OpenAI 重组：谁赢谁输，以及 Mercor 以100亿美元估值融资3.5亿美元

Rory O’Driscoll

I just can’t think of a retail IPO that would be more popular than OpenAI. To me, going from $23 billion to $30 billion, I don’t even consider it an up round. It’s not enough.

If they really have gone from $100 billion to $150 billion in less than 6 months, then I can make this conversation really quick. There’s no way they should sell. Andreessen Horowitz is the Red Army of the venture industry.

What you’re basically saying is, “Rory, you can win a deal not because I’ve grafted for 30 years and returned, frankly, billions of dollars to my investors, but because I’m on a P.”

Rory O’Driscoll

If that sentence is true, then 100% Marc Andreessen was right all along. I think Bezos would lay off half his company in a fortnight. It was the right thing. I don’t think you even care.

We have so much to discuss this week. It’s my favorite show of the week.

Rory O’Driscoll

It’s in the top 2 each week, isn’t it? At a minimum.

I would say it’s a top 2 show for sure, right?

Rory O’Driscoll

Yeah, always a top 2. Does that make you feel special now?

They say the silver medal is the toughest one in the Olympics, but I don’t know about Rory. I’m good with it here. I’m good with it.

Rory O’Driscoll

I’ll take a silver.

1. OpenAI's Restructuring: Winners and Losers

Okay. We were talking about where we were going to start, Rory, before this, and you were like, “I think we should start with OpenAI,” given the news today. Learning from the feedback that we get, I would love to start with you just explaining a little bit about the news that’s just come out about OpenAI and their structure, and we can start there.

Rory O’Driscoll

Sure. The big news today is that OpenAI cut their deal. They cut their deal with Microsoft, and they cut their deal with the attorneys general—plural of attorney general—of Delaware and California. That means they have been able to implement the restructuring, which means they can raise their capital, which means they’re out of the messy, complex trap they had put themselves in all those years ago in terms of their structure, and they’ve gotten it done. That’s the big-picture news.

There’s lots of information one level down about who won, who lost, and who got what economics, but that’s where we’re at. It’s still got some opposition. Elon can still litigate and say, “I don’t think you should do that because I gave this money to a charity.” But possession is 9/10 of the law, and once the attorneys general have allowed it and they’ve actually converted, it’s a lot harder to unwind.

As of this morning, as I understand it, OpenAI has the charity—the charitable foundation—which is now one of the most well-capitalized charitable foundations on the planet. Underneath that, there is the company itself, OpenAI, which is a PBC. I can never remember the initials—basically, a for-profit company, but it also has to take into account more than just shareholder maximization. I’m sorry, I’m having my dyslexia there, right?

That’s the entity that’s been created, and that’s the entity into which you can invest. There are other companies like that. I think Patagonia, for example, has the same status. This is not a crazy thing now.

This is not some weird thing, like the old OpenAI used to have this disclosure: “You should regard this as a donation. It can all go to zero.” This is a real, honest-to-goodness American corporation—a different kind of American corporation, but they can go public with this. They’ve gotten out of the straightjacket.

That’s the big news. It means, for example, that an OpenAI IPO is one enormous step closer. I’m not saying they have to, and I’m not saying they will, but it’s big news today.

I’ll just throw one thing out that jumped out at me on the deal. The overall structure—Microsoft owning 27%, the employees owning 20-something percent, and the nonprofit owning 20-something percent—those are all roughly what we expected, right? There are some nuances on how AGI worked that are a little interesting.

But the craziest thing in this deal, because it’s unprecedented, I think, in our lifetimes, is that OpenAI said Sam Altman will still have no shares—no shares in the combined entity. At the same time, we’ve got Elon Musk arguing that he deserves a trillion-dollar pay package so the robots don’t kill us, which I think he deserves. Okay, I think his VCs will say he deserves it.

There is something unimaginable to 99%—to almost all of us in the world: that he should have a trillion-dollar pay package. But the scale has to be relevant to the outputs. I don’t believe it will ensure the robots don’t kill us. It’s a little crazy, but on the other hand, Elon wants a trillion.

Everyone was saying, “Show Sam the money,” when he was fired as CEO over a very long weekend, right? The night of knives, or whatever. He was fired by this crazy nonprofit. Everyone had to revolt to bring him back.

Fast-forward to today: that same nonprofit is still in charge. Now, listen, there’s been turnover, but it’s still sort of in charge, and he has no shares. We could hypothesize why. It’s pretty transparent.

I actually think it gives him, in some ways, more power as well as less power. You can’t assail the man for capitalism when his other billion-dollar entities are the ones that let him finish off the McLaren collection.

But we’ve never seen someone own nothing like this, have we? Have we ever seen anybody own nothing? It’s crazy.

If we just drill down on winners and losers from this restructuring change, who are the winners and losers here?

Rory O’Driscoll

I think the lesson here is one that every attorney knows. You often hear this expression from litigators: “What’s the case worth?” They look at the filings from both sides, and experienced litigators look and go, “Okay, we’ve got these 3 points. They’ve got those 5 points. In the end, we’re going to win on the 3, they’re going to win on the 5. This is the way it’s going to come out.”

Then there’s a whole lot of human drama because humans are like that, and we yell and we scream and we have juries and all, and then typically things settle out for what the case was worth. This settled out for where the cases were worth. Let me tell you what I mean by that.

When you look at the thing—and Jason said, “Not only no surprise, but no surprise for a long time”—Microsoft had a fair amount of leverage. They used it. They got a great deal. They put in $13 billion, and they got a 10x on their money as of today.

They got a lot of AI leverage. They got some going-forward IP rights. They got a significant going-forward contract for Azure business if they want it, which we can come back to. Overall, they didn’t push it to the point of breaking, but they got pretty much what they were going for, and they still have some revenue share, which surprised me.

So they got a great deal. Again, I go back to my comment: Microsoft corporate development and lawyers deserve a gold star from their shareholders in a way that, frankly, Microsoft R&D does not. The proof of this is that this morning, Microsoft stock is up nicely. They’re like, “Thank you for the $100 billion. We’re up.”

There are 3 big winners. The second big winner is, just to step back, the charitable foundation. There are some people, even today, griping and saying, “We did this all for charity. It feels wrong that there are any capitalists involved.” I get that, especially if I’d given the seed money.

But stepping back, somehow in the midst of this, we’ve ended up funding a wonderful $135 billion charitable foundation. That’s a significant contribution to whatever good they hopefully will do with that money, and they’ve already made some announcements about AI for medicine.

So there’s $135 billion out there that’s not going into someone’s pockets to buy yachts, boats, and football teams. It’s actually going to try and solve the world’s problems. Whether they can or not, TBD, but yay, because I would have laughed in 2016.

The people who started OpenAI saying, “We want to do good for the world,” have, at a big-picture level, succeeded. They’ve built something worth $130 billion that is a foundation they can be proud of.

Obviously, the employees own a third, and now it can get liquid. Yay them. The remaining investors can kind of exhale and get a sigh of relief. SoftBank can put in its $22 billion, and the investors as a group—I mean, SoftBank will own about 10%, and everyone else will own low single digits—and everyone won. Everyone got roughly what the leverage would make them get.

The final winner, I’ve got to say—and I know this is a lot—is that Bret Taylor just wins the best board chairman of the year award again, for the second time in the last 5 years. He totally won it as board chairman of Twitter, where he jammed that down Elon’s throat for $44 billion despite his opposition, and he won here today because he unraveled the mess and set it all up for a win.

It’s a really good settlement.

There are no known losers.

Rory O’Driscoll

Well, I mean, the losers would be Elon, who feels he didn’t want any of this to happen, so he’s miffed. And the people who think, “Oh my God, it should all have stayed not-for-profit.” They feel they’re losing. There are a lot of Twitterverse comments today about that.

But pragmatically speaking, if you think about it, Microsoft put $13 billion in here, and they’ve made a 10x, which is a good return. But we’ll talk about returns. It’s not like it’s rapacious relative to the risk they took. No one else was writing OpenAI a billion-dollar check in 2019. Microsoft did, and they got that return. Absolutely not.

Yeah. I don’t think there are any losers at all.

Rory O’Driscoll

I don’t think anyone got much more than they deserved for the risks they took and the work they did.

David George

So I can tell my LPs that the little 6% stake deal I’m doing today is okay because, look at OpenAI. I don’t need to get double digits. Point to this one. Don’t beat me up on the double digits.

Totally. No, exactly. And you’re right, Jason. It just shows that rules of thumb are made to be broken. 90% of the time, your ownership target is a really meaningful metric, and it should run your business on it. And, you know, 10%, 1% of the time, who the hell cares? 1% of the biggest company on the planet is $5 billion.

David George

There you go. The 1 micro-thought I just had was that it’s very confusing. There’s all this circular financing: Nvidia giving them money, AMD giving them 10% of the company, and Oracle raising an unprecedented amount of debt.

