What this covers

This is a solo conversation with Bill Gurley about why private markets have seized up. He treats the problem through systems thinking: mega funds, closed exits and strained LPs interact, and the aggregate matters more than any one actor. The arithmetic anchors it. Time to liquidity has stretched from five to seven years to ten to fifteen, and because IRR compounds with time, a five-year delay on a $100 return needs roughly $250 once dilution and a risk-free base are counted. He argues the common retreat to DPI does not escape that problem.

The middle of the conversation is the most contested. Gurley suggests that everyone copying David Swensen's Yale model may have caused today's LP liquidity squeeze, and notes that Yale is now selling about $6 billion of private equity. He blames investment banks' IPO discount, about 25 to 26 percent plus a 7 percent fee, for the shut window, pointing out that the NASDAQ rose about 30 percent in 2024 without reopening it. He also sizes the roughly thousand zombie unicorns at about $300 billion, explains how liquidation preferences can swallow a company sold below its capital raised, and warns about opening private funds to 401(k)s.

The last stretch turns to AI. Gurley calls it a real platform shift that "might be bigger" than earlier ones. He argues that incumbents moved early this time, that falling model prices (about 1/100 in two generations) allow later optimization, and that open Chinese models could speed experimentation. He also gives founders advice on engineering network effects and on moving from pirate to navy.

Sharpest takeaway

Gurley argues that zero-rate-era overfunding, the Yale-model shift toward illiquid assets, and banks' IPO pricing have jointly stalled exits and distorted private marks, creating a self-reinforcing system with no corrective mechanism unless LP liquidity pressure forces change.

  • Stalled IPO and M&A exits plus LP liquidity needs mean private companies stay private longer, and founder and employee liquidity keeps them there.
  • Nobody in the chain has an incentive to mark assets accurately, so the reported value of roughly a thousand zombie unicorns is unreliable.
  • AI-driven overfunding forces every company into capital-intensive all-or-nothing competition, repeating the 2021 dynamics that produced the zombie unicorns.

The argument · threads15 threads · 104 claims
0.68

Abundant private capital lets companies stay private instead of going public.

6 pointscentrality 5/5
  • Branded VC firms have grown from roughly $500 million funds every three or four years to around $5 billion, about 10x, and now invest actively in late-stage deals, though Gurley argues 'late-stage' is often a euphemism for big check, as in $300 million into a 12-month-old AI company.

    “many of the branded firms I think have moved from maybe $500 million commitment every three or four years to $5 billion, so 10x”

  • There are roughly a thousand private companies that have raised over $1 billion each, around $200 to $300 million apiece, totaling about $300 billion of capital, and the group is widely called zombie unicorns, a label Gurley dislikes.

    “There's somewhere around a thousand, so these are a thousand private companies that have raised money over a billion dollars.”

  • A CEO was told by bankers to price an IPO at a certain level, but the founder could raise a billion dollars privately at 20 percent above that price, so private capital now beats public offerings and removes the reason to go public.

    “the founder said I can raise a billion tomorrow at 20% above that.”

  • The long zero-interest-rate period, lasting five to seven years and unprecedented in roughly 100 years, postponed any VC correction and created a flood of money and speculation that let companies raise sums never seen before.

    “Zero interest rates for, what was it, it ended up being five, six, seven years?”

  • Many zombie unicorns have enough capital to reach or approach break-even and could exist indefinitely without exiting, which is why the zombie label fits.

    “Many of them had enough capital to get towards break-even or near it.”

  • When a company generates excitement, investors knock with 100 to 300 million dollars, which forces founders into an all-or-nothing, swing-for-the-fences approach because a competitor raising that much can 10x its sales force and kill companies that build slowly.

    “it forces everyone to go all or nothing, swing for the fences.”

0.63

Excess capital drives startups toward overspending, all-or-nothing bets, and ZIRP-era zombie-unicorn problems.

