Thinking Machines Co-Founder Joins Meta for $3.5BN, Industry Venture's $665M Acquisition

Summary

• The week’s flashpoint: a Thinking Machines co-founder (likely Andrew Tulloch) walked away from the company he helped raise $2B for at $10B post to rejoin Meta for a reported $3.5B. Jason opens with “in the face of unprecedented wealth, I’m shocked to discover that most people behave badly”; Harry adds that “the loyalty conversation erodes pretty quickly when you enter the third comma.” Once someone offers you $3.5B, you’re playing a one-and-done game, not a multi-period one — so investors must price defection, not assume loyalty.
• The practical fixes are unglamorous: six-year vesting, cliffs, repurchase rights, penalties for leaving to a competitor — and a bigger fund, because diversification is the proposed hedge when a $10B-post “raw startup” is seven poachable minds. Jason’s seed-stage protection — “knowing the founder would never quit” — is under pressure at these sums; Rory calls the new reality “quite terrifying.”
• AI capex will be stopped by economics or nothing: scaling insiders treat “1% of GDP” for compute as a matter-of-fact to-do list (per likely Dwarkesh Patel’s oral history of scaling, which Rory read over the weekend), while Jason testifies demand is insatiable — 8 vibe-coded apps in 100 days, 12 AI agents running at SaaStr, “I could use 100x the tokens” — and only 0.1% of Salesforce customers really use AI yet. Harry’s brake: at some point “capitalism is going to say… you can’t have your $1 trillion dream.”
• Seed’s investable sweet spot “may have declined to like half an hour”: companies are “born almost instantly,” Lovable passed $170M ARR at its first anniversary, so you either invest into acute uncertainty or at $2B pre. The emerging edge: calibrate to Aaron Levie’s “diffusion rates” by industry, or hide in legal/regulatory complexity that better code alone can’t disrupt.
• Polymarket ($2B at $9B) and likely Kalshi ($5B from a16z and Accel that week) are, per Roger, “the purest regulatory arbitrage play of all time” — 90% of the business is sports betting “we’re not calling it that,” racing to get too big to regulate, with striking Trump-family proximity. And it’s proof kingmaking has limits: bettors “don’t give a damn” who funded the book, provided it pays out.
• Founders Fund’s shift from 31 growth investments to a planned 10 surprised Rory only in that “they weren’t there already” — if you can call the shots, concentrate. Roger’s counter-playbook for early stage: 20–25 names as “the farm team,” then 75% of capital into the 3–5 that prove out — justified by Rory’s data that hitting the first two underwritten revenue years lifts the odds of a 5x+ from 30% to mid-70s, in a world where the IPO bar has moved from $200M to $400M ARR.
• Goldman’s purchase of Industry Ventures — $665M plus up to ~$300M earnout against $7B AUM, roughly 10% of AUM and ~10x revenue — is a fair market price and a structural lesson: only productized GP businesses (secondaries, fund-of-funds, platforms) can be sold 100%. “The only asset in Roger’s new fund is Roger’s IQ as a stock picker.”
• Masa’s $5B margin loan against ARM to fund OpenAI is “a relatively low octane Masa move”: SoftBank still owns 90% of ARM (~$90B position), and Harry reckons he could take $25B against it “easily.” Jason’s caveat from 2002: individual Nasdaq stocks fell 90% — “it is possible the loan will get called. It’s just unlikely.”

Deep dive

1. Goldman pays market price for Industry Ventures — and only productized GPs can sell

• The deal: Goldman Sachs acquires Hans Swildens’ Industry Ventures for a $665M starting price plus ~$300M in performance earnouts to 2030, against $7B under management, after a 25-year grind that began “just after the crash in 2000.” Jason’s opening congratulation is for restraint: “congratulations on letting the ego walk it back” and not demanding a headline over $1B in “2025 founder language.”
• Rory’s valuation frame: the deal traded at roughly 10% of AUM — Carlyle and KKR, which own all their economics, trade near 20% of AUM; public equity managers as low as 1–2%; secondaries economics sit in between, in line with StepStone and Hamilton Lane. Roger, who sold asset managers in financial-institutions M&A early in his career, gets to the same place: ~10x revenue, which at 50% margins is ~20x earnings — “a straight-on-market deal.”
• Goldman’s rationale is distribution: with S&P exposure available under 10 basis points, public-markets asset management “is eroding away super fast,” so Goldman jams high-fee privates through its channel — including its Apex platform for ultra-high-net-worth clients. Jason’s endorsement is from the buyer’s seat: an ~18% historical IRR productized “every day in and out” beats “the crazy calls I get from Morgan Stanley… the ideas are so dumb.”
• The structural lesson from the discussion: productized GP businesses can plausibly be sold 100%. You can’t sell 100% of Benchmark “because then you don’t have Benchmark.” Secondaries and fund-of-funds are productizable (Greenspring sold to StepStone the same way); a small venture firm is three people. “The only asset in Roger’s new fund is Roger’s IQ as a stock picker… without Roger, does nothing.” The continuum runs through YC, General Catalyst, a16z’s “journey to bigness and maybe an IPO” — and, Rory needles, Harry’s “media company with a venture fund attached could be monetizable in a way that Roger’s fund or my fund will never be.”

