Anthropic融资100亿美元｜a16z的150亿美元基金：VC的中间地带死了吗？｜OpenAI如何归零？

[Speaker?]

In the early stage, you’re taking uncorrelated business risk, and in the late stage, you’re taking 100% correlated valuation risk. If the growth is there for one more year, it looks cheap.

This week, we have a lot to cover: Anthropic’s $10 billion fundraise, xAI raising $20 billion, and Andreessen Horowitz raising $15 billion.

[Speaker?]

I would be nervous if I were a $27 billion Cursor investor, but they’ve created something amazing.

If you’re OpenAI, are you not slightly nervous? You’re being eaten away by Anthropic, and then on the consumer side, you’ve got Gemini outperforming.

Well, luckily, you’re a nonprofit, so whoever wins is great for the global economy, right? You don’t have to worry as a nonprofit. It’s all for the greater good.

This is a world where we have ascribed the odds of a downturn to less than zero. I think OpenAI has existential risk.

It is a bet that the best of times lasts at least a decade. It’s pretty interesting that Anthropic not only raised the most capital, but on a 2x, too, I think has the strongest founder brand. You can be promiscuous at the A if you have enough late-stage stuff to cover it up.

We’re going to start with 2 monster rounds. It seems the only thing anyone’s talking about right now is Anthropic, raising $10 billion at a $350 billion price. I would love to hear your thoughts on this. Is this the last round before they go public? How do we feel about the price?

[Speaker?]

Probably yes on the first question, because they’ve stated they want to, and it feels like they can. If someone says they want to do something and it feels like that thing is doable, then logically it should get done. So, yeah, I think it probably will be the last one before the IPO.

And how do you feel about the price? Look, when they raised at $160 billion, I remember internalizing it. We talked about it, and frankly, we talked about it in our partnership. You kind of go, “Big number. Wow, that’s the second- or third-largest private-cap valuation ever.”

Then you look at the market traction and the revenue traction, and you go, on a revenue-multiple basis, it’s cheaper than some of the stuff we’re doing at $200 billion and $100 billion. This is a company that’s gone from—it's easy to remember the numbers for Anthropic because they very kindly did them in round units of 10—they went from $100 million at the end-of-2023 run rate to $1 billion at the end-of-2024 run rate, to allegedly between $9 billion and $10 billion at the end of 2025.

So let’s assume those numbers are roughly correct. They 10x’d 2 years in a row. I don’t know—next year, do they—let’s just say they only 3x, so they go to $30 billion right now.

A rule of thumb—I’m going to go now from ARR and run rate at the end to GAAP revenue for the year—a rule of thumb says, take the opening ARR and the closing ARR and calculate the average. $10 billion and $30 billion average to $20 billion. That says they do actual GAAP revenue of $20 billion next year.

So it’s 17 times NTM revenue. It’s a much lower revenue multiple than Palantir. It’s kind of comparable with Cloudflare, for God’s sake, in the public markets. You do that math, and if the growth is there for one more year, it looks cheap.

It’s the old rule: it turns out you really, really can pay up for anything that goes 10x year on year. So that’s the bet. The guys who did it at $170 billion 3 months ago are feeling pretty smart now. They’re at a 2x in 4 months. Calculate that IRR.

[Speaker?]

Only raising $10 billion is actually a sign that the unit economics are probably healthy. It’s not that much dilution, right? Anthropic has been clear that they believe their unit economics are strong.

They own not just enterprise, but code creation and application creation. They own building everything we have spent our lives working on. If you’re building with these tools, I know it’s the trite VC thing to say, but it’s hard not to believe we’re in the first inning and just getting going. It’s literally hard not to believe it.

How that works out on a spreadsheet—how many tokens, how many trillions of revenue—is complicated, but qualitatively and subjectively, it feels like the first inning.

Jason, do you think they have the enterprise market at this point?

Everything that I see at the API level—I mean, Claude has won it. Nothing is perfectly stable in AI; we should not feel that anything’s perfectly stable. But the reality is, so far, nothing’s dented it.

It’s birthed Cursor, Lovable, Replit, Harvey, and Rogo. These all—I mean, even Cursor is just a derivative of it. There are other models as well, but it’s tough to stop this trend.

[Speaker?]

Yeah, I agree, and I think you should break the market up into—I would have said 2, but as of yesterday, I’m going to say 3, and I’ll tell you what I mean in a second.

In the enterprise market, there’s the enterprise API market, which is basically selling your product to other ISVs that are building on top of it, or to enterprises building on top of it. They’ve been the premium product there for the enterprise for a long time because you’re, quote unquote, just an API.

There’s always a risk that, as an ISV is building on top of your product, they might try to use cheaper open-source models for some of the more commodified stuff. But to the extent that you need the high-end part of the product, that’s been the business that Anthropic and Claude have been able to get. That’s where they started, obviously, at the API level.

The second thing they’ve done within the last year is say, “Coding is probably the single largest use case for what we make. Let’s build a coding product.” So they have Claude Code, and that’s allowed them to, quote unquote, win at the enterprise.

They’re winning not just by being the API, but by being the app for coders. It’s not a 100% win—they’re competing with Cursor and GitHub—but you’re grabbing more money. Instead of maybe getting 50% of the revenue because you’re grossing at a 50% gross margin, you’re getting 100% of the revenue because you’re selling the product.

So that’s the second category in which they’re clearly, quote unquote, the winner. My sense is the enterprise share of coding revenue is, plus or minus, comparable to Cursor and GitHub—maybe a little lower, but growing nicely.

The third thing is, they announced the product yesterday. Caveat: I haven’t been able to use it yet because I’m actually here at an offsite, it’s early in the morning, and my coffee hasn’t kicked in. The product is basically Claude for non-coders. It’s a Claude workspace, I think it’s called. I could be wrong on that.

But basically, the idea is: if you’re doing other knowledge work besides coding, can you do it within Claude? This is the idea that the world has been going in this direction. We talked a little bit about Manus last week. There are companies like that, and there are a number of others. We have one, Obvious AI, that has launched a product in that space that’s just starting now.

Claude Cowork is obviously the dominant one. The idea is that if you’re building PowerPoint, manipulating data, or doing all the other knowledge work that those of us who aren’t coders do, instead of bringing the AI to the Excel spreadsheet—which is what Copilot tried to do at Microsoft—you bring all these tools into the Claude space, into the workspace, and maybe become more efficient.

I’ve read some preliminary reviews. Some people say, “Yes, this is amazing.” The people who’ve used it more say, “Yeah, it’s amazing, but it’s a bit janky.” But the idea is there.

The reason I mention all this is that the direction of travel is: don’t just be the chatbot for enterprise, the chat interface for enterprise, like ChatGPT for research. Be the place where you do knowledge work for the other knowledge workers who aren’t coders.

1. OpenAI Could Still Go to Zero

At a high level—and this is a zoom-out comment, but it’s a scary one if you’re Microsoft—every single knowledge worker uses the Office suite. You get PowerPoint, Excel, and Word, right? What is the AI version of the Office suite?

That’s a great part of Microsoft’s product because every knowledge worker buys it. Can you imagine turning up for work and someone saying, “We’re not going to give you a spreadsheet, a word processor, or a PowerPoint product”? You’re like, “What the fuck?”

The idea that, for every knowledge worker, there can be some product like this, some bundle like this, is a huge-ass idea. I don’t think this is it yet, but the idea is clearly that, to the extent that you are doing knowledge work using AI, you probably will need some space to be in. That’s the game they’re just joining now.

2. Has Claude Code Beaten Cursor Already

Sorry, Jason. My job is to ask provocative questions. When we look at Claude Code, you mentioned the potential impact it has on Cursor.

I speak to many CPOs as part of 20Product, and I ask them about tool usage internally. Everyone I speak to instantly states Claude Code, and the proportion of people who say Cursor has gone down dramatically in the last 3 months.

Would you feel nervous if you were Cursor and a Cursor investor?

It depends on the price I got in at. I wouldn’t feel nervous if I got in at the round of $200 million pre, because it’s not going away.

I mean, again, there’s an element of horse-race drama here. But we like to get caught in what you call provocative; I might call it getting lost in the details. There’s no doubt that Cursor has 2 large competitors, both of whom can bundle with adjacencies: A, Claude Code, where they have a strategic dependency as well, and B, GitHub. So, yeah, I would be nervous if I was a $27 billion pre-money Cursor investor, but they’ve created something amazing.

What I always say to my CEOs is the best way you know you’ve graduated from one league is when you start competing with people one league up. It’s like, you know, you’re in Division 1 of the English Premier League, and suddenly you graduate to, you know, the top division. It’s Champions. What is it now? Premier Division, right? Yeah, I’m so old. I remember when that was Division 1.

3. Anthropic's $10 Billion Fundraise

But anyway, you just get to play against different competitors. I mean, Cursor is now up against Microsoft. It’s up against its own supplier, Claude. So, yeah, they’re playing in the big boys’ leagues, but are you a little scared? Yeah, but you’re damn glad to be playing there, because of the other 10 coding agents, is any of them going to get to play?

[Speaker?]

For sure. I’d be a little nervous. Listen, if I were an investor, first of all, I’ve given up, in the age of AI, on this nervousness about competition and disruption, because what can you do? I mean, none of these products even worked a year ago. How nervous can you be holding a large position in a product that didn’t work a year ago? You can only be so nervous, or quit the game.

