Elon Musk vs Sam Altman｜Thinking Machines 的崩塌｜VC 能否撑过公开市场定价？

Team, I am excited to be back. We've got a lot of stuff to go through this week, and I wanted to start with a really optimistic view of public markets, which makes me question whether venture today as a model can still make money. When we look at Figma down to pre-IPO levels, Datadog now 20% down, even CoreWeave, and monday.com getting killed—we've talked before about PagerDuty consistently being down—we're just in the dumps with public markets. How do we analyze this, and do public-market multiples today make the venture model increasingly challenging?

Rory O’Driscoll

No, they don't. The public markets are actually almost the exact opposite. What those multiples say is that when companies go at a certain growth rate, they get much lower multiples, right? Slow-growth companies get low multiples; high-growth companies get absurdly high multiples. You could have also cited Palantir at 70 times forward sales for a 45% grower. Every venture capitalist is going to make a trillion dollars. [snorts]

What you're really seeing is the markets doing what they always do: they're sifting, and they're basically saying, “Low growth, we're going to value you very low”—and perhaps too low. We can come back to that in a second. That's what's really going on here. It's not, “Oh, tech is doomed.” What it is is a sifting and a sorting.

Things that look like they're going at a certain growth rate are getting thrown out, getting discarded at pretty low valuations, while things that are perceived as exciting and on trend are getting very high valuations. Venture is nothing if not a trend business. So I would argue that, for venture, it's pretty good. It just reinforces the dynamic of this business: always be in the hot stuff and you'll be fine. If you're in the trailing-edge stuff, you're toast.

Goodness. Here's how I think about it: just be more pragmatic. If Figma isn't good enough, what hope is there for the rest of us in software? I'm not saying that there aren't companies exploding—the ones we know, the ElevenLabses, the Replits, the Lovables, the Higgsfields, and others—but I feel like none of the unicorns are better than Figma. If Figma isn't good enough, it's not good enough.

It is not a great IPO. I'm not even sure it's a great public company. It's a great product, right? I'm not even sure it's a great public company. I look at my portfolio and think, what the hell am I going to say at board meetings this week, Rory? “Great job, guys. But have you seen Figma?” [laughter]

Rory O’Driscoll

But hang on. I'm just going to say something: boys, great job, lads. This is the problem with anchoring, right? Figma is a little bit down from its IPO, but then it's hugely down from where all retail priced it the day after the IPO. If you look at it versus the IPO, you're right: it's still down a little, but it's not nearly as catastrophic. It's still a $12 billion market-cap company. It's still growing.

It's still trading at 10 times forward sales, growing at 30%-plus. It's an awesomely good company.

That's just the value. In other words, 10×—if you invested at 8 or 6, though. What? Maybe not if you invested at unicorn valuations. It's not just—

Rory O’Driscoll

1. ClickHouse's $15BN Deal: Analysed

That's exactly right. Again, the message is really clear: when you pay up for high-growth companies and the growth slows even a little, and the belief goes out of the multiple, you're in for a long, hard haul. We've come back to this a few times before. You can be valued on free cash flow, but it's a long journey from the hope and the sizzle of a forward-revenue multiple and a high growth rate to the steady anchor of 12 times free cash flow. It's a long and tedious journey, and there's a long, flat period for the stock while that happens.

I feel like one analysis—it's likely Bill Gurley, after all—is that there are just windows when there's lots of liquidity and high prices, and then it's crap 80% of the rest of the time. Another version I've always thought is—and I don't want folks to take this the wrong way—in some ways, I feel like venture and tech is a bit of a scam.

What I mean by that is that our job is to convert very high revenue multiples into cash almost unnaturally through M&A and through public offerings, when they haven't earned it in free cash flow. Our job is to find companies worth 20, 50, 100, or 200 times revenue and magically convert that to cash. When it works, that's how we build 5× or higher funds, and that's how we make money. If we have to go to an EPS world, we're dead.

I feel like we're waiting for these moments, and it's even worse because when multiples are way down, like today, no matter what any Carta data says, there is no liquidity. It evaporates from the system.

Rory O’Driscoll

2. Can VC Survive With Public Market Prices Today

But you say that—I mean, you describe the system correctly, but you kind of imply it's called a scam. I don't think it is a scam. A more sophisticated analysis here says that everyone has long since internalized the following: in the end, the surviving tech companies at scale are astronomically good businesses.

Microsoft dominates the PC era, and going back a long way, it's an astronomical business. Therefore, you want to own that. You work your way backward, and that says, “At the point in time when I don't know which company is Microsoft, I can't wait until it's trading at 10 times EPS to buy it. So I've got to take a chance and, as you say, buy a basket of things that might be Microsoft, and price them long before they have EPS.” So you price them on forward sales, right?

The truth is, 4 out of 5 of them turn out not to be Microsoft. No one even remembers what Ballin Software does. If I put a gun to Harry's head, he couldn't tell me about Corel, VisiCalc, Lotus, and so on. All of those stocks traded high and then went down.

But it doesn't matter a damn because, in the end, as a whole, the system—the venture guys and the public markets—was correct: this is a big-ass trend, and the winner will be worth $4 trillion. If you write off half a trillion in guessing to get there, you still win. That's why it's very hard to fund an old-school SaaS company today, because people are just saying, “Look, you're great. You're going from 2 to 5. You're great. You're a profitable company.”

Again, this is the thing: people who rain down a little bit of contempt on VC, who kind of project a little contempt—“Oh, your little $100 million revenue thing doesn't matter”—can come across as a little callous. A fairer statement is this: your $100 million revenue SaaS company is an awesome entrepreneurial achievement. You are to be hugely congratulated. It's magnificent. It's just not something that we can properly finance, because we're just not going to make a public-market venture return here.

If I am a founder of Linear or RevenueCat, or any of this generation of companies—maybe 5 to 10 years old, in that broad range—and I'm at $50 to $75 million of revenue, and I'm not an AI-first company particularly, and maybe AI helps a little bit, but I'm at $50 to $75 million and growing maybe 75% to 125%, what do I do when I look at—

Rory O’Driscoll

Harry, you've been too tough. I think the truth is, if you're growing 75% to 100%, you're fine. Stop. This idea that you're not fine is wrong, because if you're going 100% at $50 million, you're going to be at $100 million. Then your growth decays to 80%, and you're going to be at $180 million. Then your growth decays to 40% or 50%, and you're at $300 million. You're going to get to scale.

So you've got your funny face on, for listeners who can't see it. Harry's got on his “I disagree” face.

It's not that I disagree—I'll tell you. Listen, I don't want to get lost in the weeds. I only disagree with you for 2 reasons, Rory. They're just structural. One is, as someone with a relatively modest amount of capital and a concentrated strategy, I haven't bought 20% of a startup in 7 years. I haven't done a round at $1 million in revenue, growing quickly in the teens, in a decade.

Now, if I could buy 20% of these companies for, in the high teens pre-money, like I used to do when Rory and I met, then I would be pretty zen about today.

Jason Lemkin

I’d be like, “Whatever the Lord brings: $500 million, $200 million, $3 billion. If I have to own 5% at a $50 million post-Demo Day valuation, the math works out in the aggregate. Garry Tan’s got it all proved.” No criticism, but it really ratchets up the pressure when the valuation’s 3 times higher and the ownership’s at 25%.

I think this is where the Figma thing gets stressful, right? I mean, you’re not quite as good as Figma. I’m going to end up owning 4%. You’re going to be worth $4 billion if everything goes well—$5 billion. Guys, I don’t know how. Teach me about crypto.

Rory O’Driscoll

I’m so sorry. I push back again. 75% to 100% for the biggest and best funds isn’t enough when you have an ElevenLabs, a Lovable, or an Airwallex. Let me answer your question, though.

Jason Lemkin

Yeah. Rory can challenge it. I agree with you. If that is all you have—and Rory will disagree—I’m very worried about you today, right? The question is your job. For some folks, you’re dead in the water. For other folks, it’s not too late. Your job is: how are you going to attach to AI trends?

If you look at RevenueCat, the whole vibe-coding-and-mobile thing, I think they tripled the number of developers on their platform in the last 3 to 4 months of the year. Their job is to do what WorkOS and others did, which is directly and linearly convert that to revenue. RevenueCat has massive tailwinds that it didn’t even have at the start of last year.

There are a lot of folks like that that have AI tailwinds. If you have none, you sure better figure them out right now, right? This is the answer. For example, I have one of my first investments. It’s been around for a decade, and it’s not even at $10 million yet after a decade. I’m the only investor. They’re cash-flow positive, and I love them. It’s a very specific use case with limited competition.

They finally added a way to do deep AI analysis on their data and reports, which for their niche was impossible until December. It blew up. They will more than double this year after a decade of 10% annual growth, with 20% being pretty good. If it doesn’t fully answer the Figma question, this is your job right now. Sit around your company and ask yourself: those kids at YC built an agent for your space. Why the fuck wasn’t that you? There’s no excuse.

We all use the same LLMs. There is no excuse for you not to have an agent as good as the new kids. There’s honestly no excuse.

Rory O’Driscoll

Agreed. Going back to your pushback, Harry, I think it’s in the framing of your question. Let me tell you precisely why I think you’re wrong. You said, if you were the CEO of X, Y, or Z and you’re growing at 70%, then you kind of did the throwing-your-hands-up-in-the-air thing. I think you’re wrong for them, right?

