Why Apple Needs a Management Overhaul & Why Google is Catching Up with Hyperscalers

Ironically, Google, whom we all piss on, has executed the best of the four. They have a model that works. Apple doesn't even have a product that works. Microsoft bought someone else's product, but doesn't quite own it.

Facebook is desperately trying to buy a product, but it's not out of success. It's out of a terrible psychological need for a product, even though they don't have a business to justify it. I don't think these guys are too powerful. If anything, I think they're a bunch of rich people on the back foot behind the new trend, desperately trying to catch up.

We had Victor leaving Benchmark and now there's only three partners remaining with obviously Eric with Chathan and with Peter Fanton. It's big news in venture. I'd love to understand how you thought about this and how you responded to it.

First, listen, if the best AI researchers are jumping from Meta to Anthropic or wherever in 6 months, or getting massive packages, there's no stability, and it should be the same in venture, right? That's thought number 1. This is not the bucolic—whether there was ever a bucolic—but the idea of Benchmark in the eBay days, of a bunch of gentlemen investors staying together for 80 years, is possibly part of the past, right?

The other tactical thing I learned myself is that if you're going to leave a fund, and you can, and you have a hot hand, it's better to do it early. Staying longer doesn't help. It might seem like it helps the fund, but it doesn't help you. You're probably not going to leave with much carry, right? You're going to have to start over.

There are definitely reasons to stay at a fund, especially a top fund, but you might as well leave the moment you can because, all things being equal, there's not a lot of economic incentive to stay, right? Starting over is not that fun.

Agreed. I think it is very interesting. Stepping back, I remember Benchmark founding in '95. If I remember correctly, they pitched us in '95. So I would have said, what's happened here? Benchmark will be fine. They'll find 2 other people to fill those slots, and they'll be fine because they've been fine for 30 years.

I think the interesting thing is, in a prior era, I would have said being a partner at Benchmark is one of the top gigs in venture. They've been able to hire people from other places and other good firms and get them to join Benchmark. We all know the strict equal-carry-focused fund, with no layers of hierarchy. It's a really compelling story, and it's been a compelling story for 30 years.

The stunning fact now is that someone can be in the best gig in venture and decide, no, it's not enough. I need something more, and they can actually probably pull that off. To me, that's the big data point. It's less about Benchmark. It's about a talented individual with obvious prior entrepreneurial success being in venture for 2 years and deciding not to go back to being an entrepreneur, a kind of founding startup entrepreneur, but to say, “I can get money and continue this business on my own on equally or more attractive terms.”

It just speaks to where the market is, rightly or wrongly, and we can talk about that in a second. For that kind of top-tier talent, it's very much a sign of the times, right? You made a comment that most people who are good investors would prefer to just do it themselves. You stick together with your partners for a combination of reasons: You hope there's some strength and diversity—or some diversification, to use a less loaded term—and for a long time there was a perception of a bare-minimum fund size and team size needed to get funded.

Clearly, all those perceptions are gone, at least for a small number of very successful, talented people. That's what's worth talking about.

The nuance I've thought about over the years is that, if it were me, if I were at a similar spot at that time, I wouldn't leave Benchmark. I wouldn't leave Benchmark because what a lot of these folks going out on their own miss—and Harry and I did it right, and Harry and I had brands behind us. We actually had brands: quirky brands, SaaStr and 20VC—but we underestimate, especially today, how powerful it is to meet a founder and say, “You're from Benchmark.” It's freaking powerful.

I would argue that you and I are at the edge. Anything less than a brand like what we have, I wouldn't do. I just wouldn't. Everyone's different, but it's not that you don't have the skills of hunting, evaluating, relationship-building or schmoozing. It's just that the best founders want the same guys that are in Figma, the same 4 funds that are in Figma, right?

20VC is above the line, but it doesn't have the storied history of Benchmark. What do you think, Harry? That's why I'm not sure I would leave Benchmark.

I think people greatly overexaggerate the impact that this has on Benchmark. I don't mean that rudely to Victor, but I don't think any firm has played AI as well as Benchmark has done in the last 18 months, despite the team churn they've had. Let's just go through this. Mccor over 100 million. Hey Jen hitting 100 million NR. Fireworks 140 million NRA. Sierra with Brett Taylor over 100 million NRA. Manis AI, the best Chinese AI team there is. Lorra, the best European AI team there is in many respects. All with double-digit ownership in 1 single fund. How could I not be an LP in that fund?

I'm going to call a stop here only because we didn't—you actually didn't answer the question. We agree. Let's stipulate Benchmark will be fine because they're awesome investors. That's just not an interesting discussion, right?

I actually think, until you cut him off, Jason was starting down the far more interesting discussion. It's super interesting with this group because I've actually never had this discussion with 2 people who've done the thing. There are very few people who've done the thing that Victor did, which is raise a solo fund. The odd thing is, I'm on a podcast with 2 of the people who've done it, both of you as brand names, right?

I actually think talking about how Benchmark is awesome is fun because they are fucking awesome, but it's just not that interesting. Would I do it if I were Victor? Yeah, of course I would. You know why I would? Because he showed in his investing that he likes to move across scale in a way that I don't think Benchmark likes or allows in their disciplined manner, in a similar way to Miles Grimshaw, who's another fantastic investor.

I think they are multistage investors and just want to intersect with founders, whether it's at the pre-seed, like Victor did with Brex, or whether it's writing a $55 million check into HeyGen, which is really outside of what a Benchmark deal would normally be, being that size check on entry. I look at it and go, “I just want to engage with the best founders. I don't want the constraints of a partnership. I can get a lot of money privately from great individuals, and I'll own 100% of the carry.” A freaking amen. I'll do that all day.

I hear you. I just have a more nuanced view now that time has gone on. When I left, I had 2 top brand-name funds reach out to me that I didn't know well. It just wouldn't have worked out, right? But I had a pretty hot hand when I did. Harry will remember, right? I didn't even take the meetings, for real, because I'm like, it just doesn't work. I can't get the math to work in my head: 100% of the carry, right? I can't get the lack of autonomy to work in my head.

I've been a founder with a modest level of success. I ain't going to go work for somebody. It was a no. I didn't take any of these meetings. But time has gone on, and I do think that if you're going to leave, you need to understand the value of the brand you're leaving behind, right?

That's my only point. If you have no brand, it's tough. I think it's tough.

100%. You're so right. It's tough. But I do know Victor, and he is in the most exclusive inner circle of Silicon Valley that exists. He's in the Sarah Guo and Elad Gil inner circle.

Yeah, so he has his own micro-brand of relationships that will make it work for him, right?

Unbelievable. I would actually say the thing that's more important and attractive than brand is scale of cash. The only thing that I find attractive about joining another firm—which I never would do, ever, obviously—is the fact that I could deploy a $1 billion check into a great company. I actually think scale of cash is almost a more attractive magnet than stellar brand.

That's the one way the world's changed, right? We wouldn't even be having this conversation 4 years ago. Even 4 years ago, we wouldn't have had that conversation, right?

Yeah. What you find attractive about being on your own, to your point, is the ability to deploy large amounts of cash. I'm just trying to understand it. What you're making the point about is that you give up on the brand if the existing brand is strong, so you probably have to come to the table with an equivalent brand. You probably do, because if you don't have an equivalent brand, you won't be able to get the big cash.

What you're basically saying is—and I agree—that if I have, let's just say, a functionally equivalent brand as an individual to the brand of the firm I'm at...

In other words, high, right? So, I’m not losing a ton on deal flow from transitioning, and I get all the money myself. Then I should just go do that because I can do what I want.

Right?

100%.

What’s interesting is the people who need to think about this a lot are the LPs. And I’ll tell you why. Because what it says is, “We will now finance you to do all the things that we spent 20 years telling every other venture firm not to do.” Right? And I think that’s a big insight, right?

For 20 years, if you think about the LP focus on discipline, stick to your lane, and do the stuff you do well, we value teams, we value cohesion, and we value the long term. And what they’re now saying—and maybe correctly; this is not a diss. This is a reflection on where the world is, and on the value, sometimes, of listening to advice and, sometimes, of not listening to advice—is what they’re showing by their actions.

We’ll talk about it out loud in a second, but they also like a product that says, “Thank you for your input on being focused. I’m ignoring it. Thank you for your input on being a value-add in terms of taking board seats. In Elad’s case, I’m ignoring it. Thank you for your input on small funds. I’m ignoring it. Oh, and by the way, after giving me all this input, I’m ignoring it all. And then you’re going to give me the money.”

Note to self: What are you doing here, people? It’s super interesting, right? What it says is, “As I say, thank you for your advice. Now, thank you for your money.” And that’s why I think it’s an interesting time.

What does that tell you then? That Elad has such a strong brand that he is able to do that—to say, “Thank you for your advice. I’m not going to take it”—over and over again? I don’t mean that badly against him. He’s incredible.