10 pointscentrality 4/5
  • The 2022 to 2023 mini correction pushed companies toward break-even, cutting the many initiatives and overfunded sales forces that had produced unsustainable revenue, which in turn lowered their growth rates and explains the low growth of the zombie cohort.

    “when you cut it back and go towards break-even, your growth rate gets hit.”

  • When companies raise huge sums they stop focusing; instead of choosing one or two product initiatives, they do seven, which creates too many participants in a single market and makes market expansion harder because three to five survivors remain instead of one or two.

    “When you overfund, you do everything.”

  • In the early Internet era, all startups first built on Sun and Oracle, and about five or six years later none did, so the timing of that shift in buying behavior is a key signal to watch.

    “the first two years everyone built on Sun and Oracle. Everyone. All the startups did. And five, six years in, no one did.”

  • In an all-out capital-fueled war, companies must go for market share and unit economics cannot matter yet, so today's negative gross margins are a rational response, with a window to be settled later.

    “Until we get to a point where unit economics matter, and they can't matter in an all-out war market that the gavage tube funding creates. You have to go for market share.”

  • A PE firm that aggressively works through the roughly thousand zombie unicorns to extract value could find opportunity by looking at them optimistically instead of pessimistically.

    “I think there's some opportunity there to look at it optimistically instead of pessimistically.”

  • During downturns, pretenders and opportunists who over-promote, over-raise, and over-participate in secondaries leave the industry, as they did after the Internet bubble burst when people went back to consulting and banking.

    “the pretenders left town when the Internet bubble burst.”

  • If capital pushes the ecosystem toward grand slam home runs only, all the small and middle outcomes will disappear, which is bad for the ecosystem.

    “I think it's bad for the ecosystem that we are going to remove all the small and middle outcomes and just play grand slam home run ball all day long.”

  • Firms that compete for founders by writing large checks and acting as best friends do not take board responsibility to challenge decisions, and fewer investors call out bad unit economics, as happened with FTX, where no one took a board seat.

    “they differentiate themselves by being the best friend of the founder they possibly can be... They're never going to tell you no.”

  • The industry learned nothing from the ZIRP days: funding AI companies the same way those zombie unicorns were funded repeats all the same problems, including rushed overfunding and loss of focus.

    “And it feels like we didn't learn anything from the ZIRP days.”

  • Gurley personally lived through the Uber-Lyft capital battle, where competing for growth forced boards to debate going negative gross margin for years to take market share, a dynamic that will repeat in every category.

    “I lived it in the Uber-Lyft situation. But we're going to have that type of capital battle in every category under the sun.”

0.63

Slowed venture liquidity and stuck exits interact to squeeze LPs and endowments, so new capital cannot fix it.

10 pointscentrality 4/5
  • Harvard announced it is selling about $1 billion of secondary positions, adding to a pattern of large endowments seeking liquidity from private holdings.

    “Harvard announced they're in the market selling secondary for a billion.”

  • The time to liquidity for venture companies has moved from 5 to 7 years to 10 to 15 years, and the share of committed capital returned in the 5-to-10-year window fell from an average of about 20 percent, sometimes 30, to about 5 percent last year.

    “We're moving the time to liquidity of these companies from 5 to 7 years to 10 to 15 years.”

  • Many LPs have a new liquidity problem tied to the lack of IPOs and M&A; US colleges and universities issued $12 billion of debt in Q1 2025, the third-highest quarter ever, to fund capital commitments and payouts.

    “in the first quarter of 2025, US colleges and universities issued $12 billion of debt, which was the third highest quarter ever.”

  • When traditional LPs were tapped out, capital came from the Middle East, which funded the late-stage AI rounds; however, Gurley warns that if Middle Eastern sovereign funds turn skeptical about private equity valuations, that would have a big effect.

    “The Middle East is the area where most of that money was coming from.”

  • LPs should dabble in late-stage private markets as both buyers and sellers to understand how the motion works, while avoiding a run on the bank, and reevaluate whether the Yale model still works if everyone copies it.

