5Y View | Bill Gurley: How the AI Wave Is Reshaping the Structure of Primary Markets

Recommended by

Liu Kai, 5Y Capital

Article reprinted from: Investment Notebook Pro (ID: touzizuoyeben)

Recently, the Invest Like the Best podcast invited renowned venture capital veteran Bill Gurley to discuss current problems in the U.S. primary market, as well as AI company valuations and where the opportunities lie. In the interview, Bill analyzed the current state of the venture capital industry, noting that mega venture funds continue to rise, with many increasing their investment scale by 10x, which has spawned some "zombie unicorns" — massive in scale but questionable in real value. Currently, neither GPs nor LPs have sufficient incentive to correct valuations.

Bill pointed out that the zero-interest-rate environment led to capital abundance, and companies that could easily raise money had no incentive to go public, causing the U.S. IPO and M&A markets to stagnate in recent years. The liquidity problem for LPs is also worth noting — you can make huge money doing non-consensus but accurate things, but when everyone puts 50% of their assets into illiquid investments, problems arise.

On AI development, Bill believes that the arrival of the AI wave interrupted a market correction that should have happened. AI is seen as a historic platform shift, driving a new round of investment fever and valuation bubbles. But he noted that most AI company revenue currently comes from compute resale, and economic efficiency is ultimately a problem that needs to be solved.

He also mentioned that China's intensely competitive environment can actually shape stronger companies, and if Chinese tech giants open-source their AI models, that would be incredibly powerful. Open models that can train on each other, that everyone can use — I think that would lead to massive choice and experimentation, and that's not going to happen in the U.S.

Here is the interview transcript:

Patrick: Our guest today is Bill Gurley. Bill was a general partner at Benchmark Capital. This is his sixth appearance on Invest Like the Best, and it's his most comprehensive market analysis yet, exploring the realities that are reshaping the venture capital industry.

Bill confronts the uncomfortable math behind venture capital returns today, especially the phenomenon of companies staying private longer. He also explains why no one — from GP to LP to founder — has enough incentive to mark assets accurately, creating a coordination problem across the entire system.

We also dive deep into AI's investment implications as a platform shift, from assessing the quality of AI revenue to international competitive dynamics. Bill provides critical perspective on how to navigate the present and future. Please enjoy my conversation with Bill Gurley.

So, Bill, this marks your reclamation of the "most frequent guest" crown, defeating our good friend Michael Mauboussin. Welcome back.

Bill: I can't think of anyone on earth I'd rather be neck-and-neck with than you.

Patrick: Interestingly, this is the first time you and I have done a solo show since 2019, which is unbelievable. Time flies. Since it's just us, I want to keep this very broad and start with your take on the current state of play. I know this is something you used to do at Benchmark, like a "State of the Union" for the market. I'd love for you to do one for us — talk about the market of summer 2025.

Bill: I'm excited to do this. Yes, I used to kick off our LP meetings with a presentation on the state of VC. So this is a process and presentation I'm accustomed to. I've noticed a lot of different things happening in the world lately. The venture capital world may be changing permanently. A lot of my presentations were based on what seemed to be inherent cyclicality in the venture capital industry, and that cyclicality has been somewhat disrupted or upended recently — it's gotten a bit messy, which we'll get to.

Before we dive in, let me offer two premises. First, Michael would agree with this — I'm a huge fan of systems-level thinking. There's a book on how to do systems thinking, and it's so cool. All my time with Michael at the Santa Fe Institute basically relates to this theory that systems behave differently than their component parts.

It's not easy to see the connections between systems — it's a difficult thing. But when we dig deep, I think a lot of components in this industry are colliding with each other, and the combined impact of all these factors is what's most interesting. So you have to step back a bit and look at the big picture from a distance.

Second, I also want to state upfront that I'm not making any judgments about the participants in these activities. There are people and companies who took actions that changed the state of this field, and I think their actions are all rational, all in their own best interest. The negative effects may not be positive for the world, but I don't see it that way. I'm not accusing anyone of malice. I want to get that out there first.

If you'll allow me, I'll dive right in. I want to talk about some market realities I'm seeing. So in part one, I don't want to over-analyze — I just want to highlight some things.

If you're in the VC market, and by the way, I think what we're going to discuss matters for VCs, founders, LPs, and anyone touching this ecosystem. This is very high-level stuff. So let me talk about the realities, and then we can exchange views on some of the interpretations.

The State of the U.S. Primary Market

1. The Rise of Mega Venture Funds

So the first thing I'll bring up, which everyone is discussing — I'm just presenting it as a key variable without over-analyzing — is the continued rise of mega venture capital funds.

In the beginning, you know, everything was bespoke. Most brand-name funds focused on early stage. They didn't participate in late stage, and their fund sizes were modest compared to today. Many well-known funds have increased their deployment from $500 million every three or four years to $5 billion — a 10x increase. They're very actively participating in what we call late stage. Though I've always felt "late stage" is a euphemism for big checks, you know, because if someone's willing to put $300 million into a 12-month-old AI company, right? So that's not late stage. That's just a big check.

Many companies have moved upmarket, and then they've also created different sector-specific funds and so on. All of this has led to a significant increase in capital under management for many brands. And there are many new players who have entered the late-stage market in different ways. Some legacy institutions participate occasionally, like Fidelity and Capital Group. But I think Atreides, Coatue, Altimeter, Thrive — they're all very active, and I think they're doing distinctive things in the market. Also, Masa (Masayoshi Son) is back. We didn't hear much from him in recent years, but he's active in the market again.

Patrick: He's his own indicator.

Bill: Yes, I agree with that. So there's more money in the market now.

2. Zombie Unicorns

The second reality people talk about, which is somewhat surprising, is "zombie unicorns." I don't love the term but it's the most used. If you look at the numbers, I like to use the arrival of LLMs as a dividing line, because it really was a watershed moment — everyone got excited about this new platform shift. There are roughly a thousand such private companies that have raised over $1 billion. ChatGPT tells me 1,250, NVCA says 900. Let's call it roughly a thousand.