I think what it says is exactly that. I think it probably says 2 things in the nuance, because they go 1 way. One thing it says is, in the end, idiosyncratic success triumphs over bland, mediocre standard advice. Winners win.

If you see something working, you’re just going to lean into it, right? Because I think they’re correctly saying, “I had this pattern-matching thing, and these are the things that I said we shouldn’t do.” And then I look at Elad—let’s move on to Elad here, because we had him on to talk about this. Oh my God, this guy was a seed investor in Airbnb and Stripe. He’s ridden the last 15 years; he hasn’t put a foot wrong. He has access to the best deals. Screw my rules. This guy’s going to make money, right?

So, on the one hand, it says capitalism works. Part of what’s going on is there’s a big signal coming here that says the things that we believed for 20 years might not be as true today. We should lean into the new. That’s the positive side. And I believe that, to be clear.

I actually look back and go—and I’m sure you’ll ask about this when I say it in a second, so we’ll put a pin in this—you look back and you go, “There are things that I thought one had to do that you look back and go, maybe you don’t.” Right? But let’s leave that there for a second.

The other side of the table is this. Sometimes rules of thumb, or heuristics, are in place because across cycles they’ve been proven to be correct. And my guess is some part of this trend makes total sense, and some part of it you’ll look back after the next downturn and say, “Oh yeah, that’s why we had that rule. That’s why we had the stick-to-the-knitting rule. That’s why we had the more-than-one-partner rule.”

And I think you’ll find some parts of this, totally free and easy, will prove to be challenging as you move across an entire cycle. And, zooming out, never forget: there’s no data on anything through a cycle yet for anything that started from about 2010 on, right?

So, again, I think the LPs are discarding some of their rules of thumb, and in some cases they’ll be right. But my guess is there was an embedded piece of wisdom in that that will come back, when you discard it, to bite you in the butt as you go through the next downturn. That’s a long answer, but it’s a super—I frankly think it’s a super interesting subject.

I think Elad is the best embodiment of the concentration of value and, to your point of picking your space—you said very wisely, Jason, and I always remember this—Josh Kushner is a master property developer in the way that he picks the block and buys the best house on the block.

Stripe in fintech, Databricks in data, OpenAI in AI. And I think Elad is very similar in terms of making sure that whatever that category leader is—whether it it's Heling or whether it's a bridge or whether it's Stripe he or whether it's Harvey—he is in that leader, and he’s in it big. I think that’s really impressive with him.

Do you know what I do? Do you know what I worry about when I see this news? Speed of deployment equals relevance. And I think about this because we’re very disciplined on 3-year fund cycles: temporal diversification. I know everything that we’ve said—maybe Jason doesn’t agree with on temporal diversification—but just doing what we said we’d do on how we invest and how slow we are, I see actually cadence of deployment can lead to relevance in a way that really benefits that manager.

Yes, it does. And I think we’ve talked about this before: you internalize that, in the short term, that’s true. And, again, the only guardrail on that is no, the only people who mind money are the people in charge of minding the money. The people who are in charge of minding the money here are the LPs.

They’re trying to figure out, on the one hand, because people who are deploying capital quickly and getting capital are, by definition, getting short-term positive feedback. No one deploys $1 billion, values it at $800 million, and then gets more money, right? So, if you’re deploying capital quickly, you can assume you’re getting markups, you’re getting success, you’re in the hot deals, right?

And what you’re left with is the LP trying to figure out: Is this a hot hand that I have to follow, or is this a flash-in-the-pan signal that could blow up in my face? And it’s not one way or the other. It’s a tricky thing to figure out.

By following, you made a comment on consistency. By doing what you said you do, by being consistent, you lower the chance of just drifting off and screwing it up, but you probably pass up on the upside of being as aggressive as someone who’s not honoring the temporal diversification and who is using that velocity, right?

All other things being equal, the person who’s doing more deals has more relevance. So, therefore, the only question is, as long as they’re deploying it well, it’s a double win. You have relevance in the short term and returns in the long term. If you deploy it badly, you have relevance in the short term and failure in the long term.

And I’ve seen, frankly, examples of both, right? I think Eli Gil is, for example, the example of the latter. I think he’s wildly successful, seems to be doing everything right. I’m willing to stipulate that he’s one of the most talented investors of his generation and will do it really well and continue doing it well. That’s one example.

I’ve also seen examples of people who’ve been relevant for 2 years because they’re putting the money out, and then the money blows up and they’re gone. Tiger is the obvious example. Now, they’re still around, but a shadow of their former selves. Their quote-unquote relevance in 2021 meant that they saw every deal that year, but they picked badly, so they got sent home.

Same thing with SoftBank. You can be relevant. Any early-stage VC worth a damn in Silicon Valley knows who the dumb late-stage money is, right? And you can tell. I remember watching—I won’t even name names—“Oh, that late-stage investor is attending that early-stage investor meeting. I know why they’re there.” That’s money—that’s dumb money—that they need. And those investors obviously blow up.

So, in the end, relevance is a short-term thing. And, in the end, as I say about this job, the only thing that’s true about this job over the medium term is you have to be right. The hot hand who’s right across an extended period of time—and, as I say, Elad, I think, is one of the best of that—you deserve to get the capital. You deserve to get the runway because you’ve proven over 10 or 15 years your ability to pick and, as you say, stunningly pick both early and late.

We can come back to what is wrong with the single-person model. But I’m willing to say the following: If it doesn’t work out, it won’t be because the person in charge of it isn’t brilliant. He clearly is. It’ll be because, structurally, there are problems with that model.

Well, the interesting thing is, when you look at Bney[?], the best-performing managers, or the breakout managers, of the last few years, I think you probably put Thrive, Greenoaks, Founders Fund, and Elad Gil definitely up there. All 4 of them are one-person-led. I’m not saying there aren’t great teams beneath them, but they are very dominantly led by 1 person. That is interesting, and it’s a very significant change from the venture partnership of old that was so predicated amongst the industry.

It is, and it’s easier, I think—perhaps wrongly—to be one-person-led the later you go. And the reason I say that is this: If you want to be an early-stage investor, right, and you’re putting up capital and doing Series A rounds, I’m going to say just doing early stage, because I recognize that all the names you cited have some amazing early-stage investments, and I want to come back to that because it’s important, right? Because there was an insight in that that the pure late-stage guys didn’t have.

But if you want to put most of your money in early-stage investments, what it means is that every partner can only do so much. If 1 partner is called the dominant partner, then all the others are, to some extent, not able to speak for the firm. You don't want to have a person doing a 10% or 20% ownership position who can't speak for the firm.

If all your dollars are going into early stage, even as a partner in the firm, I want the guy who's running 10 more deals beside me to be roughly a peer of mine. I don't want some weak person who's not that good, who's frankly a bit of a beta and is willing to work for half of nothing, running my money. You want that, which is why Benchmark, as the quintessential early-stage firm, internalized that you need to have equality so that anyone can be a premier partner.

It's totally different in late stage when you're not trying to be on the board. You're not trying to be involved day to day. Your time is not consumed by the deal; you're just making decisions. Making decisions can arguably be best done as a smaller group.

In the limit, you can have 1 person making the calls on late-stage deals, and it's much more scalable. Remember, if you're trying to put out $1 billion and you want to write $20 million checks, someone has to make 50 decisions. If you put it out in $100 million checks, someone only has to make 10, and it's hard to make 50 decisions.

So I think what you see about these, as you say, mainly late-stage funds, is that someone is capable of making the big decisions. They have structure to do other things, but then someone is calling the big shots, and they've been getting it right. That's true for Greenoaks, that's clearly true for Founders Fund, and it'll probably be true for—my guess is, fast-forward 3 years, there'll be 3 other people in the firm in junior roles, but that guy will be calling the shots, and he's earned the right.

Can I be so blunt, Rory? Does it make you question your model when you see these kinds of dictatorial firms—not in a horrible way—with massive capital sources, and then you look at yours, which is more disciplined in size and has a partnership structure?

No. I think we're all children of the era we grew up in, and we survived 2 downturns. Typically, in a downturn, the high-priced late-stage guys go bust. I'm not saying all these guys will, but it's what happens every time.

Remember, you're only taking 1 risk as a late-stage investor, which is valuation risk. When it goes wrong, it goes wrong in a 100%-correlated fashion: everything blows up in your face.

My liquidation preference always bails me out, doesn't it, Rory?

Yeah, you're being sarcastic. And, by the way, just for our listeners, you're being sarcastic there. You're exactly right: it doesn't.

To build over an extended period of time, I think you have to start much earlier, and that's been our primary focus. To do that, you have to have much more of a partnership structure. I think that's the place you end up in, right? If you wanted to build a late-stage firm, you can be much more hierarchical. By definition, almost, you should be, because there are a whole bunch of economic reasons. We can talk about that; that's a fun thing to do.

