SpaceX启动史上最大规模IPO｜OpenAI提交上市申请｜Uber裁撤23% HR

Rory O’Driscoll

But the one thing we know about Elon for the last 30 years is that when he hears the words “more risks,” he says, “Yes, please. I’ll have two.”

Jason Lemkin

I think the IPO nominally will be a dud. I don’t think it will trade up dramatically.

Rory O’Driscoll

There’s always money when people aren’t afraid. When things get scary, it’s not that money runs out; it’s that money gets scared.

Jason Lemkin

I’m contemptuous of startups that need to be fat. I’m like, “What’s your excuse?”

Rory O’Driscoll

In any business, there are only 2 things that happen. People are either making stuff or selling stuff.

Jason Lemkin

If AOL becomes the next hot thing, I mean, these guys are fucking geniuses.

Rory O’Driscoll

Anyone who hasn’t churned from AOL by now isn’t churning until they die.

Okay, we are back, and what a week it is. We have the largest IPO roadshow in history. We have to start with SpaceX. We’re speaking on Tuesday, and obviously SpaceX is going out on Thursday. There’s going to be some time discrepancy there, so what we say will be scrutinized in intense detail by the time you’re probably listening.

Rory O’Driscoll

True. Though one of the things is, there are usually 2 questions you’re asking at this point: What’s it going to price at, and what’s it going to trade at? The funny thing is, unlike 99% of IPOs, the first question has already been answered.

Elon has decided that instead of doing price discovery, where the bankers build the book and then pick the price and announce it right at the end, the night before, the IPO pricing typically takes place the night before the trade opens. Then everybody who participates in the IPO gets to buy at that price, and it opens the next day at whatever price, up or down from that.

In this case, Elon has decided in advance of getting anyone’s input that the number should be, I think, $135 a share, which values the company at $1.8 trillion. In other words, he’s short-circuited the price-discovery process. Instead, he’s saying, “We’re not doing price discovery. I’m telling you the answer, and the only question is, how much of it do you want to buy at that price?” So one thing we can’t get wrong is that, Howie.

Is that a wise move? He’s leaving a lot of room for the markets to move in between that. That’s why you normally leave it as close as you can, because you don’t want an Iran-Israel conflict or a Broadcom moving markets and then putting you in a precarious position. It feels unwise, but Elon is a master, so I’m not going to—

Rory O’Driscoll

Calling someone unwise who’s about 2 days away from becoming a trillionaire is a big call, Howie. But I think what it is is, it’s no surprise given Elon. It’s ballsy. You’ve got way more room for error. You could be wrong to the high side, you could be wrong to the low side, or you could leave money on the table. Maybe all the orders flood in and you’ve left money on the table. Maybe, on the other hand, you’re struggling to get the orders in, it feels very high, and it opens down.

It’s more risk, but the one thing we know about Elon for the last 30 years is that when he hears the words “more risks,” he says, “Yes, please. I’ll have 2.” This must appeal to him. It’s like, “I’m telling you the answer in advance, and I’m taking the risk.” That’s how he became a trillionaire. Is it wise? We’ll see on the day.

Jason Lemkin

It’s a huge amount of capital, Rory, but if it’s really only 2x subscribed or oversubscribed—I’m not even sure oversubscribed is the right word if it’s only 2x—plus Elon picking the price, that suggests to me this one won’t pop. I do believe the day traders will drive it up ultimately, but it doesn’t feel like there’s excessive demand at 2x. In most IPOs, it would be almost insufficient to close the IPO.

Rory O’Driscoll

Agreed. There are 2 separate things in that, Jason. You’re right. One is that the decision to pick a fixed price logically reduces the probability of a pop, with no other information, because the whole point of the banker process is to pick the price the night before that allows the pop the next day. You simply aren’t doing that because you don’t have the information. You’re right.

But then the second thing you added is some information that has come out, which is that, to date, the book is 2x covered. Your comment is that that feels low compared to normal IPOs, correct?

Jason Lemkin

Traditionally, you want 8x to 10x to get the deal that you want, but you’re not raising the vast amounts of capital Elon is raising, either.

Rory O’Driscoll

It’s hard to get 10x oversubscribed on $75 billion. To be really direct, what you’re saying is that you’re pricing something at a fixed price that isn’t taking demand into account, where you’re looking for a very large amount of money such that you only have a small amount of coverage.

You’re right. You look at those circumstances and say, “There’s a nontrivial chance that it pops to the downside.” Is it 30%? I don’t know. But if you think about it, normally bankers bend over backward to try to have the damn thing pop, right? They’re trying to get a 10% to 15% pop, and nearly 90% or more of the time it pops. But 10% of the time, it breaks the IPO. They get it wrong. Even trying to fix the game, they get it wrong.

In this case, they’re not even trying to fix the game. Time will tell on Thursday night whether they’re too high or too low. But there is, by definition, probably a higher-than-10% chance that on the day, people go, “Everyone who put in for it put in for it,” and it’s not impossible for it to trade down.

It’s just that if you use mechanism A, which is designed to create a pop and works 90% of the time, and now you use a mechanism that doesn’t have the information to allow you to make a pop because you’ve used a fixed price, then, by definition, the probability of it going wrong goes up.

That’s all.

Jason Lemkin

I think what will happen, if that’s accurate and 30% goes to retail, is that the IPO nominally will be a dud. I don’t think it will trade up dramatically. But I do think every time there’s great news—more satellites in space for SpaceX, more things—it will begin an inexorable rise. People will be excited, especially if the upside is tied to potentially significant revenue, as the last announcements have been with Anthropic and Google.

I just don’t think it’s going to pop that first week. I think there aren’t enough buyers out there in this universe, or at least in this galaxy, at this price, at 2×.

Rory O’Driscoll

I’m going to step back. I kind of hate that we got—and I caused this, so I apologize—into the technicalities of the IPO, because zoom out a million miles here. This is amazing. This is an amazing technical company. It’s the iconic company of its generation. It’s going public this week.

It’s a huge moment. What do you say to Elon? Congratulations. What do you say to everyone involved? Congratulations. It’s a wildly impressive company. Look, I am skeptical of the valuation, but step back. I’ve watched some of the launches on little YouTube, and I’m like, they’re just so impressive. The whole thing is so impressive.

At the risk of sounding a little partisan American—sorry, Howie—this really is an only-in-America moment, right? Who else is going to find the capital to take that kind of risk and go for it? And frankly, who else is going to have a big enough capital market to fund it and a big enough addressable market to sell to? It’s a great outcome. It’s an amazing company. It’s a real asset to America.

End-of-day-1 prediction and end-of-day-90 prediction.

Rory O’Driscoll

So you really are determined not to let him have his great moment. See, you’re like those commentators in politics who won’t talk policies. All they want to talk about is the horse race. All you want to do is talk the horse race here, Howie, because you know that’s what sells. You’re such a little media slut.

I don’t think it’s knowable on day 1. I think all 3 scenarios are equally likely: one-third, it just goes down because there are weird pricing mechanisms, so they don’t have demand; one-third, it’s flat, because whatever; and then one-third, to your point, retail enthusiasm, it goes up. There’s no information here.

Now, I will make a call, though. I think over the next 12 months, I doubt it will retain this price. There, I will make that statement. I disagree with Jason. I think fundamental value here reasserts itself. I mean, there are 2 reasons I say that.

