Summary

OpenAI’s rumored $3 billion Windsurf purchase is a 1%-of-market-cap hedge against its clearest strategic weakness: coding. Jason Lemkin treats 10% as “bet the farm” and 1% as an SVP betting a business unit; Rory O’Driscoll says a $300 billion company chasing a $2 trillion outcome should buy relevance while “no one knows nothing.” Even if Windsurf does not sell, “the mighty corporate intent has been stated.”
Multi-stage funds can dominate Seed and still face brutal fund math. Harry Stebbings cites Greenoaks leading Windsurf’s seed and doubling down at Series A, yet the reported $500-600 million return would be less than one-third of a $1.5-3 billion fund; Jason’s warning is sharper for dedicated seed managers who may now scrape 2-3% ownership rather than roughly 12% while funds grow: “If a unicorn can’t return the fund,” the model is under compounding pressure.
A 100x revenue multiple is not an investment thesis until it is paired with growth and persistence. On a company at $7 million of revenue and a $700 million valuation, the forward multiple was roughly 33x because revenue was expected to triple; Rory argues that 100x growing 3-4x with persistence can exit the danger zone within two years, while deceleration leaves investors “so screwed your head’ll hurt.” AI makes that persistence harder to forecast than SaaS’s old “apply capital and grow into the multiple” playbook.
AI is collapsing the interval between product-market fit and a multibillion-dollar outcome, not necessarily eliminating the pre-PMF search. The “walk in the wood period” can still last six months or five years, but Windsurf went from near-obscurity to a potential $3 billion transaction extraordinarily quickly. Jason’s mechanism is that perhaps 95% of early adopters are currently in market, helped by self-serve products costing $5 or $20 rather than traditional enterprise deployments costing tens of thousands.
Endowments face a cash problem more serious than the familiar denominator effect. Public-market weakness, delayed private distributions and potential losses of university funding collide with mandated spending, so Rory says a CFO facing the possible disappearance of 30-40% of revenue would be thinking “bonds,” index funds and accessible cash—not more illiquids. Existing manager relationships will be protected first, making 2025-26 materially harder for spinouts even when LPs like the talent.
AI rollups are attractive only where acquired customers are genuinely interchangeable. Rory’s objection is that legacy customers “didn’t pick you” for the AI product, making broad rollups a pile of low-multiple services and churn risk; Harry counters with a real-estate management company that reached $30 million of revenue in two years and lifted margins from 5% to 40% within six weeks. Rory concedes the model when the customer base and workflow are truly uniform.
Conviction can justify breaking price and ownership rules, but only governance determines who may make the exception. Jason will do any deal where he has “100% conviction” of a 5x regardless of ownership; Rory agrees that informed certainty is rare and should be monetized, yet argues that peer-led firms cannot institutionalize one leader’s exception card without corroding strategy. Jason instead filters for top-0.1% growth, an S-tier CEO and CTO, accepts brutal competition, and wants “one more Wiz” with a real $8 billion outcome.

Deep dive

1. Windsurf would be a 1% hedge on OpenAI’s coding gap

Rory’s first reaction to the rumored $3 billion transaction was the classic venture reflex—check the cap table, discover “you’re not in it,” then ask what it means. His strategic answer: coding is one of AI’s few visibly enormous use cases, and buying a serious position for roughly 1% of OpenAI’s $300 billion market value is rational.
Jason cautions that “we don’t know” whether the deal will happen, but offers an M&A hierarchy learned inside big tech: 10% of market value is a CEO-level “bet the farm,” as with Facebook’s Instagram and WhatsApp deals or Adobe buying Figma; 1% is an SVP saying, “I’m betting my BU.”
The competitive premise is explicit: Jason believes OpenAI has pulled irreversibly ahead in direct chat revenue, while Anthropic owns the coding mindshare and developer advocacy. Harry’s formulation is harsher—OpenAI could either “cede the market entirely” or acquire its way into contention, with Cursor likely too expensive and unwilling to sell after its latest round.
Jason disputes that Cursor was literally unbuyable: at 10% of OpenAI’s value, Sam Altman could offer $30 billion, roughly 3x Cursor’s last reported round, and test that conviction. The real lesson is scale: “The clock’s ticking,” and growing platforms must make different bets from stalled companies.

2. The app layer can stay open while platform power compounds

Jason says ownership need not immediately produce a Salesforce-or-Oracle-style ultimatum. OpenAI could put “1,000 engineers” on Windsurf while continuing to support Cursor and other coding tools, letting customers decide rather than ruthlessly forcing the owned application.
Rory agrees only on the time horizon. Microsoft supported independent applications around Windows, but after a “long 10-year grind,” only a few scaled productivity vendors remained outside Microsoft. If OpenAI both owns and continually improves a coding product, independents may retain access while still being slowly “ground down.”

