20VC：Chamath说得对吗——DPI是唯一重要的指标吗｜OpenAI到底重要吗｜Mary Meeker的AI报告：分析｜IPO拆解：Chime、Circle与Thoma Bravo的新基金

Will we look back in 20 years and say OpenAI was a fundamentally important company? Maybe.

Every company's market share is up for grabs when there's a platform shift. The hyperscalers have taken very good, cash-efficient businesses, and they would break Mr. Buffett's heart because they've turned them into CapEx hogs. Public company investors are just mean VCs on steroids.

Guys, I am so excited for this. We had so many interesting things that we want to dive into this week. I want to start by diving into the deep end on a series of Chamath tweets. Chamath basically said, “Hey, TVPI is a bullshit vanity metric. You can't eat IRR. You can only eat net DPI.” When you read this, and when you read his quite opinionated stance, how did you feel?

I hate agreeing with Chamath on principle, but I agree with him.

What's interesting is that, if you read the Wall Street Journal article on Thoma Bravo, they just raised a $34 billion fund, right? Harry, your buddy. What the Wall Street Journal said is that it's a record fund for private equity: $34 billion. It's a record. But Q1 was a low point. No one even raised a $5 billion private equity fund because of the lack of liquidity.

That ties to Chamath's point: if you have liquidity in today's world, you're going to get the capital, right? Thoma Bravo had $30 billion in distributions last year. $30 billion. I think that exceeds most guests, right? $30 billion in distributions—not paper markups, but distributions.

1. DPI Versus Asset Gathering

Look, from my perspective, this is simple: there are 2 very different games that are called venture capital, or even private capital in general. One game is actually making people money: finding companies early, making the right bets, paying the right prices, and selling. That is a DPI game, and all that matters is DPI. That's the game I like to play. That's the game I think I value and respect.

There's also an asset-gathering game, and the asset-gathering game does exist. Here's the thing: I say this as a seed investor, with full transparency. As a business, the asset-gathering game is actually a better business. If you're just in it to make money, you're an asset gatherer. That's what the market wants, and we can talk about why and the whole nine yards. The public market wants that. They care about fees, et cetera.

The problem is, I'm just an intellectual snob, and I have no respect for asset gatherers. I think it's a stupid game. I think they're just 2 very different things that are called the same thing, and you should be really sober about what game you're in and what you're trying to do.

I think it's really fun that we had this conversation, and we ended up thinking that Chamath's on the side of good. Nice job, Sam.

I'm the first to say that I think it's wild that I agree with Chamath. My default instinct is to completely disagree with him on whatever he says, but in this case, I happen to agree with him.

I'm comfortable disagreeing with him. I honestly thought it was kind of a trite comment. It's a very typical Chamath comment. On first glance, it sounds smart, but on deeper analysis, it's kind of vaguely right but not useful.

Obviously, he cited his 2013 and 2015 vintage funds. Obviously, at that point, DPI is the only thing that counts. If you're 10 years in and you're still selling promises, then you're in trouble. So that's a trivially obvious comment. But to say TVPI doesn't count—the truth is, venture is investing in illiquid assets for 5 to 7 years with the expectation of making a greater return. That means that for 5 to 7 years, you don't have DPI. You've consciously un-DPI'd yourself. You've taken money and given it to them to tie it in the ground. In that period of time, you have 2 choices as an investor gauging these guys. You can say, "TVPI means nothing. I'll stick my head up my ass and I'll look in 7 years. I'll see how they're doing." Or you can use TVPI for what it is, a proxy, a loose proxy for performance.

You can just say nothing.

Yeah.

Okay, I'll be more direct. Again, we don't know each other. Here's the upshot: you have to understand that LPs are just incentive-driven.

From an LP's perspective, what you should do as a rational human being—if it were me investing in a fund—you'd say, “Yeah, the money's in the ground for 5 years, and after that we'll see where we're at. That's all that matters: what you deliver.” I know full well, as an investor, that the marks you've made up, or the marks that SoftBank listed something at somewhere, are completely irrelevant. In fact, they're a negative signal in a lot of cases. The marks are irrelevant and stupid.

Now here's the reality: institutional LPs are people too. There's some junior guy who wrote the check, and he wants to get promoted, and he exists in an organization that's trying to deliver something. He doesn't want to wait 7 years to get promoted for making a good call or a bad call, right?

So it really is just a marketing thing where you're saying, “Hey, I'm going to give the person who wrote me the check some marketing thing they can then use for their own internal purposes.” Because everything in life is about get laid or get paid.

If you had 2 funds 3 years in, and one of them had no markups while the other was a 2X TVPI, would you regard those 2 funds as exactly identical?

No. What I would regard them as—and this is how I actually do regard them when I look at my own portfolio or things like that—is very simple: how many credible things do you have that are going to be fund returners? Just give me the list of 4 things. I don't care where they're marked. You either have a set of true, legitimate shots on goal for important companies and important outcomes, or you don't.

If a company is marked at 2X where it went in, SoftBank or some crazy person marked it up and some crazy—

Same with Sequoia, but my point about the Chamath comment is to say it means nothing. To say it means nothing is a gross exaggeration.

There almost certainly is some signal in that data. Let's take it. Let's go right down into “TVPI means nothing.” I am willing to bet that if you got some kind of machine-learning algorithm and looked at all the funds 3 or 4 years in that had a TVPI of 2X, and all the funds that were 4 years in and had a TVPI of 0.8X, and then correlated that to the ultimate outcomes, I bet there would be data in that signal that says it has some value.

It's not the only source of data. You're right, there are better sources of data if you're in a position to evaluate better data. But as an LP, in the absence of anything better, there is signal in the data, which means the comment is wrong.

It would be fun to look at. I actually bet that firms that hold positions at zero or cost, or even are willing to mark down positions, are more honest about it in reality and might actually outperform. I don't know. We can look at it. But the reality is, I think there's this incentive. As an LP in many funds, you have funds, you get their statements, and you laugh.

Totally.

Because they hold everything at ridiculous valuations—the highest water mark they can possibly come up with—or these numbers that make no sense, and you look at it, you're like, “This is kind of funny.” But you're like, “Oh, this is not real,” right? I just think that it's because everyone has methodologies all over the board. It's all marketing.

Unfortunately, I hate being with Chamath. I hate it. I've known Chamath a long time—not well, but casually—and I have to say, it drives me nuts that I have to be. But with him—

Okay.

—all I care about early is: give me 5 names that matter. All I care about after you've had enough time is: did you make me money or not?

Okay.

But I get your point. It's a zero is an extreme statement, but we live in an age of memes.

2. The Hollowed Out Middle

We said we're living in this kind of dichotomous world, or this kind of binary world, of venture. In the middle, SVB did this great analysis, and they were saying that the middle of VC funds is getting really hollowed out. In other words, your mid-tier firms in terms of size are really falling apart. What happens to them? Is it a game of the very small and the very large, do we think, when we look at this report?

Yes, there's no middle. You can't be a billion-dollar venture fund.

Is a billion the middle?

What we do is target, call it, $200 million early-stage funds. That's what we do every few years. I'm very confident we know how to deploy that. I believe there's a market where you can, from a DPI perspective, make money on that. It's not a great fee business, but it's a great DPI business.

I don't know how you make multiples on $1 billion consistently in VC. I think once you're doing $10 billion, you're playing a completely different game of asset gathering and asset deployment where you no longer have the same goals. So I personally think that the $1 billion zone is the death zone.

Gotcha. Good to know as a $900 million fund.

Yeah. It's a tough one.

But no, actually, I did look at the SVB data. First of all, I was thinking, “Are they throwing me under the bus too?” I checked the data, and they were actually saying that $200 million to $500 million was the middle zone. So now I felt even more depressed. Now I'm a small behemoth, which is even more degrading. I'm kind of a low-rent conglomerate.

I think in the end you have to be sized for the stage you're playing at to achieve the portfolio construction you want to achieve, and I'm pretty confident that our typical check size is $20 million to $30 million in $30 million to $40 million rounds. We want to get 30 checks. The math works. Thirty deals total. So I think you have to be sized for the game you're at.

It still doesn't mean you can't get it wrong, and I do think, to your point, one of the things Jason and I have been batting around over the last 8 weeks is that I do believe the impact of the conglomerates has made everything harder, including making it harder for us to make money. I totally buy that. The existence of people with $10 billion to spend makes it very hard for people with $900 million or $500 million or $600 million to spend to do that rationally in a way that—and I'd love to hear from Jason on this—in a way that if you're putting out $100 million or $200 million, you're probably a little more inured to it.

Maybe the more interesting question is for founders. I think the SVB report, which I wrote up, I don't think it was really saying $800 million to $1 billion was the hollowed-out middle, but I guess it is part of the analysis, right? That they're not raising funds.

If that is true, then every founder wants to raise $20 million or $30 million in their Series A now. No one wants to raise an $8 million Series A; it's now 3 SAFE notes. So does that mean in a couple of years your companies outside of some scales are going to be limited to mega-funds? Are they the only people that are going to be able to write Series A checks, because $20 million to $30 million isn't even a large Series A today; it's a normal one?

