Bill Gurley - The Gift and The Curse of Staying Private - [Invest Like the Best, EP.427]
Summary
Gurley sees venture trapped in a self-reinforcing system: branded funds grew roughly 10X, while roughly 1,000 private companies have raised more than $1 billion and remain difficult to value. Those companies collectively raised roughly $300 billion, while the NVCA estimates $3 trillion of venture assets sit on LP books. GPs report the prices, large-endowment venture managers may be bonused on paper marks, and founders protect perceived net worth, creating little incentive to reset marks. Gurley has not statistically surveyed how many of these companies are profitable versus doomed.
Staying private now lets elite companies and late-stage investors capture growth that historically belonged to public shareholders. Despite the Nasdaq rising 30% in 2024, the IPO window remained effectively closed; Gurley says 25%-26% underpricing plus a 7% fee implies roughly a 33% cost of capital. Private rounds give large investors meaningful allocations, founder and employee liquidity, and an opportunity to “hoard the public IPO growth years.”
Longer holding periods are mathematically destructive even when the underlying company keeps appreciating. Venture liquidity has stretched from five-to-seven years toward 10-to-15, while mature private companies may dilute shareholders 3%-6% annually through employee equity. A projected $100 return in year 10 must become $160 by year 15 at a 10% hurdle—or roughly $250 when Gurley combines a 15% return expectation with 5% dilution: “If time doubles, it is IRR.”
LP liquidity is the likeliest pressure point, but opening additional capital taps does not fix a blocked exit pipe. U.S. colleges and universities issued $12 billion of debt in Q1 2025, Harvard announced a secondary sale of roughly $1 billion, and Yale looked to sell about $6 billion of private equity. Expanding access through smaller fund minimums, Middle Eastern capital, or 401(k)s is, in Gurley’s metaphor, “just eating more food” without solving the constipation.
AI is a genuine platform shift, but some reported revenue may be subsidized compute counted several times. Wrappers can resell foundation-model capacity hosted by another provider, and some may operate at negative gross margins, making revenue quality hard to assess during an all-out market-share war. The offsetting bull case is rapid cost compression: a model two generations old can sell for “one one-hundredth the price per token,” leaving room to optimize later.
Founders cannot simply reject abundant capital when a funded rival can expand its sales force 10X or 50X. Gurley calls the resulting force-feeding a “gavage tube”: promising companies are pushed toward $100 million-to-$300 million rounds, extreme burn, and grand-slam-or-bust competition. If investors demand that game at 30X revenue, his reluctant advice is to take some founder liquidity—even though the system removes smaller outcomes and repeats the mistakes that created the zombie-unicorn backlog.
A reset would be painful, but Gurley thinks it could restore authentic company building and drive out opportunistic capital. For durable founders, the eventual requirements remain unit economics, learned leadership, and the transition from Reid Hoffman’s “pirate” to a “navy.” AI moats should make every additional customer improve the product for the whole network; consumer AI may reopen as voice and memory mature, with Gurley saying he would be shocked not to see four or five companies emerge within a year.
Deep dive
1. Mega-funds broke venture’s old corrective cycle
Gurley’s governing frame is systemic, not accusatory: individual firms, founders, and LPs can act rationally while producing an aggregate result that “may not be positive for the world.” Venture’s historically cyclical system has been “upset or kicked over,” so incentives matter more than assigning malintent.
Branded firms that once raised perhaps $500 million every three or four years now manage commitments nearer $5 billion, a roughly 10X increase. They also write “late-stage” checks—though Gurley calls late stage “a euphemism for big check,” since investors will put $300 million into a 12-month-old AI company.
Roughly 1,000 private companies have raised more than $1 billion. Gurley uses pre-LLM versus post-LLM as a dividing line when discussing the market; ChatGPT gave him 1,250 companies, while the NVCA counted 900. At perhaps $200 million-$300 million raised apiece, that is roughly $300 billion of capital; the NVCA estimates about $3 trillion of venture assets on LP books.
2. ZIRP produced well-funded survivors with stale economics
Gurley traces the backlog to an unprecedented five-to-seven-year zero-rate period. His shorthand from asking Warren Buffett what happens to discounted cash flow at zero rates: “All it does is create a lot of speculation.” Buffett’s answer was simply, “You betcha.”
Overfunding kept three-to-five competitors alive where earlier scarcity might have left one or two, making expansion harder for everyone. It also encouraged companies to pursue seven product initiatives instead of one or two and to build sales capacity that generated revenue without necessarily producing sustainable economics.
During the 2022-23 mini-correction, companies cut those seven initiatives toward two and moved near break-even. Gurley has not done a statistically significant survey of the cohort, but thinks many had enough capital to reach break-even or near it and could perhaps exist indefinitely; their slower growth leaves their 2021 marks hard to validate, helping produce the “zombie unicorn” label.
