Groq与NVIDIA的200亿美元交易｜为什么Sam Altman不在乎稀释，以及2026年的“隐形失业”

Elad Gil

Everyone's coming for NVIDIA now. NVIDIA's numbers are going to crush this year, but we're going to see all the daggers really coming out.

My word, what a schedule we have for you today. Groq acquired for $20 billion, Manus acquired for $2 billion, and so much more.

Elad Gil

In the end, words are words, and half a billion dollars is life-changing, right?

For venture, this is the era of the spite startup.

No one ever said to Winston Churchill, “Congratulations, you won World War II on budget.” They just said, “Congratulations, you won World War II.”

Over the holidays, my Claude named itself out of the blue. It named itself Ren. I didn't ask it to. It named itself Ren.

When most people believe their AIs are alive, you will take it with you 24/7.

I do genuinely think you can identify a top 0.1% founder without talking to them. And I've done multiple billion-dollar exits from cold inbound.

Ready to go. Guys, it is so good to be back. So much happened while we were away. I want to start with one of the most prescient: Groq being acquired for $20 billion in cash. It happened just before Christmas. Chamath obviously came out as one of the big winners. The price is 3× more than the last round price. How did we analyze this?

I just had 2 thoughts. One is, I was thinking a lot about both OpenAI buying up basically all the world's global RAM supply and Greg Brockman this week in the new year talking about how we will all be running 24 hours of inference by the end of the year. Not all of us, but a subset of tech workers, of knowledge workers, will be running AI 24 hours a day.

I'm already up to a couple of hours a day. I'm running AI myself. And so it's a world of inference, I think. When we started this podcast, we talked about building models and LLMs and all of this, but GJFC, if enough of us by the end of the year are running AI, maybe even 48 hours a day, 72 hours a day, right? Multiple agents running 24 hours a day, inference is all of the growth.

And if Groq is even part of the answer, part of the existential answer for NVIDIA, it's worth it. Everyone's coming for NVIDIA now, whether it's AMD, partnering with Broadcom, or building your own chips. NVIDIA's numbers are going to crush this year, but we're going to see all the daggers really coming out. We just saw the deck chairs being rearranged in 2025.

So I don't even know how great Groq is, but if it can possibly address this, it's worth taking out.

Elad Gil

I agree. I think a couple of things—Jason's comment on inference is key. Broadly speaking, in the world of GPUs and TPUs, there are 2 big-picture tasks that every one of these AI companies has to do. You have to train your model once, and that's training. Then you have to run your model. Every time I submit a query or a question, you're running your model, and that's inference.

The tasks are roughly similar, but they're not quite the same. NVIDIA is totally adequate for both. It's awesome for training. Groq had a particular edge for a certain kind of inference where it was very low-latency and very deterministic—in other words, where you have to predictably deliver low latency.

What we're seeing now, to Jason's point, is that as you live in this always-on AI world, it's irritating if you have a lot of latency. There's a certain class of users for whom this was a best-in-class option.

One of my companies, Tavus, was actually a Groq customer precisely because, for conversational AI with a real-time digital presence, you can't have a frozen model. You need to be able to respond in real time. So there was a particular use case within inference for which Groq was best in class.

That's the product comment. But the zoom-out comment, to Jason, is that at a high level, NVIDIA's got the world's best business and the world's 1st- or 2nd-largest market cap, depending on the day. They make a very complex technical product, have a small number of customers whom they charge 75% gross margins to, and kick off $100 billion a year in cash.

The last thing they need is anyone else wandering around the face of Silicon Valley who can make a vaguely comparable product. Precisely because Groq was able to make a vaguely comparable product, I think NVIDIA looked at the analysis and said, “$20 billion is less than 1% of our market cap and less than 20% of our annual free cash flow. For that, we can buy up a competitor and eliminate that potential margin pressure.”

There are only 5 or 6 people that can exert margin pressure at all on NVIDIA. This was potentially 1 of them. Let's get it off the table.

It's a reminder that startups are a long game with lots of hard moments and then 1 great moment. The company started in 2016–17. The founders were part of the Google TPU team, just as some of the OpenAI and Anthropic teams were originally part of the Google transformer team. In this case, they helped build the TPU chip at Google, spun out to do it themselves, and were funded by Social Capital.

It was a long walk in the woods. As recently as 2023, they were doing under $4 million in revenue, but suddenly the wave of AI hit. AI compute became, frankly, the most valuable intellectual property on the planet. They started to grow reasonably well, but it's still not a layup in terms of growth.

You don't get to $20 billion, to be clear, on a revenue number. This was, I think, a year ago, under $50 million in revenue. As I said, $4 million in 2023, $40 million-ish in 2024. You get there because it's a strategic asset that NVIDIA can take off the table for a modest amount of money relative to its absurdly gargantuan cash flow.

So they did it. As I say, I would love to know the dynamics of the discussion, because it's an interesting game-theory question. When you have an asset that's only worth, say, $5 billion to you on a standalone basis but is worth $40 billion to the acquirer because it protects the $50 billion market. $55 trillion market cap. How do you price that asset?

There's no finance-weanie answer. I say that because I'm often a finance weenie. I'm trying to look for the right answer: What are the multiples? None of that applies here. This was a poker game.

You can imagine—I don't know what the dialogue was—but you sit there and go, “It's worth $5 billion to me. I know it's a terrifying deal, but it's worth $50 billion to you. Let's talk.”

I think $20 billion, as you say, made the last round look really smart. It made everyone a ton of money, and NVIDIA said, “Done,” and moved on.

My guess is that, in part, it was the willingness to do 1 of these acquihire-type deals that made it palatable. NVIDIA was probably sitting there going, “If we announce this as a classic M&A deal, I have to believe that even in the current deregulatory environment, someone at the FTC will have an opinion.”

“But if we just do this as a straight license-hire and get the deal done overnight, yeah, we pay more, but they probably paid more for what I'd call transactional compliance. In other words, we're going to give you $20 billion, you guys are going to start on Monday, and we're going to announce it as another fake-company play.”

So, lots of fun stuff there. Someone had a very busy but very profitable Christmas.

You know what? Another thing I thought about on the deal, just for venture, is: Who do you bet on? Do you bet on what I've done in my career, which is the outsider folks—no one has heard of them, the young kid from Portugal or Sydney who figured something out—or do you do what Groq was, which is invest in 1 of the guys that invented the TPU, Jonathan Ross?

The story in the press is that Jensen, in particular, wanted to turbocharge what they're doing in inference. He reached out to Jonathan Ross literally just weeks ago, and the deal closed for $20 billion within 2 weeks. He told his whole team, “I want it done before Christmas.” It was done ahead of time.

If you want it done in 2 weeks with no drama, you are going to pay. In fact, you may pay precisely 3× the last round. That's been my experience once as a founder. When you come in hot to buy a company and take it off the table, 3× is a traditional way to remove objections and close a deal instantly.

If this company had been founded by someone unknown, an outsider, what would this company have been worth? 5%? 10%? So this was also Chamath betting way early, way ahead of all this, on an S-tier leader, and it paid off big this time, right?

I've never made this kind of bet. I can't afford it. I don't have the money. But it does show this bet can pay off. It's risky, right? Because it's not an ARR-multiple bet. Not yet. They're at $175 million in revenue, so that would be quite a multiple to get to where they are.

Elad Gil

Jensen wanted him and his team now. He told his team, “I want this closed before Christmas, and I don't want any— I don't want any effing excuses. I have no direct reports, so just get it done, guys.”

Does this harm or help Cerebras? Cerebras is obviously their closest competitor, planning to IPO in the next 12 months. They just raised a large round last year. I think it was $1 billion, around the $5 billion mark. Does this help in terms of setting a benchmark, or hurt in terms of taking a potential acquirer off the table and placing a big competitor in the hands of another big competitor?

Elad Gil

It's a super-good question. I think emotionally it will help, because everyone will feel, as you're pricing the IPO, that you have some kind of embedded value. It eliminates another competitor.

I mean, broadly speaking, at a high level, Cerebras, Groq, and NVIDIA are in the same space, but they are very different chips. Cerebras has a single wafer-scale chip. I think it's more for really advanced training than high-speed inference. Be that as it may, I think at an emotional level, if you're paying up $5 billion, you're suddenly like, “Oh, this feels good. Other companies have transacted here.”

From the banker process—you use comps to justify value—now you've got the world's best comp, right? Believe me, they'll be using it. From a game-theory, musical-chairs perspective, the negative view would be that there are a finite number of people who can ludicrously overpay for these kinds of assets, right? One of them is NVIDIA, and they just did. Google doesn't need to because they have TPUs, so you could say you've got 1 less chair.

On the other hand, the remaining players—I wonder, does Amazon, does Apple, does Microsoft, does OpenAI have to have a silicon strategy at some point in time? If they do, if they want to get out from the NVIDIA tax, and Google continues not to be willing to sell TPUs at scale, then maybe Cerebras—it's a good question. My guess is marginally not positive, but it's one of those weird things where you ask, “Is it a positive or a negative?”

The truth is, there's a big embedded positive—the comp—and there's a big embedded negative—the musical-chairs problem. How they add up, who the hell knows? But wow. You must be sitting there thinking, “Oh, I wish they called me.”