So if there is no one in the middle, then we're all stuck with mega-funds to fund the C companies, right? I mean, I guess it's an obvious point, but founders just better get to know folks with $5 billion or $10 billion funds before demo day.

Or they're all going bankrupt and these are all bad companies. I think the other way to look at this is that I know a lot of people are pulling out of Series A entirely because they're like, “These are completely mispriced,” and, “These make no sense.”

I think the other possibility is that we're in an era where there are good companies to be built and places to make money. If you are going into corners of the economy or funding things that other people won't, and so you're an N of 1 or an N of a few looking at spaces that are really novel, there are places to make lots of money. But I think we also have to look at the other way, which is that there's just a massive amount of capital being massively misallocated right now.

It's always funny when people simultaneously have the worldview that the big funds are going to win everything, but all the money is being wasted. I mean, at some point—

Well, I think the thing that was in my mind is that I don't even know the big funds need to make that much money because they're asset gatherers. They need to—

Yeah.

—make enough money to generate gathering more assets, and they need to justify their own existence, and it's a good business. That's different from saying you're trying to make a lot of money in venture capital.

Broadly agreed. But I actually think, Jason, I want to go back to what you said, because I thought it was spot-on. If there are only 2: million-dollar funds and billion-dollar funds, then you're exactly right. Logically, those $20 million, $30 million, $40 million, and $50 million checks are going to come from only 10 names who are writing 40 of them a year, not 8 like us.

That's possible, but it's just a weird fund construction, because those people are then going to be writing 40, 50, 60 checks of this size. You're going to have a very spread-out partner base. I don't know if they'll be able to meet the founder needs where it is.

Is that not what your insights have today, though? I mean, we've mentioned before the outcomes that they have, with Hinge Health returning $400 million on a $6 billion fund, and them having hundreds of positions—

Yes. It is the construction they have. But at some point it gets easier when you're running big sums to put big money in a smaller number of companies than try and diversify away.

If we're trying with $900 million to have 30 As or Bs, someone who's running $9 billion, if they were trying to do the same thing, would have 300 As and Bs in 3 years. It would be silly. So at some point it becomes a part of their business, but not all their business.

And then I think Jason's right. It's so funny. We're going to talk in a few minutes about—and Sam mentioned earlier—this idea that a lot of As are struggling to be raised. The only way that worldview makes sense is actually what Sam said: if you have a world where people—can I call them mid-sized funds, quote-unquote—can't survive because of the big guys, and the big guys are, quote, “stealing all the good deals,” and at the same time we're also saying many companies are struggling to make a Series A, to get a Series A raised, those things are almost opposite to each other. One says capital is scarce and one says capital is plentiful and the big guys have it all.

Well, I would say actually the way you resolve that illogical-seeming statement is quite simple, which is: every generation, there's only a few companies that matter.

Yes.

Most of the money in Series A is completely wasted. And they might want it in larger amounts; that is kind of a war of attrition, right? It's just more money being lit on fire.

The mega-funds can win because they can say, “Okay, look, I'm going to plow a gajillion dollars at almost any price in,” and because of AI or whatever argument they want to make, there's no upper bound, right? And so they say, “We make lots of money just on lots of money being deployed,” and that can logically make sense.

Seed funds are fine because you still have to be in winners, but it's just a multiples game, right? If you're in it at zero, then you can make 100 checks and make the math work from a DPI perspective.

You also have a nice benefit, which I very much appreciate. I really strongly believe private-to-private is an important future, and I love being early and small and first because I actually can sell into the private markets in a way that you can't if you've written a $30–40 million check into the Series A. You're too big, right? So I think those 2 things survive.

3. Selling Early In Secondaries

What do you think about selling early—selling in secondaries, unicorn secondaries? How do you think about the goals? Is it 1× your fund, 2× your fund, half? Is it just risk allocation? I think if you have a smaller fund, it's really interesting—

It is interesting.

Because, just the other day, I had a 1× exit opportunity to return the fund once. One of my anchors was also involved—I brought them in as part of the deal, and we talked about it. They followed me, so it was less for them, but for them it would've been very high IRR and for me a 1×, right? That sounds good on the internet, but then it's gone.

Wow.

It's all gone. There's no more returns.

At the end of the day, there are a few things that matter. The number-one rule as a fund allocator is you cannot sell the things that matter. Rule 2 is this: we were one of the first investors in Allbirds, Astra, or plenty of things that got out and people were excited about, but the thesis broke. And so I think you have to be really honest with yourself. You're going to make mistakes.

I think one of the things I always say is that, as an early-stage investor, you get hundreds of shots to buy. You get really good at buying. You get far fewer shots at learning to sell, so it actually takes much longer to learn to sell well. But it's still as important as learning to buy well.

I think you just have to be really honest with yourself: Is this an infinity shot or is it not? The second the thesis is broken, or if there's a buyer at a price that makes sense because someone else has different fund dynamics—maybe they already own a lot and for them it's about rounding down their average entry cost, or maybe they're a net capital allocator and they don't care as much about the mega-return—they have different goals. There are opportunities that work for everyone.

Rory, can I ask you: if it's not one of the companies that matter, is it lighting money on fire in the kind of binary way that—

No. I'm glad you came back to that, because I would say, again, respectfully, I disagree, and it's a matter of degree. It is true in every decade you look back and go, the vast bulk of the value is driven by 1 or 2 companies. It's a power law. We all understand the math. To a rounding error, there was a decade where it was Google and then everything else, right? That's intellectually absolutely true.

So you could say to yourself, “If I didn't do Google, I must have lit the money on fire.” But then when you make that sentence a few times, you go and look at your Schwab account and say, “Oh, I have money in there, so I mustn't have lit it on fire.” There are more wins than the biggest win. It is a power law.

I would prefer to have done Google than anything I made money on from 2000 to 2010, but I'm damn glad I did that one, too. The over-extrapolation that only 3 deals make money, so therefore everything else at Series A is burning money, is true in the sense that you'd prefer to be in Google, but it's not the only way to make money.

In just the last couple of weeks, we've had, as you've mentioned, a couple of decent IPOs: Hinge Health. We're going to have Chime; Mountain was last week. We have another one this week. Take Hinge Health. Someone made $400 million for their investors. 20% of $400 million is $80 million.

I say this every time when someone at the office says, “Oh, it's just a 3×.” Every single one of you will cash the fucking check. Every single one of you. If I left them on the counter there, you'd all take them home with you, right? $80 million is still real money in America.

Totally. And, Rory, that's almost my point, but just to put you on Chime, Chime's a great example of this, right?

Yeah.

If you're an early-stage investor, you're very sad you didn't sell last round in Chime, right?

Yes.

And I would argue at $25 billion, wherever the last round was done, when you look at that rationally, you'd say, “Look, it's a good company. Is it an infinity company, or is this a really great place where I did my job, which was to fund it early, to find it early, to fund it when money was scarce? And now we have capital allocators who are making a different assessment?”

I think the answer is you clearly would've wanted to sell last round, and I think it's incumbent on investors—

I only know it's in the S-1, but I didn't see any of the early shareholders make the cut on the principal stockholders. When I look at principal stockholders, there's dilution, but I don't see any of the seed guys on the table. So that says to me they probably sold at some amount below $25 billion, when I literally only see DST Global, Crosslink, and Menlo.

That means they did their job.

Well, not if it's a $250 billion company.

But I think you're—

Hang on. There's 2 separate things floating around, and I want to disaggregate them. First is, I'm going to fight in defense of mid-tier. Are the mid-tier investors who didn't do the seed but did the $25 billion round happy they did that round? The round that Menlo did, I think, was out of the Series B or C. It was a couple hundred million. I can't remember what the pre-money was. They're damn happy they did that round, and that's a classic example of a mid-size venture firm making a savvy bet, doing good stock-picking long after the seed, but still making good coin.

To me, that's the first point, which is the validation that you can, at that fund size—probably it was a $400–500 million fund then—make really good coin and move the needle at the fund level.

Second comment to your point, Sam and Jason: you're right. You look back and go, “Hmm, I might have been marginally smarter if I had bought at $200–300 million and then sold at $25 billion, rather than holding and selling at $12 billion.” But the big advantage they have is they'll still book probably 10 or 20×. Now, 40× is better than 20×, right? So they've made good coin in a situation where, as we discussed before, the kind of $25 billion round is going to lose money.

So that's proof that you can make perfectly shrewd Series A-to-C bets, sub-$1 billion pre-money, and make very good money in a non—quote unquote—generational company. Chime is an extremely good company, but not a generational—

I agree with all that, Rory. And again, I'm obviously trying to be provocative on purpose, but I think, directionally, here's my point: it's going to bring it full circle to DPI, though.

Yeah. Yeah.

When you think about how I relatively value DPI versus TVPI or any other metric, if I'm an early-stage fund or really anyone, and you said, “Hey, I sold my Chime stock at $25 billion for American dollars—for cash—and I gave it to you years ago,” right? From an IRR perspective and a cash perspective, that is a great move.

Absolutely.