    “you dabble in the late-stage private market on both sides, as a seller and a buyer, to see what the motion looks like”

  • Gurley believes the system promotes less liquidity, less high-quality company building, and higher burn, and that these components reinforce one another with no corrective mechanism unless something changes at the LP level.

    “all of the components that I listed, unless something happens at the LP level, I don't see a corrective mechanism.”

  • Gurley frames the market as a systems problem in which mega funds, zombie unicorns, closed exits, and LP liquidity stress interact, and the aggregate effect of these components matters more than any single actor's conduct.

    “a lot of the components of the industry are bouncing into one another, and it's the aggregate effect of all those things that's super interesting.”

  • Finding new capital sources for LPs only adds money at the top of a pipe whose outflow, exits, is stuck, similar to adding food to a digestive system with constipation, so more inputs do not solve the liquidity problem.

    “that's just putting more money in the top... Just eating more food doesn't help with the constipation problem.”

  • Gurley agrees the world is heavily private-market weighted, with 87 percent of firms with over $100 million in revenue now private by count according to Apollo, but argues that world is messier than the efficient public-market ideal.

    “of the firms that have more than $100 million of revenue, 87% now, just by count, are private.”

  • Talk of endowment taxes and cuts to NIH and NSF overhead may force universities to demand 5 or 6 percent annual payouts from endowments instead of 3 percent, pushing LPs into a more difficult liquidity position.

    “Even that is going to cause the universities to tell their endowment, instead of 3%, we need 5% or we need 6% a year.”

0.57

Abundant capital and the AI platform-shift boom inflate AI company economics and prevent the usual venture reset.

14 pointscentrality 3/5
  • Some AI revenue may be resale of compute, because wrapper companies on top of foundation models on top of hosting services may each book the same revenue, sometimes three or four times, often at negative gross margins.

    “some of the revenue growth is resale of compute... that revenue is being counted three or four times with negative gross margin.”

  • Leading AI companies spend $100 to 150 million a year in cash burn, and OpenAI is reportedly planning to spend about $7 billion in a year, a level radically different from traditional company building.

    “all the big AI companies are doing that. Maybe more. I think OpenAI said they're going to spend 7 billion in a year.”

  • Benchmark partners assess when AI companies move from experimentation or sandbox mode to optimization mode; with more capital, companies can stay in sandbox mode longer before optimizing, which makes the transition important to watch.

    “when companies move to optimization mode and how they make decisions differently once they do that than when they're in experimentation and sandbox mode.”

  • AI model pricing falls so quickly that a model two generations old sells for about 1/100 the price per token of today's model, so companies can expect to optimize prices later.

    “any AI model that's two generations old, sells for 1/100 the price per token as today's.”

  • Capital-intensive hard tech such as solar has not produced great venture returns, and the main exceptions are companies Elon Musk touches like SpaceX and Tesla, which Gurley treats as outliers tied to Musk's execution speed.

    “There's one exception to this whole rule and it's anything Elon Musk touches.”

  • Alibaba's Qwen, Xiaomi's MiMo, and Baidu's model going open source in June could give four deep-pocketed open models that train each other and generate far more experimentation than the US market.

    “if you have four open ones that can all train on each other and everybody can get a hold of that, I think that's going to lead to a massive amount of optionality and experimentation”

  • Waves of abundant capital lead to more investment in less capital-efficient businesses, and if capital tightened, appetite for such capital-heavy companies might disappear.

    “we've seen waves of capital availability lead to more interest in businesses that are less capital-efficient. There's a correlation there.”

  • LLMs were built around language and are not great with numbers, so generalized AI will not simply replace all computation; systems will need hybrid fixes such as models writing Python for math.

    “they were created around language, they're not great with numbers... The way when you ask an AI math now, it goes off and writes Python”

  • Gurley told Warren Buffett that a discounted cash flow model does not work when interest rates are zero and that it only creates speculation, and Buffett agreed.