Look, cynical comment—not cynical, realistic comment. If I had known in 2009 that we were heading into the best equity decade ever, with 0 recessions, coming out of the worst equity decade in a long time, with 2 recessions, I might have played a different hand. If you knew that stocks were going to go up pretty unstoppably—they were at an all-time high today, and they've gone up for 15 years—if you knew that the correct risk-adjusted, work-adjusted, money-adjusted play was to just do Tiger and not quite shank it in 2021, I don't know if that answers your question.

If you knew then what you know now, you might make that play and make more money, but you design to survive the cycles you've seen.

Well, we can move on, but I actually would argue that Tiger will do better than people think, given their positions in OpenAI, Scale, and many others that actually are not bad companies. You said that they picked wrongly. I would argue they didn't pick wrongly; they just picked everything. In that, there's some good and some bad.

And it's not, “Oh my God, look, they're stunningly successful investors.” It's just that, as an LP, you probably would have preferred to skip that experience, because one of the things you have to look at is this phenomenon: the average IRR is great, but if most of the funds went in at the top, the LPs who just participated in that are not so happy. I agree, it's not that they're not going to make money. I'm sure they have many winners, but it's not investment excellence as you'd want to live it.

I think—listen, I know you want to move. I think maybe it's a discussion for another time. To me, the interesting thing about Tiger is that they played the game on the field, right? Like many are doing today, they played the game on the field. What they did, though, at least in my limited direct experience, is they played it broader than most.

So today, Harry, we're all talking about piling in. What should we do? Should we do Anthropic at $100? Tiger rode the vibe of 2021, which I don't even think is worth talking about. It's such a strange world where every B2B company was a winner that was triple-triple-double-double or better.

What I think Tiger did, in my view, and SoftBank—but really Tiger and B2B—is they lowered the bar because everyone won. The average public SaaS company in 2021 was growing 70%. Public.

Yeah.

So now we're doing something different, and that, I think, was the bet on the field, right? But it turns out, to Harry's point, that in the end only a fraction of those investments are going to have the power-law benefits, which we all think is so obvious today. But they were just playing a different game, weren't they?

Everything worked in 2021, right? Today, we're kind of like, some things work and a lot of other things are getting funded, but we're not clear. We're not agreeing that everything works, are we? In the age of AI, we're not agreeing on that.

Pretty clear on that. Everything does not work. But I'm just going to make 1 other point, because it was important that I put a pin on it.

The names you cited, those excellent investors—the Greenoaks, the Thrive—1 thing that's super interesting about those firms versus the quintessential late-stage-only firms is that these guys have also made some amazing early-stage investments. By definition Peter the Foundation did the best early stage investment ever, Facebook for 500k.

It's not that they're not early-stage investors. Their real brilliance was realizing that being a great early-stage investor with an implicit $200 million to $300 million fund size allowed you to have a $3 billion fund size, where the median pooled-dollar return is on your late-stage investments, because you're deploying lots of capital at late stage. The entrée to that has been your brilliance at early-stage investing.

That's why those guys are different and, in my view, better late-stage—how to put this—more likely survivors than the late-stage players who dash in from the public markets, think it's all a financial game, and then get blown up. These guys, the list you named, have, in my view, played it perfectly, including jumping a lab in that.

Having really great early-stage deal flow and access, and then realizing that the best way to monetize that is to follow on your $10 million check with a $200 million Series E check and collect the money every way.

We'll move on, but the brilliance of Josh and Neil from Thrive and Greenoaks is that they made $500 million-plus from Carvana in the public markets, buying when it was in the dumps and riding that up. That is a fucking good picker when the world tells you otherwise.

You have to look at that stock chart to make that bet. It was a falling knife, that one, right?

And then Neil also does Windsurf at seed.

Yes, agreed. I give up. Neil, too good.

Anyway, we said something about concentration of value. Anthropic started out raising at $100. Then it got so much demand, it went up. It went up. It landed somewhere between $150 and $180. I heard it was closer to $180, which wouldn't be surprising. I'd love to hear your thoughts on how you guys read this and what you make of it.

Well, look, the interesting thing here is I believe that there was $1 billion late last year and $4 billion this year. So this is classic—the thing of what really counts as the first derivative, the growth rate, maybe even the second derivative. Not just the growth being good, but the growth rate appears to have accelerated, which is stunning, right?

Most of your expectation is that you start off growing at 300%, then 200%, then 100%. The absolute numbers are going up, but the growth rate's down. These guys appear to have reaccelerated at scale. You're doing $1 billion, you're, I don't know, doubling—more than doubling—and suddenly you're doing $4 billion and you're 4–5x-ing.

If these numbers are correct, that's amazing, and I think that's what's getting valued. We can both feel smart: last week, you pinged us on OpenAI versus Anthropic—OpenAI at $300 billion, Anthropic at $100 billion—feeling pretty good about that 1.8x in a week. If only we bought.

So, yeah, I think that's what it is. I think it's just that people are looking at the reacceleration, and it's clearly a two-horse race—brutally—with everyone else either not shipping or not monetizing and not relevant in the case of some of the smaller things. You're buying the reaccelerating winner.

I'll tell you, for what it's worth: You also asked about capping Claude Code for overage, right? This is one part of Anthropic, but my experience in vibe coding—my captain-obvious learning—is that we're vastly underestimating the revenue potential per developer, per user, of Claude and Anthropic. We're vastly underestimating it. We're vastly underestimating it.

Why, Jason? Just educate me.

Because what was clear to me is that I was on a path to spend $8,000 a month vibe coding. Even with that, Harry, what's happening now is that the context windows have gotten longer. On Replit—and I think Lovable just did this in the new release—you could have up to a 15-minute-long context window while it's thinking through debugging or complex stuff.

You know what I would love to do now? Run 4 of them at the same time, or maybe even 15. So that $8,000 bill, all things being equal, could easily be $10,000. I remember, and I didn't get it a month or 2 ago, Farhan C one of the CTOs at shop I think CTO of Shopify, let all their developers use as much AI as they wanted—no caps—and the top guys were using $10,000 a month in credits because they were running multiple things in parallel.

If we're talking about 15-minute context windows, and chatb5 has a you know whatever a million tokens or billion token windows, then it's, “Debug my entire codebase. Now do this other thing.” Now I'm doing 10 things at once. I think my learning is that every developer at a top, growing tech company is going to get $10,000 a month of AI credits. Everyone's going to get $10,000, not $200, which is what we're thinking. We thought $200 was a lot a couple of months ago.

They're all going to get $10,000 a month. Every leading tech company is going to do it. It's cheaper than hiring any human, and you can't find humans. So if that means 50x growth in per-developer spend, it's going to be a CFO's nightmare. But putting that aside, that's 50x growth from where we are today for Claude Code and Anthropic, or if they use it through Cursor. It really doesn't matter where you're buying the tokens. Fifty times per human—that's a lot of growth, isn't it?

I'm convinced. It's not even close; it's blindingly obvious to me now. Shopify already got there. $10,000 is fine, not $200. We're going to consume it. The other thing that the CEO of Replit said yesterday about this was that you can't cap it, because any great developer will consume almost unlimited tokens, no matter how cheap they are. I think we're all going to land at $8,000 to $10,000 per month. If you're a good developer, you're going to be on AI 24 hours a day, not a couple of minutes here or there.

First of all, directionally, I totally agree with what you're saying. Zooming out a million miles, I think what we're saying is these—and bringing it back to Anthropic—these guys have a model that makes developers, who are expensive assets, extraordinarily productive. Somehow the market's going to find a way to massively reward them. I agree.

You made a comment on capping, and it's been interesting separately. Anthropic did some capping for their plans, and I think Cursor did some capping and some token limiting for their plans. You'd asked the question: How is that good or bad? I think it's amazingly good, right? I think what it says is that there's infinite demand for this product right now at the price point of $200, and the cost of generating those tokens—the economics—doesn't work for the model provider.

You could look at that and go, “Oh, my God, I'm going to worry about the fact that the economic model is wrong.” As a point in time, it probably is. My guess is somebody's gross margins are troubling. But when you're selling something, we have a couple of facts. One is that the long-term trends are in your favor, because all these costs just go down over time. Thank you very much, Mr. NVIDIA. And you've got almost infinite demand.

You just price it to the point where you get these people hooked, and over time your costs are going to go down and how hooked they are, as Jason said, is going to go up. Maybe today you're selling them $200 worth of tokens. You're getting $200, and maybe you're spending $250 in tokens. So you cap them now at $170 and think your margins are terrible, but next year that $170 will be $120, even if the token count goes up. And as Jason says, you raise that $200 to $400.