Bill: Supposedly each of them raised $200 to $300 million, you know, add that up and it's $300 billion. NVCA estimates that LPs have $3 trillion on their books. I've had one-on-one conversations with LPs, and their participation and investment scale has slowly increased from 5-7% to now 10-15%. Some can be as large as half their private allocation. So we're seeing that while some private PE is much larger, venture capital is taking up an increasingly large portion of balance sheets. So this matters. I think there's a lot of question about a whole cohort of companies. One is: what are they actually worth? You know, their last round pricing is still from 2020.

Patrick: Around 2021.

Bill: Yes, when the market really peaked — the second year of COVID, if you recall, all tech stocks went crazy at that moment, Zoom went crazy at that moment, everyone performed incredibly well during that period. So the question is what they're worth. The investment community doesn't seem particularly interested in this category — their overall growth rates aren't high. I want to talk about why I think that is.

Most people probably won't believe this, but I assure you it's true: nobody has incentive to mark things clearly. So if you're unfamiliar with this world, you know, private investment — whether PE or VC — is a strange model where GPs, the people doing the investing, quote prices to LPs and mark their own prices.

There are auditors causing trouble behind the scenes, and you'll hear LPs complain. Some companies are more conservative and mark lower, others mark higher. So they're getting mixed signals.

3. Distorted Company Valuations

Patrick: Same asset, different prices from different GPs, right?

Bill: Right, but people may not realize that the managers of VC groups within large endowments have no incentive to try to adjust this number. In fact, many of them are based on book value. So if anything, they have inverse incentive to correct it.

Patrick: But don't founders have incentive to get this right? In the long run, isn't it better for companies not to play games?

Bill: That's a great question. I think there are two things that counterbalance that. One, every founder I've ever known has taken their ownership percentage and multiplied it by the company's all-time high valuation and treated that number as their net worth. I don't think that's right.

Patrick: But who cares? I mean, it's meaningless.

Bill: I'll say it again — I'm not judging. I think it's natural to do that. But to mark that number down 70% is a difficult thing to do regardless.

Then, the other issue is liquidation preference, frankly. This is another technical detail, so let me explain to listeners. The total amount raised directly impacts the importance of liquidation preference. In an M&A scenario, investors can choose liquidation preference and get their principal back rather than converting to common stock. So if a company raised $300 million at a $2 billion valuation, liquidation preference doesn't matter much. If the valuation drops to $400 million, liquidation preference could represent 75% of the company's value. You know, that's a real problem for people.

Patrick: If we went back to that list of a thousand zombie unicorns and dug deeper into the space, you'd find how many of them are profitable, and so this problem can go on forever, until they're willing to reprice, versus companies that will actually go under at some point and are ultimately forced to raise and reset their price.

Bill: Well, I have to admit I haven't done a statistically significant survey, though it might be interesting if someone did. Maybe a fund of funds, or PitchBook, or someone at one of these organizations has the data. I'll tell you, this is to tee up what I think happened. We had a very long period of zero interest rates, now called ZIRP, which was unprecedented in a hundred years, zero rates for five, six, seven years.

Patrick: A long time.

Bill: On the one hand, it delayed any VC correction; on the other, it created a massive amount of money and speculation. The funniest thing is, I got an invitation. I've only met Mr. Buffett once in my life, at a small fundraising event with twenty people, each allowed one question. I said to Warren, you know, if interest rates are zero, your DCF doesn't work. He replied, "You're right." That was it, a brief encounter with greatness.

Anyway, speculation ran rampant. That figure I mentioned, $20–30 billion, that scale was unprecedented before that.

When companies raise that much money, certain things happen. I think too many participants enter a single space, and companies that should have been eliminated earlier survive. That makes market expansion harder, because instead of one or two survivors, you end up with three to five. When you're overfunded, you can do anything. There are many articles and studies showing that constraints breed creativity, that you're better off focusing on just one or two core products. But with too much money, you do seven projects.

I think we had a small correction around 2022, 2023. But this was before AI exploded, and most people were working toward what you call break-even. So once you pivot to break-even, you cut those seven projects down to two. But those seven projects and the overexpanded sales team had generated revenue that wasn't sustainable. So when you cut spending and move toward break-even, your growth rate naturally takes a hit, and I think that's what caused the low growth.

I agree with you that many companies have enough capital to reach break-even or near break-even, and based on all my earlier discussion about the beautiful nature of traditional company-building, you might think that's a good thing. Of course, I'm for that. But there's an underlying reality: they might actually be able to persist indefinitely, and that's where the "zombie" label comes from.

Patrick: Yeah, so what's the point of any of this? Since nobody has incentive to correct valuations, does it just stay this way forever?

Bill: We'll come back to that. Let me continue — I want to get these market realities on the table first, then we can dig into what might happen.

4. Exit Windows Closed

Patrick: The next issue is exits, meaning how these companies would actually be priced in a real market.

Bill: Right, we have mega-funds, zombie unicorns, and capital markets. For reasons that are hard to articulate, IPO and M&A markets have been frozen for the past few years. 2021 was actually pretty good on both fronts, but now things are stalled. I think if you look back at last year, 2024, this is really important. The Nasdaq was up 30% that year, yet the window remained closed — that seems to be the consensus view out there. In all my history of watching capital markets or being in venture capital, I've never seen the Nasdaq perform well while exit windows were closed.

Patrick: No IPOs, right.

Bill: Yeah, it doesn't make sense. These two things used to be correlated, so something else must be going on now. I've been very focused on IPO discounts, particularly those imposed on the market by the big-name banks. But some argue that going public is too expensive, and others that being public is too expensive. Of course, money is everywhere. We'll come back to this — successful companies don't need to go public now, or at least they don't need to rush.

M&A is harder to explain. Everyone blames Lina Khan [FTC Commissioner], but she's gone, and there hasn't been record M&A in the first five months of this year either. I think it might have something to do with the Magnificent Seven. These seven companies are sitting on absurd amounts of cash, which should theoretically lead to massive M&A, and I'm sure they'd love to deploy it. But Washington doesn't want to see that, and the EU certainly doesn't want them active, so things are stuck. Nobody wants to take the risk of a deal not closing. Even a big deal like Wiz — the moment it was announced, they said it would take over a year to complete. For a board and management team, waiting a year is very hard, too hard.

Patrick: Do you think we'll soon see a private company worth $1 trillion?

Bill: How far is SpaceX from that?

Patrick: About a third of the way there. OpenAI is about a third too. Stripe is a tenth. There are several that could likely get there if they keep succeeding. What I'm getting at is, if you can be a trillion-dollar private company, do you even need to go public?