    “I said, "You know, your DCF doesn't work if the interest rates are zero."”

  • Gurley agrees AI is a legitimate platform shift comparable to mobile, the Internet, or the PC, and says it is 'certainly one of those and might be bigger,' so it doesn't need to beat those earlier shifts to justify the capital flowing in.

    “if it's a platform shift, as were mobile or the Internet or the PC, that's big enough, it doesn't have to be better than those.”

  • AI companies are priced at roughly 10 to 20 times revenue compared with a normal company, and in some cases more.

    “10X, 20X a normal company. Is that fair?”

  • AI arrived just as a mini correction was building, preventing the full reset that historically cleaned up venture excess, because excitement about AI revived capital flows.

    “The AI wave came at a really interesting time... We were headed to this mini correction.”

  • When money tightens, pretenders and opportunists leave the market, and opportunists who over-raise and over-participate in secondaries can cause crashes.

    “the pretenders left town when the Internet bubble burst.”

  • Everyone now knows about compounding, network effects, and boom-bust cycles, so investors rely on the expectation of a quick rebound, which makes them more willing to overcommit to AI.

    “everyone has studied history. Everyone knows about compounding effects, everyone knows about network effects, everyone's studied cycles, they've seen boom and bust.”

0.54

Bank IPO discounts and blocked acquisitions leave companies without a viable exit path.

6 pointscentrality 4/5
  • The number of total public companies in the US is far below its peak, and Gurley attributes much of the decline to the IPO process and banks' pricing practices.

    “the number of total public companies in the US is way down from peak.”

  • In 2024 the NASDAQ rose about 30 percent yet the IPO window stayed closed, which Gurley says has never happened in his experience and makes no sense unless something else, such as banks' IPO discount, is at work.

    “the NASDAQ was up 30% and the window was closed.”

  • Gurley believes the IPO discount that investment banks impose, especially for well-known brand-name banks, is a major cause of the stalled IPO market, with underpricing of 25 to 26 percent, and with the 7 percent fee the total cost of capital reaches about 33 percent.

    “They're up 25, 26% underpricing. You add in the 7% fee and you're at a 33% cost of capital.”

  • The Mag 7 companies hold a huge amount of cash that could drive large M&A, but they buy back stock instead because regulators in Washington and the EU make acquisitions too uncertain to close.

    “I think they buy back stock because they can't.”

  • Deals announced with expected closings more than a year out, such as Wiz, are very hard for boards and management teams to accept because low certainty of close discourages buyers and sellers from signing.

    “People don't want to enter into an M&A agreement with a low certainty of close.”

  • High private valuations, such as a roughly $15 billion AI round, make acquisitions by companies like Apple, for example Perplexity, hard to close, so overpricing of the most exciting companies also suppresses M&A.

    “It makes it hard for things to close when the prices are that high.”

0.53

Liquidity delays and long time to exit erode returns and compound the cost of capital.

12 pointscentrality 3/5
  • Yale, the institution that did the most to define the endowment illiquid-assets strategy, is now selling about $6 billion of private equity, which Gurley finds historically significant as a signal that the model's architect is trying to get out.

    “Yale has announced that they're in the market looking to sell $6 billion of private equity.”

  • LPs have increased venture exposure from 5 to 7 percent of their portfolios to 10 to 15 percent, and for some it reaches half their private equity commitment, so the zombie unicorn asset class now sits on LP balance sheets in a material way.

    “They've slowly increased their participation in venture from anywhere from 5 to 7% up to 10 to 15%.”

  • Delaying a $100 return from year 10 to year 15 at a 10 percent compounding requirement means it must be worth $160 by year 15, and layering in 5 percent annual equity dilution and a 5 percent risk-free base pushes the effective hurdle to about 20 percent, so a five-year delay requires about $250 to meet the same return expectation.

