Databricks 1000亿美元、CoreWeave的112亿美元债务押注与Nubank 25亿美元利润震撼——第19期

Snowflake had a billion dollar quarter growing 26%. Databricks just said they’re crossing $4 billion, growing 50%. It actually makes Databricks sound undervalued. Now this episode is a cracker. We discuss CoreWeave, Nubank, Databricks hitting $100 billion, and much more. One thing I’ve learned about large amounts of money: what people say they’ll do is meaningless. When you get the $10 million or $100 million offer from Meta, it rocks you back. And that’s real competition. At some point, you’ve got to match the dollars. I think we’re all going to live in AI 24/7, and we’re all going to use 10 times the tokens and 10 times the compute we are now in 24 months. Help me do the math. How much more do we need to spend on infrastructure?

Guys, it’s the holiday edition. I’m dialing in from Greece. You guys are dialing in from the office, but we still get to do this. My mother is sitting outside, and she’s like, “Are you recording the Rory and Jason show?” I’m like, “This is the one show of mine you listen to these days.” She’s like, “It’s my favorite.”

That’s because we’re mean to you, Harry.

I think she actually likes you. She says, “You bring me down a peg or two, Rory.”

It needs doing, Harry. It needs doing.

Listen, I want to start with Databricks, now at $100 billion, worth more than Snowflake. I wanted to start with how we thought about the news of it hitting $100 billion, and we can touch on that first before we move on to what it means in terms of founder companies and them inherently being better or more valuable.

Well, I’ll tell you, it’s funny. When I saw it in the Journal, it shows the times we live in; it seems small. Snowflake’s worth $60-something billion, so the private worlds and the public worlds are like 2 different worlds. If I’d seen this a year ago, I would have fallen out of my chair.

This is just for fun. Then we could talk about its relative valuation and whether it’s a good deal—Rory will help us figure out if it’s actually a bargain. But it’s funny how I don’t blink an eye at $100 billion in the day and age of AI. I just don’t even blink an eye.

Yes, that is the meta-commentary. You’re exactly right. If you’d said 5 years ago that there was going to be a $100 billion market-cap private company, you’d be like, “No way.” You’ve got Anthropic at $170 billion, SpaceX at $360 billion, and OpenAI at $500 billion. The correct response is, “Yeah, whatever.” Exactly. So, another $100 billion outcome—it’s awesome. That is the first comment.

The second thing, to Jason’s point again, is you just have to size it on the comp. I will admit I saw the announcement and didn’t have time to dive in and read the numbers, so I haven’t had the most recent update, but—

They said they’ll hit $3.7 billion ARR, or they’ve crossed almost $4 billion.

And what that means is, versus Snowflake, the growth rate is higher. I’m doing this on the way. So you have that perfectly positioned infrastructure provider for everyone who’s messing around in AI. You have the growth. It’s a healthy multiple, to say the least.

But provided the growth’s there, on a comp basis, it’s not entirely crazy. Going back to my point about what value you get in the public market versus private markets, I’ve always said that if one gets out of whack with the other, things go the other way. People gravitate toward the cheapest capital.

This feels roughly right. In other words, this feels like roughly where it would price in the public markets in a world where Snowflake’s market cap is $50 or $60 billion. If this is bigger and growing faster, okay, it’s like, “Yeah, that’s probably what it’s worth.”

Actually looking at it, I think they’ve crossed over. Snowflake did have a $1 billion quarter, so that’s $4 billion, growing 26%. Databricks just said they’re crossing $4 billion, growing 50%.

50% and 26% are pretty different. It actually makes Databricks sound undervalued, if you think about it.

This is what I was just going to say: it actually feels undervalued at 25x revenue, given the growth rate, given the fact it’s found a lab, and given the macro trend where it sits. It really feels like it’s actually very reasonably priced.

I just love it. Just another cheap stock at 25x run rate. Again, it’s all about growth and persistence. The interesting thing is, in one sense, every one of these bets is some version of: if you’re paying 25x run rate for anything, what you’re really saying is that the only thing you know for sure is that when things level out in growth rate, they get valued like normal companies.

Implicit in that is some statement about how long this company can continue to grow at that kind of rate. If you get 2 more years at 50% or 60%, you’re golden. You’ve earned your way into the company, into the valuation. If that growth ever plummets, then you’re high and dry, like many of the 2021 deals.

So the valuation totally makes sense, provided that growth rate goes 50%, 40%, 35%, 20%—not some precipitous growth-rate decline. That’s the bad part. So, again, it resolves itself to this: if the AI boom continues for 2 or 3 more years, all this will—it will be trading at, at that point, 8 or 9x run rate, and all will be fine. That’s it in a sentence.

I actually interviewed Ron Grarisco, the CRO—astonishing guy, by the way.

He was there from almost the beginning, right?

Zero to $4 billion in revenue as head of sales and CRO. He said they were 5 years ahead of Snowflake. I thought it was interesting, so I said, “Technologically, how far ahead are you?” He said, “5 years.” I thought it was interesting of him to state.

Yeah. I’ll tell you a couple of things. The one thing I thought was: it’s cheap. The other thing I was thinking is that Sam Altman said this week that, 1, they’re going to spend trillions on infrastructure, but also that a lot of folks are going to lose a lot of money in AI—that there is a bubble. He said 2 different things that aren’t completely incongruous.

But the other thing I thought when I looked at Databricks is, obviously, it feels very bubbly right now. It feels like the best of times, but it might get better—not just because OpenAI is going to continue to grow, but because let’s look at comps. Databricks hasn’t IPOed yet. Databricks, compared to Snowflake, is growing twice as fast at $4 billion.

Figma had the crazy IPO we just had, right? One of the reasons, Harry, is that Canva might be even better—might be even better—and maybe Stripe is accelerating. I don’t know if Stripe’s going to be better than Adyen, but let’s imagine it’s accelerating because of crypto, AI, and mobile.

We might get the next wave of IPOs that are even better very quickly. The amount of froth that that could create in the ecosystem, and the amount of returns to LPs it could create, could create a bubble on top of a bubble, because times might just get much better in the next 12 months.

There’s an interesting question to be had here: given what we’ve seen with Figma, CoreWeave, and Circle, if you had the likes of Canva and Databricks go out, they’re better—

The public reception would be even more inflated.

Maybe you could also argue, to make the counterargument, that a little bit of what’s happening now is scarcity value. There are very few high-growth companies in the public markets because all that growth has been annexed by the private markets as a bizarre outcome of ill-informed regulation. All the wealth creation is taking place on the private side.

If all those companies were to go public, as you say—if the Canva of this world, if the Databricks of this world were to go out—it’s not clear to me that they’d get that 3x pop. Instead, what you might have is that they probably trade at very high, quite deserved values, but I think the scarcity premium would go out of it.

For example, I can’t remember—it’s been 2 weeks now—but Figma is a Canva equivalent, and Figma priced at 15x forward. Sure, a Canva equivalent and a Databricks equivalent then popping 3x? Probably not, because at some point they would all just be wonderful public companies trading at appropriate growth-adjusted multiples.

What it really would be is taking $1 trillion off the private balance sheet and putting $1 trillion onto the public balance sheet, where it could easily digest it, because U.S. stock markets are, you know, $40 or $50 trillion. So another trillion on $50 trillion—it can digest another trillion.

You could shove onto the public markets an amount of equity that, at appropriate multiples, would, as you say, Jason, lead to a huge amount of LP capital return and be totally digestible by the public markets—probably without the frothy premiums. It would just be the return of high-growth companies in the public markets.

Let’s look at Snowflake. Databricks is growing twice as fast as Snowflake at the same ARR, which is stunning. Figma is growing 48% at $1 billion. Pretty crazy, worth 40x ARR today. That could change—I mean, it’s frothy, right?

But if Canva IPOs next year at $4 billion instead of $1 billion, growing 40%, and Figma’s at 48%, I mean, Figma is one for the ages, but that’s an even better company, isn’t it, Rory? If it’s growing 40% at $4 billion, it could be a jaw-dropper if multiples and market dynamics are consistent with today.

Yes. Again, as I say, it would trade like a wonderful 40% growth. Look, the absolute number would be bigger. The multiple would be good, adjusted.

As I say, my comment is merely on the speculative side of it—the speculative side that took it from the IPO valuation probably would not be there as more stocks come out. But it would still be the aha here at the high end of the private market.

There’s a significant absolute sum locked up in private markets that, at some point, is going to seek a public market, where a number of them will be worth $50 billion, which is unprecedented. A few will be worth $100 billion, and maybe one will be worth $1 trillion in the private markets. Your little head would hurt, because I remember—I mean, it’s only 2018, if you think about this, it’s a weird comment—it was only 2018 or 2019 that Apple touched $1 trillion in the public markets.

Seven years later, arguably—I’m not sure I buy it—there might be 1 or 2 private companies that touch $1 trillion in the private markets. That’s just bizarre.

I go back to this, and I’ve said it so many times—and I’ve said it on this show so many times—but David from Bain said that the biggest mistake we make is that we look at today’s fund sizes and today’s outcome sizes and think that is what it will be in 10 years’ time.

We should be looking at the outcome sizes of 10 years’ time and thinking, are these fund sizes rightly correlated for those? I always think back to that, because if you were to do that, you might actually have a more reasonable or nuanced mindset toward the fund sizes we’re seeing today.

Yes, you do make that point very consistently, and yet I think it’s true. But you’re probably also guilty of overextrapolation, and that’s okay.