But even little things, like Anthropic cutting off xAI’s access to Anthropic this week, make it easy for me to imagine the business-model switches right now. It’s great for Anthropic to get an extra billion or so a year from Cursor. It’s a great deal. It’s free money. They repackage the product; they don’t—I don’t believe they have to sell it at any discount whatsoever—and they get another distribution channel.

Should that change as Anthropic crosses $2 billion and $10 billion in revenue? It’s easy to imagine: one, they could cut off access. That sounds aggressive, right? Or they just might limit access to the top models. They just might limit access. They just might degrade it. We’ve seen it on a limited scale with xAI and others, and so there’s no reason to believe that the scorpion might not sting the frog who takes it across the river.

The frog and the scorpion. It’s in his nature.

Yeah. There’s no reason that Anthropic just might sting Cursor just before it gets to the other side of the river. I think it would be naive to assume otherwise, right? And there are many ways that the scorpion could sting the frog. The simplest way is to limit access to models. They’ve already done it on a limited scale with xAI and others. They could simply copy the product. I mean, how hard is it to build an IDE that’s just the same as Cursor? It’s really not that hard. They can build Replit and Lovable too. These are not the greatest challenges of mankind. So all of them are at risk of the scorpion stinging the frog, but I would still invest.

It is worth pointing out that when the scorpion stings the frog, the scorpion dies too, in the old Aesop’s fable and in The Crying Game, if you remember the movie. But, yeah, I die too.

Yeah, I love it, Harry, because the first thing you said, I think, is just really so true. It’s very helpful for me, because the comment on being scared—if you’re going to be uncomfortable being scared, you need to just go home. I’m scared all the time because these things change so much. I think we said this before: in SaaS land, you could compound for 7 or 8 years. Now there’s existential risk every 6 months. And if you can’t live with that, you probably need to find a different job.

So, I think you’re spot on there. If we’ve got to be comfortable being scared, to what extent is Google choosing Gemini for Siri over the prior relationship with OpenAI a massive deal versus a temporary moment in time where Gemini is proving to outperform?

The big-deal comments are this one: Google and Apple obviously have a longstanding relationship where the money moves from Google to Apple for placement of search, because search monetizes with advertising. Therefore, it’s valuable to get real estate, right? So they have a longstanding relationship. It kind of makes sense that you’d go with your default relationship to make it happen, right?

The odd thing is, for this relationship, the money may—I’m not clear on the money movement—but it kind of, because there’s no advertising model, maybe the odd thing is Apple might be paying Google for Gemini, while at the same time getting paid a lot more by Google for placement on search, which is why the 2 products are slightly different. That could flip. If OpenAI had a model, for example, that had ads in the thing, then maybe the dynamics of the money move can flip.

But, yeah, if you’re in the distribution business, you want to be on a billion phones. I mean, the proof that it’s worth something is Gemini. Google pays—I used to know the number. I don’t—$10 billion a year, some absurd sum of money, just to show up on the phones, because it’s the best distribution on the planet. So, at the margin, you’re sad, unless the economics were stupid. It’s not like they even canceled that contract, because it’s not like OpenAI blinks at bad economics. I mean, those guys have an economic indifference curve that would make your head hurt. So, yeah, at the margin, you’d be bummed not to be on it.

If you’re OpenAI, are you not slightly nervous? You’re being eaten away by Anthropic, who have headwinds behind them seemingly like they haven’t had before, and incredible model performance. Then, on the consumer side, you’ve got Gemini outperforming. You’ve got Nano Banana being incredible and the tailwind of Google and the machine behind Google. It feels like you’re being eaten at every angle, combined with very high SBC and high churn, and it feels precarious.

Well, luckily you’re a nonprofit, so whoever wins is great for the global economy, right? You don’t have to worry as a nonprofit. It’s all for the greater good.

You’re not a nonprofit anymore. Stop.

[Speaker?]

No, hang on. Stop. First of all, it’s confusing when Harry says, “Just to be precise, you’re not a nonprofit anymore.” Your largest shareholder is a nonprofit. So, to make it even harder, if your economic value goes down, the biggest single loser is this wonderful nonprofit called OpenAI Nonprofit, which has actually already made some interesting donations, which is very clever, by the way. Once you got that deal done, start dispensing some money as a charity to show it’s a charity, to separate the 2.

So, if OpenAI’s value goes down, the largest loser is a charity. The second-largest loser is Microsoft, which will survive. And, as you pointed out, the third-largest loser is Sam. So, it is a problem, but “precarious” is a little strong. You feel angsty and driven. I mean, that’s why they’re at Code Red. But, to Harry’s point, anyone who’s not feeling nervous doesn’t understand the game. So, of course they’re feeling nervous, because you’ve got to play the game.

But look, I saw something—I actually thought in The New York Times this morning, I could be wrong—Sebastian Mallaby, who I think wrote the book on venture one time, was kind of, “Oh, I think Google goes up as OpenAI goes to zero,” and I think that’s absurd. There’s huge value here. We all default to them. They have 800 million users. They’ll find a model. I mean, I think that there is a model there. They have subscriptions. They have a business. It’s not going to zero.

The way I keep score is not, “Does it go to zero?” The way I keep score is the relative value of Anthropic to OpenAI, which is kind of the ratio of, let’s call it, management success over the last 3 years. And the truth is, it’s gone from, you know, 10 or 8 to 1, to 10-plus to 1, to much more convergent—it’s now only 2 to 1. So, if you were in a race, the objective measure of success over the last 3 years is something like: you were in the lead 10 to 1 over the other guy.

You're still in the lead, but he's now only 50% behind you and coming on fast. So are you nervous?

Yeah.

Are you bummed?

Yeah.

You're still in the lead. Don't blow it. And I think you've got a differentiated business. For all the Gemini talk, I still enjoy the ChatGPT experience more for the kind of research I do, for example, to go on this pod.

So they've got something amazing and compelling there. They just need to, frankly, focus, knuckle down, focus, and make it work and realize its potential.

There is a very simple bear case for OpenAI, though. There is a very simple bear case that it goes almost to zero, which is that the shelf life of an LLM is less than 100 days. The half-life is very short, and if something happens—there's a macro disruption and OpenAI can't raise the capital it needs—all of its competitors we just talked about: Anthropic has much superior margins, Gemini has massive cash flow, and xAI is crazy, but it'll get a trillion of Trump contracts.

OpenAI is vulnerable to macro disruption. We joke about macro disruption. Every portfolio company that didn't hit its Q4 numbers blamed macro disruptions, but it easily could. We've never seen this amount of capital availability ever, and it is not hard to imagine something. We've had systemic shocks in our lifetimes. If this was 2007–08, or whenever, OpenAI could die in the sense that it could not evolve while its competition could.

One caveat: I understand your comment, which is why the old Bill Gates rule was always to have 2 years of cash on the balance sheet, like OpEx cash, right? Because you're right: if that's the case, then you can't. The only way you could get into a really tough situation is if the world went to shit just when you needed to raise more money.

So they're smart people. If I was the CFO of that company, your mental rule of thumb should be: raise like crazy. You've got the world's best fundraiser, and never have less than 2 years' cash, because with 2 years' cash, even if the world changes, you can tweak the thing enough to converge more quickly. You just dial down your ambition and dial up your cash-flow focus, and in 2 years—

But how do you do that if Gemini and Anthropic can keep going? How do you do that if your competitors can keep going through that? That's the thing. This isn't Workday spending a little bit less on making sure that the Windows 98 integration works properly. You die if you don't have the capital.

But I suppose you're right, Scott. I rejected your first comment, which is that I don't believe this is the kind of user base that churns at 100 days' notice. I think there is a large degree—an increasingly large degree—of consumer behavior and stickiness. So, yeah, can you paint a scenario?

Oh no, hold on just 1 second. Just imagine, okay? OpenAI needs $100 billion in the next 2 to 3 years. That is more than it has spent to date. Its spend is accelerating. Would you—let's imagine it can't raise that and it's frozen in time. ChatGPT is essentially frozen in time today.

Would you use ChatGPT from a year ago? Would you use Claude from a year ago? NFW. You wouldn't use these products from a year ago? There's no way you would use them in Cursor or for coding. There's not a 1-in-a-million chance any developer would use a year-old model today. They were so terrible. This company would deteriorate.

So it would be like Detroit. It would still exist, right? Or be like AOL and dial-up. You'd still hear the shrieking because some people don't. Grandma doesn't know. Grandma's fine with ChatGPT from a year ago because it helps her with recipes in the kitchen, but the rest of the world's moved on to broadband.

2 comments. One is, yes, it is. By the way, it is astonishing that someone just traded AOL and it still has cash flow. That was the funniest fact of the year. Literally last year, someone thought, “Wow, that thing's worth a billion bucks still, 20 years on.”

Harry, get me on the internet.

[Speaker?]

Yeah, but I don't agree with what you're saying. I understand the point, but what you're saying—let's imagine a 2x2, which is macro conditions good, macro conditions bad, and then the other side of the 2x2 is scaling laws still working, so improvement is vital, versus scaling laws slowing down.