So what I mean is, I don’t know how much equity OpenAI needs, but it seems to me that if they’re coming up on being the first trillion-dollar startup and more, and if they could IPO at $2 trillion—which is crazy by any historic standard—then, as a company with no stock, maybe they can raise another $200 billion, right? I mean, at least in theory, that’s a lot of capital to access. But my point is, this potentially unlocks 4 times more equity for them if the public markets are different from the private markets because of the potential valuation at which they could IPO. They may need that. They may need an extra $200 billion to go the distance.

I think that’s an excellent point, Jason, and you’re exactly right. It means that there’s 1 more set of winners here: all those people whose stock popped because they have a [__] promise from OpenAI to buy a whole bunch of their stuff in the future with money OpenAI didn’t have. Now, at least, they can say they can go get that money.

The problem, as you know, is that I’m skeptical that Oracle will collect the last dollar of that cloud contract. But at least you can now say, hand on heart, that the company OpenAI now has a sensible corporate structure. They obviously have an amazing business, and if they need to raise another $100 billion, it’s not crazy anymore. It’s just banking and math.

So you’re exactly right. This thing may fail for you. There may be business issues around, you know, return on the thing, but we’re out of the stupid-corporate-structure-getting-in-the-way-of-everything stage of the business.

I just can’t think of a retail IPO that would be more popular than OpenAI, right? It would bring everyone out of the woodwork to put a little bit of their life savings into it, ignoring whether the valuation makes any sense. This would have to be the most popular retail IPO of all time, right?

David George

I think you’re exactly—I think, yes, you’re exactly right, and it’s now doable. I mean, people aren’t going to be reading the prospectus and saying, “Maybe we won’t make profits.” They’re not going to be reading the thing about $250 billion of cloud commitments to third parties. They’re just going to say, “Let me get some of that OpenAI,” right?

I think it would be interesting, actually, to your point, to check on the secondary valuation pop today for OpenAI trades. Right now, the weird thing is SoftBank is closing 2 separate deals. They’re closing their direct investment at roughly a $300 billion pre-money valuation, and they’re also doing a share buyback from some existing employees at a $500 billion valuation.

Literally, you have the same security trading at 2 different prices. I think that’s true for other investors, too. They’ve locked in the earlier price. When the other investors—I think D1 is in this, too—committed to OpenAI a while back, they said, “Hey, we’ll give you money at $300 billion, but you’ve got to get your conversion done first.”

So now that the conversion’s done, they’re going to put the money in, but they’ve already had a markup before the money. Oh my God, they’ve had a markup before the money’s gone in, right? Because SoftBank is marking itself up by doing business at $500 billion in a secondary, right?

Literally, you’re going to wire money at $200 billion or $300 billion, whatever the number is, and then the next day you can say, hand on heart, “The current valuation of this is $500 billion.” So there you go: you’re 40% IRR in an hour.

Do you think there will be a trillion-dollar company in 2026?

David George

On the current trajectory, maybe. All I can answer is, on the current trajectory and without the euphoria, as you mentioned, it’s probably 2 years, because you do get some attenuation of growth at scale.

At the current thing—at $500 billion and $12 billion, let’s call them boring GAAP revenues rather than ARR—it’s kind of 40 times GAAP, and if it’s $20 billion, 25 times ARR run rate. So my guess is it would take 2 years in the normal course, but you might see that euphoria moment.

It’s not crazy. It’s not like it’s never going to happen. It’s within the trajectory; it’s within the strike zone if anything like the current growth rate continues. If it slows—I mean, as a reminder, when you’re trading, and we saw this in 2021, if you’re trading at 40 times revenues and your growth rate slows, it’s nasty and you fall sharply. But right now, they’re growing, so they can get it.

I would love a $2 trillion IPO.

David George

Anchor from here. Actually, the interesting question will be the dynamics of the negotiation. Again, when you have master corporate financiers like Bret Taylor and Sam Altman on your side, it’ll be fun to watch the bankers beg for that. They might pretty much do it for half nothing just to be on the biggest IPO of all time.

I mean, technically, Saudi Aramco had a couple-of-trillion-dollar market cap, but nobody really cares. Let’s get real. It’s an oil company in Saudi Arabia. This would be one for the ages, and I think every banker on the planet will be making decks as we speak and calling on Mr. Altman and Mr. Taylor.

Dude, you’d do it for free for the credit.

David George

You probably would.

Yeah, no doubt. Is there anything on OpenAI that we haven’t covered that we should cover? Do you think the browser, the RSUs, anything that you’re like, “Ah, we have to cover that”?

David George

I mean, it’s all second-order stuff. We can come back to it if we have time, but I don’t know if it’s—I mean, that’s the big story. They’re free.

You know, the only thing that’s mildly interesting about the browser is that I’m just waiting to see, unlike Sora and such, whether the world will really care. I just wonder how many—it’s just tough to keep track of how many stories tech VCs on X create.

It’s not that it’s not interesting, right? I mean, obviously, the browser is where we live a lot of our lives, but it could be like MCP. It could end up being interesting, but not really the game-changer that VCs and others think it’ll be. We’ll just see. I don’t know.

David George

Right. Yeah, I don’t know that every single average consumer will find it a game-changer to find a new browser rather than type ChatGPT into their browser and get enough memory. We’ll just see whether they really want it to pick its Spotify selections for us.

I just don’t know if it’s a game-changer or more of a land grab. We’ll just see.

2. Andreessen Horowitz's Raise $10BN in New Funds

Okay, we’re going to talk about Andreessen’s new funds. Andreessen dominates so much of the venture market today: $10 billion split across $6 billion in growth, $1.5 billion in AI apps, $1.5 billion in AI infrastructure, and $1 billion in defense. My word, what a big raise.

I would love to understand: Is this just a new normal of General Catalyst and Lightspeed and the mega-platforms raising like this? Is this different? How did you analyze this news?

David George

Honestly, I thought they were small. What I mean is, I didn’t think it was small until I saw the breakdown of the funds. At first, when we talked about this before, I’m like, “$10 billion. That’s unprecedented,” right?

But when I look at that and the new Sequoia fund—the $200 million Sequoia seed fund—that’s not that big compared to 20VC. What’s the 20VC? 20VC is $150 million out of $400 million or something, right?

Yeah, Ror and Rori [?] is more A and B, but it’s not $200 million. It doesn’t really get you out of bed at scale. And $1.5 billion for AI apps, when you’re investing in ElevenLabs and friends and Replit, doesn’t seem to get you very far.

The seed funds and the AI fund for Sequoia and Andreessen were smaller than I would have expected in today’s insane world, where even $50 million at a YC Demo Day could be a low valuation. I wasn’t expecting that.

No, I thought it—look, step back. Is it the new normal? Yes, it is. The business model that Andreessen and a couple of other firms have brilliantly pursued—this is the world we live in today, and as investors, will live in for the next 4 to 5 years, right?

Somewhere down the line, you’ll either have that it works across a cycle, and then this will be the norm forever, and you could envisage a world where this is a little like where venture becomes more like investment banking, where there’s Goldman Sachs, JPMorgan, and the rest of us are boutique players, right?

Or the other thing that could happen is the returns from the bigger funds are slightly disappointing, and there’s a little bit of a tilt back to more mid- and small-cap venture. But this is the dominant modality today. This is what top-dog venture investing looks like: this kind of scale, this kind of dollars at work. There are 4 or 5 other firms doing this.

Is there any excuse to say that I can’t compete with Sequoia? I mean, scale can outbid them.

David George

I think you can compete with a $200 million seed fund or a $1.5 billion A/B fund. I think you can compete. I agree that the sentence, “I can’t compete at seed with Andreessen because of check size,” doesn’t make sense, except in one very derivative way, right?

Because I agree, at the end of the day, someone’s raising $10 million, and you have $10 million, the other guy has $10 million, and of course, right, there could be one—there are 2 arguments against what you’re saying, though, right? There are 2 ways in which the fact that one checkbook comes with $10 million out of, let’s just pick it, a $400 million fund, and one checkbook comes with $10 million out of a $10 billion colossus—

Rory O’Driscoll

Right? Let’s just say there are 2 ways in which the colossus has an advantage. The first is that they can literally decide they’re not pricing this round; they’re buying an option on the next round. If Harry’s trying to make his money on seed and they’re simply trying to set themselves up to make the money on the Series B, they can, by definition, pay a higher price, because any option always trades higher than the intrinsic value of the asset in question by virtue of the time value.

So they can pay a quote-unquote stupid price because they have a different model. The second way they can win—and I’m really internalizing this now—is the wall of news. At the end of the day, when you have a $10 billion fund, you always have shit going on. You don’t talk about your bad stuff, but provided you’re modestly competent—and these guys are far beyond modestly competent; they’re extraordinarily competent—you always have some good news in the portfolio. You always have exciting things, and you’re probably going to be in some winners.

I don’t buy the wall of news anymore.

What? Okay.