I mean, listen, the IPO floodgates are open, right? Even a couple of months ago, we could have debated it, but if you think about it for a minute, the IPOs we’ve had this year, they’re just—they’re not the epic ones, are they? They’re not the epic ones. The epic ones are still to come.

We haven’t—what is the big epic? I mean, we’ve had a few: FIGS, Rubrik, Klaviyo, and Figma. But the mega ones, I think, could just be a force of nature coming.

These are small. CoreWeave has a huge market cap, but CoreWeave, Circle, and Hinge Health—these are still niche plays, aren’t they?

Tranmere Rovers. Yeah, yeah.

I got a reference that’s wasted on your audience, but yes. I don’t think they’re just niche plays. I think they’re the bread and butter of what will make most venture firms money.

I’m not going to denigrate them with niche Tranmere Rovers appellations, but you are right, Jason. In a power-law world, the longer you let things cook in the private markets, the bigger 1 or 2 of the outcomes will be.

Things have been cooking in the private market for a long time. I mean, I think SpaceX is almost 20 years old. They’ve been cooking for a long time, so it totally makes sense that there are going to be 5 or 7 north-of-$100-billion outcomes. It’s just the math.

I’m super happy with Tokyo as a company in our portfolio for detection growing at scale, many hundreds of millions of dollars, profitable, accelerating growth rate. We’re in on a wonderful basis, he said happily, about 123, so feeling very good now. And to the point, in the context of what Jason says, it’s a niche play; it’s not doing $4 billion. In the context of a midsize fund with a good ownership position, a company like that, I feel great. There are going to be 10 or 15 companies that are $10 billion-plus, $50 billion-plus outcomes. But the real point, the venture business will only work for a few funds if they’re the only outcomes. For the venture business to work for us all, you’re going to have to see more of these $5 billion market-cap outcomes. You, as you say, the Hinge, the unknowns of this world start to take place, and I think it will. If Databricks IPOs next year at $200 billion, which makes sense because that’s the implicit goal of this valuation and led the seed, how much do they make in what fund and how does that change how we think about fund dynamics? Could they own 15% of Databricks? It’s possible, right? So, if they make $30 billion on the Databricks IPO, how big was that fund size? And how much does that further reinforce the cycle we’re in? It seems like a lot.

$30 billion is a lot. Look again.

And if the growth slows and it has to go public at $80 billion, which would be a heroic win in any man’s terms, it’ll feel like a real Debbie Downer from where we are now. You’re exactly right, Jason.

I’m just thinking through what I’m trying to figure out—what fund size that would be in, going to Harry’s point about the mega-rounds and things that make sense today. We’re going back in time, because the beauty of these types of investments, like Andreessen leading Databricks in 2017, is that you get the benefit of the VC economics of its day.

They probably overpaid in 2017, right, as an inception investment, and it would probably be seen as laughable today. By the same token, earlier than 2017—because I remember—

Oh, I got it wrong. 20 years ago.

You didn’t, because I remember we were in a conflicting investment.

That was the deal in 2017. A decade ago. They led it a decade ago—the better part of a decade.

I want to say 2013 or 2014. I could be wrong, because Snowflake had just started out. But, yeah, what it says is—shock, horror—if you do the best deal, if you do the best infrastructure deal of your generation, and you hold it for a decade and a half and it compounds, it turns out you make a lot of money.

I go back to a statement that you say, which I love, Rory: Josh Krishna, a phenomenal investor, goes to the best block or the best house on every block—xAI, OpenAI, Databricks, Stripe, fintech—and buys the best asset.

Yeah, buy the quality asset and hold it. I mean, look, the only risk is—I’m not saying this to be a Debbie Downer, but just to point it out—the interesting thing is that, in the public markets, when things start going south, because there’s liquidity, everybody can have a stop-loss of 10% and get out. Then someone else can come in, think it’s going to turn, and, you know, yeah, yeah, yeah, you can hold it all the way down, but you can also get out along the way.

In the private markets, you don’t have that option, right? If you’re wrong at $100 billion, you’ll own the thing. If you’re wrong on price by 50% or 60%, you’ll own it the whole way down. So the only risk you’re running is one big, large-ass pricing risk, right?

I don’t think you’ll have preference to save you, but so far that risk has worked. It gets back to my comment on growth persistence. If they continue to grow for 2 or 3 more years and you have a run-rate business at, say, $10 billion to $12 billion in ARR or GAAP revenue, then you’re kind of money-good on normalized multiples, which is how I think about it.

Can you get from where you are, which is a crazy multiple, in 3 or 4 years to being money-good on normal multiples for normal growth rates? Provided you can visibly see that growth rate, you’re good.

If you’re doing $4 billion and you’re at 40% growth, what you’re basically saying is this infrastructure investment in AI keeps going for 2 or 3 more years, and then I have $10 billion of locked-in contracts with JPMorgan, BofA, OpenAI—everyone. Even if the world slows down from there, I’m good. That’s the bet.

Harry, not to spend too much time on this, but just thinking through it for a minute: if Andreessen invested—did the Series A in 2013—out of a $650 million fund at the time, trying to line this up through AI. It’s probably not perfect, right?

But if they did it out of a $650 million fund, obviously they probably invested in multiple funds, across funds, right? But boy, they’ve got some good funds.

Absolutely.

Exactly. Yes. Again, I think you’re probably right, because this is David’s point. For Databricks, I’m willing to bet they have the guts of $1 billion in that deal. I could be wrong, but half a billion at least, because when you have something like that, you follow it the whole way up and you put a lot in.

The interesting thing is, you’ll probably have a couple-hundred-times multiple on your first investment. You’ll have some 20-times multiple. You might have some late—let’s say 3-times multiple. But if you’ve turned $1 billion or $1.5 billion into $30 billion, you’re going to feel good in the morning.

The 1 thing I also think we don’t take enough into account is just how many people make money when these go out. There are so many LPs who are in SPVs—SPVs on SPVs. There are dentists who are going to make 5x, and I’m saying that positively. We always pick on the dentist, which is unfair of us, but the multitude of winners when the SPVs are as spread out as they are in these mega-companies will make a lot of people a lot of money, in a great way.

Yes.

Yeah, buy a house now in the Bay Area. That’s my main recommendation. Buy a house now. They’re all going up.

They’re all going up. We can talk real estate with Jason Lemkin. All I know over these cycles is, you buy a house before the next massive IPO wave.

Why is everyone paying cash?

Speaking of which, guys, I want to usher this along. Rory, you're going to freaking kill me because it was a bit of a late addition, but it's peak bubble time, some might suggest, when you see the return of the SPAC. Chamath announced his new SPAC today, and in the prospectus it says that if they lose their entire capital, they will embody the adage from President Trump that there can be no crying in the casino.

I respectfully thought this was peak bubble. I would love to hear your thoughts on whether it is peak bubble and whether I'm being too moralistic.

You won't know it's peaked until it goes down the other side. It's interesting you say that, because I'm listening to the two of you and your euphoria 5 minutes ago saying, “Oh my God, this is just the hors d'oeuvres, and the entrée is coming later in terms of the public markets.” If that's the case, then it's not peak bubble. He'll be selling into strength like he did in 2021.

That's the first comment. I don't know if it's peak bubble or not. If I did, I wouldn't be wasting my time talking to you; I'd be trading the QQQ and making money.

It definitely correlates remarkably. The return of Chamath to the SPAC market does appear to correlate with the most recent bubble in 2021, where it worked in the bubble and didn't work afterward. So there's definitely correlation there. I'm loath to impute causation.

The serious comment is that I remember Bill Gurley talking a little bit about SPACs. It's not a crazy alternative on its face to the problems around IPOs and mispricing, but I believe that when you parse through it, it's a bad structure. I think it was validated to be a bad structure in 2021: relatively few of them worked. Though, to give credit, one—I think SoFi did work. I think it priced at $10 and is now $23.

Most did not, because you have all sorts of weird adverse selection. So, especially in bubble times, it's a pretty iniquitous and expensive vehicle to raise capital in the public market. I'm just not a wild fan of SPACs.

I think you get adverse selection in terms of who decides to do the deal. The other thing is that the incentives aren't aligned, because the person doing the SPAC effectively crystallizes their entire gain once the deal gets consummated, regardless. Obviously, if the outcome is great, that's wonderful—they make even more money—but provided they have an outcome, they make some money.

The structure is all wrong. You effectively get paid as the promoter the day you get a deal done. One thing you learn in any investment-management business is that if you pay people to do deals, deals get done regardless of the quality. I just think it's an imperfect vehicle for fundraising.

Therefore, rather than personalizing it on one person, be he ever so great or not, I just think it's a bad structure. The fact that it's back is indicative. The fact that the structure is back is indicative of a bubbly time.

That was me being nice. Jason, is it below your line?

I only know Chamath from a distance, back as a founder watching him in the day, and then as an initial VC investor at Social Capital. I do have a ton of respect for him, but I've honestly never listened to it. I know David and Jason a little bit.

I don't get why he cares. This seems like a lot of energy for someone reasonably rich to be doing, to Rory's point, weird shit like SPACs. It's not a criticism; it just seems like a lot. Maybe he'll make $50 million off this. Help me do the math: is the sponsor in this?

Why not do something more meaningful with the world rather than do something like a SPAC, which is at the edge of grift? I appreciate the disclosure, but this seems to me something better for someone on the way up, like trying to get their first $100 million. This might seem like a good strategy, but I don't get why billionaires are maybe approaching it. I don't know. It's none of my business how rich he is.

