Elon Musk vs Sam Altman | The Implosion of Thinking Machines | Can VC Survive Public Pricing?

Summary

• Elon wins no matter what is the panel’s verdict on Musk v. Altman: his claim isn’t the $30M donation back but that OpenAI was “a for-profit all along,” so he’s owed what a $30M seed would own today — $70–130bn in extra OpenAI shares, diluting everyone else, and a judge has already refused summary dismissal. Rory thinks the core assertion fails (“if they wanted to cheat, they could have just done what Anthropic did” — a public benefit corporation), but the suit is an asymmetric win-win for Elon à la “Peter’s spend on Gawker.” Investors should price roughly a 10% chance of 15–20% extra dilution — an asterisk, not an existential event — and “the guy laughing with popcorn is Dario.”
• Thinking Machines’ implosion gets reframed in one line: “It happens with seed rounds… this is just seed rounds with extra commas.” The rational play isn’t a long bloody rebuild but redemption — spend $200M of the $2bn, bribe with $200M more, return $1.6bn, and recycle: $0.8bn put into the last Anthropic round six months ago would already be $1.6bn, up 60% despite losing 20 cents on the dollar. “Reallocate to success and away from failure.”
• Public multiples aren’t killing venture — they’re sifting it: low-growth companies get discarded while Palantir trades at 70x forward sales as a 45% grower. Figma at $12bn, 10x forward, 30%+ growth is “an awesomely good company” — the pain is anchoring off unicorn entry prices. Rory’s rule: “Always be in the hot stuff and you’ll be fine. If you’re in the trailing-edge stuff, you’re toast.”
• Harry’s confession — “venture and tech is a bit of a scam”: converting 20–200x revenue multiples into cash “when they haven’t earned it in free cash flow,” because “if we have to go to an EPS world, we’re dead.” Rory’s rebuttal: it’s rational basket-buying of might-be-Microsofts — “four out of five of them turn out not to be Microsoft,” and system-wide it still works.
• For mid-stage SaaS at $50–75M revenue, the investor answer is brutal — the probability of exploding from there “is rounding error zero” — but the operator answer is actionable: attach to AI tailwinds now (RevenueCat tripled developers in 3–4 months), because “there is no excuse for you to not have an agent as good as the new kids.” The grind path — 50/40/30 growth to $200M, sell at 5x, keep 20% = $200M — is fine for founders, “but it’s a grind. And grind is not in our MO.”
• OpenAI ads are inevitable and possibly enormous: free-tier conversion runs well under 5%, probably 5% or less, “there’s simply no other way to monetize,” and LLMs are now the prime discovery real estate — Google’s ~$240bn ads cash cow “isn’t the best product on the market anymore.” Jason’s math: 0.22 monetized ads per prompt at a $50 CPM is $25bn; “this clip will be used in three years’ time when they are at a hundred billion dollars in revenue.”
• The tradeable adjacency is AEO, not ad-buying tools: OpenAI will run the paid auction itself (“all the value on paid is captured by OpenAI”), while Adobe buying Semrush at ~4x revenue “only makes sense” as an answer-engine-optimization play — the trade the panel missed.
• On new rounds: ClickHouse at $15bn is “underwriting growth persistence” in a real category (OLAP), worth maybe $30–40bn if Snowflake/Databricks are 100–200 — “welcome to late stage investing.” Replit at $9bn is defensible because the product is “50 times better” than at the $2–2.5bn round with ~$250M ARR heading toward 700–900. And Sequoia holding both OpenAI and Anthropic at $350bn pre is fine — “you are Fidelity large-cap growth, just in the private markets” — while Jason calls competitive Series A/Bs at 100x ARR “the dummies game.”

Deep dive

1. Public markets are sifting, not sinking — Figma is an anchoring problem

• Harry’s opener: Figma back near pre-IPO levels, Datadog down 20%, Monday “killed” — do public multiples break the venture model? Rory’s answer: no, “almost the exact opposite.” Markets are sorting, not collapsing — ex-growth gets “thrown out… discarded at pretty low valuations” while Palantir trades at 70x forward sales as a 45% grower. “Venture is nothing if not a trend business… Always be in the hot stuff and you’ll be fine. If you’re in the trailing-edge stuff, you’re toast.”
• Harry’s gut check, delivered as a board-room joke: “If Figma isn’t good enough, what hope is there for the rest of us in software?” Almost none of the prior unicorn class is better than Figma. “What the hell am I going to say at board meetings this week, Rory? Great job, guys. But have you seen Figma?”
• Rory’s correction — this is the problem with anchoring: Figma is still a $12bn company at 10x forward sales growing 30%+, “an awesomely good company. That’s just the value.” The real lesson is entry price: pay up for growth, growth slows even a little, “the belief goes out of the multiple, you’re in for a long hard hole” — the long flat journey from forward-revenue sizzle to “the steady anchor of 12 times free cash flow.”

