Workday,” the 20% IRR becomes a game-changer — though Rory insists “it wouldn’t be in the damn base case.”
Jason’s closed-versus-open thesis: Workday is “closed AF,” and that’s a feature under agentic pressure. He runs Salesforce “entirely headless” via his own agent — which makes Salesforce more powerful but also connectable to competitors and abstractable away. “The value will extract to the agent even if the system of record is retained. But Workday’s so closed they’ve got a leg up.” Rory’s proof point: Gong, Outreach, and SalesLoft are effectively built on top of Salesforce; “you can’t name the equivalent with any ease in Workday.”
Rory’s market-wide takeaway: this is the high-water mark of the bid-ask — the best-in-class system of record gets roughly 5x; the Bending Spoons/Airtable tier gets 2.7x. “Contrast that with the AI-routing example… 70 times trailing revenues. Which game would you prefer to play?” And the prize: turn a ~$25-30BN equity check into ~$60BN — “someone’s about to make $6 billion if they can pay down this debt and just knuckle down.”

10. Jason for PE operating partner — and the Airtable warning

Rory reverses himself mid-conversation on whether Jason could run a buyout firm: what PE deals miss is “mission clarity around growth” — without it “you’re in a desperate race against the debt,” especially for firms that paid 10-12x revenue years ago for worse assets than Workday. “The only thing that matters — you can’t just stick it to your customers.”
Jason’s cautionary flip side of captive customers: hostage CIOs try to cut spend, and Airtable users are already posting “I’m lifting off Airtable now” ahead of reported potential Bending Spoons price hikes. Losing 20% of customers while tripling prices may still pencil — “but it’s gonna happen a lot faster than Workday.”

11. Lovable at $13.3BN, Higgsfield at $5.5BN: moats are accreting at the app layer

Harry flags the divergence: similar ~$600-700M ARR, wildly different prices (Lovable via Menlo, Higgsfield via DST — and Higgsfield’s deal was struck at $500M ARR, hitting $700M by announcement: “that’s what happens if you don’t announce a deal the hour the term sheet is inked”). Jason’s reframe: both were “really terrible products when we started the show” and are now “truly generationally great” — and with Cursor at ~10x forward as a comp, Lovable’s multiple is “not radically off, is it?”
The meta-learning both land on: moats start light and accrete through execution. Jason: Higgsfield went from four-second clips to full-length motion pictures; Lovable/Replit “can almost build production-grade, highly secure apps,” with automatic deep penetration testing shipped, and both are talent magnets (“the smartest mathematicians in Kazakhstan”). The layers “are not impenetrable, but they start to get thick and crusty.” Rory’s historical rhyme: Netscape and MS-DOS started “mind-blowingly simple” and accreted value — “you just have to be faster and better… you just gotta know which game you’re playing.”
The week’s mania data point: Etched raised $700M at $21BN from Jane Street, Kleiner, Sequoia, and Andreessen — within roughly a month of the $17BN mark. Jason: “You only need one great month to raise today. You used to need three to four.” His only asterisk: “Jane Street wants to be a customer too… you never quite know how it’s all tied together.”

12. The DOJ’s Clayton Act poke at Andreessen is a non-story with a real lesson

Rory did the homework on why the DOJ is probing Andreessen’s overlapping boards: Clayton Act Section 8 (early 1900s) bars individuals from sitting on two competing boards; the initiative began under the prior administration (Thoma Bravo precedent — they just took board members off), and the Fivetran/dbt merger review apparently tripped the alarm on Andreessen sitting on both Databricks and Fivetran. Likely ending: “they’ll go, okay, we’ll take a board member off.” Jason: “probably a non-story in the end… there is a remedy here. You resign.”
The absurdity both savor: the law imagines Databricks and Fivetran colluding “to raise the price of AI tools” like U.S. Steel in 1909, when the logical test would be “Founder A, are you cool with this for Founder B?” — “but it turns out that’s not the way we write laws.” Rory’s serious closing point: regulatory regimes “go on forever,” so “if we pass some law about AI regulation now, you’ve gotta be really careful the unintended consequences… decades later.”

From Restoring Sight to Reimagining the Brain, with Max Hodak Vol. 230 Industry Watch 43 | How Is AI for Science Being Deployed in Pharmaceutical Chemical Synthesis?