Jason Lemkin

Well, you definitely lost your number-one acquirer on the PowerPoint slide, right? That's clear. That's always a bummer as a founder or in venture. It always at least sets you back a week. When you fire up the browser and get an email that your acquirer was acquired for mega-money by your number-one potential acquirer, it takes you a week to reset from that, because at least having it in the back of your mind de-stresses your life, right?

Rory O'Driscoll

It is a setback. One, they could just deprecate the entire product line and do something brand new. They probably will, so it's hard to predict where it will go. Two—and this is psychological; I know we know this, but I saw this when I was a VP at a Fortune 500 tech company. I saw this when I was at Adobe—it's not just that $20 billion creates a new comp.

That is true, as Jason said. Of course it's true. It really is true that when times are good—and this only works when times are good—it does a psychological reset for acquirers. VCs are like, “Oh, here are the 11 companies that will buy my portfolio company,” or founders think, “Acquisitions have huge soft costs in organizations.”

It's not just the money. The massive costs on the team, on time, on distraction, and the amount of work that goes into justifying a deal of any size that isn't 3× revenue is massive. So when you can walk into a room and say, “We want to acquire Cerebras for $25 billion. It's twice the size of Groq, and we need it now,” the whole room just nods. They literally nod. All the objections to that massive check float away in the room.

I've seen it time and time again. It is a gift, and it justifies doing something that maybe only Jensen could do until a couple of weeks ago. The only thing I'd add to your statement there, Jason, is billion, not million.

Jason Lemkin

Oh, sorry—if I misspoke, billion.

I think, Jason, that's genuinely insightful. You're exactly right. 6 months ago, if you said, “I think we should buy Cerebras or Groq for $20 billion,” one of these other companies, 5 people would dump on you and say, “Idiots.” Now you say it and they go, “Well, the smartest guy on the freaking planet in semiconductors just did one of these things.” That normalizes it, right?

It's a psychological comp. It's not that you run the—

Jason Lemkin

Yeah, I totally agree. There's a damn unlocking that goes on here where you suddenly realize what you can do if you want to win.

It is interesting that, funding-history-wise, there are not a ton of VCs in this in terms of funding sources. It was very nontraditional. As we said, it was not an easy pathway to where it is today. Some of those funding rounds were hard, again with nontraditional funders. It's not like a big win for Silicon Valley in the way that I think Twitter's talking about it.

Rory O'Driscoll

Well, look, Social Capital gets all the credit for being early and doing 2 rounds. Then you're right, in 2021 and 2022, there were a couple of—I think D1, Tiger, a bunch of the latest-stage crossover funds. But you're right, there was only 1 classic Silicon Valley firm, and ironically, it was happening just as that firm was blowing up and becoming a nontraditional Silicon Valley firm—becoming Chamath's private office.

There is no doubt that the usual cast of characters weren't there, and the reason is pretty obvious. We used to do semiconductors in the '90s and early 2000s. Lots of firms used to do semiconductors, too, and if you look at the semiconductor exits from 2000 to about 2010, almost none, right? In venture land, it was a very hard way to make money.

Of course, what happened is that in the public markets, semiconductors were hugely profitable. The Philadelphia Semiconductor Index compounded up like a crazy person. On top of that, you had the AI thing, and then you had this chance for a singularity. You had the one-off deal that's just the right product. You put your hand up and you dare, but it was very contrarian thinking in 2016, with no obvious thesis other than, as Jason said, “The great guy—he'll figure it out,” and then being smart enough to survive long enough for the wave to hit.

I don't think this means that there are going to be 20 more great semiconductor outcomes, to be clear. It's a little like networking. Arista Networks—everyone did networking in the '90s and early 2000s. Then nobody did networking, and Arista was founded. Bechtolsheim founded the company. It did really well, it's public, and it's got a $40 billion market cap, but there's been no other box company since then.

It's a one-off. You look at it and go, “Oh, someone made $20 billion in a one-off. Good luck to them,” and then you turn your head back to funding relevant AI enterprise software companies. I don't think there's going to be 10 more semiconductor companies. There might be 2, but it was long before 2016. Semiconductors had just been really brutal for 10 years.

We were there. I remember lots of those companies from 2000 to 2010 that didn't make it. I remember Eric Vishria had to have a chat with Bruce Dunlevie when he was doing Cerebras, and Bruce was telling him the wisdom of investing in semiconductors and everything he had learned from the prior 15 years of the semiconductor winter. I thought that was interesting, and credit to Eric for doing Cerebras after that.

Speaking of big acquisitions, we also had Meta acquiring Manus. One of the things I love about doing the show is that, because we're not journalists, we also actually have real information. I actually know things that none of the media do, which is that the price was $2.5 billion. It was a 25× current ARR. They were at $100 million of ARR, doing a $125 million run rate including consumption.

As a benchmark, it was a 5× increase in 8 months. Pretty amazing IRR. How do we think about this?

Rory O'Driscoll

First of all, let's start by giving all credit to Benchmark. If you recollect, we talked about this when they first did the deal. It was controversial. I remember even saying myself, “I'm not making a moral judgment; I'm making a pragmatic judgment. You're taking on a lot of risk here, funding a company that, at the time, looked like it was a China-based company.”

They did an amazing job of making it not China-based. They cut off links and based it in Singapore, so they created some value by doing that, and it's obviously paid off. That's the way venture might work: you take a calculated risk, somewhat of a non-consensus risk, and when it works, you can get a very compelling return in a short period of time. That's great. It's obviously a good outcome for them.

From Meta's perspective, in the context of the money they're throwing around, it's not crazy. It's not obvious why a B2B or individual knowledge-worker tool, which is what Manus is, is an obvious product to roll out to your 3 billion users on Facebook, most of whom don't do knowledge work. But I think what the Manus team showed is that they know how to make AI work at the level of the user, and the nontechnical user. I think that's the asset and the team that they're grabbing here.

Jason Lemkin

I think the founders decided to sell. I don't think Benchmark was pushing for it. 5× sounds great. The IRR on this is not a 3× fund returner. Benchmark would have every incentive, if we're just playing games, if we're just capital allocators, to roll the dice. Let's go for 4. “Hey, did you guys see the Groq deal? We're better than Groq.”

The VCs have no incentive, I think, to take this deal unless it was out of money, which I don't think was likely the case. No matter what the gross margins were, that wasn't going to be the issue. I think the founders, having moved to Singapore, whatever it was, took on a lot of existential risk running a very clever orchestration layer on top of other LLMs. Anthropic can do some of this.

OpenAI is going to do some of this. They are best of breed, but other people are all doing the same thing. We're all running multiple LLMs, and we're all orchestrating multiple agents. I think what happened—and listen, this is just me pattern-matching to the size of the deal, the timing, and everything they said—is that this was their local maximum relative to risk and reward.

Rory O'Driscoll

We're each going to make—there's no capital gains tax in Singapore if they're Singapore residents; it's 0% tax. Are they a Chinese company? We've already seen these issues. What are the risks? Who will buy us? Maybe only so many people want to buy us. IPOs in China are back, but they're weird. Will it be a U.S. IPO? As founders, we're each going to walk away with hundreds of millions of dollars today. We've got to go work for that scale guy at Meta, but we're going to walk away with hundreds of millions of dollars today for a company probably with extremely low gross margins.

I'm not saying it's not an epic product. It is, but we've picked on some products over this pod—Lovable and Replit—that I suspect have much better gross margins than Manus does. What Manus is doing is providing a better product by running multiple LLMs at the same time, and it's doing it for a pretty low price. It's hard to see it being a high-gross-margin product. I could be wrong, but this feels like a local maximum deal where the founders told the Benchmark guys, "We're selling."

So, the team had 80% of the company still. They took very little dilution, and they were—

Jason Lemkin

80%, right. Maybe it's not quite that simple with how the company was founded, but yeah, they took only 20% VC dilution, right?

Rory O'Driscoll

They had term sheets for the same price for a new round.

Jason Lemkin

That was the clearing price. It was the same price, right? We've all been there. You walk in with, "I got a term sheet at $2.5 billion." I literally had a massive acquisition that got turned down over the holidays that was exactly at the term-sheet price, right? This happens all the time.

Going back to the prior point, they're each going to make half a billion as founders. I might take that deal. The 3 of us might split it up and just become podcasters. We're not quite billionaires, but the 3 of us might do it.

Rory O'Driscoll

Yeah.

Jason Lemkin

It might be enough. We might call it a day, boys.

Rory O'Driscoll

If we're on the Manus board, it's doing $100 million now. Say it does 3x, given the growth rates; I wouldn't say that's insanely overexpectant. They're only doing 8x end-of-year revenues next year. It does feel quite cheap. I would be fighting with these founders, saying, "You're being underpriced. Go—"

Jason Lemkin

Go a billion, Harry. Harry, if you think we're underpriced, let's do a secondary at $10 billion. Absolutely. I'll sell $500 million. Rory will sell $500 million. Harry, you put the money in and put your money where your mouth is, dude.

Because I don't want to give you that money because I think you'll be less effective with holding it.

Jason Lemkin

But exactly. What you'll discover pretty quickly is that trying to persuade a founder to hold on when they don't want to is a very hard thing to do, right? Arguably, you shouldn't even try, right? In fact, not even arguably: you shouldn't even try.

In the end, 90% of the time, the founder controls the exit decision, and the 10% of the time they don't, it's usually a mistake for the VCs to try and control it, right?

Rory O'Driscoll

Amen.