Patrick challenges whether founders should force honest marks. Gurley answers that founders naturally multiply ownership by the highest valuation ever achieved, while liquidation preferences make resets dangerous: if a company raised $300 million but is now worth $400 million, investors could claim 75% of a sale.
3. Broken exits let private investors capture public compounding
The capital-market anomaly is stark: the Nasdaq rose 30% in 2024, yet the IPO window was considered closed. Gurley had never previously seen a strong Nasdaq coexist with virtually no IPOs, suggesting the traditional correlation has broken.
Jay Ritter’s rerun of the data showed 25%-26% IPO underpricing; add a 7% banking fee and Gurley calculates roughly a 33% cost of capital. One founder was told to accept a public price even though the company could privately raise $1 billion at a valuation 20% higher.
M&A offers little relief. Although many blamed Lina Khan, she is gone and the first five months of 2025 did not produce record M&A. Gurley points to Magnificent 7-related regulatory constraints, the EU’s posture, low certainty of close, and the Wiz transaction’s announced closing period of over a year; lofty private pricing also makes acquisitions difficult, as with a hypothetical Apple purchase of Perplexity after a roughly $15 billion round.
Thrive and others now approach IPO-ready companies with private rounds containing founder, employee, angel, and investor liquidity. An investor in an IPO might receive only 1%-2% of the offering, but privately could take 30% of a deal—an oligopolistic chance to “hoard the public IPO growth years” and then tell LPs that access requires investing through them.
4. Private markets became liquid by appointment, not by design
Patrick describes Stripe’s market as “a by-appointment public market”; Gurley compares it to old pink-sheet trading. Selected LPs can move in and out through company-tolerated transactions, letting a great company obtain capital and shareholder liquidity without public disclosure, daily trading, or equivalent scrutiny.
That model works for perhaps five exceptional companies, Gurley cautions, not 1,500. A broader shift toward private funds charging two-and-20 would bring more obfuscation, less transparency, greater fraud risk, and higher transaction costs than healthy public markets.
Patrick cites Apollo’s finding that, by count, 87% of companies with more than $100 million of revenue are now private. Gurley sees the same direction in likely Coatue reducing a reported $5 million minimum toward $25,000 and in private-equity firms seeking permission for 401(k)s to invest in private assets.
More money entering the pipe cannot repair a blocked outlet. Gurley’s deliberately earthy analogy: if exits are constipated, raising from doctors, dentists, retirement accounts, or new geographies is “just eating more food” without addressing the underlying problem.
5. The copied Yale model is meeting its liquidity limit
Venture allocations at some LPs rose from roughly 5%-7% to 10%-15%, sometimes representing half their private-equity exposure. With IPOs and M&A stalled, U.S. colleges and universities issued $12 billion of debt in Q1 2025—the third-highest quarter recorded—to help fund capital commitments amid payout pressure.
Harvard entered the secondary market for roughly $1 billion, while Yale looked to sell about $6 billion of private equity. Yale is uniquely symbolic: David Swensen compounded at a stated 13% over 35 years by accepting illiquidity when others would not.
Gurley’s question is whether an advantage survives consensus adoption: “What if everyone copies David Swensen? What if everyone goes to 50% illiquid?” Yale moving toward the exit may reveal the consequence of every endowment adopting a strategy that worked precisely because it was once non-consensus.
Taxes on endowments and cuts to research funding could push university payout demands from 3% toward 5% or 6%. If many institutions then sell simultaneously, lower secondary prices could feed back into marks across the system; a broader change of mind among Middle Eastern capital providers could amplify that pressure.
Gurley’s tentative LP response is to dabble in late-stage secondaries as both buyer and seller, reconsider whether the Yale model still works when consensus, and look for a PE firm willing to extract value from the zombie cohort.
6. Abundant capital force-feeds founders into winner-take-all wars
Gurley’s metaphor is the “gavage tube” used to force-feed geese for foie gras. The moment a startup shows excitement, investors arrive offering $100 million, $200 million, or $300 million, forcing the company toward “all or nothing, swing for the fences.”
A founder might prefer disciplined building, but if a competitor raises $300 million and expands its sales force 10X or 50X, abstaining can mean extinction. OpenAI’s stated plan to reach roughly $7 billion in a year illustrates how far this environment sits from “your grandfather’s startup business.”
Gurley lived the dynamic through Uber and Lyft: after a rival raises another billion, boards confront choices such as sustaining negative gross margins for two more years to capture share. Those decisions are “super high-stakes poker,” not strategies found in Good to Great, Harvard cases, or Buffett letters.