    “guess how much money you need instead of $100? $250 for a five-year delay”

  • People argue that DPI rather than IRR matters, but if time doubles then IRR falls, so time-to-liquidity matters for IRR too and the DPI argument does not escape the time problem.

    “everyone loved to say, oh, it's not IRR, it's DPI, but if time doubles, it is IRR.”

  • Yale achieved about 13 percent compounding over 35 years under David Swensen with the illiquid-heavy model, but Gurley argues the model's success depended on few adopters and that its benefits erode as the crowd joins.

    “Yale, I think they say, had a 13% compounding return over 35 years under David Swensen.”

  • Everyone copying the Yale model, which shifted endowments toward illiquid assets, may be the cause of today's LP liquidity problem, and the model worked when Yale was alone but may not work when everyone adopts it.

    “What if everyone copies David Swensen? What if everyone goes to 50% illiquid? Will it still work?”

  • People argue that a fund's return excluding its big winner is poor, but Gurley says the opposite question is unasked: what if you keep the big winner and remove everything else, since that is where the industry is headed.

    “But I haven't asked anyone the question, well, what if you keep the big winner but get rid of everything else? Because that feels like where we're headed.”

  • Time to liquidity has moved from 5 to 7 years to 10 to 15 years, and because IRR compounds with time, doubling the holding period doubles the effective cost of capital, so time is a massive problem for GPs and LPs.

    “We're moving the time to liquidity of these companies from 5 to 7 years to 10 to 15 years.”

  • Physical-world hard tech such as mining, fusion, and fission requires very large capital to reach revenue and long timelines, and Gurley did little of this investing because the historical return record is weak.

    “They require tons and tons of capital to get to revenue and long periods of time.”

  • Many zombie unicorn valuations were last set in 2021 at a market peak, so their true current worth is uncertain and probably below the marks.

    “A lot of the marks for their last round was set back in 2021.”

  • Amazon went public below a billion dollars and later traded above a trillion, so public investors captured years of compounding that delayed IPOs now withhold.

    “So we all know Amazon went public at less than a billion dollars and then traded over a trillion.”

  • David Swensen's Yale model put far more money into illiquid assets than liquid ones, a strategy no one originally adopted because of opacity and illiquidity, but which produced about 13 percent compounded returns over 35 years.

    “Yale, I think they say, had a 13% compounding return over 35 years under David Swensen.”

0.38

Late-stage rounds and secondaries let insiders capture returns public investors would otherwise earn.

5 pointscentrality 3/5
  • Zombie unicorns dilute existing shareholders by about 3 to 6 percent a year through equity issuance to employees, which adds to the cost of capital investors must earn.

    “every one of these zombie unicorns is diluting 3, 4, 5, 6% a year on equity issuance to the employee base.”

  • Late-stage investors take the high-compounding years that public market investors would otherwise capture, and by delaying IPOs they then sell the story to LPs that exposure to high-tech growth requires investing in their funds.

    “Here's a super important piece that goes on top of that. They then turn around and go to the LP community and say companies are no longer going public”

  • Late-stage investors such as Thrive now approach companies already set to go public and offer founders and employees attractive liquidity, which encourages companies to stay private; Databricks is a recent example.

    “you present them with an offer that... founder liquidity encourages employee liquidity, it might encourage angel liquidity, and you basically encourage the company to stay private.”

  • Stripe shareholders have traded in and out of the stock without going public, with the company comfortable with that, which Gurley calls very new and unique, functioning like the old pink sheets where trades happen by appointment.

    “They've traded in and out of Stripe and the company's somewhat comfortable with it.”

  • In a traditional IPO, large investors who oversubscribe 100 times may get only 1 or 2 percent of the offering, while in pre-IPO late-stage rounds they can win around 30 percent, so the investors gain a bigger ownership stake than they could in a public offering.

    “They oversubscribe by 100X hoping that they may get 1 or 2% of the offering. They're not going to get 30%.”

0.38

Staying private deprives companies of public-market feedback and the leadership discipline it imposes.

