为什么需要10亿美元基金才能做Series A｜SpaceX估值2万亿美元与太空数据中心｜Groq的200亿美元交易

1. Anthropic vs. OpenAI: Who Is Actually Winning the Enterprise War?

Guys, it is so good to be back. We had a lot going on this week. I think there are a couple of places we could start.

To the arbiter of economic justice and revelations, Ramp, who revealed recently that they process, I think, about 0.5% to 1% of U.S. GDP transactions, or whatever it is that Eric used as the stat to validate themselves. Ramp data suggested that Anthropic now captures 73% of all spending among companies buying AI tools. Ten weeks ago, it was 50/50 with OpenAI. In early December, it was 60/40 in OpenAI’s favor. Are we seeing Anthropic run away with the enterprise lunch, so to speak?

Rory O’Driscoll

Just to start with the facts, they actually said 73% of new spending, right? The same graph shows OpenAI is still actually ahead of Anthropic in terms of total spend, but the marginal buyer in the last 6, 8, 10 weeks has massively shifted, which is obviously the most leading indicator.

People in the market today for a new AI went 70% to Anthropic. So, just in the interest of being precise—

Yes, the claim from Ramp. I thought the OpenAI response—it’s not, you know, that snarky comment about extrapolating from a lemonade stand—was just a bad look, right? First of all, it kind of doesn’t really understand statistics. I would argue Ramp is probably a pretty accurate statistical reflection of especially digital company spend in the U.S. I think they have a pretty diversified customer base, and they probably have decent data and good data scientists. So, OpenAI trying to be snarky, I think, was a mistake.

I think the real point is that it does represent the facts: in the last 3 months, there’s been a shift in the zeitgeist. I do believe that the marginal user, the marginal person opting for AI today, or even people switching today, is moving toward Claude and away from OpenAI. That doesn’t mean the end of the world, but sometimes the first thing you have to do in dealing with a problem is face the hard facts. I think the data was good and the conclusion is real.

If Anthropic is now maybe a $22 billion run rate, the revenue does tie to these conclusions, too. It certainly isn’t inconsistent with it, right?

Jason Lemkin

The only thing I might say—I don’t want to spend all our time on it—is that I think it’s possible they’re both right. OpenAI and Ramp are both right. What I mean is, in some ways, this felt to me like the Cursor debate.

If we walk into our portfolio companies, barely anyone’s using Cursor today. In my own portfolio—and people said it on Twitter, too—Cursor’s dead. No one’s using it. They’ve all moved to Claude Code. I would say, in that ecosystem, it’s true. Maybe if a lot of Ramp’s data is still biased toward tech, they might see this same trend.

Claude has—if you just look at it, everything’s accelerated since December. It is Opus 45 and after. It is crystal clear. As soon as Opus came out, it was another step function that was under-discussed. Everyone’s PRs exploded; everything got better.

But for the normal world, they live in ChatGPT. So I don’t want to say for sure that the Cursor experience isn’t happening here. My point is they both could be right.

But certainly, what I don’t like is how OpenAI is acting wounded, to Rory’s point. I don’t like it. When we started this podcast, OpenAI seemed invincible no matter what Anthropic did. It seemed utterly invincible, and you could just smell this air of desperation: “We’re going to keep headcount flat to manage costs”; “We’re doubling headcount”; “We’re going really deep on agentic commerce”; “We’re basically canceling it,” and Walmart says it doesn’t work. It feels very inconsistent.

The one thing that I love about Anthropic is that it’s very consistent about its ICP and goals. It has been very consistent. We know what it stands for. We know what it’s trying to do. It launches new features—it’s got its new Opus 4.6 version of Claude, launched yesterday as we record this—but you know what’s coming with Claude and Anthropic. OpenAI, I’m getting whiplash from everything. The Debbie Downerism—it’s not going to last.

It doesn’t smell like that at OpenAI today, does it? It’s just a downer to be around, and I don’t want to try their products because of it. Honestly, to Rory’s point, I actually don’t want to try their new products. Literally last night, I was DMing with our chief AI officer. We’re trying the new Claude app that just launched, but it ain’t going to happen. I don’t want to hang out with Debbie Downers.

If we put that into strategic takeaways, in terms of that pivot, Sora is now getting folded into ChatGPT rather than being a standalone app. Hardware ambitions are being deprioritized, and they’re really trying to consolidate efforts—stop having such a diverse product set. They’re also, to your point on headcount, planning to nearly double headcount to 8,000 by the end of the year, having said before that they were actually going to keep it flat.

Is there a question? No, I just want people to understand the context rather than—you know what it feels like to me, in all seriousness? It feels like when we started this podcast quite a while ago, but not a year ago, I don’t think, OpenAI was an exception to the rule. You could have massive founder turnover. You could have massive management-team turnover. You could have an unusually high amount of drama, like kicking Sam Altman out and then bringing him back in, a dysfunctional board, and it seemed to be the exception that made the rule: if you had so much momentum, you could overcome it.

Now I feel like the downside is rearing its head. This inconsistency is damaging the company day in and day out. We can see the downstream impacts of that massive turmoil.

I’m going to come in here and try to say something positive, but start by pointing out that if we go back to those first 10 podcasts, they announced the hardware deal with Jony Ive. If you recollect, I was like, “This will never ship.” I said it at the time, and I actually said, right when he was on a high, that I don’t envy Sam because the press only has 2 stories: “We love you” and “We hate you.” Once they’ve written “We love you,” there’s only 1 story left.

So, they’re just moving through the to-do list. We’ve done the “We love you, Sam”; now, “We hate you, Sam,” right? He has brought it on, and you spend a year talking to the prince of fill-in-the-blank, the president of France, instead of staying at home and shipping product. Eventually, things get to focus, right?

Things are never as good as they seem, or as bad as they seem. That’s just one of my rules. It was never as good as people thought a year ago. They still have—to your point, Jason—they still own the consumer business, right? Job 1 is figuring out how to monetize that and make that a big-ass business.

It’s hard to believe that the advertising efforts seem to be struggling now, but that’s job 1 to figure out. And then, job 2 is to figure out enterprise, in particular coding. They’re finally doing the right stuff, perhaps a year, a year and a half later, but it’s still clear to me that if you just take a big, deep breath, and you were running that organization, focus on the 2 or 3 things, get a little more sensible on your financial trajectory, you still have a comfortable chance to be the winner.

In other words, to exit 2 or 3 years from now as the largest market-cap, standalone foundation-model player, right? You blow it for another year and you won’t.

Jason Lemkin

You know what’s interesting? There are sort of 2 things going on here that I see in the data. If you look at OpenRouter’s data, it has exploded since the start of the year. What that’s saying is that folks are aggressively switching between models for cost and output, especially cost.

There are a large set of customers who are optimizing when to use Kimi, when to use Haiku, and when to use Mini, and that market has exploded, right? We can see it in a lot of our more mature customers—companies that are trying to optimize their spend.

On the other hand, so many of us have said, “Listen, Claude Sonnet and Opus 4.5 and 4.6 are so good.” If you’re not deep into coding or vibe coding, you don’t see how much better it is in the last 90 days. I have no desire to screw around when something gets its hooks in. It’s so good.

I want to build all my scaffolding. I want to build my apps. I want to build my AI agents around something that isn’t just good but now is epically good.

2. The $20 Billion "Acqui-hire": The Groq Deal Broken Down

And so, you're right: there are a lot of applications, even with the OpenRouter data. What I mean is, there are 2 things happening. On the one hand, the soft costs are very low to pick a different model, but on the other hand, there are high soft costs for managing the outputs, QA-ing it, qualifying it, and making it great. I don't want to do anything except SaaStr Nation now. I don't want to spend any time on it. It's not worth the soft costs. And I think that's where the panic is: they can smell that they're losing that, even as cost-sensitive customers will rotate through the cheapest possible thing.

I think so. And I think it speaks to that. You have these discussions: Do you want to be first to market, or do you want to be second and know more? Pioneers get arrows in the back and all those clichés. But I think the real truth is, if you are first to market with the right product, you grab that early mindshare and market share, and then it's theirs to lose.