And so that's, I think, the thing to keep in mind when we go back to all of this: we're talking pretty deep cuts on the game of VC right now, as opposed to the game of company building. But I do think when you come back to this, this is why I believe the Chamath line—unfortunately, it's going to get clipped and I'm going to be upset about it, but whatever—he's right. He's like, in the end of the day, good investors, good capitalists make people money if that's the game they're playing and they're not asset gatherers.

In the end, it's true: cash in the bank counts. Sometimes, as you correctly say, it's even true that a smaller check earlier has more value than a bigger check later, just for a whole bunch of life reasons. I like what you said about Facebook. You said sometimes it's even true that a smaller check earlier has more value than a bigger check later. That's why we have interest rates. There's the time value of money.

I remember a friend of mine. We sold his company, and I was giving him grief. Then, 5 or 6 years later, another company in the same space went public at 2 or 3 times the market cap, and I foolishly gave him shit about it. He called bullshit on me. He said, “Look, I'd been at the thing 10 years. I made a lot of money. I got married. I have a life. I have a lovely house in Spain, and I started another company. Yeah, I'd have more money if I'd held, but that was my life choice.” And he was exactly right.

Just on Menlo Ventures: I obviously don't know what their basis is, Rory. Obviously, they'll make a ton of money on Chime, right?

Yeah.

A classic great bet. They said they did the Series B, right?

Yeah. They win.

So they did the Series B at $200–300 million. Maybe we can figure it out on the fly. There'll be some dilution. They'll make 10 or 15× their money, right? That'll be a fund returner or more than a fund returner, probably.

Just under. Maybe a little under, but yeah.

I do like that team. Back in 2021, they wanted to invest in all of my companies.

They're very high alignment, right?

Yeah.

What's interesting to me today, from the start of the conversation, is that all they talk about is Anthropic. And this is a smart team. It's everywhere. They have an Anthropic fund. I follow all the guys on LinkedIn. They didn't do the seed, and they're going to make a lot of money off this, right? Because Anthropic went from $1 billion to $3 billion in revenue in 5 months, right? But it's a sign of the times, isn't it?

Probably.

I don't see them trumpeting their coming DPI from Chime, which is epic, but Anthropic is on their social media 10 or 11 times a day. That is perhaps as it should be, but I don't think they're taking any DPI in that business.

Well, they're marketing to entrepreneurs. They're not marketing to LPs. Here's the way I look at it, honestly. I think in venture capital it is unbelievably difficult to know what anything is worth. And if you're Menlo or you're anyone in the middle, my God, do you have to price things properly.

If you're plowing enormous amounts of money and sitting late on the infinity dream, price doesn't really matter, which is how you get these ridiculous prices. The reason is very simple: what is OpenAI worth? Who the hell knows? It's like one of those things where people can dream, right? You can have a trillion-dollar dream, and that means that there's very little pricing discipline. It's basically a pissing contest over who can pay more or who's willing to go further. And at seed, nothing's worth anything, right? So you just have to have pricing discipline and remember that.

The problem with Menlo is I don't know how you know. I say this as someone who started my career at Bain & Company. I think with these private companies, catching the right price and the right dynamic and being right at Series A or B is the hardest game. I think some parts of that are true. I think you are right that the seed investor—

I always say, we typically do As and Bs, and I always say we're at the first point where you can start to do some analysis versus just people and market, which is what you have to live on, on broad directionality, right? You've got at least some pitiful facts to look at and try and come to conclusions.

No, I have no facts. I exist in a fact-free zone.

Yes, absolutely. And we live in a fact-thin zone. So I do agree. I don't agree that pricing for us is, quote, “harder than pricing for the late stage.” I think late stage, on average, is much harder to price, because on average you'll be wrong. And we're seeing a whole bunch of down rounds where the last round loses money, and every other round prior to that makes money. So I actually think price discipline is even more important the closer you get to an exit.

Well, they don't lose money. They just get their price down.

Well, we've thrashed that one to death. In Chime's case, they're going to lose money because they don't have the market share.

Well, that's a fair eye roll, Jason.

But I think the more important thing—the interesting thing—is that I think Anthropic, by the way, I think Menlo doing Anthropic was a genius move, right? I'll come to that in a second.

I think it's a genius move, too.

Totally. Look, there's always some deals, and it's very surprising that even at that $2 billion, $3 billion, $5 billion valuation level, there's still another 10× from there. Now, there's not many of them. There's literally 1 maybe every 2 years. But if you do that 1, you can price like it's an A or a B, but deploy money like it's an F and make a return—a 10×-plus return—as if it's an A or a B. It's an awesome deal.

There's, as I say, 1 of them every couple of years. Anthropic was 1. OpenAI was 1. Probably Android was 1. But the average late-stage deal isn't that, right? And there's just less degrees of freedom on the average late-stage deal to compound. So I think you have to be a little—

When I look at the people who do it well, like Meritech, like IVP, there's a fair amount of shrewd price discipline in what they bring to the table.

Or what the product they're selling is—just a different product. The product they're selling is access to this pool of companies that are late-stage for people with too much money or that need private access, with a cherry on top of infinity, right? And they're actually not in the business of making that.

I think that's probably not as true for the names I cited, but I think you're right. There are people who are—there are funds who are playing that game where it gets back to the whole idea that, when all the cute, sexy stuff is private, there are eventually going to be entities whose sole goal in life is to put public investors in contact with those sexy private companies.

The one thing I will say on Chime and the $25 billion is every single round in the billions for Revolut, everyone was saying, “Well, how much more can it be? How much more can it be?”

I think everyone is realizing now that Revolut will quite likely be at least a $100 billion company—

Yeah.

—and likely $150 billion. If you sold at $25 billion, you would be grossly underestimating it and missing $125 billion of gains in a European neobank.

It also just depends what business you think you're in.

Aren't you in the business of making as much money for your investors as possible, Sam?

No. Well, I think there's another real debate, which is a tough one. I would argue that there's a strong case to be made that, as a venture capitalist, you should not have an opinion about the public markets. You're not paid to have an opinion about public markets. People can have their own opinions about public markets.

You're paid—your job is to manage the private markets or private pricing. And then once something's public, it's like, look, the game should be over. Now, there are all sorts of people that juice that or decide they do have opinions about the public markets. But I don't know. It's something I've gone back and forth on, to be totally honest with you, about whether VCs should just do that, or whether you say, “Look, we can turn over shares. You guys figure out what to do.”

But I would probably sell Revolut and Chime right now. Right now. Both of them have about 15% market share of true TAM. If you're trying to decide when to sell, or whether they have infinite runway, market share is a limiter, right? There are folks that can get to 100% market share, but for a lot of apps, 15% I find is a little bit of a headwind.

You start to get to double-digit market share, and just when it gets good, because everyone's heard about you and all the leads come in, it's like, ah. 15% market share is a lot.

Well, that's a classic.

It's a lot if it's not a monopoly.

That is the classic thing. As an early-stage investor, people come to me like, “Well, the CAC is blank, but it's going to go down.” No, CAC only goes up, right? People just consistently get this wrong, right?

Almost always up at scale, right?

It's like this is how the world works. And the kind of argument that you're going to get good at something and all of a sudden it's going to get cheaper is just wrong.

4. The Companies That Matter

So is Chime an important company? I just don't see it. It's a good company; they provided some banking products to unbanked people or underbanked people. I get it. It's not bad. They did a good job. They built a good app. But is it important?

If the $14 billion market cap was in my bank account, I would think it was a very important company indeed.

It's fine. It's not a bad thing, right? It's just—

If that is the case, if Chime is not an important company at all, then I've never done anything important in my life. I've been a decent founder. I've seeded multiple. My team has gone on to help run multiple decacorns. I've had $5 billion cash exits, but I really don't matter. I think I've helped thousands of founders build companies from scratch, but I probably don't matter if Chime doesn't matter.

I think the default is that most of us don't matter, and most of the companies in the world don't matter. But the things that matter really matter, right? And I think the job—

Right.

…is to find the things that really matter.

But do you think Salesforce or Oracle matter?

Arthur Balfour, Prime Minister of England in the 1900s, a very languid, relaxed man, used to say, “Nothing matters a lot, and very little matters at all.” And that's really what you're saying, Sam, and it's true, but of course the correct response is, what am I meant to do with that?

I think an Anduril can matter. I think a Tesla can. I'm not saying it will.

I do think killing people matters. It's very impactful.

I think that OpenAI could matter.

Microsoft, Facebook, Google—

You know—

These are companies that mattered. I think Bitcoin matters. I personally think Solana matters. I think Venmo might matter. We can go down the line of things that are actually paradigm-shifting and have an impact, and then everything else is not. You can build good businesses, but it takes a long time working for them.

So you don't think—you're not sure OpenAI will matter. You just think it's possible it could matter?

I'm not sure it will matter. I think it might matter.

That's a pretty arrogant thing to say, isn't it?

Why? The game is young. They clearly pushed the ball forward, but will we look back in 20 years and say, “OpenAI was a fundamentally important company”? Maybe.

I think the truth is most companies fail. Some companies compound to $1 billion in value. A few each year compound to $5 billion, 1 every year to $10 billion, and 1 or 2 every decade to $100 billion. So, taking away that quote, “matter,” I think market cap is a rough proxy for mattering. I say that pretty confidently because I believe the capitalist system works. The things that matter the most are valued the most.