I’ll tell you why I don’t buy a wall of news. I’ve learned this first from Dr. Harry Stebbings, and now I’ve learned it from others. Harry’s in some great, great investments that he quietly but relentlessly reminds us of—the complexities and the, how he says it with his British accent. How do you say “more” in British?

“Mur.” He elongates the vowel or something like that, right? And he tells us all the great stories. What I’m saying is, you can get coverage. You don’t have to write $10 million checks to be the with-participation fund guy. You really don’t. The scale could do 50 deals if it wanted, with participation from Rory from Scale, who we love from the pod.

So, I think the option thing is a bigger deal. I just wonder, if you have a $200 million seed fund at Sequoia, how many options can you afford before your whole fund is options? That’s the question. The one thing I will say is that the with-participation thing does not work for publications, and this is very in the weeds and granular. With participation, playing in rounds with smaller checks does not work unless you have an existing brand. For a random tier-two or tier-three firm, if you were to do with participation—not the TechCrunches, the big—

You have to end the fifth paragraph—

Rory O’Driscoll

No, they honestly don’t. They really don’t. And so, yes, we do—

He has a following now after this pod. He’s got a pretty big following. Rory now would be—

Rory O’Driscoll

Even if I did—which I would stop; it makes me cringe—but even if I did, you’d be proving my point. What you’re basically saying is that Rory can win a deal not because I’ve grafted for 30 years and returned, frankly, billions of dollars to my investors, but because I’m on a shit pod.

If that sentence is true, let me finish. If that sentence isn’t true, then 100% Marc Andreessen was right all along. He said it’s not. I don’t think, in the end, investing is a media business. But I do believe that—and I’m going to give them enormous credit—I admire people who pull off a strategy, articulate a strategy, and pull it off.

They tilted the table with their media strategy. Here I am on this podcast, in my little, tiny, humble way, trying to say, “Okay, this is the new game, right?” So I do disagree with you, Jason. I think the wall of sound is a combination of the $10 billion, the media presence they generate, and all things. I think, at the margin, the bigger funds are harder to beat. You just have to say that it’s not by any means impossible.

But what I’m saying is, I think Andreessen Horowitz’s $10 billion raise—and, by the way, I think the structure of it actually also makes sense: the 3 different individual funds and then the growth fund. It makes sense, I think, both for structural investing reasons and probably also for human capital management reasons. You can give your chief lieutenants each a little fief where they can feel in charge, which is a good way to keep them.

I think the strategy works, and as I said, provided the long-term returns are there, in terms of tilting the field of play in their favor, I think it’s been successful.

I would argue, actually, that there are massive advantages to them. I get you, Jason. We’ve got $275 million for a Series A fund versus their $1.5 billion. That is a lot more money for management fees to pay great people. That is a lot more carry. That is a lot nicer office space, which founders do get wowed by, like it or not. That is a lot more events to host where you can have serendipity.

There are a lot more things that I think scale and AUM buys that do increase.

One of my favorite expressions is from the Russian Red Army: “Quantity has a quality all its own.” In other words, when you want to take Berlin, at some point what you do is just get 2 million people willing to die and march them forward.

Andreessen Horowitz is the Red Army of the venture industry now. They got the $10 billion, and they’re going to march it forward.

But listen, we can move on. Here’s my point you guys didn’t address. Of course it’s $10 billion. It’s $10 billion of 30% carry, $10 billion of 2% fees, and $10 billion of all this. But when you break apart the funds, I really don’t think Sequoia having $200 million is that different from what Harry has.

I don’t have $1.5 billion for their AI app funds. I think it’s only twice the scale. I just don’t think air cover is an excuse. You have to work twice as hard as they do. That sounds right. You should have to work about twice as hard, but it’s not really $10 million in a billion. When you break the funds up, it’s not.

I agree with that. You do have to work twice as hard. I totally agree with that. But, as a comment here—and again, I mean, hard-nosed as a comment—the definition of a good strategy is if you have a strategy that doesn’t require you to work that hard. Therefore, by definition, having $10 billion is a good strategy.

I think the interesting thing, and we said this before, but my bigger concern is that the only people for whom this might be a net negative are the LP investors. As I say, we don’t know that yet. Maybe it’ll be a wildly successful strategy, in which case it will run the table. Or maybe it’ll be modestly successful, in which case, in 5 or 7 years, the LPs will start going, “That latest-stage fund gives me a good return, but it’s 12% or 13%, and maybe I could get that in the markets now that there are more IPOs. Maybe we should just throttle back our allocation.”

Some version of that. But until something like that happens, for every other player in the market, having $10 billion is better than not having $10 billion. That’s where the game is.

I also think, for everyone listening, they don’t often understand the stapling that is required to be in these funds. If you’re an LP and you want to be in the early-stage fund, very often, with top-tier brands, you have to put in 2 times that into another fund to get that $1 in the other. You don’t get your pick of which fund. Very often, you have to be across all of them.

Yes. When you see that, you realize that, intuitively, leaders of venture firms understand the concept of bundling at their core. Bundling and rent extraction are fully understood.

3. Mercor Raises $350M at a $10BN Valuation

Jason, you said something about this company, Mercor—I think it was. I think it might be a 20VC company, to be fair. They announced yesterday they’ve raised $350 million at a $10 billion valuation, led by the person who led their last 2 rounds at $2 billion only 8 months ago, which is Felicis.

This company has gone to $500 million in revenue faster than anyone else, I believe, in history. I think it was 17 months. Again, this is not bias—I hate freaking biased shows. I sometimes listen to them. A lot of people say it’s not real revenue; it’s GMV. I want to hear how you guys felt about this round and the speed of revenue acceleration. If it’s real revenue, I want your thoughts.

It is quote-unquote real revenue, in the sense that they ship things and get paid. GAAP requires you to book it as revenue. Maybe step back and give people some context. What Mercor does is provide humans with specialist knowledge, and their customers are the large foundation-model companies.

They bring this human talent to bear to help the foundation-model companies train the models by providing human feedback. People would have heard of RLHF, reinforcement learning from human feedback. These are the humans that do that.

If I’m OpenAI and I want to teach my latest model how to do advanced math, what I need is a whole bunch of doctorates and PhDs who understand math and are available to pose questions to the model, judge the model’s answers, and give feedback on which answer is correct and which answer is not.

By doing that, by giving this human feedback, I think of it as pounding the model into submission, where it eventually says, “Okay, I’ve learned this shit,” by adjusting the weights. That’s what’s going on here, right? It’s an astonishing amount of training.

We think of this all as happening in NVIDIA GPU chips. There’s an astonishing amount of human training that is required to make these models work.

Right. So that’s kind of the market these guys are playing into. Scale AI, which was partially acquired by Meta, is in the same broad market.

Mercor has done an amazing job because, 5 years ago, as Jason said—I love the expression—we were labeling cats. You had people, often overseas, charged with labeling cats and not getting a lot of money. I think Mercor realized that the market today is not labeling cats. It’s, in fact, answering complex physics questions, math questions, and bio questions, because the models know how to label cats now. What they need to do is reach the outer edges of human knowledge.

It’s a very different set of humans that you need. Mercor did an amazing job of assembling all these high-end folks and making that product available to the model companies. That’s what they do. In that context, the growth rate isn’t surprising because these model companies have grown faster than any company in human history.

They’re spending $300–400 billion on compute. They’re probably spending $3–4 billion on RLHF, and they were spending zero 5 years ago. It’s an explosive growth market. It didn’t happen in a vacuum; it happened because our customers want their stuff.

I’ve spent quite a lot of time in the market, which will surprise you, Rory, to hear—that I’ve been thinking about markets more deeply. It all started with talent acquisition: if you got the talent and could provide it, fantastic, we’ll pay you. That’s pillar 1. Pillar 2 is data acquisition: you provide it to us; we don’t extract it. You provide it to us, and then you measure the quality of it, too.

Now we’re adding the third pillar, which is the implementation layer. We now expect you not only to do those first 2 things, but also to implement them efficiently and make sure our models get off the ground more effectively. With that, you also see increased pricing and the willingness to spend much, much more from the model providers.

The other thing I will say is that you have a concentration of buyers unlike any other industry. 2 buyers are 50% plus of every one of these labeling providers’ revenue. There is real revenue concentration there. They’re high-quality customers, but you do have that dynamic as well.

David George

Agreed. That was well put because it actually makes it an interesting question from an investing perspective. On the one hand, there’s nothing better than a customer who has an urgent and compelling need, who’s well-funded—we just agreed they’re well-funded—who wants you to grow with them and wants you to help them get a bunch of stuff done.

What you’re saying is exactly right. OpenAI and all these companies had simple requests 5 years ago. The more the complexity goes up, the more you meet that complexity as a vendor. Mercor has done that in spades. The more revenue they’re going to give you, the more they’re going to be shoveling money at you because you’re solving their problems.

They’ve got a lot of shit to solve. They don’t need to be thinking about this. “Mr. Mercor, if you can make this go away and get me 5,500 doctors, this data, and this answer, and integrate it into our system, I will pay you money because money I have in spades. Time I don’t got.”