You're right: in a world where scaling laws are still massively working, so the next model is infinitely better than the last model, and where macro is shit so they can't access the capital, then in that corner-case scenario, you're right. You can play that scenario because you can always paint a bad scenario. That's what you learn, I think. It's just the little—

Let me just add 1 more point, and I don't want to take too much time here. You're the boss. I think we have returned to a moment in time—it feels like late 2020, 2021, or maybe you, Rory, can pick some other times in our careers—where we have ascribed the odds of a downturn to less than zero in venture and everything.

We are deploying, we are raising funds, we are deploying capital, and we are doing up rounds weeks after the last one. Underlying that bet, essentially, is a 0% chance of things not— And look, we see it in data-center use. We see it in power use and water use and ramp. But that's fine. We're not paid to mitigate downside in venture startups.

But I think OpenAI has existential risk. It is a bet that the best of times lasts at least a decade. And I think you can tell me the history of downturns; they're usually shorter than we think. But we don't have a 10-year cycle. That would be a long one historically, right? 10 years with no downturn.

I do have to say, I do think on the consumer-retention element, I think you're wrong, Rory. I think people are a lot more promiscuous than we give them credit for.

Speak for yourself. Excuse me, since the latest Gemini models, I definitely am. I'm a total slut for a new model, but since the new Gemini models came out, you've had a 22% drop in ChatGPT usage.

My son dropped it.

Okay.

Yeah, he pays for Cursor, and Gemini's free for him. He doesn't want to pay $20 a month for ChatGPT. He pays for Cursor out of his own pocket—out of his own pocket, he pays for Cursor—but he doesn't pay for ChatGPT anymore.

Summary is this, Scott, and I think it's fair. So if things go to—

And remember, invest in whatever your kids do. This is how you get into Snap and all these hot deals. You just do so.

So if my son's off ChatGPT, Rory, we've got to short. Let's get on Kalshi and just short this baby. Come on.

Got it. Okay. It's to Harry's point. It is that the next generation is fickle.

This discussion is like—I don't know if you ever read failure analysis of things like airplane crashes or anything like that. What you always discover is that any crash is always multifactorial. There's always more than 1 cause, and I think what you're saying is correct: if macro goes to shit at a point in time when they don't have a ton of capital, and at that time the market for this product is still incredibly fluid at the consumer level, then if all those things happen at the same time, you have trouble.

4. Andreessen Horowitz's $15 Billion Fundraise

I just argued that, so it's not a stupid comment to say it can happen. It's just you concatenated probabilities that I think are fairly low. I think what's much more likely is that, yeah, you have to moderate your ambitions and just execute on the consumer space, make it happen, and build a world-class business on that.

Speaking of moderating ambitions, there's 1 firm that is not moderating their ambition: our dear friends at Andreessen Horowitz. $15 billion for the new funds. I believe it was 22% of all of the dollars raised from venture in 2025 going to them with this fundraise. It's enormous.

How did we react to it? And a subsequent, really underlying question: do you have to go mega-big platform or tiny boutique to play the game in 2026?

Listen, on the one hand, so what? We've been talking about this since this pod started, right? We've been talking about massive funds, and all you have to do is look at the Databricks and Anthropic rounds, and it's pretty easy to see why you'd want to do that playbook.

I would say what's pretty interesting is that Andreessen not only raised the most capital but, on a 2x2, I think, has the strongest founder brand. That's hard to do both. It's hard to do both, and it has evolved.

Andreessen—I mean, I've been around long enough to remember vaguely when it started, and it was cool from day 1. It was cool from day 1; it wasn't what it is today, right? But I remember I had a subtenant. Marc Andreessen came into our office to meet with them to fund them, and it was a god moment: “Oh my God, is that Marc Andreessen in the office? I mean, it sure looks like Marc Andreessen.”

And they have invested in the brand at many levels. I don't know how they've done it, in some ways, but they have, and it's gone a little bit up and down. I remember I had 1 founder who was pretty hot who was bummed that he got a term sheet from Andreessen and not Sequoia, but that doesn't happen today. That was a brief—that was the 2008–09 version of Andreessen. That was a brief moment.

And you have returns. The returns were published. The returns are top decile. So you have the biggest fund, the top-decile returns—or quartile, whatever, top tier—which used to be a knock. And founders love this brand. Whoever was talking about fund versus firm or platform, it's hard to do all of those at scale.

[Speaker?]

Founders love it. They love it. It’s defensible. So you might as well hoover up 51% of the capital and then just shut down your competitors.

[Speaker?]

Agreed. I’ve thought about this a lot in terms of the question you asked, and I have a lot to cover on this. Just a heads-up, right? I think I’ll give you the summary. They’ve won, and they’ve won really well, and the only thing that might impact them at this point is misexecution internally.

But now let’s unpick this, because the first question in your little notes was: Can they make a 3x or a 5x on $15 billion? That’s the question you asked, right? Everyone always starts with that question: “Oh, there just aren’t enough exits to justify that,” is what they say. And it’s the wrong way to think about it, because I think you have to break it apart and say, first of all, is the industry at a stage now whereby that amount of capital can earn a return in total?

In other words, the total capital going in—because remember, if the total capital going in can overall earn a decent return—then it doesn’t matter from the industry’s perspective if that $100 billion of invested capital goes all to 1 firm and they invest it all, or it all goes to 100 different firms and each invests $1 billion. Right? The first macro question is: Is the overall market in equilibrium such that you can get a decent return here?

And then the second question is, given that, if it is in equilibrium, how much of that total money can they take and profitably deploy? In other words, are there diseconomies or economies of scale, and can they execute it well? Fast-forward to my 2 comments on this: First, I think the industry is roughly in equilibrium, so they can do it, and in fact the numbers are moving in their favor. And the second comment is, for the argument on deploying it at scale, I think it can make sense.

Right. So let’s do the first.

Yeah, they raised $15 billion, but they do 20% of the total. So it means the industry as a whole raised $75 billion, right? And everyone goes, “Oh, there aren’t enough exits for that, right?” Well, rough and tough, 3x the value of exits this year—which wasn’t an amazing year for exits, including healthcare, by the way—was around $300 billion. So, not perfect. That’s not all owned by venture, but you’re kind of roughly there.

And the other interesting thing is, presumably next year, with the caliber of—I mean, if it’s $300 billion this year and Anthropic alone goes public next year, it’s $500 billion of exits. So the industry raised under $100 billion this year. I mean, if they really raised $15 billion and they really are 20% of the total, that implies $75 billion of venture raised. It seems to me that’s a kind of number that can be digested and yield a 3x return overall.

So it’s not stupid, the amount of money, and it’s actually getting better, because in the last couple of years venture has deployed a couple hundred billion a year and only raised about $60 billion to $80 billion a year. Now, some of that is because some of the capital being deployed is nontraditional venture, but it’s getting harder for newer funds to raise money.

If you move on to the second question—can they deploy 20% of venture successfully?—the macro trends are moving in their favor, because they’re raising more money at a point in time when other people are raising less. So they’re in a nice position, provided they can deploy it. I think overall the industry is getting into equilibrium.

Then the second question is, how can I put it out? That is 20% of the money last year, right? But you’ve got to think of it over 2 years. That’s roughly, if they’re raising every second year—though they may raise in 2024 and then 2025; let’s just say every 2 years—that implies it’s 10% of the money on a sustaining basis. Agreed? It’s like you’re putting out 10% of the capital.

So basically they’ve got to get 10% of the exits. They’ve got to get 10% of the Series A’s, et cetera, et cetera. Now, interestingly, that work that we talked about way back last year, which the partner from DST did, showed that over the last decade Andre did roughly 10% of all Series A’s that became $5 billion outcomes.

So it’s kind of their market share, right? They’ve got to get 10% of everything. They’ve got to get 10% of the great Series A’s. They’ve got to get 10% of the great Series B’s. Provided they can execute that all the way up the stack, they make it happen, right?

Way to summarize it.

[Speaker?]

Yeah. And what I realized when I did that, Harry, was that I literally did it this week because I’m getting ready for a slide. So I’m looking at exit data. I mean, there are 2 risks, and we’ll talk about them in a second. But you look at it and go, it’s not crazy.

And as you say, it’s in part because they’ve done it. I think we all—a lot of us—come into this business as investors, and I think they came into it as engineers and as company builders. They did a great job of solving the system, and there’s a lot of leakage along the way.

I read the Packy McCormick article and all that. One of the things that was interesting in that article was that you make some mistakes and deal with a lot of negative knocks along the way, but it doesn’t matter provided the model works overall. Again, I repeat: They’ve got to get 10% of everything, right?

Now, maybe 2 or 3 things go wrong—maybe 3. The first is, just when you get bigger, if you have to do 10% of all Series A’s, it becomes a lot of deals, which means a lot of people. Which means: Is the marginal investor any good? Can you stay good when you have 20 people writing checks versus 10? It’s just hard, right? That’s a management problem. They’re good managers.

The second thing is, you say you’ve got to get 10% of all exits, right? The total value of all private companies right now is about $3.6 trillion, which, pleasingly, by the way, if you say 3x on invested capital, is $1.2 trillion over 10 years. That kind of says it’s got $100 billion a year creating $300 billion a year of value.

If you chop off just the top 3 deals, you’re down well north of a trillion bucks. You go down to $2.6 trillion. So the bigger the firm, the more capital you raise, the math all works, but it’s very top-dependent. Again, I thought the Packy McCormick article was very good.