I know.

What I'm disconnecting with a little bit is—and so, I love what Jason said. I just wanted to come back to this: I love what you said. I'm at peace with the fact that if the definition of mattering is $100 billion, I may well complete another 10 years in venture, do it for 40 years, and, quote, never matter. I'm actually okay with that. I'll take it up with my therapist. I'll be fine.

But I know I'll have invested in perfectly good companies that have built real value, that create perfectly good businesses that, in some cases, are worth $5 billion or $10 billion, and that's okay. The problem with only doing the things that matter is, in the end, it tends toward some kind of nihilism, which is that nothing else matters.

I do think that people should wake up every day and have the opinion that they want to do things that are going to matter in the world. I'd say the most insulting thing you could ever call me is a market participant. I find that incredibly insulting.

You're like, “Yeah, it's a pretty efficient market. Capitalism is pretty efficient. You pay $150 billion, $120 million, who cares? It's fine. You have a billion-dollar company that does some stuff, and so, fine, I'm glad that happened. That's capitalism working.” But then you're really just a cog in the system, right?

I think that's not a recipe for outsized returns, but it is a recipe for doing fine. It's just not my goal.

I am a cog in the system. I might want to be more than that, but I think—

No, but you're a hollowed-out fund that doesn't matter.

Harry, you should have asked Aaron Levie when you interviewed him yesterday if he mattered.

Yeah.

Sure, but—

I don't think he matters. I don't think Box matters. I mean, he has a—

Aaron Levie? Oh, Box. I love Aaron. He's a good buddy.

It's less—it's nowhere near OpenAI.

He—

I don't think Aaron Levie would say Box is more important than OpenAI. I'm confident he wouldn't.

For what it's worth, I love Aaron. He's a good buddy.

But his company doesn't matter. It's only $4 billion, $5 billion, $6 billion. It's irrelevant, right?

I completely agree that his company doesn't matter. In fact, one of the things I told him when it went public was—

So you think he should have just shut it down tomorrow, his $5.5 billion company?

No.

Should he leave the keys for some random person to come run it from some hedge fund?

Aaron is very excited about AI, if you didn't know, right? And I think part of that is—

I did know.

There's an opportunity for him and for Box to matter, right? Because Box is—again, he's done an incredible job. I'm very pro-Aaron Levie. What I said to him when the company went public was, “Look, one of the things I most respect about you is you just worked this to make it happen,” right?

This was not from a first-principles, important-company perspective. This is not a sexy company, but you just ground it out, and I really respect that. I think that's an incredible thing to have done.

But he's wasting his life. I mean, Aaron and I started together, so he was 20. If his company doesn't matter—

Well, no, he's living a great life.

I don't think his company does.

I don't think his company does—

When he started, he was a wonder kid. Now he's the greatest CEO I know. He's wasted his whole fucking life, is basically what you're saying, because it doesn't matter.

He's not wasted his life, but I do think that—

He could have sold his company to Citrix. Rory was there. He could have sold his company to Citrix and made $100 million with retention payments—

Right.

—in 2008. So, according to you, he wasted the last 16 years of his life. He fucking threw it, flushed it down the toilet by doing a B2B company.

He made less money at the IPO than he would—

Flushed it down the toilet.

—I would. But look, I do think there's this thing going on right now—

Wasted his life.

—where there's a lot of people—

Wasted his life in Citrix. Wasted his fucking life, the poor guy.

For what it's worth, I think—I don't know, I think you probably know the math, but my sense is that he actually probably would have made more in that deal than ultimately going public. Now, I don't know where he is now, but from that, I think he—

Cash in 2008, plus his shares in Stripe and Gusto and others. Yeah, he should have just founded an Allbirds instead of doing Box.

If he were a fund manager, he definitely should have sold then. He's not. He's a person, and he wants to run it. That's fine.

The second thing I say is I do think there's a lot of exuberance about AI right now from a lot of people who are in these positions of, “This is a company that's kind of been meh, but maybe there's an angle to not be meh,” right? And that's exciting to people. I think that's awesome for organizations to try, but I think it's the race to matter that makes it interesting.

That's why I think the AI stuff is so interesting in so many cases. Some of it's real, some of it's not, but there is this new opening where people are like, “Oh my God, does this now matter?” That's kind of the big game to play.

Anyway, guys, with that—

Yeah.

I'm actually heading into my partner meeting, so I'm going to drop.

Right.

But that was fun.

I hope I have my therapy session set up for tomorrow. I don't matter. I'm hollowed out. But I'll get through it somehow.

I'm sure you will.

As my wife says sometimes—

You will. Yes. Cry—

“Go get some Allbirds, Harry. You'll feel better.”

No—

“Go to the mall at Stanford and get a few pairs of Allbirds, and you'll feel like a winner.”

Cry into your bucket of money. Okay.

5. The AI Infrastructure Reckoning

The thing that I do want to go to now is actually a phenomenal report, which, Jason, you did brilliant work on: Mary Meeker's AI report. I want to go to it, though. You posted an incredible thread—

Yes.

—with 10 fantastic takeaways.

Yeah.

I want to hand over to you on which takeaways you found most striking and why, and just start there.

Some of it was obvious. The first one I started off with—which is obvious, but I'll tell you why I put it in there anyway—was that ChatGPT is the fastest. Even though Sam Lessin said he's not sure it's important, it's the fastest gain of users in the history of the world: 0 to 800 million in 17 months.

Yeah.

It took Netflix 15 times longer. It took TikTok 5 times longer—the rate of adoption. I talked to so many folks, especially in B2B here. They're like, “You know, I don't know. I don't know that AI can really replace a good, hardworking marketing manager or a good PR person.” Or they say, “There are too many...” You know what they say, Rory?

What?

“There are too many hallucinations.” To me, it was a reminder of the rate of change.

Yeah.

I don't think, as humans, we can process—

Yeah.

—the rate of change of AI. It's so fast that we don't understand almost how quickly it's changing. Whatever you think AI could do 3 months ago, it's super dated.

The second Mary Meeker takeaway showed that infrastructure spend is unprecedented in the history of the internet, right?

Yeah.

The Big 6 spent $212 billion on CapEx.

Yeah.

And, yeah, the top line's great, but the spend here is insane. I'm still a little confused about where it's going because it's early at the application level. At the application level, is this all going to coding and support? Is it all going to custom data analysis? Is it all going to subsidizing our ChatGPT $20 a month?

I just don't know where all this infrastructure is going. I think it's laying a foundation that is transformational.

It feels so far ahead of the application level.

I think you're right. My takeaway from her thing was that the hyperscalers have taken very good, cash-efficient businesses and broken Mr. Buffett's heart because they've turned them into CapEx hogs, right? Microsoft, Amazon, Google, and Facebook have all had CapEx as a percentage of free cash flow go up significantly. Now, as she pointed out, the good news is that, in the last 4 or 5 years, free cash flows themselves have grown, so your cash flow hit hasn't been enormous. It's only been about a 10% decline in free cash flow, right? It hasn't been horrific.

In a less generous economic environment, the shareholders of Microsoft might well be screaming and saying, "Wow, $60 billion and nothing to show for it, guys. Come on here." It's been perfect timing. It's actually a little, in that respect, like '98 and '99. The overall economy has been favorable, and the performance of the hyperscalers' existing businesses—not their AI businesses—has been pretty damn amazing.

As a result, the market has been able to say, "Guys, knock yourself out. Spend $260 billion. I'm sure something will come of it soon." The $600 billion question is: Where are the apps? Since then, it kind of implied, though it didn't say it, that there was going to be a correction. Obviously, since then there hasn't been, right? Everyone is just plowing on. They're investing the capital, and the revenue's coming.

But relative to the spend, if someone said to you, "You can finance a $600 billion CapEx business, and 4 or 5 years in, you'd have $10 or $15 billion in revenue and $400 billion to $585 billion in losses," you'd probably say, "No, I don't need that right now." Right?

Yeah.

But that's where we're at. We say that on the revenue side, but then I think by the end of next year, OpenAI is going to be at $25 billion to $30 billion.

Yeah.

We saw Anthropic cross $3 billion in revenue, up from $1 billion just 5 months ago. That's 3× growth from $1 billion in 5 months. It always takes time when we're laying infrastructure. Revenue comes later.

The variable that we're missing in this discussion, in my view, is time. I totally believe that all the apps will come to fill the space available for them with all this CapEx.

Yeah.

I believe the LLM guys will make amazing shit. It'll be impressive, and businesses will find a way to use it. But to Jason's point, I think if businesses adopt it in 2 years and then the revenue comes quickly, it'll all be fine. If businesses take 4 or 5 years to get there, you could be looking at a more extended period where you're having the costs but not yet the revenues. I think that's the as-yet-unknown: how quickly will app revenue fill the gap?

When you look at the percentage of free cash flow, you said they're at 10%, and investors are actually okay with that.

Yeah. Yeah.

There's a ceiling to what they're okay with. If it is 5 years out and that is now 30% or 40%, is there a ceiling to what investors are okay with Satya, Larry, Sergey—you name it—spending?