Bill: Kind of crazy. We'll get to that. One last factor that might be affecting M&A is overvaluation, like what we did in 2021 — we're still pushing valuations for the most exciting companies to historic highs today, and that affects M&A too.

Patrick: In your view, why does this keep happening? Is the feedback loop essentially what we just discussed?

Bill: I think ZIRP was the main driver before LLMs. After LLMs, everyone believes AI is the biggest technology platform shift of our lifetimes. So if you believe that... there's another thing: I recall thirty years ago, when Mauboussin and I were at First Boston, network effects and compounding weren't fully understood or appreciated.

Now everyone completely believes in them, so people who watched Google or Meta go from $12 billion to $3 trillion — if they think a company might reach those heights, they feel they can't miss out — which is rational for an individual investor. If everyone thinks that way, the market prices in that expectation, but we'll see.

5. LPs Facing Liquidity Problems

The next thing we need to talk about is that many LPs are facing liquidity problems. This is a new phenomenon, related to the IPO and M&A windows being closed. There's also a unique data point: in Q1 2025, US universities issued $12 billion in bonds, the third-highest quarter on record. If you're using debt to fund capital calls, it's because your endowment doesn't have enough liquidity to pay out the 3% or 5% annual spending it used to.

Then just recently, you may have seen Harvard announce it was selling $1 billion of private equity assets in the secondary market. More interestingly, Yale University announced it was seeking to sell $6 billion of private equity in the market. Yale is the institution that does this — that's very important and very interesting. From a historical perspective, no institution has had greater influence on endowment management strategy than Yale. David Swensen [former Yale CIO] pioneered this model.

Patrick: He's the godfather of this model.

Bill: Yeah, without question. Under David Swensen's management, Yale achieved a 13% average annual return over 35 years. He's famous for the Yale Model, which invests far more in illiquid assets than liquid ones. Originally, nobody did this because of the lack of transparency, lack of liquidity, and the difficulty of managing these things. But he did it, and it worked. What I'm saying is, what we're seeing now may be the result of everyone copying the Yale Model. You know, Howard Marks has a famous line about how you make great money doing something non-consensus but correct. But what if everyone copies David Swensen? If everyone puts 50% into illiquid assets, can that still work? I think that's a very challenging question, but that may be what's happening. And Yale, the institution that led us to this strategy, is trying to get out — I find that very interesting.

Patrick: If you consider the LP liquidity problem, could this be the key to breaking the stalemate you just described?

Bill: It's possible. If I can get past these realities.

6. The AI Wave Interrupted the Market Correction

The AI wave came at a convenient time — this is point five of my six realities, I think. We were heading toward a small correction. You remember, Patrick, everyone was tightening their belts, laying people off, pursuing break-even, worried about whether they could raise again.

In my thirty years in venture capital, every time the industry overheated, there was a correction, then everything quieted down. I've seen Morgan and Goldman open offices on Sand Hill Road and close them. I've seen Fortune and Forbes cover Silicon Valley and pull back. I've seen it several times.

But there was no full correction this time, because AI arrived and everyone got excited. I'm not saying they shouldn't be excited — if this really is the biggest technology platform shift of our lifetimes, you have to be excited, and that affects the zombie unicorn cohort and everything else.

But suddenly, investment enthusiasm surged. What are the multiples for AI companies on revenue? My god, 10x, 20x for an average company, right?

Patrick: About that, some even higher.

Bill: Right. Despite traditional LP money being tight, they've been able to find capital elsewhere. The Middle East is the main source. How many of your friends have gone to the Middle East in the past twelve months? A lot. They're all talking to fundraisers, so the money found its way, and everyone's chasing this opportunity, nobody wants to miss out. This is a very important component of the whole situation.

7. Companies Prefer to Stay Private

The last reality, which you already mentioned, is that there's a new dynamic in the late-stage market. I think Josh and the team at Thrive led this, though certainly not just them — they've just been at the forefront.

They'll go to companies that were preparing to go public, that the media had reported were going public, and make them an offer they can't refuse. Founder liquidity encourages employee liquidity, and possibly angel liquidity too. And they basically take on the burden of persuading your company to stay private.

The recent example is Databricks. Stripe's Patrick and John have also mentioned on different podcasts and in speeches that initially they were saying "maybe we'll go public, but no rush," and later it became more like what you said, "maybe never." I've spoken with some LPs who've traded Stripe shares, and the company has been somewhat accommodating about it. That's very new and unique in our world.

Patrick: These companies can get the capital they need, whether for employee liquidity or early investors selling shares — it's basically like a "by-appointment public market"?

Bill: Right, like the old pink sheets trading, by appointment.

Patrick: Stripe is undeniably a great company led by outstanding founders. If you can have your own private market, why take on the extra work, regulation, data disclosure, and let competitors know your business? It makes sense for everyone, so I wonder if this model will just keep going. If LPs can get liquidity by trading Stripe shares, doesn't that solve the liquidity problem too?

Bill: We're about to dive deep into all of this.

I want to add something — these investors who encourage companies to stay private have another motive. In a traditional IPO, banks are extremely careful about allocation. If a major public fund or private equity firm applies for shares, they'll typically oversubscribe by 100x hoping to get 1-2% allocation; they could never get 30%. But when these investors do large private financings, they can get 30% of the company, far more than in an IPO. This way they get a higher ownership stake than they would through the traditional IPO process.

It's kind of like an oligopoly, capturing the IPO growth premium away from public markets. Amazon went public at under $10 billion, now it's worth over a trillion — public markets enjoyed all that compounding. If you delay going public and get a high ownership stake early, these investors do better than buying after the company lists.

Another important point: if they turn around and tell LPs, companies aren't going public like they used to, if you want exposure to these high-growth tech companies, you have to invest with me. That's very persuasive.

What needs to improve in U.S. capital markets?

1. IPO costs are too high, companies don't need to go public, tokenization could be the solution

Patrick: You understand the market realities now, and I want to go deep on all of this. My framework, what I find interesting, what I find interesting is — I've always wanted capital markets to function well. U.S. capital markets are one of the most important innovation engines in world history, they've driven countless innovations.