If you miss even 1, I can’t make my math work and not get any top-10 exits. You simply can’t make this kind of math work without getting those top exits. And you don’t have to get in at the A. You don’t have to do the A of SpaceX, but you better show up on the cap table before they hit $1 trillion, because that’s a trillion dollars of value that you’ve got to get. So that’s the mission for them. They’re doing it, and that’s why—

That part to me seems the easiest part, right? You simplified it in a great way: They need to do 10% of Series A’s that matter each year. Okay, that’s doable if you have a top-2 brand, I think.

And then if you have 1 of the top 3 brands and a large enough team, their job is—I mean, even at Insight, I learned this from Teddy back in the day—you get fired if you don’t see every deal. It’s a different question of whether you win it. We get fired if we don’t see every deal.

If Insight can do that and Vista can do that, why can’t Andre see every single deal? They should have relationships with every seed manager that matters. They’re out everywhere. They’re close to Gary Tan and the rest of the world. Why shouldn’t they see every deal?

I mean, there’ll be some from left field, right? Of course. But if your brand’s strong enough, why shouldn’t you still see them? And then the interesting question is: Why can’t this scale? This was the question a couple of years ago that I remember Andrew Bialecki from Klaviyo asked. He’s like, “Well, why not 90% market share? Why can’t Andre have 40%, 50%? Why not?”

I mean, there are conflicts, of course. Let’s put conflicts aside, though. Why can’t your math scale to 50%? If conflicts weren’t an issue, why can’t it scale to 50%?

It’s actually an interesting question, because if you think about where I started, you’re right. You start—look at Sequoia. One of the things I always say is, no other venture firm has a billion dollars. You’re like, “Why are we letting them have that?” We’d really just prefer to be all us, right?

And if every year the technology industry gives entrepreneurs the chance to give the venture guys the money to turn $100 billion into $300 billion, the entrepreneurs, at some macro level, don’t care if that’s done by 1 firm doing all of it, or half of it, versus 100 different firms doing it all. Right? There’s no obvious economic reason. So I think the—

Especially if there’s no downside to it. If all I get is upside, I don’t get any drama if I sell my company. I don’t get thrown under the bus. They do my pro rata. The worst case is I’m treated well and I get to go to these cool events. That’s the worst case. Why would I not take their money?

So you’re right. Exactly right. It’s an interesting question: If they can do 10%, why can’t they do 20%?

Right. So why can’t they?

Yeah, exactly. So I think there are really 3 things that could go wrong. It’s an interesting speculation, because I think you should assume that when you have that scale, you see all the good deals at the Series A—but remember, you also see all the bad deals.

So, the more pickers you have to have to do more of the deal, the harder it gets to have all those pickers be good. Your mistake rate goes up, but you can cover for that if you do enough of the A, right? So that’s the first thing.

Then the second thing is, as you get later, you just have to concentrate in the winners, right? You can be diversified at the A. But going back to the comment, if it’s $3.6 trillion of total private value and the top 4 companies now—if SpaceX really was worth $1 trillion—you could argue the top 3 companies are now $1.88 trillion, right? You just have to make sure you concentrate down on those, and if you slip on missing them, it just gets harder to execute. That’s the second big risk: you don’t concentrate.

Every 2 years or every year, they don’t have that. I think your job is to get good at concentration, like you own it. I think Andreessen should target Ben and Marc. I actually did not WhatsApp Ben and Marc on this, but I think your math, Rory, is so powerful to me: 10% of Series A’s, combined with Andrew Bialecki’s “Own 80% of your market or you’re a failure as a founder,” and, from the CEO of Clio, “Own 51% of venture.”

I believe conflicts are a super-solvable problem for founders, like law firms figure it out. You just isolate it. We have 3 teams, and Andreessen becomes known as the gold standard. There are no conflicts. You can have direct competitors at Andreessen, and they have solved this; there is no leakage. They have solved this problem.

This is traditional VC. Even Sequoia has the issue: “We don’t do conflicts,” right? I remember in the early days, when we met, we referred some stuff, and you guys were like, “No, HubSpot’s our winner. We can’t have any conflicts.” But I think it’s a solvable issue. Then you get 51% market share, then you own it, and then Sequoia and those General Catalyst guys get the scraps.

If you want to build a firm and not a fund, this is what I challenge my friends to do: 51%. Because your math just—I think you can solve all the other issues. I genuinely think you can solve them. And it hasn’t had a fund below 4× gross. It hasn’t had a fund below 3× net.

Yeah. What I like about doing this for you, Noam, is how incredibly I can go in expecting to have to make one set of comments and end up on the total opposite side, because I was expecting to have the, “Oh, they can’t make the math work at 10%.” And clearly I convinced you they can. So now you’re like, “Fuck it. If you can do 10, why not do 50?” So what causes you not to try it if you can access the capital?

Well, that’s actually an interesting caveat. I think there are 2 or 3 risks, right? One is, if you’re doing Series A’s, the more you do, the more people you have to have. At some point, because the more capital you’re deploying, you only have 1 or 2 moves. You either do more small deals or fewer big deals. If you’re doing—instead of 20 Series A’s a year, you’re doing 60 Series A’s—you need X number of GPs.

I think quality goes down at scale. Let me give you proof on that. Andreessen’s market share is higher than Benchmark’s in terms of what worked. I wish Rothman, the guy from DST, did that. It’s higher in terms of the great Series A’s, right, as a market share, but the hit rate is much lower.

So, as you get bigger, you get more done, but the quality rate goes down. At some point, your hit rate goes down, and therefore you probably have a lot more fails, right? So it’s hard to scale. If you scale from 5% market share to 10%, your hit rate goes down by a couple of points. If you go from 10% to 20%, now you have the next 10% being written by less-good investors. The pressure to do deals goes up. My guess is your hit rate goes down over time.

So that’s the way it happens on the Series A side. I think there is a natural limit to this, because if you look at public investing, index investing is a scale business. Stock-picking is not. Right now, we can talk about whether, in the public markets, indexing is the right answer, which is why all the big money managers in the public markets are indexing. But stock-picking, in general, gets hard. You can’t be smart in a room with more than 5 or 7 people in it. So I think there are inherent limits.

If you are indexing, Sequoia Index reduced the fund size that they went out and raised. They raised $1.5 billion. Sequoia actually has quite constrained fund sizes, I think. The seed fund is around $200 million, and they don’t have billions and billions per vehicle.

Do they have to embrace scale and say, “Fuck it”? Andreessen have set a precedent. This is a money-wall game. There’s no doubt that you can pursue a really great seed and Series A strategy with plus or minus $500 million—plus or minus $500 million to $1 billion. There’s no doubt about it, right? Index can do it, so you can do it. The math is clear. You can have 5 partners doing deals.

I don’t believe, no matter what VCs tell founders in their spiel and pitches, that founders highly value the fact that VCs can fund you through every stage. Every big fund tells you that. Every Index, every Redpoint, everyone comes in and says, “The good news is, if we deem you worth it, we can shovel cash into you if we believe you’re one of our best companies.”

I don’t think founders go skipping down the street from South Park or Sand Hill when they hear that. They just say, “That’s not my problem, right? Getting help now and giving me the capital on the amount and on terms.” So I just don’t think that is as defensible as winning all the A’s. That is just an output of a combination of pro rata and winning, winning, winning, winning the right to do beyond your pro rata.

Put me down for a no on that, because I think there are 2 ways it helps, Jason, right? I hear you. I don’t think it’s dispositive for the founder, but I think it helps at the margin for a couple of reasons.

One, it helps you tell the founder a story: “Oh, look at the last 2 years.” To me, they do a really good job of that. “Look at Vanta. Or, oh, look how much we owned at the exit because we were there the whole way through.” And at least tell that good story with data: “Oh, look, we’re there to hold.” I think at the margin, that helps, right? More money is better than less.

I don’t think an average Series A founder is picking you because you diluted the founders of Notion to 5%. I don’t think that’s the most compelling story I’ve heard at a founder pitch.

Oh, that’s harsh.

They’re picking me because I believe Marc, Ben, and the team are going to help me build a $100 billion company.

The second argument—because I think the third argument is the important one. I’ll give you the second argument. The second argument is, you can use all the growth-stage fees to fund all the platform stuff, and you can decide how much or how little you believe in that.

I think the third argument is the really compelling one. Watch this: when I have a late-stage fund, I can afford to do more. I can decide I’m not clever enough to be like Benchmark and pick just the good ones. Fuck it. I’ll just do more of them, and some will be great. And even if I make errors at the A, I will be able to get so much money in my winners that I can cover for my mistakes.

I’m not saying that’s what any of these firms are doing, but it’s clear in the math. The more scale you do, the more errors you make, right? And therefore, the only way you can make more errors is if you have a way to come back from them. The easiest way to come back from them is to know that if I get 1 good Series A, I’m willing to get 3 or 4 of them wrong, because in that good Series A I’ll do the B, C, D, and E, and the other stuff gets lost in the noise.

That’s actually the real power of the late-stage fund. It’s clean-up on Aisle 5. Yeah, we made some whoopsies. We made some misses, but we’ll just clean it up, right? We’ll go, “The other 3 Series A’s that went bust? Cost of doing business.”