Just contrast the 2 markets, because it's pretty clear the 2 potential mega-markets here are the consumer market, where OpenAI grabbed mindshare with ChatGPT. And despite the fact that they haven't monetized it yet, no one has really taken that away from them at scale. The other mega-market is not just enterprise, but within enterprise, coding.

3. Win or Die: The Terrifying Reality of the Unicorn "Dead Zone"

You're right, Jason: the scary thing is that maybe 6–12 months ago it was up for grabs. Today, your description is right: it's half up for grabs. People are starting to lock in. If OpenAI allows Claude to become the default for another year and the perceived best for another year, I don't think you get to show up after a whole bunch of people have made enterprise decisions and say, “Oh, now we finally got our shit together. We're good too now, I promise. Please pick me.”

There is a moment. “There is a tide in the affairs of men,” as Shakespeare says, right? This has been the last 6 months of coding lock-in, just the recognition that coding is the mother lode app within enterprise spend. You're right: if you let Claude run away with that for another 6 or 12 months, you've probably sacrificed value that you'll never get back.

Jason Lemkin

Yeah, let me just give you one small example, because the models are so much better since December. Since then, we've built an AI VP of Marketing and an AI VP of Customer Success for real, and they're really, really good. But here is the meta point: our AI VP of Marketing defines every day, every single marketing activity. It wakes up in the morning and gives us Slack updates. It runs our weekly team meetings.

Our AI VP of Customer Success—we have, like, 200 sponsors for SaaStr, and all the humans would quit because it was too much work. It does it 24/7, and the sponsors love it, okay? These are great. It runs on Sonnet 3.7—I mean, sorry, 4.7—and maybe a little bit of Opus. There is no way we're going to switch the model. This is dialed in; it works.

Now, we're going to have to deal with QA when it goes to 4.8 and 5.1. There's a little bit of QA, and it does change. But my God, there's no way we're going to switch these apps, which we rely on every day, to Codex, because it took us weeks to dial it in. You have to train it, and you have to do it.

Now that they're great, once there's a certain level—I'm not saying that other folks won't—but that is a lock-in from the latest models. We will not invest the time after we've done it, because the models today are so good. That's a risk. I would have a code red on this.

I agree. And again, just donning my economics-of-industry hat, you're exactly right. You're an individual enterprise, right? Maybe if you were a SaaS vendor of these products to 1,000 enterprises, you might have a big enough engineering team where it might be worth your while 6–12 months from now to evaluate new models. But you're right: you've built a business that worked for you, and unless they're extorting you on token costs, it ain't broke, so you ain't going to want to fix it 6 months from now.

Jason Lemkin

I agree. And that's why, just like on the consumer side, every time you lock in muscle memory—I mean, I'm using Claude all the time now in Cowork, but I will admit, when I'm doing my random research for this sort of thing, I still go to ChatGPT. I'm used to that. I've got a lot of stuff in there.

There are applications, if we just stick to B2B and AI for a while, that are very sensitive to token costs. We've all—even things like support are super sensitive, right? Because they're using so many tokens. But I got to tell you, there are so many applications like the ones I described above that were not that sensitive to token costs.

If you use $200, $400, $2,000, or $10,000 of tokens for a month for these, it just doesn't matter. And so there's the OpenRouter world where costs are super sensitive, but there are plenty of applications that will deliver epic value on these LLMs where it's not worth it. If you want to reduce my token cost from $2,000 a month to $1,500, I'm not going to—go leave me alone. Leave your gimmicks and all these levers; I don't care. I've got 99 problems. This isn't one of them, right? And there are going to be more of those apps than we think.

You're exactly right. One of the things I've been thinking about for us, for our software-app investments, is just having this mental model of what's the token spend as a percentage of revenue. And you're exactly right, Jason: there's a ton of really interesting apps that, for 5%, 7%, 8% of revenue on tokens, are building huge value, which my sense, at least, is very different from the coding apps, where you might be at 40% or 50%. And if you're at 5% of revenue and you're growing really quickly, you've got a lot of better things to be doing with your life than over-optimizing the models. And I think if you get around it, you might use some open source.

Jason Lemkin

Exactly. I think that metric—I actually meant to do this work, and if someone has done it out there in internet land, I'd love to see it—just looking at a couple of hundred AI apps and literally looking at the AI token spend as a percentage of revenue across them all. I'd love to know what the pattern is, because I totally see very different percentages depending on the token intensity.

4. SpaceX at $2 Trillion: Elon's Insane Plan to Build Data Centers in Space

SpaceX, Terafab, potentially $2 trillion. We're reaching new heights. We started off at $1.2 trillion, $1.5 trillion. Now Terafab and the $2 trillion number is being mentioned. Roy, why don't we start with some context from you? You're the best at providing some kind of context.

Okay. The big-picture context is that Elon made an announcement that they're going to build a fab—effectively build the equivalent, I think, of almost 70% of the volume of all of TSMC, right?—in the US near the Gigafactory. Across the chip need for Tesla and the perceived and potential chip need—I'm picking my words carefully—for SpaceX, to the extent that they build data centers in space, he doesn't think TSMC will be able to make enough chips to support his needs. Therefore, continuing a pattern of vertical integration they've had for an extended period of time, they're going to build a fab. Right? Not just any fab, but the most advanced, modern fab on the planet, for probably a capex cost of $25 billion. That's the announcement.

And it's not clear who owns it, by the way, but I think it's about the usage. The idea—I saw some number like 20%—is that it's some kind of joint Tesla–SpaceX venture. 20% of the volume in the end will go to Tesla; 80% will go to SpaceX and data centers. So that's the story.

The second piece of context is how is SpaceX now being talked about at $2 trillion? Let's be clear what you're saying about that, because I'm going to push back strongly. Polymarket said the probability of SpaceX being worth $2 trillion in the IPO went up to 50–60% from a lower number, right? And that's what Harry is attributing information and signal to. I would point out that Tesla stock didn't move. So if this really is—even if it's 80% SpaceX and 20% Tesla—in the market, nobody blinked, right? Where actual money is, where actually significant money is changing hands.

So, I am significantly more skeptical that 6 months from now people are going to attribute another $400 billion in value to a statement that I'm going to build a fab.

Unless they saw your math—the 80/20—and really thought of all the value going to SpaceX, right?

I mean, yes, okay. So again, it's back to the same eternal Elon discussion. At every point in time, with every one of his companies, you have things he's already done that you can value on a revenue multiple, things that have been announced that were in process and are in various stages of doneness. In that case, you have to assign a probability to getting it done.

If the probability is 100%, then announcing a fab means you own the fab. If the probability is 1%, then announcing a fab means you own 1% of the fab, right? Everybody gets to pick their percentage in that continuum.

5. The Broken VC Math: Why You Need $1BN To Do Series A

Right now, clearly, TSMC itself is just over $1 trillion in market cap. It's basically saying, if it really popped up by $400 billion of value, it's like basically saying TSMC has spent 30 years building the most modern fabs out there. You've announced that you're going to do the same. You do have customers for those chips in the main. So I'm going to give you a 50% probability of getting it done. It's a pretty high Elon-attributed probability number.

But do you think that's unfair?

I wouldn't bet against it. I mean, it's Elon. Yeah, so again, it's back to the same eternal Elon discussion. At every point in time, with every one of his companies, you have things he's already done that you can value on a revenue multiple, things that have been announced that were in process and are in various stages of doneness. In that case, you have to assign a probability to getting it done.

If the probability is 100%, then announcing a fab means you own the fab. If the probability is 1%, then announcing a fab means you own 1% of the fab, right? Everybody gets to pick their percentage in that continuum. Right now, clearly, TSMC itself is just over $1 trillion in market cap. It's basically saying, if it really popped up by $400 billion of value, it's like basically saying TSMC has spent 30 years building the most modern fabs out there. You've announced that you're going to do the same. You do have customers for those chips in the main. So I'm going to give you a 50% probability of getting it done. It's a pretty high Elon-attributed probability number.