The juice doesn't seem worth the squeeze here. The drama doesn't seem worth it.

Agree, because I think you're right. By the way, my mental model is that the promote is 20% of that. So, you get $50 million, but you have to gamble $5 million up front on fees. That's it: it's a 10x. The day you get a deal done, it's $50 million, and then you hope like hell it goes up, not down.

You probably get out. If it goes down, you probably get $25 million, but that's it. You're right. It seems like a lot of pain and drama.

Also, on the disclosure—again, I'm going to get moralistic—at one level, it's cute, because I do like the fact that there's humanity creeping into these disclosures. I remember back when there was no founder letter and there was nothing except bland legal speak. So, I love the comment about crying at the casino, but it's not a casino.

The point of Wall Street—I mean, it may feel like a casino to some of the marginal players, but the point of Wall Street, and even these IPOs, is to funnel money from the savings of ordinary people, managed by people like Fidelity, into profitable, long-term investments. That's why we have the whole damn thing: Wall Street, all of us getting highly paid to allocate capital.

It's not a freaking casino, which is a zero-sum game where only the house wins. It's about allocating about $2 trillion worth of savings every year into great new investments in companies like Figma. It's not a casino.

I personally would never let anyone manage my money who thought it was a casino. Excuse my language, but it's not a casino.

Is stuffing retail investors' 401(k)s with private-equity investments a casino issue? They don't even understand the investments. They have no financials and don't know what the markups are. Is that at the edge of crying in a casino?

I understand what you're saying, which is that the more unsophisticated the investor and the more complex the instrument, the more likely it is that the investor will lose and the promoter will win. It's one of those life rules.

It's why, even with IPOs, there are complex things, and you kind of go, “Whenever it's complex, your rule of thumb should be: whenever a bank is explaining something complex, you're losing money,” right?

I don't think it's as bad as, quote, the casino, because, on aggregate, in a casino the median expected return is negative because the house takes a rake, and the median is a loss. Even private equity—the long-term returns in PE are good. The long-term returns in venture are good.

So, in theory, allowing individual investors to access those returns might have meant some individual investors would have gotten the Databricks exposure that Andre got. So it's not a bad idea. But to your point, Jason, the problem is often that the more complex the instruments are, the more likely it is to go wrong, the more likely it is to be adverse selection, and the more likely it is that the good parts of the PE and venture value chain don't get allocated to retail investors.

I don't think it's casino-level. I do think it requires a fair amount of disclosure and monitoring.

There was an example on X, formerly Twitter. I wish I'd saved it. I think I saved it. I'll look.

It was a PE firm that had to do public disclosure about a secondary position it had bought. So, it had bought another position with 50 investments in it and disclosed it, and got it at a 20% secondary discount. The next day, it marked it up.

Yeah.

Right, and the disclosures were in 3-point font, with asterisks and daggers at the end of the disclosure. The point was that a retail investor would think there was—I mean, the IRR would be insane for the next week—but a retail investor would think they had a 20% gain this year.

They bought Scale AI or Saster One[?] at a 20% discount and immediately marked it up because they had to mark to market, and the fund was up 20%. How would a retail investor even understand what the hell was going on there, or whether that's even legitimate?

I think the interesting thing—and, again, Harry, you're right, I am annoying to you because I haven't thought about this at all, so I'm working on the fly here—but I think it gets to an interesting philosophical question.

There are 2 approaches. There's the “if I disclose everything, it's buyer beware” approach, which in this case—and this is an example of that, and SPACs are an example of that—means it's true to say that every single thing you needed to know to stop yourself from getting screwed over was available in the prospectus. So there's one school of thought that says, “On your head be it. You made your own decisions.”

Then there's another school of thought that's a little more nanny-protective and says it's just not realistic for individuals to do that level of analysis on something managing their money, where they spend 3% to 5% of their time. Therefore, individuals need to be protected or regulated.

I think what we're seeing right now is probably, by virtue of the overregulation in the last couple of years, a feeling that we're going to wrench the needle back the other way, perhaps to a lot less protection. I think it's a happy medium. It's a hard thing to decide because, in protecting those individual investors from the grift, as you put it, Jason, of a fund that has an arbitrary markup that's not there, you also protect them from being able to invest in that Databricks round at $500 million pre-money that's now 100x and getting that.

I mean, rather than doing what I think we do a lot of, which is this black-and-white thing, it’s pretty tricky. It’s a tricky regulatory balance: how do you allow people a fair amount of freedom to pursue what they want to do, treat them like adults, and assume the disclosure is there, while at the same time avoiding those horrible stories of little old granny giving her money to XYZ Wire Co., and they just lost all her savings and now she’s penniless on the street? It’s tricky.

I am going to take it back to regular programming, Rory, because I don’t want to piss you off too much. My mother can’t be too thrilled to see me getting a beating.

I want to go back, actually. We said something there about the extended period of privatization for companies, SPACs being a different mechanism of getting liquidity. There’s a $6 billion secondary for staff at OpenAI, and a lot of them are going to make great money at scale. What do we think of staff secondaries at this level of scale, which is in the multiple hundreds, if not low thousands now, getting this level of secondary? Start there.

Good for them. It probably speaks to willing buyer, willing seller, and an informed buyer. Again, it speaks to the competitive pressure that Meta has exerted on AI salaries by being willing to offer people cash packages—allegedly, and I haven’t heard whether it’s $10 million to $100 million.

If you’re sitting there in OpenAI, you can talk about the mission until you’re blue in the face, but it’s going to help a lot if you can also take off $10 million. So you’re going home. I mean, it’s all very—

One thing I’ve learned about large amounts of money, as I see this in acquisitions, is that one of my rules is this: what people say they’ll do when faced with a large amount of money is meaningless and bears no relevance either way to what they’ll actually do. Predicting who wants to sell their company based on what they say in the abstract is a waste of time.

When you get the $10 million or $100 million offer from Meta, it rocks you back, I’d say. You go home and talk to your spouse: “Do we want to do this? This changes the rest of our lives.” That’s real competition. If you’re OpenAI, you can preach the mission, but at some point you’ve got to match the dollars.

If you think about it, $6 billion is only 60 $100 million engineers; it’s only 600 $10 million engineers. Stunningly, to Zach’s comment, it ain’t that much.

Wow, Jason, he’s turning into you. This is the full-scale transformation of Rory.

Well, you know what? There’s the competition issue, which is critical. Rory, I just think, stepping back for a minute, it is nice that as companies get bigger and stay private a little longer, if this is literally institutionalized.

We talked about tender offers and all that. I think more so if they actually act like the public markets, where every month you vest, right? You’ve got to stay through your cliff, and all that stuff works. But if it’s pseudo-public, and you make it to 13 months, you sell a 48th of your stock, or a 60th, or whatever it is, that sounds audacious, but if you step back for a minute, that would be a nice outcome.

It would just work that way as a late-stage company. Forget about tender offers; it was just almost automatic. Every month, you could sell your vested shares.

Agreed. To use a silly example, Apple had an $800 billion market cap going into 2018. So let’s go back to about 2014: it had a $500 billion market cap. You didn’t have an objection to those people selling in the public markets.

The only difference between this $500 billion market cap and that $500 billion market cap is, for stupid, absurd reasons, it’s still private. So I’m totally with Jason. If we collectively want to recreate the entire mechanism of the public market at massively more expense and massively more transactional cost, then more power to us. Part of that is being able to sell.

When suddenly a large portion of your employees are multimillionaires, does that impact the cadence of their work?

I think money brings out what people really are because it stops you from having to do things you don’t want to do. The people who love their work are going to work just as hard, or harder, because they’ll be able to throw money at all the other little problems that take up your daily time. There will be people, and then other people who may have other dreams in their life will go do other things.

That’s what always happens. It’s not a binary thing. People ask this question, but we see founders get rich and then just keep on trucking, not sell a single share, and compete aggressively the whole way through. Then you see other people go, “I never thought I’d have $20 million. I have $20 million. I want to go teach high school.” That’s just the way it’s going to be.

I do think, to Harry’s point, that this is a cultural or tech change, and I don’t think saying that is bad. The handcuffs of staying at a startup for a while so that you could see it through a liquidity event are weakening. You would stay longer if you left because you might not have the money to exercise. If you left, other things might happen. I think at the margin, those things are all weakening, as are implicit tenures for employees. It’s all changing.

As VCs, historically, we’ve used the pool as a retention weapon. “If you stay, you’ll make a lot of money. If you leave and you can’t afford your exercise price, well, great. I get half of it back. I get all of it back for my pool to give to the folks that stay.” Some PE firms are extreme there. They have a lot of leaver clauses: you leave, you get nothing. This is kind of the opposite, and I don’t know if it’s bad, but it is different as a way to lock employees down with illiquid equity.

I totally agree, Jason. It is different. I love it. You’re right. The idea that we’re all locked into this made sense when it was a 4-year sprint in the mid-’90s and you would go public at $50 million in revenue.

I don’t think it’s realistic to have that attitude in a world where you can be 12 or 15 years in. This evolution, to allow people to access their capital, is just more realistic. Refresh grants get into the same thing. It’s not just a 4-year sprint anymore; it’s a 15-year marathon, and you can’t expect people to put 15 years of their life on hold and not be able to do the things they want to do in life. You have to have these kinds of events.