You’re down $60 million, $20 million on each. You have $20 million in the good one, you put $1 billion in, and you just 2× it. You’ve covered your nut. That’s the real strength. You can be promiscuous at the A if you have enough late-stage stuff to cover it up. That’s the argument.

And it gets back to the core thing. I think it was insight. I think if you approach your business as an investor—and I think Benchmark are superb at that—you just say, “I’m trying to pick the best,” and I naturally gravitate to that: let me try and be smart and pick the best.

I think when you approach it as an engineer, those guys say, “How do you engineer an overall system such that it works?” And you say, “I can take a little loss weight here, provided the overall system can cover.” It’s just an approach.

5. The Middle is Dead: Boutique vs. Large Platforms in Venture

For me, the truth is that the ballooning of your growth assets means you have ever-increasing price elasticity on your early assets. And so, for us playing the early game, they can just come in and bid $300 million when we're bidding $150 million. It doesn't freaking matter, because David George is going to put in a $300 million check at $3 billion or $4 billion.

The more you have here, the more elasticity you have here. That's the real alpha that you get from this. And that was my point, which Alex Rampell said on the show that we released on Monday.

Very simply, the middle is dead. Like every other asset class that matures, you see a boutique specialist and you see a very large platform play, and the middle hollows out. I mean this in the nicest and most loving way, Rory, because I think you're utterly brilliant. Are you not the middle? And how would you respond to that?

I think, first of all, in that context, you are the middle. But I think if you're going to do it crudely on AUM—and I don't—and I do think that there is pressure when you have firms that can raise $15 billion. That definitely creates additional pressure, and you'd be a fool not to say it.

I think you have to focus, because I think the word “boutique” doesn't just mean small. Given the stage we invest at, we can only do 20 to 30 deals per fund. We can only focus on enterprise software. We don't do consumer. We don't do crypto. You have to be good at something.

At the stage we invest at, we couldn't be a $250 million, quote-unquote, boutique, because at a Series A and a Series B, you're going to have to be writing $20 million to $30 million checks with 50% reserves. So what you have to do, at a minimum, is focus on a specific set of areas and be the best at that.

Let's examine what Andreessen does, right? Enterprise, consumer, fintech, crypto, defense, blah. If we were trying to cover all those grounds, we'd be doomed. And it's interesting today, Alex says that. But on the other hand, it's very noticeable that they've split the fund up into 4 funds, roughly our size, to put right at you: American Dynamism, over $1 billion; AI and apps, about $1.5 billion; infra, $1.5 billion.

Implicitly, by the way, I think it was a brilliant strategy—very Alfred Sloan, if you read the founding of GM. What they're doing is saying, they're recognizing you couldn't run this as a single thing, because I think you see deterioration of investment quality.

What they've done is they've given Martin his sandbox, they've given David Ulevitch his sandbox, they've given Alex his sandbox, and each of those funds is a freaking boutique fund at $1 billion, just like us. So, no, I don't buy that, right?

A focused $1 billion fund—what they do have with the $15 billion that you don't have as a $900 million fund in the same market as their $1.5 billion AI and apps fund is the air cover of the brand, and they have the cleanup of the $5 billion late-stage fund to cover for their misses. So, yeah, you are competing. That's the advantage they have.

But I think to just simplistically say everyone else goes away is interesting as a comment, but it's belied by the way they've structured their business. You have to be damn good, because you have to get up every morning and say you're competing against someone who will see almost everything, who can really lean into what they want because they have it, and they have the brand and the late-stage money.

So you have to get there earlier. If you wait till it's consensus or anything close to consensus, you're probably going to lose.

I mean, it's the Peter Thiel question, because I always think if you look at the 2 biggest entrants in the last 30 years—really, since Benchmark in '95—you know, Sequoia figured it out as founders who were engineers, and they systematized it. And I think Founders Fund—even though the name is Founders—figured out founders who were incredible investors and figured it out from an investor perspective.

So the lens is, I think they thought it through, whereas engineered and managed are true, and they bought, obviously, the 2 successful scaled entrants. I think the big question is: why do you know something that no one else knows?

If you're doing a quote-unquote boutique or a focused firm, you have to know something and have an area that the general funds don't have. Otherwise, you're toast.

Exactly. You have to see things earlier. It's hard. It turns out to be hard to make money.

Did you hear that spoiler? That was Rory saying the next fund is going to be $1.5 billion.

It's not.

Soon.

Absolutely not.

I think all of this—I mean, it's all true, right? Obviously, Andreessen down to YC will squeeze out a lot of players. You can't argue with that.

The meta question is: can you still find acorns? Can you still find diamonds in the rough? Are there any good startups that don't go through YC? Are there any of them? Are there any that Andreessen won't see at the A?

Anthropic is on fire, but one of the co-founders said that the very first time they tried to raise money, 22 out of 23 VCs said no. Now, almost instantly later, everyone put money in. Can you find that moment in time?

If the markets are so efficient in venture from the bottom end—from YC to South Park Commons to HF0, to Project Europe—if those have become so efficient in discovery that the only thing left is inception, there are a lot of VCs that have been doing this a long time who think the only thing left is inception investing, right?

Because you can't compete with YC and Project Europe and HF0 and South Park Commons. They've all locked up the market, so inception is all that's left. Maybe there'll be a new fund, perhaps, that locks up the pre-inception market.

We'll go to middle school or grammar school. That feels like it, but there is truth to that.

And is there still—can you find—here's my way I think about venture. This is the only thing, because otherwise I would quit. I would retire. Can you still find a $10 billion gem outside of the boundaries of this system or not?

This is the meta question. If you cannot find a $10 billion gem, then this is all a game of fees, or writing down the downturn of an industry. If it is still possible, and your fund or firm—that differentiation between fund and firm can actually still find one of those outside of the boundaries of this system—then you can make an insane amount of money.

But if not, it's all performative. It's all little checks. It's all $25,000 checks into hot YC companies, and it's all a lifestyle joke on Twitter. That's the question.

Can you—and will this market, as it matures, and it has goodness gracious matured a lot in the last couple of years—ruthlessly create discovery for all asset classes, to pre-inception? It certainly started down the path to doing that, right?

6. The Future of Venture Capital

Here's the question to Garry Tan and friends: can you find a great startup that won't go through YC and friends? Can you even find one anymore?

But I—just in 2 ways, there are 2 types of founders: the ones who are young and want YC, and then there are the serial entrepreneurs who want money and a good price and people who won't get in your way. Andreessen and Sequoia back multiple rounds before anything comes public, all swallowed by Sequoia and Andreessen.

So, just to add to your point, can you find any founders that don't go through either YC or Sequoia and Andreessen with big money very early behind the scenes, because they're in those insider networks?

I don't know. I got in trouble last week because I tweeted, “The worst place to be investing is Series A. You either need to be pre-seed or pre-IPO today to make money.”

There is 1 segment that will always exist in venture. For what it's worth, when I look back, this is where I've done a lot of investments accidentally. We used to call it a second seed. You can call it whatever.

It's when there's a glitch in the matrix, when they stumble a bit, or when no one sees the reacceleration. It's hard, right? But there are moments in time where someone is the hottest company at YC. It has a couple of great months, it reboots, and all of a sudden it reaccelerates 6 months or 12 months down the road.

I just invested in one that, because of Anthropic and friends, reaccelerated 2 years after YC. It can happen, and that is a niche, but it's a narrow one still. That's a hard investing ground. I credit you, and you're brilliant at it, Jason, but doing the glitch in the matrix—“I see what others don't”—that's tough.

Yeah. Owner was a glitch in the matrix when Redpoint didn't see it and I did the seed, and then they came in and put in every single round since. There's a lot of glitches in the matrix. They happen because the progress is not linear in the early days, right?

But if progress is linear, man, I don't think there's any hope for boutiques and buddies.

It's not. It's not, but it almost is today. That's kind of the weird thing with AI companies.

It is almost.

You know, all of your comments here, right? A lot of that is true. And just to cite some numbers, roughly 20-something-odd percent of unicorns have gone through Y Combinator; 80% haven't.

And then, is it all going to be done by quote-unquote the good investors? We track this by round. Typically, taking 10 names as being impressive, hard to beat, as I'd call them, right? Where you kind of go, “Hmm, if I'm up against them, I might lose,” right?

We have a mental list of hard-to-beats. At the A, it's 40% to 50% of total deals. The interesting thing is, it climbs steadily, and by the C, it's about 80%.

In other words, let’s call it the very hard-to-beat mental list. By the time you get to the C, 80% of the time, one of them has one of those names in the cap table. So the market is pretty efficient, right? As you pointed out, venture does a stunning job of missing the turn.

Salesforce struggled to get a dime from venture, and it didn’t. Anthropic and OpenAI, with the amazing exception of Khosla, didn’t get venture either. The first venture round at OpenAI was at a $23 billion pre-money valuation with Thrive, and the first venture round at Anthropic was at a $4 billion pre-money valuation with Spark and Menlo.

We’re sitting here saying we’ve structurally solved all our problems, we’re amazing, we’ve got all this coverage. But in the end, it turns out picking matters, and it is noteworthy and interesting how many of the dominant companies, because they were unusual, struggled to get venture acceptance. It’s not just a given that if you cover enough ground, you get it right.