One is, again, I always go back to the base rate. The base rate on IPOs in general is that you do see quite a lot dip. The base rate on IPOs at more than 10× forward sales is even more of a dip. The base rate on IPOs at 70× forward sales—there hasn’t been one, but you’ve got to believe there’s a dip. So I think valuation reasserts itself over the medium term, and the probability of it being higher than the IPO price in 12 months, in my gut, is lower—significantly lower.

So I’d say I haven’t a clue about day 1. It’s a tactical thing based on the mechanisms, and I think over the medium term, this amazing company might, shock horror, only be worth $1 trillion instead of $1.7 trillion, and it’s still a huge win.

Jason Lemkin

Two thoughts. I don’t know what you guys think. One is, listen, there are many great IPOs, like Facebook and Google, that IPO’d with a whimper, right? It would not surprise me if this IPOs with a whimper at the end of the day. It doesn’t matter for SpaceX. We will have multiple layers of generational wealth created.

Elon will get his liquidity. It’ll all be great whether it’s a nothing-burger IPO or not. I guess it might hurt OpenAI the most because they’ve been so aggressive on their valuations and so aggressive on their capital raises. If that means they have to cut back their aspirations for the amount of capital raised, the valuation maybe doesn’t matter as much, but they are related. That could be the biggest negative effect. They need so much capital, too, that it could take some of the wind out of OpenAI’s sails.

The other thing I’ll just say briefly: before I got on a plane—I’m in Hong Kong as we record this—I spoke to one of my LPs, who’s getting lots of cash here, got cash in Cerberus, and is getting cash in all these other deals. It ties to a conversation we had before: the expectations are so high now for performance, and I think that will permeate through the ecosystem.

I do think it’s a minor negative, but I do think it’s something for founders and others to understand: it’s not a free lunch. The bar will continue to go up after these events when LPs are looking for 7–8× routinely from GPs, which is hard to do outside of anomalous periods of time. The expectations that GPs will have from founders continue to go up.

As this LP said to me, “I don’t know that little $5–$8 billion IPOs really make the math work anymore.” We’ve talked about it, but to hear it from a large LP, it echoed in my ears how the bar goes up.

Rory O’Driscoll

I don’t think you can take a once-in-a-decade event and start extrapolating it as a norm. I think in life you should take this as the once-in-a-decade, maybe—

Jason Lemkin

But there are 4 or 5 of these once-in-a-decade events. There’s going to be Anthropic, OpenAI, SpaceX—

Rory O’Driscoll

Well, hang on. Go back. It’s interesting that you say that, but of course, the opportunity—the once-in-a-decade…

I mean, SpaceX was once in a decade. It was last decade. Reminder here: Founders wrote that check in 2008, right? It’s now 2006. It’s 18 years ago. So for that kind of huge return, I mean, yes, there’s been a 10× since 2008.

Jason Lemkin

When was the Mercer seed check written again? Remind me.

Rory O’Driscoll

The Mercer seed check? 3 years ago. Yes.

Jason Lemkin

So maybe they do happen more than once a decade. They seem to be—these decades seem to be shrinking.

Rory O’Driscoll

I think you’re going to have 1 $1 trillion outcome from the last decade, and 2, it looks like, from this decade if it all happens according to plan. But my point is, yes, you probably can’t assume 10. You don’t run your business on the expectation that every check you write is going to be a trillion-dollar outcome. If you’re really smart and you get 1, you should say, “Yay.”

So I think $8 billion outcomes will make everyone perfectly bloody happy. Obviously, unless you have a $10 billion fund, in which case it doesn’t. Fund size dictates the amount of market cap it takes. It’s a Josh Kopelman thing from ages ago—the venture arrogance index, whatever. The bigger the fund, the bigger the deal there has to be to make it work. There’s nothing surprising here.

Will this have knock-on effects in terms of LPs investing more directly and an increase in fund investments from LPs? You’ve got Ohio Teachers, who I think will make over $10 billion from their SpaceX.

Rory O’Driscoll

To be clear, I know you think the entire Midwest is the same, Harry, but I think it’s Ontario Teachers’, right? At this point, you’re conflating Canada and America, which is an easy mistake to make because we’re making it ourselves, starting with the president. It does begin with O and it’s kind of in the middle, so I understand your ignorance. But let’s go back—

Jason Lemkin

Maybe he’s just a big Fallout player, too. Who knows?

Rory O’Driscoll

Easy. But the bottom line is, yes, Ontario Teachers’ Pension Plan nailed it. I mean, they’re going to make a magnificent return. And there are a bunch of others. It’s great. University of Washington has an extremely savvy CIO who…

And by the way, Washington, Harry, just to confuse you further, is not in either Washington State or Washington, D.C., but we’ll keep that for now. But yes. Look, by definition, these are going to be the best co-investments ever because it’s the best deal ever. I mean, there’s nothing surprising in it.

But I think our LPs are going to come back and go, “Those that will get liquidity from this go, ‘Hey, we’re going to reinvest more,’” and will all that brethren be like, “Hey, we’re going to join this because we want the next generation, even if we didn’t have them?”

Rory O’Driscoll

Yes, because everyone’s just going to go, “Wow, that looks amazing.” As I say, again, it’s back to the extrapolation from the unique event. Of course they are, because it’s going to be amazing.

I saw, I think, in The Journal this morning—not University of Washington, Washington University. Again, I’m now getting confused. It’s 10 or 15% of their endowment. It’s awesome.

Yeah.

Rory O’Driscoll

It’s awesome. This is the best venture capital deal ever in terms of absolute return, and anyone involved is going to do really well.

Speaking of a once-in-a-lifetime or once-in-a-decade moment, as Rory very articulately put it, another once-in-a-decade moment is obviously OpenAI filing to go public, not so confidentially. Anything to say here that we haven’t covered?

Jason Lemkin

The thing I don’t understand is—maybe it’s a question for Rory—because I don’t get it. Other than the Captain Obvious element, what’s the point of hedging your bet on the timing but filing? I mostly get it, but I don’t totally get it. I only half get this: “We may want to stay private; we want flexibility, but we’re going public.”

Rory O’Driscoll

I think all they’re doing is being a little smart and managing expectations, finally. I read that as, “We’re filing to go public.”

In a perfect world, we'd love to go public as quickly as we can, but if it's delayed for whatever reason, we don't want to have a whole bunch of negative stories saying, “See, it's slipping.” So if you preemptively manage expectations and say, “We're filing, but we're not committing to a timeline,” we're not all going to be sitting here in late October going, “They said they'd be going public in early November. WTF is going on?”

The big aha here, and we said it 2 weeks ago, is everyone's suddenly gunning for the door. At some point, you need the capital markets—the public capital markets—because the scale involved is such that that's where you have to go, and everyone's just hit that point and they're going for it. I think, going back to your comment earlier on SpaceX, it feels like the market is very risk-on. We had that little dip last week, and then everyone got over it in 2 days, so it's as good a time as any. You keep cranking while you can and see if you can get it done.