3. OpenAI’s bias to action matters more than perfect foresight

Rory frames the acquisition as a bet under radical uncertainty, not a claim that Windsurf is defensible forever. Microsoft’s bought-in elements became part of Office; Excite@Home’s marriage of search and cable infrastructure was “dumb as rocks.” At this stage, nobody reliably knows which precedent applies.
The AI narrative itself has already cycled from “models are everything” and applications are wrappers, to models becoming commodities and apps holding the value, to models accumulating enough market capitalization to buy the apps. Rory’s conclusion: “No one knows nothing,” so refusing to act is not prudence.
Altman’s strength, in Rory’s telling, is a “bias to action”: direct chat, tick; coding, next; customer success, later. “Not making a move is, like, akin to losing” when a platform is competing for one of the handful of eventual $2 trillion positions.
Even a failed transaction changes the market because “the mighty corporate intent has been stated.” If Windsurf remains independent, every other coding application now knows OpenAI wants to own a major endpoint; Rory expects the line outside its office to go “around the fricking block.”

4. Multi-stage seed wins expose a harsher ownership equation

Harry reads Windsurf’s cap table as evidence for multi-stage dominance: Greenoaks reportedly led the seed and doubled down at Series A. Its cheaper capital and ability to support companies across rounds make the seed product so formidable that Harry told an LP asking for a San Francisco seed manager, “I wouldn’t touch it.”
The fund-return math tempers the victory. Harry estimates Greenoaks’s relevant fund at $1.5-3 billion and says its Windsurf position reportedly returned $500-600 million; even against the smaller figure, that is less than one-third of the fund. Jason’s deadpan calculation: a 5x fund would need roughly 15 comparable outcomes—“Man, venture’s brutal.” He also questions whether a $1 billion outcome can still return many seed funds, let alone produce a 3x net return.
Rory pushes back on making one cap table a systemic verdict. He sees “extraordinarily good picking” by a highly connected investor reaching down from later stages, and insists that seed, Series A/B and growth all look easier from somebody else’s seat. “The dirty little secret is it’s hard everywhere.”
Jason nevertheless accepts Harry’s structural concern: seed managers may now scrape for 2-3% rather than roughly 12% while managing funds twice the size. Harry has seen whole seed rounds compress from 15% dilution—12.5% to the lead and 2.5% to angels—to 10%, split 7.5% and 2.5%.

5. Growth persistence, not the headline multiple, prices risk

Harry’s investment committee had just reviewed a company with $7 million of revenue valued at $700 million—“return of the 100X.” Jason immediately asks for the forward number: roughly 33x, with revenue expected to triple. Rory calls Harry a hypocrite because every seed investor begins at an effectively infinite revenue multiple.
Rory’s core equation is stage, growth rate and persistence. A partner refuses to discuss revenue multiples without growth because it is “an incomplete equation not worthy of discussion.” At 100x with 3-4x growth likely to persist, two years can de-risk the entry; at sub-2x and declining, “you’re so screwed your head’ll hurt.”
The 2021 error was not simply paying 100x; it was paying for growth that never arrived. If $7 million becomes $20-30 million and then compounds through another strong year, today’s frightening price can approach a conventional multiple. If growth fades, the investor has no protection.
Harry’s pushback is that AI product-market fit and revenue may be transient. Rory agrees this is precisely why classic SaaS now looks like a golden age: sticky revenue made sales-and-marketing inputs predictably generate outputs, allowing investors to “apply capital and grow into the multiple.” Today’s dispersion is much wider.

6. AI preserves the product maze but compresses escape velocity

Harry contrasts today with Klaviyo, UiPath and ServiceTitan, which took years to navigate the “idea maze” before reaching $1 million of ARR. Jason notes that Cursor nearly died, Bolt nearly died and Windsurf existed as Codeium before its breakout; the experimentation period has not disappeared.
Rory calls that phase the “walk in the wood period”: it may last six months, one year or five, and seed capital finances it as long as founders do not run out of money and still want to continue. What changed is the post-PMF trajectory—from a company that may fail to one considering whether to reject $3 billion in perhaps 90 days.
The accelerants are cumulative: rapid adoption, easy distribution and universal belief that the prize matters. Unlike the internet four years into its development, AI reached a point within roughly two years where virtually every company believed, “Shit, I gotta do something here.” Boards therefore reward big bets rather than caution.