So to me, well-functioning, healthy capital markets price risk well. So I'm for anything that does that. I'm curious, from that perspective, given these market realities, where do you think the system most needs improvement, and how would you want it to change.

Bill: Yes, I I share your wish, I think we'd be much better off with more companies participating. One thing I didn't mention in reality that most people also know, the total number of U.S. public companies has dropped significantly from peak, fewer and fewer companies are going public.

I think a large part of the reason is the IPO process and those well-known investment banks. I asked my friend Jay Ritter to reanalyze the data, now IPO underpricing is around 25%, 26%, plus 7% fees, your cost of capital reaches 33%.

I know a CEO preparing to go public, talking with investment banks, the bank said you should price at X, the founder said I can raise $1 billion in the private market tomorrow at 20% above that.

As you said, if private markets are this liquid, flexible, optimal, why go public? I don't know what needs to change. I think as long as it involves fundraising IPOs, everyone will bypass that part.

I know of a CEO, there's a record of him talking with bankers, the bankers said, we think you should price at X. The founder said, I can raise $1 billion tomorrow at 20% above that price. To your point, since private markets are so flexible, liquid, and well-optimized, why go public? I don't know what needs to change. I think as long as it involves fundraising IPOs, everyone will bypass that part.

Hester Peirce wrote an article titled A Balancing Act of Creativity and Cooperation, about eight pages, very much worth reading. She was the longest-serving commissioner at the U.S. Securities and Exchange Commission (SEC). Currently, she is one of the four most crypto-friendly commissioners. The article proposes that blockchain technology could be the way to fix the IPO market, a somewhat controversial view, but one I personally support.

Patrick: What did you mean by "free trading of private assets"?

Bill: What I mean is, tokenization of securities. In the future, people won't allocate to crypto the way they do IPOs, they'll turn to distributed ledger (DL) technology. In fact, initial coin offerings (ICOs) are already operating this way.

Indeed, this is a very interesting topic. I'll continue following it. Right now, regulatory pressure is very heavy. We've seen some strange "workaround acquisitions" in artificial intelligence (AI), like signing a licensing agreement first, then hiring people, but we haven't seen a truly big case in a long time, this seems to be a circuitous strategy.

And when assets are overpriced, deals are hard to close. For example, some AI companies are raising funding rounds at $15 billion valuations. I understand why Apple might want to acquire a company like Perplexity, but such high valuations make transactions difficult. Regarding capital markets, what you mentioned is very interesting. Many people claim we have the most well-functioning capital markets in the world, the envy of the globe, but I don't entirely agree.

Patrick: You mentioned the Middle East has been very active in this wave of technology, they're working hard to participate in the most interesting companies, technologies, and infrastructure. Any other capital market innovations you find interesting?

Bill: I don't know if you'd call it innovation, but there was a recent announcement about Coatue that caught my attention. I haven't spoken with Philippe about it, but I think they've lowered their minimum. In the past, their previous minimum was $5 million, now it's dropped to about $25,000, and they'll work with an investment bank to handle the related matters. This is similar to what I mentioned earlier about pitching to LPs, but this approach uses a pool of capital, sometimes called "dentist and doctor money," that couldn't invest in a fund like Coatue before, but now can.

I've heard similar things are happening in PE, a large private equity firm is lobbying in Washington to allow 401(k) plans to invest in private equity, to open up different sources of capital.

Interestingly, when I was testing this series, someone pushed back and said, U.S. institutional money can be found elsewhere, but that's just letting the top earn more. You could compare it to a pipeline with imports and exports, where the export is blocked. I can't think of a better analogy offhand, the human digestive system might be the best one — eating more food doesn't solve constipation.

2. Companies are burning cash massively, but only a few stand out

Patrick: When you talk with limited partners (LPs), you have to mention specific names. What are they saying to you in private? What do you think they're not saying that matters?

Bill: I think awareness of market realities has increased. They have to make decisions. For LPs, this is a long-term decision. If you work at an endowment, you don't have much time to make decisions, because your feedback cycle might be 10 or 15 years. It's hard, but you have to start thinking about whether these changes we're discussing are temporary or permanent.

If they're permanent, you need to change how you do things. For example, one LP I spoke with has been in and out of Stripe, knows the person at the company who handles capital markets, and is starting to consider that this might be permanent, also thinking about how they need to prepare for such a world.

Patrick: Apollo published an interesting report showing that among companies with revenue over $100 million, 87% are now private. Of course if you look by market cap, public markets have higher share because of the big tech companies. This phenomenon is dramatic, even the minimum threshold of $100 million revenue, many companies have reached that level. So I'd say, we live in a world where private markets are very active, that's undeniable.

Bill: I want to adjust the order, I have about five analyses. I think this is a more chaotic world. The best world you mentioned is one with well-functioning capital markets, where capital markets are efficient, going public is easy, liquidity is strong, transaction costs are low. I do think that world is better.

If we move to a new world where ordinary consumers enter high-growth tech, putting their 401K, individual retirement account (IRA) money into venture funds charging 2% management fees and 20% carried interest. I just think information will be more opaque, less transparent. Fraud will increase, transaction costs will rise, that's the inevitable result.

Take Stripe or other companies, that's just one company, in this example we might mention five companies, but what we're really worried about is 1,500 companies, they can't all do what Stripe does.

Patrick: You taught me something years ago: you have to play the game on the field, but also think about future rule changes, prepare for the future. But if we take this more chaotic, more private-market-dependent, liquidity-reality field approach, I'm curious, what do you think different groups should do, starting with founders, all the way to real value-creating entrepreneurs, funded by these capital markets. In the AI world, if they can raise at a $15 billion valuation, what would you advise them, what's the optimal choice under current rules?

Bill: They're forced to act under the rules on the field. This is also what I find worst about this world. I recently learned a word, "gavage tube," do you know what it is? The French use a gavage tube to force-feed geese to produce foie gras. In this world, the reality was 2021 was like this, as long as there was some trend, someone would come knocking, trying to give them $100 million, $200 million, $300 million.

I think for founders who've struggled their whole lives to raise money, this must sound like the most absurd comment, but that's the reality. I went through this with the Uber-Lyft competition, now capital wars are unfolding across every sector.

You mentioned the concept of traditional company building, not spending $100, but burning $100 million or $150 million per year, all the big AI companies are doing this, probably consuming even more.