2. “Venture is a bit of a scam” — and Rory’s basket-of-Microsofts rebuttal

• Harry’s provocation: “Our job is to convert very high revenue multiples into cash almost unnaturally through M&A, through public offerings when they haven’t earned it in free cash flow… If we have to go to an EPS world, we’re dead.” And when multiples fall, “no matter what the Carta data says, there is no liquidity. It evaporates from the system.”
• Rory’s pushback — correct description, wrong label: surviving at-scale tech companies are astronomically good businesses, and you can’t wait to buy the next Microsoft at 10x EPS, so you buy a basket priced on forward sales. “Four out of five of them turn out not to be Microsoft. No one even remembers what Ballin Software does” — but write off half a trillion in guessing and the winner is worth $4 trillion; the system was right.
• The corollary for old-school SaaS, stated with unusual care: “Your $100 million revenue SaaS company is an awesome entrepreneurial achievement… It’s just not something that we can properly finance, because we’re just not going to make a public-market venture return here.”

3. The mid-stage SaaS playbook: attach to AI, or learn to run without capital

• Harry’s case study — a Linear or RevenueCat-generation company at $50–75M growing 75–125%, not AI-first: Rory says stop worrying (“100% at 50 → 100, decay to 80 → 180, decay to 40–50 → 300 — you’re going to get to scale”); Harry disagrees structurally — he hasn’t bought 20% of a startup in seven years. Jason says at 4–5% ownership, even a $4–5bn outcome doesn’t return the math. “I don’t know how. Teach me about crypto.”
• Jason’s prescription is urgent and specific: find your AI tailwind now. RevenueCat tripled developers on its platform in 3–4 months off vibe-coded mobile apps; a decade-old, sub-$10M, cash-flow-positive portfolio company added deep AI analysis in December, blew up, and will more than double after ten years of ~10% growth. “Those kids at YC built an agent for your space. Why the f* wasn’t that you? We all use the same LLMs. There is no excuse.”
• Rory’s harder truth for the $50M company growing 50–70%: “the probability of a mid-stage SaaS company exploding into something amazing is rounding error zero” — so “live in the world you now find yourself” and run the business assuming no more cheap venture capital. Grind at 50/40/30 to $200M revenue, sell at 5x, own 20% — that’s $200M, top ~5,000 richest people in the world. “It’s okay, but it’s a grind. And that last sentence explains exactly why venture guys aren’t investing. Grind is not in our MO.”

4. Thinking Machines is a seed round with extra commas — hand the money back

• Two co-founders gone in a week (“Barret’s off”), a $50bn last valuation, and Jason’s entire analysis fits in a sentence: “It happens with seed rounds.” Rory upgrades it to the episode’s insight: the number one cause of failure at seed is founder incompatibility, “and this is just seed rounds with extra commas” — people discover after a year they don’t want to be doing this.
• Jason’s deeper skepticism, stated as a permanent rule change: on paper, likely Andreessen and Sequoia putting in $2bn to re-run the Anthropic playbook was a 15-minute yes — but the CEO’s background is non-technical (his telling: an arts degree from Colby College, engineering PM at Tesla). “I don’t know how you run a lab if you’re not Ilya… I’ll never do an investment again where the CEO isn’t one of the greatest technical visionaries in the industry.”
• Rory’s mechanics of the graceful exit: redemption clauses trigger when more than X team members leave — spend $200M of the $2bn, take another $200M “to bribe everyone to go along with it,” and return $1.6bn. “That was a risk that didn’t work. I only lost 20 cents on the dollar” — far better than “the long and bloody march trying to fix this thing.”
• The kicker math on recycling: get the 0.8 back inside the fund’s investment period, put it into the last Anthropic round six months ago, and “that 8 is now a 1.6 — on your initial billion you’re actually ahead 60% despite having lost 20.” “At heart venture is a capital allocation business… reallocate to success and away from failure.”

5. Researchers pick the mission — venture built an efficient labor market

• Jason on why the AI talent war isn’t about comp or brand: “the best researchers in AI only want to work on what they want to work on — and they will leave a lot of money behind.” Hence OpenAI’s move to no-vesting: “come to OpenAI, you will lose nothing.” The rule: “You have to provide the job the researchers want. You don’t get to decide what they do” — an environment 99% of software companies can’t create, and one Thinking Machines may no longer offer.
• The live test: [likely Yann LeCun] raising $500M at $3bn to hunt a new vein rather than “grinding out the fifth marginally better LLM.” As a money person, Harry calls it “a lot of risk” — the industry spent 2017–2022 unlocking LLMs and he’s betting that’s the wrong vein — but to a researcher, “maybe I can figure out the 2026 equivalent of ‘attention is all you need.’” Downside is cushioned if a talent acquisition recoups the 500; Jason’s open question is what redemption or rescission terms now get written for the if-the-talent-bails scenario.
• Jason’s wry close: VCs made capital an efficient market and “were shocked to discover that labor makes it a very efficient market for labor in response… Carl Marx would be kind of glad. This is what successful labor empowerment looks like.”