Rory O’Driscoll

For sure. But I’ve got to imagine you guys—and especially Harry—would know better than me. I don’t think it’s stress-free. I have a portfolio company that was doing a vaguely similar attachment to a large AI business model, where the contract was growing to 8 figures. That renewal was stressful AF.

Yes. [Laughter]

David George

Nobody, once you get to 8 figures, says, “I know these guys have more money and time than engineers.” There’s a point where you turn around and say, “Maybe we should do a little something ourselves.” OpenAI is building its own chips. I’m not saying this happened at Mercor or Scale AI; I’m just saying Harry’s point about having to radically go up the value chain means more revenue but also more stress.

I don’t think anyone’s just shoveling money at Scale AI and Mercor without even thinking about the margins.

Rory O’Driscoll

I totally agree, and that’s what makes your venture fun and challenging: you’ve got 1 enormous positive, a great big, honking market growing like a weed. You’re exploding, and then you’ve got 2 negatives.

The first is that your margin profile isn’t amazing because the gross revenue is $500 million, but you give 70% of it to the doctors and mathematicians who are doing all of whatever it is. Some percentage—I’m not going to speculate. The second fact you have against you is massive customer concentration.

At some point, they’re going to say, “I’m giving you $200 million,” which means you’re making 30% on that—$60 million. Hell, maybe you do it for $40 million, right? From a long-term value-extraction perspective, you prefer to be OpenAI and have 800 million customers rather than be Mercor and have 2.

When you look at those positives and negatives, what you say to yourself is that this is fundamentally a bet that the AI capex train will keep running for 2, 3, or 4 more years. If the AI capex trade slows down and OpenAI is only—

Enough is 2–3 years at least. You’re right, Jason—at least 3–5 years. In other words, it’s a permanent new thing in growth because OpenAI isn’t going to focus on getting efficient until it’s dealt with hypergrowth.

As long as it’s not getting efficient, you probably can lean in. If it slows down, then the positives—the growth rate—go away, and all the negatives come back to bite you. If you were to say to a public-market or hedge-fund guy, “Find me a bet that has the maximum exposure to hyper-AI-capex growth,” this would be right up there with Nvidia. I’d say, “Yeah, I like this risk, man.”

If you’re doing this at $10 billion, what are you underwriting it to? What does that math look like?

Rory O’Driscoll

As we’re going to discuss later, you should be underwriting anything to a 3x. You’ve got to be at $30 billion to make it worth your while. At $30 billion—even at 5x revenue—my God, that’s $6 billion. You’re underwriting a lot of training data. That’s a sobering thing.

In the short term, though—in the short term—

David George

If they’re at $500 million at $10 billion, growing at an unprecedented rate, we’re all ignoring gross margins in 2025. It’s only 20x revenue. This is actually where we’re seeing revenue compression, right? We talked about this with Cliff and others. He doesn’t like the revenue compression at Canva, right?

Where this 20x ARR was—was your last deal lower than that, Harry and Rory? Higher or lower than 20x ARR on the last deal you did?

It was higher and higher. [Laughter] And, Jason, you’re exactly right. You actually said something else there that was really insightful. Your comment is that there are 2 modes of thinking about an investment.

There’s 1 mode that says, “Pencil me out the next 5 years. How do I think about the end state? What market do I have?” And therefore, what return are you underwriting? I think if you assume 5x revenue and you want to be at 3x, you’re underwriting $30 billion, which means $6 billion in training revenue. That makes a man pause.

On the other hand, you can say it’s growing 5x, and if the multiple just stays constant and it grows 5x for another year, I’ll be 5x up. In fact, that’s just what happened on the last round. It was at $2 billion—I don’t know—at $100 million, and now it’s at $10 billion at $500 million.

Near-term revenue traction is saying to everyone, as long as this keeps happening, you can go quickly. When you look back now, we’re talking about whether it’s worth $10 billion, but go back and give them credit: you look at the $2 billion that somebody paid 6 or 7 months ago, and you’re like, “Oh my God, that seems cheap,” because they’re at $500 million already.

When hypergrowth happens, it’s tempting, and it often pays, to lean into that growth. The Anthropic round at $67 billion earlier this year now looks dirt-cheap. If the hypergrowth comes at the size of growth we’re dealing with now—which is, to your point, Jason, triple, triple, double, double—I don’t know what the word is, but if you grow 5x or 7x, you can grow into almost anything.

This whole thing, top to bottom, is 1 big-ass bet on AI-capex hypergrowth. As long as it keeps happening, I don’t want to quote Chuck Prince, but we all know the quote.

David George

No, we don’t. You were young then, Harry. Chuck Prince famously said in 2007, in some version, “As long as the music is playing, you’ve got to stay on the floor and keep dancing.” It turned out you should have sat down and not danced anymore, obviously, given the way 2008 happened.

In other words, when things are working, everyone just tends to lean in.

Well, when things are working, everyone just tends to lean in. Maybe Ramp is a fundraising machine. They raise every few months.

Rory O’Driscoll

Yeah. I mean, every—it’s nuts. Reports say they’re raising a new round at a $30 billion valuation, reportedly. What are they doing with all the money? Do they need it? Is this a game of customer acquisition and brand? Do we just do rounds now to continuously stay relevant?

To Jason’s point, we talked about this a few weeks ago on the last raise, and I have nothing to say that I feel like I didn’t just say in those comments.

David George

Yeah, there does appear to be a combination of insane demand for the product, for the company, for the stock, and an ability to use that to create this aura of inevitability. To some extent, as I remind you, this is more than most companies: this is one where constant growth requires lots of capital because you're in a capital-advancing business.

You probably have—every dollar you add of revenue takes $5 of capital because you've got to finance purchases, because you're effectively recreating Amex. So it may well be—I haven't seen the numbers—that there's a larger demand for capital here than the average deal, right? I'm sure they can leverage some of that, but if you're going to $1 billion, you probably have a $5 billion balance sheet. I remember doing the math at one point in time. So if someone's lending you $4 billion of that, they're probably going to want a $1 billion equity cushion.

Going from $23 billion to $30 billion, I don't even consider it an up round. It's not enough. Let's say I'm a seed investor. I've probably had a ton of dilution. So, let's imagine I'm down to 2%, which would be great. I have a $400 million position at the last round. Right now, it's worth $480 million with dilution after this next round.

I mean, it's a lot, but it's not doubling my position like a classic round is. These little rounds—$22 billion, $25 billion, $30 billion—and then, with a lot of dilution—not maybe at Ramp, but with a lot of AI companies—you could see 10% annual dilution in these companies or more. You might go from $22 billion to $30 billion and have the same price per share. It's possible, right?

I just don't really care about these micro-step-ups that look great. You're like, “$30 billion.” Well, if the last round was at $10 billion, impressive, right? If the last round's at $3 billion, like Mercury, impressive. Here I'm like, “Meh.”

That's cute, but I'm pushing back for real.

David George

For real. For real.

No, I know it's for real. I know you.

David George

But another spin on what you're saying is, with the exception of the new AI companies, that probably is the kind of IRR you should expect to get in a mature growth private company, right? Remember, one of my insights over the last—I mean, sometimes I just realize the obvious—is that there's really early- and late-stage venture, and then there's venture for companies that already could comfortably be public.

I think we need a different word for that. It's not even late-stage; it's private-as-public. It could just as easily be a small- or mid-cap stock at this point in time, public, right? You don't expect mid-cap stocks to gap up 3x in a year. You expect the overall market to go up 11%; the best companies grow 30%–40% year-on-year.

So I look at this and go, there's no reason to assume that the return to the stock should be different just because it's held in a different corporate structure. The company is the company, independent of whether it's public or private.

The reason I mention this at such pedantic length is I think, as you think about these late, super-late growth-fund things, the return they will realistically get, absent the AI lift, is some version of the small-cap, high-growth public-market return, right? For those guys, because you're going in, Jason, with your venture rule of thumb—and even on a late-stage company doing $50 million, going to $100 million—it should be that 2x–3x step-up, financing event to financing event.

When you're doing $1 billion, you're near profitable. You should be public. You're going to see these much smaller percentage step-ups because you really are in a different asset class. You're public stocks hiding in private, which is why Stripe's most recent change was from $91 billion to $110 billion. It's the same kind of thing, right?

But you just said we're still underwriting. Harry asked you what we're underwriting to. You said at least 3x. So if I'm going to underwrite to 3x and I did the last round at Ramp at $23 billion, it's not really getting me to my 3x.

Rory O’Driscoll

No, listen. You are, dare I say it, confusing on your pronouns, right? You said we're underwriting, or you're underwriting, to a 3. I'm trying to underwrite to a 3x.

David George

If you're running money and doing companies at a $30 billion pre-money valuation, you're probably not underwriting to an overall 3x on your fund. Now, you might get—let's be fair—the most iconic company of its generation, OpenAI, will give you a 10x.