What is true is this: if you’re at $50 million, growing 50% or 70%, you may struggle to get venture capital because you’re right—everyone’s correctly focused on the things that have the embedded upside of potentially exploding to hugeness. You’re probably at a stage now where that kind of mega-growth is not going to reaccelerate from there. So the probability of a mid-stage SaaS company exploding into something amazing is a rounding error: zero. As an investor, that’s not attractive.

But as the person who owns that asset—and maybe you own 20% of it—you can’t just say, “I wish I’d done something different with my life, and now I wish I was an AI company.” You’ve got to play the hand you’re dealt, and Jason’s exactly right. The first thing you do is say to yourself, “I may not be able to raise much more venture capital on attractive terms. Run my business accordingly.”

I am now just like—I thought we were going to talk about the public SaaS stocks here, like you hinted in your questions, but you took us offline. Live in the world you now find yourself in, where capital is no longer free for your sector. If you describe that negatively, you can say it’s because venture capitalists are fashion chasers. If you describe it in a logical way, you can say it’s because the kind of extraordinary growth has vanished from SaaS and has now reappeared in AI land.

Run your business so you don’t need capital. Then, as Jason said, look at what these next-generation people are doing and find a way to attach it to your business. If you grow at 50%, then 40%, then 30%, and you get to $200 million of revenue and sell at 5 times, that’s a $1 billion. If you have 20% of that, you have $200 million. It makes you one of probably 10 million—I’ve done this math—about in the top 5,000 richest people in the world. It’s okay, but it’s a grind.

And the reason it’s a grind—and that last sentence, because it’s a grind, explains exactly why venture guys aren’t investing—is that grind is not in our MO.

3. The Implosion of Thinking Machines

We’ll come back to new rounds. You mentioned the grind isn’t in our MO. When the grind isn’t in your MO, you move on. Yes. Have you ever seen the Home Alone movie cover, which is Macaulay Culkin screaming at the bad guys?

I feel like someone needs to do that for poor Mira Murati because I think she’s the only one left in Thinking Machines after everyone else left this week. We’ve seen the implosion of Thinking Machines, with 2 co-founders leaving. Barret Zoph is off. How did we analyze this news, and what’s left of the team?

Jason Lemkin

It happens with seed rounds.

That’s a great answer.

Rory O’Driscoll

It’s just part of the risk.

That is a great answer. I was going to come up with a whole bunch of other things, but Jason is exactly right. You have to remember, despite the big deal, Jason, that’s a great answer. This is a seed round, and I know why Y Combinator does this thing: the number-one cause of failure at the seed stage is founder incompatibility. This is just seed rounds with extra commas, and they discover after a year that they don’t want to be doing this and they want to go back to the big company.

Jason, you’re exactly right. That’s the big insight. It’s a seed round. Just treat it like a seed round.

Jason Lemkin

I’ll tell you one niche thing I never understood about this company. On paper, I get it. likely Andreessen and Sequoia put in $2 billion. You’re basically taking—Mira is taking an OpenAI co-founder and a whole bunch of the team, and it’s just going to be Anthropic again, right? They’re going to recruit some of the best, and they’re going to do that. In theory, you should do that bet in 15 minutes if you have the capital, right? Just take a bunch of the best guys and go do a modern Anthropic. It worked at Anthropic, right? Good God, it worked at Anthropic.

The part I never got—here’s the weird part, and maybe this has nothing to do with the tensions—is that her background is not technical. This is the part I never got: a degree in arts from Colby College and an engineering product manager at Tesla. I’m not saying she doesn’t have 50 IQ points on me. I think she does, but I don’t know how you run a lab if you’re not Ilya in the early days of OpenAI, or Greg at least.

I don’t know how you get the respect of the team if you’re not on their level as a researcher. It’s a challenge for a lot of other B2B companies. How do you recruit this S-tier AI talent if your team isn’t at that level? Who the hell would want to go work there? Compensation aside, right? Why do folks leave OpenAI after a year and leave $5 million or $10 million? They’re just so smart. They just want to work on smart problems.

I felt this one was unstable. People loved her, I think, but I’m super skeptical of technical companies where the CEO isn’t one of the greatest technical visionaries in the industry. I’m just—I’ll never do that investment again. I’ll never do an investment again where it’s not like Databricks, with the founder as CEO. I just won’t do it.

What happens in this situation?

Rory O’Driscoll

There will be some version of the following clause that says, if more than X team members leave, then you have the ability to call for redemption and basically wind the company down and say, “You spent 20% of the money.” If you can’t get an M&A outcome where you get 1×, then you can wind the company up.

You give them $2 billion, they spend $200 million, you take another $200 million to bribe everyone to go along with it, and you get $1.6 billion back. You say to yourself, “That was a risk that didn’t work. I only lost 20 cents on the dollar. I can recycle that money into the next OpenAI round. Everything’s fine.”

That would be a far better outcome than commencing the long and bloody march of trying to fix this thing and trying to hire new people. Jason’s right—it’s such an insight. It’s a seed deal that went wrong. What do you do when the core premise you invested in turns out not to be true and you’ve spent 10% of the total money?

If you’re smart and you don’t have another compelling idea, you give the money back. You earn the respect. Everyone respects you. “Okay, I got it. You moved on.” You call bullshit.

From an investor perspective, I wonder whether they’re sitting there thinking, “It would be totally fine to take a 20-cent haircut and be done.” Do you think they’ll do that?

Rory O’Driscoll

It probably felt safer when they made the investment that, worst case, Meta wanted to buy them. Worst case, we exit for $5 billion, $6 billion, or $10 billion to acquire the team. We don’t quite make as much as you might think on paper because of how the deal is structured with employees and retention. But worst case, we make some return on this deal.

Jason Lemkin

So, it’s not as crazy as we might think from the outside. Because of the quality of the team, there’s a high chance we get our $2 billion back. It seems like lower odds today. [laughter] The way I described it, if I were on the board, we’re not going to get our 2x from the M&A outcome. But if we get our 8x to a rounding error, it doesn’t matter.

Especially if you can quickly recycle the capital, because this happens so quickly. It’s all in the current fund cycle. Just so listeners know, if you’re in the investment period for a venture fund, you wire $1 billion into investment A. Instead of waiting 10 years and getting 0.8 back, if you get 0.8 back in 12 months, you can reinvest that 0.8 very quickly and just move on. It’s almost like it didn’t happen.

So much better.

Jason Lemkin

So much better. It’s like—

Forget about IRR. You can put that money to work. You can put that money right out the door.

Jason Lemkin

Literally, ask yourself—

You might not even care. The loss is a bummer, but if you’re trying to get 10x outcomes, you almost don’t care. Just give me the money back and let me invest it tomorrow.

Jason Lemkin

Watch this. If you put out a billion and got the 0.8 back 6 months ago, then put it in the last round at Anthropic, that 0.8 is now 1.6. So, on your initial billion, you’re actually ahead 60%, despite having lost 20%. The whole trick—and remember, this is why, at heart, venture is a capital-allocation business—is that you are trying to stuff your money into the place where it will grow the fastest. Once they start going down, venture guys, it’s brutal, but the rational thing to do is reallocate to success and away from failure.

I think we underestimate the challenges and the wars for AI talent. We talk about it being compensation-based and brand-based, but it’s mission-based.

Jason Lemkin

Yeah.

Yeah. Well, it’s even more. The best researchers in AI—the best, the ones you need to win—they only want to work on what they want to work on.

Jason Lemkin

Yes.

And they will leave a lot of money behind. That’s why we’ve all moved to no vesting anymore at OpenAI, because we want the folks who are like, “You know what? I just don’t love my job at Thinking Machines. Come to OpenAI. You will lose nothing.” And folks want the smartest people in AI, in applied math, and elsewhere. They only want to work on the intellectual challenges they want to work on.

Jason Lemkin

And if you can’t deliver that and the right boss, there’s so much portability that they will just leave. That is an environment almost impossible for 99% of software companies to create. Maybe Thinking Machines does—I don’t know. Maybe it doesn’t have that environment anymore. Maybe it just can’t provide that environment any longer, because you have to provide the job the researchers want. You don’t get to decide what they do.

I think you’re right. And, you know, we had it on the thing. Let’s talk now about Yann LeCun raising money at $3 billion. That’s an example of—I personally look at that and say, wow, that’s a scary bet, because somewhere between 2017 and 2022, we collectively—the industry—unlocked LLMs, and we are mining that vein. It took 5 years to unlock it, and we’re mining it now.

This guy is making the big call that says, “Nope, we’re going to find a different vein. That’s not the right vein to go down. We’re going to do something new.” From a probability-of-success perspective, if you’re a money person, you kind of go, “Hmm, maybe, but that’s a lot of risk.” But you’re right: from a researcher perspective, they go, “Wow, that’s way more interesting than grinding out the 5th marginally better LLM. I want to do that.”

He’s been able to attract talent. He’s been able to attract capital. I think you’re exactly right, Jason: people want to do the things they want to do. And to the downside-protection point, raising $500 million at $3 billion, you’re basically looking at that and going, “Do I think I’m going to recoup my $500 million on a talent acquisition? If so, I’m just riding upside as an investor.” It actually makes sense.