Yes, there is—that's the answer. It seems inconceivable today because everyone's like, "Rah, rah, go for it." But public company investors are just mean VCs on steroids. We turn on a dime from, "I can't believe you're not spending more," to, "What the frick do you mean you're spending so much money?"

We've lived through it in '99, where you'd call the team in and say, "Double the burn." Without blushing, you'd call them in 6 months later and say, "Why are you spending so much money? Let's cut." If the economy slows down, or if your core growth rate starts to decline or slow down significantly, I think you'd see mild to reasonable levels of pressure on all the hyperscalers. An interesting example would be in a different market: Meta changed its name and wanted to go all in on VR.

They're 2 years in, and there's no return. There was a little bit of pressure, and I think Zuckerberg heard it and pivoted nicely. I mean, he's still spending a lot on virtual reality, but not as much. There was a little bit of belt-tightening there. When the pressure comes on, I think CEOs respond.

I said this when we chatted last time: one of the things I admire about Sam Altman—some things I don't understand, but admire—is how he's telling you the future, even if you just listen.

Yeah.

When he started talking about Stargate, saying that he alone needed $500 billion just to get going, and up to $5 trillion, we were kind of—I was like, "What's Stargate? What do you mean, $500 billion?" But he's socializing the whole market to the idea that it's going to be an order of magnitude bigger, right? He's socializing the market here. This isn't Amazon.

Mm.

Not only is the size bigger than Amazon's classic investment losses, but you need everyone to go all in on this. You need all the big participants to say—

Really?

—we're going all in on this big bet. It's not just 1 company making a big bet. This is all the big companies making a massive bet.

I think the interesting question is that there was a time when Amazon stock was perilously low, and a whole bunch of banking analysts were sneering and saying they'd go bust. Obviously, in retrospect, they didn't, and all those banks did. Jeff Bezos got to say, "Neener, neener."

Intuitively, you just know that when you're spending half a trillion dollars in advance of revenue, at some point in the next 2 or 3 years, there'll be at least 1 scary moment. That could actually be an interesting opportunity to invest, but it's hard to believe it's up and to the right from here.

What is a scary moment? Can you just help me visualize that? When we see the commoditization of models, they're all becoming very efficient, and they're all becoming really bloody good. What is a scary moment?

Maybe if OpenAI just misses a growth plan.

Yeah.

If it's just 30% or 40%, if they have 2 rough quarters for whatever reason—a sort of saturation we're not anticipating—that would set off a mini panic, I think.

Agreed. I was just looking at their numbers. It's interesting: their numbers are exactly 20 years apart from Google's numbers 20 years ago. Literally, 2002 is the same as OpenAI's 2022. 2003, I think, is about $1.3 billion for OpenAI, $1.4 billion for Google 20 years ago. This year, 2005—uh, 2004, sorry—$3 billion and $3 billion, so roughly the same. Now what's interesting is that the projections for OpenAI, still unrealized, start to pull way ahead of Google. In other words, OpenAI was the Google of its day until now, and it's projecting to become twice the Google of its day for the next 4 years.

Maybe it will, because I think it is a more impressive piece of technology when you use it, but maybe it won't. Let me make the statement clear: if all OpenAI is is as good as Google, then it's going to miss its number next year by about 40%.

That's amazing.

That would be a scary moment. I'm simply saying that the expectations are so high that anything less than freaking amazing could feel like a fail, even though in any logical terms it isn't. Just as Amazon wasn't a fail in 2001 or 2002. It was, in retrospect, minor growing pains that a panicky market interpreted as, "Oh my God, they're going bust." No, they're just building a business where it's going to take $3 billion to $4 billion of losses to get to cash flow break-even, and then you're going to make billions every year.

Jason, just to be clear, we're going to do a visual of Rory looking mean with "OpenAI doesn't matter." No one else in the frame.

You're such a dick with your captions, Harry. I just want to say that, and you'll get me into trouble, but I've given up.

Oh, you've got to do it. You want to do the third point because it's kind of interesting—the shit I brought up?

Yes.

You could have opinions on Elon Musk, but now that he's out of government, man, the guy is direct, right? He was direct today. He said this spending bill is ridiculous for all parties. And he's been talking about China with AI.

Mary Meeker's third point was that we're missing what's happening in China, which is natural. We're not even connected to them on the internet. We have a different internet. Her point was that 6 months ago, we thought DeepSeek was going to change the world. It didn't. But it had 93% of the performance of OpenAI's o3-mini at a fraction of the cost.

Alibaba, which we can't even use here, outperforms both. Baidu ERNIE, which I don't even know, costs 0.2% as much as GPT-4.5. People smarter than me are worried about China and AI. That was kind of her third point, and we've stopped talking about safety. I think we've stopped talking about China.

I mean, the simpler point is probably some version of: I think ChatGPT will be the Apple-level-quality product, but there'll be a whole bunch of Android-type-quality products out there that will just keep pricing honest if they're commercially available universally. I don't think, to Jason's point, that we'll be accessing them here if the models are built in China, even if we were legally allowed to. My guess is corporate America wouldn't want to.

The meta point about DeepSeek was that it was possible to get quite close quite cheaply, and you're going to see that independently of China, just in terms of competition here. But I think the undeniable fact is we ain't a monopoly anymore. It's not 2 or 3 companies monopolistically producing LLMs. There are 4 or 5 companies on the other side of the pond cranking them out.

Jason—

Yeah.

Were there any others that you think were specifically relevant for B2B?

Maybe just 2 more, and then you can shut me up. The second one, which we all know, but I think it's very useful to see simplified and compressed into a chart, is that folks still get this wrong in B2B today: token costs collapsed 99.7% in 2 years. Now, if you're a developer, if you're sitting in Windsurf or Lovable all day, you get this. It's not this simple because you're consuming many more tokens, right?

Yeah.

But all these ill-fitting blazer-and-jeans SaaS guys who are saying it's too expensive—

Oh, man.

—or the mediocre VP of Engineering who doesn't code anymore, who hasn't coded in 5 years, who is constantly telling me why AI's too expensive and doesn't work—find a new job, right? If tokens collapse 99.7% in the next 2 years, you can tell me, Rory, but I don't know where they're going to be in 2 years. It's at least going to be an exponential decrease in absolute cost per token, right? We'll use more, right?

Training costs will go up. But even today in June, I just want to jump off the roof when I hear a VP of Engineering who doesn't code anymore saying it's too expensive to do enough AI in the product. “We'd like to do this, Rory, at our B2B workflow company, but we can't. It's too expensive, Mr. Board of Director.” I just want to tell—I’d tell the CEO to fire that guy.

Yeah, you should pummel him to death. Because you're exactly right. There are some things I'm angsty about in terms of the road ahead, in terms of the economics of it, but just from a raw knowledge—intelligence, for lack of a better word—is just going to get cheaper and cheaper, faster and faster. Intelligence is available at a price that's declining by log orders of magnitude every fricking year. That's just huge.

As I say, on one of my boards, I have a wildly smart—call it a core AI model—as a fellow board member. It just repeats over and over again: “You can't do it today; build a product. By the time you've built it, the stuff will be cheaper, it'll work, and if it doesn't work, then it'll work 6 months from now.” That's the big trend that you're all leaning into. The price of a unit of intelligence is plummeting every month. Therefore, anything that you control with intelligence will get cheaper every month. Just lean into it, and the models will make it happen.

I call it the AI slow roll, and I think—

Yeah.

—it’s the number 1 thing killing B2B companies: the AI slow roll.

Yes.

We're talking about it. We're going to roll something out in Q4. We're going to do a limited release—

But—

—in Q4, Rory, and then next year, if it goes well, we're going to roll it out to a little bit more of our base.

That's a—

They're just going to be slaughtered.

That's actually a fun discussion because it is so hard to watch companies turn on a dime and push towards the new thing, right? We were talking when Sam was on early. I give Aaron huge credit. He's like, “We're going to do this. It's going to be top-down.”

A quintessential example is Facebook, just after the IPO, deciding mobile is it: strap ourselves into the office, and we're just going to get it done. It takes top-down leadership. And I agree, if you don't do that, and if you don't do it well, you're just going to be slow toast over the next 2 years. It's just hard to do. It's just hard to—

It's hard—

—shake companies out of their inertia because you'll always feel that doing a little means at least you're trying, so that gives you points.

That's the problem with startups: thinking they get points for doing a little, right? Here.

Yeah.

I thought it was fantastic. I had Varun, the founder of Windsurf, on the show. It released on Monday, and he said, “Listen, startups beat incumbents because of existential dread.”

Yes.

If you are in a startup and you don't ship great product that converts to sales, you lose. If you are in an incumbent—

Yep.

—and you don't ship great product that converts, and you're a great engineer, you're definitely getting reassigned and reallocated. You're not going to lose great engineers. You're just going to put them on different products. See it all the time in—

That's the problem with B2B today versus where Varun is. I don't see enough existential dread in most B2B startups. I don't see it. Aaron has it. Yamini had it when she was on stage.

Yes.