What it says is that there's an astonishingly strong demand pull for Claude Code and related products, either directly using Claude Code or indirectly using the Anthropic models with one of the other providers, like Cursor. That's going to reflect in the Anthropic numbers, which is reflected in the valuation, and it's going to reflect in a whole bunch of pushing and shoving around gross margins for a year. But in the end, you'll stabilize at something that works for everybody. So, yeah, it's astonishingly bullish on market size and growth.

Moore's law shows that the reduction in cost over time has pretty much always been an immutable law. When you look at that immutable law and the impact that will have on token cost, surely that impacts what we said about per-developer spend being $10,000 per month, because in 3 years' time it'll just be so much cheaper on a per-token basis.

Yeah, but the products—the whole reason your Lovable investment works, Harry—is that these products are so much better than 6 months ago.

Maybe, maybe, but you're right—maybe it all plateaus. I think a lot of—

It's shooting from the hip. I just think these products are getting exponentially better constantly, and we'll consume a million tokens, a billion, and a trillion. I mean, the token value will be more. We have inexhaustible demand to run multiple processes in parallel 24 hours a day for development. Our agents don't sleep, Harry.

This is the thing I didn't get until I ran 4 different AIs for SaaStr. Our AI SDR brings in—we were DMing yesterday, right?—our AI SDR brought in a $70,000 deal, and it doesn't sleep. If you're an engineer or developer, your AI isn't going to sleep anymore. You don't want it touching production code; I have learned that lesson. But if it's debugging, working on complex issues, building out an algorithm, or refactoring part of your codebase, you can consume it. It can work for you 24 hours a day.

We will use all of it. There may be an end limit, right? Maybe this investment will peter out, but I don't see it today. I think, good God, we're going to be coding 24/7 with our agents.

Everyone's saying that it's so hard to get a job in software today. If you look at The Wall Street Journal, it's not remotely true. If you have the skill set today, if you're super smart in math, computer science, or physics, and you're in any school today—my son's a freshman, and he's off the charts in math—he already has offers, and he's not even looking.

This is what we know, but I can see in the real world that if you're a top-tier developer, you just want to arm them with every tool you possibly can, because no one wants the folks who went to DeVry Programming School like they wanted in 2021. That market is dead. We're going to arm them with $10,000 a month.

Listening to everything that you say, then, Jason, Anthropic is at least a $2 trillion company.

Maybe. Right. I mean, it's just—there's so much investment here right now. Are there issues? Obviously, OpenAI is coming very hard at them right now, right?

I'm not signed up for $2 trillion, but I am signed up for this being an extraordinarily valuable company. I don't have to be signed up for $2 trillion to say that. This is going to be an extraordinarily valuable company.

There are only 4 or 5 companies worth more than $1 trillion, and it's worth pointing out that only one of them is really selling just to developers: NVIDIA, obviously, which is pretty much a monopoly selling 100% market share in the most valuable commodity you have. All the others touch every human being on the planet: Apple, Google, Facebook, and Microsoft.

I don't think you get to more than $1 trillion if you're just selling stuff to developers, which isn't to say I don't think you do extraordinarily well. If Anthropic ends the year at, what, $7 billion in revenue run rate, and let's say the average Anthropic developer customer is paying $100 a month now and that goes to $5,000—

Yeah. If you—so what's 50? What's 50? Everything's amazing.

Yeah, but I'm not being facetious. This is my captain-obvious epiphany: This is bigger than the $4 a month we spend on Jira at Atlassian. All the VCs like to talk about how AI is going to tap into the human budget, and some of it is just phony baloney. I've learned this is completely true for developers.

$10,000 is nothing compared with a $500,000 fully burdened developer who quits in 7 months. They're quitting for the $100 million offer. I'm not being facetious when I think it's 50x. It may not add up to $2 trillion, but I think the TAM—the true TAM—is 50x what it is today, without question.

This is the greatest case not just of making developers more efficient, but of radically, truly tapping into that nonexistent human budget that people have. There's no one to build this software out. There's no one to build it. Outside of Cursor, no one can hire anybody. Even at Cursor, it's hard to hire, isn't it, Harry?

It is. They have a big war chest, and so—

Even at level it's hard to hire Right now. Lovable doesn't win every developer they want on planet Earth, do they?

We saw Microsoft and Google roll out Lovable competitors. The question is: Is it too late? These are very established brands versus Replit and Lovable, in this case. Is it too late, or is incumbent distribution so overwhelmingly strong that they will be the victors with these products?

It’s a funny one. The Microsoft competitor to Lovable, Replit, and Bolt—now, this will change; it’s beta, it’s limited. Everyone shares 1 database, so you may have seen some issues I had with my database, but at least I had my own. Everyone shares the same database.

Microsoft has a clear warning: Be careful what you put in the database. Maybe you don’t want to build Tea. What’s the app that just had the terrible leaks today, the one we were talking about before we started?

Yeah. Don’t build Tea, because everyone’s sharing the same database.

What that says to me is that Microsoft can catch up, but it’s not taking this seriously enough. Or it’s being so panicked that it has to put something out there, which is probably more the case. They’re so panicked that Replit and Lovable are at $200 million in 6 months, so they’ll put something out there that no one in their right mind would use.

Yeah, I agree.

They only bought 1 database. They could only afford 1 instance at Microsoft. They could only afford 1 database at a time.

They could only code 1 app, but yes.

The answer to the direct question is: Is it too late? In other words, are Lovable and Replit already too established for Microsoft to maneuver? I actually think Jason’s is the right answer. It’s more a question of whether they’re going to focus on this thoroughly and comprehensively enough to win.

If they put all their eggs in winning this basket, they probably could. But it seems odd, for example, to angst as Microsoft about, “Oh my God, I’m losing $100 million of revenue to Lovable,” when you should wake up and realize, “I’m losing $900 million to Cursor” in a product I’ve already been shipping for 5 freaking years with GitHub Copilot. If the thing you delivered first can’t beat the newcomer, how are you going to beat the even more trivial newcomer when you’re later than them?

Brutally, I think the question is whether Microsoft is chasing shiny objects. Is it focusing on the next thing—which is now, “I have to do a Lovable thing”—or is it putting enough effort into making sure its Cursor competitor is as good as it should be?

I’m not fully formed on this yet, so I’m going out on a limb. For all the Microsoft-is-amazing-in-AI story, they’re better than AWS of the 3 hyperscalers, but they don’t have their own model. The relationship with OpenAI is a bit tortured. Cursor is now at $900 million, and developers—which is their core kind of raison d’être—in a world where Microsoft is at $500 million with a 5-year start. I don’t think they’re playing a perfect game, and I don’t know if trying to do still more things is the way to win here versus actually doing some things well.

That’s the trap, right? Trying to do even more things usually doesn’t work, no matter who you are—startup or Microsoft. It usually doesn’t work.

Sadia made the ever so funny comment that he wanted to make Google dance. Fast-forward 1 or 2 years, and they appear to be dancing pretty well. Microsoft owns OpenAI in many respects. Amazon owns Anthropic in many respects. Google owns distribution and Gemini in a way that is very, very powerful and important.

You have to buy 1 and short 1. Which one do you do?

Which of the 3 stocks are you talking about?

Buy Google, short AWS—Amazon. They’re just—I mean, I’m not fully developed on that yet. I haven’t spent time thinking about it.

Google has the advantage of underestimation, right? It could lose its search business; it knows that. There’s a lot in this, because remember, I was just talking about the hyperscalers. If you look at the execution of the 3 hyperscalers, then it’s easy. Let’s just do the hyperscaler first, because otherwise we have to think about retail and search, which are big things. I admit that.

On the 3 hyperscalers, 5 years ago, Amazon built the category and was the dominant one. Microsoft was number 2 because it could leverage over and into Azure, so it was kind of limping onto the field with a bunch of relationships. Google was nowhere, to the point where there were credible stories about it getting out of cloud.

Fast-forward to this quarter, and Google has nailed it. Great growth in Q2. As of the recording, we don’t know where Amazon and Microsoft are, but by the time you produce this podcast, people will know, because I think they’re reporting tomorrow. My gut, just looking at the trends, is that Google is pulling ahead. It’s the smallest but by far the fastest growing.

Amazon is growing—sorry, Microsoft is growing—roughly equal in scale to AWS, admitting that there’s always some weird accounting around Azure, and has a slightly better growth rate. So, by definition, of the 3, AWS is the hyperscaler underperformer, and I’m 100% certain I could make that case.

That was a lot. I don’t—obviously, if I had to make the bet today, I would make Rory’s, but I won’t make it for the Captain Obvious reason. One of the reasons is that there’s unlimited demand, and Amazon has its own TPU. As Amazon licks its wounds and recovers—which it’s already working code red on—when it gets back into the sweet spot of what its customers need, demand will be unlimited.