Jason Lemkin

Again, I would say the 2 things I would—what I’d love to know, and I don’t—is what terms have been written in on behalf of the money to protect against what we just talked about, the if-the-talent-bails thing. I wonder: are there evolving standards on that around redemption or around rescission events?

But you’re right. If you say to yourself, you’ve got 1 of the 2 or 3 godfathers of AI, plus whatever team he assembles—I believe, I think we talked about it last week, he’s chairman and someone else is building it—you’re making the same bet.

You seem to have a monopoly on talent outside of the Valley, which is quite powerful. He does have a monopoly on AI talent in France in a way that—

Jason Lemkin

Well, I’m not sure it’s a monopoly, but as we were discussing before you came on, one of the interesting things about the current political environment is the extent to which non-U.S.-sovereignty-based AI plays probably have had a vindication in the last few months. I don’t know how that factors in.

Yeah, I mean, as I said, as a capital provider, you look at it and go, it’s a much higher-risk bet than, “We know how to build LLMs. Let’s just build a slightly better one.” From a risk perspective, you go, “Wow.” But to Jason’s point—I mean, we kind of got the conversation off a little, but I think Jason’s point is the right one.

The big attraction you have is that you’re selling to the researchers, who are the people you need to appeal to, the chance to do something wildly new. If you can figure out the 2026 equivalent of the Transformer paper, and if you can figure out the equivalent of “Attention Is All You Need” and publish that, that’s what you’re selling. If I’m a researcher, that’s what motivates me.

I’ll just struggle along on my $5 million a year, whatever I’m getting from OpenAI, but maybe I can change the world here. That’s the logic. You’ve got to play to that.

You’re lucky if you’re not Thinking Machines or OpenAI and you have 1 of these people at your startup. You’re lucky if you have 1. It probably has to be your CTO, because she or he probably has to do it themselves. They probably have to be your chief AI officer and researcher, because unless they’re incredibly charismatic, your company selling vinyl mugs online or whatever—it just, none of these guys are going to join you.

Even if you could afford $5 million a year, $10 million, or $20 million, they just don’t want the job. They just don’t want the job.

4. Elon Musk vs. OpenAI: The Legal Battle

Jason Lemkin

No, it’s a very efficient market for talent. I mean, you know, the venture guys—it’s funny, it’s kind of the old capital-versus-labor thing. The venture guys have tried and succeeded in making it a fairly efficient market for capital, and then were shocked to discover that labor makes it a very efficient market for labor in response. And, you know, go team. Carl Marx would be kind of glad in a weird kind of way. This is what successful labor empowerment looks like.

We’ve spoken about the breakup of teams. We touched on OpenAI a couple of times there with Mira Murati. I have to jump to the ultimate breakup of breakups, being Sam versus Elon going to trial. How do we see this playing out? Who wins? What do they actually win consequentially? What does this look like?

Jason Lemkin

Look, it’s going to be—first of all, let’s be honest—this is going to be the gift that keeps on giving. If you’re the kind of person who slows down at a traffic accident—in other words, if you’re like 90% of humanity—you’re going to be slowing down every time the depositions come out. It’s going to be great.

So, let’s start with an utterly different, idealistic comment as we get into this whole mess, and we will for a few minutes. It’s worth pointing out that they did all start by making a charitable donation for something they passionately believed in, where they weren’t trying to make money. Elon chipped in $30 million-plus, Sam Altman chipped in $10 million, and Reid Hoffman chipped in.

They genuinely believed—actually, we’ll come back to that statement, because some might say it’s a lie—but it appears that everyone genuinely believed at the start that they were doing something for the good of the world. They were doing something for charity, and they were trying to understand what AI could do and head off existential dangers at the pass.

As is so often the case in life, I think everyone’s intent was pure when they forked over real money to try and do something. I want to start with that: thank you, everyone, for trying really hard to save humanity. And then the old rule applies: no good deed goes unpunished, right?

So, the whole thing is now a mess. Just to give the zoom-out comments here, it was a not-for-profit company. At some point, it became obvious to the management team around 2017 that the costs to build what OpenAI was to become were such that you couldn’t keep going as a nonprofit. You had to become a for-profit entity.

During the period when that started to happen, Elon had his demands on how he wanted that to happen. Sam and Greg Brockman had their perspective on how they wanted it to happen. They ended up, as it were, breaking up. Elon resigned from the organization.

Fast-forward past the drama of 2022: in late 2025, the conversion to a for-profit finally took place, and it was approved by California and Delaware. Now OpenAI is a for-profit corporation, and the largest individual shareholder is the foundation. The argument that Sam and Greg would make is, “Hey, you invested money in a charity, and that charity now owns 30%-plus of one of the most profitable, one of the largest companies on the planet.”

Rory O’Driscoll

So, you kind of got what you paid for. That’s the argument they’re making, right? And Elon’s argument is, therefore, you’re entitled to nothing. You made a $30 million donation. You created a $300 billion—well, probably $150 billion—foundation. Congratulations. You’ve helped humanity. You got what you paid for.

And Elon’s claim, which is going to be tricky to prove but is going to be messy, is not just, “I don’t like that,” but, “All along, you guys were planning to cheat me and planning to make it a for-profit company.” Therefore, his ask is not just, “Hey, I want my $30 million back.” That’s chump change in the back of his couch, right? His argument, which is very aggressive, is that you guys planned this all along, and therefore my damages claim is not just, “Give me my $30 million back,” and it’s not, “Stop the conversion to a for-profit,” because that’s already happened. That can’t be stopped now.

It’s, “You guys took my $30 million on false pretenses. Therefore, I’m owed roughly what $30 million would own out of that company now,” which is $100 billion of value. He’s basically saying, “If we’re going to go for-profit here, guys, and you were lying to me all along, then I want my share of that now.”

So you’re all going to have to take dilution such that I get my $100 billion. The damages claim is $70 billion to $130 billion, which would come in the form of extra shares of OpenAI to Elon, and everyone else would have to take that dilution. That’s the ask. That’s what’s going on here. In the end, it’s an economic argument on the basis of fraudulent intent from day 1.

Is Elon doing this to slow them down? I don’t think he’s doing this for the $70 billion to $100 billion.

Rory O’Driscoll

Yeah, I think he’s doing it because he can and it’s fun. Look, billionaires are going to billionaire, and near-trillionaires are going to near-trillionaire, right? He feels shafted. It’s a win-win. He feels shafted because he does feel that he was mistreated in this process, and you’re right, it’s good for likely Grok if it slows it down. He might win $100 billion, and there’s not a ton of downside other than a bunch of legal fees because they’ve got no claim on you, right?

So it’s an asymmetric win-win situation for him, provided you have the couple hundred million dollars of legal fees it’s going to consume here, because we are litigating over $100 billion. The other thing that happens—and he’s lived through this on the Twitter litigation—is that deposition and discovery are a sucky process, because you write down stuff in your email or your diary, and then it suddenly comes out, and it’s always embarrassing, right?

The truth is this: If I spewed out your last 10 years of emails, I’d find some embarrassing stuff, right? It happened to Elon in the Twitter litigation. Some of those texts just looked juvenile. You think, “Oh, wow, you’re the richest man in the world. You sound like an idiot. That sucks.”

In this case, if you look at what they’ve already got on the record, they’ve got poor Greg Brockman, who kept the diary in 2017. He writes in his diary, as one does, “What does it take to get to a billion dollars?” And now that’s come out, and now they’re going some version of, “You were cheating me all along. You wanted to do a for-profit.” I mean, can you imagine, Harry, having to get your diary from 8 years ago?

And then you had Ilya, who had to do a 10-hour deposition on the 2023 CEO drama, which is not really relevant to this because the alleged fraud happened in 2017 or 2018. But again, it’s just a fun chance to get all the mess out there. So they deposed Ilya for 10 hours, and we finally got to hear what he thought about the great fiasco. It’s kind of embarrassing for everyone.

Everyone looks like an idiot in 2023. The board looks stupid. Ilya looks a bit naive because he relied on Mira Murati. She looks a little naive. It just looks like amateur hour everywhere. So if you’re Elon and you’re like, “I can torture these folks and make them look stupid,” maybe it postpones things, if there’s any kind of credible case.

And remember, a judge was asked, because OpenAI did go for summary dismissal, which is what you do. You say, “Hey, there’s no case to answer here. Just dismiss the case”—the charges, I should say; they’re not charges. The judge said, “No, there is a case to answer here. I’m not saying it’s right. I’m not saying it’s wrong, but there’s a credible discussion here.”

So now they’ve got to go to trial, right? I don’t know how that impacts fundraising, but for the next 1 or 2 years, there’s the potential that every financing of OpenAI now has—you might have to take 10%, 15%, or 20% dilution if they lose a jury trial. The intent of the other founders, Sam Altman and Greg Brockman, wasn’t, on day 1, to convert. I think Elon’s case is built on a slender conspiracy thread, but it’ll sound compelling to a jury when you also have all this dirt about, you know, like—who the hell knows.

Rory, what happens? And Jason, what happens?

Rory O’Driscoll

It drags on a long time, and it gets in the way.

And who wins?

Rory O’Driscoll

Elon wins no matter what.

Yes, that’s the right answer.