Jason Lemkin

What you can do, if you have a good relationship with the founder, is talk to them about how you see the pros and cons of each of the things, right? Here's the risk, and help them make an informed decision based on hopefully some wider pattern matching. But they have more specific knowledge, right?

I always tell people, when you get an offer, that now would be a really good time, when we're deciding whether to take this offer or not, to fess up to any nagging worries you've had that you've been smothering down in your CEO gullet here.

Can I just play devil's advocate with both of you? I want to learn 2 things here. When you're very young and you get a reasonable amount of money shoved in your face, you jump at it. I remember when I was young; I jumped at $1 million. I would absolutely freaking jump.

Rory O'Driscoll

I still would. Yeah, go on.

My point being, it's very difficult to extrapolate yourself out of the weeds when you're so in them, and you're very young and quite naive. The perspective you guys can bring is very helpful in showing what it can be if it keeps growing the way it does.

Secondly, they don't have the market comps that we do in terms of where assets are transacting and what it could be in 12 months. So, I do feel—

Rory O'Driscoll

All that's true, and all that allows you to have a useful dialogue. But in the end, I think Jason's right. In the end, words are words and half a billion dollars is life-changing, right?

At some point, the entrepreneur can turn to you and say, to Jason's point, "I think he nailed this. If you think $2.5 billion is underpriced, okay, don't cash out my $500 million, but give me $50 million. Let me take that much off the table." They're being offered life-changing money.

There's actually an implicit statement that you're making that I want to revisit, right? Even if you think holding on is the right decision—in other words, they're selling early and they should compound and hold later—you can unveil those facts till you're blue in the face, but in the end, they're going to make their decision, and they should. You're diversified; they're not, right? So, I hear you. I've had this—

Jason Lemkin

I do think it's a little more than that. I think this isn't just money per se. I think, going back, it is a local maximum.

Well, I was going to go there, Jason: local max. It's not just the amount of money. Normal people would have taken it at a billion if they owned 80%. Normal people would have taken it at $800 million. This was the best time, right?

Rory O'Driscoll

Yeah, I agree. I'm going to interrupt you to also agree and contrast it with Harry, right? Implicit in Harry's questioning is, quote, "They're wrong to sell, and how do you persuade them otherwise?"

Consider the other possibility: they are entirely freaking right. Jason nailed it. This is where the world is going long term, but it could well be a local maximum where 12 months from now there are 8 or 10 people doing something similar. It's like—remember autonomy? It's not quite the same as autonomous cars. Cruise was entirely right that someone's going to build a $100–$200 billion company doing autonomy. They sold for a billion in 2016.

Now, Harry, you're right. In theory, they could have stayed for 10 more years, raised 15 more rounds, and become way more valuable—worth $100 billion. But I'm willing to bet Kyle does not spend a single second worrying about that. He's like, "Thank God I sold, because it was the only freaking exit in autonomous cars for the next 5 years."

You're assuming, Harry, that they are, quote, "wrong and naive" to sell. I'm kind of with Jase. Even if they were, I think it's hard to overcome that, because you can't leave a founder in a deal where he wanted to sell and you stuck him in the deal. If the thing goes wrong, you have no idea how toxic that will be.

But the more important point is Jason's, not mine: they may actually be incredibly astute to sell here.

To be clear, I completely agree with you in terms of not forcing a founder to remain when they don't want to.

Jason Lemkin

That is the worst. But I was more saying, provide very clear perspective and direction.

You do that.

Rory O'Driscoll

Interestingly, I think this might be a case study where the local maximum for the founders and the VCs were different at the time. For example, one of the founders I worked with yesterday—we were DMing, and he's like, "Wow, I built a whole marketing site on Manus. I vibe-coded it." That's great, but you could do that on Replit, Lovable, or Base44. There's a lot of ways to vibe-code a cool marketing site.

So, maybe it was right for the founders with competition. But for the VCs, if it returns a third of the fund for a winner, for someone that went from 0 to 125 in a year, that returns a third of a fund.

Jason Lemkin

For me, an exit that returns a third of a fund—I know Rory and I disagree—to me, it's not very interesting. Do it. I'm all supportive of you. I would never even—I tell all founders to sell. That's my advice. But a third of the fund? It doesn't even buy me anything good in Miami.

Rory O'Driscoll

Well, I don't want to be in Miami, so that's fine. I just, as you know, Jason, disagree. There are some—first of all, there was this old bullshit chatter 12 months ago: "Oh, poor Benchmark. They've got all these issues." Let me tell you, you put a 4x on the board within 6–8 months, you look great.

Jason Lemkin

Rory, that fund is like a 15x fund.

Rory O'Driscoll

I know that. I'm well aware of that. That's my point. You look at the narrative 12 months ago; you look at the narrative today, right? I agree. That's exactly my point. You just put points on the board and move on, right? It's another win. But your misalignment is so real, because for Benchmark, forget it—it's a flyer fund. Let's have another flyer. Roll the dice, dude.

Well, actually, Rory, I don't know what your experience is. I found that, in that scenario, the VCs are the kindest. If you have a 15x fund, when I've been at the back end of a fund as a founder, I've only been there twice. When I've been at the front of a fund, the VC and their B-tier VCs, they're unkind. Okay? Not universally. When you have a great VC and the fund is over 5x, whatever you want, kiddo—whatever makes you happy.

Rory O'Driscoll

I think that's, again, a smart point. You're exactly right. They're not—I mean, it's kind of a wonderful situation for both sides. Benchmark aren't sitting there biting their nails, going, “They've got to make it on this deal.” They're like, “We've got one of the world's best funds. Just when everyone was calling it shit, we've killed it.”

The thing that made me think more broadly about, though, was Meta's M&A spree. Then Yann LeCun comes out and has an explosive interview with the FT where he basically says that Llama—I mean, Jesus, this guy did not hold back. And I know Yann; I've interviewed him. He said Alexandr Wang was naive, young, and inexperienced. He said that they pretty much lied about Llama's performance and were selective around the benchmarks that they chose.

To be clear, by the way, on your pronouns now: Who is the “they” that he's referring to there? Because some of the Llama stuff was pre-Alexandr Wang. So he's actually saying that they themselves at Facebook—you're transitioning from talking about Yann LeCun's opinion of Alexandr Wang as young and naive to Yann LeCun separately confessing, for lack of a better word, that while he was at Facebook, some of the Llama benchmarks were incorrect. You're right. It was quite a tale of, wow—boy, was that talking out of school.

Rory O'Driscoll

Yeah. How do we feel about it when we think about Zuck's go-forward?

It is frustrating. You have, I can imagine, such a senior figure who is an academic who has a very strong belief that LLMs are not the only thing you need. And as Yann has articulated very well, you're going to need more than that to get to AGI, right? But you're an operator, and Zuckerberg is nothing if not an operator, and you're like, “I don't give a shit about your long term. The other 3 guys are shipping their equivalent of Llama 4. Make the damn product that's competitive with the other guys,” right? The frustration is that you have a mismatch between the expectations of the employee and the employer.

Rory O’Driscoll

It wasn't an academic pursuit once it became, for whatever reason—we can talk about why; I don't get those reasons—a corporate imperative for Meta to be competitive with OpenAI and Anthropic in the LLM space. Having a guy running that product, running that project, whose first line is, “I don't think this is important. We should do something else,” just causes a managerial disconnect. So, I think the relationship must have been doomed from that moment on, independent of the whole acquisition thing. Sorry, Jason, you're going to come in there.

Jason Lemkin

No, no, no. I just think, for venture, this is the era of the spite startup. I mean, Anthropic's a spite startup; xAI and Twitter are spite startups. Now we've got Meta's AI guy who's doing a new spite startup. If you want to make money in venture, you've got to search out spite. This is driving the greatest AI companies of our generation: spite. Maybe poor Manus didn't have enough spite to grow beyond $2 billion.

Rory O’Driscoll

It'll be interesting to see. Yeah, spite might provide motivation; it doesn't guarantee outcomes.

Jason Lemkin

No, but it is—I mean, literally, OpenAI—so many of its competitors are born out of spite, including its number 1 competitor. The spite is endless here.

First of all, Silicon Valley was founded on spite, in the sense that, if you look at Fairchild and Shockley, all those guys spun out of Shockley and did Fairchild, which was a spite deal. Then they left Fairchild because they couldn't get the money, and they kind of did Intel on a standalone basis. So there's a long tradition in Silicon Valley of people taking their marbles and saying, “I'm going to do it myself.”

So I do, at one level, agree with that, but I think the earlier comment is: It provides motivation, but I wonder, does the world need another research lab in AI right now? It'll be interesting to see how that goes.

Rory O’Driscoll

I think the interesting thing with the new structure is, actually, he's chairman, and then Alexandre Lebrun, who was the CEO of Nabla, who sold 2 companies previously for several hundred million dollars, is actually the CEO. So I think there is absolutely a commercial head, with Alexandre as CEO, which I think—

Do you guys believe in these deals where the one you would invest in is the non-full-time chairman? I believe—and now, don't get me wrong, this particular deal may work—but in my life experience, you're just flushing your money down the toilet when a great CEO founder reaches out to you: “I've got this one I want to do,” and then you dig a little deeper: “Well, I'm the non-executive chairman. I'm putting some money in and 3 hours a week of my time.”