But I don't think you are underwriting to a 3x when you're doing ultra-late-stage, billion-dollar revenue run-rate, $30 billion valuations, and you're definitely not underwriting to a 30% IRR, right?

I'm with you.

Yeah.

Rory O’Driscoll

Well, listen, my only meta point is—and Harry's made this in the early days—the massive dilution at Anthropic and others, right?

Totally.

David George

I'm just less excited. I found that the things that move the needle for my little portfolio—some of these headline rounds don't always do it. It's not always that simple, right? It could be years have gone by. There could be massive increases to the round, massive dilution, and you might have forgotten that the last round was pretty high, too, back in 2021.

You might even be flat or down. It looks great, but they're not always as sexy for the IRR or the markup as you might expect. That's all.

Rory O’Driscoll

I, by the way, totally agree with that, and I think that's because you're busy trying to turn $10 million into $100 million or $200 million, but there's someone else out there who's very happy to turn $1 billion into $2 billion.

David George

Very happy indeed. And it's going to make more money than you, just to make it even sadder, dude, right? Way more money. And keeping your IRR above 40% is not easy in the later life of a fund. It gets really hard, right?

Rory O’Driscoll

Agree. But for those guys, again, I go back to—and for those guys, a different game. So, yeah, the 2 biggest changes of the last 3 to 5 years have been the advent of what's happening in AI and the transformation of the late-stage and IPO marketplace to this ultra, ultra-late stage, where companies are way beyond the IPO threshold and still private. Those are the 2 biggest changes in the game.

I completely agree. And on the second, I go back to actually Jason's point on the modest size of the Andreessen funds, because when you take the $6 billion growth fund—minus fees, you're at $4.8 billion—you've got 22 $200 million checks. It's not that much. It actually feels very reasonable in terms of size.

David George

To their credit, in that chart that went around, they did do a good job of recycling.

Yes.

They did an excellent job of recycling.

And recycling is good. So we can assume they get the full $6 billion out, but fair point. It's not a lot of checks.

4. Spray and Pray: Does it Work: Data Breakdown

David George is very, very good. I'm always impressed by him. I want to move to spray and pray. There are different models of venture. You can be concentrated, or you can be spray and pray. Spray and pray, for those that don't know, is obviously having a broadly diversified portfolio, investing in lots of different companies rather than a few with more money.

There's always a question of whether spray and pray works. There was some data released by Carta. What did the data say, and how do we think about spray and pray today?

Rory O’Driscoll

I disagree. I read the data. I also read, Jason, your blog post on disaster, and I think your characterization of spray and pray is wrong, to be very direct. I think the data was awesome. Let's start with the comment: the Carta data was awesome and very pleasing.

Stepping back, what it told people—for listeners—is they looked at all 547 2018 Series B investments, and then they did a histogram of where they came out: less than 1x, 1x to 2x—

Oh, that was Series B.

Rory O’Driscoll

So, 547, and then 18% of them were greater than 5x, about 10% of them were greater than 10x, and 1 deal—Figma—returned 100x, right? That's a pretty sizable chunk of information. What's it telling you?

I found it really interesting because, in fact, we typically do A's and B's, and I was very happy. By the way, it was exactly what the distribution for our fund model—we think—has to be, which is 30%. They were actually 35%; we would have said 30% less than 1x.

We do wider buckets: 1x to 5x, 50%. They broke that into 2 buckets, and greater than 5x, 20%. So the distributions on 547 deals match pretty much what we're saying, right?

That's a long answer. So if you hit those 3 buckets correctly, the blended return—I know it from our fund model—is 3.7x gross, 3x net to the LP, right? So if you look at that business, if you look at all those things, and you get enough slots in the buckets—in each of the buckets, the good buckets, the 5x bucket and the 10x bucket—you end up with a 3x net to the LP.

So the first piece of good news is, if you do it right, the return was available to you. The interesting thing is, to your point, is the right strategy spray and pray, and I don't think that's what it—and that's why I was jumping back on it—I don't think that's what it said.

I think it says you have to pick very carefully because, obviously, if you do a lot of deals—I mean, Jason picked on the negative. Looking at the same data, Jason picked on the negative, which is two-thirds of all deals are less than 2x, which means they just don't help.

Correct, Jason. That was the point you made in the blog, right?

David George

Yes.

And what it says is that picking is so important.

David George

But did it say that?

I know you didn’t say that, but I think the data says that.

David George

But first of all, you’re right. To me, the data was shocking as someone who is a concentrated investor, because this is scary risk for me, but it did blend out to 3x net. Plenty good, right? So, you’re right. At Series B, everyone thinks they’re a great picker, don’t they? This isn’t pre-seed. Everyone’s a great picker at Series B, right? So, does everyone have the same log-normal distribution across these deals? I don’t know.

Rory O’Driscoll

No, they probably don’t. I wouldn’t say everyone. I’d say 3 things: everyone thinks they’re a great picker, not everyone is a great picker, but the third sentence is you have to be a great picker to win, right?

Because if you look at it, you kind of articulated those 2 strategies. You can, again, be faced with this opportunity set of 547 Series Bs in 1 year. That’s a very clear data point, right? I think if you don’t pick, if you’re not a good picker, you will end up with more.

Remember, Jason’s a different set of buckets, but his buckets are more scary. He said 2/3 of all these deals are less than 2x. If you skew, instead of 66%, 75% or 80% in that less-than-2x bucket, your math doesn’t work. It’s not that hard to be that bad because, on average, it’s 66%. My point is that it requires a fair amount of discipline and picking to pull this off.

I don’t think the spray-and-pray strategy would work here. It might work at seed. I’m not as familiar with seed, but here the cost of spraying just gets too high. The picking doesn’t work. The only way it doesn’t—and I can see Harry doing his “I don’t agree” face—is that I think there are actually 3 approaches: there’s picking, there’s spraying, and then the third one, which someone big like Andreessen can do, is optioning.

5. The Role of Option Checks Venture Capital

David George

Right? You can spray if you have option value, if you’re just doing options. But if you spray and that’s the only way you make your money, the probability of being wrong is just too high.

So, that’s exactly what I was going to say, which is actually in reference to an amazing graph that was released a month or so ago about seed bets by multistage firms. Andreessen did 72 seed bets, compared to Sequoia at number 2 with 27.

Exactly to your point there, Rory, you said you can’t win without being a great picker. I’m not saying Andreessen aren’t great pickers. I’m not saying anything against them, but you can if you have 72 option bets.

Rory O’Driscoll

Agreed. There are 3 strategies: spray, picking, and optioning, right? To do the optioning strategy—and the beauty of when you can option is that you can afford to spray more. Optioning allows you to spray more, right?

No, I think it’s super clear at every stage. Again, it goes back to structure. Once upon a time, you thought seed was all about option value and anything beyond that wasn’t because we were dealing with gargantuan sums of money. It is now plausible that, for some people, As and Bs are partially options.

Really, if you’re going to deploy $200 million in the growth round, you don’t want to be totally slipshod at the A and the B, but you can think of it as more option value, right? That is absolutely a superpower that a wall of money gives you, right?

I can’t afford to do that because most of my money goes in on my initial round. At most, 50% or 60% comes in follow-ons. So, I can’t afford to be wrong on 2/3 of my money to be right on 1/3. That’s not going to make me a dollar, right? So, poor me, I’m stuck having to pick.

Does the ever-expanding size of the outcomes we’re seeing today favor a spray strategy? Before, $30 billion was an insane valuation for a company. Now, we’re kind of like, “Meh,” with Ramp, $10 billion with Miro, and we’re not blown away by it. The outcome sizes are getting so much bigger. Does that not favor a spray strategy?

All that you need to do if you’re early is get into the winners. Who cares—not who cares, but $500,000 or $2 million, it doesn’t matter. Spray the $500,000, because the only thing that matters is Miro and Ramp, and everyone is in yours.

Rory O’Driscoll

No, because what you’re doing is taking a plausible theory and extrapolating it to the point where it no longer holds true. You can spray as much as your bankroll will allow you. The more bankroll you have, the more you can build up option value.

Well, step back. You could be talking about 2 things, Harry, and you’ve got to break them apart. Are you saying simply spray because I’m not going to make my money on follow-ons, but it’s the seed argument? I have to be in the very best deal, and I have to cover wide versus concentrated to do that.

Mhm.

Rory O’Driscoll

Right. That’s 1 thread. The separate thread would be: how much easier is the spray constraint if I also have option value at the back end, right? That’s easier. There’s no doubt. The more option value—let’s agree on the following: the more option value you have at the back end because you have a $10 billion fund, the easier it is to spray because you can amortize the cost of the losses over the 1 winner, provided you get to stick $500 million in the winner. The rest is noise.

Mhm.

Rory O’Driscoll

But if you’re a seed fund and you’re David Tisch today—and he won’t mind me calling him out—he explicitly does, respectfully, spray and pray. He does 50-plus companies in a portfolio with low ownership, but he is in some of the biggest companies, including Ramp, consistently.