Is it a 15-year marathon when Hopin’s founder raises $130 million and walks away?

It’s only a marathon if you finish the marathon. Harry, it turns out that if you punch out after the first mile, it’s not a marathon anymore. It’s a failed marathon.

How did we think about that? It was a big announcement: $130 million raised, and he leaves pretty abruptly.

Well, look, it makes sense in this era. We should see 10 or 15 Hopins, not just one. Harry, it just makes sense. We should see it. Hopin was just a prominent one, but there are other Hopins we can talk about. We can think of several folks in the SaaS boom of 2020 and 2021 who walked away with too much money.

There should be double digits in this era. There should be 10 or 20 founders who walk away with 9 figures, whether they’re WeWorks or Hopins. There have got to be 20 or more of them. It’s a cost of doing deals over the—

I find all hot deals now are done on Saturdays. They’re all—this is the new founder vibe. You get an email, even if I know them: “What’s going on?” “Well, we’re taking meetings on Saturday.” I’m like, “Can we—?” I work hard. We all 3 of us do, but this Saturday I can’t. I’m like, “Well, I mean, if you’re going to close every deal on a Saturday, we’re going to lose a few hundred million here or there.”

I find that hilarious. I’m with you on the Saturday deals. There have got to be 10 Hopins, right? Just mathematically, there have to be.

Maybe, stepping back, you should explain, because most people might be like me, where I haven’t spent a second thinking about this story. Implicit in that is that it’s a company that raised $135 million. It sounds like the founder took off significant money and, lo and behold, as of today, he’s not at the company anymore. Is that the summary?

He has a new passion. He wants to start something different, Rory, all over again. He might conceivably sell it, sell the rest to Windsurf or somebody, but he’s moving on to his passion project.

Yeah, and now you’re being snarky. I actually thought your first response was the correct one, which is, “I didn’t even know about this. There are going to be tons of these.” I didn’t even break a sweat thinking about it, Harry. It’s like I’m shocked to discover that some deals don’t work, some founders end up either leaving or getting transitioned out, and some money gets lost.

To me, it didn’t even rise to the level of a story. It’s fun to snark, but literally, Jason’s right. Your typical venture loss rate is that 30% of your deals don’t work out, and that’s where we play, in As and Bs. It’s probably 50% plus in seed.

I remember in 2001 and 2002, the loss rate spiked to 60%. So the fact that loads of companies that have raised $100 million go bust and, you know, under those circumstances, yeah, whatever, didn't even rise to the level of interesting.

Speaking of rising to the level of interesting, one for me that really did—which I love—is that Nubank reported earnings on Thursday. Blowout quarter: net income reached $2.5 billion, up 42% year over year, with 123 million customers. It's incredible. I'd love to hear your thoughts on this blowout quarter for Nubank and how you feel.

Sure. I'm not going to focus so much on the quarter as just stepping back and talking about these fintechs. It's interesting—and obviously, as people I'm sure know, Nubank is a dominant fintech, which is a next-generation bank in Brazil, Colombia, and Mexico. It's interesting to kind of sweep them up with Revolut, a dominant fintech bank and next-generation bank in Europe, and maybe even Chime in the US.

You pinged me on this, and it got me thinking: What do all these have in common? What do all these have that's different? I think what they all have in common is that, in the last decade, they found the weak, soggy, badly run, and profitable parts of the existing banking infrastructure and attacked them voraciously.

If you look at the 3 companies and the outcomes, Nubank is public and worth $60 billion. Revolut is private, maybe worth the same, and Chime, in the public markets in the US, is worth $10 billion. The interesting thing is they're all different. They all found a different weak spot, and because there's a different dynamic on each of the 3 continents, in my view, it's no surprise that Nubank is the most profitable and, perhaps we can talk about Revolut in a second, the most stable.

Frankly, the incumbent banks were weakest in LatAm. They were old. They were stodgy. They weren't serving their consumers, and these guys came in over the top. They didn't just build a deposit base like Chime did. They've done the whole full monty: deposits and lending. They've basically built a next-generation bank. They've cleaned the clock on the average middle-class consumer in Brazil versus the old guys.

As a result, depending on the day, they're the largest or second-largest bank by market cap in LatAm. I think what I like about Nubank in the context of the other 2 is that's what winning looks like, because banking is inherently geographical. What you can now say truthfully is that a fully developed bank that doesn't just do deposits but does the full gamut of financial services can end up with a market cap equivalent to the largest bank on the continent, although it's probably still 5 times smaller in terms of assets. That's winning.

If you contrast that—I mean, it's going to take time—the other extreme is an $11 billion public outcome. Great outcome, great company, but the US banks are pretty well run. Their niche wasn't as big a niche there. So where were they able to thrive? 2 things.

One is, with the Durbin Amendment, they had this arbitrage on deposits, and they were able to make money on debit cards and therefore without ever doing lending. The interesting thing is, unlike Nubank, they've built a profitable business without ever doing lending, which is the bread and butter of most banks. They've built a perfectly good business on deposit accounts and debit cards, living off the interchange.

Now it's at $11 billion, despite the US being—I don't know, depending on the day—a 10-times-larger GDP country. Chime is only an $11 billion market-cap company because they were playing in a less profit-rich environment. There just wasn't as much money. JPMorgan might have ceded some of their mid-tier customers, but these guys didn't roll over everybody because the banks here just weren't as inefficient as in LatAm.

Then Revolut, obviously, had an edge in Europe with cross-country money movement initially. Now, obviously, I've seen the numbers, Harry, don't worry—they're roughly similar in size to Nubank: less lending, more FX, and more crypto. That's another wonderful business, because I think the European banks, frankly, weren't as efficient as the US banks, and the opportunity for Revolut, although slightly different from Nubank, was big.

If you look at all 3 together, you kind of go, “Wow, big outcomes here.”

I think when you look at all 3 together, you do actually go, “Wow, Nubank is an incredible business.” Chime, respectfully, is just a much different scale. The 123 million customers for Nubank really shows the scale difference.

I would say that, as you said, FX and crypto are the core revenue drivers for Revolut, versus Nubank now moving more and more into secured and unsecured loans. That's really hard to replicate and such a defensible growth driver for them, which is so impressive.

Ironically, the pun is that Nubank is old bank. It's a mainstream bank. I mean, both Revolut and Chime exist, and I don't mean this on the fringes. The opportunity was Nubank saying, “No, we're—if Giannini, the 1908 founder of Bank of America, went to Brazil today, he'd be like, ‘I know exactly what these guys are doing’: the bank for the middle class, top to bottom.” It's an incredible story.

Interesting, as you size up the Revolut opportunity, it's a very crude, simplistic metric, but the fact that Nubank has a market cap equivalent to that of the largest-market-cap bank—I think Itaú Unibanco—in LatAm just speaks to the opportunity when you can make it happen at $63 billion. Do you not still think it's actually quite underpriced, given where it could be?

On an earnings basis? I can see why you say that. In the end, again, back to the only negative about being an old bank: if you're not careful, you end up trading like an old bank. Traditional banks, once they become mediocre, mature businesses, trade on a multiple of book, and Nubank will be very overvalued on that basis.

I do think there's upside there because they're only in 3 LatAm countries. There are other countries too, so there's more room to run. From a revenue and operating-income perspective, yes. You do wonder: When do the dynamics of just being a bank start to catch up with them? Do they get regulated? Do they start doing bad loans? Do they do all of that?

Invariably, banks at scale do dumb stuff. It's kind of their genetic disposition. Maybe the better answer is this: If they can continue to do what they're doing and avoid doing dumb bank stuff, you could see a 2x from here, which would be wow. It would be just a wow, wow.

I agree. I actually think when you look at Revolut's product expansion versus Nubank's product expansion, and how few products Nubank has compared to Revolut, Nubank could very much have the same. Also, in terms of revenue potential with each of them, wow. It could actually be a $300 billion to $400 billion company.

These are at the edge of consumer investment. It's not my strength. But the one thing I think about, and why I'm especially impressed with the seed investors, is that Nubank has 35% market share of Gen Z in Brazil, and Revolut is all across Europe. If almost 20% of everyone in the UK has a Revolut account, you don't want to be number 2 or number 3.

You can compare it to Chime, but if Chime has less than 5% of US holders having an account, it's very different. These are incredible outcomes. Predicting this at seed or pre-seed may be beyond my skill set. It's pretty impressive because you want to see massive market share in the largest part of the economy.

This is like a W on the 2-by-2. You want to be number 1 in the entire plumbing of a nation's financial ecosystem. That's an S-tier bet to make at seed, isn't it? You're not just rolling up to the latest plumber software and seeing 18% month-over-month growth in their AI-fueled agent and saying, “Let's do this one.” This one is knocking it out of the park from the pre-seed and seed stage.

But I think the smart thing all of them did was, obviously, you have huge aspirations, but you find a very profitable entry niche. For Revolut, it was definitely cross-border money movement. The way I describe it is, you just find something that the banks do so badly, and they're just robbing you blind as an individual.

I'm not a Revolut user; I'm a TransferWise user when I'm sending money back to Ireland. It's so much a better product than the banks. You type it in, the money goes right, and Revolut has done really well by starting off with a niche product that's very gross-margin-rich and LTV-positive. Frankly, it gets a bit of consumer love because you just hate getting ripped off when you're moving money, and then suddenly it's good.