I'm sure that they made that caveat—“We'll take our time”—but they made that statement in the press department. My guess is that in finance and legal, the mandate is, “Get this puppy done as quickly as possible so we have maximum optionality.” It's worth pointing out, by the way, that the SpaceX S-1 went through the SEC very quickly. Normally, that's a painful process with multiple iterations, and it seemed to happen here extraordinarily quickly, probably because we don't regulate anything anymore. So go team. This may all process through really quickly, in which case, brace yourself for a fun fall.

Rory's on fire this morning, eh? Gosh. I really want to touch on something beneath the product layer for OpenAI, which is that Sam Altman has been driving toward persistent and always-on AI. They shipped Dreaming V3, the biggest memory architecture upgrade since launch. I'm intrigued, Jason in particular, to hear your thoughts on this. Is the future of AI continuous, persistent, 24 hours a day, a fabric of life, always on, in your mind? How do you see this?

Jason Lemkin

I think we all believe it. We can make fun of Apple this week, basically repackaging Gemini and giving up on AI, if that's the way we want to view it, but that's a little piece of wanting, ultimately, AI to be persistent 24/7. We do want this. We already live little hints of it, and it's pretty silly that AI, for the most part, lives in our browser, right? If you think about it, that's very dated. It's so dated that we still use browsers. I mean, who would have thought we'd still live in the era of Netscape in so many ways?

I do think it's exciting. I do think, as this show continues, whatever we would want to call it—the tokenpocalypse, I think—it will morph into just standard business practice, right? At some point, the IT budgets can only be so large. There's only so much. Even if we lay off half of the employees, employment keeps growing, so we're going to have to manage spend. There is a conflict, but I do think we're going to look back in 2 years and think of this nonpersistent AI as almost archaic, almost sort of desktop-like.

Rory O’Driscoll

Yeah, because you threw in a lot there.

Rory, you pulled several faces there. For the audience listening, Rory's facial nuances—

Rory O’Driscoll

No, it's just that Jason, as he often does, covered a lot of different things, and I'm just processing through it more slowly. On the memory thing, and kind of what Jason has said, it just totally makes sense, right?

The question is, if you step back, you have the core models, and then you have what people are calling the harness, which is all the stuff around it to make those models effective. Part of that—it can either be in the model or, in theory, it could be in the harness—is just understanding memory. The impact of that, and I think why Jason went to the token economics part of it, is that part of the benefit should be that you get better answers with memory, and part of it should be that it's more cost-effective in terms of tokens, because you're not passing through all the context all the time. I think a lot of the trend with these harnesses will be adding stuff to minimize your cost on frontier models, and part of that will be having memory, right?

It also leads to a better experience, right? I actually just went in and tried to see: has it been switched on in mine yet? It makes a ton of sense. You should know who I am after I look up 58 20VC podcasts. I'm probably here to look at my 20VC podcast research, you guys, so it just makes a ton of sense.

Rory, most people know who you are now as well.

Rory O'Driscoll

No, I'm saying, but—

You're famous.

Rory O'Driscoll

My OpenAI sometimes doesn't. So, yeah, it's absolutely one of the necessary to-dos, and they're doing it, and it's great.

Rory, you pulled a face when Jason said about Apple giving up on AI with Gemini. You all right there?

Rory O'Driscoll

Yeah, moving on to that, that was an interesting one. Ben Thompson at Stacker did a really good piece on it this morning that I was reading. To some extent, they're giving up in the sense that they're paying Google $1 billion to use its model as the default model, but as a reminder, Google pays them $20 billion to be the default search engine, so it's a minor offset.

I give them credit. I actually think that they're making some progress. Yes, it would be better if they'd had their own model, but they're making progress on the use cases that make a ton of sense for the consumer. I think the amount of context you have when you're on someone's phone is such that they can deliver a unique and compelling consumer experience for the kinds of things they demoed around knowing context. It's like your memory comment, Jason: knowing which Rory it's talking about, knowing your calendar, knowing everything, and delivering a much better experience.

Now, should they have been able to do it with their own model? Yeah, but the bottom line is that they control the handset, and for the consumer it's a pretty powerful product. So I think they're in a good position to make progress. I don't think they're giving up. I actually think they're pragmatically saying, “We kind of screwed up by not having our own model, but that's actually not what matters for us at Apple. What matters for us at Apple is delivering an amazing experience to our consumers, because if we do that, they'll keep buying handsets, and if they keep buying handsets, we can probably afford to give someone $1 billion a year.”

I give them credit for getting their shit together. It is stunning that Siri has been so bad for so long, so I think actually trying to fix it is just awesome. I give them credit for taking a step in the right direction—that's my takeaway from it. So, it's the opposite. I don't think they're giving up. I think they're doing what it needs to win coming from behind, and they have a great position.

I think it's interesting. The person you have to think about this a lot with, obviously, is if you're OpenAI versus if you're Anthropic, because Anthropic has made the enterprise bet and OpenAI, in part, has made the consumer bet. I like my OpenAI subscription, because I sit at my desk and I do research. But for a lot of consumers, it was a great line, and credit to Ben Thompson: he said it very clearly. I've thought it, but he said it clearly. He said, “Consumers don't want to work. There's not a big market for consumers in their non-working life to do a whole bunch of complex research or use AI for productivity. They just want delightful experiences because they want to relax and be entertained.”

I think, actually, the consumer space is going to be a tougher space for OpenAI. The enterprise space has really been validated, because an enterprise is all about automation and efficiency. In a consumer space, it's about experiences. Apple's well-placed to do that. OpenAI has to compete with that and compete with Google, and it's a tough space, especially if Apple is getting its shit together.

Well, speaking of consumers not wanting to work, soon they won't have to. Uber cuts 23% of HR.

Rory O'Driscoll

It's just to make Jason happy.

Sorry, I'm so sorry. It's obviously people losing jobs, and it's terribly sad, but I'm the one who fucking said, “No great CEO likes HR,” and everyone got angry at me. Then everyone starts cutting HR. Anything of note here from Uber cutting 23% of HR, remote work rescinded, 3-day in-office mandate? The company denies AI played a role, despite 95% of engineers using it daily. Anything of note there?

Jason Lemkin

Well, look, HR and recruiting—which, let's consider them different—are the easiest things to cut. You always see any big tech leader stumble a little bit, and they lay off 30% of their recruiting department. Well, you often want them back. I mean, it makes sense on paper, right? The HR one will be interesting.

We've put out a call for someone to report to our AI VP of marketing, and I've gotten my head bitten off a lot on social media for that by people not really listening to what I'm saying about that.

But I do think HR is one of these areas where many parts of it will be better managed by AI. I think an AI can be a better VP of HR for certain parts of the job than a biased human. I think there are advantages to having an AI VP of HR—not that I want to get rid of all the humans or even lay people off, but an AI VP of HR can evaluate every single thing you’ve ever done, every little bit of your work, and all of your issues.

An AI VP of HR can figure out, hey, maybe it really is your idiot boss, Jason. Maybe that really is the problem. It’s not you. An AI VP of HR can find out a lot of things, process them, and ask questions. I think it’s an under-discussed area compared with other areas, but it should be massively disrupted.

Rory O'Driscoll

The big-picture question in all these areas is: how much efficiency do you get? My gut? It felt like 23%. I doubt everyone is automating and saving 23% using AI in the non-engineering departments, because adoption there isn’t as strong as in engineering. Do I think there’s some? Of course I do.