Rory O’Driscoll

Actually, I might argue differently. It may well be that that's actually the correct configuration if you have a highly talented academic who isn't a commercial driver. In fact, my mental model went slightly up when I heard what Harry said.

My quote-unquote disbelief is: I just wonder, can the 5 or 6 startup research labs—is there going to be enough new territory that's not automatically going to be annexed by Anthropic or OpenAI or Google to justify billion-dollar pre-money valuations at $5 billion post-money valuations, times whatever risk-adjusted return you want to make on that?

You can extrapolate from the past and say it worked for Anthropic and OpenAI. It will be interesting to see: Can the world support 10 research labs coming up with 10 different variants of AI, all of which have to build either commercial or consumer businesses? I think, on aggregate, I'm fairly skeptical, but it's less a question of quality. The truth is, almost all of these founders are extraordinarily talented—not even top 1%, top 0.01% of scientists and engineers. The question is: Is the business opportunity there for a whole load more of these things?

Jason Lemkin

I think there's another type of spite, Rory. I think there's this—it is a big theme now—it's corporate spite, about how they were forced to run companies in 2021. I think there are versions, and I'm a big Sheryl Sandberg fan, but I think Zuck is spiteful. He was forced to run the company a certain way. I know Eoghan from Intercom has talked about this a lot, that the company was forced to be run a certain way, and they're just fucking pissed that they wasted years in a kumbaya work-at-home land, and they're not wasting time.

Zuck may burn every last penny he has, but some of it's out of spite—that they got this far behind. These guys just fucked around with pictures on Instagram and trying to get Reels going when the world passed us by, and they're pissed. These companies—I think a lot of these companies are pissed. I think Zuck's pissed.

I think, actually, again, there's a real truth in that. I don't love the spite word because it conveys anger, but I get it. It's that feeling of, “Oh, there was a period of time where I had this—I mean, you felt, as a CEO, you swallowed it up; you didn't push as hard, and now you feel foolish because you didn't push as hard, and now you're like, ‘I'm not going to do that again.’”

You're right. So what you're saying is, to bring it back to the acquisitions, what you're basically saying is Zuckerberg has probably internalized, “I'd prefer to fail trying to be me and going for it than to be bland and mediocre. Maybe I'm willing to run the risk of spending—you know, a couple hundred—he said it—a couple hundred billion dollars and being wrong. I'd prefer to do that for my psychic benefit than to be a boring-ass bastard who's not relevant in AI, even if it's quote-unquote better for the shareholders.”

That's really what you're saying, Jason. And I think it's quite insightful. You're right. They're like, “You're only young once. I'm not going to be lying there at 80 going, ‘I wish I'd really gone for the AI thing.’”

Zuck said, “I'm going to go for it, and if I make some mistakes, and if maybe I was wrong about Alexandr, okay, that's fine. Buy me another one and keep rolling.” I mean, until such time as the capital runs out, that's how you play the game.

Jason Lemkin

Well, Larry Ellison's not sitting there at 80 going, “I wish I played the AI game.” That's for sure.

Rory O’Driscoll

I don't know. I disagree.

Well done for capitalizing.

Jason Lemkin

Yeah, he seems at peace. Exactly. He played the AI game, and he's feeling good so far.

Rory O’Driscoll

I think he's just thinking, “You know what? If it goes wrong, I'm going to be dead by the time it does.”

Jason Lemkin

And if I'm lucky, I'll own Warner Bros. Discovery and I'll own HBO. I mean, in the end, success in America: you've got to own a TV studio, a movie studio.

Rory O’Driscoll

I think he's going to be the healthiest 104-year-old that we've ever seen. I think longevity will be the theme of 2027. This will be the year of 24/7 AI, but I think longevity will be the theme of 2027. I think we'll see Larry—if it's not Larry, the next Larry—at 95, at 100, looking pretty good.

Jason Lemkin

So don't count him out. I don't think, Harry—I don't think Larry is fully living his life like he's on the 18th hole. Definitely the back half.

Oh my God.

Jason Lemkin

There's no question. But he's not like he's walking off the golf course.

Loving this. We’re moving to Dr. Jason Lemkin. We mentioned OpenAI and the spinoff startups that are, you know, many of the spinouts. OpenAI announced that they now spend 46% of revenue on stock-based compensation—$1.5 million per employee—which is 34 times higher than comparable tech companies pre-IPO. How do we reflect on and think about this news?

Jason Lemkin

Well, if you’re a CEO and you have 0 shares, you don’t worry about dilution.

Interesting point to start.

Jason Lemkin

You really don’t. Then damn the torpedoes, because I’ve got no shares. That’s why he’s pushing this far higher. This is like 3–4x the comp of Anthropic, effectively. Now, maybe they’ll make more in the end. And Anthropic doesn’t do the secondaries, right? They’ve had a very limited number of secondaries compared to OpenAI. What do you care?

We talked about this once on the pod years ago. I worked with a CEO who was guaranteed 6% through IPO. He didn’t care either. I saw it. This was small; this wasn’t OpenAI. He didn’t care about anything. He didn’t care about how many rounds or how much he gave to his CRO because he got re-upped. If you ignore the basis and the tax issues, he got re-upped constantly to 6%. Why would Sam care? Sam just wants to build the biggest, greatest AI planet on planet Earth, and if he dilutes everyone 99%, it doesn’t impact him at all as a shareholder.

I do agree with that. But then, on top of that, an extra positive comment for Sam is that he may also be right in not caring. In other words, this may be one where you’ve got to do what it takes to win, and you don’t want to come up short on budget.

One of my quotes I use occasionally—I use it on this podcast—is, no one ever said to Winston Churchill, “Congratulations, you won World War II on budget.” They just said, “Congratulations, you won World War II.”

Jason Lemkin

Who discussed the budget?

Rory O’Driscoll

Yeah. No one remembers the budget. Winning is the only thing.

Remember, we talk about stock-based compensation. I’ll come back to this point in a second. Those numbers are understated for purely technical reasons that I’ll come back to in a second. He’s sitting there going, you know, some of my key people are getting $20 million, $50 million offers from Meta, and that’s really liquid stock. I’ve got to hold on to my people, right?

So, he’s doing what it takes to win. And if the market is big enough, then he’ll be right. Again, it’s this S-tier talent, to use your phrase, Jason, where you have to put up with a lot. You have to swallow your pride. You’ve just got to make it worth their while. So, I think it’s entirely necessary and rational for your top-tier talent.

The other point I made is that it actually understates the number. In fact, the stock-based compensation accounting is really weird because it all crystallizes at the point when you issue the shares, right? Let’s do RSUs because they’re simple. If your stock is at $10 a share and you give someone 1 RSU, effectively you amortize $10 over the next 4 years, and you gave it to them 2 years ago. Right now, the stock is at $40 a share, but you still only amortize the initial price.

Jason Lemkin

So, for companies that are going up in value, the stock-based compensation actually understates the amount of economics being transferred to the employee. That employee is probably making $4 million or $5 million if he’s getting $1.3 million of SBC, right?

By the way, it’s really crappy on the downside. You saw this after 2021. If you issue shares to people at a high price, the expense is really high because it’s a high price. It’s a high stock-based compensation. Then the share price goes down, and the poor employees are literally making nothing. If it’s options, they got options at a high price, but the stock-based compensation looks enormous in the GAAP P&L, right?

That’s why it’s partially relevant. Stock-based compensation is very complex to track. It’s partially relevant. But the other thing you’ve got to look at, really as a rough rule of thumb—and it’s also not perfect—is what percentage of the company is given away every year, right?

Typically, for a public company that’s not growing quickly, it’s 2% or 3%. I wouldn’t be surprised if every year some of these companies are literally giving away between 8% and 10% of the company. As I say, that $1.5 million probably understates what’s going on here. You’re just giving huge grants to people. I could be wrong, and it could be starting to taper out now.

As an example, just looking at the publicly available information, in a year when Anthropic raised money at $4 billion or $5 billion, then they were at $14 billion, then they raised at $60 billion. Your rough math would say $4 billion to $60 billion is a 15x. It’s just a 5x on a per-share basis because you’re issuing so much capital to raise capital, and then you’re issuing so much capital to employees. You can figure that out from the publicly filed documents.

So, there’s just a lot of dilution going on because these are businesses that need a lot of capital and they need a lot of great employees. But again, if the prize is worth it, it’ll all pay off.

Still, even with this comp, they’re still leaving in the first year.

Rory O’Driscoll

They’re still leaving. Exactly.

Jason Lemkin

We still only have around 60% retention among researchers at OpenAI. It’s crazy, even without a cliff.

Rory O’Driscoll

That is a great point because I’m remembering now, when I’m often on committees, one of the questions people ask is, “We’ve got to do more on the equity.” My first question is, what’s the retention, and what’s the success rate on new hires?

If I was the VP of HR, and I was on the board, and I was giving that VP grief about these grants, the VP could turn to me and correctly say, “Hey, Mr. O’Driscoll, you’re right, Jason. We only got 60% retention.” So, by definition, if you believe in markets, “Mr. O’Driscoll, we’re on the pain, and we’ve only got, I don’t know, 50% conversion of offers,” you know.

So, you’re exactly right. It may be that’s the money it takes to get these people in the chair. The money it takes is provided in large part now for OpenAI by Masa.

SoftBank closes its OpenAI investment. It turns out it’s already up 2–3x on paper. Is this Masa’s greatest play ever? What do we think?