2. The Thinking Machines defection: “you’re no longer playing a multi-period game”

• The facts as discussed: a Thinking Machines co-founder (likely Andrew Tulloch) leaves the company that raised $2B at $10B post — after less than a year at Thinking Machines — to rejoin Meta for a reported $3.5B, after a career of roughly 14 years at Meta, under a year at OpenAI, under a year at Thinking Machines. Jason’s disbelief: “people are checking out of $10 billion seed companies now… when I was a founder, there was no way I would leave no matter how tough it was.”
• Jason’s moral case, kept sharp: “You leave the people that brought you to the dance because you see a prettier girl over here… if my kid did that, I would not be happy with my kid. Something’s broken in the way that we’re evolving as humans if everything ultimately reduces to what’s in it for me.”
• Rory’s game-theory rebuttal is the episode’s spine: “Once you’re not playing a multi-period game — and when someone’s offering you $3.5 billion, you’re no longer playing a multi-period game. You’re playing a one-and-done — you’re going to get bad human behavior.” Harry’s summary, which Rory accepts: “Big boys rules.”
• The pure financial call isn’t close. Rory to Roger: $2B of unlisted Thinking Machines stock or $3.5B of liquid Facebook stock over five years? “How obvious.” Though Roger flags the embedded option — Thinking Machines “could be a $500 billion company” — and the context: after ~14 years at Facebook, “the dude has got a hundred already.”

3. The protections that actually exist — and the diligence that didn’t happen

• Rory’s checklist for founders and investors alike: extended six-year vesting, cliff vesting, repurchase rights, and “if you leave for a competitor, something really bad happens.” His framing is founder-first, not VC-defensive: if seven co-conspirators quit safe jobs together, “you got to run the game theory of how will I feel if one of my seven bails on me.”
• Harry’s uncomfortable question about the round itself: in a deal done “on a Saturday” where investors weren’t even told what the product was, “did any of them even meet him in person?” The blunt rule that lands in response (attribution muddy in the crosstalk): “As a VC, you should be fired if you write a $100 million plus check and you don’t meet the co-founders. Period.”
• Jason’s confession of what this breaks: “My liquidation preference has always been — and I put it in quotes — knowing the founder would never quit. That’s my protection as a seed investor.” Rory’s conclusion is darker: with proven evidence that the core asset — “those seven minds” — can be poached for billions, “it makes it real how risky those investments are… it’s quite terrifying really.” The half-joking answer from the table: a bigger fund. “Diversification. You don’t want to be too concentrated with these deals.”

4. Masa levers ARM for OpenAI — “a relatively low octane Masa move”

• SoftBank reportedly secures a $5B margin loan against ARM shares to invest in OpenAI. Rory shrugs: “Masa rules. Nothing new to see here… when he has a feeling, he goes all in. All chips, max risk.” Rory checked the collateral: SoftBank still owns 90% of ARM, trading around $90B — “$80 billion of equity there. He can lever up some more” — and Harry reckons he could take $25B against the position “easily.”
• Jason, initially uneasy, comes around to it as smart leverage — no capital gains triggered, and “that loan’s not going to get called under any scenario, probably.” Jason’s caveat, exactly as hedged: in 2002 the Nasdaq saw individual stocks fall 90% from the peak — “so it is possible the loan will get called. It’s just unlikely.”