Again, going back to the Carta data, what I like about Carta is it’s actual data, not words. There were 547 Series Bs in 2018. I’m going to guess the graduation rate is probably about that, which implies 800 As. That probably implies 1,600 seeds.

So, even what you are pejoratively calling “spray and pray” is doing 50 deals out of 1,600. There’s still a huge element of picking involved in that. The only people on the planet who have a structural business where they can deemphasize picking because they can write option checks at scale is Y Combinator. They have a structural advantage in terms of their economics. They are the only people who’ve built a mass-production seed business.

But for everyone else, if you’re trying to pick even 50 and there are 1,600 places, right? We know that at the B, only 500 of them get to the B. There are 1,100 places to put that money. That doesn’t work out.

It’ll always be true that if you pick the single largest outlier in any vintage—if 1 of those deals was Figma, where the B made 100x, so probably the A made about 200x and the C probably 400x or 500x—I don’t know, right? Of that order of magnitude. The numbers are available.

Even if it’s 300x, if you did an index and did every deal, you made a 0.3x. If you did all 1,600 deals equally, it’s only a 0.2x return. You understand what I mean? If you put $1 into everything, your Figma check doesn’t, quote-unquote, return the fund.

If you did the whole industry, maybe OpenAI would. Maybe there is 1 deal so big that it would literally return the vintage, such that if you just bought the entire vintage, you’re good. That probably happens maybe once every decade or 2. Most of the time, even at the seed stage, if there’s not an element of picking to it, you can’t just say spraying is my point. There’s an element of picking up and down the stack.

It’s funny. I was more on the Jason headline: only 1 in 3 deals double investors’ money, per Carta deal. That’s sobering. It makes you remember how much of this business is disappointment.

You remember your deals. You remember your 10x’s and your 15x’s and your 20x’s, and you just forget. And I think, actually, the best investors—

David George

Just learn to become numb to it, Rory.

Rory O’Driscoll

I don’t. I get all sad. I’m a bit lame. I had a shitty outcome this morning, and I’m really upset by it.

Yeah, you get vested.

Rory O’Driscoll

Well, you know, I had the worst, which is when a company gets bought by another company and you get stock at some insanely high price.

David George

That’s not the worst. No, the worst—let me, dude. The worst is you’re on the board, you try to sell the company, it doesn’t close, you’re on the hook for shutdown costs, and you have to wire $300,000 or $400,000 just to pay severance costs, which you should do as a board member and you’re probably legally obliged to do, and then write that money off straight away because you waited too long. That’s the worst.

Wow. I thought my day was bad. That’s perspective for you, kids.

David George

That’s why I just do everything on a SAFE. I have no rights, no visibility, no financial statements, no understanding of anything. But if it goes south, I could just ghost them.

Are you serious, though, Jason?

David George

What?

About just doing everything on a SAFE?

David George

No, but the SAFE does have some comforts. I do like the post-money cap as a seed investor. I don’t have to worry about the pool or other things. And I like the fact that it’s a medium commitment.

It’s like I’m not the only guy. I don’t always love being the only guy on the cap table, the only director. The SAFE is like, “Listen, I’m only sort of committing, guys.” Good luck to you. If it goes great, I’ll invest some more. If it doesn’t, I’ll write cheery responses to your monthly updates.

But you didn’t ask me to get married and commit, so I’m not committed.

You think I'm joking, but the stuff Rory talks about is awful, right? This end-of-life stuff, or a decade of a struggling company where you're on the board. I'll take the SAFE over that—the unsafe SAFE—because the commitment is low. Founders should be founders: if you're going to raise on a SAFE, you can't expect too much.

Speaking of commitment and marriage, I'm really intrigued to hear your thoughts on this one, Jason, because you've said, “Oh, I encourage founders to sell. I encourage founders to sell when the offer comes in.” So, there's no regrets?

Synthesia, one of the hottest, fastest-growing AI companies, is now at $150M in ARR. It reportedly turned down a $3B offer from Adobe to be acquired, and it's going to be raising a new round at well north of $3B following this. Jason, if you were on the board, would you have told them to sell for $3B?

David George

I mean, yes. I'm not saying it's the right decision, but, one, he's been at it for a while. Synthesia is one of these stories, like Replit and Vercel, that blew up with AI, but it's a journey, right? There is competition, and it's been a while. I would tell him to take it unless you're 100% sure you're going to build a $20B public company, a $10B public company. And if you are, do it. I like to stress-test it, right? I'm not saying that's the right thing to do, but I want to be the guy that gives that advice.

$10B or $20B? Those are 2 different outcomes.

David George

Even $10B is not really worth it for a founder. It is worth it for the VCs. Accel gets another play, right? $3B to $10B for Accel. Here's the issue: $3B to $10B for Accel is a huge difference, right? So, let's say they own 15%. Out of a what, $450M fund? Who knows? Let's make it up, right? Instead of a 1x fund return, it could be a 2x or 3x. You make that bet as a VC all day long, right?

For the founder, let's say they own 10%. What the hell's the difference? There's no difference. We've had billionaires on ours. You've had billionaires, Harry. I don't think Jeff Lawson would be living a better life with half the money or twice the money. It's the same dude. There is a weird disconnect between $3B and $10B here for VCs versus founders, because 1x your fund to 3x—that's a lot more carry.

So, I would say with Accel and Index, you're absolutely right. They'll absolutely take the risk for another turn because they have the luxury of doing so. They don't need to return cash to LPs in the same way that maybe other managers do. If you are not a tier-one GP, though, you will want to provide DPI to your investors.

David George

I don't believe that. I don't see it. I've asked my LPs. I've asked a few others. If you're a strong manager, okay, if you have a track record, and their goal is not a 3x seed fund, but if they really want 5x or north from you, the game is you've got to keep playing another card.

No matter what they say, I've asked my LP. I guess it's a small set. “Do you want more money back?” I even asked my most conservative LP, like a university endowment that isn't that big, that is small, right? It has stress. I ask, “Do you want your money back?” I mean, with a gain, they're like, “No, we're just telling you we're really worried about it. We really want DPI, but we don't want it. We want you to play another card.”

I think if you tell them, “I seeded into Synthesia, and we've got a chance to 3x it,” they're going to tell you to go for it. The fund—we're going to trust the fund manager's discretion here. I just don't know that there's this massive DPI pressure if you have a hot hand. I think there's DPI pressure if you have a mediocre hand, in which case, so what? Because it's not enough DPI.

What's the point of returning 2% or 20% of your fund? Hooray, I got an exit today. I returned 2.1% of my fund. That's not the full job, friends.

The first question really should be—it's actually what Jason said—is, Victor, how do you feel about the business? I admire Synthesia enormously as a company. We tried to contact them; they didn't get back to me. They got the deal done. I'm so disappointed. We have another investment in the space. I love the space. I admire Synthesia a ton and many of the investors there.

I think the first question is: how do you feel about the business? And you threw out a statistic there. They said they're at $100M in ARR when they took the Adobe money in April. I checked, right? If they really have gone from $100M to $150M in less than 6 months, then I can make this conversation really quick. There's no way they should sell, because that thing's exploding, right? It's a great category, and it probably is. That probably sounds right to me.

The other reason I tell CEOs this: I would tell Victor to sell at $3B. It's okay if he says no. I would do it because I wanted to be the guy. Even if it wasn't in my direct interest as a gambler, I would tell him to sell. I want to be the guy that has that conversation with him.

There's another thing that's underdiscussed, and I've had this conversation twice with founders in the last 12 to 18 months in similar situations: are you really an IPO guy? There was one where I loved everything about this company. I loved the founders. I loved everything. There's nothing negative, and I told them to take the offer even though I didn't want them to as an investor.

I told them to take the offer because they said, “Are you sure? You're such a great set of founders, but I don't know if today I see you living the life these public companies see. I just don't see it—the way you have to do it, the constraints, the stress. I just don't think that's what you want to do. So I think you should take the deal, because I don't know if, if it's Synthesia or Synthesia Prime, you're going to get another offer like this before the IPO.”

That's the thing. You have to assume it's IPO or bust. And if you're not that guy, then you're going to bring in an outside CEO, or you're going to—it's just a mess. So that's the question I have: do you really want to run a public company for real?

That's an excellent question, Jason. Genuinely, that is really an excellent question, because you're right. At $3B, there's not going to be a ton of people coming to you offering $5B. You either have to cash out, or persevere and settle down and just become the Collison brothers and compound privately forever, to universal bewilderment, or you have to go public.

And the fact, by the way, that it is perceived by so many CEOs as a pain in the butt is, in my view, the thing about being public that has to change, because what you're doing is allowing this thing to get in the way of people's entrepreneurial ambitions. I think you're right that it's a pain, and I think it's a shame that it's a pain, and I hope it changes.

I mean, I was at this Dreamforce Marc Benioff dinner. It was great, and I saw maybe 10 public-company B2B companies that we know, that have all been on Harry's show, and Rory's invested in some of them. I got some hugs. Believe it or not, I don't give the hugs. I like to get them.

