Summary

Mega-fund venture math no longer requires a single fund-returner; it requires repeated $500 million wins and one radically concentrated position. Rory O’Driscoll models 20 deals as 30% losses, 50% returning 1–5x, and four winners averaging 10x, but says the upper tail determines everything. At $6–8 billion scale, the viable strategy may be putting 20–30% of the fund into the best company: “return that $8 billion in paltry $500 million chunks” and ensure one position returns another $2–3 billion.
Hinge Health shows that expensive preferred stock can no longer reliably block an IPO or preserve a flattering mark. After a 2021 round at $6 billion, Hinge went public around $2–3 billion; some investors negotiated a conversion, while roughly $200 million of preferred remained outstanding until the stock reaches $77, versus an IPO in the mid-$30s and early trading in the $40s. Chime’s documents imply the harsher outcome: above a $6 billion IPO valuation, later preferred automatically converts into common, potentially crystallizing a large loss.
YC has “won” because it combines the scale of Walmart with the aspiration of Chanel. Accelerators and incubators now represent 24% of VC deals, while YC has expanded batches, tilted into AI, and retained the pull of Harvard, Stanford, or MIT. Rory calls it “one of the greatest equity businesses ever”: its repeatable machinery may produce roughly 6x where an equivalently staged seed fund earns 3x, because YC converts raw founders into marketable companies in three months for about 7%.
Seed remains easy to sell, but Series A is brutally bifurcated and ownership economics are deteriorating. Harry’s distinction is “Walt Disney” at seed—tell the story—versus “Jerry Maguire” at A—“show me the money”; the few hot AI companies face mass competition while roughly 75% of candidates struggle for capital. Jason compares RevenueCat at a $7 million pre-money valuation in 2018 with a similar-risk YC company at $30 million today, then warns seed investors may suffer two-thirds dilution by IPO without pro rata.
OpenAI’s $6–6.5 billion Jony Ive transaction is simultaneously a hardware hedge, talent purchase, and fundraising narrative. Jason expects a subsidized $20–50 device within a year that could expand ChatGPT engagement from 20 minutes toward all-day use; Rory predicts the usual platform-company “hardware paranoia” will probably end in a three-to-five-year fizzle. Harry’s sharper capital-markets read is that “the guy who did fucking Apple” gives Sam Altman a fresh story for raising the roughly $50 billion he says OpenAI needs to spend.
Corporate leaders are converging on deliberately bland AI messaging while disagreeing sharply about the employment timeline. Duolingo’s cited figures are 140 courses made with humans over ten years versus 140 made with AI in one year, yet backlash forces CEOs to append, “In fact, we’re hiring.” Jason expects mass layoffs within 24 months and says companies above 500 employees privately believe 30–40% of staff are unnecessary; Rory expects a slower 60-month adjustment, closer to 2–3% less hiring each year.
Only 20–30% of America’s 646 tech unicorns may still merit a $1 billion valuation, and realized outcomes could be worse. The emerging IPO bar is roughly $200–300 million of revenue, about 30% growth, and profitability or proximity to it; even two successful IPOs a week would take more than six years to clear the backlog. Jason’s counterweight is that 2021 handled roughly one IPO a day, so market capacity exists—but only if companies “reaccelerate to growth.”

Deep dive

1. Builder.ai is painful, not existential, for Insight

Jason Lemkin initially found Builder.ai’s collapse shocking enough that it “almost seemed like fraud.” He said he read that the company reportedly raised roughly $500 million, projected about $200 million of revenue, delivered approximately $45 million, and was shut down after missing projections under its debt arrangements. The important distinction is that aggressive plans routinely miss; the severity lies in the scale and alleged gap, not merely failing to hit 100% of plan.
Harry’s scale framing: an Insight exposure north of $100 million is enormous in ordinary terms but approximately 1% of a reported $12 billion fund. Rory’s conclusion was that someone does not automatically lose their job for one failed position—especially when senior investors have generated billions elsewhere, including through Wiz. “Some of those $100 million checks don’t work out”; persistent losses relative to wins are the employment problem.
Insight’s Hinge Health result sits on the opposite side of the ledger: roughly a 5x and $400 million returned. That is a good venture outcome, but against a $6.2 billion fund it is not transformative, forcing investors to abandon the reflex that every successful deal must return the fund.