This isn't your grandfather's startup or venture capital. This is a completely different world. If you're a founder, ignore all this, build your company the way you want. But if your competitor raises $300 million, and their sales scale 10x, 50x, you're dead before you know it.

You have to play a game on the field, the good news is, because these investors are all eager to give you money, you can probably get founder liquidity. I think this is detrimental to the company's long-term success potential, but because it fits their strategy, they're all encouraging it. So I think, if someone's willing to pay 30x revenue, and forcing you to play a game you don't like, burning hundreds of millions per year, then you probably have no reason not to cash out a bit.

I think if we remove all the small and medium wins, only thinking about hitting home runs all day, that's harmful to the ecosystem. But if that's how I feel, it feels like we learned nothing from the zero interest rate policy era, all the zombie unicorn problems we talked about, we're just repeating them with AI companies. It's like, this might be part of why we haven't experienced a market correction, but the way we're funding these AI companies now is exactly the same as how we funded those companies back then.

Most AI company revenue is mostly compute power, economic returns are the ultimate deciding factor

Patrick: I want to start with something really important, which is understanding your view on AI as a new general-purpose enabling technology — that's the key difference between now and 2021. We've never seen a technology wave like this, and we've never seen companies with this kind of revenue growth.

I know you're like me because you love technology. I use this thing every day. It's the most amazing technology I've ever encountered. So I want you to riff a little on what I'd call the bull case, which is that people aren't actually being irrational, because we really could get to 5% GDP growth or whatever crazy number. Because this really is a different class of technology, arguably comparable to the internet itself.

Bill: First, I agree with you. I would never take the position that this isn't a legitimate platform shift. If it's a platform shift, like mobile internet, internet, or PC, that's big enough — it doesn't have to be better than those.

Patrick: Even if it's just another platform shift?

Bill: It's definitely one of those, and probably bigger, which leads to everything we're talking about. I'm not judging any participant before it begins. I think things are what they are. I have this thought in my head, and I haven't fully worked through all the implications, but some of the revenue growth is actually compute resale.

A lot of companies in the market are basically repackaging and reselling foundation models and cloud services. Many companies are actually negative gross margin. The companies buying these products may be getting them cheaper than buying the models or cloud services directly, and that revenue is being counted three or four times over, with negative gross margins.

Until we truly care about unit economics, but that doesn't matter in the all-in phase of a capital war, everyone can only grab market share. The window of opportunity in front of us is wide open until things get resolved. I have no doubt about this — even setting aside foundation models, like what Bret Taylor is doing at Sierra, I have no doubt that AI will substantively change every enterprise. I think a lot of what's happening is a rational response to what's going on.

Patrick: What about the technology itself? You've been through many of these technology paradigm shifts and invested through them. What excites you most about this one, especially compared to others you've participated in?

Bill: My answer is personal, and it relates to what you just said. I probably do 40 or 50 searches on AI platforms every day — more than I search on Google. It's almost like a very rapid way of learning, like super-fast learning of details, things I forgot I didn't know, and it's happening every day. I think to myself, for people who are natural self-learners, the speed at which they can get things done and advance is just amazing.

Then I think beyond large models, from Tesla's FSD to other problems solved with traditional AI. Those are super interesting to me too. Maybe more profound.

I do worry about limitations with large language models. This might be solvable, but their language is as hard to understand as potholed ground. They're not great with numbers. When people say AGI will replace all computing — I don't think they have to fix certain things or merge it. Now when you ask an AI a math problem, it runs and writes a bunch of Python code. You'd have to do a lot more of that kind of work to get to that level. If you support that. But isn't that the real argument? I can't refute that.

The Problems Facing GPs and LPs

Patrick: Let's take one more step forward and talk about GPs. Same question: what's the rational move under the current rules of the game? There are two versions here, a "Spock" rational answer and a "Kirk" emotionally-driven answer.

The Spock version is: since the market is what it is, I'll start my own firm and rationally maximize investment returns.

The Kirk version is: if you were to restart a venture firm today, what would you do? Would you do a small fund like you did at Benchmark? Or would you build a fund that invests everywhere with a different fee structure? I'd love to hear your answer from both perspectives.

Bill: I want to emphasize one of the last two things, which is that time is a huge problem. We've extended the time for these companies to reach profitability from 5–7 years to 10–15 years. I don't know the exact numbers, but every LP recognizes this problem.

I think I've forwarded you this NVCA chart showing the percentage of committed capital returned by venture funds within 5 to 10 years. It used to average about 20%. It peaked at 30%, hit a low of 5% last year. It's now roughly in the 5–7% range, and that's exactly where the LP liquidity problem lies.

Time is such a huge problem because of cost of capital, internal rate of return (IRR) constantly eroding returns — everyone loves to say what matters is distributed to paid-in capital (DPI), not internal rate of return (IRR). But if time doubles, internal rate of return (IRR) — that's what really matters. Beyond time and cost of capital, there's equity dilution. Every zombie unicorn issues 3–6% in employee equity compensation annually. When you combine those two, it becomes a real problem.

For example, you originally expected a $100 return from an investment in year 10, and now you're pushing it to year 15. If you just think about 10% compounding, it should be worth $160 in 15 years. You know, if you think about these people. If you're investing in a project for outsized returns, your cost of capital isn't 5%. That's the risk-free rate. 15%, then 20% annually, 15% plus 5% from equity dilution. Now, if you wait another five years, guess how much you need to replace that $100? Delaying five years to hit people's expected returns for this asset class — the return needs to be $250.

So I think this is a big problem. I think a certain number of companies genuinely achieve acquisition or IPO at some stage, and then entropy sets in, and all companies face difficulties with long-term growth.

People like to discuss what fund returns would look like if you stripped out the biggest winners. But I haven't asked anyone this: what if you keep the big winners and strip out everything else? Because it feels like we're heading in that direction. So I don't know, after all that, I really don't know the answer to your question. I've spent my whole career in the early stage, and I still love that period. I think it's a time window where you can make the biggest bets and get the biggest returns.

I really don't want the next generation of GPs to see every company go through what Uber and Lyft did. When you walk into the boardroom, you know the other company just raised another billion dollars, and the decision becomes: should we sustain another two years of negative gross margins to capture market share? You won't find this in a Harvard case study.