Aaron and Yamini at SaaStr this year, and Rory at Scale Venture, all said the same thing: “We're so excited and we're scared.” I don't see enough existential dread in B2B startups. I don't see it. I see a little bit of it in some discussions. I want to walk into a board meeting and see a little bit of shaking because there's so much going on in the world, and they just had yet another hackathon this last weekend, and they're rolling out AI voice agents on Wednesday. If that's not you, I honestly—

Got it.

—think you're going to fail. You want dread. Dread. Varun's right, because Varun's had 3 different companies in 18 months, right? And every month he's at risk of being displaced. Every month, right? He's honest about it. He was honest on 20VC, right? We need more of that in B2B, not this slow roll.

So the other point she made, which we knew, but I really think is super interesting for B2B, is this. A lot of the stuff she said, like pricing, is worth discussing, but this is one to slow down on. And Harry sometimes jokes that he's old, but this is where Harry is old, for real. Her point was the next 32% of the world is just coming online right now. They will all be AI-first. She says they'll use voice agents, agent-driven interfaces, and natural-language interactions. I think that is 2025.

What I do know is that this generation will not use the internet remotely like we do. They will not use leads, contacts, and opportunities. They will not use files. Files are disappearing overnight. Kids do not know what a file is. It does not matter. Everything will be MCP'd or AI'd.

This is an existential threat. If I'm Aaron or Yamini, these are S-tier SaaS companies, I have to worry about AI today, right? I have to worry about agents. I have to worry about automation. And then I have to worry about what happens when no one uses these applications remotely like they use them today. You really think that the next generation is going to even understand what a Salesforce UI is? It won't even make sense to them. They'll never use it.

My son helped us at SaaStr Annual this year. We have a SaaS platform we use for tickets and stuff, and—

Yep.

—he's like, “Wow, this is the first time I've seen software like this.”

Wow.

And he codes every day.

I do think that is true. I think in front of every system of record, there's going to be some system of work that does it, and it'll sit on top of the system of record for now. If it does a good job, it will gradually displace, make irrelevant, or replace the system of record, if they allow that to happen.

The AI that's closest to you is the worker. If I'm the worker, if the thing that I interact with every day that helps me do my job, over time, that just becomes the most important thing in my life, and anything behind that doesn't matter. So if I'm a sales rep and I have a lot of automation—voice automation, AI SDR, whatever—telling me what to do, doing things for me, where I'm acting as a controller of the agents, that's just going to become, mentally, my model of what a CRM is. And back behind that, yeah, there might be some database called Salesforce, but I just won't care.

Yeah. And there might even be multiple ones—

Yes.

—you won't care if the company's using Salesforce and HubSpot and Attio and its own database.

No.

You won't care.

No.

You won't care.

I think smart companies can reimagine what those workflows look like and build to it, and that's obviously what they're trying to do. But if you fail—if you're not relevant in terms of how your next generation of workers use AI—you will eventually be displaced. I mean, it might take a long time. I think, as I've always said, Salesforce instances will be there at scale long after I'm dead. But the value accretion will all go to the technology that's helping me do my job, not the technology that's counting, keeping score on how well my job's been done.

6. MCP Changes The SaaS Stack

The one I'm really struggling with today, and I have asked Aaron and others this, right? It's early, but I think MCP is an existential threat to almost every SaaS app. And now you can MCP into Notion, you can MCP a little bit into HubSpot, but it's very limited, right? You can MCP into Google Calendar. I just don't know why, if I can just talk to my AI—my Claude or ChatGPT—and do everything I want to do at HubSpot elegantly, I don't know why I would ever log into HubSpot or ever even learn what HubSpot is.

That's Harry's perspective, though, isn't it, really? In an agentic-first world, or an agent-first world, you basically have all of these applications that become databases, and then agents basically become the data transporters—and you never need to engage with the core database then.

Yeah. I thought it was too nerdy when he started saying it, right? And too technical. Too Microsofty. I thought it was too Microsofty. But time goes by, and now I can see MCP applications just starting. The problem today is you need to—listen, I'm not an expert, but everything that I've tried, you need a key.

Yeah.

A key is annoying, right? I have to go to some website, get a secret key, and add it to my thing. But when those keys go away and I can just talk to all my apps through my ChatGPT or Claude, man, I just don't think we're ready for this world.

I'll give you a personal example. Harry knows there's a company I invested in that I love called Mangomint. It's next-generation SaaS for spas, doctor's offices—

Yep.

—and the like, okay? They're coming up on $25 million. I love the CEO to death and would do anything with him. He was all over MCP the day he could use it, okay? And he's like, "Here's the problem for me. Let me be clear." He's the best in his space. That's why they were able to do it. Not a huge TAM, lots of issues.

He's like, "Now, let's say I want an appointment at Watercourse Way in Palo Alto, right? Where Harry goes to decompress after a tough pod."

Tell it.

But right now, he might have to figure out which application they use. Do they use Mangomint? Which of the various applications do they use? MCP can just abstract it away, and you'll have no more relationship with this application. He's like, "I could become a pipe overnight." And he's already completely changed his value proposition because of that.

If I don't change what I'm doing today, no one will even know which of these applications they use. And Harry's invested in some adjacent companies, not in this space, but you might not know. You might not know which restaurant application it is, which spa application. It might not matter. It's very difficult for me to imagine you can increase your prices in that world.

Maybe you can increase your prices.

Sam, I can tell you one thing, dude. If you're doing SaaS for spas, you definitely don't matter according to Sam Lessin.

No. I mean, I wouldn't even—yeah.

Restaurants, you're just not even in the conversation, dude. We're both fucked.

Yeah. Doctor's offices, spas. But he was, in a good way, an instant wake-up call. He's like, "If I don't add even more value, I will be partially obsolete because people won't even know who I am anymore. They'll just use AI to go out and pick it for them."

Anytime anyone starts talking about agentic this and that, I start to roll my eyes. It's too nerdy. The real world doesn't think about agentic this, but the world does think, "Can I go to ChatGPT and get my haircut?" Harry's upgraded his haircut for the show. It looks good, right?

Thank you.

But why can't my ChatGPT just figure out where the best haircut in the Mid-Peninsula is and do it for me? Why do I have to do these apps?

The closer you are to what the customer or the user wants to do, the more right you have to swallow anything that's behind that, and that's really what you're saying. Do you want to be the person who keeps track of who's coming to the spa, or do you want to be the software that helps decide who's coming to the spa? I want to be the decider.

Yeah.

I don't want to be the tracker, because tracking on its own is inherently just not as valuable a thing to bring to market. The long-term trend is clear here.

It's very interesting that Dharmesh is on the bleeding edge of this for HubSpot, and I think HubSpot is threatened by MCP. HubSpot is a system of record too. It's a very valuable one. Does HubSpot really want everybody abstracting away all their structured data and using it however they want in their own applications, in their own CRMs, in their own marketing applications?

The incumbents will last a long time, but that doesn't matter, because what matters to us is where the next new dollar goes. We're all about the next $100 million of new ARR coming on, and it's hard. That could comfortably go to these next-generation products, while at the same time SAP can compound for the next decade happily. But they won't be getting the new money.

Our job is to find the companies that get the new dollar. You don't have to have a world—and one of the bigger ahas for me has been that, quote-unquote, "last-generation companies" can compound happily for years on their existing base, on expanding within those customers. While at the same time, effectively, you're signing new customers who are saying that that old thing is not where the world is going anymore.

But it just has such momentum, it just keeps on compounding. That's okay for that as a public company. For you as a venture investor trying to find the new, new thing, it's all about the much smaller dollars in budgets that are available for the new stuff, because that's the only money we can live off.

We can't take the SAP budget for our new product for the next 10 years. We have to take the new companies or the new functionality that sits on top of those apps, be it SAP, Salesforce, Oracle, et cetera. And it's hard to imagine in 2025, knowing what we do about AI, saying, "I've been thinking we really should build a screen just like the other SAP screens." That's not where it's going to go.

Every company's market share is up for grabs when there's a platform shift. The very biggest ones don't go to zero; they just slow down. But you're right, the mid-tier ones can really hit a wall. And if you're running one of these companies and you're not afraid, and you're not 110% focused, you're almost certainly going to fail. The only question is over what period of time. It turns out making money is hard; keeping money is harder.

7. Public Markets Reprice Venture

Guys, can I just roll to one thing we said about not being a public-markets investor? Everyone loves the analysis we do on your Chimes of the world that we have done in the past. There have been a lot coming out in terms of M&A and IPOs. There's Chime, obviously, this week. There's Groww, which filed for an IPO. It's an Indian IPO. Shein is moving its IPO from London to Hong Kong, maybe finally getting that one out. Omada Health is coming out at a $1.1 billion price. Circle filed for an IPO at $8 billion. Salesforce acquired Informatica for $8 billion.

Which one do you think is most interesting?

I'll hit two. I do think Circle's interesting just because it's a fun business model. It's basically a Bitcoin-enabled money market fund. They get your money, they're able to invest it in treasuries, and they get a nice yield. They have to share some of that money with customers and distribution partners, and they build a perfectly nice business there, right?

A valuation at around $18 billion, plus or minus, is reasonable. It's just an interesting business. It's like a $43–$44 billion money market fund where they're getting roughly 4% or 5%, so just under $2 billion in revenue. Then they have to share a lot of that with distribution partners, and they have a net interest margin. It's just like a little mini bank-type thing.