I’ll tell you what was interesting to me. I watched something yesterday on YouTube. They had Jay Leno, who was the first one to test the new Tesla Semi, and the team from Tesla came over—Franz, the head designer, and the head PM for the Tesla Semi. They were talking about it, and the designer said, “Yeah, we strongly believe across all of Tesla that the future is fusion. It is fusion to power our trucks. We just believe the fusion’s from the sun. There’s no point in doing it on Earth, and we will soon power all of our semis through fusion.”

This is a thoughtful lead designer saying this vision that has been there, and they believe it. I think it’s a great story that can happen: that we’re going to build more power than, I guess, exists in the world today, and 80% of it is going to space. Eighty percent of these chips that come out of this are going to space to power fusion.

You could mock that or say it’s going to take 9 more years than we thought, but it does create a pretty powerful vision for the IPO and beyond. Now you see it all coming together for SpaceX, for real, for the first time, rather than, “We’ve got internet satellites and spaceships.” It sort of made sense, but when we’re harnessing the entire sun—because it’s pretty doable, because we’ve built a lot of it already—it’s starting to sound cheap at $2 trillion. Who else can harness the sun?

Jason Lemkin

Just to push back a little on one word: I think it does all make directional sense, but I’m just going to call you on the word “mock,” because I didn’t mock.

Yeah, I was very clear I didn’t say mock.

Jason Lemkin

Right. I think you’ve just got to say, look, someone can be 10 times more accomplished than you as an entrepreneur and a human being, but when you’re investing money, you’re still entitled to say, “What probability do I ascribe to that 10-times-more-accomplished person being able to do the next thing?” Therefore, you have to look at this and say, for how long will this be supported by a future statement, and when will it be worth something on 20 times free cash flow?

But you know why it’s interesting if you really believe in DCF and free cash flow for real in the public markets—and I still get confused. If Starlink really has 53% profit margins and is wildly profitable, the fact that this extends the Starlink vision 2 to 5 orders of magnitude is actually a reason to say, “Hey, if I believe in this at all, my DCF has gone up.” How much, I don’t know. It’s gone up because Starlink is so profitable at scale—jaw-droppingly profitable, right?

Jason Lemkin

Two comments. First of all, big picture, you are correct. The reason Elon can do it and no one else can is that he’s going to articulate these big step-function stories. One of the investors in one of his companies pointed out to me—it’s a great point he made—that they’re not like software companies that incrementally grow every year. They’re kind of step-function technical challenges that you accomplish maybe every 5 or 7 years, and then you harvest on that while you’re building the next step-function challenge. Then that gives the next lift.

I think Starship’s a great example of that. You had, “Hey, I launch rockets and all I do is get government contracts.” Then you’re like, “No, I launch rockets, and now I have a cellular service for remote areas.” The next turn of the crank is maybe, if I can get Starship working, you can have cellular everywhere and data centers in space.

So, you are right. These are big, chunky visions, each of which, if realized, gives you—just to pick a number—$100 billion, $200 billion, or $300 billion of net present value.

But you’ve still got to go back to what’s the probability of it happening, when does it happen, and what’s your cost of capital between now and there, right?

Yeah, and I would just argue that if you’re the classic optimistic Wall Street analyst, you can probably justify the $2 trillion valuation by saying the odds that this occurs are 80%, but we’re describing only a 30% chance it happens on time. There’s an 80% chance that within 5 years it achieves similar profit margins to Starlink. You roll it all back, and you can justify $2 trillion over $1.x trillion, right? I think you can do it on a spreadsheet, and that bet will be available to you. Have at it.

6. Jeff Bezos' $100 Billion Fund: The End of "Doing It the Hard Way"

Well, in the week of bold $100 billion bets, why are we doing seed-stage investing, Jason? Jeff Bezos seeks $100 billion to buy AI and transform manufacturing. The Wall Street Journal broke this one: Jeff Bezos is raising a $100 billion manufacturing-transformation fund to acquire companies across semiconductors, space, and defense, inject AI into their operations, and make them much more efficient. He’s apparently been touring Singapore and the Middle East to charm some sovereign wealth funds into giving him the money. How do we think about this? Again, it was another week of, “I feel irrelevant at early stage.”

Jason Lemkin

I think it’s a great classic Indian Creek Island investment. So, you’re sitting in Miami in your couple-hundred-million-dollar home. You’ve got Jassy and team running the hard business. You don’t have to do that that much. Luckily, they’re doing the hard work, and now I get to think big.

I love you, Jason.

Jason Lemkin

I get to think big at Carbone on the yacht, and I don’t want to go small anymore. I’ve already done it. I’ve already built Amazon. So, what do I want to do? I’m going to remake some industries. I was with my friends at Pura Vida getting our smoothies, and we’re all going to remake industries. This is the Indian Creek Island bet, and I get it, right? You don’t want to screw around anymore in the billionaire’s bunker. You just don’t want to.

Do you think he’ll be able to raise $100 billion?

Jason Lemkin

He could just sell stock. If he wants $100 billion, he can get it himself. So, I’m sure he’ll get some significant slug of capital. It’s hard. I don’t know about $100 billion. It’s so out of my pay grade, but the only problem is SoftBank seems tapped out. They’re hitting their debt limits. They just announced this week, right, that they’re flashing above their covenants. But I don’t think he would announce it if he didn’t believe he could do it, right? So, you’ve got to announce a decent probability.

I want to come back to what you said, Jason. I thought that was actually very insightful, the Indian Creek comment, right?

Jason Lemkin

You do see this: I did it the hard way. I’m now 50, I’m 60, I’m not 22 anymore. I’ve got more money and less time, so I’d like to insert myself further along in the value-creation process to make it happen quicker. That’s the logic.

You’re right, because I was reflecting back: AI is this cool new technology that could transform all kinds of industries. Just like 20 or 30 years ago, the internet was this cool new technology that could transform a whole load of different industries, right?

When Jeff Bezos was starting out, there were 3 different plays you could make. You could say, “Hey, the internet’s going to transform retail. Let’s focus on retail. I should build software and sell it to retailers so they can move onto the internet. Build Shopify.”

The second thing you could do is say, “Hey, the internet is going to transform retail. I should buy Walmart, because I’ll kick ass and I’ll make them become an internet company, and I’ll do it that way.”

Or the third thing you could do is say, “I’m going to do the hard thing for the most amount of money. I’m going to transform retail myself by building full-stack retail. I’m going to call it Amazon, and I’m just going to kill everyone.”

The last one was, it turns out, the $2 trillion opportunity, right? From zero. Your IRR is from effectively no money, and you make a couple of trillion bucks. From a value-creation perspective, Shopify is roughly a couple hundred billion dollars, because that’s the best e-commerce technology provider. To be fair to Walmart, if you had half a trillion dollars lying around at the time, you could have scored a double because it finally adopted the internet.

Jason Lemkin

And you make a 2x on a lot of money and make half a trillion bucks. Because now Walmart has a market cap, plus or minus, of $1 trillion, and they're very much a winner in the internet age. I was just thinking, those were the 3 games that you play. Back when you're 25 and you have incredible drive, you don't have half a trillion dollars lying around, so you do Amazon.

But you're exactly right. If you're in Indian Creek and you're like, “I don't have 25 years of working out of a desk. I'm like a 2x on $100 billion—taking $100 billion and buying a bunch of companies and injecting AI into them, like I could have injected the internet into Walmart. Maybe that's the play.” It's inherently less disruptive and more financial engineering than doing either the Shopify play or the Amazon play. So, yeah, I agree. I like the framing: it's what you do when you have too much money to want to do it the hard way.

If you talk to billionaires today who aren't pulling their hair out because they're running public SaaS companies, the vibe is similar. They want to do something huge in AI right now. They don't necessarily want to run it themselves. They don't want to be CEO again, right? But they're very motivated to do one of these plays.

We're going to see a bunch of these plays. It's what everybody wants to do. It's logical. It's logical from a billionaire's perspective.

I love seeing Sergey Brin rock up to a random hackathon in Miami. I'm not sure if you guys saw this, but in Miami, he came out at the end and was a judge, standing onstage at a very grassroots hackathon in a random part of Miami. Maybe it's the only place left in the country hospitable to billionaires, so we'll watch it accelerate.

Jason Lemkin

Oh my God, the oppressed species of billionaires.