It's like playing a unique deck of cards with super high stakes — a strategic poker game you won't learn from Good to Great. This isn't traditional company management, nor is it what Warren Buffett writes about in his shareholder letters. None of that applies in this world of capital wars.

Patrick: Now I want to talk about LPs, and the trend of capital seeking the highest risk-adjusted return. Generally speaking, from a rational perspective, over time capital pools flow around seeking the highest return for the risk taken. That's the whole point. So I'm curious, what do you think is preventing this from happening? In other words, what should LPs do now? They're capital owners, they represent capital owners. On the surface, their job is to maximize risk-adjusted returns. What should they do, and what's preventing them from doing it?

Bill: This is probably the last point I want to emphasize. You raised a very illuminating question at the beginning of the podcast: could the LP liquidity problem become some kind of catalyst for change in this world?

There are many factors pushing this change. Time is a problem. LPs are still levering up. There's widespread discussion in Washington about endowment taxes on funds, which would create more liquidity pressure. This is something they've never experienced before. There are also research funding cuts — not just the radical cuts at Harvard, but even normal NIH and NSF grant reductions.

These cuts to indirect costs or other portions also lead universities to demand more from their endowments — for example, needing 5% or 6% annual payout instead of 3%. These changes could put LPs in an even tougher position. Yale University may have been the first to enter the secondary market, which seems exciting. If you're a small endowment that never invested in Sequoia Capital, you can now get a piece through Yale. However, as more and more big players join, the aftershocks of secondary pricing could have chain reactions across the entire market.

Additionally, I think another big thing to watch is whether the Middle East changes its mind. The chief investment officer of Qatar, Sheikh Saud bin Nasser Al Thani, has stated that private equity's time is running out. He joined investors increasingly concerned about the industry's valuation methods. If this view spreads and becomes consensus, influencing other players, that would be huge.

So I think these are the areas to watch. If I were an LP, what would I do? I think I'd definitely be making two-sided bets in the late-stage private market, observing market dynamics to get a feel for things. I don't want to cause a bank run, but you might really need to reassess whether the Yale model still works. I think it definitely worked when only Yale was doing it, but I don't know if it works now. I think it would be interesting to find a private equity firm to actively look for opportunities among those thousands of zombie unicorns, trying to extract value. I think you could view that optimistically rather than pessimistically. So I'd probably be interested in that too.

Patrick: If you're just thinking about the return piece, as your former partner Andy Rachleff used to say, to make the most money you have to be contrary to the crowd, you have to look for opportunities the market is neglecting. Is part of the answer possibly in private markets outside of AI, where pricing and supply-demand dynamics are completely different? Like, if you go to a regular company, capital markets aren't particularly enthusiastic about investing in them, and to some extent, they evaluate these companies with very strict calculation standards. It's completely different from what they're doing in AI — should you pay more attention to these places?

Bill: I even think some of the later-stage investors I mentioned earlier are thinking this way. They're thinking, if I can find a traditional company that may not realize AI will enhance it, but we can do that ourselves — maybe that's a disruptive way of looking at things.

Howard Marks originally came up with "non-consensus and right," and I read him a lot. But this idea conflicts with platform shifts. Because platform shifts are now consensus, to be contrarian you'd have to not invest in AI, which sounds absurd. So it's hard to do both at the same time.

One interesting thing about AI is that the big companies seem to be moving remarkably fast. If you go to ServiceNow's website, AI is everywhere. Microsoft's earnings call transcript mentioned AI 67 times, and Satya spent two hours talking about it. That's odd. In Crossing the Chasm or The Innovator's Dilemma, we read about how big companies should be slow to respond in mobile, in internet — sluggish. That creates opportunity for startups. But this time I think a lot of big companies started paying attention very early.

Patrick: Do you think it's just happening in a different form now? Maybe, for example, theoretically Google should be in the best position to dominate every AI use case. And yet, basically no one I know uses Gemini or Google for co-generation, or as a tool they use every day, for daily LLM work. Frankly, not for much else either. Instead, they're using startups, Cursor, Anthropic, OpenAI. So even though big companies are moving fast, the same phenomenon is still playing out within tech itself.

Bill: I think there's data on both sides, and I think your argument is interesting. Apple is an interesting argument. You know, Microsoft missed once but survived, which made it more capable of being vigilant about the next opportunity. Right? I saw an interesting interview where Friedberg interviewed Sundar, asking if he had read The Innovator's Dilemma, and he admitted he hadn't. So when your company is thriving, these theories seem like they're for other people — though maybe it's time to read it now.

Patrick: When evaluating an exciting new AI company, the nature of its revenue may differ from traditional models like enterprise SaaS. As an investor, how would you assess the quality of revenue for a new AI startup?

Bill: I think it's very difficult, for the reasons I mentioned earlier. You might get a $1 million order with negative gross margins. On the other hand, you'll find that two-generation-old AI models now cost one percent of what they used to. You might have confidence in optimizing pricing in the future.

The Benchmark partners have been studying and evaluating when companies enter optimization mode and how they make different decisions compared to being in experimentation and sandbox mode. With your amount of capital, you can run sandbox mode for a longer time. Before entering optimization mode, I want to emphasize that in the first two years of the internet, all startups were building on Sun and Oracle, but five or six years later, no one was doing that anymore, so paying attention to this transition is very important.

If Chinese Tech Giants Open-Source Their AI Models, It Would Be Incredibly Powerful

Patrick: How do you view the international competitive landscape in AI? In some other technology platform shifts, this competitive dynamic was relatively weak — mostly American or Western technology was at the forefront. China is obviously a region to watch, especially projects like DeepSeek. Now there are also more and more Chinese startups launching impressive products. How do you view the international factor in this race, particularly the US-China AI competition?

Bill: There's a very interesting development in China that's worth watching. When DeepSeek launched and grew rapidly, we were all watching the American response, American models, and Washington policy.

But in China, Alibaba open-sourced Qwen, Xiaomi now has its own model — I can't remember the name, maybe MiMo — also open-source. Baidu's Robin Li's model was originally closed-source, but he said it would be open-sourced in June. If this level of competition ends up producing open-source products from four well-funded companies, that would be incredibly powerful. We've learned that these models can train on each other, help each other, get better. So if there are four open models, they can train on each other, everyone can use them, and I think that would create enormous choice and experimentation that won't happen in the US. That's the most fascinating part of the international AI narrative I've seen.