6. Elon’s $100bn claim: “it was a for-profit all along”

• Jason insists the story starts pure: Elon chipped in $30M+, Sam $10M, Reid Hoffman more — genuine charity to head off existential AI danger. “Thank you everyone for trying really hard to save humanity. And then the old rule applies: no good deed goes unpunished.” By 2017 the costs made the nonprofit untenable; Elon and Sam/Greg split over how to convert; the for-profit conversion finally closed in late 2025 with California and Delaware approval, the foundation holding 30%+.
• OpenAI’s defense: you donated to a charity that now owns 30%+ of one of the largest companies on the planet — a ~$150bn foundation built from $30M. “You got what you paid for… you’re entitled to nothing.” Elon’s counter is far more aggressive — fraud from day one: “if I was the $30M seed in what was a for-profit company, then I want what a $30M seed would get” — $70–130bn in extra OpenAI shares, everyone else diluted. A judge denied summary dismissal: “there is a credible discussion here.”
• Why he’s doing it, per Rory: “billionaires gonna billionaire” — he feels shafted, it’s good for likely Grok if it slows OpenAI down, “and you might win a hundred billion” for only a couple hundred million in legal fees. And discovery is the show: Brockman’s 2017 diary (“what does it take for me to get to a billion dollars?”) is already on the record, and Ilya sat a 10-hour depo on the 2023 board fiasco where “everyone looks like an idiot… amateur hour everywhere.”
• Jason’s Brockman theory — the human read on those journal entries: Sam recruited him out of Stripe (fourth employee) when it was worth $3bn, and he watched the fortune he left behind compound. “How would 99% of humans feel? ‘If I just stayed at Stripe and played Mind Sweeper, I could be worth 10 billion.’ I think that haunted him” — the diary isn’t greed, it’s haunting, “and I think that creates a whole bunch of bad facts the deeper we go.”

7. Elon wins no matter what — and how investors should price the asterisk

• On the merits, Rory thinks Elon loses: he must prove they intended the swipe while raising charitably — “a high bar” — and “if they wanted to do that, they could have just done what Anthropic did, the totally sensible interim stage of a public benefit corporation, and saved all this freaking drama.” In an Oakland civil courtroom of jurors who “don’t like any of these people,” Elon must prevail: “as long as there’s one person on the jury who hates him more than he hates Sam, he ain’t going to win.”
• But normal-people logic doesn’t apply. Jason: this is the sequel to The Social Network — a normal plaintiff settles on the eve of trial for 4–5% “like the Winklevii got,” but “they ain’t going to settle… he’s going to go to trial.” Rory agrees on the psychology: Elon is “post pain when it comes to public shame and obloquy” — sometimes a billionaire just spends a couple hundred million to grind the other guy, “like Peter’s spend on Gawker.”
• The panel splits on whether Sam should fold: Harry would take the dilution to end three years of distraction (“bring the enemy inside”); Jason notes OpenAI arguably has a fiduciary duty to have explored settlement, “but you can’t settle if he doesn’t want to settle.” Jason’s operational warning: litigation that subsumes the company is the real danger — the GC’s order should be “other than depositions, none of you talk about this ever again” — while Harry piles on: Gemini is killing them on consumer, Anthropic on enterprise. “The guy laughing with popcorn is Dario.”
• Pricing it, per Harry: a new investor asks “am I paying 600 or 800 billion — do I need to say it could be a trillion, because it could be 20% dilution?” The answer is probability-weighting — maybe a 10% chance of 15–20% extra dilution, an asterisk rather than a discount to zero, because “greed trumps fear.” And the counterintuitive close: post-conversion, “OpenAI from a structure perspective is less risky now than it’s ever been” — the true existential risk was the conversion failing, and that cleared last October–November.