7. Cheap tools have put nearly every early adopter in market

Jason relays Marc Benioff’s distinction between customers already “all in” and a much larger enterprise market that remains extremely early. His own explanation for current growth is not that everyone has deployed AI, but that “100% of the early adopters are in market.”
Self-serve pricing makes experimentation nearly frictionless: Windsurf at about $20 per month, Higgsfield at roughly $5, versus perhaps $20,000 merely to engage with a traditional enterprise tool such as Atlassian. These products make ordinary B2B software “look like a frigging rip-off.”
Jason therefore reframes apparently impossible growth as a temporary concentration of demand: instead of 5% of early adopters shopping, perhaps 95% are. The remaining 90% of enterprise is still largely experimenting through incumbent platforms, so explosive adoption and an early market can coexist.

8. A hot hand should either get inside the tent or spin out

Discussing Bucky’s departure from Kleiner Perkins, Jason’s rule is blunt: “If you have a hot hand in venture and you’re not running the place, I would leave the next day.” Tomas Tunguz’s ability to raise roughly $700 million independently illustrates the alternative to earning $400,000-$1 million and waiting “22 years” for carry.
Rory refines the point from control to partnership. Talented investors want to be “true partners,” with compensation and influence commensurate with value; leadership must pull a successful younger partner “inside the tent as quickly as humanly possible.” Otherwise the organization has failed its central succession task.
Departures are not always under-reward stories. A rising investor may be performing well inside a firm whose older portfolios lost money, leaving the next five years devoted to “digging out of someone else’s hole.” After illiquid recent vintages, the expected value of staying can rationally look poor.
New firms also let investors “sever” mixed old records “like a stage of a rocket” while retaining stories about their winners. Rory cites Fred Wilson’s unsuccessful Flatiron period before his later success as evidence that a reset can reflect lessons from the cycle, not merely clever attribution.

9. The LP barbell narrows just as new managers need it

Rory sees two simultaneous truths: the mass of newly invented 2021 funds is way down, while a smaller group of proven mid-career investors from top firms can still raise. LPs know those people, can reference them and find the spinout legible in a way that anonymous three-person seed launches were not.
The allocation barbell is psychologically convenient: place $200 million with a mega-platform because it can absorb capital, then give $20 million to an emerging star and “feel good” about backing the future. Rory rejects size as destiny, however: “Our number one job is to be competent,” because performance eventually overrides category.
Harry argues the timing has deteriorated sharply as endowments confront possible fines, tax-exempt-status risk and liquidity uncertainty. Rory agrees that 2023-24 may have been unusually receptive; in 2025-26, even excellent spinouts might “cling to the lifeboats” because LP desire cannot substitute for cash.

10. Endowments have moved from allocation risk to cash risk

Rory’s stress stack begins with down public markets, illiquid portfolios and weak venture distributions, then adds an exogenous policy shock. A university CFO contemplating the possible disappearance of 30-40% of revenue would tell anyone proposing more illiquids to “get the frick out of my office” and prioritize bonds, index funds and cash.
Harry’s objection—voiced on behalf of more than 50 LPs who contacted him—is that skipping a fund can destroy a decades-long relationship. Rory agrees preservation wins first: an LP with one remaining commitment will probably retain Sequoia rather than fund an exciting newcomer, especially if withdrawal means losing access permanently.
Yet that choice may be overruled one level above by an institution demanding liquid rather than illiquid assets. Rory’s phrase is that “liquidity premium” means nothing until it means everything; when scholarships, professors and research require cash, institutions surrender upside, stop making new commitments or sell existing assets.
Yale’s reported effort to sell roughly $6 billion of assets is therefore symbolically important because Yale was the intellectual “godfather of the endowment model.” Rory offers two interpretations: it may simply have misjudged cash-flow variability, or it may believe private markets are structurally overfunded. The latter would be much more consequential.

11. Private equity’s liquidity drought is venture’s bigger cousin

Jason says LPs were already admitting before the latest political shock that distribution and cash planning were wrong: paper returns and IRRs looked acceptable, but nobody expected the drought to last this long. The larger strain is private equity, where institutions may deploy five or ten times as much capital as venture.
Venture is often “juice”—smaller commitments intended to add basis points—while private-equity holdings such as Zendesk and Anaplan consume far more capital. If those deals do not go public and return cash on the expected timetable, venture suffers because its “bigger cousin” has exhausted the liquidity budget.
Rory distinguishes three escalating problems: private returns arrive later than modeled; the expected 17% return, perhaps 600 basis points over small caps, falls toward 15%, 14% or 13%; or the institution “actually need[s] the damn money.” The denominator effect is real, but this combination of timing, compensation and mandatory cash needs is the crisis.
Harry was surprised that more than five major endowments he approached reported over 30% in private assets—and, in some cases, venture—rather than his expected 6-10%. Rory says centuries-long universities can rationally carry illiquidity, provided they are paid for it; what they did not model was a sudden transformation of their operating cash flows.