2. Mega-funds need concentration, not mythical fund-returners

Rory’s operating model uses 20 investments: 30% are bad, 50% return 1–5x, and four companies—the remaining 20%—return more than 5x with an average near 10x. Those four contribute about 2x the fund, while base hits contribute another 1.5x; success therefore requires several half-fund outcomes, not one miraculous position.
The only variable capable of radically changing the result is the size of the best outcomes. Moving a loser from 0.2x to 0.8x barely matters, and middling deals are definitionally unable to rescue the portfolio; every three funds, a forecast 10–15x winner might become a 20x, 40x, or 50x, but assuming that miracle every vintage is “a fatal error.”
Large-fund arithmetic eventually collides with market supply. If a strategy requires six $10 billion exits annually while the market historically supplies only four—and multiple funds need the same outcomes—equal-sized diversification cannot work. Rory’s answer is to “stuff money into the very best company,” as Founders Fund has with Anduril, until one position represents 20–30% of the fund.
Concentration is more defensible late because the company has revealed more information; at seed, “you know jack.” A scaled late-stage franchise must harvest many $500 million gains, then make certain one heavily weighted winner returns $2–3 billion—an entirely different business from assembling an early-stage portfolio.

3. $100 million ARR is a magnet for talent and capital, not a verdict

Rory calls speed to $100 million ARR an important but incomplete signal: beyond seed, traction is the best available proxy for commercial success, so ignoring it would be foolish, but weighting it at 100% would be equally foolish. The decisive refinement is “$100 million with low churn—that’s very bloody meaningful.”
Jason’s concern is that slower, less dilutive companies with stronger moats may look attractive mathematically yet lose the surrounding contest. AI has intensified the “moth to flames” effect: elite engineers want OpenAI, Windsurf, Cursor, or other frontier companies, while even strong B2B employers such as Rippling must compete for the same people. He also warned that many B2B VCs will not fund a company without a credible hypergrowth path, so slower companies need their own recruiting, capital, and operating ecosystem.
Rory defends that career choice as rational. Joining the defining wave offers five or ten years of frontier experience, relationships, and knowledge that may shape a 30-year career, much as joining SaaS around 2004–05 offered a durable professional runway.
Compensation committees should judge dilution alongside two marketplace signals: employee attrition and offer close rates. The cited two-year retention figures—67% at OpenAI versus 80% at Anthropic—suggest materially different talent dynamics; if people leave or offers fail, dilution “maybe isn’t high enough,” however painful that conclusion is for investors.

4. Hinge Health and MNTN prove the IPO window is open

Hinge Health and MNTN are facts on the ground against claims that companies need $500 million of revenue or that the window had already shut. Both reached public markets with roughly $200–300 million of revenue, solid growth, and profitability or proximity to it; Hinge’s cited growth rate was approximately 48%.
These were not speculative shells but “great classic venture outcomes”: multi-billion-dollar listings of businesses growing roughly 30–50%, with meaningful value for founders and early backers. The new practical threshold is no longer $100 million of revenue, but it remains attainable below $500 million.
The uncomfortable implication is how few private unicorns resemble either company. Rory calls it “actually terrifying” once the bar is stated clearly: hundreds of marked billion-dollar companies remain far below the revenue, growth, and profitability profile that public investors are currently rewarding.