8. Ads are inevitable — LLMs are the new prime discovery real estate

• Rory’s why-now: Google (2001–02) and Facebook (2005–06) each agonized about a year before accepting “there’s simply no other way to monetize.” ChatGPT’s free-tier cost-to-serve is higher than either had, and consumer conversion runs “well under 5%, probably 5% or less.” “You can hate ads till you’re blue in the face, but America wants free stuff… rip the bandage off, especially if capital is going to get more expensive.”
• Jason thinks the ads will be additive, not extractive: he already does vendor discovery on Claude, and a ~10:1 analysis-to-ad ratio with auction-matched intent is “a win-win for everybody” — early-Google ads before “enshittification at scale.” His Gemini-assisted math: at a $50 CPM, 0.22 ads per prompt — one in five interactions monetized — is $25bn of search-style revenue. “This clip will be used in three years’ time when they are at a hundred billion dollars in revenue and you’ll look back and go, wow, we underestimated this.”
• Rory’s caution cuts the other way: OpenAI is doing around $20bn of revenue, so a billion of ads is only 5% — “it better do a billion dollars pretty damn quick,” because against this capex even that is a drop in the bucket. The bigger signal is where dollars come from: Google’s ~$240bn ads cash cow “isn’t the best product on the market anymore… search is not the best place for discovery anymore.” His own proof: five or six ChatGPT queries to pick a TV, then a walk into Best Buy — and Amazon’s intent-driven ads already generate all the retail side’s gross margin.
• The investable adjacency: Jason floats “let’s all put a bunch of money into likely AppLovin and the Trade Desk” if OpenAI opens to third-party buying — while calling today’s GEO tools “snake oil.” Rory kills half the trade: paid placement will be an OpenAI-run real-time auction, so “all the value on paid is captured by OpenAI” — you don’t need AppLovin for Google or Facebook either. The free side — answer-engine optimization — is the real business: likely Profound, AirOps, Evertune, Harry’s Peak.

9. The AEO trade already happened: Adobe bought Semrush for exactly this

• The missed trade, per Rory: Adobe acquired Semrush at ~4x revenue, and “that acquisition only makes sense if their plan is, as quickly as humanly possible, to introduce an answer-engine-optimization product.” Both guests have sold companies to Adobe, and both say the same thing about why it wasn’t a Peak-sized startup instead.
• Jason’s version: a conservative acquirer buying $15–20M of GEO revenue at 4x overall “is the kind of accretive deal that makes people comfortable” — then Adobe walks it into CMOs rebranded at $20k a month, “and it just works… versus buying some crazy startup that could blow up on you. Toby could buy a company like that; it’s just not going to be Adobe.”
• Rory’s Omniture lesson from the board seat: “when you want to make a big move as a big company, if you buy something small you’ll just smother it, you’ll stomp on it” — you need critical mass and heft so the asset “can go do its thing and not get swallowed by the machine.”

10. New rounds and the late-stage question: ClickHouse, Replit, and Fidelity-in-privates

• ClickHouse at $15bn (Cerebras’ $22bn round got a pass): born inside Yandex, spun out in 2021, it nailed the open-source-to-proprietary-cloud conversion just as AI blew up — Tesla was early, “fast forward, Anthropic and everybody needs it,” and it was doing only ~$50M a year ago. Jason’s summary of the bet: “you’re just assuming everyone uses it… just go find your ClickHouse.”
• Jason’s framework for the price: paying up means “underwriting growth persistence” — “two to three years at 3 to 4x growth at 350 — pencils out” — plus a real category: OLAP is the Teradata/data-warehouse analog, historically an appreciable fraction of the transactional market. If Snowflake (80) and Databricks (110) anchor the core, “maybe you get a $30–40bn outcome… if it turns out to be a smaller market than you think, you’re high and dry. Welcome to late stage investing.” His risk stack: technical/founder risk, then go-to-market risk, then “valuation risk expands to fill the gap.”
• Replit at $9bn (Lovable at 6.5, roughly neck-and-neck on revenue): Jason’s defense is product, not comps — “this product is like 50 times better than at the $2–2.5bn round.” On V1 he couldn’t finish an application; over the holidays he built a working startup-simulator game in ~100 hours. With ~$250M ARR plausibly running to 700–900 by year-end, “this is a much less risky investment today than it was at 2.5.” Rory’s own replit epiphany: “it’s not programming at the programming level, it’s describing at the describing level” — the code resists being found, and you have to let go, like trusting the spreadsheet to add the numbers.
• The Sequoia question — into Anthropic while holding OpenAI — gets waved off: “no one gives a damn about competitive investing anymore in 2026.” At $350bn pre “you are Fidelity large-cap growth, just in the private markets” — a billion dollars buys 0.3% and no information rights, “about what you’d hire a director for at Series B.” You couldn’t do both Series As; at the F it doesn’t matter. Then the stage fight: Jason calls competitive Series A/Bs at 100x ARR for 10–12% ownership “the dummies game” — he’d rather own 15–20% of a pitch deck or pile into likely Airwallex at $5bn. Rory’s rebuttal: markets tend to rational equilibrium and “you can duck around in the box, but you can’t do violent switches” (soccer to golf) — though Thrive earns credit for excellence at both ends. The zoom-out both accept: “the public markets ceded another 3 to 5 years of growth to the private markets,” which is why LPs hand mega-funds 2-and-20 money saying “I can’t get me that Anthropic in the public markets. Go get me some.”

Verification Notes

• The raw captions are ambiguous on the ads-per-prompt figure (“22”/“222”); the digest’s “0.22” cannot be verified.