Then, building on what Harry said, you just build year after year, for a decade-plus, and add all the other stuff over time. You wake up one day and realize, “Oh my God, it's not my account for moving money overseas. It's my account for money. Then it's my account where I do my crypto, if you do it.” Suddenly, for that next generation, it becomes that primary bank. It takes a generation, but it's unstoppable.

When I had Nick on the show from Revolut, he said, “We think about banking like snacks. We want you to start with a little snack, which is FX, and the more you use it, you see that this snack's great. I can add this really routinely, and then suddenly the snack becomes the meal, the meal becomes the whole meal.”

Now it's reported that 35% of users have it as their primary account, which is interesting compared to 50% of Nubank users, but still really impressive.

And what’s also interesting about this is that, in a world where banks were tied to physical infrastructure, you couldn’t have a bank for a specific demographic because you had to have a bank for a specific geographic type. If you’re the bank for Northern California, you’re the bank for Northern California. You deal with old people, you deal with young people.

On the internet, instead of slicing by geography, which is how banks have been for time immemorial and how U.S. banks are regulated, you could slice based on demographics or based on propensity to buy a certain product. It really just turned the table. You see it in Europe as well: instead of there being a bank for the U.K. and a bank for Germany, catering to Harry, his mother, his grandmother, and the local SMB, someone was able to carve out this niche: people who want to move money from A to B, regardless of what country they’re in, within reason, obviously, given some regulatory issues.

The big aha was the acceptance of the internet, which has taken, frankly, 15 years longer than all the people who funded the neobanks in ’99 and 2000 thought. I remember being there. It took 20 years longer than people thought—maybe 10 years longer for people to become accepting of it—but once they did, you had, just as you say, Jason, this opportunity to significantly disrupt one of the top 3 market-cap sectors, fintech and financial services, on the planet. These have been great compounders.

Final thing, and then we’ll move on, but it astounded me that Revolut internally has 26 new products that they’re baking in their venture lab. They have insane testing metrics every week with Nick to determine what makes it to the main stage, and they run them like little venture teams with different funding goals.

That is where Nik is truly world-class. Of all the founders that I’ve interviewed—and I’ve interviewed hundreds and thousands, and Jason, you and I have both interviewed thousands—he’s the best I’ve ever met. Unbelievable.

I’m pretty impressed with this playbook and how many of the S-tier founders can run it. They get 20, 30, 40 ex-founders together and have them be little GMs of their product lines.

We’ve all had one or two of these. As founders, everyone has found one founder and tricked them into coming and running something for them. But the Ripplings, even Oura, where I’m on the board, and others where you’re seeing double-digit numbers of CEOs join to be GMs—it sounds like Revolut does it too. It’s a disruptive way to build a company at scale, and getting them to join is a superpower. It’s a superpower to get CEOs to do this. It’s not easy.

Yeah, absolutely. We’ve seen some of that in the portfolio, too, getting 10 or 20 of these guys to join.

You said something about the edges of consumer there. Consumer fintech has been super hot for a while now. We’re seeing it with Revoluts, your Nubanks of the world, your Circles, and it will continue. Consumer’s been fucked. Everyone has been shitting on consumer, especially consumer goods, and On seems to be the anomaly that just keeps on giving.

They have $4 billion in sales, up 38% year over year, software-like margins of 61.5%, and they’re worth $15 billion. Is this a shining light and a beacon of darkness for consumer, or is there a pathway for consumer brands in venture?

I don’t know how you get from On to brands in venture. Look, it’s clearly an amazing company. You look at the story: founded in 2010, public in ’21, and just killing it in the high-end running shoe marketplace. I think it’s a thing. I don’t know.

I don’t know how applicable that is. I don’t know if the insight from that is that someone should raise a fund focused on consumer goods, or whether they’re going to conclude that not everything in life is knowable, predictable, or lends itself well to venture-type investing. I think some categories are categories of one, and they’re very hard to access. That’s the kind of deal where you just go, “Good on them.” I’m not sure you build a repeatable venture fund around that.

Before we go to CoreWeave, I actually had a better thought there because I lost my thought. I should think a little more about your comment on On versus Jason’s comment on fintech. I think one of the things that venture—most venture, not all but most venture—needs is some kind of megatrend that’s pushing these companies along, where you can say these things are happening at a wider scale, and therefore companies will be formed in this space. You can have an a priori belief that they’ll be big.

Even though it was hard to predict, I think the Chime and Revolut opportunity was knowable. You had digitization, you had the internet, and you could see there was this forming consensus that there would be activity in fintech. I remember looking at Simple, which was a predecessor of Chime. It didn’t work out; Bank of Santaare bought it. But there was a sense that this thing could happen over the next 10 years, and venture could think about it and form a little thesis. Someone would be there, ready to intercept the Chime or Revolut wave because the trend made sense.

I think something like On is very hard to predict. There’s not a whole set of trends. It’s very much a category of one. You can say, “Running is a trend, or high-end shoes are a trend,” but it’s just not as scalable. That’s perhaps a better explanation of why consumer fintech can be an investable venture trend, while high-end running shoes are probably going to be a one-off anomaly. Sorry, you can go back to CoreWeave now, but that was just so much more interesting than high-end running shoes.

CoreWeave is up to $11.2 billion in debt. We’re also seeing a load of other players join them, or join the model—Nebius being one. Mistral is moving more and more closely to them. First, how do we feel about the news that CoreWeave is up to $11.2 billion in debt?

The quick answer is: duh. If you just look at their very succinct Q2, Q3 investor presentation, it was like, “Here, we did $1 billion this quarter. We’re going to do $4 or $5 billion for the year.” Duly noted: a loss of $800 or $900 million, not that much. Then you look at the capex plan for the year: $22 billion. Okay. Once you decide to spend $22 billion in capex, you’re going to have $11 billion in debt.

The way you think about CoreWeave is that they’re at the pointy end of the bet that Meta, OpenAI, Anthropic, and all these companies are making. They’re saying, “We want to deploy a ton of GPUs and a lot of data centers.” CoreWeave is stepping up and saying, “We will be the financing vehicle to make that happen.”

I love the Barron’s headline. The Barron’s headline said it all: “CoreWeave Borrowed $5 Billion in Debt. It Won’t Be the Last Time.” This model—and there’s nothing wrong with it—is a real estate company. They’re a very sophisticated, clever, risk-managing real estate company. They’re in the business of borrowing large amounts of money, building data centers, and signing long-term leases.

If you’re going to grow revenue from $5 billion to $10 billion, you probably have to grow capex from $10 billion to $20 billion. They want to be doing more of this.

Does the debt matter? When I interviewed the founder, he said, “We borrow when we have long-term demand.”

What’s the issue?

I think, obviously, if he can match his debt structure with his demands from customers who will do take-or-pay, then it doesn’t matter. Risk management is all a matter of degree. There are no hard-and-fast rules, and people who try to do this with hard-and-fast rules—it’s just not doable. It’s absurd. You can’t do this business without debt, right?

But what will kill you—and I can’t remember who did it on Latent Space, but somebody had a really nice discussion about this—is if you have long-term expenses and your customers are short-term inference customers. Then you’ve done the classic banking thing: you’ve lent long and borrowed short. You can be caught short. Your customers can evaporate in a month, and you still have a 10-year lease.

If these guys really are signing up Microsoft and OpenAI to a 7-year contract, and they have a 7-year debt profile, then provided they match it, they can manage. Which of the 2 is going on? I don’t know. I don’t have visibility.

That’s the bet you’re taking. I’m sure they disclose things like what percentage of their long-term fixed charges are met by take-or-pays. Implicit in that is that the company on the other side of the take-or-pay is willing to continue to take it or pay it. If times got tough and people didn’t want all that capacity, it’s been my experience that people fight really hard to avoid paying it.

It’s not riskless, but they strike me as having a fairly good understanding of the risks they’re taking. Whether they can, in fact, fully hedge it, I don’t know. Sorry, Jason.

My only question—or my only question is, I’m just watching and learning. They need the stuff. They’re going to take another $10 billion this year; they already said it, right? They’ve got to buy the GPUs, and the hyperscalers are offloading some of this risk to CoreWeave.

They're guaranteeing it as part of it. There's a slightly cynical view of offloading your risk but guaranteeing it—put it in air quotes. I'm really just here to learn. I see it as the canary in the coal mine.

OpenAI will find the capital it needs. Microsoft will certainly find the capital it needs, right? But if CoreWeave struggles, if it can't service its debt, if it can't raise more debt through this weird arrangement of the hyperscalers guaranteeing it, that's where we'll see it. Even if there's just a bump in this cycle that doesn't last, I think we'll see the bump in CoreWeave before we see it in the folks one level up in the market.

You're not going to see it in Microsoft. OpenAI isn't public, right? You're not going to see it in Google. They're too big. But you could see that canary in the coal mine in this public company due to its exposure.

That's very true. Or even in the, to Harry's point, wannabe CoreWeaves, because CoreWeave at least has public liquidity. But yes, in a downturn, it's amazing how quickly liquidity evaporates for this kind of product.

Yeah. So if they have a really rough quarter, we may see something there that you wouldn't otherwise see in the end customers. We might see it a quarter or 2 early, and it might not show up in the end customers.

I'm sorry, teach me, Rory. Why would you see the supply of cash go away in a down market for this type of product, given the quality of customers guaranteeing revenue, being Microsoft and OpenAI?