5. Preferred protections are becoming isolated piles of underwater capital

MNTN appears to have completed a conventional IPO without a blocking preferred round or an obvious down-round conflict. Hinge was the revealing case: it raised at $6 billion in 2021, then listed around $2–3 billion, forcing investors and public buyers to confront expensive late-stage preferred directly.
Coatue apparently negotiated—selling some shares back and buying common—to permit conversion. Other preferred holders did not convert: roughly $200 million remained outstanding with a conversion threshold near $77 per share, even though the IPO priced in the mid-$30s and traded in the early $40s.
Those holders have not technically “lost money” because their 1x preferred claim remains, but they own a non-interest-bearing, illiquid instrument inside a now-public company. Economically, it is underwater and worth less than 1x on a discounted basis: “That late-stage money is stuck at a 1x, 0% IRR for the next three years. Knock yourself out.”
The public market’s message is that a messy capital structure can be priced rather than cleansed. Founders and early investors can obtain liquidity and continue building while stranded preferred remains on the balance sheet, substantially weakening the assumed power of late rounds to stop an IPO.

6. Chime may crystallize what Hinge could postpone

Rory’s reading of Chime’s pre-IPO articles found a different auto-conversion term: above roughly a $6 billion valuation, the later preferred converts into common. A hypothetical holder carrying a $25 billion-round investment at 1x could therefore become common at, for example, a $12 billion valuation and record roughly 0.5x immediately—though Rory stressed that he did not know the eventual trading price.
The comparison exposes three possible outcomes for late capital: preserve a nominal 1x in stranded preferred, negotiate a make-good and convert, or convert automatically and recognize the loss. The result can turn on “one little term deep in the bowels of the liquidation-preference auto-convert terms.”
Jason connected this to other supposed protections. In one bad investment, an acquirer closed with only 80.1% shareholder approval and did not seek the other 19.9%, contrary to the assumption that buyers demanded near-unanimity. “Anything you can get around, people are gonna get around” when liquidity is at stake.
Rory’s broader explanation is that capitalism must process roughly $2.7 trillion of private assets and about 600-plus unicorns. Transactions will accept more structural noise because capital needs a home; complexity will not disappear, but it will become a price rather than an absolute veto.

7. YC has become both Walmart and Chanel

Accelerators and incubators account for 24% of VC deals, and Jason’s default conclusion is that YC has won. It now runs four larger batches, received a major uplift under Garry Tan, and pivoted aggressively toward AI; for young founders, its pull resembles Harvard, Stanford, or MIT even as competing accelerators multiply.
Rory deliberately calls YC a business, not merely a fund. A conventional investor is “only as good as your last game,” whereas YC’s machine keeps working even if Paul Graham is “walking around the cute little bookstores” in England: it industrially turns founders from London, Sweden, or the Midwest into financeable companies over three months for roughly 7%.
His rough return comparison gives YC a structural 2x advantage. Where a competent seed fund selecting at the same stage might make 3x, YC’s locked-in access and published hit rates could imply 6x. It fulfilled a real market need—making startups easier to begin—at enormous scale, so the economics are earned rather than accidental.
Harry’s formulation captures the moat: venture may be won by Walmart’s breadth or Chanel’s exclusive aspiration, but YC is “Walmart and Chanel.” It scaled supply without surrendering brand; if YC vanished, another institution would have to fill the gap, whereas the ecosystem could lose one of 600 venture firms and simply continue with 599.

8. Seed sells the dream; Series A demands proof

Harry describes seed as Walt Disney—“tell me the story”—and Series A as Jerry Maguire—“show me the money.” Many credible founders from excellent companies can narrate a compelling future; far fewer demonstrate sustainable, high-quality economics that a Series A investor can underwrite.
Rory agrees despite Carta data showing seed-to-A conversion deteriorating. The apparent contradiction is a bifurcated market: roughly 75% of available companies are dying or capital-starved, while the few AI companies showing explosive early traction attract nearly every venture firm.
His own firm lost two competitive processes in emerging AI categories—“outpriced in one, out-beauty-contested on the other.” That creates a possible contrarian opening: the profitable Series A strategy may be finding a non-obvious company among the neglected majority, though doing so does not remove the fundamental difficulty of identifying quality.