Jason Lemkin

This is a jury trial. Again, we’re not experts, but a jury trial is unpredictable, and there are bad facts on both sides. This is The Social Network 2. I know they’re making another sequel at Meta. This should be the next The Social Network.

There are bad facts on both sides. If Elon was not the richest man in the world, he would settle for $30 billion—or $5 billion, whatever the Winklevii got. He would do the same deal: “Give me 5%.” They would settle on the eve of trial for 5% to Elon. They would say, “We’re just doing it to move beyond.” They’d give him 5% or 4%, and if it were about money, he’d move on, just like the Winklevii got their 4% or 5% of Meta, Facebook. That’s what would happen if it was about money.

The fun thing about the sequel to The Social Network is they ain’t going to settle for that. No, he’s going to go to trial, and he’s going to win. Elon has bad facts, but this man has already been bad-facted the last couple of years in public.

His worst fact—forget about that—is that Oakland is very liberal. Elon’s worst fact was that Trump hated him for about 6 months. He fixed that issue. Now he’s back in the inner circle. Get a couple of MAGA folks, Republicans, on the jury. We don’t know.

Sam Altman was fired by the OpenAI board for reasons not fully disclosed. I have one strong theory. Greg Brockman—and do you know what I think really happened with Greg that maybe people missed? This is my theory.

The guy really partially regretted leaving Stripe as the CTO and as the 4th employee. He leaves, and Sam recruits him from Stripe when it’s worth $3 billion and says, “Don’t worry. We’ll make it up for you at our nonprofit. Come do this thing. We’re going to change the world,” to Greg. It’s very exciting, and Greg really doesn’t want to work on the Stripe API anymore. He’s already gotten payments to work. It’s kind of boring, right?

And he leaves in 2015, Rory, when the world seemed very flat and simple, right? He leaves, and he turns around and he’s like, “My God, I left billions behind.” How would you feel as a human being who didn’t make the money Sam had already made? Sam had already become—he wasn’t a billionaire then, but he was on the path, right? He’d raised hundreds of millions at his own venture fund to invest in YC startups. He personally owned 2% of Stripe.

How would 99% of humans feel when you’re like, “Fuck, if I just stayed at Stripe and just played Mind Sweeper, I could be worth $10 billion”? And I think that haunted him. I think he’s haunted by it, and that’s where these journal entries come from. It’s not him being douchey and saying, “Man, I want to…” It’s him being haunted by Sam getting him to leave Stripe at $3 billion.

Haunted by it. We’ve all—I’ve been haunted by some mistakes I’ve made, too. I think he’s haunted by it, and I think that’s going to create a whole bunch of bad facts the deeper we dig, that Greg was haunted by leaving the money at Stripe.

First of all, interesting. And, yeah, I’d be haunted too. I’m not going to speculate whether I’d be haunted if I left a couple billion behind.

Jason Lemkin

Yeah. No, one of the things that’s also sucky about this when you’re in litigation is suddenly everyone, including Rory O’Driscoll, who I think has never laid eyes on you, has a fucking opinion about you. That’s the really sucky thing.

Rory O’Driscoll

I’m not sure I agree with you on Elon in terms of the jury. I want to come back to that in a second. But I do agree on one thing: He is past pain when it comes to public shame and obloquy. Where we are, there’s nothing left to do for him, man.

Basically, everyone else is going to have to get down in the muck, and he’s already so far in the muck that, on various different parts of the last 3 years, 40% of the country has hated him, and it’s been a different 40% each time. So there’s nothing left for him to do.

Which actually gets to the question: How do you think it ends? I think, Jason, you’re right: If it were normal people, they’d settle. If it goes to a jury, I think there are 2 comments I will make from a facts perspective.

I think, in the end, Elon has to prove his case, and it’s a high bar to prove. You’ve got to prove that when they were raising the money charitably, they were all along intending to swipe it and build a for-profit. I think that’s very hard to do, because Altman’s money went in as a not-for-profit, too. So I think Elon’s wrong on the core assertion.

Jason Lemkin

I don't think this was a cunning device to cheat everyone, because if they wanted to do that, they could have just done what Anthropic did, which is the totally sensible interim stage of a public benefit corporation, and saved all this freaking drama. So, stripping aside all the bad facts, I think the core assertion that Elon is making—which is, they misled me into giving them $30 million for this thing that was never going to be a charity, and therefore I get my money back—I think that's wrong.

Right. But to your point, you still have to prove it to a jury. Now, the other point is, remember, he has to prevail. So, as long as there's 1 person on the jury who hates him more than he hates Sam, he ain't going to win. Actually, in a civil trial, it might be 10 out of 12.

But I think in the end, it's going to be hard. First of all, it's going to be a very unattractive cast of characters in an Oakland courtroom, if it is in an Oakland courtroom. You get Sam to come in and say, “Hey, talk about how you said this is going to make everyone unemployed.” Then you get Elon to come in. They're going to hate the whole being-on-the-stand thing, which everyone's going to have to go through, right?

And I think the jury will get, if they get into that room, they're going to go, “Wow, I don't like any of these people. Why should I?” The problem is this: for Elon to win, he has to get them to all vote for him. A hung jury is fine for the prevailing wisdom, but unless Elon prevails, he doesn't get it.

So, I think way down the line, 2 or 3 years from now, at the end of a long and bloody trial, the probability is he doesn't win the case. But Jason, you're right: he's already won if what he wants to do is get psychic revenge. This is going to be the best. This is like Peter Thiel's spend on Gawker, right? Sometimes a billionaire just wants to spend a couple hundred million bucks—in this case—to grind the other guy down and make him sorry.

I think he wants every bad fact about Sam to come out. Every bad fact in this trial. We haven't even heard why the board fired him, right? If someone subpoenaed my diary for the last 10 years, and if you keep notes on days when you're feeling like shit, you're like, “That's not great.”

That's the problem with litigation. That's the problem with convoluted structures. You end up in litigation. That's why keeping it simple was step 1. And once you didn't keep it simple, trying to keep everyone in the tent—when you don't do that, you end up being sued by the richest person on the planet who's angry and mad at you. It's a tough place to be.

If you're Sam, do you not go, “Hey, I'd rather have the dilution, not go through 3 years of distractions, and actually get him off my case? Bring the enemy inside. I don't have stock anyway.”

Jason Lemkin

Don't settle at $5; settle at $100, too.

I don't have stock anyway.

Jason Lemkin

This is true.

Well, sort of. I think that was dishonest.

Jason Lemkin

Yeah. Let's not even worry about that. But you're right. Because the truth is this.

Actually, the big question is, if this is getting in the way of the next financing and it's being seen as a credible risk, then would you be better off, even if it's extortion, to cough up? Which, again, is why Elon is in an asymmetric win-win situation and OpenAI is not. The only way it doesn't matter is if every investor—and I don't know if every investor looks at this and says, “I'm not worried about the risk. In the end, Elon will lose, so therefore we can ignore it.”

Jason Lemkin

Wasn't that what they said—that our exposure is capped at Elon's donation of $30-some-odd million? That was the public announcement, right? Or pseudo-publicly leaked, right?

And that's why his aggressive claim is, on the facts, very contestable. I mean, it's a real reach. The claim is a reach to say it's not just my $30 million back; basically, what Elon is saying is, “It was a for-profit all along. You guys just didn't tell me.” And if it was a for-profit all along, and I put in—if I was the $30 million seed in what was a for-profit company—then I want what a $30 million seed would get.

It's a stretchy claim. You're right, Jason. OpenAI are going to say there's simply no way, from a point-of-law perspective, we're going to concede that. And even if we lose a jury trial, we'll go in on appeal, right? They're just going to litigate it the whole way down.

Jason Lemkin

But you're right, that's fine. And this actually ties, funnily enough, to the ads coming and everything else. What I don't know—the key question that I don't know—is how will the investors look at it? If you're writing a check right now, do you think, “Am I paying $600 billion pre or $800 billion pre? Do I need to say it could be $1 trillion because it could be 20% dilution?” I don't know, right? Or do you say—

You said last week there was no existential risk to OpenAI. I said there might be some structural economic risk. You said—

Jason Lemkin

But this isn't existential. To Harry's point, in the end you can fold, give the man what he wants, which—

If you can't—if it's hard, I don't think this will make it any harder for them to raise capital, because greed trumps fear. But if it did, it could create some risk, right? If it made it harder to raise capital—

Jason Lemkin

I think you have to fold, because the winner in this case—I know it's ridiculous to say—but the winner in this case, the guy laughing with popcorn, is Dario.

Yes. Going great.

Jason Lemkin

But you can't fold. I don't think he'll let them fold. I could be wrong, but there's no chance that there wasn't some lawyer discussion talking about what it would take to resolve this, right? In a quiet room off the record, there's just no way that OpenAI doesn't have a fiduciary obligation to find out what it would take to settle it, right? Arguably, they have to.

So, Elon didn't want to take it. He just wants his day in court.

And that's why you have to settle if you're Sam, because—

Jason Lemkin

You can't settle if he doesn't want to settle.

No, I mean, to be fair, you don't have to. As I think about it as an investor, what would you say? You'd say, “Here, let's just say this is a round at $700 billion pre. There is a probability of an additional 15% dilution, but it's not 100%; maybe it's only 10%.” Right?