Agreed. I actually think you’re right. Then again, Oracle sells. I look at this a lot because there’s a little part of the contrarian in me that says, “Could you hit the point where that stops?” If it does stop, the correction is going to be super interesting.

I’m trying to find some evidence of overshooting on demand, and to your point, Jason, you’re exactly right: I can find none. You kind of go, “I can’t remember—was it Amazon or Microsoft?” One of the hyperscalers said, “We could sell more [bleep] if we had more [bleep] to sell.” That is a new idea, and it is the best thing in the world.

The simplest thing in the world is to be selling something that everyone wants to buy. It’s just a beautiful place to be in right now, and all those guys are there.

The Wall Street Journal had an article yesterday about the number-one KPI for public-company CEOs. It’s reducing headcount and growing revenue. Reducing headcount—that’s the number 1.

From Bank of America on down, it went through all these big companies, not just the little tech startups. Everyone wants less headcount, and the answer at Meta is AI. There’s unlimited demand to drive headcount down and revenue up. I guess it’s going to end, but when we’re all just living in the Matrix—

I don’t think it ends that way, to be honest. I’m just going to go out on another limb. I don’t know; I’ve had 2 coffees today. I don’t think it ends when everyone’s automated and demand stops.

I think what’s really interesting—and I’m keeping an eye on this because there’s a trade to be done when it happens—is the creation of AI. It’s happening at a hyper pace. There’s $600 billion of capex, whatever—just a huge amount going on. Enterprises’ ability to digest it is still relatively slow. Even though demand is growing, you’re still early.

Super early.

One of the weird things would be: How long can you continue investing $600 billion a year against a $30 billion growing revenue line? At some point, for example, even Microsoft—it was interesting that they did a bunch of layoffs, and Satya did his comment on the enigma of growth.

I didn’t quite get it. I think what he’s really trying to say is, “Yeah, it’s pretty cold-blooded to fire a bunch of people when you have hypergrowth and 40% operating margins.” But what he didn’t say is that part of the dirty little secret is, “We’re going to have to start depreciating half our free cash flow in capex through the income statement over the next 3 years. So we’re going to have to take people out just to be at the same level of efficiency.”

There’s a lot going on.

Yeah. At some point, someone's going to depreciate all these NVIDIA chips, and someone's going to have to pay for them, right? So I do think there might be a slowdown at some point in time, but right now there's no evidence of it whatsoever. Everyone did the, “Oh my God, NVIDIA's in trouble at 140,” and now it's at 170.

The only trend worth a damn in the last year has been: lean into the hypergrowth. Lean into the math.

Just so I understand, Rory, you're saying depreciation is the primary reason why you'd see that dampening?

No. I'm saying not just depreciation. If you continue to spend and run large capex through your income statement without an associated revenue line, eventually the world says, “Dude, maybe you shouldn't be doing this,” right? The truth is, right now, you're getting the depreciation; you're not getting the revenue commensurate with the spend. You're getting revenue growth—Jason is right, there is large corporate demand—but even with the best will in the world, they can only digest so much software in a year, right?

David Khan from Sequoas—I love that piece because, honestly, I thought about writing something like it, and I was like, “Oh, it's done now. I don't need to say that.” But the interesting thing is, it's a great quote: “The market can stay irrational longer than you can stay solvent.”

If you traded that position a year and a half ago, you'd have said, “The market's going to top out.” You'd have shorted NVIDIA at 100, and you'd be one sad puppy at 170. And forget even the stock prices—the capex spend by the hyperscalers has gone up since then. So I think the thesis is actually correct at some point, but that's—excuse my language—fucking useless if you're a stock trader. Is it correct today? That's the only question worth a damn, and I don't know when that happens.

And I think when you bring that back to Jason's point of the $10,000 per developer—Jesus, fuck—I just want to buy more and more and more NVIDIA today.

And I go, “Hmm, fully priced for that, baby.” So I'm not in the same place, but I go back to this: I think Jason is right, fundamentally, that insatiable demand is there. The only question is, how quickly can it be met—quickly enough to cover the nut?

Can I ask you, do you think—and I sound like a communist here—but we have never seen incumbents at this size and scale before, with this dominance, where they can drop $15 billion on a team and it's 40 days of free cash flow. Bluntly, who gives a fuck about the money? It's worth the chips on the table. And when you look at Google's asset base, it's insane.

Are these companies too powerful?

No. I'm just going to go straight out to no, because whenever people have been equivocal on that—one of my bigger bugbears on these kinds of questions—you end up with people coming up with dumbass things like the whole AI regulation, like the whole FTC process.

There are some things about these powerful corporations I don't quite like, right? I wish we could do nuance and fix some of the problems without going crazy. But if you have to pick 1 of 2 worlds—either idiot regulation or just let free capitalism run wild—I'm voting for the latter, because I think in the end Mr. Market will take care of these things.

I don't know, dude. Mr. Market says that Adam Smith's invisible hand—and you're seeing that now with Zuck hiring with an unlimited budget and making B2B hiring, to Jason's point, the biggest problem in startups that will just get worse. The interesting thing is, we have to decide if oligopolies are okay.

Yeah. Now, the case for monopolies is clearer, right? Maybe monopolies are good. Monopsonies are even more complicated, but we could argue the monopoly dynamic here. All of these that we're talking about—the reason they're interesting is they're oligopolies. We have 3 to 4 players, right? Oracle is all of a sudden competitive, massively competitive with Google, with Microsoft, and with Amazon. We have OpenAI, Anthropic, Grok, and friends.

So if you want to give up on oligopolies, I quit this venture. I quit this business. Let me return capital. I'm going to do an OpenView. I'm calling it a day. Here's the rest of the money. I'm going to ride out my winners, because if oligopolies aren't okay, I quit.

And the reason, you know, if you read classic textbooks on oligopolies, they compete on features, not price. There's a lot of hints of that, right? These products kind of are the same price today. A lot of folks believe that's good for an economy because oligopolies maximize innovation. They don't maximize discounts; they pour all the money back into R&D because they're competing on features, not price, right? If you believe in technology, you might almost want oligopolies, right? 3 to 4 players. 3 to 4 players.

I totally agree with Jason. I just want to come back to your point on “too powerful,” because I think they're 2 actually slightly different comments. All the incumbent older companies, the trillion-dollar companies, are in their own market monopolies. Apple, I would argue, and Microsoft, Facebook, and Google are monopolies in old markets—iPhone, corporate, social, and search—though they'll resist that definition forever. And whatever Microsoft is—corporate, right?

So, in the old businesses, they're monopolies. But Jason's exactly right: in these new businesses where they're putting their money and where the new startups are coming in, what you're seeing is oligopolies, which is what you'd expect. Now maybe fast-forward 10 years and they've ground down to monopolies. But right now, in the markets that matter, for all the power that Apple, Facebook, Microsoft, and Google brought to the table, it's OpenAI and Anthropic that are making the running.

I would argue that says there's simply no need for these whining, worrying FTC people to sweat it, because the truth is, none of those companies executed well. I mean, ironically, Google, whom we all piss on and say that's good, has executed the best of the 4. They have a model that works. Apple doesn't even have a product that works. Microsoft bought someone else's product, but then doesn't quite own it.

So it's kind of weird. Who else is there? Facebook is desperately trying to buy a product, but it's not out of success. It's out of terrible psychological need for a product, even though they don't have a business to justify it. So I don't think these guys are too powerful at all. I think they're a bunch of rich people on the back foot behind the new trend, desperately trying to catch up.

Rory, what do you do if you're Apple? You're looking at that assessment, which I completely agree with.

I think, as a board, you say to yourself, “Do you have a management team that's too old?” Because that's the only button you have. And I love Tim. I've made a lot from Apple. Apple is my largest single position. I have agonized about selling it for a decade and a half, and I haven't. So I love that man. Thank you very much. My little house owes him that much gratitude.

But you do wonder: is this the team to grasp what's going on in AI, which they absolutely should have a product in? At some point, do you say, “You just have a bunch of folks who aren't figuring it out,” and do you make a change? That was the comment, rather than some kind of tactical, “Hey, you should buy X, Y, or Z.” Look in the mirror and say, “Are you getting this done?”

I think it's very hard for a board to do. I think it's very hard for a board full of superannuaries to do. It's very hard when the numbers are still good, but you haven't grown in 5 years. You've done magnificent financial engineering. Thank you very much again for your dividend, but you're not winning. At some point, the buck stops at the top.

Well, do you think AI is a legitimate threat on its own to the App Store?

No. Because if I look at Apple today, 25% of its revenue is from the App Store, 40% of its profit, right? And 75% of its revenue is iPhone and App Store. So if AI just enables more apps to be purchased and Apple keeps its tax, right, maybe you should stick to your lane in the short term.