If you have a 7-year contract with Microsoft that's take-or-pay, and you have 7-year debt that's perfectly matched, and the contract's ironclad, you're right: there's no risk on that contract. Let's just say you're 3 years in and Microsoft isn't using its take-or-pay, but it's still paying its commitments because it's Microsoft. It's the only AAA credit out there, better than the government. It pays its deals, so you're fine on that asset.

But I'll tell you one thing you won't be doing: you won't be building another one. That's my point. Perfectly matched businesses will be fine, but if there's some up-and-comer who's maybe built another data center a little bit on the come because they thought they'd get some contracts, because Microsoft was going to continue expanding, and suddenly that's not there, that's the guy who'd be struggling.

To be very clear, if they're well matched, then despite the debt they'll be fine. But the point is, it comes right back to the Databricks discussion: all this depends on 3 to 5 more years of continued AI capex expansion. If that's the case, everything works. If that's not the case, all bets are off.

Yeah. Or even if—listen, I obviously think a lot of this is disclosed in a way that's most favorable to the company. If they say all their customers are 4-year or longer contracts and 98% are take-or-pay, that sounds great to the public markets. One, going to Rory's point, there may be asterisks and daggers in those agreements. No matter what it says, they may not be as ironclad as they look.

Two, even if they're not, if they pay rather than take, that's a canary in the coal mine. If the quarterly release says OpenAI has determined it's cheaper to pay rather than take, that would just be an interesting canary to see that maybe there's a bump, maybe there's a bump in the road. Even paying is a canary rather than taking.

And you always feel a bit uncomfortable even raising this risk, or at least I do, because I know that the trend is so strongly toward expanding compute. People are talking about next year being bigger than this year in terms of hyperscaler scaling. So I'm not saying it's the most likely outcome, but it is simply true: you have to at least encompass it as a possible outcome.

I mean, you have to have 2 scenarios: continued enormous growth, and what it would look like if it didn't. When that happens, you just don't want to be the marginal player, because not everyone will go to the wall, but the marginal player will. You want to be Goldman Sachs, not Lehman Brothers.

I was speaking to the CEO of one of these largest players. I can't name them, otherwise they will literally sue my ass, but he was like, “Harry, you literally just don't understand. The only thing that is limiting me now is supply—supply of compute. I am so voracious. You don't understand. I feel like a drug addict, chasing, chasing, chasing. Nothing is enough, and the scale of the money demands is so much more than anyone knows.”

Yes, that would be true today. So if we were to extrapolate that to a little game, Rory, because I know you love games: is CoreWeave over or under a $60 billion market cap next year? It's at $43 billion today.

I don't have a meaningful answer to that question.

Well, you're a venture capitalist. We don't always have to.

But actually, one of the things I'm learning, especially on this podcast, is not to opine, or to feel the need to have an opinion just to fill the space. I would say the logical answer is there's probably a 1-in-3 chance it doesn't, right? It's not 100%. I'm not absolutely certain, but I can contemplate at least a 1-in-3 chance that sometime in the next 12, maybe 24, months, demand slows down just enough that people penciling stuff out to the sky suddenly realize maybe it will taper down. At that point, a lot of these multiples could compress.

I'm not going to say it's going to, because that's the question you asked. But I am going to say the probability of continued infinite AI demand—the probability of Sam Altman's trillions—might be high, but it's not more than 2/3. There's at least a 1-in-3 chance that we end up in a period of time where demand slows, and at that point some of these stocks probably wouldn't trade at those valuations.

That's all. Not that I'm an expert in public-market investing, but I would just say, look, obviously CoreWeave is riding the right trend. It is inherently tied to Microsoft as its dominant customer and OpenAI, which has guaranteed it $12 billion of revenue. So it's got to run.

But this thing just went public, and it peaked on June 20th at $183 a share. It's half that today—half that today in 2 months. So the volatility here, the beta, the alpha, it's very confusing.

My knee-jerk response is, sure, bet on CoreWeave. The trend's there. Sam Altman just said he's going to spend $1 trillion on infrastructure. He's got to send a few of those nickels to CoreWeave. But it's already down 50% since its peak, even though the macro trends are strong.

We're all investors in QQQ. Whether it's just in our 401(k)s or our private investments, we've all been making so much money from AI, just by being in Nvidia, Microsoft, and the top 7. Why doesn't CoreWeave seem to be doing better if it's down 50% from its peak?

Maybe we're fine just all living in our QQQ. The weird thing about the stock market today is that's where all the gains are. The long tail is not doing well in the U.S. It's a weird world.

This whole economy is being inflated in a good way. We're all living in the bubble. It's not just the VCs. Anyone with exposure to VTI or QQQ is living in an AI bubble, whether they realize it or not.

We're all benefiting from this because the hyperscalers are doing well. We all have Microsoft stock, Nvidia stock, and Google stock. We even have Meta stock, even if we don't realize it. I think Nvidia is 15% of QQQ or something like that.

Nvidia and Microsoft are 15.8%.

15%, yeah. So we're all deep in it.

And it's 35% or 36% for the top 5 or 7. I can't remember which. Yeah, no, it's astonishing.

And you know what? What do you do with that information? I mean, the question is—

Do we need CoreWeave?

Well, no. I mean, what do you do with that information? The serious question is, what do you do with that information? You know what you basically said, Jason, is that the stock-market bet is entirely predicated on the AI bet, right? Which is astonishing.

For example, the Sam Altman quote—so clever. They're going to spend trillions. I mean, do you think that's someone who sat down and added up the numbers? They've raised $50 billion so far, so they've probably spent some amount less than $50 billion. Do you think that trillion is a real number or a metaphor? Let's start with that.

No company on the planet has spent $1 trillion in capex ever, right? Do you think that's a budget estimate for the next 5 years, or do you think it's just some out-of-my-ass conceptual number?

I think everything he says is very thoughtful. I think he needs a trillion. I think we're all going to live in AI 24/7, and we're all going to use 10 times the tokens and 10 times the compute we are now in 24 months.

Help me do the math. If we're using AI 20 times more a day and 10 times more tokens, if all things were equal—which they're not—how much more do we need to spend on infrastructure to support 200 times what we're doing today? What's 200 times today? I mean, I know there's not enough capital in the world, but don't gloss over that minor problem.

The odd thing now is we're at the stage where the size of this single trend is capable, as you correctly point out, of influencing macro trends, like global aggregate trends. It's now—as you say—37% of the stock market. It's a significant percentage of total domestic capex investment.

I went and checked. It turns out we spend about $1 trillion; corporations invest a total of $2 trillion, of which roughly $1 trillion is on buildings and structures, and roughly $1 trillion is on capital equipment, right? So basically, we're talking about taking a third to a half of everything we invest in the U.S.

I don't know. I think we may have reached the point where, as I say, these are numbers as metaphor rather than numbers as real, doable numbers. The aha here is we're going to spend a lot more than we ever did. It's hard for me to imagine it's $1 trillion, and it's even harder for me to imagine that, if it is $1 trillion, it pencils out.

Now, Sam is so, so smart, and he even worked into the same comment—he said all the economists will be wringing their hands. Maybe I'm just doing that and wringing my hands. He's very good at anticipating the counter-comments and preempting them. He's such a talented guy for that.

He did that—he did the “people are going to lose money” thing. Because when people say it's not a bubble, it attracts antagonism, and people want to argue with them. It's very disarming to say, “We're going to spend $1 trillion.” Pause. “Economists won't like it. It might be a bubble. Oh, but by the way, we're fine.” It's like, “I see all the negatives, but just give me a trillion.” It's so freaking clever. Wonder who's a fundraising genius.

But the question is, can you take this trend to the bank? Because literally every discussion we've had, starting with Databricks all the way through CoreWeave, is how long can you take the trend to the bank? I don't want to be the cynic, because I do believe the common adage that cynics aren't smart and optimists die rich. So, on aggregate, you should be an optimist, but you've got to keep your eye open to the downside.

But help me with this, Rory. You're so good. I don't mean to interrupt, Harry, but if Amazon, Alphabet, Microsoft, and Meta are going to spend $365 billion in infra today—$365 billion today—and if OpenAI goes on its current trajectory—and we talked about consumer; it's the most successful consumer app ever, right? Software app ever—if they're doing $365 billion this year, can't OpenAI consume $1 trillion in capex?

No. I mean, it can't raise it. Other folks have to do the investment, but can't it be the engine of a trillion if they're at $365 billion today? And this 10x—this 10x is what we need.

No, because the $365 billion that's being invested by the 4 most profitable companies on the planet is already straining their balance sheets. Meta is stunningly having to start borrowing money, right? Microsoft's capex as a percentage of revenue has crept up to 25%. It's all knowable, right?

Yeah. It's huge.

Yeah. We're reaching the limits of what people want to finance. Maybe over 2 decades you get to a trillion dollars. I mean, even over a decade, investing a trillion dollars is $100 billion a year. $100 billion a year in a world where the most profitable company on the planet—well, I'm sure Google is more profitable than Microsoft—but all of them are spending $50 billion to $80 billion.

That implies a non-profit-making company can invest more aggressively than a $4 trillion market-cap company that kicks off cash. I don't buy it. So again, I'm not arguing the direction. My point is, I think it was “trillion” as a metaphor for lots, not “trillion” as an FY26 capex plan by quarter.

Yeah. But the math makes sense, right? Maybe it takes a decade to get to the trillion, and Sam did say we need some new financial instrument to finance it, which is telling, right? But if we're going to use AI 10 times more often in a year or 2 in our lives, and we're going to use 10 times more compute, that's still a lot of infrastructure we need, isn't it, for 100x more AI?