Patrick: Do you find yourself more loyal to certain groups, hoping they win over others? What are you most rooting for?

Bill: That's an interesting point you raise. I've noticed that some of the most aggressive Chinese investors are actually betting on a new generation of VC-backed defense companies. I hate that you might become a war profiteer, but I know it can happen because when I invested in Uber, I would defend it at all costs. It's natural — like your child, you protect it, so your position shifts with your investments. I still feel that way about anything Benchmark-related, and I'm not sure I'll ever not feel that way. That's reality, that's how the world works.

As for the technology itself, I find some non-LLM directions very exciting. I'm looking forward to seeing what robot intelligence will accomplish. I hope we can make progress in healthcare. I don't think all diseases will disappear in ten years, as some AI founders claim. I think that's too exaggerated, but the process will be interesting. As you said, I use these things every day. The pace of change is the fastest I've seen in my career. If you don't read the news for a week, you feel like you've entered another world.

Patrick: You mentioned the defense startup ecosystem just now. I want to extend this to the physical world, the hard tech ecosystem — much of which actually has nothing to do with war, like mining companies and so on. What do you think of this category of companies? They're undoubtedly tech companies, usually operating in very large markets, but they're capital-intensive and take a very long time. Areas like nuclear fusion, fission, and so on. What do you think of this kind of private market and tech investing? I know you haven't invested much in these projects before, maybe you don't like them?

Bill: Generally speaking, if I were a professor, I'd say you can study this mathematically — the returns in these areas are usually not high. You can look — 15, 20 years ago there was massive VC money pouring into solar, and the results weren't good. Of course, there's one exception to this rule, which is that anything Elon Musk touches will have returns. So SpaceX and Tesla are data points, but they're actually exceptional cases. And they're both associated with Musk. So I think we need to see four or five non-Musk people do this before we know if it's viable.

From what I understand, hear, and have researched about his execution capabilities and the speed he's demonstrated at these companies, I'm not sure others can do this, or have the capacity to. If they succeed, it would be a great thing for the world. By the way, we've seen that capital abundance has led to greater interest in capital-inefficient businesses, as if there's some correlation between the two. So another thing to watch is, if capital becomes tighter, will the demand still be there? A lot of these businesses involve regulation.

Patrick: It seems like this is about to happen — companies that got venture funding, early private market support, and entered heavily regulated large industries will get sorted out. I'm curious, a company like Anduril, maybe valued at $30 billion, not quite at SpaceX's level, but it's an important data point — does it make you feel we do have the capacity to execute in capital-intensive companies?

Bill: No doubt, from a regulatory perspective, that's true. So I think, historically, it's been very difficult for companies to break through in these industries, and the main reason has been regulation. Before Tesla, there were about 7 electric vehicle manufacturers, right? There were other attempts to make cars, but none succeeded. I think a lot of people got blocked in this process from a regulatory perspective.

Anduril getting certified by the Department of Defense and actively selling to the military is undoubtedly a new milestone for a startup, and impressive. I don't know if that means every VC should jump in. You know, it's hard. If you can start a software company, or as people say, start a social network company, make it thrive, achieve high-margin revenue growth — that's a much easier way to make money than what we're discussing.

Patrick: What other areas in the ecosystem that we haven't discussed are you particularly interested in — types of companies, investment strategies, or dynamics?

Bill: If I were still an active GP, I think I'd be thinking about AI verticals and where AI performs exceptionally well. AI is very powerful with language, and coding is really just refined language, so AI is even stronger in programming — these areas are all important. Legal, customer service — there have already been lots of explorations. But I think there are still areas that haven't been fully mined. That fit is very interesting to me.

What If the System Corrects?

Patrick: Returning to the LP perspective and capital markets system-level issues we discussed at the beginning — you've laid out the actual situation, the various incentive structures, the lack of motivation for change. What do you think will happen in the next five years?

Bill: My intuition is that we're in trouble. Although I've had some success in venture, I've always been more of an analyst than an optimist. But I started in security analysis — I'm naturally inclined toward critical thinking, so my bias is there. Someone could certainly push back and say Gurley's always predicting the next recession or whatever.

Today's system leads to worse liquidity, less traditional high-quality company building, faster burn rates. In my view, that's not a good combination. It's all self-reinforcing. All the components I listed — I don't see a correction mechanism unless something happens at the LP level. I think we're getting more and more stuck in that loop.

You may have seen a great video where Josh Kopelman simply explained some GP math, from his perspective. But I find it hard to disagree with what he did there. From the prices we pay, the amounts we spend, and what would need to happen for VC returns to match historical levels — it all seems like a very difficult knot to untangle.

Patrick: So if a reset happens, what happens on the other side? Let's imagine a simulation scenario where we apply public equity pricing scrutiny or mechanisms to every available asset, and the result is a massive pricing reset. If we go through a bad period, what are the pros and cons after the reset?

Bill: I think most people would find it scary — having been through a few of these resets. But I did find that as an active GP, I was actually calmer and happier in reset environments, and found my work more fulfilling, more efficient, more productive. Some other people probably prefer bubble periods — those with sales talent who enjoy being in the middle of it. But I found that conversations about traditional company building happened much more efficiently and authentically in these windows.

When the dot-com bubble burst, the pretenders left Silicon Valley — there were B2C and B2B. The joke was that when making money was no longer easy, people left Silicon Valley and went back to consulting and banking. I don't like those opportunists — I think their motives are impure, they tend to over-promote, over-raise, over-participate in secondaries, and then exit when things might turn south. When it's moving at that speed, it's part of the world.

Back then there was a joke: when making money got hard, B2C and B2B became "Back to Consulting" and "Back to Banking" — because when the money dried up, the opportunists left. I don't like those opportunists. They tend to over-promote, over-raise, over-participate in secondaries, and then exit when things might turn south. I don't like it, but in a world moving at that speed, it's part of the game.

If the market corrects, people will go find new opportunities. One reason we're in this situation is that everyone has studied history, everyone understands compounding, network effects, cycles — everyone has seen booms and busts. Remember how long the COVID market drop lasted? About three weeks. Then everyone started buying the dip. So I suspect confidence in AI is high enough that even if people feel it's overvalued for six months, it'll bounce back quickly. Of course I would think that.