When I think about this, the New York QSBS and other stuff—we don't have to do it. What I do think, when you saw Sergey there, is that he didn't even move to the billionaire bunker; he moved to a different part of Miami Beach. I do think what we're missing is that it may be the only place in the U.S. over the next couple of years that is welcoming billionaires. Texas does, but even Austin doesn't. Austin has mixed views.

Say what you will, and I know it's controversial, but Florida welcomes billionaires. Look at the OnlyFans guy. It's very controversial subject matter, but where did he live? Pompano Beach, Florida. Florida said, “You be a billionaire here,” okay? We're underestimating that it's not just taxes. With Sergey, it is that you may only feel comfortable if you're Bezos or Sergey in Miami soon.

Why would you feel comfortable in California, Washington, or New York? Seriously, why would you feel comfortable? Maybe Utah. You just don't want to be attacked constantly. No human being wants to feel that way. You want to go where people will let you just be yourself and raise $100 billion or do whatever. “Leave me alone. Let me live.”

That's what I think people are missing with the Golden Goose. They're making billionaires uncomfortable, and it's not just the money. It's being uncomfortable. That's Sergey. He's gone. He ain't coming back except for staff meetings and hackathons in Mountain View.

I'll tell you, yesterday I was here in Utah, and I saw Ryan Smith. I love him, right? Founder of Qualtrics, okay? People love Ryan. He took his money and bought the Jazz, as well as a hockey team. He had a rough first year with the Jazz because he traded some top players, but they love him here. They love Ryan here, and I don't think in the Bay Area anyone is beloved; the billionaires are vilified. Why would you stay? It's only going to accelerate.

Here's the thing: it's hard to predict the outcome. I really don't think adding QSBS to New York is going to lead to the exodus that people wanted to see. They did it in California, and that alone didn't work. But when you are uncomfortable living somewhere, you leave. Ryan Smith is beloved here in Provo, where I am today.

If Sam Altman is vilified, and Dario gets a pass, maybe because he gave away 80%, who wants to live where you're vilified? Who the hell wants to live there? Cut poor Sam some slack. He apparently has no OpenAI to give away.

Jason Lemkin

Well, Mark left. I mean, he's old, but he did leave.

Speaking of the world hating billionaires, Grok announced essentially the debrief on the $20 billion deal with Nvidia. It had less than $100 million in ARR when it was acquired. Jonathan Ross, the founder, is going to make about $950 million off what will be a double-taxation structure. It's quite costly in terms of the taxes on the IP and team acqui-hires that we've seen.

How do we think about it? Chamath also reportedly made $950 million, to which you responded on Twitter, “He made much, much more.” How do we think about the analysis of this breakdown?

Jason Lemkin

I think there are 3 different things you said we should talk about, and I think you should break them apart. One is: how often do you see these kinds of hot, sub-$100 million ARR businesses going for this kind of value? The second thing is, what was the structure, and why does it result in double taxation? Then maybe the third thing is, why does poor Chamath need to tell us he's rich all the time? It's okay, we believe you're rich. You may have made other people poor, but we stipulate that you are rich. Even you are smart. You did a great deal here. Let it go, poor guy. Therapy will help.

Let's go back to the first one. The first question you asked in the notes, which you always ignore when it comes to the show, is: when do you see these kinds of transaction prices where revenue doesn't matter? When does someone pay $20 billion for $100 million in revenue?

I was thinking about that. It's easy. The answer is when the value to the acquirer is so high and they have the market cap to do it. Nvidia, with a $5 trillion market cap, can pay $20 billion for something that's valuable.

There is another good example I thought of. At scale, it happens a lot at small scale. There is loads of tech M&A where some shitty little company is doing less than $1 million in revenue and someone buys it for $100 million, which is a 100–200x revenue multiple. But we don't make a noise about it because it's just so small. The reason they're doing that is because they can run it through their channel and convert that $1 million in revenue into $20 million, $30 million, or $40 million very quickly. Or it has strategic value to them.

The number of times it happens at $20 billion is low, but WhatsApp is the other great example of that. Facebook paid $16 billion for WhatsApp, and it didn't have a dime of revenue. It was a great deal, and it's still a great deal. So it does happen, but there are only a few buyers who can afford to do that.

The Grok one—they just announced at GTC last week that it's going into production, so that's different. I'll tell you what I find interesting in general for venture. They had $100 million in revenue; they proved the concept and proved it sort of works. Jensen Huang said, “Within a year, we can get this into production.” That's worth billions, right?

It illustrates how weird M&A is. This deal probably was a multiple of the last round. The last round was at $6.9 billion. Classically, rounds would be 2–3x the last growth round. That has collapsed for some reason recently, but my second startup was acquired for exactly 3x our round. Literally, our acquirer downloaded our certificate of incorporation, found our per-share price, and showed up unsolicited with an offer at 3x. That's what I think happened here.

What is so weird, though, is that it's a reminder that so much of M&A, one way or another, is focused on revenue multiples, either directly or potentially, and then they're just abandoned after the deal. You're worth 10.2x ARR, and then we fire the sales and marketing team after the deal closes and roll it up into our core product. It's an odd thing—necessary, but odd. There are so many weird ways they are valued.

Agreed, and yes, it's a weird thing because there's often a huge gap between what you're worth. What you're worth standalone might be $2 billion or $3 billion; what you're worth to the acquirer might be $20 billion. It's a question of how that $17 billion of value gets allocated.

Obviously, sometimes the buyer is trying to grind you down to $1 more than your standalone value. Then sometimes, like in this case, they're like, “Hey, we will pay you a fair amount of what it's worth to us, which is way more than you're worth on a standalone basis. But in return for that, you're going to use this structure, which is wildly tax-inefficient, but it's the only way for us to get this thing done quickly and without government review.”

The other point you're trying to make is that no one's going to cry for someone who made $950 million. They're not even a billionaire, so no one will hate them. That's great. I mean, it's actually a win.

Good. Another $50 million, and, Jason, the pitchforks would be out.

Jason Lemkin

But these transactions are very capital- and tax-inefficient, because what happens is the company sells the assets to NVIDIA, books a gain because the assets were on the books at less than $1 billion, and now they're getting sold for $20 billion. So you have to pay tax at the company level on that, and then you dividend or redeem the money out, and individual investors have to pay tax on the gain.

So you're probably wasting plus or minus $4 billion or $5 billion on a $20 billion transaction. And for the founder, Jonathan, it's roughly a 60% effective tax rate. You can't hold NVIDIA stock. You're cashed out. There are so many inefficiencies in this deal.

One, you don't even get the IP, right? You don't get the company. You don't get anything. And it's a 60% tax rate to the founders, with no ability to roll over the stock. Avoiding antitrust—is it worth it? Probably, for the $950 million. But, man, this has got to be the most inefficient, convoluted thing ever to avoid antitrust. It's so expensive.

And it is worth pointing out: it's quite terrifying from a government perspective. You now have a process whereby the government makes the rules that enforce antitrust, and you've basically got 2 choices. It would appear you either lobby extensively at the highest levels of administration and get a waiver from the top down. Read that Wall Street Journal article this week, which was pretty good.

Option B is you pay double. You do it this way, and then you pay double taxation, and the government wins either way. You know, wire me the money and I'll let you off, or wire me the money after it closes, but wire me the money either way. It's a really perverse incentive.

Whoever comes into the antitrust division next time and says, "I think we should clean up the rules and make them much more transparent," it's probably going to cost the government $20 billion or $30 billion in terms of some combination of kickback and tax avoidance. There you go.

I have to say, I do like it. Jonathan was in the desert in the dark for many years, and he is a founder who's been a real resilient cockroach who's gone through the hard times. He's gone through the criticism. He's also just a good dude. I like him a lot. So it's nice to see good people win.

Jason Lemkin

Yeah, I know. I just like Chamath, another good guy who's been through the wilderness and has now got his $950 million. So there, I'm sure that's the point you're trying to make, Harry. Much more.

7. Figma's Death Spiral? Why the Markets Are Terrified of AI Disruption

He needs money, Rory, okay? He can now buy some more Loro Piana. Rory, you brilliantly said to me, "That's great, but what about me? I own stock in Amazon and I own stock in Figma, okay? Google launches Stitch. Figma tumbles."