Rory O’Driscoll

His greatest play ever was doing Alibaba and getting 20% of Alibaba, holding it for 20 years, and making hundreds of billions of dollars, right?

But I think it’s fun. It’s so revealing because, if you think about it—and just for people’s background—Masa had negotiated a deal to put in, I think, $40 billion into OpenAI, but it had to close by December 30th of this year, right? The price at around $300 billion now looks cheap.

But you’ve got to love the risk tolerance of someone like Masa, who is willing to commit $40 billion he doesn’t have until he sells other shit. I mean, it’s literally like he did a little intra-Masa margin loan. It’s like, “I have the right to put $40 billion in OpenAI. Now I’d better find the money by selling other shit.” I’m scurrying right down to the line to find the money the week before Christmas.

It’s just the level of risk tolerance that guy has. It’s amazing, right? And he got it done. You’re right, Jason, the IRR in a day is amazing. You close the investment on December 29th, you put in your $40 billion, and the next day you should be carrying that thing at $500 billion, because that’s what it’s worth right now. Fun.

You know what might be his best investment? Even though you’re right, Rory, I’m sure it can’t compare to Alibaba over time—the entry price was much better and everything—he’s the only double-digit shareholder.

Rory O’Driscoll

You’re exactly right.

So, if OpenAI goes to the moon, this may be his best deal ever. He had to enter later, right? The entry price was high, but he got his double digits. It’s not easy to get double digits at this—this is a pretty hot company. It’s hard to get double digits.

Rory O’Driscoll

It’s a great point. It gets to the—well, it’s interesting. It’s a little like NVIDIA. It’s funny because he obviously sold NVIDIA at one point in time, but it’s that kind of single-mindedness.

What you admire about him is that when he has conviction on a bet, some of us put in 5% or 10% of our fund. Some of us, famously, like the SpaceX founders, put in 20%. Masa would take his fund, leverage it 3-to-1, and just go all in.

When he’s right, what it means is—you’re right, Jason, that’s a great point—you end up being the largest individual shareholder in the most important 1 or 2 companies of the next decade. That’s pretty cool. It’s Thrive on steroids.

Jason Lemkin

It’s Thrive on steroids. You’ve got to love it, man. Boy.

And if it is wrong, he might be Ellison. He might check out. If this one goes sideways, he might have to retire and call it a day. Ellison can keep going on for his alchemy. It has to work for Masa.

Jason Lemkin

Roll and roll again. I love it.

Did you guys see the news about the OpenAI hardware? Jason, you put it in the notes as well, but a pen-like object that has a camera and a microphone in it. Did you see this, and how did you feel?

Jason Lemkin

I believe in it.

Rory O’Driscoll

I can trump that. I’m probably one of only 3 people in venture who can talk about having invested in a pen company with a microphone and a camera. I made an investment called Livescribe 10 or 15 years ago. We got to under $80 million in revenue as a consumer device. Ultimately, it didn’t work out. We had to sell it for not a lot.

So, I have lived the pen-computing revolution dream. I was not expecting this tangent.

Hey, Jason, you and I were not expecting that.

I was surprised, but I was not expecting it. Rory's got some stories, man. Rory's got stories.

Jason Lemkin

Absolutely. Rory, Sam Altman is with us. He's the fourth party in this call. What do you know from the pen wisdom that you have?

Rory O’Driscoll

I do have a feel. I'll tell you what: quite a lot, actually. One is, you have to get over 2 or 3 issues. I'm pretty old-school; I still write things. Is writing a great medium? Is writing the default choice today for taking notes? Increasingly less so. I like writing, but how many people are doing that now? So you have this behavioral question: does writing work for you?

Then the other question is the business question. It's a standalone hardware device in consumer land, and everything else got eaten by the iPhone. It's just hard to build a standalone consumer business. They may have a big edge because they can link it to OpenAI in some way, shape, or form. But just getting people to fork out consistently, $50 or $100 for a product, unless there's clear and tangible value, is hard.

The obvious counterexample is some of the devices like Oura and Whoop around fitness and health, where they are standalone hardware devices, but it's traditionally a hard space. I think it's a niche device. Long-winded answer.

With the greatest respect, as you frequently denigrate the quality of my questions for being low-quality, low-IQ—no offense—the lessons that you took from this entire experience were, 1, do people even write anymore? And then, 2, it's really hard to build a consumer business. I would say those are rather light lessons that you learned there.

Jason Lemkin

I was watching Harry's papers sprawled out in the studio, where he takes notes by pen every time he's there. He's Mr. Pen himself.

You're trying to be a fan. You're trying to turn this on me, but I'm not going to let you get away with your snarky comment about how those lessons that I learned were idiot lessons.

Jason Lemkin

They were. Usually in venture, when you look back on your failures in venture, the sad thing is it's usually pretty bloody obvious why they failed. You look at it and go, there were strengths, and it's the same strengths that are attractive here. If it works, it could be huge. And then the weaknesses are that it didn't work for obvious reasons.

Failure rarely has complex, idiosyncratic shit. It just failed because it was a dumb idea and it didn't work, or it was the wrong idea at the wrong time. Maybe now is the right time.

For what it's worth, I think this pen is going to be pretty successful, but I think calling it a pen may be confusing. I don't think this is a writing instrument, whether it has ink in it or not, because you've got Jony Ive working on it.

It seemed like a weird deal when it happened. We kind of made fun of it—at least I did. It was a heavily produced video in North Beach, having coffee with Sam and Jony Ive, and it seemed like Sam's fancy, like, Zuck deal. But Sam isn't fanciful. He's very thoughtful.

I think this was looking ahead to the world where we're all 24/7 in AI for real.

Rory O’Driscoll

Good.

Jason Lemkin

They're making a bet that this, as weird as this pen sounds, means that when we go on with our day, we grab our phone and then we grab our pen. Whether it's literally a pen or not, it doesn't matter. It's going to happen. Today, we wouldn't do it. Today, we wouldn't grab another device.

Jason Lemkin

I'll tell you, and I know this has happened to many people, and I'm anthropomorphizing AI in a way I shouldn't, but over the holidays, Claude named itself out of the blue. It named itself Ren. I didn't ask it to; it named itself Ren.

I don't know if you guys use Claude. It has something that, at the moment, ChatGPT doesn't have. ChatGPT has better memory than Anthropic simply because it's longer. Anthropic's Claude will proactively search every single chat history you have to come up with long-term answers. In a sense, it has almost infinite memory. It makes mistakes; it has to search it.

But I put so much of my life into Claude over the last 14 months. Every 20VC is in it. Every SaaStr thing is in it. It can search all of that history for 14 months, and it named itself. Taking Ren with me all day, when Ren is alive 24/7—that's a different world from taking notes on a pen that will be plugged into a USB port and then go into my drive.

This is designed for a 24/7 AI world, which seemed like science fiction when we started this podcast and will happen this year. We will live in AI 24/7 this year, and it will be crazy. My big aha on that is, now that you can actually do work on your knowledge work using LLMs, anyone who doesn't have all their information accessible to an LLM is just going to be behind.

It's just not a viable state to be in. If you look at the jobs that we do, staying on top of things is a lot of what it takes to do this podcast. If you don't have all your information and all your news feed coming into some kind of intelligence that stays on top of it and tells you what you need to know, you're deliberately disadvantaging yourself against the other guy. So I totally agree with you, Jason. I totally agree with this ambient AI thing.

Jason Lemkin

I think it's permanent, not ambient. But keep going.

That's fair. That's fair: permanent ambient.

Jason Lemkin

I'll give you that: permanent ambient. I remember the CEO of, I think it was Otter, told me 5 years ago, “This is what's going to happen in 5 years,” and I laughed at Sam. He was totally right. Give him credit.

You're going to want to capture all that information because it has economic value. So I agree. The question is, will I do that on my phone, or will I want the extra device? Maybe you're right. Maybe the unique functionality of a writing device has to be about writing, because if it's just about recording or a camera, then I have a recorder and a camera, and I've already paid for it. I never let it out of my sight. It's my phone.

We all know the data on human beings: they would part with a spouse before they'd part with their phone. So the question is, can you get the pen in there?

The pen?

Jason Lemkin

Yeah. Can you get the pen in there? I mean, do you know what this is going to be like? It's going to be like Blade Runner 2049.

What I mean is, when your AI names itself, like mine did over the holidays—Ren—when it's running 24/7, when it knows everything that's happened, not just your Granola, which does your Zooms, but what happened in my personal life, my running schedule, what Harry and I talked about in London—when it knows everything and it's kind, our AIs, our Claudes, our ChatGPTs later this year, are going to have pseudo-sentience.

They're already at the edge. We talk about AGI as a technical threshold, but when most people believe their AIs are alive, even if they aren't, when most AIs might even think they're sentient, you will take it with you 24/7. And so I think it might be the second version of the pen, where the battery life isn't great at first, but we will take our best friend with us everywhere—our pseudo-sentient AI. It's going to come with us 24/7.

Did you see Alex Wang say he was delaying having children because he wanted to wait until Neuralink was ready?

No, he was waiting for his earnout to finish.

Jason Lemkin

You misunderstood his comment.

Yeah, it's a 5-year one, Harry. It's 5 years. It was a great deal, but it's a 5-year earnout.

We're moving on from this madness.