If you couldn't get Siri to work after 27 years, and the App Store is a monopoly, to your earlier point—and it's not—not everything's under threat from AI, right? Not everything. If it's a beneficiary of AI, then maybe they have the last laugh. Maybe they have this high-margin last laugh: all this AI revenue gets more and more of it routed through the App Store, and they take their 26%.

Fair.

Could be worse. Could be worse than taking 26%. By the way, I agree with that, Jason, because you're right: when you make the hardware, and the logistics of making the hardware, the thing you're amazing at, and hopefully design, your whole business doesn't go away. You're exactly right.

You don't have—like, you can talk about Google, ironically, and say existential threat to the downside if there's no search. No matter what stuff we're using for AI, I think we'll be using it on our iPhone. So I think that's a very fair pushback. It's not as existential, right?

But the brutal thing about capitalism—it's like, “What have you done for me lately?” Congratulations on building the best consumer product in the last 50 years. Thank you for the App Store business and the services business, which is fucking awesome. But did I mention what's the new new thing, right?

Do you want to run the risk of another—you know, I'm waving my arm furiously here—OpenAI-type personal companion that knows everything about you? Do you want to let anything get between this and you as a user for your 1 billion-user base? So I agree.

It's not existential. It's not like—but it's a shark thing: if you're not moving forward, you're dying. And they're not moving forward in AI. Google pays them to be the search provider, and maybe they just collect their tax that way.

You're exactly right. Then OpenAI can do the same thing. Maybe Apple says, “We build great hardware. Our hardware has such a dominant place that people will continue just to pay us money. We got it from Google and we're fine. Even though the courts are trying to stop that now, we're just going to get it from OpenAI.”

You're right. That's the argument that says, “Stick in your lane. You make $20 billion from other people giving you money.” I don't know if I buy it, but that's the argument.

I don't know how much OpenAI pays Apple today. Let's assume it's still zero, right? Which it used to be. I'm skeptical. But if it's—I mean, Jesus Christ—$108 million going through per month, I may have that wrong. Even if there's a sweetheart deal here, if Apple can just keep 25% of all of that, it's a good model.

One of the challenges I would push there is that owning the hardware is a great bet, and it has a long inertia factor. In other words, long after you stop innovating, you can keep monetizing, right? Which is, I think, the stage they're at now. The question is, at some point, do you need to do more than that to keep not just surviving but growing? Maybe it is, “Stay in your lane and just get paid by everyone else, effectively, for product placement.”

I love that debate, by the way. That was a fantastic discussion. For me, as an observer, it was brilliant. But a tough one—you said something there about the management-team switch, with Zuck hiring another stellar hire from OpenAI, a chief scientist or whatever it was. He is going all out. To what extent are we, like, 80% of the way there in his mind in terms of his acquisition and talent spend, versus this being 5% of the way there? When does Zuck stop this siege?

I don't have a clue. He seems—I mean, his argument, when you hear him speak, is: If I'm spending $30 billion on capex a year, or $40 billion, whatever it is now, do I need to skimp on the 5 or 10 people who can spend that most correctly? It's Jason's point: The super-scarce engineer is spending so much money that you don't care. My assumption is that, at some point, the marginal return on the next hire just gets negative, so you probably stop spending $100 million a pop.

But I get what he's done. He's said, “I'm going to spend $40 billion or $50 billion in capex. I better spend $1 billion, $2 billion, or $5 billion to make sure I have the smart people to implement that program.” It makes sense.

You know, it's not even that big of a deal because if he wants to—which isn't going to happen—he can always just stop. It's only in the past. He can stop tomorrow and say, “We're just going to harvest classic Facebook ad revenue. We're going to become the next Yahoo, and we're going to just be a cash cow.” It doesn't matter. If you're a public company that's massively profitable, the past doesn't matter, does it?

It doesn't matter, right? You can write off anything. You can fire all the people and you're still obligated to these massive contracts, but you can also write them off as one-time expenses or stick them in another bucket. They don't matter. If you have massive cash, you can always do a mulligan, at least every 8 quarters.

Yes. Obviously, they did the mulligan on virtual reality with the whole Meta thing.

Yeah, it's a mulligan.

But the interesting argument is, again, he gets to the question: Are you buying insurance, or is it existential? Implicit in that statement is that if you do nothing—if you're unsuccessful in AI—your core business just continues to compound, right? As long as that's true, you're right, Jason. If the core business is kicking off—what's it, $100 billion a year of revenue? You know, 40%?—if the core business is kicking off $40 billion or $50 billion and you make a $40 billion or $50 billion mistake, you just go in a penalty box for a year and move on. You're exactly right.

Yeah.

But of course, part of the reason he says he's doing it is because he says it's existential. The horrible outcome is you misfire on the new innovation and the new innovation eats your core business, right? I would argue that for Google, if they misfired on AI, they wouldn't have search. Whereas I think your point was more right than mine: If Apple misfires on an OpenAI-type AI product, they'll still be making hardware, and it won't be existential. And to your point, if Facebook misfires on Meta, they clearly still have Facebook. If they misfire again on AI, as long as they still have $40 billion of free cash flow, nobody cares. So, you're right.

And I think it's even more than that. Zuck is telegraphing to the market, “I'm making this bet,” very clearly. When I was a VP at Adobe, you're trapped with your 40% net margins because you can't invest in anything. The market is expecting cash to rain down to the bottom line.

Shaoo did a good job when they were moving to the cloud of getting some credit once for a massive generational shift. But Zuck can not only spend all this money, he can say to the market, “Give me a little time. We're going to get back to those classic margins.” And the market won't punish you, because most public companies are trapped in their margins once they're profitable. You're stuck. You can't go back.

It's almost impossible for most public companies to spend the cash they're generating. So Zuck's got a triple down here because he's also getting a hall pass. The market is saying, “It's okay to spend everything.”

Remember, it's not only that the market won't punish you. The more important point is that the market can't punish you because you have, in the case of Meta, voting control.

Well, yeah, your stock price could go down.

But it's worth pointing out that if you're a quote-unquote normal company, your stock price goes down and the vultures circle, whereas if you're Meta, your stock price goes down and you're sad. Your employees don't do as well, and there's dissatisfaction, but you still have control. Same thing with Oracle.

I think you're right: Those 2 players can swing a bat secure in the knowledge that no one has a voice other than them, which must be—

And they can actually spend their cash, which is otherwise trapped.

It's trapped with these large public companies. It just builds up in weird areas and it's unspendable. All you can do is repurchase your stock, right?

Most of what you can do is buy something, make an acquisition, and do some accounting shenanigans, right? Or repurchase your stock. Most steady-state, boring public tech companies are trapped when they're profitable. It's not the worst thing—it's great because you're not losing money—but laypeople ask, “Why don't you invest it?” It's not that simple. Your EPS goes down, and no one likes that with a normal, mature company. No one wants to see Zoom's revenue or EPS go down. That's the whole value of the company today, isn't it?

And this is why Mr. Buffett is right. Once a company becomes ex-growth, the most important thing to assess about management is their ability to rationally allocate capital, including sending that capital back to the shareholders, which, as we said before, Oracle has done in spades.

Anyway, what else, Harry? What else have we got?

Final one, and then we'll do a quick fire. It's just Figma's obviously going to IPO this week. I do think we have to talk about this. I don't think enough people are excited enough about it.

I remember when this would have been the most exciting thing in tech this week, and it would have dominated the news. It looks like it'll price at, you know, $18.8 billion. I think it prices at 5:00 p.m. either today or tomorrow, which is yesterday when this is out. Are we excited enough about this? How do you expect this to pan out? Is a pop baked in? Help us understand it as an audience.

I think it's unfolding. If anyone listened last week and this week, by definition, they're a glutton for punishment. But it's unfolding exactly as we said last week, which was that the filing range was, I think, $24 to $28 or something like that. They drummed up demand, which felt low. They've drummed up demand just in the last 24 hours, and they've refiled to, I want to say, $32 to $35.

So, it's worked. The bankers are doing a great job. They've whipped up enthusiasm now that they know they're multiple times oversubscribed. They've raised the filing range. I'm sure the rest of the drama is pricing at the high end or maybe above the range, and then a pop. So, the whole movie is unfolding exactly. Mission executed.

There's been a lot of talk about how they're trying to make people be more specific about their demand to better allocate the shares and avoid that pop. But I think the structure of these approaches means that that's almost unstoppable.

My gut is—and this is horrible because you're predicting something, and you're breaking the first rule of predictions—you either predict a number or a time, but never both.

And I'm predicting something now that, by the time this thing comes out, people will know if we're idiots or not. But my sense is it's hard not to imagine this thing pricing and then popping very well, because the structure of how they went at it is such that they will almost certainly have that money-left-on-the-table feeling. But I'm wildly excited about it. I think it's great. I mean, it's a wonderful company. It's got compelling numbers. It's a great story of a guy picking himself up after a difficult acquisition and showing the world it wasn't just going to be an acquisition-or-nothing story. And by the way, I think everyone gets excited once it prices, because it's all just talk now. But the reality of 4 VC firms each making $1 billion, that'll focus the mind.