5. The capex question: economics is the only brake on AI

• Harry’s bull framing — we’re building more data centers than office buildings — gets dismantled by Rory as “clever but trivial”: of course you don’t build offices “because there’s no one in them.” The real anchor is likely Dwarkesh Patel’s new Stripe Press oral history of scaling, which struck Rory for its matter-of-fact tone: the scaling law “has been proven to hold for six, seven years at a high degree of accuracy,” and the field’s smartest simply accept that “of course we’ll need 1% of GDP to invest in computers… and somewhere along the line we’ll get AGI.”
• Harry’s model of what could stop it: not technology, not demand — “it will be purely and simply, at the margin, the marginal capital provider says the economic return isn’t holding.” The scaling law may be log-linear, but “every economic phenomenon tends to be diminishing marginal utility — at some point capitalism is going to say… you can’t have your $1 trillion dream because we just can’t afford it.”
• Jason’s demand testimony from the front line: he’s vibe-coded 8 apps in 100 days and runs 12 AI agents at SaaStr that replaced almost all of the sales team and the entire content team — and still, “I could use 100x the tokens… I got to wait 20 minutes to build one feature.” He cites Replit’s Amjad: “you’ve got it backwards — everyone will consume every available token.” With only 0.1% of Salesforce customers really using AI (per Marc, Dreamforce week), “it’s 100 times 100 times something.”
• Roger probes the one relief valve — could step-change processing efficiency shrink the infrastructure need? Jason: no. Engineers whose code is 50% Cursor-built don’t take the rest of the day off; they ship another feature. “The better that gets, the more tokens you’ll consume. I don’t think there’s this great efficiency coming.”

6. Seed’s sweet spot has compressed “to like half an hour”

• Jason on why seed is suddenly brutal: “that company probably didn’t exist 30 days ago” — when this group started, startups were “never good 30 days in,” and now companies are “born almost instantly.” Lovable crossed $170M ARR at its first anniversary; six months ago it was raising at a couple of billion. Rory’s arithmetic: the window “where it’s working but it’s not obvious and you can invest… may have declined to like half an hour. You’re left with the choice of do you invest into acute uncertainty or do you invest into two billion pre.”
• Roger’s answer — and his thesis as he returns to the field with a new fund — is a third door: businesses insulated by legal and regulatory complexity (financial infrastructure, media rights, copyright, IP), where winning is “more nuanced than am I able to develop the next base model.” Acute uncertainty “does not trouble me in the least when it’s expressing a deeply held, well-researched thesis.”
• Rory adds the framework, borrowed from Aaron Levie: diffusion rates differ radically by market — a Lovable-type market is “done and dusted in 6 months,” while a regulated vertical may take two years to the first lighthouse customer “but then it’s boom” and the next five land in six months. “Setting your investing thesis accordingly, varying it by the diffusion rate, will be one of the key skills here.”

7. Prediction markets: the purest regulatory arbitrage, not kingmaking

• Polymarket raises $2B at $9B; that week likely Kalshi raises at $5B from a16z and Accel. Roger’s read: “This is the purest regulatory arbitrage play of all time” — the cumulative market cap of regulated sports betting has dropped in direct response to two venues not subject to the same rules, and their strategy is to “run as quickly as they can to get so big and so powerful that they will not face the parallel regulatory scrutiny the legacy companies have suffered through since PASPA.”
• Jason strips the euphemism: political markets are 10% of revenue; “90% of their business is sports betting but we’re not calling it that… everyone in America wants to bet on the NFL and they’re cleaning up.” And he flags it as the counterexample to last week’s kingmaking thesis: “I don’t think anyone betting on Poly or Kalshi gives a damn how much money they have, provided they can pay their bet” — unlike enterprise software, where brand VCs create herding (his example: Brett Taylor’s Sierra). “There’s no anointing here. It’s just you’ve got money to play the game.”
• Harry, “outside the borders,” names the coincidences: Eric Trump on one board, another Trump invested in the other, Howard Lutnick’s son running “the fastest growing investment bank.” Harry’s zoom-out is a policy stance: regulate as little as possible, because regulation creates incentives to capture regulators — and “the efficiency of the regulatory arbitrage has definitely gone up… no trivial $500,000-a-year jobs after office. Just give me 5% of the company, it’s so much quicker.”
• Roger’s market-plumbing addendum: with structural state budget deficits and differential gaming tax rates, every time Illinois jacks up rates, regulated books cut investment, handle falls, and customers leak to offshore unregulated operators — Bovada, Crypto.com, Stake — “and these companies make billions and billions of dollars.”