Figma tumbles is an understatement. I saw a tweet—it was actually an announcement that someone posted—that Sequoia were buying $35 million of Figma stock, and I was like, "If Sequoia and Andreessen Horowitz are buying $35 million, I'll put in some of my money. That's a good sign for me." I'm down 22%. Twenty-two percent.

Yeah, it's $21.66 a share. I remember, look, when it was at $108 and you asked me what I thought it'd be, I said $35, and I was wrong. You laughed at me because I was so pessimistic, and I was wrong. It's $21. Wow. No, does this have a floor?

Jason Lemkin

Of course it has a floor. I mean, stop. It has a floor based on its cash flow, which is strong. It has a floor based on the growth rate of that cash flow. The third element is the probability of disruption, right?

The tricky thing about the equity business in the short term is everybody gets to speculate on that probability of disruption. So you take the cash flows and, if you love the story, you apply uplift—see Elon for details. If you're really scared about the story, you apply terror downlift, which is what's going on here.

Now, in the end, you said, "Is there a floor?" In the end, if the business is worth it, it will grow and it will generate the cash. They'll get to do what the Palantir guy does, which I so love: every earnings call, he basically slams all the haters and says, "Basically, you—I'm making money, and you all were wrong."

In the short term, the market gets to moan and have its opinion, but the floor doesn't come because the market changes its mind. Maybe it will, maybe it won't, but that's not really in your control.

If you're running Figma, the floor comes if you execute, demonstrate that you've been able to adapt to an AI-first world, generate the growth, and generate the cash flows. Eventually, it'll come. But that's why drawdowns are terrible. That's why drawdowns are hard.

Well, on the Stitch-Figma thing, I mean, I've used Stitch, of course, as you would imagine. I think most folks that chimed in on this never used it. I'm confident 98% of them never used it. So, for those who don't know, what is Stitch?

Jason Lemkin

A new design tool that Google launched, and the bar at Google to launch a new AI tool is pretty low. They try a lot of stuff, and they abandon almost all of it because then they focus on a few core products.

So you literally cannot take it seriously when Google launches a product, because you have no idea whether they will stick to it. They launched a Sonos competitor—is it Sonos? Sorry, I keep getting this one wrong. The audio one, right? What's the one—the audio one? Oh, it's Sonos. It's Sonos.

Yeah, so Google launched a Sonos competitor. It's cool, but it's not nearly as good. It doesn't really work. And will they keep with it for 5 years? I'll bet you dollars to donuts they don't, because it's not core, okay?

I used Stitch. I've used all of Google's design products. I think the odds that they decide to build a Figma competitor from this for a decade approach zero.

Jason Lemkin

On the one hand, massive market overreaction to a proof of concept. Give me an F-bombing break if I'm Sequoia or whatever. On the other hand, the markets are saying, "We are extremely worried about disruption. We are extremely worried about disruption. You better prove to us, Figma, Atlassian, Salesforce, that you are ahead of disruption and not behind it."

The market said, "We don't believe it." And I agree with the markets here, because Figma Make is one of the worst products I've used in the last 6 months. But Stitch, on its own, is at least better than Make. At least it can take context from a website and not hallucinate.

But the market should be like, "Show me the money." Again, where is Figma's $300 million of revenue from disrupting Replit and Lovable, like we talked about before? If you haven't delivered like Palantir or started to deliver like Salesforce, the markets are going to freaking panic that your revenue is not that durable.

This is what I think it all—like, I didn't get the 2026 panic for a long time. I was slow. Now I get it. The markets are rationally saying, "We no longer believe this revenue is particularly durable."

SaaS—old SaaS people—PE doesn't believe it. Qualtrics couldn't finish its debt offering this week. Salesforce barely got its debt done. And why? The markets just don't believe that Figma is a bad company. They just don't believe this revenue's going to last a decade anymore. They don't believe it.

So you're going to see more of these panics for anyone not accelerating. They're just going to panic every time it happens.

Jason, we just released a show with the CRO of Figma, and I asked him how he's seeing AI implemented into your sales teams. He said, "Honestly, we don't really have that ability, and we haven't done it yet. Oh, and we're hiring a lot more in sales, by the way. We're not reducing headcount at all. We're not seeing that. And, oh, by the way, we're not seeing pricing change at all."

How do you feel when you hear that? Because you're just serving up the—

Jason Lemkin

Ironically, Harry, ironically, that one does not worry me as much as the product. I will tell you what I've learned. You and I—you formerly on the other 20VC, me and Fari—I know a lot of the CROs and CMOs at leading AI companies.

The ones I'm close to are pretty good, but I also see tons of folks I call recycled mediocre. They are folks that bombed out of old B2B companies that barely did anything there, but because they have the right logo, got hired to a super-hot AI company. They're all over the hot AI companies. So many recycled mediocre people.

They're going to hire 250 reps and not train them. They're going to build infographics. But the products are so strong and the demand is so strong, it doesn't effing matter. If Figma Make was so great and their AI product was doing $500 million, you could sell it with folks fresh out of a nontechnical junior college. It'd be fine, right?

My point is, even though I talk a lot about AI go-to-market agents and I believe they're great and they work, they don't fix product-market fit. Sales tools and sales do not fix product-market fit. We're seeing broken—we're seeing decaying—product-market fit. That's why the market panicked on Figma.

They're seeing hints—just hints—of decaying product-market fit in the AI era, and you should panic.

Yeah, I'll just pile in on that, because I actually have these conversations with my companies all the time. I totally agree with what Jason said. When my companies come and say, "Hey, you've got this product. We want to talk about AI," and they say, "Well, hey, look at us. We're using AI in go-to-market," or they say, "Look at us. We're using AI to build engineering," I'm like, "That's great, but nobody gives a rat's ass."

Rory O’Driscoll

That’s like jack shit. That’s not solving the core problem. The core problem, unless you’re making something like cars where it doesn’t matter, is back-office efficiency. If you’re a software company, the number-one question is, “How does AI change the end product you deliver to your customers?” That’s what’s going to determine success or failure.

So I thought I’d have fun. I was wrong to pile on to that guy and kind of bludgeon him while he was down because I thought Harry had served you up a softball for you to say, “Hey, the idiot is not using AI,” and I argued with you, but I think you nailed it. Look, you could be using AI well or badly in go-to-market. You can be using AI well or badly in engineering. It will catch up with you over time if you’re not using it well, but that’s not what’s driving 30–40% price declines.

What’s driving that is exactly what Jason said: the market looks at this and says, “There’s disruption risk here. I don’t know the terminal value here. I’m nervous, so I’ve got to be paid for that risk.”

And he didn’t take my softball, did he, Rory?

Rory O’Driscoll

No, he didn’t. Normally, he’s so disciplined.

Jason Lemkin

He did. But Rory’s got the important point. You’ve got to have the right AI or the company’s going to decline. Listen, would I— I’m sure the Figma guy is great, but honestly, if I interviewed a CRO today and they didn’t have any AI agents that he or she had brought in, I would recommend to the CEO, “Don’t hire them,” okay?

For sure. But would I be much more worried than if I talked to a CTO and the new CTO they wanted to hire didn’t really believe in using agentic coding? Then I would ask for my money back. Can I have all of my investment back at 1X? You can keep your markup. Just give me my $5 million back.

I’m going to go one level more than that. You’re right: not using AI in go-to-market is bad. Not using AI to build the product, maybe I want my money back. But if you’re a software product and you don’t think AI is going to disrupt not just how you build but what you build, then you actually probably want to actively short it.

For example, they’re not doing that, but if Figma or Salesforce or someone was to say, “I don’t think AI is relevant for our customer base, and they don’t want to use AI in design,” if someone was to take that position, you’d be like, “Oh my God, you’re just going to be left behind now.”

Figma’s insensitivity to how mediocre Make is really worries me.

Rory O’Driscoll

That’s actually an interesting point. If Dylan or the team said, “Listen, Make isn’t good enough, but give me time. Give me 6 months. It’s going to be great,” that would be one thing. But saying this product is the worst vibe-coding tool I have used in the last 6 months, the fact that there’s no public awareness that this is an issue really worries me.