David George

Yeah, some crushing weight there, too. I saw it. The weight of being a public company is so heavy today, right? We could see it when we had Jeff Lawson on, and Cliff didn't have any of it at Canva. He didn't have any of the weight. It's just crushing.

I have Mike Cannon-Brookes on the show from Atlassian.

David George

Yeah, some crushing weight there, too. I saw it. The weight of being the unreasonable man—his co-founder calls him the unreasonable man—the guy who wills things into existence. You could see the weight of that on his face.

And when I look at Ramp and Mercury, these are epic companies, right? But I don't see that weight yet. I'm not saying it's not the hardest job in the world, don't get me wrong, right? But I don't see the deep crow's feet, and Aaron Levie going gray at 28 or whatever. I just don't see that level of anvil-level Wile E. Coyote weight on their heads yet.

That'll go to something about Anduril and capital efficiency. Rory, I just have to ask you: you said there, “Hey, we tried to get in contact with Synthesia. They didn't get in touch.” I've learned over time, if you want to be in Anduril, the next best thing is just to pay the higher price for Anduril, not to try and be in the next one.

You said you've got another player in the space. Why not just be in Synthesia? And do you not think this market takes the way most markets do in terms of composition, with the winner accruing the most value?

David George

I do. I think it's a broad market. We have a very different product.

It's kind of real-time interactivity versus asynchronous avatars. So there's a big difference in terms of the interactions. A company called Tavus, basically: instead of pre-recording an avatar saying something or having it be real-time, you actually have real-time interactivity with the avatar. Right? So, at a high level, it's the same conceptual trend.

But I agree, because we looked at a Synthesia wannabe at the time. I'm not going to mention the name, and we decided Synthesia was the one. I'm not—you're not in the business of doing modest number twos.

David George

Agreed.

So you're playing the broad trend in a different part, in a related but different market.

David George

I agree. You never want to do the number one. They wouldn't have me; I couldn't get into them, or they had just raised a round. You had to pay up for the later round. I made that mistake in 2013.

I won't mention the 2 companies. We had missed the rounds that were the typical-scale round for company A. We actually had a chance, with a lot of networking, to do a very late-stage round in company A, or we could do a more normal-scale round in company B in exactly the same market.

These were the 2 most head-to-head companies I've ever done. We did B, we got a 2X, and we didn't do A—we left a 15X on the table. Literally, every time we discuss this at the off-site, one of my partners—I love them dearly—the canonical example has become that decision.

I'm like, "Am I going to hear about this till I die, people?" Right. I know I got that one wrong, but they're right to mention it.

True. Exactly. You never want to do the way-behind, exact-same-category number two if they're way behind. It's just too hard.

6. Amazon's Struggles: How Do They Return to Greatness in AI

David George

Yes, I've made that mistake. I remember the week I got that learning. I can tell you exactly where I was when I took those 2 phone calls. Don't do that.

Okay. We can do Anduril capital efficiency. We can do Amazon layoffs. We can do the Amazon cloud wars or Oracle debt. You choose.

David George

I feel bad for Roomba, but you choose.

Oh, you want to do Roomba?

David George

Kind of. Kind of, because this whole talk is about how everything's up and to the right. Poor Roomba gets an offer from Amazon to buy them for $1.7 billion—

—and then gets blocked by antitrust.

David George

They raised $200 million of debt to finance the gap, and now it's all spent, and they're probably going to go bankrupt. This is why maybe, maybe, maybe you take the Synthesia offer.

I've owned maybe a half-dozen Roombas over the years. It's a tough, tough end to a founder journey, isn't it? Sorry, maybe that's not the story you wanted to pick. You go ahead.

David George

Do you sue the government in that case?

No. I think in the UK it's called Crown immunity. It's hard. I mean, I can't remember, because you can't sue the government. In the UK, you have Crown immunity, where literally the governments are protected from incredibly dumb acts because they're the government. I think over here we have more rights, but still—and I'm winging it here now—I can't remember: If the DOJ blocked something on antitrust grounds, can you take them to court? Those DOJ decisions have been overturned.

I can't remember: In this case, did they take it to court, or did they just decide to fold? If they decided to fold, then you probably can't sue, because you had a statutory remedy and you chose not to take it.

I mean, so again, zooming out for people, the story here is that iRobot had a deal to be sold to Amazon, and the FTC, led by Lina Khan, chose to block that deal because, as I think Benedict Evans said, of an incipient monopoly in the household vacuum-cleaner marketplace. Very tongue-in-cheek.

It was an absurd decision at the time, and it only got more absurd since, because, unfortunately, the poor company—which is a consumer hardware business with all the gross-margin profile that that requires—struggled on a standalone basis and now faces the risk of going bust. It's a horrible and unfair outcome for which the government, Lina Khan, and the FTC are entirely responsible, based on an outdated, stupid, and foolish paper.

There's just nothing but a belief—and a foolish belief—about how things work, beyond doubt. It's also another reminder that there's the whole "how antitrust killed the company" story. It's the edge of the no-politics rule, right, but not past it.

It's also a reminder that when you're going into the M&A process, sometimes the acquirer will pay a multiple that only sort of kind of makes sense—a revenue multiple that makes sense for them, but you couldn't get it otherwise. When you get one of these deals, it's tragic that it died, but you've got to take it. You've got to take it.

If you're Synthesia, you're looking at Wiz and Figma going, "God, do I want to put myself in that potential 18-month waiting period?"

Exactly. By which time, I'll be $400 million in revenue, being acquired for $3 billion. You think you're getting this great multiple—oh, they're paying me 30 times revenue—but the damn thing's going to close in 18 months, by which time you might be down to 10 times, and it's just not going to feel that much. I don't think Wiz has closed yet. I could be wrong. I think it was the crazy one that hasn't closed. It's crazy, right?

If you think about it, everyone's like, "Yeah, you got a 2.5X in 6 months." No, you didn't. You got a 2.5X in 2 years, right? This prolonged antitrust process really is kind of sand in the gears for a lot of these M&A decisions and, at the margin, probably pushes people to either A, push on, or B, in the case of the crazy deals, when the acquirer only wants the people, they do the Silicon Valley acqui-hire routine. But when you're buying the vacuum-cleaner company, you want the freaking vacuums, right?

7. Why IRR is a BS Metric and What Matters More

That duration period, though, is also why I think it's our responsibility as early-stage managers to be much more proactive in secondary markets, because we get cash back way sooner. The age-old thing of a 4X fund over 17 years is the same as a 2.5X fund over 10, and duration matters and time matters.

IRR is king. It's not the only king, because, obviously, back to Jason's comment—he was just sneering at a 4X, because, look, in this very show Jason was saying, "Oh, it's only a 30% IRR ramp from the $22 billion to $30 billion." From an IRR perspective, that looks amazing. So I don't think IRR matters. I actually think, for the record, the formulation of the optimization function is the maximization of the multiple subject to a constraint on a minimum IRR, just to be a total geek here.

Basically, you should know what your target IRR is. Let's just say it's 25%. You want to maximize the multiple, provided you don't dip below 25%. That's actually what you're trying to do, right? So, for example, a 30% IRR in 1 year isn't as good as a 25% IRR for 4 years. But if you hold on too long and that 25 starts dipping to 19, 18, 17, then you've gone to a different place. That is the rule, right?

In the end, you want the maximum amount of capital to invest, and you get that because you're held accountable at the investor level, on the IRR basis. At some point, they're looking at you, they're looking at the public markets, and they're saying, "Risk-adjusted, I need my 20%." So it is the constraint, because it's what prevents money from coming down the spigot to you, but you're actually trying to maximize your multiple.

So the shit is hitting the fan at Amazon before we do agree or disagree: 10%, the largest layoffs in history; 10% of white-collar employees, which is very sizable. They're falling behind in the cloud wars, from 50% of cloud revenues in 2018 to 38% today. Raymond James sees Amazon's AI cloud share falling to 7%. Then we had the outage—billions of dollars in damage. I mean, this was a bad fortnight.

You know what? Maybe it's not fair, but what I was thinking is, contrast this with Sergey Brin coming back to Google and everything else happening. Maybe this was a tough time for your founder to leave and go to Miami. Maybe this wasn't a very stable time to do a transition, right? I mean, Jeff Bezos had a long time as CEO, but maybe this wasn't the perfect age for stepping down. Just before AI hit was suboptimal for Amazon.

David George

Or maybe it was brilliant for Jeff, because you're implying that, had he stayed, all these bad things wouldn't have happened. It's plausible, just given his world's top 2 or 3 entrepreneurial achievements of the last 3 decades.

But it's just worth pointing out what the 2 big problems they have are. The first problem is that, in their retail business, they overinvested for COVID, and now they're trying to replace people with robotics because the technology is there. That's just something that had to be done. That's just more of the same.

In cloud, it's less that their core AWS business has folded up. It's that all the new compute—which is 10X and 20X larger in terms of demand for these customers—is AI-related compute. You neither built something compelling standalone, nor have you partnered—except, to be fair, a little bit with Anthropic—and you haven't found a way to get some of that compute, right? It's not clear.