So, really risk-adjusted, you add 1.5%, but you don't assume you're assigning a probability to a capped, large, and uncertain event, which is there is some chance that I take 15% or 20% extra dilution, but it's not 100%. So it doesn't drag the financing down to 0%. It merely means there's an asterisk risk on it.

Jason Lemkin

And to be fair, anyone who's financed OpenAI today, if you look—the funny thing about OpenAI is, even though this looks very risky, I'm going to make a weird statement: OpenAI, from a structure perspective, is less risky now than it's ever been.

If you look at all the risk, the first money went in and it was, “Hey, it's for-profit and nonprofit.” Then you had that disclosure: “Hey, we're never going to make money.” So what can you do? In the last year, they've been able to convert to a for-profit company.

A huge step took place last October-November of 2025 when OpenAI—I mean, the real danger would have been if they hadn't been able to convert to the for-profit company, because at that point the whole thing was at risk. When they got that done, they took a big step-function decrement down in risk.

So, even though this feels very risky, it is less risky than the risk under which they were able to raise $200 billion, plus or minus. It'll be fine. It's just a risk. It's going to be a monstrous pain, and it's going to be popcorn time for everyone else.

5. Can OpenAI Win Ads?

I think you're being way too nice. At a time for OpenAI when you have Gemini killing you on consumer, Anthropic killing you on enterprise, you're just lumping everything in together, you know—

Jason Lemkin

Because you have to. This is—

Yeah, but I mean, look—

Jason Lemkin

The litigation—I can't isolate things like that. But actually, funny enough, you have to, in the sense that whenever you have that kind of litigation, I've been at companies that occasionally have it. What you have to do is say, “This is the one thing.”

The really dangerous thing about litigation is if it kind of subsumes the whole company, because the thing about litigation is it gets your blood up and you start trying to win, and you just get emotionally vested in it. If they're smart, they'll have a great GC who says, “Other than depositions, none of you talk about this ever again. We will deal with it and our $100 million worth of lawyers,” right? Because you can't let that get in the way.

So, let's assume you do that. And now you're still running a wildly successful business with a whole bunch of competition. You're right: you've got to deal with Gemini. You've got to deal with Anthropic. You've got to get ads out the door. You've got to start having a convergence plan on profitability, which is, I think, why the ads are coming out.

Yeah, you still have lots to do, but with the exception of the litigation from the world's richest man, nothing's changed from 2 weeks ago. Is ads coming at just the wrong time, as Gemini is killing you on consumer and producing better and better models, and Google is really feeling tailwinds to come out with something that does deprecate the product, even in a small way? Is it coming at just the wrong time?

Jason Lemkin

But I think you do. I mean, I think it actually speaks to something. Why do you introduce ads? If you look at it, the 2 biggest ad businesses are Google and Facebook, and both of them agonized for about a year before introducing ads.

Rory O'Driscoll

Google in 2001/2002ish and Facebook in 2005/2006. There were 3 big ad businesses—the 2 biggest ad businesses were Google and Facebook, and both of them agonized for about a year before introducing ads.

Then everyone goes to the following logic: There’s no other way to monetize. There’s simply no other way to monetize. The cost to serve a free ChatGPT customer is higher than either Facebook or Google. So what are you going to do for a business model? Facebook and Google didn’t have a paid tier.

One argument could have been that a certain percentage of the free people would just convert to paid, and that would be enough to make the business work, right? But the truth is, consumer conversion tends to run well under 10%, probably 5% or less. You just don’t get enough conversions to serve the free tier. You’ve got no choice.

Once you recognize you’ve got no choice in the end, just like Facebook hated ads and Google hated ads, you can hate ads till you’re blue in the face, but America wants free shit. The only way consumers get free stuff is if you run ads. So there’s no choice. I think it’s inevitable. Rip the Band-Aid off, especially if capital is going to get more expensive. You just have to go do it.

I think the ads will be great.

Jason Lemkin

Yeah. I genuinely think they’ll add value to ChatGPT. I don’t have the numbers on it, but I know for myself, I do my vendor discovery using LLMs. I do my vendor discovery on Claude and a little bit on Google. That’s when I want to find a new tool or a new product. That’s where I start. I start with LLMs, especially if I’m using the free product.

If I get my rich response on which auth product to use in FounderStack, and at the end there’s a little ad from WorkOS, and I choose to click on WorkOS instead of the one I chose—and that was a great option, too—that’s a win-win.

I think it’s a win-win for free people. I suspect, in the beginning, unlike Google—we wanted to blow our minds out with Google because it was 10 blue links before the 1 natural result, and now it’s so polluted with ads we can’t even figure out what’s an ad. If the ratio is, say, 10 to 1—a lengthy analysis of what’s best for you and a little ad in a different color—I think that’s a win-win for everybody.

It makes the product more profitable and gives it better margins, and we get some value. Advertising is not valueless to consumers when it’s perfectly executed. It’s not valueless.

Rory O'Driscoll

I think that’s a great point, Jason. I really do. There was a period of time when the Google ads were awesome. Early on, when there were only a few paid links, you remember on the site, it was like, “Okay, that was marginally additive.” You’re exactly right.

Right now, obviously, they’ve swamped it, and you can barely figure out what’s going on. All these things tend toward that wonderful word, enshittification, at scale. But you’re right, this could be the period where 1 or 2 ads at the bottom are additive in terms of information, especially when you auction.

I mean, the beauty about the auction process for the ad placement, right—it’s going to sound weird—is that because Google and Facebook run these very efficient auctions, you actually end up selling the ad to the person who values that real estate the most, which usually is someone who’s got something very precise to sell you.

To your example, Jason, you’re exactly right. If you write this long query on some kind of auth product, at least the ad you’re going to see is someone who says, “Dude, you should buy my auth product instead of that one.” There’s net information added here, right? That won’t be true when they have 40 of them, but that’s 10 years from now.

Plus, I think we’re underestimating how much better LLMs are for discovering what to buy. I find Google unusable for discovery. Unusable today. It is all random ads. I can see the vendors, but I can’t figure out which product to use or buy. Google’s useless.

Amazon is frankly more valuable, but it only works for the goods Amazon is selling. It’s still exhausting, right? LLMs are a gift, if you use them properly, for discovery.

Instagram’s full of ads, TikTok’s full of ads, but it’s pretty mediocre for discovery. It’s just well-targeted, right? This is brilliant, I think. Bring it on, bring it on.

Jason, do you think it will be an unbelievable, massively significant needle-mover in terms of revenue very quickly?

Jason Lemkin

I think it—listen, first of all, there are a lot of products we’ve talked about, like OpenAI’s web browser, that we’ve never discussed again. We may never discuss Claude Cowork again. I don’t know. They try stuff, okay?

I think we are way under-discussing the power of discovery in LLMs. I think this is the way we will buy everything in the future, as we are embedded in LLMs. I don’t know why I would use anything else other than the best of ChatGPT, Claude, or Gemini to find a product to buy. Why would I use anything else? It’s so powerful. And so the ads will be great.

And you think within a quarter this will be a billion-dollar-plus revenue business?

Jason Lemkin

I’m not—I haven’t run the math. You’d have to give me a moment to run the math, but why can’t a billion dollars go to it if there’s any ROI? It only has to be, like, 1% of ad spend going to TikTok and Instagram to move over there. Marketers will just have to try it. The first billion may not be impressive because you’ve got to try it.

Rory O'Driscoll

Agreed. And a couple of comments on that. There’s just a ton in this. First of all, yeah, it’s hard not to imagine it’s a billion dollars very quickly. Jason’s right. Remember, it’s only 5%—they’re doing around $20 billion in revenue. It’s only 5%. It won’t be a needle-mover.

What are the differences between this? When Google and Facebook added ads, each of them at the appropriate time for them—2002ish and 2005ish—they had significant but manageable cost structures, no other revenue source, and very quickly they became profitable, right, on profit and capex.

Even a billion dollars is a drop in the bucket compared to the spend here. So it better do a billion dollars pretty damn quick, because if it doesn’t do a billion dollars pretty damn quick, it’s not going to do $20 billion reasonably quickly, right?

My gut is that it is, because they’re just going to find a way to make it work. I think Jason’s right: It is prime real estate. It’s worth pointing out, on Jason’s comment, that we’ve all been in this—everyone kind of went through that, “Oh my, Google is doomed,” a year ago. Then Google executed on a bunch of things, and now we’ve all gone to, “Oh, Google’s amazing, and likely Coatue was so stupid for leaving Google out of their amazing AI companies.”

I think they are amazing in terms of their AI, but Jason’s comment is really significant: He doesn’t go to Google anymore for search.

Jason Lemkin

For discovery, for what to buy.

Rory O'Driscoll

For discovery. You’re exactly right: discovery real estate. So even though Google’s doing amazing, it is worth pointing out that their cash cow, which kicks off $240-odd billion a year of revenue, is the ads business, and it isn’t the best product on the market anymore.

Some of those dollars will go to OpenAI because it’s better real estate. So it’s not all one-dimensionally good for Google today, just like it wasn’t all one-dimensionally bad for Google 12 months ago. They’ve done an amazing job of getting relevant in all the spaces and winning some of the product wars, but they still face the core problem, which is that search is not the best place for discovery anymore. Like Jason said—

Okay, just for fun, I asked Gemini, so that we’re not biased in here for OpenAI, to do $25 billion in search revenue. That’s really Rory’s point: to open the floodgates at a $50 CPM, which I think is possible because it’s about discovery, right?