Jason, can you help me understand that? Dylan is a good CEO. I don’t know.

Excellent CEO. Well, yeah, so help me understand this, then.

Jason Lemkin

Listen, we don’t know for sure, okay? But I think everyone at scale has a trap, which is that their installed base is a trap. It is an opportunity and a trap. It is the greatest thing in the world to sell your agentic product—to sell Agentforce—to the $44 billion of ARR. It’s the greatest opportunity because you don’t have to earn that base, right?

But it is also 50 years of debt, 50 years of features, 50 years of offline integrations, and non-agentic apps they want. It’s 50 years of endless work, and if you’re not careful, it will consume 98% of your resources—that installed base, right? They need so much attention.

And I do think, literally, Make is the only vibe-coding product that I have used where you say, “Build me a website for 20VC. Use 20VC.com as a template,” and it can’t go to 20VC.com and pull the context to make the website. Now, Replit couldn’t do this in June 2025, Lovable couldn’t, but anybody can do it today, right?

That includes—who could do it pretty well was Stitch. Stitch got the context of the design projects right. I get it. And that just shows me whoever’s running it doesn’t care about these use cases. They just don’t care. It is falling behind, and my guess is there’s a small team on this. They’re still growing, what, 35%, right? I mean, this is one of the best out there.

But here’s the trap, and we all have portfolio companies like this. The trap of earning that 35% can imperil your agentic growth. It can consume more than 100% of all the product, engineering, and customer-success resources you have, right?

Even Mike Cannon-Brookes, when he was on the show, alluded to it, right? Then they did the layoffs. He’s like, “I’ve got to get these resources because otherwise Jira and Confluence and everything are going to suck it all up, and I have no people.” Super interesting. It’s a trap. It’s a trap.

Think of some of your portfolio companies at scale, north of $100 million. They don’t describe it this way, but you know it’s a trap, especially when they have a mediocre VP present at the board meeting and the mediocre VP is like, “I want to do it, but I don’t have the people. I need another 700 people to build that.” You know, like, “Ugh, you’re trapped.”

Rory O’Driscoll

It’s hard to say, “I’ll allocate to the new thing first and allocate the residual to the old thing.” Your instinct is to say, “I’ll deal with the old thing, and then I’ll find some people for the new.” It’s a mind shift. That’s an interesting point.

I don’t want to overstate the Intercom example just because it’s enough already, okay? But if you want to use it as a case study, it does illustrate what you have to do. No one was clear: we let our core business go into partial decline.

It’s very hard, almost impossible, for a public company, though. It’s lucky they’re private because you’ve got to have a lot of guts to say, “We’re going to let a $1-something-billion design business decline a little bit so we can build our agentic product.” It’s tough when you’re public, right? But it’s what you’ve got to do.

Atlassian is at $21 today, with a $12 billion market cap. Would you buy it today? No. I don’t see any evidence of these agentic investments working. I would rather miss out on the bounce right off the hard deck than invest in something that may fall below it.

The reason is there’s too much change. My God, there’s so much change. It’s so fast. Make would have been a great product in August 2025. It’s just not today.

The problem with the bounce—and it’s a first-world problem—is, say you buy at $21, the concerns are overdone, and you’re back up to $24, $25, $26. So you’ve made some decent money, but you sell it, you get ordinary income, blah, blah, blah. Net-net, after taxes and risk-adjusted, is it worth it, right?

The only way you buy is—and I’m not saying I have the answer to this—when you buy stocks because you want to hold them for 5 or 10 years. You just do better, especially tax-adjusted, than when you buy things because you think there’s a short-term bounce.

I personally think that almost all of this stuff is overdone, right? There probably will be 5%, 10%, 20% bounces across a number of these companies, but to want to buy it, I haven’t done the work that Jason’s done. But he is right: you only should buy things if you think 5 years from now they’ll be a winner in the future.

I think it’s totally plausible Figma could be, but Jason’s right: until you’ve used the product, until you’ve seen what they do, I wouldn’t make that decision, especially at scale.

For the record, that’s an excellent test. Yeah, you can’t—even Figma can’t charge for it. It’s a bad, very, very bad sign going into the middle of 2026. It’s a terrible sign. Terrible sign.

I’m naive here, so forgive me, but what is Notion charging for it? You’ve spoken before about Notion’s brilliance and the integration of their AI capabilities. Is Notion charging additionally for that AI capability, or is it just a superior product?

Jason Lemkin

Well, first of all, if I said I thought it was truly brilliant, I misspoke. I use Notion all the time. I think that they’re good. I don’t think it changes the game in the way other products do.

But the simple answer to your question is they have been able to—look, they’re not public, right? But if you combine what they’ve said with their pricing page and their announcing of acceleration and the fact that they have effectively doubled ARPU from it.

Here’s a way to look at an SMB product today: is your ARPU 50% or higher than it was pre-AI? It’s a really simple test. Can you drive ARPU up 50% or more? It’s very different at Salesforce or ServiceNow.

It appears that Notion has done that, in which case they pass the test, right? But I want to see 50% or more ARPU growth due to AI, or it’s just a feature. It’s just a feature, like an integration. Not great—hard work, good job, guys. Let’s have beers, but it doesn’t count.

Yeah, I mean, reacceleration is the aggregate test, to your point 2 weeks ago, Jason, right? Something’s got to be going better across the marketing KPIs. It can be charging for your AI product, it can be reacceleration, it can be ACV, but something has got to be working and getting traction.

Otherwise, you are falling behind, because the AI-first players are delivering utility to their customers, and that’s manifesting in those customers giving them money. So if you can’t make the same phenomenon happen one way or the other, you’re falling behind.

Jason Lemkin

Well, obviously, the market rejected Microsoft Copilot products as things they didn’t want to pay more for, right? That’s the example of stumbling. Notion with AI is $20 a month; the basic is $10. If everything ties—and private companies have an incentive to be honest, but polish the numbers a little bit, right?—assuming there’s at least a substantial amount of truth in what they’ve said, then it’s working, right?

They are able to charge $20 a month for their agentic product not because they splashed an AI label on it, but because it’s so much better that it is worth it to talk to your docs and to have stuff flow into your database autonomously. It’s worth $20.

I suspect we’re going to find this—I don’t love the pricing—as the year goes on. We’re going to see the same thing at Slack, because most folks haven’t used it. We’re on it. The AI version of Slack, for certain use cases, is so much better. I don’t love that they’re not just charging another $4 a month, but I wouldn’t be surprised if it’s a drama, if it’s like Notion, and they’re able to do it, right?

Speaking of Figma, I’m going to get a little off-piste here, but I think it’s why, Jason, we were joking about it on Twitter earlier. Mamoon announced raising the new Kleiner Perkins funds: $1 billion for the early stage and $2.5 billion for growth. Growth seems small.

I mentioned meeting him for the first time at SaaStr, Jason, if you’ll believe it, 10 or 11 years ago in a side room. He was kind enough to give me time. My takeaway is, honestly, you can’t do early stage, and you’re going to kill me, Rory, with less than $1 billion if you’re going to compete to lead Series As.

My reasoning for that is Series As now are $30–40 million, and if you want to lead them, you need to be able to write $25–30 million checks. I think you need 20 across a firm. So if you need 20 $30 million checks, you’re at $600 million. I would argue that your fund scale is as small as it could be to lead Series As today.

Rory O’Driscoll

I don’t entirely disagree, in the sense that, yeah, it’s a $900 million fund, and you’re exactly right. I mean, the math of what you throw out is correct. Series As tend to be—you’re writing an average of a $20 million check. Our reserves have consistently—

Jason Lemkin

I don’t even think $20 million is enough these days.

Rory O’Driscoll

Just an average, right? If you hustle and find some deals where it’s a little off the beaten track, I’ve written As where we’ve gotten 20% ownership for $12 million or $15 million, right? So it’s not all that, but let’s just go with your average, Harry, rather than arguing that, right?

We have 50% reserves. Over 6 or 7 funds, we’ve been at 50% reserves. So you end up with $30 million in the average deal.