8. Founders Fund concentrates; indexers index; diversification is “buying time”

• Peter Thiel has shifted “from caution to concentrated AI bets,” heavily via OpenAI — the opposite of DST, Lightspeed, and GC’s index-the-wave approach (Mistral, Anthropic, OpenAI). Harry’s take: being too diversified in AI today “is buying time. It’s not knowing, not having the conviction” — defensible as a plan B, but Thiel is “40% of the capital in Founders Fund plus his own capital. Making little teeny bets doesn’t get you there, does it?”
• Rory’s math is unsentimental: diversification narrows variance both ways — “it’s the central limit theorem, it’s not a great insight here, people” — so the more certain you can call the shots, the more focused you should be, and Founders Fund has both the evidence and the confidence. His genuine surprise from the reported numbers: Growth Fund I held 31 investments, Fund II mid-to-high teens, Fund III aiming for 10. “The only surprising thing was they weren’t there already.”
• The stage caveat matters: that’s a growth-fund strategy for “effectively public companies that are just private.” At the “will this thing even work” stage, you need meaningful diversification first and concentration later — “it doesn’t lend itself to a one-dimensional answer.”

9. Can you pick winners early? Harry’s Clubhouse problem vs Roger’s farm team

• Roger’s returning playbook: 20–25 portfolio names “to create the farm team,” then deep concentration on second and third checks — historically 3–5 companies end up as 75% of capital deployed. And the entry prices still exist: he just wrote $1.5M at a $10M post for 15% of an analytics company; Harry adds that it has multiple six-figure ACVs. Harry, wide-eyed: “Five on 50 and I’ll do it.”
• Harry’s pushback from his own portfolio reviews: six years into fund one, the best performers “were not obvious early and the early outperformers did not signify enterprise value — Clubhouse, Hopin, BeReal. If you think you can pick your winners early, I think you are wrong. Am I wrong?” Roger’s answer is that fund-returners take radically different paths: Trade Desk had “multiple near-death experiences” and no product in market for eighteen months before it hit; Wise was the smoothest up-and-right he’s seen ($750K first check at a $5.5M post, compounding to 13% at IPO, alongside 17% of Trade Desk); Datadog he could only fight his way to 2.2% at IPO — still enormously valuable on a $40B outcome.
• Rory quantifies the middle ground: seed information isn’t zero, and post-revenue it gets strong — his firm’s data shows that if a company hits the first two underwritten years of revenue, the probability of a 5x+ jumps from 30% to mid-70s. Meanwhile the finish line has receded: “we’ve moved from a world where an exit is 200 million in ARR to a world where an exit is 400 million in ARR at an IPO” — two or three more years of holding — so his firm is stretching from under 20 deals per fund toward 25.
• Jason’s self-diagnosis as the concentrated-from-first-check outlier — roughly 8% of fund per deal — is that the model forces painful triage: “I have to turn away something where I don’t have certainty even if it’s cool… you’re not going to do the Clubhouses because it’s wacky, but it might be great. You got to just find the Wises and go all in.”

10. “Everything in life you can price is an option”

• Roger’s follow-on discipline: “We look at every check independent of the check prior. Period.” The proof case is Trade Desk — after ~$2M across four early checks, an air gap, then a $3M check at a $280M post out of a $50M fund into the $20M round; that check alone returned $40M. Rory’s tribute: “no one I know understands options value better than Roger” — and Roger owns it: “I walk through life. Everything looks like the Greeks.”
• Jason’s worry — what if every follow-on round is priced at 300 or 500 when you entered at 8? Roger: then don’t write it. “If the worst thing that happens is my initial check gets marked up and I don’t need to chase the money, I’m money good… provided every check you write is money good, in the end you’ll die rich.” Roger adds Peter Thiel’s rule: when a big, reputable outside investor leads a markup in your deal, “do everything you can in it” — there’s real signal in that round, risk- and information-adjusted.
• Jason’s counter-scar is reserves: with the standard put-10%-into-winners rule, three or four breakouts can exhaust $30–40M of a $100M fund fast — “I have come to regret some of my third checks. I just wish I had more flexibility in the midlife.” Roger’s structural fixes: recycle to 110–120% invested, and deliberately build parallel LP bases across funds to enable cross-fund investing without conflict — turning a $100M fund into effectively $260M-plus of firepower for a proven winner.
• And a stewardship line worth keeping, from Roger: if you own double digits, “ethically you got to be there till the end. Otherwise, you’re dead weight on the cap table — you don’t get to check out after 24 months and show up once a year as an observer.”