So my bottom line is, I think my guess is that some portion of this is, quote-unquote, AI automation. I doubt it’s 20%, because I’m always calibrating off what percentage—it’s the Dario number—of quote-unquote knowledge work is going to be automated. And it’s knowledge-work tasks and then knowledge jobs. Is it 5? Is it 10? Is it 50, as Dario has said? Twenty-three percent felt like a lot, but whatever. Again, what you don’t know is how much of it is just that there are too many folks there and they’re partly rationalizing. So it’s a data point.

I think the other data point from Uber is far more interesting, which is not the AI for HR but the AI for autonomous driving. They continue to make progress on autonomous driving. They’re actually rolling out more autonomous-driving experiences in Europe, in Madrid, I think, right? They’re partnering, I think, with WeRide or some of the technology providers.

If you want to talk automation, driving is one of the biggest targets in terms of the number of humans who do that job. When you see Uber making experimental progress on robotaxis in Europe, that’s obviously something to keep an eye on. It’s worth pointing out that this stuff is still moving way slower than I think people anticipated. It hasn’t been Waymo in San Francisco resulting in Waymo being everywhere within 6 months. It’s been a long, steady progression for Waymo.

Uber’s doing what it should do. The Travis Kalanick devotees would say the cutting of their autonomous project in 2016 or 2017 was a fatal error for Uber. I’m not sure. I think 10 years later they can pick up the thread, which is what they’re doing, and catch up on that, because it’s not like the technology tipped like a domino.

I think they’re smart to now start pushing robotaxis and partnering with technology providers. The question on the Uber stock is always, “Oh my God, is robotaxi existential?” That’s the bad scenario. The good scenario is that lots of people build robotaxi technology, and Uber is in a wonderful position to be the coordinating thing because it’s the app we use. If they just add 10,000 robotaxis to the fleet, then things continue just fine.

Frankly, it’s good to see the Europeans do something. I say this respectfully, Harry, but typically Europe is the slow technical laggard, especially on stuff like that. So, go Madrid.

Should we discuss Revolut at $115 billion? $115 billion, you know?

Rory O'Driscoll

Amazing.

Yeah. Thank you.

Rory O'Driscoll

I think you’re doing that defensively. You felt I was dissing you and Europe, and you’re basically implicitly saying, “Oh, look at Revolut. It’s amazing,” correct?

Correct.

Rory O'Driscoll

And it is amazing. And you know why it exists? Because the European banks, unlike the American banks in general, are so crappy. There’s a reason that Revolut is worth $115 billion, because the incumbent European banks were fat, dumb, and happy, making margin off their customers. There’s also a reason why Chime is worth $5 billion. That’s still a great outcome, by the way. That’s because the U.S. banks, on average, are a little more efficient.

That’s also why New Bank is such a valuable business—

Rory O'Driscoll

Agreed.

—because it’s multinational.

Rory O'Driscoll

Because the Brazilian banks were inefficient. I think all these fintechs really are proven markets. It’s a function of how egregiously priced the incumbents are. Europe, especially when it had no single currency, had all this foreign exchange. Of course, you guys aren’t into Europe. You had the FX charges and all this transactional bullshit, and Revolut just blew a hole through that.

I think it’s amazing, and I know you’re a big fan of the CEO. I wish him all the best, and I hope he pounds those old-school European banks into the dirt. At some point, we’re going to have to deal with the fact that the largest bank by market cap doesn’t do much lending, and that’s actually going to be a real problem in the aggregate, because the whole point of banking is to recycle savings into lending.

Right now, Revolut’s not a long-term lender, but that’s by the by. They’re killing it.

I’m fascinated to hear Jason’s thoughts on this one. What’s dominated my Twitter over the last week is Greg Isenberg’s original tweet about a horror story of a venture fundraise. It led to hundreds and hundreds of founders sharing horror stories, including the Cloudflare CEO, who talked about his experience with Khosla and Vinod Khosla. Jason, I’m really intrigued to hear your thoughts on this one. I’m sure you have some. How did you feel about this slew of founders bluntly saying how terrible a VC experience they had in certain cases?

Jason Lemkin

Well, I’d say a couple of things. First of all, when I was in the most intense phases of being a founder, I had those stories, too. We forget how deep some of these things cut—these slights. Folks who are friends of ours now, whom we co-invest with, I thought terrible things of them at the time. Literally. One that we both know really well would constantly use me just for due diligence on another investment. Constantly.

Now I’m pretty zen about that crap. To founders, I’m like, “Just take the meeting and do reverse intel.” If you’re just being used for a competitor, then sit down with them and find out about your competitor. Get the exact information. But, man, that stuff really burned me.

First of all, with the whole thing involving the CEO of Cloudflare, just remember: founders hold grudges. I still do. I’m just getting over my founder grudges now. Founders hold grudges in a way that VCs actually, I think, don’t, because if VCs miss the deal, they’ve got to find another bus, right?

Having said all that, get over it because it’s sales. The only thing to really hold a true grudge over is if you got fired. I think the folks from Uber who hate Benchmark deserve to hate Benchmark. I think there are others. But if you were treated poorly during the fundraising, get over it. It’s sales. Have you never sold? This is what I say to people.

Have you never sold anything? Have you ever thought a customer deal was going to close and it didn’t? Have you ever had a prospect tell you, “Rory, of course we’re going to buy by the end of the quarter,” and then you send them 28 emails and 87 texts and the deal never closes? How is selling stock any different from selling anything else? So there are a bunch of issues to separate: the grudge, the firing, which is a niche issue, and learning to sell, man. Grow some.

Rory O'Driscoll

In one sense, you’re right. But I think the difference for the founder—and I think a ton of what you said was super insightful—is that the founder in this case isn’t selling their product. They’re selling themselves. So I think you’re right about one thing: the rejections cut deeper. There’s no doubt.

Even on my side, I remember a VC 30 years ago saying to me, “You never forget the LP turndowns.” Thirty years later, he’s so right. You remember those people who turned you down. It’s just a personal thing, because you’re not just selling your product. You’re not selling Ford cars on the dealer lot; you’re selling yourself.

So I totally agree with you, Jason, that you do have to grow a pair. You do have to get a thick skin. But I totally get the way founders feel. Even if something doesn’t go wrong in the process, I totally get it. Rejection sucks. And so, that’s the founder side. I thought you were super sympathetic there.

Just to put the other side of the table, every venture person is in a business where we turn down 99 out of 100 deals that we look at. Rejection is our default M.O. That’s why I always wrestle with these ratings businesses—the kind that rate VCs. It’s doable, and I think there actually are appropriate ways to do it. But you do have to remember that the default is a no, and it’s really hard to have high customer satisfaction when 99 times out of 100 you’re going to tell the customer no.

It’s why no one ever loves the bank they apply to for lending money, because a well-run bank turns down 5 out of 6 customers. No one likes that experience. Rejection sucks, so it’s set up for failure out of the gate. Sometimes, in the course of turning down 200 or 300 people a year, you get some stuff wrong.

Jason Lemkin

What was interesting is that Matthew was really upset that Vinod asked him to consider getting rid of Michelle, whom we know and who is great, and his CTO, and giving him the shares.

It was not stealing her shares, which I think was misinterpreted. He made it a suggestion and a pitch. Listen, I’m a super fan of Michelle. I would not make that suggestion.