Coinbase, I think Andreessen did either the B or the C; Union Square did the A, very thematically, in 2012; and I think Initialized and Y Combinator did the seed. So there is an element of picking here.

There’s only an element of picking if you believe that company growth is nonlinear and will continue to be. If you believe that company growth has changed to being linear and signals are clearer than they’ve ever been, then picking becomes less important.

Yes. Yes. If it’s incredibly obvious to everyone and then you rank—to be very direct, you’re right. If it’s incredibly obvious to everyone and then you rank-order on, for lack of a better word, a beauty-contest basis, you’re probably going to rank lower than some people who have $15 billion and the guy who invented the browser. Oh, well. You can’t fight that if that’s the way it is, and you’re right: there is a little bit now where it does feel like a time when it’s “very obvious.”

Generally, my observation is: be very nervous when you think everything’s going to work, just as a comment, because that’s usually when you’re—

I just look at the best in Europe, which is Lovable, ElevenLabs, and Luma. I think you’ll probably say those are the 3 breakouts right now, and the growth has been insane. It’s been entirely linear. There’s been no faltering in execution or growth, and that is different from years gone by.

Which actually segues nicely to—remember I said there’s one other risk here, right? About all these strategies, let’s call them, that involve excellent early-stage investing as part of your overall strategy, but then a huge number, maybe 4 or 5 times that number of dollars, going into the growth rounds, right? The risk in that strategy is that even if the execution is good, the pricing bet is still the remaining, as-yet-unresolved question here, right?

In the sense of my comment, when everything becomes obvious in terms of market and business opportunity, valuation expands to fill the gap, to fill the vacuum. Put it another way: when it’s obvious, people pay up, because the only risk left to take is valuation risk. So, brutally, even though, yes, the best firms win the beauty contest, they win it at the top price. You don’t get a mega discount. No one’s giving Lovable—Lovable is not saying, “I’ll take $6 billion when I could get $8 billion from someone else.”

The best firm might win the round, but they pay the market price. So the remaining embedded risk here is that all these late-stage valuations are 20 and 30 times, and the growth persists. If you were to do the postmortem 3 or 4 years from now, and many of these assumptions were wrong, what would that look like? I’m not saying it’s going to happen. I’m simply saying, what would that look like?

You’d say to yourself, all these growth rates attenuate just a little bit and multiples come down a lot, and you’re just in a different place. I’m going to pick, in my view, one of the best companies out there: Databricks. It’s doing $4.5 billion, right? It’s valued at $100 billion, it’s got a growth rate of 40%-plus, it’s cash-flow-positive, and it’s a superb company. It’s top 4. It’s one of the top 4 companies out there. At $100 billion, it’s 25-ish times revenues, right?

If growth slowed to just 20%, across the last 2 decades, 20% growth companies that are cash-flow-positive trade around 6 times. 6 fives are 30. So they grow 20, but 6 sixes are 36. All the math here is predicated on these kinds of valuations.

If the growth stays, I think the valuations stay. If the growth slows down even slightly, then you have a dislocation to the downside. I think then some of those strategies could feel a little painful, because you’re taking this utterly correlated—

What I say to people is, in the early stage, you’re taking uncorrelated business risk, and in the late stage, you’re taking 100% correlated valuation risk. When it goes wrong, it’s going to go wrong for all of them, right? That’s the embedded assumption: you’re assuming it’ll just be fine. Yes, it’s clear, obvious, and linear, but if it’s not—because it’s been so clear, obvious, and linear for 3 years, everyone’s leaned in so far that if it dislocates even slightly, the pain impact will be magnified. To me, that’s the—

You want to hear a small, fun example? You talked about the best ones in Europe—Lovable, ElevenLabs, and Luma, right? So I started using ElevenLabs for real this week. I vibe-coded my favorite thing today. It’s a game for founders. It’s called Founderscape.ai. Try it. I put 200 hours into this.

Founderscape does everything from picking your accelerator. You can join YC, you have batchmates, you struggle, you build the team, you go public. It simulates everything: fundraising. Let me know if you want 20VC and Scale in it as funds you can raise from. It simulates the whole thing. A couple hundred folks have played it. It’s kind of addictive. I can tell you why.

This week, I wanted to go to the next level. Your CTO joins you, too, and I added ElevenLabs. Your CTO talks to you the whole game: “The team’s struggling. Get your NRR up. Do this with your product.” I added ElevenLabs, and it was—it’s awesome. Your CTO talks to you the whole game. It’s so fucking good. It’s a 99-out-of-100 product, and I burned through $30 in credits with just a couple of people in 3 days.

Imagine thousands of people playing this game. I don’t have those resources. So my point is, it both shows why these companies are so explosive and also why they could be fragile. It’s hard to predict, right? ElevenLabs ended last year at $330 million in revenue—they just said it—from nothing, right? Motti is such a great CEO on so many levels: so charismatic, so good.

But for my game, if I could do something a tenth the price or 50th the price that was close to as good, I would have to switch. I burned through $30 of credits on ElevenLabs in 48 hours with 20 players—30 players. I don’t have those resources. I was already thinking, maybe I need to find another one, maybe I need to try the cheaper ones in my first week.

Harry’s so good at math. Imagine I have 10,000 people playing this game for hours on end. I need a lot of fees to support that, right?

Everyone watching this, let’s make Jason have a massive 11Labscape.ai.

And the fees do come down at scale, in all fairness, right? But my point is, it both shows why these companies are so explosive and also why they could be fragile. It’s hard to predict, right?

Jason, would you invest in ElevenLabs at $11 billion?

I would. Listen, I mean, I wouldn’t.

You would not.

At $11 billion—

They’ve gone to $330 million.

And listen, I’m not a late-stage investor. I’m not Andreessen. I don’t have the funds. Would I invest in the CEO? Absolutely.

If you’re an Andreessen, would you invest?

I would. Mati is the kind of guy I’d just want to bet on no matter what, right? Even if the ship went down, I’d bet on him. So if that’s all that matters going forward—if price doesn’t matter, if market dynamics don’t matter—I’m in.

I would have loved to invest in any round just to be on the journey together, but at $11 billion, I’m not smart enough to take it, because I already want to substitute out in less than 1 week. I don’t even care, Harry. For the most part, I don’t care what my Replit bill is. I don’t even look, and I spend a lot on Replit, but I was already thinking, maybe I need to find another one, maybe I need to try the cheaper ones in my first week.

That’s why I wouldn’t invest at $11 billion, without more work, because I think there’s an underlying fragility to it. I think that’s why Mati’s such a good CEO, because he knows this. He knows there’s risk, and he’s going 1,000 miles an hour to destroy the competition and not be a replaceable product in 24 months.

Jason, I’m pushing you. Is there a 3–5x on ElevenLabs from $11 billion?

Yes. If the whole world uses voice the way all the VCs talk about it and they can maintain some of their unit economics, of course there is, because what people don’t get is that ElevenLabs lets you have conversations like we’re having with AIs. That is a massive accomplishment, and anyone that wants to build an app that lets you talk to a restaurant or talk to a game can do that. It’s brilliant. The API is beautiful. I implemented it literally in 90 seconds. It’s such an elegant product.

If you believe in voice for AI—which at least all VCs do—of course you can make the math work. It’s $330 million in 1 year, so what’s 100 times that? A lot, right? We’ve only scratched the surface.

But I do worry. I think this will be the year—probably the back half of the year—where we have to take substitution risk seriously in AI. This is the first time I’ve done it. We’ve talked about these risks before, but they haven’t impacted us as investors: substitution risks.

But at some point, we're not going to want to pay all of these AI fees. I can give you another example on top of that, right? The other thing I added to Founderscape: Replit includes free single sign-on, free login. It works. It works in 1 click.

So does Lovable. So does everybody else, but they all are not as slick as using a native Google product. I went to buy Clerk. Clerk and WorkOS are 2 of the hot products out there that use it. WorkOS was out in the desert for years and blew up this year like nobody's business, because it just works for vibe coding especially.

I tried Clerk, and it's pretty good, and it's $30 a month. Was that a lot of money for a product that 100 engineers probably built for a decade? No. But Replit's only $30 a month, so I was like, should I use the native one? When Replit launches a product as good as Cursor or WorkOS, I'll immediately delete it. I'll immediately delete it, right?

As these platforms expand, my only point between ElevenLabs, WorkOS, SSO, and Clerk is that we just haven't had to deal with any Lovable or Harvey. There's no risk we're going to substitute them for a cheaper product, or ElevenLabs, or even Claude Code. No risk. With Cursor, we haven't had to deal with substitution.

I think as we go on this year, this will be a stress in the system. Marc Benioff will actually be right. We will rotate out for cost.

[Speaker?]

I think you will. I mean, do you see some of that, though? It's funny because, circling back to ElevenLabs, we'd looked at another company in the voice space 5 or 6 years ago. Fun story: way back in the day, we were investors in Nuance in the late '90s, so we'd made money in speech once.

Yeah, I know. I like to occasionally remind you of things that were around before you were born. We looked at that, and we did a bunch of reference calls. All their customers, especially the bigger ones who were spending $2 million or $3 million on this other speech company, were like, “And we're going to swap it out.”