They just need 2.2 ads per prompt. They don’t need 10 blue links where you can’t find it. They need 2.2 ads paid per prompt to do $25 billion at their scale. Does that sound implausible?

Every 1 in 5 prompts—every 1 in 5 ChatGPT search interactions—has to be a discovery ad monetized.

Jason Lemkin

No, it does not sound implausible. At a $50 CPM, it has to work. This can’t be garbage ads. I think it’s pretty plausible. We may be upside-surprised. It may turn into Facebook or Google back in the day. Within a year, our jaws might drop.

I think this clip will be used in 3 years’ time, when they’re at $100 billion in revenue, and you’ll look back and go, “Wow, we underestimated this.”

Rory O'Driscoll

No, I’d push back and say I don’t underestimate it. I actually think that this is the core. This is the money, because if you look at intent—look, intent with the exception of Facebook, because it’s about consumer knowledge—Google was about intent.

Amazon ads, which have exploded—we don’t talk about it because it’s buried in their bigger business—but all their gross margin now comes from ads. The retail business is just an excuse to sell ads. That makes all the margin, on the retail side to be clear, not the cloud side. That’s exploded to, I can’t remember the numbers, tens of billions of dollars. It’s sub-$100 billion, but tens of billions of dollars, because they have intent. You’re right, Jason: At least 1 time in 5, I’m just looking at my search—not my chat history—I bought a TV.

I hate buying a TV. I just did 5 or 6 queries on ChatGPT: What’s the best TV? Why is it the best TV? How should I mount it on the wall? How big should it be? Here’s my room size. I got a name. I remember the name. I went into Best Buy and said, “Do you have this thing?” They said, “Yes,” and I bought it.

If they’d given me a click, I’d have probably said, “Can you deliver it?” and paid the extra money. It’s prime real estate because it is the best way to interact for complex purchases. It’s not because it’s inventing anything; it’s just synthesizing the shitty internet into the actual answer. Sometimes the answer’s wrong, which is a little bit terrifying, but most of it was right on the TV.

I think this is a great business for them, and I think you’re right: they’re going to go at it. Jason, by the way, I’ve got to give you a little push here. It is why the deals that you hate—the AEO, the answer engine optimization—are going to be a vitally important business. Harry and I like this market. He’s got a play in it with Peec. I don’t, because I lost out on one of the deals with Profound, to my sadness. All those companies are going to matter.

Jason Lemkin

I agree with you that someone is going to build a massive business connecting these ads to the LLMs. I’m not convinced it’s from the snake oil that I have seen in the GEO products that I’ve used to date. I believe there’s snake oil. They’re telling you to do very basic things that maybe work, whereas I tell you what we should do.

This is risky: let’s all put a bunch of money into AppLovin and The Trade Desk right now. That’s what we should do. I don’t know whether they will listen. I’m not a total ad-tech expert. I don’t know if they’ll be fully open to OpenAI, but if they are—if they are open to ads, if you can run your ads across these platforms, especially The Trade Desk—maybe you make a lot of money. The Trade Desk is beaten down. I don’t know if I buy that, just to be clear.

Rory O'Driscoll

Just to be clear, you’re saying they have access to OpenAI, so they’re able to do effective routing across OpenAI?

Jason Lemkin

If they’re allowed to. I need to think about it for the next week, but that’s a bet you might be able to make. If you step back—and maybe I’m wrong, maybe Rory’s right—who can we bet on today that ideally is public, because we’re all traders? If the OpenAI platform ends up being somewhat open for ads, and it does $25 billion or $100 billion, who could we bet on today where we make money because they’re beaten down by the start of the show, by the SaaS sell-off?

The bet just vanished, and Adobe did something smart.

Jason Lemkin

Hang on. There are 2 ways to make money when you’re selling software to people doing ads, right? It’s the same as Google. There is helping them show up in the free part, which is about optimizing how you appear—that’s what the AEO guys do. And then there’s helping buy the paid part, which is what AppLovin and The Trade Desk do.

I’m going to say something: I don’t think you’ll need AppLovin to help buy ads on ChatGPT, just as you don’t need AppLovin for Google or Facebook. You need AppLovin and all that for everything else, mainly, right? I think—well, maybe not AppLovin, but The Trade Desk in particular. ChatGPT, OpenAI, will make it very easy to buy ads on OpenAI because everyone knows exactly how that’s done.

You have this auction process, and you auction against Jason’s intent. If Jason is online talking about OLED, you run a real-time auction process, and whoever wants OLED the most is going to advertise against you. So, I think for the paid ads, all the value that’s going to be captured is going to be captured by OpenAI, and there’s not going to be an optimization engine.

Then you have the free content. In other words, how do you make sure that the LLMs say nice things about you? I think that is an interesting business. As of today, I think in the US, the Profound, the AirOps, the Evertune—all those guys are super interesting. Harry’s got Peec in Europe, right? But to your point on The Trade Desk, we missed the great trade because Adobe took it up.

Semrush, which is the king of search engine optimization for Google, got acquired recently by Adobe. I think that acquisition only makes sense if their plan is, as quickly as is humanly possible, to introduce an answer engine optimization product, because they bought that thing for 4x.

Rory O’Driscoll

That’s why they bought it. That’s why they bought it.

Jason Lemkin

And that’s why they bought it. I will admit, I was talking about why that’s a better decision than genuinely buying Peec, with a much smaller, more concentrated team that’s much more focused and better for a new AI world.

Why would Adobe buy Semrush over Peec?

Jason Lemkin

I can tell you why.

Yeah. Why?

Jason Lemkin

Well, I worked there a long time ago. When you have a conservative company like Adobe getting $15 million or $20 million—whatever Semrush had—of GEO revenue at a 4x revenue overall, that’s the kind of accretive deal that makes people comfortable.

They could buy Peec for $50 million, $40 million, or $80 million, but I don’t know if that returns your fund, Harry. They’re not going to spend that kind of money. Adobe’s done very few deals where they’re spending billions of dollars on something with a token amount of revenue for technology in the future. It’s just not the DNA, right? Other folks might do it. It’s just not going to be Adobe.

I’m just saying Semrush are not best positioned to win.

Rory O’Driscoll

“They’re not best positioned” is an interesting phrase. First, I agree with Jason. This is an odd combination, in that you have 2 guests on your show, Harry, both of whom have sold something to Adobe. Jason—as a CEO, I mean, because I was on the board of Omniture back in the day, which was the first cloud company that Adobe ever bought.

They were all about desktop graphic tools, and they didn’t even have a marketing cloud. They effectively bought Omniture as that. When people were like, “Why are they doing it?” I thought that was very zoomed-out and high-level. When you want to make a big move as a big company, if you buy something small, you’ll just smother it. You’ll stomp on it, right?

You want to buy something with enough critical mass and heft that it can go do its thing and not get swallowed by the machine. I think that if I was Adobe buying Semrush, provided Semrush had some story around what they’re doing in AEO, it’s more likely to be successful than buying Harry’s little startup company and hoping that somehow it gets rolled out in the system. Jason knows so much better than me: if you don’t have the customers already and the distribution, it’ll just get lost in the big company.

Yeah, it’s great. They just walk it around to their CMOs—Adobe does—and say, “We have this. We bought Semrush. It’s now been rebranded Adobe Marketing GEO, SEM Cloud, and we know this is one of your top issues.” They drop by: “We have a solution. It’s $20,000 a month, and it just works, and it delivers a report every day that is then forwarded to their team,” right?

That is insanely valuable for 99% of the world versus buying some crazy Lemkin guy’s startup that could blow up on you. There’s just so much risk for a non-founder company.

Rory O’Driscoll

Totally.

Jason Lemkin

Tobi could buy a company like that. You need somebody like Tobi or maybe even, I mean, a Satya or something like that who can take some crazy bets. Most companies don’t want to take that bet.

Jason, I’m waiting for the day when Rory goes, “You know, I remember being on the board of the East India Company.”

Rory O’Driscoll

They didn’t mean that long ago.

Jason Lemkin

It’s not the same space, but it’s pretty good, this East India Company.

Okay. This East India Company—they had a great 500-year run, dude.

Rory O’Driscoll

They did. They really built a monopoly. You’re right.

There are a couple of big rounds that went down that I do want to hear your thoughts on. There’s ClickHouse at $15 billion, there’s Replit raising at $9 billion, and there’s Cerebras raising at $22 billion. Which one do we want to focus on?

Rory O’Driscoll

Listen, I wish I was a total ClickHouse expert, but I’ll tell you what is kind of interesting about it. Maybe we could gloss through it. You asked: it’s an extreme example of how folks take advantage of AI tailwinds.

ClickHouse is basically an in-house, open-source product built at Yandex, which is now Nebius, built in Russia. Just like everyone has an open-source product that has gotten scale—LinkedIn and others—it’s a very clever way to mine massive amounts of data and make conclusions from it.

They built their own for themselves. I mean, Yandex was the Yahoo or Google of Russia. They built their own, and it works. Then the timing is perfect: they spin this thing out in 2021. They take an open-source product and turn it into essentially a proprietary cloud product, and then boom, AI blows up.