5 pointscentrality 3/5
  • Anduril being cleared by the DOD and selling to the military is a new and impressive data point that venture-backed startups can break through regulated, capital-heavy industries historically blocked by regulation.

    “Anduril being cleared by the DOD and actively selling into our military is certainly a new data point”

  • Silicon Valley was absent from Washington when Gurley gave a speech on regulatory capture a few years ago, but now the tech industry is heavily present in Washington, making regulatory access part of the investment landscape.

    “no one from Silicon Valley was in Washington at the time. And now they're all there”

  • Because investors have studied compounding, network effects, and cycles, a conviction in AI would likely lead to a rapid rebound even after a six-month period of doubt, as happened after the COVID correction.

    “I suspect that the conviction in AI is high enough that even if we were to have a six month period where people thought AI was overstated, it would start rebounding very quickly.”

  • The best CEOs, such as Barton and Benioff, and even Mark Zuckerberg at Meta, believe being public makes them run sharper, and staying private forever deprives companies of that feedback.

    “They believe being public makes them run sharper. And one other negative of these companies staying private forever is they're not getting that feedback.”

  • Sundar Pichai admitted he had not read The Innovator's Dilemma, illustrating that when a company is crushing it, the lessons about disruption feel like they belong to someone else.

    “he asked him if he had ever read The Innovator's Dilemma and he admitted not.”

0.36

Misaligned incentives keep private valuations inflated and hard to reset.

4 pointscentrality 3/5
  • Founders multiply their ownership percentage by the highest valuation their company ever reached and treat that as their net worth, so cutting that figure by 70 percent after a down round feels psychologically very hard.

    “every founder I know has multiplied their percentage ownership times whatever the highest price their company was ever valued at and thought about that number as their net worth.”

  • No one in the private investment chain has an incentive to get marks right: GPs set the price LPs see, endowment VC managers are bonused on paper marks, so the incentives run toward inflating rather than right-sizing valuations.

    “no one has an incentive to get the marks right.”

  • The world now believes AI is the biggest platform shift in anyone's lifetime, which leads investors to assume that today's richly valued companies could become giants like Google or Meta and therefore to refuse to be the one who underpays.

    “the world believes AI is the biggest platform shift in anyone's lifetime.”

  • Investors can take their liquidation preference equal to the total capital raised instead of converting to common; if a company raised $300 million and sells for $400 million, the preference could take 75 percent of the company, so it matters greatly when valuations fall.

    “If the valuation is now $400 million, then the LiqPref could take 75% of the company in a sale.”

0.33

Founder leadership and choices, not capital, drive outcomes, including engineering network effects as companies scale.

9 pointscentrality 2/5
  • Founders should design systems so the 2,000th customer gets a better experience than the 1,000th, using data exhaust and shared learning, because network effects can be made more real if deliberately engineered.

    “that 2,000th customer should have a way better experience than the thousandth. And how do you design that into your system?”

  • Founders are not born capable of leading a thousand-person organization; the capability must be learned and the founder must want it, which is why Ben Horowitz's 'we only back founders that go all the way' advice carries a condition about wanting to learn leadership.

    “there's nothing about a founder in the way they're born that makes them capable of leading a thousand-person organization.”

  • In a vertical AI company, one customer's learning can become learning for the whole group, and human-in-the-loop failure points improve the model, so an early leader can build an even bigger lead over time.

    “If the learning of an individual customer becomes a learning for the whole group and everyone benefits, that's pretty powerful.”

  • Reid Hoffman's metaphor says startups are born as pirates but must eventually become a navy; founders must accept operating discipline at scale, which many treat as red tape, once they pass roughly 100 million in revenue and approach a billion.

    “all startups are born as pirates, but they eventually have to become a navy”

  • For a new investment firm, the most valuable brand is one that helps founders make better decisions and calls bullshit when needed, not one that competes by being a founder's best friend.

    “It's helpful to have someone along who will call bullshit when it needs to be done, who will push on unit economics.”