Yeah, and then the last comment is a portfolio-concentration-versus-diversification question. I think, given the higher time to exits, you probably tend to be nearer to 30 than 20. So yes, I think there’s a certain fund scale required to play meaningfully in the Series A business.

Rory, can I ask you? You’ve got $720 million of investable capital. You can’t have a reserve of 50/50. You’re wrong.

Rory O’Driscoll

$720 million? No, I didn’t say 50/50. You didn’t pay attention, Harry. I said 50% of it, because there are 2 ways of expressing reserves. It’s simple math. Let me help you.

One way is expressing it as a percentage of original capital. In other words, if I put in $20 million, do I reserve $10 million? That’s 50% of initial capital. Mathematically, that’s the same as saying 2/3 of the money goes in on the first check, and 1/3 goes in on the second check. You’re right: I’m not reserving 50% of the total amount. Do you understand me?

I now do with that clarification. Thank you for clarifying that.

Rory O’Driscoll

You used the prepositions correctly the first time as well, but okay, it’s good to be clarified. But to be fair to you, people do express it both ways, and sometimes precision is important.

I always think of it as the amount of money you’ve put in, and then the amount of money you have in your back pocket to defend that money if you need to, or expand on that position if you want to. So I always think first check. And what it says is—and this is another insight—unlike for seed, where you also have a big fund, for that kind of Series A game, the Series A is not, for us, especially for a firm that just has checks for As and Bs of this size, an option value on putting a ton of money in later.

Most of your value is made on that first check, right? Whereas these folks who have seed or even A funds, where their real plan is to put $200 million in at the B or C when it’s working, to some extent the A is an option value. But just as, Jason, the seed check is the check, for us, the A or B—because we’re roughly about half and half—is the check. You’re not saying, “Oh, I’ll put in a little now and write more later.” That’s not the way the game can be played.

Rory, are you finding it increasingly uncomfortable with the expansion of Series A rounds because our checks are getting stretched bigger and bigger and it’s ruining our math?

Rory O’Driscoll

It is somewhat uncomfortable, but I think you have to find the deals where that’s not the case, or decide in some cases to reach. I’m going to put both sides of the challenge out there. There may be some deals that are too capital-intensive, and you decide, “I don’t want to do that because I’m just not getting paid for that risk.”

But you’re right: the average round—I looked at this math just last week. The average round that we play in has crept up, right? Not just the amount we do, but the average round that we play in has crept up over the last year or year and a half. So yes, it’s not to the acute level of, “Oh my God, we’re perfectly happy with the checks we’re writing and the round sizes,” but yes, there are times, especially—for example, we haven’t done any at scale of this—the kind of new AI lab seed checks—where you’re just so outclassed with your $30–40 million check that it’s barely worth playing.

How big is the Hummingbird fund they just announced?

Rory O’Driscoll

It’s $800 million.

$800 million? Where have they grown their fund size from over time? From doing inception investing?

Rory O’Driscoll

It’s $200–300 million, I think, of initial capital. It’s the same fund they always had, but adding a growth fund to a rounding error.

Right. And then didn’t Balderton lead the seed in Revolut?

Rory O’Driscoll

Yes.

And then who did they just do at $1 billion this week? The game does change. We were really good at doing Revolut at 4. Now billion-dollar rounds are a good entry point for us. It’s just a lot of change, isn’t it?

Rory O’Driscoll

Yes. Look, we track it on aggregate, right? Versus our 2010 checks, you’re probably up roughly 2x in terms of the valuation, and I think that’s significantly lower than the industry as a whole.

His point’s hard to argue with, which is, if you want to own 15% of a company, or whatever your target number is, and deal sizes have inflated, there’s basic math you have to do, right, with reserves. That math is, in many cases, highly stressed in the market today. It’s just a fact.

You used to solve for your fund size very easily before, right? It was $60 million per partner for seed, or maybe $100 million for A or B, and then it would go up a bit. The math was so simple that all these funds were—

But that math is broken today, and it creates interesting— I don’t find it super interesting, but it is a little bit broken, right? And so if you’re Garry Tan, you say the $100 billion outcomes make it irrelevant.

I don’t know how they work, but I was just looking and writing something up on Wiz. If Cyberstarts owned 4% at exit, okay, that model still works for a seed fund at 4% of $30-some billion. I don’t know whether that’s dilution or his model back then, but I think we’re going to see more and more of those: dilution over time.

We’ll see more and more seed investments eroded to 3–4%, right? And if they—I’m just doing one example—they’ll be eroded. If the exits are north of $10 billion, it’s okay for a seed fund, right? But it’s tough. I don’t have the answers.

I just think the old math was easy. We needed to do 15–20 investments per fund, with $15 million as the average check size, and you could roll up into fund sizes that didn’t make you fall out of your chair.

At the risk of being a dick, the math is easy: 4 times, because math is independent. You just don’t like the answer, right? Multiplying a pre-money of $50 million is no harder than multiplying a pre-money of $500 million.

Rory O’Driscoll

It creates more risk unless the outcomes are massively higher. We’re leveraging up our risk, to Harry’s point. We’re leveraging up our risk.

I agree. I was just being a mild little jerk because I think I’m getting grumpy.

Rory O’Driscoll

Exactly. It’s not that the math is complex; it’s just that the consequences are unpleasant. And it’s what you say. Everyone—we’ve said it over and over again—is pushed out on the risk continuum.

But it does feel like times are changing faster than fund sizes are able to adapt in a way that they haven’t done in previous cycles. Our normalization of billion-dollar-plus rounds, hundred-million-dollar-plus rounds—

Rory O’Driscoll

The hard part is that it’s happening so much quicker than fund sizes.

Jason Lemkin

That’s actually true. That’s actually a good reflection: the pace of change. I’ve lived through ’95 to ’99, the run-up and the billion-dollar funds, and I also lived through the unwinding, watching all the funds being unwound in ’02. I remember—and then, obviously, 2007, and even 2021—this feels faster than that, right?

The action on the table and the need to play at a high level—yes, I think that’s why every VC is stressed right now. No matter how successful you are, no matter how well your last deal, a deal that you did 3 years ago, is doing, everyone wrestling with doing deals now is grumpy, stressed, and feeling the pressure.

Rory O’Driscoll

But I think there’s one risk that’s under-discussed. All this fund math can lead to concentration risk or ownership risk. At least for a smaller fund, I have to deal with concentration risk or ownership risk. That’s the simple trade-off. That’s my math, right?

Jason Lemkin

Agreed. The one that we are just ignoring because times are so good on paper is: who the hell’s going to buy these companies if they don’t IPO? What the hell are these companies—likely Replit, Legora, Harvey, and Level—worth in the last round? $8 billion? $6.6 billion?

Okay, let’s be honest. Who the hell is going to buy them if they don’t IPO? You really think Google’s going to buy them for $32 billion? It could. Don’t get me wrong, it could happen, and acquisitions are not up, right? Dollars are up.

My point, which is under-discussed, is that we are doing these post-billion-dollar deals like candy. At least in the bubble of ’21, there were private equity exits. There were many exits out there. We have outstripped any current ability for these companies to have any exit.

You are going all in on the IPO without any worry, rhyme, or reason, and I think so many folks in venture are going to get their asses burned because they’ll end up in the dead zone. They’ll have great companies without great IPOs and zero M&A opportunities. I’m not saying don’t do it. I want my markups on paper. I’ve got to do an LP announcement this week. I want my TVPI and my IRR to look good, but it’s also terrible.

It’s terrible to raise north of $1 billion. It’s terrible because they’re just not Salesforce and Amazon. I mean, Grok is great, but a lot of the stuff is not Grok.

I agree with everything you’re saying, and what’s hard about you saying it is that the strategy that has worked has been precisely the strategy of doing those rounds above $1 billion and getting the round beyond it. But you’re right. It’s all predicated on being able to exit these companies in an IPO.

I will say, just to remind you, you can imagine a world of IPOs where some of these companies exit, but if the last rounds don’t have a block, it’s not necessarily going to be above the last rounds. There are some awesome AI application companies where you kind of go, “I totally see how that’s worth $5 billion in 3 years based on fundamentals,” and right now it’s valued at $10 billion. Three years from now, I could contemplate it going public at $5 billion, and you kind of go, that last round of $10 billion is going to have to convert and take 50 cents on the dollar unless it has a meaningful block, in which case it will get more of the company.