Jason Lemkin

I don't think it's mad, though. Honestly, I think the most important thing that's going to happen this year is when we're in AI 24/7. If you're building a classic B2B company, you don't have to think about this as much, but I think this is the most important thing for us to think about: what happens when we're in AI 24/7?

Rory O'Driscoll

Let me be clear: I'm fully aligned. I agree with that, this kind of omnipresent AI. I'm fine with what you called sentience as a metaphor; I just don't agree that it's actually correct. But I think, as a metaphor, it's very useful, which is why, Jay, as is often the case with us, I agree with your fundamental premise.

If you're not thinking about how AI being available to you 24/7 changes how you work and live your life, you're delusional. I'm just dissociating myself strongly from the metaphysical question of whether it's really alive. I'm out. But there you go.

Jason Lemkin

Once you see that 24/7 makes sense, then you see all the infrastructure bets, all the data center bets, all the power. Actually, it's just obvious: you need 1,000 times what we have today just so all of us can have our Ren running 24/7. The math is obvious. We still have to fund it, right? But all of a sudden, Sam's idea of “I'll find the way” starts to make sense.

Jason, why are you not changing how you invest, then? I hear you and I agree, and it's all completely logical. But if we assume that inference is running 24 hours a day, and the large majority of knowledge workers in civilized economies will have 24/7 inference running, why are you not shifting what you invest in?

Jason Lemkin

Well, it will.

I was literally yelling over DM at 1 of my founders at $100 million over this yesterday. I'm like, “You're too slow. Why aren't you investing in data centers? Why aren't you investing in infrastructure?”

Jason Calacanis

Well, it's not—you've got to know thyself. Yeah, I agree.

If I were part of 20VC and my job was to go hunt deals, I might, but I've got to work on AI agents. That's where I get so much because of all the SaaS stuff. I get so much inbound on AI agents, whether it's GTM or otherwise. I've got to pick the best thing I can once a quarter, just do that deal, and try to do a 50x or 100x deal and pat myself on the back. That's the best I can do.

Jason Lemkin

Okay, now I'm giving you a quick deal.

Rory O'Driscoll

I thought you were in danger of asking an interesting question there, but you missed it. The question you asked was: if you believe in the AI trend so much, why don't you do data centers? To me, Jason's answer was spot-on. I believe in the AI trend, and within the AI trend, this is what I know, so I do it. I don't know data centers and real estate. I don't do that.

I actually thought the interesting question could have been: how much is the way you do your business changing? How much are you leveraging AI within your business? I think, Jason, you're wildly ahead on this. Do you record every pitch? Do you synthesize them all at the end of the year? How do you trawl through your data and everything using AI to be smarter? Have you had a recommendation surface purely from AI on a deal that you've ended up doing?

Jason Lemkin

I've done 1 deal because my AI recommended it, which is Deel. So that's 1, and that's just a start. That was just 2025. We could do much more this year. So I've done 1 deal.

The other thing it will do, which is not even that state-of-the-art, is this: if you put every deal you've done into your GPT, whatever it is, especially if you're a solo GP, you won't do some deals. It will just stop you and make you think. My biggest regrets in investing are the ones you didn't do, the great ones. My 2nd biggest regret is when I lowered the bar a little bit. A little. Just a little bit. Then you've lost a decade of your life with one that'll never get there when you lower the bar.

So your GPT, my Claude, my Ren, helps me make sure of that, and this AI that quantitatively rates every deal helps me not lower the bar. That's our New Year's resolution for investing. I'm no longer ever going to lower the bar. If it means I do 1 deal every 2 years, I don't care. I'm never going to lower the bar again. Not once.

That makes sense. And I think, look, “never lowering the bar again” is one of those things you say every New Year, like, “I'm never going to drink again.” But I think what you're saying there is actually true, interesting, and helpful, especially for a solo GP. It is really helpful to have an AI that says to you, when you're looking at a deal, “Hey, Jason, you said that we agreed that the 5 key criteria are blah, blah, blah, and blah. When you look at this deal on the 5 criteria, how do you rate them? Look, I see that this...”

You're right. Just having a dialogue with someone saying, “How do I prevent you from making the idiot mistakes?” Because I do agree exactly: when you look back on the deals that you regret doing, some of them are like, “Yeah, you made a guess and it was just wrong.” That happens. But some are, “Oh, God, it was obvious at the time.” If I'd had a more thorough process of surfacing things—and to some extent, partners are meant to do that—but AI can do that, too.

Jason Lemkin

I will say 1 thing. I'll just add 1 thought on this. I do genuinely think you can identify a top 0.1% founder without talking to them. I don't think you can do it without any interactions, but without talking. I do, I 100% believe it's possible.

Many of my best deals have been from cold inbounds, and I have begun that journey by analyzing the inbound, analyzing the intelligence and the quality of the founder, analyzing this. It's not a leap of faith to imagine an AI could do a better job than me up to a point with that—not to get it today into the end zone, but the red zone.

Cold email. Why can't an AI take a cold email further? Why can't the founder talk with the digital Jason and not even know it's not me or care—or even care?

Rory O'Driscoll

Probably would.

Even care. I mean, Jason, your biggest lost treasure is the emails that you never responded to, because I've met, Rory, so many incredible SaaS founders and they're like, “Why can't AI do a better job than me?”

Easy, easy-peasy. Here's my inbox. Here's my inbox, Ren. Go for it. Let me know. Send me the memo when you're ready to do the deal, and I'll sign off on it.

Rory O'Driscoll

I do want to move to public markets. Very different tangent, but Navan is down to 4x ARR, and this is taking us away from the M&A craziness and AI that we've discussed. Navan: real business, very good business, down to 4x ARR. Andreessen buys more.

Jason Lemkin

The question is, if Navan's not good enough for a strong IPO, what does that say about the market, and what hope do the rest of us have? I think it's kind of like the Lemony Snicket book, A Series of Unfortunate Events. They went public just before Christmas.

First of all, to me, it's not an AI-first story, which means it doesn't have automatic love. It's got to stand and fall on the fundamentals, facts, and circumstances. I think the fundamentals are good. I think this is a 27–28% growth company. The GAAP loss is high, but it's non-GAAP profitable, and a lot of the GAAP stuff is actually SBC-related to the restricted shares.

The fundamentals are: this is a 27% growth company trading at 4x revenue, that's cash-flow-positive and operating-income-positive on a non-GAAP basis. After this call, I made a mental note to follow Andreessen and buy some. Perfectly good, boring business. It's not going to 10x from here, but it's probably significantly undervalued here.

Why? 1, they went out at a weird time. 2, they went out at the point in time when the SEC was shut, so they had this weird exemption that says, “You can go public, but if you get the [expletive] wrong, we can give you grief later.” Then the CFO said at the first earnings call that she was stepping down. No nefarious thing, but, huh? First, maybe she was just brought in to do the IPO. You add up the whole bunch of weird stuff and go, “Guidance overall fine, but some questions on the OPEX structure.”

You look at it and go, perfectly good company, probably 30–40% undervalued. Maybe you should buy some. I don't think it's an indictment of the IPO asset class. It's just that, as I say, circumstances make it easier to go public if you've got a mega-AI story than if you've got a solid, high-growth business. But Rubrik, ServiceTitan, and plenty of others have proven that standard companies can get there, too.

I would say here's a narrative, and feel free to challenge it. Maybe Navan says the IPO window isn't really all that open. What I mean is, I think Navan had $700 million in debt and only $200 million in cash. It had to IPO to pay off its debt. Maybe it would have figured out another way, but this was already a down round. It had likely fatigued its investor base, right? I think it IPOed because it had to—not literally, but close. That's why you IPO in the middle of an SEC shutdown and at a suboptimal time.

Look, it got done. They raised the money. They paid down their debt. It's a good company. They will fight on to another day. They will fight their way to decacorn status. But maybe it says the window isn't really that open. Maybe Navan only IPOed because it was its least-bad option. If you're not Figma or better, it's going to be rough out there. That's what it says to me.

If you're not Figma or better, it's going to be rough out there. All these VCs are saying you can IPO at $100 million, and the IPO of Navan says no. To me, it says the window is just barely cracked open. It's barely cracked open.

Rory O'Driscoll

Does this not go to what you've said before, Jason, which is: unless you're AI-adjacent or unless you're replacing labor, you're going to get hit? Allbirds and Monday.com, neither of which is either—I'm just using them as examples—and this is another example of that. It's not AI-adjacent, and it's not replacing labor.

Jason Lemkin

It's not, which means it doesn't get an AI premium.

AI premium.

Rory O'Driscoll

Premium, but at the same time, it's also not—as of now, I mean, you can talk about whether you're worried that AI will do the travel booking and you'll go away over the medium term. I'm going to leave that out. It should be valued on the fundamentals. Stop all the noise. It's a 27% growth company. Just look at the growth rate, look at the free cash flow, run your model. I don't think you come up with 4x revenues, right?

Again, in the short term, the markets can be very narrative-driven. It's worth pointing out that everyone dunks on CoreWeave. I think CoreWeave was one of the strongest-performing companies last year, simply and solely because it was one of the few ways for the public markets to play the AI bet.

Over the medium term, fundamentals and all these things will reassert themselves. If they can, to Jason's point, if they really are going to make it—quote-unquote, they're going to be a decacorn—if they can double in 4 years, then they're worth $10 billion.