  • Gurley says the early-stage window remains the time when investors can make the biggest bet and get the biggest outcome, which is why he still loves that period despite his career in it.

    “I think it's the time window where you can make the biggest bet and have the biggest outcome.”

  • Founders should consider taking secondary liquidity when investors pay 30 times revenue, because the aggressive-spend game requires burning hundreds of millions, although Gurley thinks founder liquidity is bad for long-term company success.

    “If someone's going to pay 30x revenue and force you to play a game that you're not comfortable playing by burning hundreds of millions a year, you should probably take a little off the table.”

  • Without founders doing their work, none of the capital, technology, and market dynamics matter, so founder decisions drive all ecosystem outcomes.

    “without them doing their thing, basically none of this matters at all.”

  • Gurley found that he was calmer, happier, and more productive during resets than during manias, and that conversations about traditional company building were more efficient and authentic in downturns.

    “I found as an active GP, I was much calmer and happier, and found my job more fulfilling and more efficient and productive in the resets, than in the manias.”

0.27

Industry actors act reasonably, yet their aggregate effects may harm the world.

6 pointscentrality 2/5
  • Gurley attributes no malintent to industry participants, who he believes act reasonably in their own interests, but says the aggregate effect of their actions may not be positive for the world.

    “The aggregate effect may not be positive for the world, but I'm not ascribing malintent or anything to anybody”

  • Investors' loyalties follow their portfolio companies, so VCs who back defense or China-hawk companies may become warmongers, and Gurley admits that Benchmark-backed companies always draw his loyalty.

    “your allegiances are going to go where your investors are. And I still feel that for any company that's got Benchmark attached to it.”

  • The next generation of GPs will face board meetings where competitors raise another billion, forcing them to decide whether to run negative gross margins for two more years to take market share, a decision not found in any Harvard case study.

    “Because you go into a board meeting and the other company raises another billion, the thinking that you're forced to do at the table was, should we go negative gross margin for two more years and take market share?”

  • Investors force-feeding capital to any promising company, like a gavage tube for geese, pushes founders into all-or-nothing growth and makes founder liquidity look attractive, even though it hurts long-term success.

    “the minute there's a company that has any amount of excitement about it whatsoever, someone's knocking on the door, trying to give them 100, 200, 300 million dollars.”

  • Silicon Valley is now heavily engaged in Washington, and regulated, capital-heavy sectors like defense and mining have become more attractive to venture investors partly because regulation can keep China out.

    “mining gets better if we're not allowed to use China. And so I hate that part of this world.”

  • Founders can run burn rates now, but unit economics must eventually matter, and two-generation-old models costing 1/100th as much per token mean companies can plan a future cost reduction.

    “unit economics will matter one day... the two-generation-old models cost 1/100th the tokens. You can plan that we're going to move to that.”

0.24

Private-market returns and public capital-market structures face hard, unproven constraints.

5 pointscentrality 2/5
  • Investing in run-of-the-mill private companies away from AI, which capital markets evaluate in strict calculator terms, could be a contrarian place to find value, with some late-stage investors already exploring buying traditional companies and adding AI themselves.

    “what if I can find a traditional company that may not understand that AI would enhance it, but where we can go do that ourselves?”

  • Opening private funds to wealthy individuals and retirement accounts, such as 401(k)s, could bring more capital, but Gurley warns it moves a high-fee, low-transparency structure to ordinary investors, with more fraud and higher transaction costs.

    “There will be more fraud, there will be higher transaction costs. It would be the nature of the beast.”

  • Josh Kopelman's simple GP math on the prices being paid and capital being deployed shows that the returns needed to match historical VC performance are hard to achieve.

    “There's a great video you may have seen where Josh Kopelman just walks through some simple GP math”

  • Gurley is less convinced than many people that the US has the best functioning capital markets in the world, despite the common view that they are the envy of the globe.

    “I'm less convinced personally.”