You know, it's funny. I just remember September 2022. TechCrunch Disrupt annual was in September then, and the Adobe acquisition happened, and 10,000 people talked about nothing but Figma.

Correct.

Nothing. Now, this was after the 2021 bubble had imploded, had crashed, right? And all of a sudden, Scott Bellski and the team are heroes. They're going out in 2022 and paying 2021 prices for this scrappy startup, Figma, and people's jaws dropped, right? It's like, this was the deal of the century, right?

Fast-forward to today, and this is in my ecosystem—people aren't talking about it. Listen, 4 VCs making $1 billion. Great topic for the pod, right? Don't get me wrong. But it's just boring to people today. There were 10,000 executives there. The crowd was hushed in September 2022. There was nothing to talk about. And now, let's see, I might burn $6,000 on my Claude credits this month. This

World. Yeah.

Lovable did what again? I mean, it's just not that—hopefully this is a $30 billion market cap company for all of our sakes—but the zeitgeist has been lost here in traditional software. Even the best of the best of the best—Figma is S-tier. But it's not capturing our imagination the way it was. And maybe that's okay. B2B software used to be boring. Maybe it's super boring again.

First of all, I just want to cue the Scotty Schlafer interview, the 5-minute interview, when he says, “You win the Masters and then, 5 minutes later, the press is asking me, ‘How are you going to play next week?’” This feels like that. It's like venture returns $30 billion and all Jason's got to say is, “Well, what's going on in Lovable?” Right? There is, but I think you're right. I mean, this is—

It's the vibe. We're all vibers now.

A couple things. One, there's no doubt that once you announce a $20 billion acquisition, everyone's mentally run the cash register. So, you don't get to take the victory lap twice, but I totally think they deserve it.

But I think the more salient fact is what you're saying: there's still going to be a few more pre-AI IPOs at scale because there are a couple of other companies like that, like Canva and stuff. But you're right, the hype cycle has moved on to the new story around what's going on in AI, which, by the way, doesn't mean that a massive amount of money can't be made in other areas, right?

I mean, I always think—and even in... I think if you look back, semiconductors became uninvestable by venture in about 2004, and in the subsequent 20 years in the public markets, people made out like bandits. It was the best-performing sector ever, even excluding NVIDIA, right? So, I think what's happening is startup activity in the space that Figma plays in is not going to be a thing anymore. You're not going to build a non-AI collaborative design tool. You're not going to build non-AI B2B software of any kind.

So, as you say, Jason, the excitement—the new new thing—is not that, but that's a function, by the way, of it taking 12 years to go from a startup to an IPO. By the time you get to an IPO—and I think we're going to see this across the board—by the time companies that are 10 years old get to an IPO, you probably are half a tech cycle or a tech cycle behind. It's terrifying, right? I said this to an LP: our holding period is now longer than the tech cycle. That's a terrifying fact.

Figma's holding period is 12 or 13 years, and you're now in a world where they talk in the S-1 about, “We're going to have to adapt to AI.” Now, it's going to be an amazing company. It's going to be great, and I think they will adapt to AI, but we're talking here about intangible things, not money things. I usually prefer to talk about money things, but you're right, Jason: from a buzz perspective, the caravan's moved on.

But on the other hand, it's moved on and deposited $20 billion to $30 billion. I think everyone involved will get over their sadness very, very quickly when they look at the stocks.

And when Figma nails vibe coding, man, that's going to be what we all want.

Well, I mean, they're going after it with Figma Make.

I know, but it's what we really want. We don't really want—because if you look at the Replit and Lovable products today, and the world's going to change, and I'm a superfan of both, right? A huge use case today is prototyping, if we're being realistic.

It's prototyping apps, and it's very powerful and very fun, and the pixel perfection is terrible. The apps look—you can smell these apps—and they're not what real apps look like. So, as soon as I go into Figma and do this and it looks like I really want it to look, I'm out. I'm out of anything that's not pixel-perfect. I just don't want it. It's just like a Squarespace site. I don't really want something that looks like Squarespace, and I can smell a Lovable or Replit app within 6 seconds of hitting their website.

And so, Figma isn't going to disrupt programmers, right? Like the way Lovable and Replit are. But this prototyping—I don't really want to do it in these apps. Listen, you're closer to it, Harry, than I am, but I can't imagine this isn't a very interesting war in 2026.

You mean the war between vibe coding and design tools like Figma? Interesting. Right now, they're great. You can go into Bolt, Lovable, and Replit, in different ways, and in some cases, in 1 click, go from Figma back and forth. It's great.

But today, they're great partners. It doesn't make sense that they will be great partners for prototyping in 12 months. It makes no sense. And Canva saw this early, right? I don't know whether they'll really get there, but they saw it. That's why they tried to go really early into this. They got it right.

I think Canva's product will compete much more with Lovable, being the much more consumer-friendly product—children doing science projects, mothers doing businesses. I think the designer-to-really-average-consumer market, bluntly, is a very different product paradigm to design for, and I don't see that competing.

It makes sense that Canva is a bigger competitor at the low end, right? And you know, that's why Wixbot base 44 whatever. But to the extent—I see on my social... You know, we've got to be careful of X; it's a weird cross-section of the world—but notwithstanding all the complaints, a lot of developers are using these products to prototype.

The folks that have helped me as a vibe coder—these are CTOs. I had a CTO reach out to me at $80 million. He's like, “I just didn't want to distract my team. I'm building my own app in Replit. Here are my 10 tips.” So, I'd rather do that in Figma for that use case.

The competition is at multiple levels, right? Some part of the Figma user base is more picture-perfect, full-on built-out apps, but some of it is rough: “Mock me up what this thing would look like.” And your point is, you might want to do that now using a tool that actually builds you a vaguely functioning prototype.

Well, even more—listen, mad respect for Figma—but as an end consumer, I hate it when I get a Figma link because the thing doesn't work.

Yes.

I don't want to click through from a static design. This is why working with designers makes me want to blow my brains out. I don't want static stuff, okay? When you send me my Figma, I want it to work. I'm not expecting that it works in production. I want it to work. I don't want design elements.

And so, as soon as my Figma works out of the box, it's vibe-coded, man. That's what I want. Then that's all I want, right?

And then the fun thing for those guys—I mean, look, the very best companies, run by founders who see this, who've demonstrated and, in my view, clearly have the ability to bounce back from stuff, are going to be all over this [expletive].

And you're right, 12 months from now, if you don't have a billion dollars at stake in it, it's going to be fun to watch what kind of products these guys roll out to be competitive with parts of either the coding, Cursor, or the prototyping marketplace, and how that design-all-the-way-through-code marketplace shapes out. Because look, there's a big pot of money here, and everyone's trying to make sure they get that bit right. We're going to have some fun here, okay?

So, we're going to do a Koshi quickfire, but I'm just going rogue on this one. And I tweeted this last night because I was just super freaking intrigued to see where everyone landed.

And so, the tweet was, “Let’s play a game. End of next year, over/under?”

Under, but close.

Well, what do you think, Harry? Help me multitask. We’re coming up on $1 billion today, right?

Yeah.

And the bet is $4 billion next year, right?

What does just my Lorm model say? Just trailing velocity says that gets me to almost $3 billion or so, doesn’t it?

Yes, you get to $3 billion. Yeah.

Okay. Now use my math of spending at least twice as much per developer, bare minimum, right?

You get there. I’ve got to go. Yes, you get there. You’re exactly right. I buy that. By the way, it does, and that, by the way, is also proof. If they’re doing $4 billion—and I don’t believe the gross margins are as negative as people say—but if the gross margins are 50%, that’s another $2 billion coming Anthropic’s way. So it’s been good for them too. Next.

Yeah, I don’t think the gross margins are negative. It’s—

So even 50% gross—

Margin, I can tell you, as someone who keeps paying these Replit checks, they ain’t negative. I guarantee you Replit’s gross margins are pretty good. They’re pretty good. I bet they’re in the 60s even today. I could be wrong, but they’re not losing money. At least at the gross-margin level, there’s just no way they’re losing money with what they pay and what they charge, right? It’s a high markup to give you this environment. It’s a high markup on those tokens.

Lovable hits $400 million ARR by the end of next year.

End of next year.

Mm-hmm.

$400 million? I have no doubts.

Yeah.

That’s easier to do the math. One to 100 in 6 months. I mean, that would be a colossal fuck-up if it doesn’t get to $400 million. As crazy as it is, right?

Where does it price then, Jason?

What’s that?

Where does it price?