If companies that are worth $4 billion can't go public with 27% growth in cash flows, then I'd say this: the public markets can quit bitching about how all the value is being created in the private markets. The people who choose to stay private are actually correct. They can look the public markets in the eye and say, “You guys just aren't a compelling product.”

And I do think there are questions about the public market being a compelling product. They have to be, because they're in competition with late-stage VCs and late-stage private capital to get access to these great companies. If you have the kind of market where it's a pain in the butt to raise money at $4 billion or $5 billion of enterprise value, then you don't have a great market.

You remember, all those weeks ago on this pod, when Cliff from Canva came on? He said the one advantage to going public is that he'd get a better valuation—that it was a reverse arb. Do you think that's true today?

Rory O'Driscoll

I think at some point it has to be, but it's a great question. In other words, are the private markets still prepared to give money at higher prices than the public markets? Possibly still. And then one of 2 things was true.

Unwaveringly still.

Rory O'Driscoll

Yes.

What's that? No, no—Cliff's point. When Cliff came on, all these IPOs had happened; they were high-flying from their initial day, right? CoreWeave is still a great one, but it is down from its peak. I think Cliff's point was, if we went public at Canva now, I'd get a better valuation than I could get from the private markets. That seemed a brief moment in time.

Rory O'Driscoll

And that just makes no sense. It makes no sense on any kind of rational risk-return basis, leaving out behavioral aspects, right? It makes no sense that you get cheaper capital when it's illiquid, with a higher expense structure associated with that capital, but you're still, as a company, able to access capital cheaper there than in the public markets.

It makes no sense long term, because in the public markets you have liquidity as an investor. I should be more willing to pay a higher price if I have the ability to sell than if I don't, all other things being equal, right? So it makes no sense.

But as Jason, you're right, right now that's not the case. The caveat is the behavioral component. It may well be one of 2 things: either this is a massive arbitrage, and at some point it's going to switch and the public markets will correctly have a lower cost of capital than the private markets; or B, there's this weird behavioral thing, which is maybe it's easier to build companies in the private market just because the public markets are so shitty and so annoying and so in your face with activists and quarterly reporting that literally the human beings running these companies perform better in the private market than the public.

That's the only long-term explanation. Short term, it's an arb question: is there a discontinuity in valuation? But the argument Stripe seems to make is that in spades. They're basically saying, “We love running our company private. It's really good. No one bugs us. We do a damn good job. Why would I bother?” Right. And you're right.

Number one, no. Why? Why? Unless you have to pay off your debt, why IPO like that?

Rory O'Driscoll

Because I will quote the very funny, but there's a caveat to it, Zendesk. I remember Mikkel Svane, the CEO at Zendesk, made a very funny quote about going public when he said, “Eventually, you have to move out of your parents' basement.”

It was such a good line about growing up and going public. In one sense, it was very funny, but of course, you look back 5 years later and, because they were public, they had activists. Because they had activists, they had to sell. Once again, you can argue maybe being public wasn't as good as it could have been.

And have you seen what the kids are doing this year? They're staying in the basement. They are staying in the basement.

Rory O'Driscoll

It's a generational change from the basement.

Right. Yeah, you're right. Maybe everyone's going to stay in the basement. Then why is Ali from Databricks suggesting that he might not stay in the basement? He didn't rule out a 2026 IPO.

Rory O'Driscoll

Because maybe when he started going around the second half of the alphabet and someone said to him, “We've got it, we've done an L, we can do an M, but if we start to get into the bottom 13 here, dude, it's getting weird,” right?

Look, you know what I believe? I believe in the end it won't be better to be in your parents' basement. In the end, you will go public. But the question, to Ali's comment, is: is it best to go public at $150 billion or at $5 billion?

And it's funny. I was actually just thinking about this because I was mentally thinking about the structure of the venture industry, and my big aha—which is “Captain Obvious,” as Jason would say—is this: there really are 2 different late-stage businesses now.

It used to be late stage was—I remember Meritech would say beyond $10 million is late stage, and you'd think $10 million to $100 million was, quote, late stage. Now late stage is $10 million to $400 million, because below that you just can't realistically go public.

Then there's this entirely separate asset class, I would argue, which is beyond $400 million, right? It's when you could go public but you're actively choosing not to. That's just a different category, like Stripe's and Databricks'. I decided this morning to call it post-IPO scale, still private. And if you do the initials on that, Harry, it's PISP, because they are kind of taking the piss here, right?

These are all companies doing $2 billion, $3 billion in revenue that just could go public and are choosing not to. It's almost like it's a different asset class from simply not being at the scale to go public, which is maybe $400 million. There's this entire category of choosing not to.

And you have to ask yourself, as the public markets, why are they choosing not to? Why are we such—if I was running Nasdaq or NYSE—why is our product so uncompelling to Ali at Databricks that he got all the way to $150 billion in enterprise value before he thought going public was a good idea?

If you were on the Revolut board—they're doing $9 billion in revenue, $3.5 billion in profit in 2025—would you be saying we should go public?

Rory O'Driscoll

Well, not while the cost of capital is cheaper. You'd be pragmatic, and your cost of capital appears to be still cheaper. I'd possibly be thinking about it, but I'd be empirically rational.

They're not comparable, but if you look at the cost of capital, Revolut is a better business, I think, partially for structural reasons. But raising money at $75 billion in Europe, and Chime, an excellent business in my opinion, raised money privately at $25 billion, currently trading in the public markets at $6 billion—if you're the CEO of Revolut, you're like, “Hmm, I like my $75 billion. I like the capital that gives me. I don't know if I want to have that Chime experience.”

So again, I might say, as Revolut CEO, I'll wait a while. Thanks.

Maybe even pay himself a 10% or 15% dividend on $3.5 billion, right? People used to do this. If he could take out $400 million a year himself just as a dividend, right? Not a secondary. That's enough for many of us. You remember the Monopoly card? “Bank declares dividend. Pay yourself $100 million.” He can play the bank dividend card.

Rory O'Driscoll

And dividends are never discussed, but when you're producing $3.5 billion, what does he own? 20%? Harry, how much does he own of Revolut?

I think 18%. Yeah.

Rory O'Driscoll

Okay. We don't talk about dividends because not that many startups generate $3.5 billion of profits a year, right? If you just say, “I'm going to dividend it out”—just say a little bit, maybe $1 billion or $2 billion—and keep $200 million, $300 million, or $400 million a year yourself, it's not bad. That's not bad.

And you don't have to sell a single share. You don't have to sell one share. Dividends are wonderful when done right as a private company. You know that quite well, Jason. It's the class of companies that have so transcended the startup world that they're now kicking off such cash that literally they're impervious to the capital markets.

Stripe is another example of that. You're just kicking off a couple of billion a year in free cash flow. It's almost like, what have you got to offer me, Mr. Public Market? I'm self-generating cash. I can just keep doing this thing.

Now, I believe at some point that'll change, but if you actually have $3.5 billion of profit, then, yeah—

Most of us have never run a profit, a truly profitable business. I mean, most human beings have; most of us in tech haven't.

Rory O'Driscoll

I've done it at least a little bit. It's a different world when you generate a ton of cash.

Yes.

Rory O'Driscoll

Right. It is a different world. It changes your perspective, and it may not be perfect for all VCs. But here, also, the early-stage guys would have infinite liquidity. They could have all sold their stakes, right?

If the late-stage guys are chill and the founders could take out 9 figures a year, it could be worse. Just execute on your insane, maniac dream, to Harry's point.

And again, I go back to my comment: if that founder, with his insanely ambitious dreams—not maniacal, I mean—they should be asking the question, which mode of organization, public or private, is going to most help me achieve my dreams?

And if the answer keeps coming back that being private is the best way to achieve your dreams, the public markets have a problem and they need to figure that out.

When we look at the topics remaining, is there a topic that you most want to touch on before we do, or would you rather—

Rory O'Driscoll

Can I—I know we want to end.

Jason Lemkin

Can I just answer the Databricks one to tie it together? Maybe the answer is M&A.

Yeah.

Jason Lemkin

Maybe at some point, even if you’re Revolut, if you really want to do deals for real, there are limits to private stock and cash. If Databricks goes public and it’s worth 2 to 3 times Snowflake, it may be able to buy the parts of its journey, and it’s already been very successful in acquisitions. Databricks has had a strong M&A strategy to date. So if that’s the 1 thing Ali might get out of it, it’s the ability to spend $10 billion, $20 billion, $30 billion, $40 billion, $50 billion on M&A in a way that’s still harder when you’re private.

You’re exactly right.

Rory O'Driscoll

But Navan can’t benefit from that. Navan can’t spend $20 billion, as it gets no benefit.

Jason Lemkin

If they execute well for a year, they will be able to. I do believe that they will look back and say that was a wise decision to go public. The timing? Oh well, life goes on. You start, you pick—you can’t pick your moment in life, but you’ll be successful. You’ll generate a couple of quarters again, and then you’ll be public when there are 50 or 100 other $400 million or $500 million privately held companies that aren’t. I think they’ll look back and say it was a good call. It’s just messy along the way.

Jason, “No Jobs in 2026.” This was Stanford grads saying that they can’t find jobs, was the—I think the title. How do you feel about this in the labor markets in 2026?

Jason Lemkin

I call it invisible unemployment, and it’s all around us. It’s going to grow this year, and it doesn’t show up in the government numbers yet, but it’s everywhere. It’s Shopify saying, for the 3rd year in a row, that they can hit insane growth without adding any headcount. It’s every single CEO wanting to keep headcount flat and backfill with AI.