Then they have OpEx, and they make a couple hundred million bucks a year. Perfectly good business. It's the boring version of crypto. Crypto, but safe. I think it'll be an interesting IPO. I think it's bounded in terms of value because there's no magic explosion. It's very clear how the model works. Anyone financial can figure it out and value it, but I think it's interesting.

To me, the most interesting, although I'm still trying to fully understand it, was that Snowflake bought probably an almost pre-revenue startup called Crunchy Data, which is a Postgres implementation, right after Databricks bought—

Yeah.

—its own Postgres database company, Neon, for $1 billion. Okay? I assume Neon had some real revenue, right? $250 million for Snowflake suggests to me they're buying some distribution.

But trying to understand, in this conversation, how AI shifts the battles that Snowflake and Databricks now need to be database companies at some level, in a year or two, could be a fundamental change, and all these lines are blurring. Instead of Snowflake figuring out how to work with all these different Postgres providers, open-source and forked versions, and all the different databases, now they're database companies.

Yes, they need their own custom databases so their agents can efficiently work on all the data that's been baked and stored here, but it's pretty interesting: within 60 days, including one yesterday, they both became database companies. It's a big change.

What it really is is feeling the need to support all the different types of data structures and, as you say, morphing from a world where typical Snowflake data was data-warehouse-type data to now the kind of data used for AI, so feeling the need to support a wider variety of data types and data structures.

Yeah. I nearly said, “Those are the two publicly traded companies,” and then I remembered Databricks isn’t publicly traded. It’s just as if it was a public company, and it should be, of course, and it will be. But you’ve got Snowflake as public, you’ve got Databricks, and then, Harry, you’re gonna ask about it in a second, but companies like ClickHouse, which is an exploding company in the database space—a columnar database—are really seeing explosive growth and just raised a big late-stage round, I think around $6 billion from a host of who’s who.

The AI plumbing business is a great place to be, right? If you’re even adjacent to it, I would argue that 5 or 6 years ago, Snowflake wasn’t in it; they were adjacent to it, so they’re trying to fearlessly move into it. I think the Databricks advantage was that they were in it from the start. It was kind of their thing: moving data and then ultimately using it for AI.

So venture guys make really good thematic bets about the next 10 years and are pretty horrible at assessing how the market’s gonna react to the next quarter’s data. Hedge fund guys are the exact opposite. Some of them, the overnight traders, don’t even wanna think about the big-picture trend. It’s just, “Is this EPS number over or under estimate, and will the stock go up or down?” That’s just a different skill set. I’m no good at that.

Proof that we’re all no good at it. It’s a fun thing you can do when you have a public company. You’re on the board of a company, and you have some interesting results you’re about to announce. Some things are better than expected, and some things are worse. I’ve seen this over and over again.

You can go around the boardroom and ask people, “Here’s what we said. Here’s what we’re going to announce tomorrow. Will the stock go up or down?” At best, 50% accuracy. I’ve seen it over and over again. It’s like, “Hmm, we think these things are great,” and then you go out and everyone gloms onto the other thing.

Sometimes the stock goes down and you thought it was going up and you wanna cry. Other times it goes up and you’re like, “Hmm, okay.” It’s kind of like 2 worlds talking past each other sometimes.

8. The YC Ownership Problem

Can I ask one more final one for a Kalshi quick fire? I see, obviously, the new batch is out this week and they’re raising. Everyone’s doing 10% dilution for the round total. You guys are much more experienced than me. I’m very lucky to learn from you. I appreciate our relationships, really.

How do we play in a world where the total round is 10% and our fund sizes are bigger? Really, you’ve got 7.5 for a lead and 2.5 for angels. Jason, how would you advise me? You do YC very well.

Well, first, I think at YC it’s not 20. I think the new deal they’re trying to do with a lot of startups is at least 50 to 60 post, even for pre-revenue. There’s a ton of deals in this batch at that level. So that’s what they’re pushing toward.

No criticism. You’ve got to see YC as a business or this stuff will drive you nuts. You have to admire the game that’s being played on the field, right? When YC went to 15, people thought it was crazy. Then 20, and now they’re trying to get everyone to 50 to 60 if they’re AI, if they just have AI—which is 70% of the batch.

Plenty of folks are raising at 50 to 60 or more with essentially no revenue. But it doesn’t change the ownership question, and it does make it all even harder, right? Because then the A has to be at 200 for the math to really pencil out.

This is what I thought investing was like: for seed investing to really make sense, the next round should be 3X to justify the risk. And for A or B, it’s gotta be 2X or you should just wait. You should just wait for the next round, because it’s not that simple, but intellectually it seems about right.

You can either play that game and buy as much as you can, or you can cry about it on the internet. All I can tell you—my little analysis of YC—is that I’ve ended up owning a little less than half of what I would’ve. That’s one guy’s analysis, because I bought more in RevenueCat—I bought a lot more in RevenueCat—and I bought more in Algolia and more in others that have done reasonably well.

You can buy more in the next round, but it only works—unless you have a massive fund—because there’s only so much you can buy in the next round, right? I mean, you run out of money, don’t you? So you’re gonna end up owning maybe half. But Gary would tell you, since 30% of the batch is unicorns, it doesn’t matter.

Jason, first of all, I could not agree more. I think it’s what you said right at the start: they’re doing their job, and they’re doing it well.

Their value proposition to founders is, “We’ll take you, 2 unknown guys or gals from an unknown place, and we’ll turn you into something that, God bless these people in Silicon Valley, they’ll value at 20 to 40 to 50 million pre.” And their success is a function of the more they can do that, the more people are gonna wanna come in, and the higher-quality people they’re gonna be able to attract.

They’re doing their job, and they’re doing it well. So you’re right. Bitching and moaning’s a waste of time, right? All you can do is do your job well in response.

It’s obviously not the stage we play at. We typically play at a round, a round and a half later. But the only weapon you have on your side of the table if you’re gonna pursue those deals is picking. You obviously have to be way better at picking, or you have to find other sources of deals.

By definition, what’s happening is, as the pricing goes up, they will be correct. On the winner deals, it won’t matter, because it never matters on the winner deals. But if your advice is only “do good deals,” it’s not really that actionable. I’m trying to only do good deals, but it’s really hard in practice.

The truth is, when you’re paying significantly more per unit, it’s just a lot harder to make money. Your picking has to be much more fine-grained. They’ve done a great job representing their side, which is the Paul Graham-stated intent: make it easier to found companies.

He’s making it easier and more profitable and more attractive for entrepreneurs to found companies. More companies will be founded.

I guess it’s the same math, right? But there’s plenty of ways to justify paying 60 million in a seed round, okay, if it’s a generational company. The low ownership crushes you, right?

You can say, “Listen, this is low risk, and this is the best team ever.” But when you end up with 3% ownership with a large fund, that’s where I think there’s the slight fallacy in the YC data. I love Gary and I love the team, but that’s a model that works better for folks who are either angels, have tiny funds, or don’t need to have a fund returner or a material impact. The math works better if you don’t tie that to fund size, I think. I don’t know how you do it to return a lot of funds with 3%, though.

Right.

And so you can say, “Listen, this is low risk, and this is the best team ever.”

Yeah.

When you end up with 3% ownership with a large fund, that’s where I think there’s the slight fallacy in the YC data. I love Gary and I love the team, but that’s a model that works better for folks who are either angels, have tiny funds, or don’t need to have a fund returner or a material impact. The math works better if you don’t tie that to fund size, I think. I don’t know how you do it to return a lot of funds with 3%, though.

At the end of the day, the only thing that makes this whole process work, normalizes errors, and corrects for overvaluation is ultimately returns. Correction on this kind of issue is a lagging indicator. If these, in fact, are on average overpriced, then 5 or 7 years from now, the people investing in them will slowly realize they’ve made a mistake, and pricing will come down.

But the weird thing about venture is it just takes a long time. It’ll take 10 years to discover if the $8 billion funds work and deliver an acceptable return. It’ll take 8 years to figure out: can you pay 60 million pre for an AI Y Combinator startup and make money on average?

Along the way, there’ll be an anecdote of 1 company that worked, but the real test is: does the overall asset class work for those prices and deals? And so far, Garry’s exactly right: he cites the data. The data from the earlier batches has been obviously amazing.

Yep.

But not for YC.

Well, it—

And the founders get 6 million instead of 2 million. There are a lot of conceits in venture, and so many VCs are like, “Hey, guys, take less money at a lower valuation. It’s less risk.” There’s a VC conceit in that.

One: a third of the money is riskier if you don’t burn it. Assuming you don’t burn it, 3 times as much money for the same dilution de-risks your life as a founder.

Absolutely.

Right?

Indeed.

The second thing that the grouchy people miss is one of the other things that YC has quietly normalized: if you do this round next week at 60, or last week at 60 post, and you have to do the next round at 20, it’s okay. It’s actually—they’re all—

Yeah.

The SAFEs are a beautiful vehicle. Every time SAFEs convert, there are 11,000 different prices, right? There’s my best friend who invested 3 weeks before the batch. Then there’s the alumni price. Then there’s the price Harry paid. Then there’s the price Lemkin had to pay on demo day, and then there’s the poor price Rory had to pay 2 weeks later, right?