12. Capital-efficient companies still absorb oversized rounds

The apparent paradox—AI lowers the cost of building companies while rounds grow—is simple to Jason: investors most want companies that do not need their money. As valuations inflate and founders remain insensitive to financing risk, those companies absorb capital up to whatever dilution threshold they will accept.
Jason tells founders that a $100 million valuation is the final moment to stop if they doubt they can become public companies. Above $1 billion, $2 billion, $3 billion or even $10 billion, he sees a younger generation behaving as though there is no risk, particularly when the market is fighting to enter the same deals.
He points to Accel figuring out how to buy 30% of Atlassian and recalls the ideal of owning 20-30% of a bootstrapped company as its only investor. Rory captures the contradiction: VCs seek founders who neither want nor need them, then try to invest in those businesses “in a capital inefficient way.”

13. AI rollups work only when customers are interchangeable

Rory begins with a success: SpeechWorks, which became Nuance, bought small medical-transcription businesses from roughly 2005 to 2015, injected speech-recognition technology and created value. His conclusion is still that the generic model is “crappy.”
The acquired customers were selected by mom-and-pop operators, not because they suited the buyer’s AI. Perhaps three or four of ten convert cleanly; the rest have mismatched needs, churn or require extended services. The buyer accumulates low-multiple revenue without building genuine new-business muscle.
Jason embraces Rory’s phrase—“they didn’t pick you”—as the decisive issue. Buying existing revenue and attaching a new technology to it predates AI; if customers have no durable attachment to the new product, the transaction remains financial engineering in most cases.
Harry’s counterexample reached $30 million of revenue from zero in two years by rolling up real-estate management businesses whose customers need essentially identical services. It raises margins from 5% to 40% in six weeks. Harry says acquisition price, payback speed and margin improvement are the key variables. Rory concedes that narrow, technology-aligned use cases can work; broader rollups are far less likely to convert uniformly.

14. Venture edge comes from maps, conviction and one real outlier

On uncrowded territory, Rory offers only a hedged answer: investors are abandoning consumer, while core “triple-triple, double-double” enterprise SaaS remains attractive. Jason sees less competition in gnarly large-enterprise problems and vertical SaaS, where domain expertise limits the field even as legal and sales tools attract hundreds of AI entrants.
Rory says investors at his stage must map markets where competitors have $1-3 million of revenue, even if timing prevents meeting every company. Losing to a known rival is normal; discovering 12 months later that the leader was a company you never heard of means the diligence process failed.
The outcome structure differs by layer: consumer and infrastructure skew closer to winner-take-all, while enterprise applications form oligopolies because submarkets have distinct workflows. HubSpot and Salesforce could both win CRM—and Salesforce even invested in HubSpot—because the apparent single market fragmented one level down.
Rory reports four or five realized investments above 10x; Jason says three, perhaps four. Rory rejects “only”: early in 1995, after his first five deals, he feared losing money on four and spent three years barely sleeping. He endured a later decade without a 10x and is prouder of 2-3x outcomes from 2004-06 than easy 10x gains into 2021. His recurring crises of confidence are functional—mistakes usually reveal a skipped analytical step.
Jason’s exception rule is categorical: “Any deal where you have 100% conviction you’ll 5X, I’ll just do it,” regardless of ownership or valuation. Rory agrees informed conviction deserves extraordinary weight, but only a leader-controlled shop can use that exception freely; a seven-partner peer firm risks turning one deviation into permission for everyone.
Jason says he has never reached out to a competitor for diligence. His own process substitutes extreme selectivity for exhaustive comparison: Talkdesk went from $1 million to $15 million in five quarters, Algolia grew about 20% monthly for two years, and Pipedrive was the fastest-growing company in its category. He requires top-0.1% growth, belief in the CEO and now an “S-tier CTO,” while relaxing the competition test.
The trade-off with growth-fund diligence is stark: Harry says this level of work is now an entry ticket for growth firms seeking a meeting; Jason saw Marratech’s process, which spoke with 100 customers, but says he himself sometimes calls only two, after the term sheet. He admits he never fully understands competitive positioning before investing because genuine market comprehension can take six months to a year.
Bay Area density fits that relationship-led style. Citing Henley & Partners’ count of 82 local tech billionaires, Jason says, “SF is so back”: a new-generation billionaire can DM him and meet downtown the next day. His automated intake now reviews decks, benchmarks TAM and growth, explains check and ownership ranges, and alerts him when metrics cross his threshold.
The portfolio objective is unapologetically power-law: “I’m only in it for one big massive win.” Everything serves the search for “one more Wiz”—not an $8 billion paper mark, but a real $8 billion outcome. Competition, missed emails and smaller wins are tolerable if repeated exposure to outlier founders eventually produces that single distribution.

No Priors Ep. 112 | With OpenAI Deep Research, Isa Fulford Bitcoin’s Bull Run & the AI Arms Race: What You Need to Know w/ Salim Ismail | EP #166 [REUPLOAD]