  • Hester Peirce's paper argues that blockchain could fix the IPO market by tokenizing securities, and Gurley thinks that shifting capital raising to tokenized trading would immediately change how IPOs work, though he is still skeptical.

    “argues that maybe blockchain is the path to fix the IPO market. Which is provocative.”

0.22

Incumbent AI responses and language-heavy verticals shape where AI investment pays off.

4 pointscentrality 2/5
  • AI is exceptionally good at language, and coding is an even tighter form of language, so vertical applications in legal, customer support, and similar domains that depend on language are the most promising places for AI investment.

    “It's exceptionally great in language. Coding is a tighter form of language, so it's even better at coding”

  • Howard Marks's non-consensus, accurate investing framework is hard to apply to AI because platform shifts have become consensus, and investing in AI is now the consensus, so one cannot easily be contrarian and accurate at the same time.

    “Because these platform shifts have now become consensus, you'd have to not invest in AI, which sounds outlandish. So I don't know that you can apply those two things simultaneously.”

  • Unlike mobile and the Internet, where large incumbents were slow and gave startups an opening, large companies have paid attention early to AI, as shown by ServiceNow's website, Microsoft's transcript mentioning AI 67 times, and Satya Nadella's two hours on AI.

    “Microsoft's earnings transcript had 67 occurrences of AI, and Satya just talked for two hours about AI.”

  • Google should have dominated AI use cases, yet few people use Gemini for code generation or daily LLM work, and they turn to startups such as Cursor, Anthropic, and OpenAI, showing incumbents can still re-demonstrate innovator's-dilemma patterns.

    “Google should have been in the very best position to dominate every AI use case, and yet basically no one I know is using Gemini”

0.20

Pulling retail and foreign money into venture reduces transparency and invites fraud.

3 pointscentrality 2/5
  • Coatue reportedly cut its minimum commitment from $5 million to about $25,000 and will place investments through an investment bank, tapping a pool of wealthy individuals and professionals that were previously shut out of late-stage managers.

    “They used to have a $5 million minimum for a commitment. They're taking it down to 25,000 or something”

  • When traditional LPs became tapped out, capital was found elsewhere, mainly from the Middle East, and fundraisers in the region supplied the money that kept the boom funded.

    “The Middle East is the area where most of that money was coming from.”

  • If the answer to getting everyday consumers into high-growth tech is funneling endowments, 401(k)s, and IRAs into venture funds charging 2 and 20, the result will be less transparency, higher transaction costs, and more fraud, which Peter Lynch would never have wanted.

    “There will be more fraud, there will be higher transaction costs. It would be the nature of the beast.”

0.11

Consumer AI awaits better voice and memory before learning effects create lasting advantages.

5 pointscentrality 1/5
  • Early LLMs were not ideal for consumer products because voice and memory were weak; as they improve, switching costs and learning effects can emerge, as Character.ai lacked because it did not really get to know users.

    “two elements of the first wave of LLMs that I think make them not perfect for consumer. One is voice being really good... and the second one being memory”

  • Gurley does 40 to 50 AI searches a day, far more than his Google searches, and believes self-learners will be able to accomplish things and move up quickly because of very fast learning.

    “I'm probably doing 40 or 50 searches a day on AI platforms, which is more than I ever did Google searches.”

  • Consumer AI was held back in the first LLM wave by weak voice and memory, and as these improve, network effects, learning, and switching costs will emerge, making four or five consumer companies likely within a year.

    “two elements of the first wave of LLMs that I think make them not perfect for consumer. One is voice being really good... and the second one being memory”

  • The pace of AI change is the fastest Gurley has seen in his career, so missing a week of news makes the world look different a week later.

    “The pace of change is the fastest I've ever seen in my entire career.”

  • Gurley does not believe all disease will be eliminated within ten years despite AI founders claiming so, though he is excited about healthcare gains from AI.

    “I don't think all disease will be gone in 10 years like the AI founders are saying.”