So when these notes convert at 11 different prices, and then there’s another set of notes at—

14 other prices, nobody cares. Nobody cares that one was at 60 or one was at 20, as long as Rory gets his ownership in the Series B.

9. Normal IPOs Are Back

Are there any other IPOs—anything that we've missed that we should discuss?

Honestly, I think the most salient fact is that they're happening. Normal IPOs are taking place. Normal service has been resumed.

I know we glossed over it in the beginning, but I do think Thoma Bravo raising a record fund this week is a good sign too.

Oh.

That's $35 billion to buy B2B companies.

Right.

Almost entirely B2B.

Something a friend of mine used to say years ago, and it's always struck me: it's just so true—price clears all markets. In other words, you can't get your deal done at $25 billion, which is the price you paid for the last one. But there is a price at which most decent companies can get public, above a certain critical mass. Price clears all markets. That's what price is meant to do.

What you're seeing now is a degree of realism creeping in, and as a result of that, transactions taking place. It's totally healthy.

10. The Kalshi AI Bets

All righty. We're gonna do a Kalshi quickfire.

Sure.

As I said, this is the prediction marketplace my team adores. So we're gonna go with the first one: Will the OpenAI–Jony Ive device have a screen, yes or no?

Your team didn't ask when, right?

Yes. Thank you, Jason.

You got it.

So here's my thesis. If you read the internet, it says that it'll be a pendant on your neck, and it'll have no screen, and even Sam's alluded to that, right? It's gonna be a voice interaction device.

I think the only thing that makes sense, and this is why you buy Jony Ive, is it's everything. So it's a pendant on your neck. It's an app in your phone. It is listening on our computers. It is listening on our laptops. It is a ring like Oura and, importantly, it's an AirPod, right? Because AirPod is great. AirPod already does all this.

And if you build all of them—and the sunglasses like Meta—if you build all of them eventually, some combination, and a watch. You gotta have a watch. People do wear watches. If you do all of them and they're all Jony Ive-level, you solve this problem.

Because we're not all guys sitting in a coffee shop with 3 things on the bar and our triple espressos. My son wears AirPods 12 hours a day. Rory probably will do the necklace, because he likes that kind of open-button shirt, hang out with the necklace, gold chain kind of look, right?

The Hawaiian shirt vibe. Yeah.

Harry will have the sunglasses.

Yeah.

The Polaroid sunglasses with his AI.

In the land of no sun.

So, yeah. Yours will have AI. Now, there may be a screen in it. So I think the answer—the first thing that comes out will be no, but ultimately it will be all of it. It'll be yes. It'll have audio. It'll have a screen. It'll be everything. It'll be in those sunglasses.

Because I don't think anyone's gonna figure out a new paradigm. He's just gonna figure out how it's all elegant.

You're saying yes?

The family of devices will have a screen. Rory?

Yeah, I'd probably go with Jason's answer in an uninformed way. I think the when question is more interesting—

Okay.

—and will it work? I mean, the interesting question would've been: Will the device sell more than 5 million units in the first full year of shipping? And I would take the under.

It might be $50, though. That's the thing. They have enough money to subsidize it.

Oh my.

This is true.

Dude, I would take the over on that.

Okay. See, that's a fun bet, right?

I have to do that. You wanna do that as a bet?

Yeah, fine. Done.

What are we gonna bet?

We'll figure it out.

You know what? You can buy me the device when I beat you.

Sold.

There we go.

Done.

Jason, you—

When they sell 5 million in the first year, I will buy you the 5,000,001st, absolutely. Done.

Oh, that's wonderful. There we go.

And if not, you can buy me dinner at a restaurant of my choosing. Done.

Will Meta release an AI model that isn't open-source this year?

They don't appear to be able to release a lot right now. They appear to be having some kind of gears-grinding internally. My guess is that would be a decision based on some perceived national security issues around open source.

Other than that, I see no reason to change the bet they're currently on, which has been the open-source bet. So no, in the absence of D.C. intervention is my call.

We underestimate the importance of Llama in portions of the developer community. It's already not fully open. We don't even know what dataset it's trained on.

The one thing you know from these guys going in is that they're incredibly competitive, but Zuck's on another level—hypercompetitive. And if releasing a fully closed-source model makes them more competitive, I think they will. So I think it's more than 35%.

I think it's a yes. I think Zuck does what he needs to.

There we go.

I wouldn't bet against Zuck on that. Final one: Elon Musk is back into Tesla and everything in between. Will Elon be out as Tesla CEO before 2027?

To be clear, I read that as an odds bet. They're saying it's only a 1-in-3 chance that that happens, correct? I don't think it's likely. I don't think it's the base case, but you never discount the random event.

I mean, I don't know the guy's motivation from Adam, but it's been kind of painful on every dimension. And coming back, it's probably not gonna be fun, and so you can never discount a rage quit, I suppose, would be that. Especially when you have the fun of SpaceX and xAI to go to.

So you don't do it because you think it's likely; you do it because you think the odds on the Kalshi bet are low enough, or high enough in terms of return, that it's not a crazy thing to do.

Jason?

Listen, he's not the CEO of his other companies for a reason. I don't think he wants to be CEO of Tesla. I just don't think he could find anybody.

He had his CTO—I forget how you pronounce his name—JB Straubel. Is that how you say his name?

Yeah, yeah, yeah. JB Straubel.

I think he wanted him to be the CEO of Tesla when he was ready. I think the problems were huge. He needed to be the Sam Altman of Tesla. He needed to be the face, and the problems were to do everything—electric, AI, automation, trucks. It just was so intense. He had to be the CEO, but he wasn't the founder.

I think if he could find someone better, now that he's out of that administration and now that he's reflecting on life and working harder, I think if by the end of 2027—

Yeah.

—I think this is the question. By the end of 2027, or by the end of 2026 is the bet, before '27, can he in the next 18 months find someone better than him for the job? I say actually more than 32%.

Yeah. You see, exactly.

I'll go 50%.

Yeah.

So that means I gotta say yes on this bet? Is that how it works?

Yeah, that's what it means. You think that—yes. It means—

I mean, he didn't even pick someone very good for Twitter. What's her name? She's terrible, right? Linda, I mean, he—

I know.

And he could also find a nominal CEO for Tesla. He could find someone who's better than her, right? But he could still run the company like he does Twitter or X.

So I think it's more than 50%, because he can't scale as a human being CEO of Tesla and Neuralink and SpaceX and going to Mars. You just can't do it.

Now that he's out of the administration, I think he will try to recruit somebody, just like Marc Benioff will. Marc Benioff's tried 4 or 5 times to get out, too. He's hired 3 other co-CEOs.

Yeah.

It just hasn't worked. It just hasn't balanced the right way for Marc.

I do think—I just want to point out that there is the complexity of the great comp package that has been held up by—

For sure.

—this Delaware judge who clearly just has a thing about stopping him, and that's gotta be run to ground.

But, yeah, there's enough complexity there that, just as a pure betting person, I'm kind of with Jason. You look at the odds and you go: It might not be the right thing to happen. It might not be what you want. But, hey, it's not a crazy outcome just given the amount of change that's going on right now.

And also, to be honest, from the Tesla brand perspective, it might allow them to put some distance between their customer base and a fairly controversial person. So I could totally see it happening.

I also think when you look at the time difference there—18 months now till 2027—you look at how xAI and Groq are going for him, their recent raise as well.

Yeah.

18 months is so long in AI. Elon is a master of moving to the most impactful thing in the world that he has.

Yeah.

It could be that. And, yeah, I think you see the performance issues at Twitter and the realization that he is a human still. And that realization comes home to roost, and he actually goes—

Sure.

He recruited Ilya to OpenAI, right?

Sure.

The guy can recruit.

And let's not lose sight of that.

He can recruit.

Well, don't hate on Twitter. That was an impossible job. Gwynne Shotwell appears to be one of the most amazing managers ever down there at SpaceX.

Seems to be.

Sending rockets up and most of the time having them work, and building an amazing business, that's just been stunning management.

Well, you have to think he wants to reproduce that if he can, now that he has—

Sure.

A moment to reflect. He's like, “Can I have a Gwynne?” For a different company, it's public, there's drama, there are comp issues, but he's got to have thought for a decade, and I think he did try with his CTO. But now that he can try again, I gotta find my Gwynne. I can't survive—

Yeah.

The 58 kids and the 11 companies if I don't find one, right?

The kids are a pretty big load, to be fair.

What? Good God.

Guys, no politics, no personalities. Let's leave families out, okay?

Yeah. Fair point. No politics, no families.

Totally agree.

But it's just a cognitive load, that's all. It is a cognitive load.

Yeah. Totally. Even Elon only has 24 hours a day, it turns out.

Neuralink had a round this week. I thought that was fascinating. But guys, listen, I always love this. You know it. This week was a spicier week. We made it through. We still all love each other. This was much more cordial. So thank you for doing this. As always, it's the highlight of my week, and I really appreciate you both.

All right. To infinity and beyond. Thank you, Harry.

Are we wrapped?

Awesome, guys.

Good.