So we didn't do the deal, right? You fast-forward 5 or 6 years, and they're all still on the platform. They never got around to substituting it out, right? Which is just an interesting learning. The risk is there. At $2 million of spend, I don't think people bothered. The question is, at $10 million of spend, they probably would, right?

Therefore, the question, going back to your ElevenLabs: you're doing a stunning company, so bummed. I wish we'd seen it. I wish we'd been in it. I like that market a lot, as we considered some of the others, and they've just killed it.

I'm just going to try and take on the question in real time. I haven't done any preparation on this. At $11 billion or $12 billion, you want to make a 3x, so you've got to be worth $30 billion, right? At scale, you're going to trade at 6 or 7 times, because that's the way human life is. Dude, get over it, right? So that's $5 billion of revenue in speech, and that's a lot.

Microsoft Office and Windows—I used to know this—is $50 billion or $60 billion. It's a big slug of revenue, right? If that revenue comes from Epic Games putting voice in all their games and paying ElevenLabs half a billion dollars, Epic Games is going to design them out, or they're going to grind them on cost, or there's going to be a competitor. They have the best product, but it's going to be a competitor.

If, on the other hand, it comes from literally tens of thousands of people using voice, where no one person is spending more than $20,000 or $30,000, or even your consumer spending $500, then you have much more ability to build a defensible business. To do the deal at $11 billion—and I haven't thought about it until literally on the fly—you have to believe in a very distributed market where there aren't just high-end people.

It's not just the total dollars, but are the total dollars concentrated in a small number of people, where you're a white-label provider and you do have some pricing pressure? Or are there lots and lots of people with voice? There might be. It's not crazy, to be clear, because they've proven the exciting thing about the $330 million of revenue. They've already proven large numbers of adopters, not small numbers of super-customers.

It's kind of like you're allowed to say something exists if you've proven it already exists by doing it. Basically, you've got to believe that trend continues. It's not crazy, but it's a lot of annual spend you have to believe in 3 or 5 years from now. I don't know, but it's a lot of momentum.

Listen, I just think Nuance was a while ago. What ElevenLabs is literally the best API. ElevenLabs is the best. I mean, there are a lot of good ones out there today, but it's the best API I've worked with, without question—the best API.

The fact that I could implement ElevenLabs myself, as not a developer, in less than 5 minutes, maybe even 3 minutes, says to me that I and Replit might be able to add another vendor, even split it. If I could do it in 5 minutes, right? I just asked Replit: at the current usage of my game, it's $1,320 a month that I would be paying to ElevenLabs. That's not nothing at this scale, right?

But you made the interesting assumption that I thought you were going to say, because you said how easy it was to adopt. You're right: if the other product has just the same ease of adoption and quality, then yes, it's easy to swap between them. But maybe you picked ElevenLabs precisely because it was the only one that had that ease of adoption. This is the advantage of a great product. If it's easy to adopt, then the other guys—

That's why I did $330 million in a year. It's the best product. It's great. It's just—it may be fragile.

Yeah, what you're saying—and I think it's a good point—is, when does the best product stop being enough? How far can you get on absolutely the best product and absolutely the most ease of adoption? Stripe would say a pretty long way. They got to $5 billion on that, right?

That's the question I worry less about. Going back to your early point on gross margins, the cost to them—I do think a lot of these non-gross-margin-positive things, the good thing is they're all digital products, so time will take care of a lot of that, along with cheaper compute.

7. The Impact of Wealth Taxes on the Industry

The final element I do want to discuss—and Rory, you can bounce when you have to, because I know that you've got to run to your offsite—is all over Twitter. I don't want us to move into politics, so I want us to stay on startups around this. I'm going to deliberately point this to our industry.

We've seen, obviously, the entrepreneurs' tax being implemented. Brin joins Page in leaving California. How does this impact, very specifically, our industry, and how significant is this actually?

You mean the entrepreneurs' tax? Not the wealth tax—the entrepreneurs' tax.

Sorry, the entrepreneurs' tax. Yes, just so we take politics out of it: the entrepreneurs' tax. We've seen Chamath say that now $1 trillion—I think it was; you reported $700 billion of $2 trillion—now is gone already. How does this impact our industry, and how significant is this actually?

[Speaker?]

I actually meant to read the text, and I didn't get around to it yesterday, but I would make 2 comments. One is, all wealth taxes underperform what people project they'll raise because wealth tends to be very mobile. It's very hard to tax that, and people can move. Norway, France, and a bunch of people have introduced them. They invariably unwind them because you get much less than you think.

The second specific comment—I've read it, but I haven't read the core text—is that one of the weird things about this tax is they estimate your ownership based on your voting control. What that means is, because a lot of these founders have these super-voting shares—and that's something I didn't agree with 10 years ago, but I've changed my mind totally—I think it's good in the public markets that founders have voting control.

They're getting assessed as if they own more than they do. Instead of being 5% of what they actually have, it's 5% of your voting control. If you've got 10x voting power, that's now 50% of your actual money. So are you going to sit in California if you're worth $2 billion and say to yourself, “I'm going to give away $1 billion for the privilege of living here”? I don't think so. You're going to leave.

I think it's going to be fairly pernicious to what we're doing here. And look, I think invariably it's unsympathetic, the sight of rich people leaving a state just because they don't want to pay more money at a time when people feel strapped. It's inherently an unsympathetic thing, right? It's easy if you're the rank and file to say, “Screw those guys. They should pony up.”

But I think this is in the category of dumb ideas that, in trying to overreach, will end up getting less. Especially in taxation, the way you should approach it is not ideologically—“Oh, we'll make them pay”—but much more, “How can I cost-efficiently milk this cow?” I think this is going to be inefficient because I think the super-rich will lose.

I think it's much more clever and worse than it looks. It's much worse than it looks because you have to read what's happening. This is a Trojan horse. This is not about a one-time 5% wealth tax. The goal of the proponents of this bill, and of everything that has been put behind it—this coalition, which has already passed similar propositions in the past, Proposition 55 and others—is that this will then transition to an annual tax.

Of course, it will. You cannot solve an annual health care gap with a one-time tax. It sounds good. So, first they need to get through the issues here and pass it once. Then the goal is 1% or more forever.

The goal—it has already been written, and this has already been attempted to be passed 3 times—is to lower it in phases to a $50 million and $25 million threshold, so that if you have $25 million of paper wealth based on the last-round price of your startup, $50 million to $25 million, you will pay a 1% wealth tax. That is the end goal. This is just stage 1 of the ultimate plan.

As bad as it is, it’s already going to fail. If the only goal was economic, it’s not going to work, right? Because we’ve already had 4 leading billionaires leave. But the goal is for this to become an annual wealth tax on $25 million to $50 million of paper net worth.

And so I say this will end up being “leave for the Series B.” Leave before the Series B, because if I’m the founder of Glean AI or ElevenLabs and I’m doing a Series B at $500 million, I’m going to pay the wealth tax right now if the goal is to implement it. I’m going to pay it.

And you can say, “Well, Gavin Newsom says it won’t happen,” but no, the voters in California are going to vote all this stuff in. What I think happens, if you just want to know what I think, is that this is much deeper than it looks. It’s not just that people feel bad for billionaires, okay? Very few people actually feel bad for billionaires.

The goal is to hit folks with paper wealth of $25 million. And I think if it passes and the next bill gets put up, it will likely pass. I think founders will begin to massively exit in 2027, before the next one goes up, because there’s going to be a second if this passes, and a third. So this is not one and done. This is just the start. It’s just the start of what the coalition behind this wants to do.

It’s crystal clear. They’ve already put a bill up 3 times to lower this to $50 million. $1 billion is a retrenchment to get it done this year because they couldn’t get a $50 million wealth tax passed. A $50 million wealth tax is a disaster. If this actually happens, people will finally flee.

Yes, they will. And I think the voters—I’m going to be optimistic here—I think 1 of 2 things happens. Sense prevails and it gets shut down. If that happens, it’s already been an own goal because we’ve lost people who’ve left California in advance of this, and that’s just stupid.

The second thing, you’re right, Jason, is it gets passed, then a bunch of people do leave because then it gets real and you start seeing other people leave. And then the voters face this other choice 2 years from now, when they put up another bill to lower it a lot, which will cause even more people to leave, and they ought to vote for that or not.

And you know, you can’t—I don’t believe you can—you can’t stop stupid, but I actually don’t think the voters are stupid. I think the default is—I mean, look, I don’t love the California referendum system, but it’s crazy propositions.

Hang on, hang on. We do, but it’s worth pointing out that most of the time they say no. I think the default California voter goes in to say no. So my guess is—and maybe I’m being optimistic—my guess is, in the end, this loses, right?

But you sit back and go, even trying to do it has had an economic cost, because if you have those kinds of assets, you’re subject to it. This is not an idea that maximizes revenue. It’s not even revenue-maximizing.

If you were a revenue maximizer—let’s just say you hated rich people. You hated them. But at the same time, you also passionately wanted to fund health care, and you believed those 2 things, and you were a rational human being, when you looked at a wealth tax, you would say to yourself, “If my goal is to fund health care, I don’t do the wealth tax,” right? Because it’s actually not the rational way to get more money. There are lots of things you can do to tax people at the point of sale, et cetera, et cetera. We could talk about it if we had more time.