9. Entry prices and dilution are quietly rewriting seed returns

Jason invested first in RevenueCat in 2018 at a $7 million pre-money valuation; a recent YC company with what he considered a similar risk profile was priced around $30 million. RevenueCat later raised at approximately $500 million in a one-hour process, but the entry comparison still asks whether today’s seed investor needs four times the fund, accepts four times the risk, or both.
Rory adjusts the nominal comparison for GDP growth and inflation, calling the present risk per dollar perhaps 2–2.5 times worse rather than 4 times. Investors still must “play the game on the field”: his firm has maintained roughly 10–11% initial ownership since 2009, but checks have risen materially to preserve it.
Long holding periods make annual employee issuance compound brutally. Rory tries to manage mature-company refresh dilution near 3–4% rather than 5–6%, while Harry cited an investor’s LLM-company example of employee stock running at 9–10% annually; the work is “low joy, high impact” because founders and essential staff need renewed incentives deep into years seven, eight, and ten.
Harry enters deals assuming 40% dilution, but Jason says that is too low at seed: without pro rata, total dilution through IPO may approach two-thirds, versus roughly half historically. YC’s fixed initial stake and thousands of outcomes offer the cleanest dataset—and Jason argues its post-money terms, added ownership, anti-dilution, and follow-on investing show it recognized the change early.

10. Vintage and ownership made yesterday’s modest exits exceptional

MNTN was reportedly founded around 2009, and early Bonfire investor Jim Andelman still owned roughly 9% at IPO. At a $2 billion market capitalization, that stake approaches $180 million against what Jason estimated was a $20–30 million early fund—an exceptional outcome even without a giant headline valuation.
The same deal today might deliver only about 1x the fund rather than 5–6x because seed funds are larger, entry prices are higher, and ownership at exit is lower. Harry cited Michael Kim’s example: Erik reportedly generated about 12x DPI for Mucker through Honey and returned $280 million.
Rory’s qualification is “horses for courses.” MNTN and Hinge built in defined markets without fighting every trillion-dollar platform; foundation-model companies compete against Microsoft, Google, and OpenAI while one participant says it may require another roughly $50 billion to reach cash-flow break-even. “The wars that you choose to engage in dictate what it has to take to win.”

11. OpenAI is paying for hardware insurance and a new capital story

Jason’s first surprise in OpenAI’s roughly $6–6.5 billion Jony Ive transaction was that Ive would not join full time; he would continue managing his design firm while OpenAI bought the startup. Paying that amount without securing the central figure exclusively was itself “a sign of the times.”
The bullish case begins with engagement: ChatGPT reportedly averages 20 minutes per user per day. Jason thinks a cool, heavily subsidized $20–50 device could launch within a year and turn “20 minutes to 200,” eventually making AI ambient throughout the day; he also recalled a suggestion that shipments could reach 200 million.
Rory’s prior is the opposite. Every major software platform develops “hardware paranoia”: Microsoft bought Nokia and built Surface, Facebook pursued VR, and Google produces Pixel phones at little margin. He expects OpenAI’s effort is sensible insurance but statistically more likely to fizzle after three to five years than become a meaningful hardware-revenue business.
Harry’s capital-markets interpretation reconciles the bet: a Jony Ive hardware chapter makes another enormous raise easier to sell when funding yet another model is becoming harder. At roughly 2% of market cap, it is not a Hail Mary; it is also a stark hierarchy—one investor writes a $6 billion check while a 55-person team receives comparable ownership for scarce design talent.