But he had 4 years. Andy Jassy took over on July 5, 2021. Bezos checked out right at the peak of the last era, when products were frozen in time for a decade, when AWS was the same product for a year.

So were most of the companies we invested in—all 3 of us—in 2021. They were the same products as in 2015. It was a great time to go to Miami, because nothing was changing in tech. Stock prices and revenue were going up, but the products were the same.

So why wouldn't you retire? There's not going to be any change.

David George

Punched out at the top. No, he gets an A++ for marketing.

No, I think he gets an F.

David George

Oh, God.

If he sold his company, he'd get an A+.

David George

That's true.

But my last deal, Salesloft, was in December 2021, at $2.5 billion. That was the last deal of the era. Kyle and the team get an A+ for timing.

I want to congratulate Kyle. Exactly. This is not—

David George

This is not the same. This is punching out of 2021 and not punching back in, like Sergey.

Cynical comment here, but you're right. When you've got a couple hundred billion dollars, my guess is you're not maximizing money. You're maximizing psychic pain and joy. My guess is his psychic joy over the last 3 or 4 years, doing what he's been doing, has been significantly higher than the psychic pain that would have been involved in realizing, one, you've never done a big acquisition in your life; and, two, you've got to do a huge corporate deal in AI to matter.

All those people you hired in 2021, trying to do the right thing for core and expand, you've all got to lay them off. So I don't think it would—

David George

Jassy laying off half his company in 4 years was the right thing. I don't think you even care.

Well, he'll have that chance.

David George

But again—

I mean, Chegg had to bring poor Dan Rosensweig out of retirement to run Chegg after laying off 80% of the company. They had to bring him back. They've got to bring Bezos back, I think.

And to be fair, one thing we shouldn't do here—and you saw a lot of it with Google, too—is overcompensate. If you look at the two, there was a bunch of bad news in 1 day, and it is bad, right? There's a lot going on with the layoffs, but going back to your position in retail, it's broadly good. You're a little schlocky in terms of the shopping experience, but you have dominance because of your distribution, and you're doubling down on that.

You're reinvesting in robotics, you're cutting costs, and Amazon wins in its retail business because it can deliver shit faster than anyone else on the planet, pretty much anywhere. You're doubling down on that, so that's a win.

On the compute business, on the AWS business, your problem is that you're just not relevant in the new world. So knuckle down and figure that out. To be fair, of the 3 hyperscalers in the pre-AI world, Google was able to be relevant because it had its own model. Microsoft went and rented a model from OpenAI, and now the contract is nearly up. They did it to make a lot of capital gain, but they didn't actually, in my view, develop something compelling that they own from it. And you did nothing, so you lose.

They've got a set of problems on the AWS side. They have to get AI-relevant without, frankly, doing what I fear Oracle is doing, which is taking on a whole bunch of subpar economic transactions. They do have a good slug of Anthropic, not as much as Google, which I think owns 14%.

Google owns 14% of Anthropic.

David George

Google. Yes, Google wins. Again, we should be promoting all these corp dev guys. They're doing great.

Unbelievable. I think people forget that. Guys, we're going to do agree or disagree. I've got 3 statements. Tim, you love this. Come on, you can hide your excitement, Tim. It's too much.

Rory O’Driscoll

I hate it.

No. Oh, I know, when the previous hour and a half wasn't—

I did research. That's the difference between us.

Oh, dear. Very fair point. Can't disagree with this.

First one: I would rather own Brex at $13 billion than Ramp at $30 billion. Ramp is at $1 billion in revenue. Brex is at $700 million. They're both around break-even to moving toward profitability, and you've given me all the information except the only information I need, which is the growth rate.

Brex said 50%. They just said they're growing 50% now. I don't know what Ramp is. This is where my cognitive bias—

I'm going to go for Brex because it's lower.

I can't help myself. You think I'm kidding. I can't help myself, for real. If you gave me a great deal at $100 million—if you gave me Ramp at $100 million and Brex at $30 million or $40 million—I'm still going to do Brex, probably.

I could assign a homework assignment that would make this an interesting discussion. We have all the data points we need: the relative size, the valuation, and the growth rate of Brex, right at 50%. The correct question is: What growth rate should Ramp have such that you're indifferent between those 2 prices?

If Ramp is at 100, you probably want Ramp. If Ramp is at 50, you definitely want Brex. It's actually the fundamental question in all of venture: How much extra do you pay for how much extra growth?

If you could buy Brex at whatever it is—$700 million—growing 50% for $13 billion, and Ramp at $1 billion at $30 billion, there is an equilibrium growth rate that would make you indifferent between those 2 prices. It's actually a very fun exercise because it's the core problem you face over and over again in venture: How much extra do you pay for 70% growth over 60% or 50%?

You pay almost 2 to 2.5 times the revenue multiple. How much more growth would you want from that? I'm missing $700 million. I—

This is your job on the show, Tim. You've got to tell me.

Yeah, I know. It's hard to think and talk at the same time.

Keep going. You weren't relying on us for analysis, were you, Tim?

Yeah, I was. Keep going.

Come on.

Andreessen Horowitz deserves the prize for best-performing mega-platform of the last 12 months.

They're going to get the prize. No matter what I say, it doesn't matter. They're getting the $10 billion. That's prize enough for those guys. They'll be fine.

I've been impressed by them. I've invested with them across the board. One thing people often say about Andreessen Horowitz, with absolute respect to them, is that they have so many people, and that almost suggests that they can't all be good. They have so many people, so, whatever.

I have worked across the board with the different partners. They are as good as the partners at smaller firms with 3 to 5 partners, and they have been fantastic. I am a big bull on Andreessen Horowitz now, where I was not before. I agree they've been operationally excellent.

What they brought to the venture business from 2007 or 2008 onward was solving the problem from the founder back, with operational excellence. On that, they've delivered in spades. They figured out what the founder wants, and they've aggregated and delivered it.

I've always said the question was whether the return profile of that would be excellent. I've seen the numbers; they've been excellent. That was the question, and they pulled it off. I agree with you on the human-capital thing, and that's why I think the way they have it is smart.

Obviously, they just lost—I can never pronounce the German name—but a very impressive guy. I've heard him speak. He's going to do his own thing.

They have a constant churn, and I was thinking that even that's okay because they have a platform that transcends. They'll find other good people. They have the little pockets to play with, so they can give you your little information world, they can give Chris his crypto world, and everybody can have their thing.

It's just like investment banks, where you're the second co-head of North America, everyone's got a great title, and Marc and Ben sit at the top and life is good. It's a well-functioning scale machine. The take 15 years ago was that venture couldn't scale, and it's to be determined how far it can scale, but they've scaled it far more than anyone else would have thought 10 years ago. For that, they get a prize.

When we sit and look at Series A, we don't sit and get frightened by Index or Accel or the big European players. We sit and get frightened by Andreessen Horowitz coming into Europe. They're the ones who beat you.

Okay, good to know.

Well, look, I think you want to be in the top 2 choices for a founder at your stage. That's when you win in venture. It would be nice if you're in every stage and every deal, right?

I think Andreessen Horowitz has elements of that today: a top-2 choice in so many stages, from so many types of founders. Y Combinator has that in a sliver, as do others. If you're not, it's just a different game than when you're a top-2 choice. That's what you want.

I agree with you. There's only 2 people who can be in the top 2, right?

Final one: Of all the companies that could be public but are private late-stage companies, I would rather be a shareholder today in Anduril, the hottest, at $50 billion. Agree or disagree? And if you disagree, which would you rather own?

No. One, I'm not interested in the company personally. At this point in life, I just want to do things I'm interested in, to a fault. I'm not interested in building weapons and stuff like that. It's just not my vibe.

Two, I don't go to enough parties in San Francisco because if I did—and this sounds facetious, but you know what? I'm joking. I'm never really jokingly joking, as it does—there's no better brag at a party than, “Anduril, you did it right.”

It's the ultimate Founders Fund brag. It's just the ultimate brag. And if 1 of the psychic benefits of investing is bragging, right? At many levels, there are different types of bragging. There's bragging because I have a sense of worth in my life that I don't have; I'm bragging because I wrote a tiny check, but I'm validated in my soulless job. There are so many levels of validation, but I just don't go to enough of those parties since 2020.

So if I did, I'd want to have that one on my chips to talk about at parties, but I don't care. And I'm not into bombs and weapons, so I'm out on this one. [snorts]

The amount of self-knowledge embodied in those last 2 minutes from Tim is just stunning. I mean, the therapy is clearly working, dude. But [laughter]—

It's just called being happier.

I think it's actually called getting older and wiser. I love it.

Rory O’Driscoll

Maybe. Maybe.

Listen, guys, on that incredibly hearty note, thank you. This has been fantastic, and I've loved it.

All right, rock and roll. Thank you, Harry. Thank you, Rory.

Thank you.