So, you’re right, Jason. The only thing is, this is not a rational act by people trying to maximize dollars. This is a lash-y thing. I’m very optimistic that it gets voted down, but it’ll still have had a cost, and it’s kind of dumb.

And on that note, I’ve got to duck out, guys. I’ve got to go and actually work and figure out how my poor little boutique firm can survive in this harsh and cruel world that we live in.

Okay, you go. Dude, I do just want to stay with you on this one because you said there are a couple of things I really want to understand, because I don’t understand this. How likely is this to actually happen, Jason?

Strange things have been passed. Strange things have passed. And the only thing that stops them from getting passed is that we’re all kind of lazy and we all vote no in general. But you whip folks up into a frenzy, and it doesn’t matter what anyone in the governor’s office or the legislature says. It only needs 50 plus 1.

It is direct California democracy. It’s wonderful and terrible and crazy. It has a type of direct democracy that the rest of the country doesn’t have. So all you need to do is go around everybody, get people upset about billionaires—and many people should be upset about billionaires—and you just need half plus 1, and it passes. And so that is why, no matter what people say or think, you just need half plus 1.

Okay, it happens and it passes. What happens then?

Well, this is my point. And listen, I’m not—as you know, I’m not a billionaire, and I’m not going to get there. I had a chance, but I won’t be a billionaire. I don’t have the same perspective as Chamath and Sacks, but I do think everyone’s mostly missing this point, which is that this is not a one-time thing.

There is a group of folks behind it that have been working on this for many years. Of course they have. This doesn’t come out of the blue, right? They’ve been trying to pass a version of this for 5 years. They finally figured out this is step 1: let’s make it all about the billionaires, right? This is the easy one.

It’s easy to bash the rich billionaires when we have a very bifurcated time, when the wealthy are getting wealthier and jobs are also going away. It’s an easy one to win. That’s why I think Rory’s wrong, because everyone feels like the rich are getting—the billionaires are getting richer. They were all in St. Barts over the holidays, but our companies are doing layoffs. It doesn’t feel very good, so I’m going to vote to tax those guys.

And if it was just 1 tax, then, as bad as it is with Larry, Sergey, and Peter Thiel leaving, it would at least be a bounded thing, right? It would be a bounded thing. But this is just phase 1 of the plan. Phase 2 is that it happens every year. Of course it’s not going to happen once. This is how you put a bow tie on something to make it look good: it’s just once. Of course it’s going to happen every year.

The prior versions of this bill, and the 1 they want to keep passing, have already lowered it to $50 million and then $25 million net worth, and it is on illiquid assets based on the last round in venture. How many deals have you done, Harry, where the last round was at $250 million or more and the founders had material ownership—like, a lot—in the age of AI, right?

And so I do think if this goes as far as the folks that have backed it want, you could have a Detroit in Silicon Valley. When it becomes a meme to do YC or to do South Park Commons, but then build up, get your money, build your team, and then leave, that could be the meme.

Come to Dogpatch, do YC, stay a year, build up your team, and then leave. Is it hard to imagine that being the new SF? It’s not hard to imagine. It’s not that it would go away. It’s just that you leave after a year.

Who wins from this? In any loss, there is often a winner. Is there a state where everyone goes that wins?

The answer is the ones that almost won in 2020 and 2021. It’s that simple, because there wasn’t enough gravity to get people to go to Miami outside of some hedge funds and others. And there wasn’t enough gravity to go to Austin because it’s really not that nice there. But obviously they will win because we already saw it happen, right?

It was just the yo-yo bounced back up when AI came out. It wasn’t worth it to be in Miami or Austin when AI came back out, right?

Jason, would you leave?

Well, first of all, I’m not starting from scratch, right? So bear in mind, I think that is an important point. It’s crazy to me that Sergey Brin, because he’s driving AI at Google, is based in the Bay Area, right? Larry Page I get, and Peter Thiel is managing money. I don’t know.

I’ve thought about it every year since 2020, when it didn’t matter for 2 years where you were. I’ve thought about it. I’m on the edge. I’m not a billionaire, but the financial cost to me to remain in California is super high. What I pay to live here in taxes and others—it’s worth it, but if I had to pay a wealth tax when it goes down to these lower thresholds every year, I would leave. Of course.

Now, does it matter? You could argue. What I thought about a lot is that it would push me over the edge, because then every year I’m paying this massive tax on top of the 50% tax that I pay in California already, or 40% on long-term capital gains. It’s the highest taxes in the country already, and then there’s a wealth tax on top of that. Every year I’ve got to pay 1% to 2% of everything.

One year, it’s actually not—who cares? One year. But what if it’s 10 years? That compounds to a lot, right? We need Rory to do the math, but that compounds to 15% to 20% of your net worth being gone until it gets increased.

So, you got to go to Miami or Austin at some point in your career, right?

I think so, but what I worry more about—and I wrote this, and it already had 500,000 views in a day—is that it just makes sense to leave after the Series B. You should just leave. I don't—and I think YC will get their 7% and the funds will still stay, but you'll just leave. You'll just leave.

It's a terrible idea to leave SF in the age of AI, but I don't think it's as bad as going to Monaco or Dubai or weird stuff like that. I don't think that's the best way to build a startup from Dubai or Monaco. I'm pretty sure that's suboptimal.

But we may go back to distributed teams. We did learn how to do it. It's suboptimal. We may not be all RTO. We learned a lot of skills that we're now putting on the back burner, but we know how to do these things.

We know how to build distributed teams. We know how to work remotely. We know how to do these things. They're not as good, but if it becomes what we do, we will just adjust. We will adjust. It's not that big a deal in tech.

We're not dealing with a meta issue: the wealth gap is just going to spread in the age of AI. It's going to get vaster and vaster, and the social implications are significant. We were all really worried about layoffs and AI taking people's jobs, but people are—

When you and I first met, Harry, a billion dollars was a good exit. When you and I first met, it was a great exit, right? Now $100 billion doesn't feel like that much, does it? It's crazy, but that's also 100 times more wealth for the founders—maybe even more, realistically, right?

That is just a gap that we kind of hide from. I brought it up on the pod. We didn't do it. But when every billionaire was in St. Barts competing with their yachts over the holidays, man, when that gets retweeted, it's hard for anyone not to feel like they want to tax the fuck out of everybody. It's gross. It's gross, right? It's tacky. It's gross.

When you think about the labor displacement, which you spoke about in terms of it really showing up in labor numbers this year, I think you have a real problem. I think that's probably one of my biggest concerns right now, especially in the UK. You see the disparity of wealth just between London and everyone else.

It's going to grow. It's terrifying.

But it's worse. It's actually worse than that. I think, and we can't do anything about some of this, that even for B2B, grounded in what we do, I think we're going to normalize around $1 million to $2 million per employee. That's, I mean, 200 employees at $300 million in revenue. How many does ElevenLabs have? We can look it up. It's probably not that many, right?

As we're able to do startups with a fifth of the headcount we used to, that's just going—even in our little ecosystem, it's going to lead to malaise because we just don't need that many people. It's not about displacing people or AI replacing people. When we can get to $1 million to $2 million in revenue per employee, we just don't need that many people.

Do you juxtapose that with the number of millionaires created by NVIDIA's market cap today who are employees, and does that not pose a dispersion of wealth because of the expansion of market caps?

It is somewhat dispersed. I forget—we can look it up—how many millionaires they have. 20,000 decamillionaires or something like that.

Something like that.

Yeah. And it has already perverted housing markets in the Bay Area and lots of things, but it's not happy.

One in 3 employees at NVIDIA is now worth $20 million or more. One in 3. 18,000 folks at NVIDIA are worth $25 million or more. I'm in Palo Alto now. There are literally 0 houses for sale. There are none, because they're instantly bought up, right? There's nothing.

On the one hand, you can say, “Great, there are 20,000 more people worth $20 million at NVIDIA.” But what it also means is that there are so many types of inflation. There's financial inflation and life inflation: the types of education those folks can afford, the types of housing they can afford, the way it changes the wealth at the Stanford Mall in Palo Alto. It is nothing like just a couple of years ago.

That's going to breed a lot of contempt. If you're one of the folks at NVIDIA and you made $20 million or $30 million, you're feeling great. If you just got laid off from a SaaS company growing 15%, what are you going to do, Harry? What are you going to do when you get laid off from a previously high-flying public SaaS company growing 4%? What are you going to do? Who's going to hire you? The problem is nobody.

We've got to tap into this zeitgeist and this wealth generation, but I do think there is a level of social unrest that will grow over the coming years. It's worrisome to me. I think this bill is part of it, and I get why. That's why I think it's only the first one, because each year that goes by, people are going to be more and more angry at the AI decamillionaires' incentives. They're going to get angrier.

They worked just as hard. “I got laid off from Zoom, and it's been 12 months. I can't find a job. I was a VP.” It's not going to feel very good, is it? It's already not feeling good on LinkedIn, right? I'm already seeing folks as we record this. You know what happens in January? Everyone says, “I've moved on from my company.” Those are the folks that got fired. You can see them.

“I've decided, after 27 years at Microsoft, that January 15th is my last day.” I know everyone they're congratulating in January—they were moved out. It's part of life. But what happens when the next job is impossible?

I'm worried. On that sunny disposition, dude, it's always a pleasure. I so appreciate you.