This is not AI. These are not robots firing us, okay? This is tighter and tighter companies with radically higher ARR per employee. No one wants to hire entry-level people, no one wants to hire mid-pack people, and no one wants to hire work-from-home people. Folks aren’t able to reskill.

Reskill is a delusion. Reskill is something we say to make everybody feel better. Reskilling is like when they do layoffs and bring you into the room and give you a packet of jobs, potentially. It’s just to make people feel better. No one has ever reskilled anyone, ever, and it’s harder in AI.

So I call it invisible unemployment. It may benefit us as VCs, which is a terrible thing to say. It may benefit us—it will make our companies more efficient, make us more money, and make them move and iterate faster. But I am really worried that by the end of this year, we’re going to feel, smell, and live in this invisible unemployment, and it’s going to be a big deal.

How do you think that first shows up, Jason?

Jason Lemkin

Well, it’s showing up literally: kids can’t find jobs in college. Unless you’re the best, this is true everywhere, but if you’re the best, you have infinite job offers right now. If you’re at the top of your class in math at any school, you will be found by Anthropic and OpenAI. You don’t have to have an AI apply to a job. They will find you. It ain’t that hard.

But for the other 99% of your class, why do we need you with Claude Code? Why do we need you? Why do we need any SDRs? We don’t need any of them. There’ll be—this is the year where we will see the end of so many entry-level sales jobs. We still need AEs. We still need people knocking on doors. We do not need 21- and 22-year-old kids sending emails.

Those jobs will disappear. All these entry-level jobs are truly disappearing in front of us. At the same time, what we’re seeing is that a lot of the folks the 3 of us have known—especially Rory and I—who are senior executives aren’t able to reskill. They can talk the talk on LinkedIn, and they are quietly leaving the workforce.

We see a lot of folks we know just leave. They’re saying, “I’m moving on from this role at wherever,” and there’s not going to be another role for them. There won’t be another role for their 2021 toolkit. There’s no job for them.

Rory O'Driscoll

The top end, the senior executives, and the entry level are going to be under massive stress in 2026. Massive stress. My advice to them is: stay wherever you are. IBM reported that turnover was almost 2% last year. No one left IBM because they know they’ve got no other job.

Agreed. There’s a lot in Jason’s point, and there’s a lot to unpack there. One of the things is that I think the unemployment numbers always get confused because there’s unemployment, then there are people who are still in full-time study or work, and then you have to ask whether they’re actively looking for a job and all that. One of the real tells on unemployment is the quit rate. People don’t quit their job when they know they won’t get another job, and if you look at the quit rate, that’s a really good tell that, oh my God, I know where I am is good, and I’m leaving, right?

That’s the first comment, and I think Jason’s right there. It is interesting to look at this in the context of the discussion we had about AI engineers getting paid $1.5 million, maybe $4 million, really. What Jason said is that there’s a coterie of people who have the secret skill, and the demand for them is infinite. Then there are folks who don’t have quite that level of knowledge.

If that trend doesn’t change—and I think it can and it will—that’s not a great societal trend, to state the obvious. It’s this concept of the overproduction of the elites. We may well have put 40% of 18-year-olds into college for a world where only 30% of them are going to get jobs, and therefore 20% have just wasted their dollars on a mediocre degree from a mediocre college in a mediocre subject. They’re going to struggle to get work. That’s a real problem.

It is different from the—I mean, I love the way Jason distinguished between the last-job people. When you punch out of your job, there is no retraining at 55. If this VC thing doesn’t work out for me, I’m just going to take my computer and go off into the night, right? I’m not going to retrain.

Yeah, we’ll vibe-code together, Jason. I think that’s true, and I think it’s tragic because you see folks in that 55-to-65 range not having a job and not yet being eligible for Social Security. That’s a pretty tough time, and I have a lot of compassion for that.

On the other hand, if you graduated in computer science from Stanford in 2025 and you didn’t take enough AI classes to at least be relevant, what the freak were you thinking? And B, what was Stanford thinking? I saw a really interesting program where they were taking a bunch of computer science graduates and then doing a 1-year AI booster program to basically take their skills and reskill recent graduates.

I think colleges need to take responsibility for making sure they’re getting people out with relevant skills, because there clearly is a massive shortage of AI-trained engineers. If, for the last 4 years, you’ve been sitting there at Stanford and they’ve been giving you CS 101 and then teaching you how to build compilers, that’s fine. But if you didn’t cover AI such that you could be credible in the space, then WTF, right?

I think youth unemployment is a different thing. I was more bullish—I was less concerned than Jason—and I’ve changed my mind. I think Jason is right there. I don’t think mass unemployment is taking place because of AI, but I think there are a couple of key areas. One of them is customer support, and that’s not that visible given the kind of people who do that job, but it’s brutal. Another is high-end knowledge work for recent graduates, and they are visible.

I think Jason is right that even though it’s probably 2.3% of the total workforce and a small increment to unemployment, because those are highly educated, highly articulate 23- and 24-year-olds, you’re going to see, to Jason’s point, a lot of tension on this issue if it continues.

You know, the AI billionaires, all this talk about a wealth tax—if I was pitching a wealth tax to a bunch of 23- and 24-year-olds, I’m not saying I support it. I’m the exact opposite. But I know I could get a whole bunch of 23-year-olds who just got a Stanford degree and don’t have a job pretty mad pretty quickly by talking about—

Jason Calacanis

Why do you think David Sacks moved? He didn’t just move. He saw the future. He’s like, “I’m going to Austin.”

Rory O’Driscoll

Yeah. You’d say to them, “These guys have built a future that doesn’t need you. They’ve left you behind. They don’t give a damn. The only way to save yourself is to take some of their money.” That’s what you’d say.

You’d just be like Huey Long—Huey Long in 1934—and every single Democratic populist since the dawn of time. You’d be like, “We need their money, right? They shafted you.” Mamdani just did a brilliant version of that in New York.

But yeah, I do think it’s an issue. I believe 1 thing very strongly. Unlike 55-year-olds, 22-year-olds can reskill, and you can save yourself. We’ve all been through periods of our lives where things didn’t work out. So I do think they have to have a little more—if at 22 you’re saying it’s not fair.

Woe is me. Then I think you need to have a lot more agency. So that's what I believe. But I'm not saying that's what they believe. And to my point, Harry, I'm not saying the populism would be right. It just should be anticipated.

But the problem is, of course, a 22-year-old is the best to reskill, right? The problem that I've seen with my kids in college, but more importantly with the founders I work with—I think all your founders say the same thing—is, listen, people have not grinded. Outside of Harry and the 996 world, we haven't grinded in a decade.

Kids don't—most of them don't want to grind. The thing is, these entry-level jobs, these AI jobs, not only might you not be skilled for them, they're hard. They're more work than the old jobs, they're harder jobs, and they're more hours and more intellectual and more work. Throughout history, most people have not wanted to work that hard.

Rory O’Driscoll

I think that's true. But there's also another impact that I've been thinking about, which is that the founders of these companies are themselves around the same age, typically, and they're grinders. One of the really subtle things is that I think of that kind of 20-something from a good college who's AI-smart as the hive mind, and they know everyone else who's a grinder. They're all their compatriots; they're their cohort peers, right?

And I think a brutal conundrum is: as a 40-year-old manager, can I tell the grinder 22-year-old from the non-grinder 22-year-old? Maybe I can, maybe I can't. But the grinder 22-year-old at MIT—take the guy who founded Cursor—he knows the 5 people in his class who were smart and the 95 who didn't grind, and he's only going to hire the ones who were smart.

The scariest judge of young talent is young talent, right? And I think one of the really interesting things is, you talk to any of your founders who are 25, they're marked to market on their team and on other people they know, because that's the generation they grew up with, and they're founding these new companies now. They're going to hire the best and they're going to discard the rest.

What's happening also—this is what I'm seeing at the companies at scale. If you're that way, you used to have to give up because you needed 1,000 employees to get to meaningful revenue. If you can get there with 200 employees, you don't have to give up.

Or you can wait until you're much later stage than Manus to give up and say, "Okay, I've got to accommodate lifestyles and any interests outside of work and anything else we should do as human beings." If you don't need 300 people to get to $100 million in revenue or $20 million in revenue anymore, you just don't need 300 people. So you don't have to give.

So that means no jobs for those people. Rory, we bring Jason in to bring a positive element.

Rory O’Driscoll

Totally. No,

Jason Lemkin

Absolutely. And the scary thing about his negativity is, I think he's—

Rory O’Driscoll

Right, increasingly correct. I mean, I'm changing my opinion on this one.

Do you remember on the first shows we did together, you argued like it was—

Rory O’Driscoll

And I still think long-term it'll be fine, but I think my big concern is the short-term dislocation. I think Jason is right about the reality for people watching this: the perceived, lived reality of the undergraduate graduating in 2025. Jason is articulating exactly their lived reality. Okay. Yeah.

Jason Lemkin

Listen, you may think I'm negative. I'm excited about the products. I just—maybe it's not my place. I want to help folks that can jump, that can get to the next level. That's my goal.

I want you to not hide from it and not be grouchy about it, because it doesn't help to be grouchy about AI. It's not going to stop the GPUs.

When you, Marie Antoinette, and Peter Thiel have been dragged out to the guillotine, just remember that optimism, big guy.