But let’s step back for a minute. We’ve all had those meetings with founders where the team is very unbalanced. Am I Vinod? Would I say it that way? No, but you might know me well enough: I almost would, in a different situation. I almost would say that to a founder. I just wouldn’t do it during a pitch. I would just say it’s not a fit for me.

But I find myself constantly, post-investing, being the only one who would say things like, “What are you going to do with your co-founder? Rory’s just not committed enough. He’s not getting it done.” I think his directness is interesting. It bothered the CEO of Cloudflare so much, but in a way, it was just his read of the team. I think it was wrong, at least for one of them, but it was his read of the team.

Rory O'Driscoll

And by “wrong,” you mean incorrect relative to the subsequent outcome?

Jason Lemkin

Well, I know Michelle. I don’t know Michelle that well. I think she’s a great founder, so I would keep her. But the fact that VCs go in and see that the founders are not equal in terms of their commitment and skill set, right?

Rory O'Driscoll

Look, it’s clear, given the superb outcome, that whatever Cloudflare had, it shouldn’t have been touched one little bit. It should’ve just been left to do exactly what it did. It’s a great outcome.

But you’re right. Again, Jason, you raise a good point. You go in, you see things, and especially at the earlier stages, if you think the team is wrong but you want to do the deal, then that’s a really tricky conversation. As well as when you’re as successful as Vinod, you’re like, “I could take 3 meetings and slowly and delicately get to this point, or maybe I’ll just say it.”

It’s also worth pointing out that he said very clearly he doesn’t believe that happened. So I think, stepping back, I don’t know if it’s a useful way to rehash. The more successful you are, the more meetings you’ll have. The more meetings you have, the more likely some of them go wrong, especially if you’re direct, and Vinod is nothing if not direct. So stuff happens.

As someone pointed out, he was on the Midas List the first time for Juniper, and he’s on the Midas List this year for OpenAI. There are 30 years between those 2 events, so he must be doing something right overall. That’s still not to say that on an individual day, you can piss people off.

I’m sure I look back across 300 or 400 turndowns a year for 30 years, and I know there have been some where I wish I’d handled it differently. There have been 1 or 2 where, literally at the term-sheet level, I wish I’d handled it differently. It happens. It’s not ideal.

If you’re aware of it, you apologize later and say, “Look, I got that wrong,” and you just have to move on. Some element of breakage is inevitable.

I have to admit, Rory, I disagree with you. I’ve been turned down by lots of LPs. The best way to have revenge is to forget they even existed. I’m being a dick here, but a lot of them ping me now. I’m like, “Wow.” When they turned me down when I was 21, I’m like, “Whoa.” And you’re like, “Who are you?”

Rory O'Driscoll

Yeah, maybe early on you remember, but you’re right. Over time, to Jason’s point, you develop a thick skin. And you’re right, I remember much less the turndowns in Fund 7 than in Fund—

First independent.

Rory O'Driscoll

The second fund was Fund 3. I do remember in Fund 3, our first independent fund, which we foolishly timed literally for the week of the Great Financial Crisis in November 2008, getting turned down 3 times in the space of an hour. So I do remember that pretty vividly, but life goes on.

Okay, so again, big milestones for Lovable and Cursor this week. Lovable, literally just before we came on, hit $500 million in ARR. Cursor has hit $4 billion, and it’s targeting $6 billion at the end of the year. Jason, you’re the man of the hour for this one.

Jason Lemkin

You’re the coder.

Any thoughts on this?

Jason Lemkin

I think there are 2 different things you said. One was about the scale of these companies, which we’ve talked about. I do think the headcount thing is something that we’re still learning about.

When we started this show, we were in an area where folks were very lean and growing very quickly. But the question was, does this normalize over time? As you approach scale—as you approach $100 million, $200 million, $500 million, or $1 billion in revenue—will startups get fat again? Do you just need these layers?

I can think of a number of hot AI startups that are getting pretty fat, especially on go-to-market teams and others. But we’re seeing more and more examples to the contrary, and it is disruptive on many levels if you can stay as efficient as these guys are.

It is disruptive to investing. It is disruptive to employees because it will shrink the number of these great roles, and it will increase compensation, right? To the ClickUp point, to Zeb’s point, I’m doing layoffs to give a million dollars to a handful of folks. Lovable can pay its team whatever it wants. With less than 200 employees, it can pay whatever it wants.

But, man, if this becomes the steady state for startups—and maybe it was in the old days, maybe in the old days of Microsoft it was true—it’s just so different if they’re not going to reflate. That’s what I think about, because it’s not a lot of people. It’s not a lot.

What people don’t understand—I know Replit a little better than Lovable, but they’re the same—is that they’re pushing out a lot of code. One thing you could say is, “Oh, it’s easy because they only have 1 product,” right? That would be a comeback that I think works a little bit. You don’t have to have 22 products like Datadog or 7,000 like Salesforce.

Well, maybe, but these are pretty complicated products. You’ve got a database, hosting, management, SEO you’re running, and more. These guys are pushing out more features than any of us did in our entire lifetimes a generation ago because it’s the most brutally competitive space there is.

I don’t think we’re dealing with people who aren’t working. These folks are working incredibly hard, and they’re incredibly productive. And if you want to have some contempt for VCs, tying this together, I’m kind of contemptuous of startups that need to be fat. I’m like, “What’s your excuse? What do you need another 200 people for?”

When I’m at a board meeting and a VP says—or, they’re all C-levels now, right? A C of something. There are no VPs anymore in startups. They’re all Cs—and they say, “Well, I could do that, but I need another 50 or 100 heads. I need another 10 or 20 or 40 million,” I just think that person should go.

Rory O'Driscoll

First of all, I broadly agree, but the only pushback I’ll make is this. We’re saying, “They’re amazing that they can do this with only 146 employees.” But remember, if you’re spending 50% to 70% of your revenue on intelligence from Anthropic or OpenAI, you don’t have the option to also have 50% to 70% of your revenue go to employees because there’s just not enough room in the percentages. They’re different businesses with different business models.

Jason Lemkin

They are, but you have the choice of who you invest in or who you work for, right? We vote with our legs and pocketbooks, right?

Rory O'Driscoll

No, of course. And this is actually one of the core challenges many of these other companies are going to have. If you can be 1 of the 146 employees, that is, I agree with you, Jason, 100%, getting leverage from this AI such that your economics are compelling because you’re 1 of a small group of people making a lot of money in a business that’s leveraging technology to have a very high revenue per headcount.

It means we can pay you a lot. That’s a far better place to be as an employee, you’re right, than one of 90,000 employees at Salesforce. You’re exactly right, because you’re not getting leverage from the models and intelligence.

And this is the question: how much will be labor, and how much will be intelligence? This is kind of the question of what the split will be. What you’re seeing, to your point—and I’m sorry I’m rambling on this, but it’s clear in my head and I want to get it across—is that in businesses that are using a lot of intelligence, and I’m using tokens as a proxy for that, small numbers of people can achieve a lot and make a lot.

Those are better places to be as an employee, and often as an investor, than slogging it out with 10 times the employees, not a ton of new leverage from AI, and being stuck in the 2010s ground game.

Jason Lemkin

Which sucks.

Rory O'Driscoll

That’s what you’d want to do if you could, as a founder, as an employee, as an investor. You’d want that. If you could, that’s the model you’d want.