Unfortunately, Harry, it might be a down round to the recent price. Unfortunately—

There’s only one hitch in the analysis. It’s just as we regress to the real world, as we regress to the mean for valuations. Unfortunately, unfortunately—

It’s interesting when I look at those 2 bets. They’re not dissimilar in the sense of Lovable being at $400 million versus Cursor being at $4 billion, right? I’m just trying to assess the relative likelihood of the 2. I think the Cursor bet is a little more sure. Look, the trajectory of Lovable, as you know better than me, 100% points to that. The only thing that could bite you in the butt would be some kind of churn explosion, and you obviously have the data and I obviously don’t. So I’m just trying to come up with this: it’s your job as an investor to think what could prevent that from happening.

Yeah. The bull case is easy to make. To pull out of our bull case is easy. We could do that.

The bear case is that you were wrong about the long-term duration of the subscription. And you would argue, Jason, that’s not the case because you’re now a guaranteed vibe-coder subscriber.

Final one. We’ve chatted a lot about predictions. The final mother of predictions: OpenAI, is it over or under on $800 billion?

What do you mean, $800 billion for what?

An $800 billion valuation for a company.

When?

By the end of next year. Now it’s reportedly raising between $360 billion and $380 billion.

Under. I’ll tell you what I’m doing mentally. There’s no logic. I’m just looking at the step-ups in the rounds. As you’d expect, they’re getting smaller, right? You get 2 or 5 times, and then you go from $80 billion to $150 billion, which is almost a double, just under $300 billion to now $350 billion. I don’t know if they raised at $300 billion and there’s now $350 billion going on. I’m just guessing. Do you see another doubling round by the end of next year? It’s a lot to have happen, right? It might be worth it, but another priced round above $800 billion? I don’t know.

I’ll take the contrarian view on you there and say that because of GPT-5 and because Codex represents 2 very significant breakthroughs, they’ll have enough. Plus, Johnny IV coming out with something a first product. The 3 combined will be enough to get enough excitement over $800 billion.

Just 2 quick thoughts on it. One, I think there is so much money invested in this company. Sometimes these valuations are just willed. Sometimes, too, people are too deep. Everyone’s too deep into OpenAI. If OpenAI needs $800 billion to survive, OpenAI will find $800 billion. It will find it. It may be through investors that get other things. It may be through existing investors. It may be through weird combinations of offshore money. No one can afford for it to lose anymore. There are so many stakeholders that if that is what needs to be solved, $800 billion will be solved.

And I actually think the meta issue is more important. The CEO of Anthropic had this thing this week where he was explaining—because they’re supposed to be the good guys, right? We left OpenAI to build something safer—and he was explaining why they have to take money from the Middle East now. I don’t know if you saw this, right? His point was, “We have no choice.” I guess the flip side is he didn’t want to admit that this was lowering their pristine ethical values by taking money from the Middle East, but his point was, “We have no choice,” right?

So I think the bigger issue isn’t $800 billion because people are invested. The bigger issue is when all the capital on Earth and Mars will be exhausted. There is a finite amount, right? It is a finite number. Trump has already lassoed into Stargate with David Sax.

Let’s talk about a more nuanced version than that, because I do think it’s actually an interesting question. I think Paul Kadoski did a piece looking at the capex investment in AI as a percentage of GDP and comparing it to 2 other booms: the dot-com boom and the railway boom in the 1840s. I love that, because I was trying to think through what the comparables are, and the railway-track analogy is a good one.

So we’re running now at about 1.2% of GDP, which is basically 1.2% of GDP going into making capex data centers. That’s a little above the 1.1% of GDP that was going into bandwidth at the top of the 1990s and 2000s. So we’re already slightly bigger than the bandwidth boom, and we know that ended. But what I did not know is that it was 6% of GDP in the railroad era. At peak railroad mania, it went to 6%.

Now, as they pointed out, in that case you were building an asset that lasted 150 years and, in fact, just today, it did a huge acquisition. In this case, you’re building an asset that appreciates over 3 years. It’s not like we’d like it to be, but just to give a sense of the outer bounds—to your point, Jason, of how crazy this could get—we’re already a bigger boom in terms of capex than the dot-com boom, but we’re nowhere near as big a boom as the railway boom.

The pleasing thing about that answer is it allows both parties to continue to feel great. If you’re a nervous guy like I sometimes am, you’re like, “Shit, this is crazier than 2000. It makes me nervous. The internet was big. I don’t know.” If you’re an optimist, you’re like, “Hey, we’re only spending 1 quarter of what we spent to build the railroads, and intelligence is better than railroads, surely, so we’re fine.” It’s a nice piece of analysis that literally doesn’t allow you to conclude anything, right?

My gut is, all joking aside, that I don’t think you can get to 6% of GDP.

Apollo was 4.4%.

Apollo? But that’s—

The space program. Just the space program.

That’s the total. Well, now I’m winging it, but I’m totally willing to do it. That’s 4.4% of GDP across 8 or 9 years, which is an annual spend of about 3.4%. It’s not a single year. There’s no way we spent 4% on Apollo. I think we’re only spending 6% on defense, so no, it’s per year.

If I believe Reddit, which is always correct, it reached 4.5% in 1969 and was just under it in 1968—4.4% of our whole economy. That, in all seriousness, is why—

I think that’s why Sam Altman named it Stargate. I think everything he says that seems simplistic and calm and almost cute—when he says “Stargate,” he’s saying, “Listen, guys, we spent 4.4% for several years on Apollo. We should spend more on AI.” I don’t think it’s a coincidence. It took me a little while, but I don’t think anything he says is flippant. I don’t think it’s off the cuff. I think he’s doing his own nerdy style of clear communication. Stargate is Apollo. We need it all, and David Sax is making it happen.

Right. So 6% on railroads ended in a bust. I’m going to give you your 4% for 2 years, and I’m going to check it. But let’s say you’re right. It is worth pointing out that in 1970 they just stopped funding NASA like crazy because a—

Lot crashed. If you look at this chart on Reddit, the crash is bad, man. It actually peaked in 1965. It peaked in 1965.

Look at America. Fantastic. We’ve got the moon. We’re done. Let’s recycle the dollars here. So, yeah, 4%, then it crashes. You can look at 1.2% in bandwidth, but then it crashes. We’re clearly above the possible crash limit, but below the outer bounds. I don’t know what to make of that, right? I don’t know if you spend as much per year for 5 or 6 years on this as you do on railroads. We’ll see. And this is why, going back to what I said earlier, you can love the long-term trends, but at the same time, you can say to yourself that at some point there could be a break in the spend.

To bring it back to your point, it’s also why you have to go to sovereign wealth funds, because you need—I mean, we’re now so irrelevant in the context of financing this that it’s almost laughable, right?

I think it’s—I’m going to get in trouble for this—it’s laughable that there’s a moral debate of, “Oh no, I’m sorry, we have to go to Saudi for money.” No shit, you do, because that’s where the money is.

I also, to be fair to them—and I’m trying to don my—I mean, obviously, from day one, Anthropic was set up very much to be more angsty than most about trying to do good. And it’s interesting if you look at both companies. If you just read OpenAI, it’s been a long journey from “We only want to do good” to “Oh my God, we need to raise a lot of money.”

It’s almost touching. It’s starting off with innocence and then coming to grips with the fact that it turns out it’s very hard to raise $200 billion for charity, whereas $200 billion to turn into $800 billion, you can find pretty much anyone to do it.

It’s basically watching a whole bunch of folks realize that there’s a reason why capitalism exists. It’s actually very life-affirming as a capitalist. You can go, “These are people who, a priori, would have been appalled by all this, but when they went out to actually pursue their dreams and wanted the capital to make those dreams possible, they had to join the system.”

Yeah. Welcome to capitalism.

And to tie it back, that’s why I think if OpenAI needs to raise at $800 billion—not wants to, if it needs to—it will solve for it, whether it’s sovereign wealth money, whatever marked-up stuff, warrants. We’ll see.

That’s actually an interesting point.

It’ll solve for it.

That last one will be the—I mean, we didn’t talk about it; we didn’t have time. But you’re right, Jason: part of the asset you have to monetize is your ability to tell large sovereign wealth funds, you know, “I think OpenAI is a way of telling a domestic-sovereignty story. We’ll give you your model, we’ll give you a thing.” People aren’t going to put an ARR multiple on that. They’re going to, you know, for the right—

I don’t care.

So I think you’re right. I’m not sure it’ll be worth $800 billion, but I’ve—again, one of the reasons I like doing this is I change my mind when I hear good arguments. You’re right, Jason. They can probably find a way to back into it, which is different from saying it’ll price the day it goes public in a public market, which would be a different discussion.

Tim, I’ve got to be honest: I think this is the best show we’ve done. For me, as a spectator, I’ve enjoyed it the most. The debates have been fantastic. Bravo, both of you. Really amazing fun. Thank you for this, guys.