Patrick: If you were starting a brand new investment firm today, what do you think would be the most important component of building that firm's brand? We're in a different era now. Some of the newer private markets firms — Green Oaks, Andreesen Horowitz, Rivet — they were founded around 2010 and within a few years became very large, respected brands with their own way of operating. We're in a new era. What advice would you give to emerging investors starting firms this year who want to be industry leaders in 12 years?

Bill: What you just said reminds me of something unrelated to your question. Another negative consequence of the systemic issue is that some firms elbow their way onto cap tables by writing $300 million checks and differentiate themselves by becoming the founder's best friend. That's easy for me to say because I'm not writing checks anymore — no one's going to keep me off a board because of what I say, so it doesn't matter.

But they don't take on the responsibility of "helping you make better decisions." They'll never say no. An extreme example was SBF and FTX. No one joined the board, everyone believed he wasn't misappropriating funds, and it collapsed. Actually having someone who can call a timeout at critical moments, who can push on unit economics — that's useful. And I worry that's happening less and less.

The best CEOs — I often put people like Barton (former Zillow CEO) and Benioff (Salesforce CEO) in this category. Even Mark at Meta has said they believe going public made them operate more efficiently. Another downside of staying private forever is that you don't get that feedback.

Now, let me try to answer your question. I don't know, because given everything I just said, I can barely imagine what embarking on that journey would look like. So I really don't know. I'll have to pass on that one.

Patrick: Well, I'm glad my question sparked other thoughts. Maybe to close, you could leave everyone with some thoughts specifically for founders. I always like to come back to founders. It makes sense. Because without them doing things, none of this matters.

Bill: Yes, if you're fortunate enough to get into a hot company, you'll be in the world we just talked about, and I have a few thoughts.

First, unit economics will matter someday, but that doesn't mean you have to sharpen your pencil right now. As I said, two generations ago models cost one percent of what they do now. You can plan to shift to that model eventually, that's fine. I think it's okay to burn money now, but unit economics will ultimately be critical. Eventually, you have to scale the company and operate efficiently and productively. I think when you're going through this all-out battle, you can lose your way. I find many founders thinking about this. They think, that's what big companies do. It's too "bureaucratic," it's not what startups are about. But when you're above $100 million in revenue, approaching $1 billion, you can't operate that way anymore. This is actually advice for any cycle, but especially important in times of capital abundance.

My favorite, Reid Hoffman, wrote an article about Uber using the "pirate-to-navy" metaphor — that all startups begin as pirates and eventually have to become the navy. It's true. For some people that transition is uncomfortable, but you have to find your own way through it.

A related thought: Ben Horowitz wrote a great blog post saying they only want to back founders who will go the distance. It's genius because founders love hearing that. Actually every VC in the world feels that way, because the success rate of replacing a CEO is only 50% — why would you take that risk with your portfolio? But there were two or three paragraphs in that piece that said, "of course, provided the founder is willing to learn how to lead." I think our industry often glosses over this — no founder is born knowing how to lead an organization of a thousand people. Some people spend their whole lives studying how to get good at this. There are only a handful of founders, maybe 30, who worked with Bill Campbell to help guide them on how to do this. But it's not innate, it's not free. And you have to be willing to do it. There's a category of people who struggle with this. I had a great conversation with Michael Dell about this. He did it, even though he felt he didn't want to, and eventually found a way to do it and be happy. That part is hard.

Two final points. First, network effects are real, and if you pay attention to them, you can make them stronger. If you're in a hot market with tons of business and gross margin everywhere, it's easy to overlook network effects. But you should think: does your business have some kind of "data byproduct" or other mechanism where, if you have a thousand customers and grow to two thousand, the two-thousandth customer's experience should be better than the thousandth's? Can you design that mechanism into your system? If you can, it will profoundly impact your company's long-term success.

AI Consumer is Where the Opportunity Lies

Patrick: What do you think about the AI era? Is it mainly a data question? You just want your product to naturally generate more data, which then improves the product.

Bill: Suppose you're serving a functional vertical. If one customer's learning experience can benefit the entire team, that's incredibly powerful, and I think it's entirely achievable. In legal, for example, there are AI companies — I'm not involved with them — they're studying all the information you'd use in litigation, plus all the precedent and case law history. The AI does certain things, and if there's human involvement, you catch failures and continuously improve the model. These factors could let early leaders extend their advantage over the long run.

Patrick: Do you think AI will make consumer internet investable again? Since the mobile internet era, tons of amazing consumer-facing businesses have been built, but U.S. VCs have paid little attention to the space, and little capital has flowed there. Do you think AI could make this area attractive again?

Bill: I've noticed China has made significant progress in certain AI applications, which might be a signal. While most U.S. AI investment has focused on the enterprise side, there may indeed be underexplored opportunities on the consumer side. We actually had an early attempt with Character.ai. But first-generation LLMs had two problems that made them unsuitable for consumer: first, voice capabilities, which are improving; second, memory, which is also improving now, though implemented outside the main model — that's fine, it'll eventually be integrated into the context window. As technology advances, these shortcomings are gradually being addressed. If AI can achieve more natural interaction and more personalized experiences in the consumer space, then the consumer market could become a hot area for venture capital again. I wouldn't be surprised if four or five AI companies break through in the consumer space in the future. This might be exactly the contrarian opportunity we've been looking for, since most resources are currently concentrated on the enterprise side.

Patrick: Bill, this has been so much fun. Maybe we do a market update every few years — since you're not doing this directly for LPs anymore, we can do one for the whole industry. Thank you for sharing your experience and insights with us.

Bill: Of course, I know the LP community pays close attention to your content, Patrick. If anyone has feedback on what I said, wants to correct me, or has any suggestions, feel free to reach out. I love this industry, and I hope my sharing has been useful. So I encourage anyone with feedback for me, or who wants to correct something I said, to get in touch. I'd love to engage and learn more.

Patrick: We'll send the signal out. Bill, thank you so much for your time.

5Y Capital seeks out, supports, and inspires lonely entrepreneurs, providing them with support from the spiritual to all operational aspects. We believe that if the "crazy" you in others' eyes begins to be believed in, the world will become a different place.

BEIJING · SHANGHAI · SHENZHEN · HONG KONG

WWW.5YCAP.COM