Rory O’Driscoll

Yeah, for sure. I just worry there’s some ratio—you can name it—of potential acquirers divided by unicorns, and I think we’re at the lowest ratio of our careers. Yeah, no, absolutely. The lowest ratio of potential acquirers divided by unicorns—unicorns plus up to decacorns. They’re just not there. We’ve lost all of private equity, and Nvidia isn’t buying 100 companies. Microsoft is not buying 100 companies.

So, what is the subsequent thought then, Jason? Sell aggressively into secondary markets, which are more liquid than ever?

Jason Lemkin

Well, probably for early-stage investors, sure. Sell, sell, sell—that’s the Hopin lesson. I know you love to talk about it, Harry. Maybe that’s the lesson.

I don’t think it is. Can I just push? I don’t want to use the Hopin example because then people would say, “Well, my company’s not Hopin.” No pun intended. I think you’re saying that even for your great companies—excuse my language—great companies where you know, in the end, they’re not going to be Hopin, they’re not going to vanish in a puff of blue smoke, you’re still right: those last rounds might be at prices you’ll never see again.

What should you do as a seed investor? What would you do if a company where you invested at $50 million, you thought it was worth it and could be an M&A at $2 billion at some point in time, suddenly got a round at $5 billion of new money, and there was a secondary opportunity? You’re a starter for 10—what do you advise?

Jason Lemkin

Yeah. I mean, listen, maybe just as we’re now okay with down IPOs, we’re okay with it—it’s just mathematically true.

Yeah, we are.

Jason Lemkin

Maybe we’ll all be chill with down M&As, and all these folks raising at $9 billion and $10 billion can sell for $1 billion or $2 billion in a few years, and everyone will be happy. I’m just not so sure that they will, and I’m more worried there’s just not going to be a lot of acquisitions.

They’re very specific. They’re specific to either being desperately behind on AI or needing to jump forward years. That is a very narrow subset versus making sure accounting software at $5 billion is better for accountants. I believe that’s a great market. I just don’t believe the hyperscalers are going to buy these companies.

[Speaker?]

No. And yes, because by definition, at the application layer, you’re not getting the hyperscalers. The odd thing is—I should just realize this for the first time—the odd thing is, because of this story of “eat the work,” the TAM is meant to be larger. That’s the whole point of these application players, and I believe it. But, Jason, to your point, what it means is you are definitionally therefore the new company is larger in terms of value than the old company you’re replacing, which means the old company can’t afford to buy you.

Yeah, no chance. If you’re Harvey and you’re worth $10 billion—and, well, it’s not really; it’s worth more—but if the old practice-management legal software is worth $2 billion, they’re not going to buy you, right? You’ve got—it’s basically win or die.

To your point, I think you’re exactly right because these TAMs are bigger. When the TAM becomes bigger, the new company gets marked up and has more potential to be a bigger outcome. But what it means is that once it gets marked to that bigger outcome via a late-stage round, it precludes the prior generation from buying in.

If I was running one of those prior-generation companies, if I was the system of record, what you need to be doing is buying the not-marked-up, $500 million-valuation company and getting something out there, right? But no matter how hard you try as the old-generation legal software company, you can’t afford to buy Harvey at $10 billion. So, it doesn’t matter what you want: they can’t afford to buy them, and they can’t afford to sell to you.

Rory O’Driscoll

I think you’re exactly right, Jason. I just wish—listen, even if IPO markets are barely, barely functional, they’re open, but they’re barely functioning. If the M&A markets are on fire, I’d be good with everything. If we were turning around and these companies were being bought for $10 billion in cash each week, I’d be like, “This is awesome, man.”

But it does worry me a little bit that it’s so easy to get a $9 billion valuation. It’s so much easier to get a $9 billion valuation than a $1 billion exit, which should be terrifying if you’re the people giving the $9 billion valuation. You’ve got to play the game on the field, but sometimes you can stick your money in your pocket and not play the game on the field. That’s okay.

But you’re playing the game on the field, Rory?

Rory O’Driscoll

Part of it. The bits I haven’t played have been—we haven’t done any deals at $1 billion, but we kept it at $100 million to $300–$400 million pre-money valuations. Right now, that’s been the wrong play, just to put it out there, right?

I mean, look, I give all credit to, for example, Spark’s Yasmin Razavi, who broke all the rules and said $4 billion for a pre-revenue company. Anthropic—that’s paid off.

You did it for Matt Murphy at Menlo. Matt has—

Rory O’Driscoll

Absolutely. Matt did it, too. Totally, yeah. So, my point is that some of the late-stage bets have paid off. On average, I think it’s a tough one. I share the concern. To your question, I’m playing the game on the field. We’re playing part of the game on the field, but we’re a bit scared of certain parts of the valuation curve.

I think that will probably, across the cycle, be the right decision. I'm not sure in the last year or two it's been the right decision, which is always the challenging thing about valuation. I shouldn't be so honest, but I think we'll see at the end of the day here.

My biggest regret with our Series A fund is not being more elastic, disregarding the Series A mandate, and just saying, “I'm going to leverage the brand and the access that I have”—arrogantly, forgive me for that—to get into super-hot companies, like your ElevenLabs, your Legoras, your—you-name-its, your Lovables, much earlier, and being a momentum investor in a hot environment.

I'm going to paraphrase what you said. My current plan is to write $20M checks in Series A investments, where I get 10% or 15% to 20% ownership, right? And I can write 20 of those checks. In retrospect, you're saying I should have taken 5 of those checks, the full $20M—not a bitty check, but a full ticket—or maybe even doubled down and used a 2-ticketer, and done some deal at $1B pre-money or $2B pre-money, because those deals with momentum had even more momentum.

The interesting question is—because I wrestle with this a lot, so I know it's an interesting question—if your mandate is to do 20 Series A investments where you get 10% to 20% ownership and you stick to your knitting, right, should you have taken 5 of those slots, used the same check size, or maybe even a larger check size, and done Series D and E rounds at $1B or $2B pre-money and gotten 2 markups already, right?

Rory O’Driscoll

And the truth is, in some cases, not only would you get markups, but you would get returns commensurate with Series A, but at much lower risk. Anthropic will be the definitive version of that forever—not actually the round Spark and Menlo did first, because that really was risky, but the rounds at $14B and $60B were risk-adjusted freaking awesome.

Now, back to your point, I wonder whether all of those rounds in those other AI app companies will be as good, for precisely the reason you articulated, which is that maybe the TAM doesn't support that exit. But right now, Harry's right. You look at Lovable at $6B. Maybe you could have done it at $1B or $2B. You'd have a 3X step-up with no hassle, no fear, no early stage.

The sweet spot of investing—I mean, it's something you said, Harry, and it was true but tautological. You said, “I wish I'd been a momentum player in a rising market.” The definition of a momentum strategy is that it only works in a rising market, right?

But yes, in the market we've been in for the last 3 years, it's hard to distinguish momentum players from very shrewd players. Both of them have worked out really well. That's the kind of step back for me. I don't think shrewd players have, actually.

That's my point. I think shrewd players have appeared disciplined, remained thematically diverse—like you and me, Rory—and remained with our high ownership targets. Momentum players have absconded with prices, gotten fast markups, and, in a lot of cases, DPI, because that buys you an insane price.

Rory O’Driscoll

Brutal comment, Harry. Then maybe you and I aren't shrewd. Look in the mirror.

Right? Very possibly. And that's the self-reflective question as I introspect, which Marc Andreessen doesn't do, clearly.

Rory O’Driscoll

Yeah, I know. Introspection here, please.

No, that's the ultimate question. Look, the truth is that is the question everyone wrestles with: when do you stick with your strategy, and when do you break it, and what's a good reason to break it?

Rory O’Driscoll

Absolutely. No, we all agonize with it. And that's why Jason and I said that the kind of trivial breaking it—just to be silly—isn't that interesting, but sometimes breaking it could be the right strategy, and that's the hard one.

There we go. Okay, boys, it has been a pleasure. Thank you so much, as always. Jesus, that was quick. We ran out of time.