The good news, Jason, is there's some cool calls that you can buy tomorrow morning—

Right now.

That right now will give you the joy of that bet in financial terms, baby. Knock yourself out.

Sorry. To me, it all goes back to this very wise statement that you said, Rory, which is that the only thing that matters in this next wave is that we're going to see the transition from technology budgets to human-labor budgets. And if you see that transition, you've got $10 trillion of value unlocked, and then spending trillions over a long period of time becomes a lot more feasible or rational. If not, then absolutely not.

It's just early. That one is so interesting, and, listen, I'm only so smart. I was skeptical a few months ago. I was skeptical. I'm not seeing it in B2B software. I'm not seeing that we're going to have Marc Benioff, which will be great, right? But I don't think he's going to tell us half of Salesforce's growth is going to come from replacing humans with Agentforce today, right?

So if we look at our portfolio companies, I know we can make up some stories. We're not seeing trillions of human replacements with AI robots today. Not yet. But, man, I can see it much more than I could 90 days ago. I can see it so much.

We have 10 AI agents at our little team that have replaced 5 humans, and it's going to go up. We only had 1 person at our stand-up yesterday because it was all AIs and us. It's early, but this wave may accelerate.

Six months ago, this was VCs talking their book, right? In 6 months, this could be what all CIOs want. All CEOs could be like, “You've got to be AI-first,” not just the Tobies and the others. They all could be like, “I'm not giving you a dollar of budget until you find me an AI solution first. Then we'll talk about humans.” This could become mainstream.

I think it's a pace question. I will admit, one of the reasons I like doing this is I come on and change my mind. And Jason, your comments last week on Shopify were the bomb—the numbers, and then your comments on your own organization and how you've been able to automate have really made me pause my thinking.

I've become convinced that, at the pointy edge, the really smart companies are definitely seeing this labor replacement, right? And I agree with Harry: if you can contemplate that at scale, then you can talk about trillions of dollars. I think it's a pacing issue.

What I try to take away from here is: what am I paying attention to right now to figure out my investment strategy? It really is, for all my companies that are selling AI products—B2B products—how well are they doing with forward-leaning customers, overall in terms of driving adoption and then driving efficiencies?

They're doing well, and you can see it. It doesn't happen quickly, though. You have the high-propensity-to-buy customers like you, Jason, and then you have big-dollar customers with a much longer time horizon to ramp in corporate America. So I think the trend is real, but the pace is, to me, the big question at this stage. And it turns out, if you're spending $400 billion a year, pace really matters, because the time value of money is going to eat your ass at 4%.

I'll tell you something interesting just for venture that I realize I'm a little slow here. Now that we have 10 real AI products replacing humans, consolidation may happen before the pace accelerates.

What I mean is, we added up the nominal list price of our 10 tools, right? $500,000. $500,000 and going up, right? So you could imagine that could be $1 million by the end of the year. You know what folks are going to say? Maybe we'll just use the ones we have. Let's use them, because there's a weird—

We did the math: $500,000 already that we're using. Every B2B startup out there raised with these gigantic growth rates—not the prosumer stuff like Lovable and Replit, but a lot of the ones we look at—they're all trying to charge $60,000 to $100,000 a year. That's the list price. It's like $60,000 a year to start, plus $40,000 for onboarding. So that's $100,000, and then we want to upsell you in year 2, right?

It's so exciting, and there's so much budget out there because this is the only thing CIOs are putting budget toward, right? These AI projects. But we may see a wave of consolidation like we saw in 2022, even next year or the year after, because there are just too many AIs that are 6 figures and up. It's too many, right?

And for us, $500,000, $1 million bucks. I mean, Harry, if your team was spending $1 million a year at 20VC on AI agents, would you be cool with $1 million, or would you want to have a meeting? And you might say, like, this Gong-ification consolidation—you might want to do it.

We may see a lot of these folks who are excited about our portfolio get consolidated out sooner than happened in the SaaS wave.

I think—I think that actually, again, I try to come out of here with insight. I think that's a real insight, Jason, because I'm wrestling with that. You're seeing exactly that: everyone thinks their thing is going to save labor, but you can't all get credit for the same labor.

So you might see stories like the Rippling story happening much quicker, which is: we're going to give you 5 agents and an orchestration layer.

We'll make it all happen, because if we just do one, it's not enough. I think that's actually an interesting point, which may mean a much quicker consolidation story across these plays.

In sales, do I need an AI tool, an SDR tool, a BDR tool, a CRM, an AI CS tool? These are all 5 folks trying to get $600,000 from you, I think. We're already seeing this consolidation in e-commerce, if we want to talk about where they're becoming one tool already. I think it's a real risk for all these cool ones. We just don't need 5 of them.

Going to Rory's point, Rippling is a good example. We may want everything from one vendor in 12 months, and maybe most AI startups can't keep up. It seems so easy in the beginning, but I don't know if they can keep up with that Rippling pace.

Yeah. By the way, that also goes back to the myth of the 1-person, billion-dollar company. I think economics works really well at competing away excess profits. This trend basically says, you think you might be able to get away with that, but this other company is going to have 5 of these agents and compete against you.

I think it speaks to the fact that nobody allows billions of dollars of enterprise value to go unclaimed and uncontested. As I'm thinking now, and genuinely what many of my AI-forward B2B companies are wrestling with, this resonates: having enough surface area of value.

One of the things we learned in robotics, which was very clear there and maybe applies here too, is that it's not enough to say, “Here's this product that can automate this particular problem and reduce the labor by half or two-thirds. It's wildly efficient. It pencils out in terms of ROI, right?” But in a warehouse with 300 people, you're only automating 2 workstations. It doesn't matter.

One of the reasons we have done very well with Locus, one of our companies, is that picking is actually the number 1 thing people do in warehouses. So your automation attacks a much larger percentage of the labor. My learning has been that it's not enough to be ROI-efficient.

There are only so many things you can get your head around as the CFO or the chief of operations in a year. So you have to have a quantum—you have to be able to not just have a high ROI, but have a high ROI on a fairly large quantum of headcount. Does that make sense? Because it's just brain death to do it.

Yeah. I think everyone, for the last couple of years, has been cutting the number of SaaS apps in their organizations. No matter how well they're doing, they get around the table each year and say, “We've got to consolidate.”

For the last 18 months, it's gone the other way. They're saying, “We have to consolidate our SaaS apps, but do the AI stuff, guys. Go pick a few of those. Here's our budget.” I think it's either next year or the year after that they start to consolidate them. It's going to happen faster in this trend. It just has to happen faster.

It's not 10 years. It's not going to be in 10 years.

Yeah. Now I've got 200 of these AI apps. My team bought 100 SaaS apps and 200 AI agents. It's too much, guys. Pick your favorite 50 and we're done. You have to cut 150.

It wouldn't shock me, because there are so many paid agents running around now—more than SaaS. It's soon going to be much more than classic B2B apps. Everyone wants an agent for their department, don't they? Or 5. They want 5 or 6 agents.

Does this increasing trend toward consolidation of budgets and assets that you want to engage with harm vertical SaaS in a way that should make us think about investing in vertical SaaS today?

I don't think so. Some people often use the word “vertical” to mean functional. Sales is not a vertical; it's a function. Vertical means insurance or health care.

What I think the trend is, once you become the dominant solution, it's actually easy for you to vertically integrate. For example, Veeva in health care, in the prior SaaS generation, started off with a single-point product. Once you earn the trust of a vertical, where there might typically only be a couple hundred key customers, if you're selling them A and you have a really good product team, it's much easier to sell B, C, and D as well—all tailored to their specific market.

So I think what you'll see in these verticals is that it's good for the dominant player, and they'll just be tucking in and adding in other products, for lack of a better term. If you're the king of revenue cycle management in health care, you just have to add a whole bunch of other insurance-claims-processing-related stuff.

Ditto across all of them. We're seeing a lot, actually. It's interesting: voice bots, for example. Obviously voice is now doable. Voice is a wedge in hundreds of different verticals, where you can go because the ROI is really compelling. You used to have all these phone calls, some of them don't get answered, and now you add a person, or you add a voice AI, and it deals with all that.

It's really compelling—the insane, fast growth—but in every case, we believe that 2 years from now this is just a wedge point. You better add way more functionality such that 2 years from now you're not just handling the call; you're handling the booking, you're handling the refund, and you're doing whatever it is the customer wants on that call.

So I think it gets back to the same thing. You find these cute little wedges, but then you have to scurry really fast to add breadth to the product in a world where your competitors are doing the same thing.

If you're in Abridge, are you shitting yourself when you see Epic announcing their transcription plans to go against your core product?

If you went into Abridge thinking Epic wasn't going to do this, you are a baby, and your money will be parted from you sooner than you can say, “Gone, Baby, Gone.”

Abridge is the most impressive profit-extraction machine in the health care industry. They partnered with Epic—I think it was to have this new scribe product that Abridge has. But I'm sure Abridge went into this eyes wide open, thinking, “At some point, Epic is going to announce their own product, and you better have the best product,” right?

Are you nervous? You're darn right you are. But there's only 1 thing worse than partnering with Epic and getting smacked around by them, and that's not partnering with Epic and not even getting a chance to get smacked around. Because if you're not with Epic, you don't have access to 40% of the market share.