That’s where I’d want to go work. I want to go work somewhere where I’m empowered, where I’m—oh, I’m 1 of 172 people at $500 million in revenue? I’m happy.

Jason Lemkin

Yeah, but to your point, I do think—and I want to call it out—I do think as you start to develop an enterprise motion, and you implicitly said it, you’re probably talking about the foundation models that are building big go-to-market machines, because they have to.

Rory O'Driscoll

We are going to see way more buyers. I don't buy that there's not going to be an infinite number of—

Jason Lemkin

No, Rory, I, it's not just, I mean, it's the Gore, it's Hobby—

Rory O'Driscoll

Agreed.

Jason Lemkin

It's your Sierras.

Rory O'Driscoll

Once you're selling to enterprise, this idea that 157 people can do it on their own is not going to be true. I think for products, because remember, at the end of the day, someone wires the check.

Jason Lemkin

It's not true.

Rory O'Driscoll

Hang on. Agreed. In any business, there are only 2 things that happen—

Jason Lemkin

People are either making stuff or selling stuff. If they're not doing either of those 2 things, they're just overhead.

To your point, if you're selling stuff via PLG, then you only need people to make stuff, so you can be pretty lean. Once you start selling to law firms, once you start selling to corporates, then you do end up with a big-ass sales force.

One of my theories is that this doesn't change from cycle to cycle. The Anthropic sales force in 5 years will look like the Oracle sales force, the Microsoft sales force, and the IBM sales force 50 years ago because—

But here's the thing: I don't know that that's going to be true, Jason. First of all, there's—I don't mean to go back to, well, if we compare Replit and Lovable, I know Replit's hiring 250 sales reps this year, so that's going to look very much like a traditional organization. Lovable isn't, and it's different DNAs and different goals.

But the majority of Anthropic's enterprise sales are not allowed to talk to a human. My point is that we can't all be Anthropic, but founders are choosing to have leaner go-to-market teams, leaner sales teams. They just don't want this crap. They don't want 250 people running around, and they're willing to trade off some marginal revenue.

Anthropic has fewer than 5,000 employees, right? They're just saying culturally—so I don't think they're all going to—I thought they would all look like SAP, Oracle, and Salesforce. We're not seeing that.

It's not going to be 147 people doing $500 million when it's enterprise sales.

Rory O'Driscoll

Nope, but what you might see is 2 to 5 times the level of efficiency, and it just changes the culture, the headcounts, and where people are. That's the difference, right? It doesn't really matter whether it's 0 or 4×, right?

Jason Lemkin

Agreed.

Rory O'Driscoll

Yeah.

Jason Lemkin

It will be better. No matter what happens, when you start with a clean slate and leverage intelligence, you just become way more efficient. I agree: on average, these companies will be way more efficient.

When you do a comparison, it's over $3 million in ARR per head versus a Salesforce, which is $350,000 per head. It's 9 times more efficient.

Jason Lemkin

Yes, you're right, but I'm just going to say it here: Salesforce is enterprise-heavy, R&D-heavy, with no AI costs, right? Remember, they have $300 million of— they just said it—of tokens. Let's do it here. We did the math. That's roughly $10,000 or $15,000 per engineer, and engineers are only about 1/5 of what they have.

Remember, that $300,000 in ARR is probably only 1% tokens. Do you understand me, Harry? In other words, Salesforce has $300,000 of revenue per head, which means if they're going to make money, they can't pay anyone more than $200,000, and they're probably spending 1% of revenue per head on tokens.

Contrast that with your example of Replit. They're getting $2.3 million per head, but they're probably spending 70% of their dollars on tokens. It's just vastly different businesses, and one of them is more aggressively leveraging the new enabling technology.

So, to Jason's point, it's probably a sweeter spot to be 1 of the 147 people in that gig than 1 of the—I used to know the headcount, now I don't. I probably could do it if I'm at that 20,000 or 30,000 people in a much larger organization where you don't have the leverage. They're just different businesses.

Rory O'Driscoll

But this, to me, is much more interesting than layoffs in these stories. I think every founder today—forget about older companies—wants to run a startup that's at least $1 million in revenue per employee or more. They're targeting $2 million. They want to be at $1 million, and they want it because they want great teams, lean teams, and the best people.

They want to work this way. They want to go to work with people they look up to and respect. They don't want bloat. My sense is that, roughly, over the coming years, startups will be half the size they used to be for a given level of revenue, including enterprise.

This is very much B2B-focused. That's a much bigger change than whether this company does a 10% or 15% layoff. If everyone's half the size they used to be, it's a much bigger change.

Jason Lemkin

Can I make a comment here? By definition, if you invent something that's meant to augment humans and make them more efficient, and that thing is called AI, and it does $1 trillion in revenue, by definition you need to see $1 trillion of efficiencies. The way efficiencies show up is fewer humans per unit of task.

You're exactly right. That's the bet. If it wasn't happening, the entire thesis of the case would be bullshit. So you're right, Rory, it's got to be happening. If the people who sell AI can't be efficient with AI, then what chance is there for the rest of them? I agree with you.

Did Elon have the acquisition of the year buying Cursor for what will be 10 times end-of-year revenue? It looks like a pretty prescient buy if they're going to hit target.

Jason Lemkin

It was a clever deal on every dimension. When I was thinking about this, because I'm always skeptical of the valuation, Elon did such an amazing job of meeting the AI moment.

You look back and go, he obviously founded OpenAI, and then all the drama happened. But in the last 24 months, he moved from ground zero to building Colossus, building Colossus 2, and failing with his model.

Just because he had the guts to show up and spend that kind of money—for fairness, he does have the cheapest cost of capital on the planet—he found himself with gigawatts of capacity just when everyone needed it, was able to sell it to them, and then did the Cursor deal to backfill the space.

Everything stems from the fact that he had the big-picture conviction that AI mattered, and he was willing to put $20 billion to $30 billion of capital in the ground in advance of revenue because he felt this was the trend to back. At least right now, it looks like a great trend.

You're right: prescient is exactly the right word. He found 2 of his biggest competitors who want to buy from him. He's getting $2 billion a month—$1.25 billion from Anthropic and $950 million from—no, the other way around: $950 million from Anthropic and $1.25 billion from Google, or the other way around.

It's $2 billion a month, $24 billion a year in compute revenue. On top of that, he has Cursor coming in at the back end to fill those servers. So he is the most efficient CoreWeave, with the lowest cost of capital.

Separately, it doesn't mean you have a foundation model; it means you're just a better CoreWeave. But, oh my God, did he turn a loss into a win in the space of 3 months.

On January 1, you could have said, "Look at all those data centers, and you don't have a foundation model. You're screwed." Here we are, June 9, and he can say, "I have a $24 billion outsourced business, and I have this other business that's coming in that's going to be doing $6 billion that'll run on my servers." I thought, "Great move."

Incredible transition. There were 2 large private rounds. Ramp raised $750 million at a $44 billion valuation. We've discussed Ramp a lot: it tripled in a year, crossed $1 billion in ARR, and is free-cash-flow positive.

Then there's Suno, the AI music creator company, which raised $400 million at a $5.4 billion valuation, teasing its first licensing model. BOND led that one. It was double the previous valuation just 6 months ago. Anything on either of those?