Folks are already using it. Tesla already figured out that this works because they were early in AI. Everyone’s already figured out this is best of breed, but they nailed the conversion from free open source to closed-source hosting. Fast-forward to today: Anthropic and everybody needs it.

But this is not a product that was born at the start of this year. This is a product that was already around. I’m not a total expert, but, man, just go find your ClickHouse. It’s easier said than done, but this is not a brand-new product.

That’s your Replit—it’s 10 years old, too. These are old products that found their AI tailwinds and blew up, right? It was only doing $50 million a year ago, right? ClickHouse or something like that, or less. Blew up.

Jason, when you’re doing ClickHouse at $15 billion, it does feel pretty expansive when you compare it to Snowflake or even Databricks in some respects. What are you underwriting it to?

Jason Lemkin

I guess Databricks is your comp, right? Which is, I guess, a large assumption that it’s even better because everyone in AI is going to use ClickHouse. You’re assuming, one way or another, you’re going to get almost 100% market share, and some will be directly competitive with Databricks or even Elastic, where I think this is the core, and others will be quite complementary.

You’re just assuming, I think—the bet is—I don’t know, I didn’t see the deal at $15 billion. Maybe Rory did, but I think you’re just assuming everyone uses it, which maybe is a rational bet. Once in a while, these products come into the market where just every single person uses the product.

And open source is common. It’s just hard as heck to monetize it the way ClickHouse does. That’s the clever part, right? I’m not even an open-source expert, but it’s brutal if everyone just goes around and hosts it themselves, builds their own version, and forks it, right? They nailed this. They nailed it.

Whenever you pay up for these high-growth companies, what you’re effectively saying—in math terms, you can use 3 different words to say the same thing—is that the most recent growth rate is going to continue for a long time. You’re paying an absurd revenue multiple, but it’s not an absurd revenue multiple if the growth rate continues for 2 or 3 years.

So, you’re basically underwriting growth persistence, right? The next level down from that is: What does that mean? You’re basically saying this is a category, they’re the winner, and it’s a big enough category to keep going for 2 or 3 more years at least at this growth rate, and then decelerate slowly, right? Is that crazy?

2 to 3 years at 3–4× growth on $350 million. That makes it pencil out, right?

Jason Lemkin

And again, we have examples of this absolutely happening if you look at Anthropic and companies like it. Then we have examples in SaaS of people where you want to underwrite a growth rate and, suddenly, that growth rate deteriorates and you’re high and dry. It’s that simple. There’s no other magic.

What I always say to our investors, and I’ve probably said this on the show before, is that first of all, you have technical and founder risk. Then you have business go-to-market execution risk, which is typically what we invest in. And then, at the end, you have valuation risk, right?

Valuation risk expands to fill the gap. Once the other risks are taken out of the deal, you’re left with valuation risk, which is all about growth persistence and market size. The good thing for ClickHouse is that it is a category, right? Step 1, what you don’t want to do is find the third-best random database.

But if you look 1 level deeper, what they do—OLAP, or online analytical processing—and this category has existed in prior generations, too, right? Back to the ’80s and ’90s, Harry, right? Way back in the 2000s, when Oracle was the relational database king, there were some obscure OLAP-type databases even back then that would be equivalent to Databricks today, which is obviously a bigger category.

Then you have these transaction and analytical processing databases, like Teradata. Think about that: It’s not about writing transactions to a database. It’s all about having 1 million, and now maybe 1 billion or 10 billion, data elements there, and wanting to quickly scan down a column and add them all up. How many clicks? How many people traversed your website? Large amounts of typically read-only data.

If you use a standard Snowflake database, it’s fairly inefficient because that’s written to be able to write transactions. You write a transaction like, “Here’s my debit or credit.” Snowflake is optimized for that, and analytic databases like ClickHouse are optimized for analytics processing. It’s a different category of database.

Data warehouses back in the ’90s and 2000s were the same thing. You have your production database where you run your system—your banking system, your ERP, whatever it is—and then you have this analytical place where you put all the transactions, because some analyst is going to wake up and say, “I have a really obscure question. How many people bought this product in this district 2 days after they did this online on the website? I need to know how many people did that because I want to do attribution.” Some weird corner case like that, right?

In AI, the number of those queries has gone to infinity, because if you have that information in AI, you can use it to predict things. So Jason’s exactly right: You have this optimized, specialized tool for a very clear use case that’s different enough from Snowflake and Databricks that you have this separate category. It’s become important because AI eats that shit up.

So you have this explosive growth, and it’s a category. The only question is whether the OLAP market can support a $30–$40 billion outcome. You squint and look at Snowflake at $80 billion—it’s a transactional database—and Databricks at $110 billion. Maybe it can.

Typically, if you look back at the prior generation, the data warehouse category was significantly smaller than the relational database part of the marketplace—not the transaction-processing part, but the transaction-processing part of the marketplace. Your airline, for example: You’re running your airline reservation system. That’s a transaction-processing system that can’t go down for a single minute because you lose billions of dollars.

The analysis at the back of that—American Airlines wants to run an analysis of how premium customers flew last week—is something that could be a little less performant and therefore not as big a market. These markets have existed before. It’s typically not a multiple of the Databricks or Snowflake-type marketplace, but a—what’s the word for below 1?—a fraction of it, but a pretty appreciable fraction.

So if Databricks and Snowflake are worth $100–$200 billion, maybe you get a $30–$40 billion outcome here. That’s the bet. Sorry, long-winded answer, but it’s a category. They’re the leader. Demand has gone up because of AI. You squint 1 way and say 2 or 3 more years of growth, and you feel like a hero. If it turns out to be a smaller market than you think, then you’re high and dry. Welcome to late-stage investing.

Would you rather be in Databricks at $130 billion or ClickHouse at $15 billion?

Jason Lemkin

Basically, what you’re saying is: Is the OLAP category more or less than 10% of the core? Databricks isn’t really relational. It’s much more about AI-enabled data manipulation. When you look at it like that, you say it’s not crazy that the subcategory is more than 10%. It might be 20% or 30%.

So, yes, if they were both public, you could have a fun macro trade of shorting 1 and going long the other, but that’s not the way the world is now.

Okay, final 1. Actually, fuck it. Let’s do a quick-fire and just cram a couple in.

Rory O’Driscoll

Oh, God.

It’ll be fine, Rory. Don’t worry. Sequoia is going big into Anthropic. They’re also in OpenAI. Are we seeing the end of competitive investing?

Rory O’Driscoll

No one gives a fuck about competitive investing anymore in 2026.

Well, didn’t we talk about this last week, when I said Andreessen should target 50% market share, not 10%? To Rory’s insight, I said they can get past the competitive issues.

Rory O’Driscoll

So, yes, Sequoia has passed it. [Laughter.] Different partners, different funds. I don’t buy any of that.

But I think the truth is, at this point, when you’re piling in at a $350 billion pre-money valuation, you’re not on the board. You don’t have meaningful information rights. It doesn’t matter.

Again, going back to first principles, this is really Fidelity mid-cap growth—well, actually, large-cap growth now—but just in the private markets. I repeat that it’s absurd that we’re in this place, but whatever.

Given that Fidelity Growth is always going to buy, if they believe in a category, they can totally buy 2 or 3 different things with no conflict, because they’re not in the room when decisions are happening, right? I think these late-stage investments, if you’re not in the room, don’t matter, because you don’t have meaningful information rights. You’re not getting the board deck. You’re not there for the strategic stuff.

You couldn’t do the Series A of OpenAI and the Series A of Anthropic. That would be stupid. I would assume 1 of the CEOs would stop you. But getting in at a $350 billion pre-money valuation, where your $1 billion gets you 0.3% of the company—which is about what you’d give a director at Series B, right?—with no information rights, no nothing, it doesn’t matter.

I mean, nothing. You get 1 email a quarter from each of the 2 companies. You know where they are. As I say, you are just a public-market investor in private assets.

6. Replit's $9BN Deal: Analysed

Jason Lemkin

Jason, I’m intrigued by your thoughts on this 1. Replit at $9 billion.

Rory O’Driscoll

I want to hear Jason.

Jason Lemkin

That’s Lovable at $6.5 billion as the last data point.

I think they’re neck and neck in terms of revenue. Really, let’s imagine they’re at $300 million this year, and when they did the round before, they were at $100 million or $100-something million. You can justify the step-up based on that. You could also challenge it using Rory’s math.

What I think people might not get is that this product is 50 times better than it was at the $2 billion or $2.5 billion round. It’s not a little bit better. When we started doing the show, I was trying to use V1 of Replit. I could not finish an application. I couldn’t finish anything, and it almost blew up on social media. You might remember, right? It was unfinishable.

Rory O’Driscoll

It just wasn’t there. One of the founders of a competitor called me up and said, “None of us are there in the industry. None of us are there.”

Jason Lemkin

Over the holidays, for fun, I built an incredibly complicated game. I’ve never built a game in my life. I built a startup simulator that simulates everything from going through YC to IPOing to controlling all the global power, GPUs, and tokens in the world. People love this game. What’s interesting isn’t that I built it; it’s that it works.

Rory O’Driscoll

Yes.