So, again, it turns out that no one rolls over and dies and gives you $5 billion of market cap. The whole issue, in general, will come up, maybe, for these platform risks like you talked about, and there are many of them. These are risks we used to shy away from in venture. These are 1 of the many risks we're ignoring today because the growth is epic.

We're ignoring so many risks and platform attacks all over AI B2B. There's so much platform risk, and we've given up worrying about it.

That's actually a great point, Jason. You're exactly right, because you should be sweating that. I love that platform risk is everywhere, and this is just 1 example of it. It's the whole Cursor and Anthropic discussion we had before.

Yeah, we're just—we don't—we're going to say, “Listen, gross margins. We've got to grow the top line.” The top-line growth is so attractive, we're going to ignore the top 10 traditional venture risks today. That's what everyone's doing. I'm not saying it's bad; it's just so different.

You'd debate this for weeks in 2019. “Will we be at the partner meeting at scale? Rory, I'm really concerned about Abridge. What if Epic goes and locks their API access down? Let's pass on this one, Rory. I just don't see it. I've never done well on these platform-adjacent risks. But I've heard good things about Clio, but Clio and Veeva and Salesforce—we've got to pass on this one, Rory.”

That conversation doesn't even happen on Mondays now, does it? It's stunning to me how good that was, Jason, and how true that was. You're exactly right. I remember those sentences, and you're right: we're not having those sentences as much today.

It speaks to something I think we said a few shows ago. We're way out there on the risk curve, every one of us, right now. The only thing between us and Armageddon is AI adoption. If AI adoption keeps happening, this whole wagon train keeps rolling, we're all going to be fine, and we're going to get through to California. If AI adoption slows down, then it's going to get ugly really fast across the board.

You're exactly right. Abridge's bet is, “Yeah, Epic has a product, but our product is so good that the doctors will want it.” They do, and they'll love it, and therefore they'll be able to get past the platform risk, right?

Intrinsic to that is this belief that the magic is so good that you'll be able to overcome the inertial resistance of bundling. It's not a crazy bet, because I think doctors do have a say in this. We have a play in the kind of smaller medical space in a similar thing.

I think doctors give a shit about this because it impacts their time. The entirely rational bridge bet is: if you have a good enough product, maybe medical will say, “We’ll stick with this. We won’t go with the bundled product because our poor doctors are spending hours taking SOAP notes.” But it’s all dependent on the magic and the end-user adoption being a strong enough lever to, as you pointed out earlier, pull the entire US economy up the hill.

Boys, speaking of platform risk—and we mentioned Anthropic and OpenAI—we’re going to do a Kalshi quickfire round. As we know, Kalshi bases these bets and predictions on weighted variables. Will Anthropic release Claude 5 this year, Rory? I’m giving you odds: $100 if you bet yes gets you $322 back, and if no, $100 gets you $17 back.

I would say yes. Given those odds, the need for speed, and GPT-5, I’d say, weighted, that’s a pretty good bet.

By the way, it’s interesting because it’s always hard to do on this kind of sound-bitey podcast. What you’re really saying is that in a 50/50 shot, probably not, because of the pace of their prior things, but the odds are so good, right? The only problem is then people go, “Well, you said it would happen; it didn’t.” It’s hard to remember. It was, I agree, probability-adjusted. You do it even though it probably won’t happen, but the odds push you that way.

There’s no chance, I think.

Because—and listen, I say this only because I’m a Claude power user. I love ChatGPT, but I’m in Claude at least 90 minutes a day, maybe 2 hours. I’m vibe coding the other 2 hours, so I’m either in Claude, or I’m in Cursor, or Windsurf, right?

And my point is just that there are 2 different worlds. Anthropic’s revenue is so driven by programming and coding, all that piece, right? They don’t need their—and listen, Claude, the consumer app, does feel a little dated compared to ChatGPT. Its memory barely works. Image gen doesn’t. There are parts of it that are dated, but I think the market share is so low.

They don’t need this big consumer press release like Sam Altman. What they need is Sonnet 4.5 to crush the limited advancements that ChatGPT—that OpenAI—has made, right? So that’s what they need: to win in their core market. I don’t think they need a consumer stake in the ground to do that. They need the next version of Sonnet to be awesome. That’s all they need. They’ve already won that market. Now they need to keep up. Right now they need to keep up.

Jason’s vision is ChatGPT calls itself 5 but doesn’t become awesome. Claude Sonnet calls itself 4, whatever is next, 4.6, and is awesome. Got it. You may not be correct. It’s not a crazy answer.

It doesn’t seem like they really care about the consumer app, even though I love it. They care, but it doesn’t seem like there’s a lot of—I don’t know. I follow them. They just added a Twitter account for Claude, the consumer one, in the last 30 days, and it said, “We don’t respond to this account on Twitter.” That suggests to me it’s not a priority at the moment if it’s brand-new and they’re not going to respond.

Mistral: will any company acquire Mistral this year? Yes, $100 gets you $420. No, $100 gets you $113. So again, pretty shitty odds on no.

I think if you’re not even at $100 million in ARR, you’ve got to make a play right now, right? You either have to sell, or you have to do the inverse, like with Windsurf, and buy. You’ve got to do something to get scale.

If you’re at less than $100 million in revenue and someone wants to buy your team for $20 billion today, I say take it. Spend the weekend thinking about $20 billion for your $100 million in revenue, but I would still take it. Even if you leave the VCs and the sales team behind, I would still do it.

There are moments in time, right? These moments are not going to last. These crazy deals are not going to last, and you’ve got to take them if you’re not crushing it.

It’s very telling that Jason’s question is whether they would take it, and my question is whether they would get it, because I agree with you: if you get it, you should take it. That, to me, is actually an interesting question. If you’re not at critical mass, I mean, you could be wrong—Cohere just raised a bunch of money—but it’s hard to imagine getting critical mass as a fourth or fifth player. Can they get it? I don’t know. I don’t have the data.

If they got it, they would take it. I think the question is whether they’ll get it, to your point.

I would agree with that.

100%.

Well, there are just going to be some weird deals. Scale AI's bought for whatever we call it, $15 billion or $30 billion, depending how you look at it, with billions in revenue that’s instantly abandoned, right? But yet it’s the anchoring point for price in the last round.

There’ll be other folks that are sub-$100 million in revenue, if that’s what Mistral is, that get a similar multiple or something. You’ve got to take that arbitrage. None of these things quite make sense. You’re abandoning the revenue. Does the revenue count? Is it proof of concept? What does revenue even mean for some of these deals?

It’s confusing what the revenue even means if you’re leaving it behind with the sales team. Is it just to justify the deal? Because sometimes M&A is about justifying a price. Rory, we’ve all been through a lot of deals. So many times, the deal is a justification: 2 times the last round, 3 times the last round, 10 times revenue, 50 times. But it’s just a justification, isn’t it?

Final one, Jason. This is for you, but be nice. I’ve told you we have to be nice on this one.

I’m nice.

Will Deel or Rippling IPO first? Final one: $100 gets you only $121 if it’s Rippling. $100 gets you $174 if it’s Deel.

Well, I’ll just tell you Rory probably has better thoughts. I’d want to know your thoughts, Harry, because you know the folks overall even better than I do.

On the one hand, Deel says it’s profitable. That certainly helps to go public, right? I mean, Rippling is in investment mode. It’s hard for me to believe—I mean, Parker is an incredible CEO—but it’s hard for me to believe they would IPO in anything less than 24 months because they’re doubling down on investing, right? The top line over the bottom line.

If Deel’s profitable, you have more flexibility. The quality of that—I’m not sure about Deel. I’ve always been a little unsure; I don’t have the financials. I don’t know how much is software revenue versus PEO revenue versus other revenue. I’ve heard different things, right? So the quality of the revenue might be lower, but at the end of the day, the bottom line is the bottom line, right?

Deel’s quality of revenue probably isn’t going to change that much, unlike, say, Palantir, which just before it went public had like 20% margins and then went to 70%. Assuming Deel’s consistent, it might as well IPO now while it’s profitable. But Rippling needs another 2 years in the oven to get to that. They’re getting to break-even this quarter. I don’t think Parker’s going to be there for 24 months. That’s just a guess, and it’s not a criticism. It’s just investment mode.

I think Jason’s assessment of the 2 companies is correct. One of them has a more transaction-oriented business that is more profitable more quickly. The decision to go public with that one or not is a function of internal readiness. Let’s be honest: any pending litigation angst matters. But if you could get liquidity on that one, you probably should.

I think in Rippling’s case, you’re right: they’re in heavy investment mode. It’s a very ambitious task. There is an argument that says that’s precisely why they maybe should go public if they are worried about continuing to need lots of capital. If you worry that the private markets get weird, having public capital wouldn’t be a bad thing. But you’re right: it would be less of a fully baked story and more of an ambitious, continuing-to-grow story.

So if you believe that the markets stay as they are for the next 2 years—broadly receptive—then the logical thing is what Jason said: Deel goes now and Rippling goes when it’s ready, right? If that were to change, it would be because Rippling decides preemptively, “I’d just like to take one risk out of my life and have $1 billion in cash and a tradable stock.”

Again, as I say, I’m always a little—you’ll see a continued bias here—a bit of a scaredy-cat, and I always err when the capital markets are this frothy. There’s a little part of me that always wants to de-risk anything.

Chaps, brought to you from Greece. This was a pleasure. My mother is beckoning me. Otherwise, I’m going to get—

**Speaker 3**

Have some baklava on us. I’ll Venmo it to you.