Jason Lemkin

With Ramp, we've said it before: it kind of gets to the Revolut point. They'll trade like financial services companies, but they'll be adjusted for growth.

We always have this example: when Brex slowed down to—I can't remember what it was—30% or 40% growth, they sold for 6×. Here we have Ramp; I've heard they're actually as much as $1.25 billion, so they're trading at 30 to 40 times, right? Whatever the number is, of that order.

It's a growth bet. If the growth keeps up, this will be a smart round, and if the growth goes down to anything like "normalized growth," it won't be. It's the same bet with Revolut. They're raising at 100 and something; they're doing $4.5 billion in revenue and $1.5 billion in operating income, which is freaking amazing.

These companies are great. Banks don't trade at 40 or 50 times earnings. They trade at 12 times. So on both of them, it's really just—people always say, "Will this trade like a tech company or like a financial services company?"

It'll trade like a financial services company, but it will be adjusted for growth. Ramp is getting the growth, and they just seem to do a very good job of riding the zeitgeist, their AI story, and their adoption story. They just seem to do a good job on all that.

For now, they've got the growth, and as long as they've got the growth, the math works. It's a big TAM, so we'll see.

Jason, are you trying Suno, the music AI company? Do you have Suno playing AI music in your personless office?

Jason Lemkin

I do like Suno. I pay for Suno. It's one of those services that, if I were more cost-sensitive, I would cancel my subscription because I think I pay $15 or $20 a month for 3 songs. There are certain apps that I think are fragile for certain users because I'll continue to pay them, but barely. The utility's there, but barely.

It is amazing. Even though I've been a customer and user for a while, I'm not smart enough to see it yet. The rate at which that valuation doubled, and the $20 billion outcome for it, I'm not smart enough to see it yet. So it feels a little bit like risk-on, right? The revenue justifies it, the growth justifies it, the stickiness justifies it, and the brand justifies it. You can't lose in AI, but I don't know. We'll see. We'll see at the IPO.

I just don't know where all this money's coming from. No, I'm saying, with all the IPOs—and then you mentioned Revolut again, Rory—is that targeting $750 million with the secondary sale they're doing at the $115 billion valuation? And then, with all the IPOs, we say, “We should know where all this money's coming from.”

Rory O'Driscoll

There's always money when people aren't afraid. I always say the converse is important: when things get scary, it's not that money runs out; it's that money gets scared. In the same thing in a bull market, it's not that more money's being made; it's that people are brave. There'll always be money when people are brave, and there'll be nothing but treasuries when they're not.

How long will they be brave for, Rory?

Rory O'Driscoll

If I knew that, Harry, I wouldn't be sitting here talking to you. I'd be trading QQQ. I don't know. At some point they won't be brave, but right now it feels like everyone's risk-on, so I think people are brave.

Jason Lemkin

But we've all convinced ourselves the rules have changed now, right? You can go from 1 to 100 in a year, and so many other things have changed. We throw these growth numbers out as if it doesn't require a massive change in externalities to justify them. Everyone thinks all the best startups go from 1 to 100 in a year. Going from 1 to 20 in a year is pretty good today. You want to be doing 5 to 8 by the time you get out of YC. The rules have changed, and they have changed, but there's a limit to how much the rules can change, right? There is a limit. It's called GDP.

Rory O'Driscoll

Yes, and it's also called human nature. I think that the rules have changed what's doable, but what we do is, in the face of these increased opportunities, we all get more aggressive. We keep on getting aggressive until the only thing that stops us being aggressive is someone gets burned.

It's the whole Minsky analysis: you're going to do what you're going to do, and it's going to continue, and the only thing that will stop it is overreaching. The skill is to figure out when you're at that point.

It was funny: last Friday there was a little dip, and you never know why stocks go down when things are overpriced. The narrative was that the employment numbers were good, so rates won't go down, and therefore stocks went down. Intellectually, I generally find things don't go to hell in a handbasket because employment is good, right? That's not going to be how this thing ends.

Well, it was because of chip guidance, to be clear. Chip guidance was $16 billion and missed the $17.2 billion number.

Rory O'Driscoll

Yeah, you're talking about Broadcom.

Yeah, which triggered it.

Rory O'Driscoll

Right, got it.

One that's amazing, which we may not have commented on, but it is amazing, is Bending Spoons. This is a roll-up play on traditionally consumer companies. Some of their properties are very well known: Evernote, Vimeo, WeTransfer, AOL, and Eventbrite. It's a massively executed roll-up strategy—$1.3 billion in revenue, and they're filing to go public at $20 billion in the US. From Italy, I hasten to add. It's one of the few large Italian success stories, to be very blunt. I thought it was an amazing success story. I don't know if you guys have a comment on it, but I thought it was fantastic.

Rory O'Driscoll

I did read it in detail because I was super interested. “Turned around” is an interesting expression. What they do—their MO—is they buy these things, cut all extraneous expenditure, including a lot of the acquisition expenditure. It's a little ironic; it's like the Vista playbook in enterprise. Then they raise prices massively.

I actually tried to figure out the organic growth rate of each enterprise because they're growing nicely overall, but a large part of the reason they're growing is that they're adding new companies. By definition, revenue goes up. You're trying to figure out what they did in terms of growth by entity. That's the next level down.

Even then, they get a pretty good growth rate out of the gate, to your point. But then you go one level below that: how do they do that? It's mainly price increases. It's very hard to get any sense of unit growth by individual product.

Take Evernote as an example. It's doing $200 million in revenue. They just cut all the marketing initiatives other than the high-ROI stuff. They take out 80% of the marketing spend, focus the team on features, and raise the prices 80% over the course of 2 years. Ten percent of the existing users go, maybe 20%. Their net retention is reasonably decent. It's below 100%, but it's reasonably decent.

So they raise prices, and the people who really want it stay. It's really hard to grow new businesses, but what it means is that it kicks off cash. Let's get real here: anyone who hasn't churned from AOL now ain't churning until they die. You can raise money on that. You've had 2 whole decades, people. It is 26 years since the AOL Time Warner acquisition. You've had 26 years to churn off this thing. You're going nowhere.

So they have very sticky, inertia-driven customers, and they stick it to them. It's an excellent business. The big 3 properties are AOL, Eventbrite, and, I want to say, Vimeo. It was interesting: the top 10 are about 80%, I think. Evernote, which I use, is in the top 10, but not the top 3.

You still use Evernote?

Rory O'Driscoll

I don't use it, but I have a bunch of stuff in it, so I paid for another year. I need to get it out and figure out where I'm going. It's a long story, but I'm not using it. I'm using ChatGPT, but I have to get all my stuff into one place. It's a long story.

I mean, Rory, what was the Last Supper like?

Rory O'Driscoll

Let's focus on the business. I looked at it and thought, the odd thing is that this is a consumer internet version of early Vista and Thoma Bravo. Buy those companies, cut the costs, raise the prices, and probably tap them out.

The question is, is it a great business? Absolutely. Should it go public at 20 times revenue or 15 times revenue? Maybe not, because you're relying on acquisition for growth. I mean, you're not getting organic growth. You're getting a profitable business, and you probably have to look at the sustainable profit.

It's hard to value it on a growth multiple, and you might be leaning in a little at $20 billion. But I think it's a great story. Everyone was playing in the enterprise space, and what these guys realized is that there's a similar opportunity on the consumer side.