His uncomfortable founder advice follows the incentives: if investors pay 30X revenue and require hundreds of millions in burn, “you should probably take a little off the table.” He still views founder liquidity as harmful to long-term outcomes and laments a market eliminating small and medium wins in favor of grand slams.
7. AI is real even if its revenue is temporarily distorted
Gurley refuses the premise that AI enthusiasm is baseless: it is at least a platform shift comparable to the PC, internet, or mobile, and “might be bigger.” He has no doubt that AI work such as Bret Taylor’s at Sierra is real and will materially change the companies it touches.
Personally, he performs 40-50 AI searches daily—more than he ever searched Google—mostly for rapid learning. For inherently self-directed learners, the increase in what they can accomplish and “the speed at which they’ll be able to…move up the ladder is breathtaking.”
Yet some revenue may be resale of compute: a wrapper sits atop a foundation model hosted by another provider, potentially sells below cost, and causes the same revenue to be counted “three or four times with negative gross margin.” All-out funding wars delay the moment when those unit economics must clear.
Cost curves complicate the critique. A model two generations old can cost one one-hundredth as much per token, making later optimization plausible; Gurley compares it with internet startups initially standardizing on Sun and Oracle before abandoning them five or six years later. LLMs remain strongest in language and coding, while numerical work often requires calling Python.
8. Time and dilution quietly consume venture returns
Time to liquidity has moved from roughly five-to-seven years toward 10-to-15. In the five-to-10-year fund window, the percentage of committed capital paid back historically averaged about 20%, sometimes reached 30%, but fell to 5% last year and remains around 5%-7%.
Mature private companies also issue perhaps 3%-6% of equity annually to employees. If $100 expected in year 10 arrives in year 15, it must become $160 at a 10% hurdle; using a 15% venture return expectation plus 5% dilution, Gurley says it must become approximately $250.
Hence his rebuttal to “it’s DPI, not IRR”: “If time doubles, it is IRR.” Delaying exits also exposes companies to entropy; Gurley wonders whether venture is heading toward keeping the single winner while losing everything else rather than merely measuring funds without their winner.
Private duration can also weaken governance. Investors winning allocations by becoming a founder’s best friend may never say no or challenge unit economics; Gurley’s extreme example is FTX, where investors took no board seat. Leading CEOs have told him public-market feedback makes them operate more sharply.
9. AI competition rewards openness, adaptability, and selective skepticism
China’s most fascinating response to DeepSeek, in Gurley’s view, may be internal competition: Alibaba made Qwen open source, Xiaomi released an open model he tentatively recalls as “MiMo,” and Baidu’s Robin Li said its proprietary model would open in June. Four deep-pocketed open models able to train from one another could create “massive optionality and experimentation.”
Incumbents are not uniformly asleep. ServiceNow’s messaging “drips of AI,” Microsoft mentioned AI 67 times in an earnings transcript, and Satya Nadella spoke about it for two hours. Patrick counters that users he knows choose Cursor, Anthropic, and OpenAI over Google; Gurley concedes evidence exists on both sides, with Apple another data point in the debate.
Hard technology remains a tougher return category: substantial solar venture investment 15-20 years ago did not work, while Tesla and SpaceX are outliers attached to Elon Musk. Anduril’s ability to clear Department of Defense barriers is impressive, but not a reason every VC should jump into capital-intensive hard tech instead of capital-efficient, high-margin software.
If Gurley were still investing actively, he would search for verticals where AI’s language strength creates exceptional fit—coding, legal work, customer support, and still-unexplored categories. Another potentially non-consensus strategy is upgrading traditional companies with AI rather than paying perfection for obvious AI-native assets.
10. A reset could restore discipline, but durable founders must prepare now
Asked what may happen over the next five years, Gurley’s instinct is blunt: “My gut is we have a problem.” The current system reinforces less liquidity, weaker traditional company building, and higher burn; without pressure at the LP level, he sees no internal corrective mechanism capable of stopping the loop.
Resets are awful for many participants, but Gurley found them calmer, more productive, and more authentic than manias. When easy money disappeared, “the pretenders left town”; even so, conviction in AI is strong enough that a six-month period of skepticism might produce a rapid rebound.
For founders, temporary burn can be acceptable, but “unit economics will matter one day.” Scaling past $100 million toward $1 billion in revenue requires Reid Hoffman’s transition from pirate to navy, plus a willingness to learn leadership—the ability to run 1,000 people “doesn’t come natively” and is not “for free.”
Gurley’s moat test is whether customer 2,000 receives a materially better product than customer 1,000 because learning spreads across the network. Better voice and memory could likewise reopen consumer AI: he invokes Her and says he would be shocked not to see four or five consumer companies emerge within a year.