Jason Lemkin

Nothing worked at $2 billion, $2.5 billion, whatever the round was. Nothing worked. The joke was all over the internet: everyone’s got a project they 80% finished in Lovable or Replit. Everybody did. They couldn’t finish it. It was a joke because they thought they had 80% finished it, but there was no chance to go from 80% to 100%.

Literally, I built this thing over the holidays, and I probably put 100 hours into it because it was over the holidays, right? This was not a one-shot deal, but it’s magic. If the revenue growth justifies it and the product is literally more than an order of magnitude better, I would argue—I know this is the math that we’re going to look back on and make fun of ourselves for—but I would argue this is a much less risky investment today, whatever it is within reason, than it was at $2.5 billion, when I didn’t even think this was a stable product. I thought it was cool, but it didn’t work that well.

By the end of this year, it’s going to be even better, right? We’re just getting so many benefits from these improvements. It’s like we’re—if you watch the Ben Affleck and Matt Damon one on Joe Rogan, pretty fun, and Ben’s like, “Well, nothing’s really improving anymore in LLMs.” Maybe it isn’t in Hollywood, although I doubt it. We’ve just started with agents, and it’s so much better.

I think Replit is 100 times better. I know the venture math doesn’t work this way, but if it’s a 100- or 1,000-times-better product, it’s probably worth 2–3 times more if the revenue is growing at outlier rates. They’ll be at $250 million ARR now, and if you assume, given their growth rates, that they’ll realistically be at $700–900 million by the end of the year, that would be a—

Rory O’Driscoll

Yeah, well, God, you’re paying—

Sounds like a better deal than a lot of other portfolio companies. If it hits the number—if it hits the number—it sounds like a pretty good deal, right?

Jason Lemkin

Yes. Implicit in that is a huge amount of belief in growth persistence, but yes, agreed. If they do that, then yes.

What was your question going to Jason?

Rory O’Driscoll

Actually, genuinely, as a Replit user, I too Replited it over the holidays. I’m inspired by your example, honestly, and it was fun. What I did find, though, is that I used to use it when I was doing my Python lessons. They’re very much driven now to a much higher level. It’s all agent-based. Basically, you don’t even need to see the code, Mr. Designer. It’s all about interacting with the chat.

I’m actually trying to get down to the developer level and understand the codebase, and it’s almost hard to find. It almost resists being found. The interesting thing about it is you have to embrace a world where you don’t need to know the code, Mr. Designer. Once you do that, just describe what you want. It’s a whole new skill. It’s almost like it’s not programming at the programming level; it’s describing at the describing level. Once you embrace that, you get stuff done, but it’s really hard to go back to first principles and say, “How does it work?” Right?

Jason Lemkin

Yeah, you can view the code, but you’re right. You can view it and some folks use it that way, but it’s basically abstracted away. It’s hidden, right?

Rory O’Driscoll

Agreed. You can find it, but it resists being found.

Jason Lemkin

The more you do it, the better you’ll get at understanding how the agent thinks and works, right?

Rory O’Driscoll

Exactly. This may be like a dude when spreadsheets were invented saying, “I like to add the numbers myself to know they’re right.” Maybe I just need to let go and say the spreadsheet’s right, and just stop worrying. Don’t be an idiot. But yes, it was super interesting and extraordinarily powerful once you use it.

The final comment, because you had it in there: Sequoia doing Anthropic and Sequoia doing ElevenLabs. I was just reflecting on this. Obviously, given the changes they had, clearly the mission of the new management in charge is, “We’re going to execute on AI.”

Rory O’Driscoll

New management of Sequoia, you mean?

Yeah. I mean, they had a transition. I think part of the reason was—my point is this: if I was a late-stage company looking to raise capital right now with a very compelling AI story, it’s clear there’s an appetite to buy there.

Jason Lemkin

We’ve talked about it before, but I think the thing that people haven’t discussed enough is the promiscuity of traditional early-stage firms across rounds.

Rory O’Driscoll

But I think, in general, you have these multistage firms and you make a multistage bet. The whole point of the late stage is to clean up on your early-stage misses, and they’re doing that.

They’re—I mean, I give them credit. Everyone’s executing the mission that they said they do, right? Everything stays private longer. Get into good deals. It’s best to get them in at the A, but if you have access to infinite capital, getting in at the F is okay, too. Or, in the case of Databricks’ D round, just stuff the money into the good companies.

Jason Lemkin

Obviously, as a trend, it will only stop when it reaches success. I love the Jason the Fog Rags [?]. It’ll only stop when it reaches success, because that’s the nature of economic behavior. People are going to do this until it doesn’t work.

Harry, you had Alex Rampell from Andreessen on the pod the other day, right? He made the point: listen, this is how you win at Andreessen. You either own a lot early—

Rory O’Driscoll

Yeah.

Jason Lemkin

—or you invest whatever you can get into if it’s a guaranteed winner like that. That’s it. That’s the playbook. If, in today’s world, you have LPs—especially if they’re in SPVs or annex funds—and they’ll give you billions and billions for access to these things, you could argue whether you want to do that or whether you don’t like it, but it’s almost free money.

This is where you’re going to hate me for going off on a tangent, but fuck it. At the end of the conversation, the game has moved to pre-seed and pre-IPO. But do you know what? The dumbest game, I think, right now, is playing the super-competitive Series A’s and B’s in venture, where we’re paying up super early for little signs of product-market fit in intensely competitive markets at 100 times ARR. I’d way rather have 15–20% of a pitch deck and 2 people, or be piling into Airwallex as I did at $5 billion, and have a little bit of something that’s really fucking working.

Rory O’Driscoll

But I disagree. I hear what you’re saying. Anecdotally, you might be right, but in the end, over the long term, markets have to tend toward rational equilibrium. That means as you take less risk, you get less return. There can’t be bad stages. This whole “X stage is really bad, Y stage is really good” thing—I’ve heard this for 30 years. It doesn’t make sense.

In the medium term, you have to assume that as you go down the risk curve, which is what stage is—going down the risk of—you slightly go down the return curve. There are interim anomalies at various points in time. You’re exactly right: anecdotally, Series B can be really hard now, and people say, “Oh my God, Series B is really hard now,” and so on. But you’re building a firm over an extended period of time. You’ve got to be cognizant of where the market is at any point in time.

It’s very hard to say, “I’m doing Series A’s and B’s. I’m getting 10–20% ownership. I’m going to wake up this morning and that whole sector is bad.” What does that mean? It’s just a weird comment.

Jason Lemkin

It means risk-adjusted. My dollars are better suited elsewhere. If you believe that’s going to be true over an extended period—if you believe that disequilibrium is going to persist longer—then, you know—

Rory O’Driscoll

I’m not saying that, but I’m saying play the game on the field. Right now, the field is showing you, to Alex Rampell’s and Jason’s point, own a lot of something.

Jason Lemkin

If you say, “Play the game on the field,” and you play the game on the field, what happens if you thought you were playing soccer and suddenly you start playing golf? It gets a little weird at some point in time. It is a little weird to say, “I’m now going to do ultra-late stage.” So yes, there’s got to be—I’m tacking and adjusting.

But, for example, if a late-stage firm said—let's just say someone was doing the other extreme. Someone was just doing ultra-late; they're putting money in 10 companies at $10 billion to $50 billion pre because there are 10 companies, not $10 billion to $50 billion pre, and that's their thing.

They're basically saying it's the pre-IPO last round, right? And that's what we underwrite. Then they came to you as an LP and said, “Right now, those pre-IPO deals are priced wrongly. I'm going to do seed.” You'd say to yourself, “You don't have the—it's too far from what you're doing. You can't get there from here,” right?

They might come to you and say, “Hey, these late-stage privates are not priced right. I'm actually going to do public equities.” It kind of makes sense; it's adjacent, so you can duck around in the box, but you can't do violent switches—not with most of your money.

Jason Lemkin

I'm sorry. I think Thrive have proven that to be wildly wrong. I think Benchmark have proven that to be wildly wrong.

Rory O’Driscoll

Well, I disagree with the two. I think Thrive have proven it to be wildly wrong. I think Thrive have executed that really well—excellently. You're right. They've probably put 60% to 70% of their dollars into high late stage, but at the same time, they have excellent early stage. Yes, they've pulled that off. You're exactly right. Give them credit for that.

I think some people, for example, in AI, have done really well by stuffing a bunch of money into Anthropic all the way up, and that's been a brilliant decision. Look, I do think where you're right is that the new space that didn't exist before was this late and ultra-late stage, and people have found a way to profitably fill that. If you found a way to profitably fill that by hiring the right growth folks, yes, that's been a good opportunity too, right?

I mean, the zoom-out comment is this: The public markets ceded another 3 to 5 years of growth to the private markets, which meant market expansion for venture and market contraction for small-cap public equities. Not the best decision in the world, but that's why everyone's allocation to private has gone up.

That's why LPs, to Jason's point, are giving money to these big mega-firms to say, “I can't get me that Anthropic in the public markets. Go get me some.” As I say, I'm not sure why the world has decided to give even more money on a 2-and-20 basis when it could be giving on a 50-bps basis. But that's the world we live in.

Because Marc and Ben need it. Okay, why would you—why would you?

Jason Lemkin

Everyone has needs. Just simple, humble billionaires.

Alex Rampell

Yeah.

All right, team. I love that. Rock and roll.

Jason Lemkin

Rock and roll.

Okay.