The whole idea was that, in these verticals, no one's going to change their car-dealer accounting system because they put prices up 20%. In the same way, the default consumer is going to stay. So it's a totally sensible and orthogonal play to what everyone else was doing. They deserve the prize. Should the prize be $10 billion or $20 billion? That's a different question, but it's a great story.

Does it diminish what we've previously said about the bar to go public today? Don't get me wrong, they're at fantastic scale. It's $1.3 billion in revenue, which is awesome. But we have said that we're seeing this kind of bifurcation and that you need to be huge.

Rory O'Driscoll

No, you've said it. I haven't said it.

Jason Lemkin

But they're growing what, 70% or 80%? What are Bending Spoons growing?

Rory O'Driscoll

By acquisition, yes.

Jason Lemkin

I mean, Rory would know better than me. I'm not even convinced the markets care as much as we think about whether it's organic or inorganic.

Rory O'Driscoll

Salesforce itself—the balance of it is inorganic at some point, and then it becomes organic. We don't even think about a lot of these products as inorganic or organic. Does anybody really care? As long as it works, if they can keep finding these targets for the right price, if they can do what they did with Evernote, which is raise the pricing from $75 to $250 a year on average, and if they can find enough of these without just running out of affordable targets, going to the founder's letter, it sounds better to me than starting something from scratch.

Just go—there've got to be 800 unicorns to buy. Just go buy those ones.

Jason Lemkin

I really liked his letter. He said finding product-market fit is just a continuous mission of luck in some ways, and then the execution machine built after that requires no luck at all.

Rory O'Driscoll

Totally. Absolutely.

Jason Lemkin

It's just traditionally you bought—the Constellation version was 1 to 2x revenues, right? I don't know what Bending Spoons' blended price is, and maybe it's not revenue-based. I just don't know.

Rory O'Driscoll

Yeah. This is Constellation for consumer with a much higher valuation, because right now software is under pressure and this stuff isn't.

Speaking of big enough to go public, guys, Databricks came out today. No, no, we're going to do another round: $165 billion, up from $134 billion earlier this year. But obviously not going public with that announcement anytime soon. How do we think about that?

Rory O'Driscoll

The argument we've said for why the big model providers are going public is that they have a huge capital need. I think it might have been Diamond, or maybe the Goldman guy, who said recently there are 3 reasons to go public: you want capital, you want currency to buy other things, and you want to get liquidity for your shareholders. If you don't have 1 of those 3 things, then do you want the hassle?

I think Databricks, unlike these guys, for now at least, may well be in the position where their capital needs are still manageable. For context, the last private round at Anthropic was $30 billion, and the last private round at OpenAI was $122 billion. So this is less than—oh my God, it's 0.1% of the last OpenAI private round. What that says is, if there's money to fund OpenAI, there was money to fund Databricks privately. So they can do it for longer because it's just not the same need.

It's a software company. So they don't, quote-unquote, “have to.” I personally think you should at $4 billion or $5 billion in revenue, at the margin. I think in the end, logically, the cost of capital should be cheaper in the public markets, but right now it's not. Databricks can get capital at a higher revenue multiple because their growth rate is higher than Snowflake's, and on hassle-free terms.

I also think the other argument he made, which does resonate a little with me, is the idea that this is just going to be a noisy year. You've got SpaceX by Friday, you've got the 2 big model companies by the end of the year—there's just a lot going on. It may well be that next year is a clean deal.

The bigger point was that they don't need it. The amount of money that you need to build a foundation model is 2 or 3 orders of magnitude more than anything else, so the imperative for those guys to go public is just different.

All right, boys, is there anything that I've missed that you think we should discuss? Other things that made it to the top: SaaS now trades at a discount to the S&P 500 for the first time in history. Wow, that's sad. Meta is weighing tens of billions more for CapEx spend, following suit with Google. Zuck, attaboy.

Jason Lemkin

I'll tell you about the 1 small one I'll pick just for fun, if we're breaking. I think it's actually a more important story, but maybe it takes time to track it. Microsoft's new models that it launched, right? I think it said they were in beta. I found it very interesting that the models can't even search the web.

There are certainly use cases where that's not important, but it's interesting to me that you would launch a model that can't extend its knowledge by searching the web. It's a flashback to when this show started, when basically every time you chatted to ChatGPT, everything was 9 months ago. I don't remember. My memory's only through September 2024.

We think everyone can catch up. We think Microsoft can catch up. We think DeepSeek and open source can catch up. But if Microsoft launches these models and they don't even search the web, can we really keep up with the pace at Anthropic? I mean, the pace of change is so rapid. It's so impressive. There's so much progress. I just can't predict. I can't predict where it will play out over the rest of the year and next year.

Rory O'Driscoll

But you are right, Jason. It did matter because it was the final recognition that, frankly, they got caught out in mobile. They never caught up in search. They did catch up in cloud compute with Azure and, who knows, here. But you're right, it is the 1 that matters.

One of the big questions, between Microsoft and the open-source vendors, especially, is whether it's going to be a non-Chinese, U.S. open-source vendor that's even within spitting distance of the frontier models. That matters a lot from a pricing perspective.

There's a lot of open-source models today that are within spitting distance, no?

Rory O'Driscoll

There are, but mainly Chinese, and the question then is whether that's sustainable. A lot of our companies are using them; is that sustainable? Even though it's open source, is that sustainable over the medium term?

If your only plan is that you can download Kimi or DeepSeek and fine-tune it, that's great, but, A, some of those Chinese companies are themselves going closed source. I think what happens in terms of an open-source competitor in the U.S. matters, and obviously you've got, I think, Reflection AI and Poolside doing that.

But, to Jason's point, sometimes you get caught up in the stories, and you're the worst for that, Harry, because you just love the gossip. But Jason's right. What really matters is whether this is going to be an oligopoly or whether it's going to be 4 or 5 players in foundation-model land 2 years from now, which is why what Microsoft did matters.

I just did a show with the founder of Nebius, and he said the single biggest threat to Nebius is consolidation of models. If we have concentration of model winners, we are in a tough space, and we want an ecosystem, not a monopoly.

Rory O'Driscoll

There's a reason, yes, that everyone other than Anthropic and OpenAI is shoving money furiously at anyone else who can help erode that competitive advantage.

I just did a show with Aaron at Perplexity, and he said that export controls have actually hurt the U.S. in many ways because it meant that they've innovated on architecture that they wouldn't have needed to, and really built muscle that they wouldn't have had to. Combined with the open-source model capability that they have, it's now a competitive threat that's even stronger. It was an interesting discussion.

Rory, I have to say, we'll wrap. My mother texted me after our last episode and said that your quote, “Making money is like sex,” was the favorite moment of any Trio show that she's heard, and I got about 50 texts from people saying, “That is the quote of the century.”

Rory O'Driscoll

I've got to tell you, I think it's not in direct format, but there's a version of that either in Fred Schwed's Where Are the Customers' Yachts? from the 1960s or in Reminiscences of a Stock Operator from the 1920s. One of those 2 investing books hinted at that, but I always remembered it.

I'm not the original author, but the books are 3 to 5 times older than you are, Harry, so it's kind of like the Bible as far as you're concerned.