12. San Francisco’s moat is density; London’s is concentrated talent

Jason says San Francisco’s AI-founder density may be higher than in 2019 even though the broader Bay Area ecosystem is smaller. In Dogpatch, founders repeatedly encounter YC peers and industry leaders; the distinctive motivational product is feeling that “no matter how well you’re doing, someone else is doing better.”
Harry rejects external comparison as a prerequisite for exceptional performance. London offers cheaper access to deep AI talent around DeepMind, while ElevenLabs, Synthesia, and Granola have created a concentrated proof set and local recruiting ecosystem. The best founders, in his view, run on “internal fire,” not the need to feel like failures beside their neighbors.
Rory separates individual quality from system quality. European entrepreneurs who succeed despite a less conventional path may possess unusual determination, but America offers stronger mechanisms for recovering from failure, raising again, and converting talent into scale: “Exceptional people rise to the top anywhere”; the US superpower is making even mediocre people “damn successful.”
Project Europe is an attempt to build those mechanisms locally. Harry reported 8,000 applicants, with 300–400 looking “pretty fucking awesome”; Rory’s YC-derived logic is simple—make starting easier, accept that most companies will be mediocre, and let a few extraordinary winners “cover a multitude of sins.”

13. AI efficiency is real, but CEOs cannot say plainly who loses

Duolingo’s cited productivity figure is the clearest specimen: humans created 140 courses over ten years, while AI was associated with producing 140 in one year. Yet its leadership, like Klarna’s, faced backlash after describing an AI-first posture too directly.
Jason believes such walk-backs retreat to “70% of the truth.” Public-company CEOs privately tell him they do not need 30–40% of current staff once organizations exceed roughly 500 people; he expects mass layoffs within 24 months, although he expects net headcount to stay flat.
Rory agrees on the messaging but not the speed. Corporate language will promise efficiency to Wall Street, reassure employees that no mass layoffs are coming, and add, “In fact, we’re hiring”—“the current state of the lie.” His economic forecast is a slower 60-month grind, perhaps 2–3% less hiring annually, rather than Jason’s 12–15-month break.

14. The quick-fire bets expose where definitions and marks can move

On AGI, Rory refuses the technological over-under: because AGI is ill-defined yet contractually relevant to Microsoft and OpenAI, he expects it to be declared when one party can use the definition for economic leverage. Jason predicts it will feel like AGI in 2026 and that very smart people will agree around 2028; Harry takes under 2030.
Rory says the US corporate rate will not be cut this year: the 21% rate was made permanent in 2017, and the discussed bill changes secondary international-tax provisions rather than the headline rate. Jason separately estimates that lost California pass-through deductions plus a state increase could raise his personal burden by roughly seven percentage points, while acknowledging “no one’s gonna cry for us.”
Rory bets the first $500 billion fortune arrives well after 2027 because high starting equity valuations imply lower long-run returns. His caveat is Elon Musk: private holdings such as SpaceX can receive paper step-ups “untrammeled by any form of reality,” enabling a half-trillion-dollar mark without public stocks compounding enough to produce it.
Jason sides with the nearer, more bullish outcome, effectively betting on renewed double-digit Nasdaq growth despite admitting it conflicts with mean reversion. He found the market’s roughly 38% probability intuitively plausible, then disclosed that his own positioning was “all in on the market.”

15. Most unicorns will fail the newly visible public-market test

Of 646 US tech unicorns, Rory estimates only 20–30% still deserve a $1 billion hard-cash valuation. The rough qualification is at least $100 million of revenue, more than 20% growth, and something approaching profitability; the other 70% have value, but are “sub-scale, sub-growth, sub-profitability.”
Jason fears realized outcomes could be half even that estimate because marks do not create buyers. The viable proportion already appears lower than GP markdowns imply, and today’s private-equity and acquisition liquidity is weaker than in 2021.
At two successful IPOs per week, the market would process about 100 annually and need more than six years to clear 646 companies. The investable bar shown by Hinge and MNTN is harsher still: roughly $200–300 million of revenue, 30% growth, and profitability or proximity to it.
Jason’s pushback—worth keeping—is that 2021 sustained approximately one IPO every day; TechCrunch reportedly stopped treating $1 billion listings as inherently newsworthy. Capacity is not the binding constraint. The “culling of the herd” depends on how many companies can reaccelerate enough growth and operating quality to satisfy the market, while aggregate returns are ultimately driven by only five or six extreme winners.

No Priors Ep. 116 | With Sarah and Elad Nico Rosberg on the Booming Business of F1 | All-In Live from Miami