Anthropic 对决五角大楼：谁赢？| Cursor ARR 达到20亿美元 | Block 裁员40%

Rory, last week, this man predicted one of the biggest layoff announcements in history before it came. He is that good at predicting the future, so this week is going to be even better.

1. Anthropic vs The Pentagon: Who Wins

We're going to start this week with Anthropic going to war and being banned, and the issue with the Pentagon. Rory, do you want to provide some context for everyone? What was actually happening with the battle between the Pentagon and Anthropic before we dive in?

Rory O’Driscoll

Okay, yeah. A reminder of the context: Anthropic actually has a $200 million contract with the Department of War. In a negotiation around that contract that culminated in a rupture last week, Anthropic was seeking to impose a couple of conditions on the use of its model that the Department of War was not willing to sign up for.

Specifically, the 2 issues, as I understand them—and again, I'm going to preface all this by saying, “Unless you're in the room, you should assume that everyone who was in the room is spinning a version of this that makes them look good.” With that caveat, the 2 issues appear to be Anthropic wanting to specify 2 things. First, that you could not use the software—their models—for mass surveillance. Secondly, you could not use their software for autonomous weapons.

Those were 2 restrictions they wanted to put on the use of their models. The Pentagon's counter was, “We want to be able to do anything that is legal.” That was the bid-ask spread, and it didn't converge. On Friday afternoon, the Pentagon broke off the discussions and effectively threatened—though, as yet, I understand it hasn't been done—to, at a minimum, cancel the Anthropic deal, but also to issue a supply-chain threat such that other vendors selling to the Pentagon or even to the U.S. government couldn't use them.

Depending on how that threat is interpreted, it's either a mild loss of a contract or something approaching thermonuclear. It's not yet clear how that happened or, if it happened, what it means. Those are the facts, simply put, as distinct from the editorial comment.

Was it the right decision by Dario to have these principles and to stand by his beliefs in this way? Sam has since swept in and done a deal, and we can get to the repercussions on the consumer side, but Sam has done a deal in replacement.

[Speaker?]

I think he got swept up in a tough position involving his team, his employees. Dario, clearly, since the day he was the lead safety researcher at OpenAI, has felt that safety is profoundly important. In fact, they refused to release Claude before ChatGPT because he didn't think it was safe. They only released Claude under duress.

Finally, he goes to his team and says, “Listen, at least we can make weapons safer. It's going to happen one way or another. There will be AI in weapons. This is my understanding.” He convinced the team that what they would do was make them safer.

That was the justification to a lot of folks in the AI space. Even Elon said a little while ago, “I wish AI hadn't happened, but since it is, I'm all in on xAI,” right? So if these AIs are going to be out there in the military, we'll make the safe one.

I think it just got pushed to the point where—I think he has his personal beliefs, but I don't think he could go to his team and say it was true anymore. It was just pushed too far to say, “At least we'll make it safer than our competitors. At least this will be safer than xAI or ChatGPT or some tortured open-source thing, because at least the good guys will be in there and we'll make sure we don't do the wrong sort of mass surveillance, and that when we do target folks—which we've been doing with some form of AI for years—at least we'll minimize the collateral damage.”

I don't think he could say it with a straight face. So he said, “I've got to step out on this one. I'll lose my—Not only do I not believe in it, but I'll lose my team.”

If that's the framing, then I can answer the question concretely. I think you're right, Rory, which means he was wrong. I totally agree.

One of the organizing principles of Anthropic has been this almost messianic belief in the threat of AI, and yet the willingness to develop AI because they're the only people who can be trusted with it, right? I find that weird. But I'd be an idiot not to recognize that this has been an extremely successful unifying principle of the company.

That's why they still have the 7 founders, when the other guys across the street have lost almost all but 1 or 2 of theirs. Whatever it is that's in the water at Anthropic, it's working and it has created unity, right? He would be a fool, in terms of running his business, to blow that unity, and the man is clearly not a fool.

So I get it, Rory. You're right. He had this belief around AI safety that is not just a widely held belief in his company; it's a core organizing principle and arguably even the engine that has driven the enthusiasm to build that company.

Given that, I think it was naive to try to sell to the government at all—to the Department of Defense at all. Because they, too, unlike a lot of us who are just ordinary people, have an organizing principle, which in the case of the Department of Defense, or Department of War, is defending the United States of America, right? They believe, correctly, that that's the task they've been charged with.

The idea that someone's going to waltz in and say, “I want a $200 million contract, but I'm going to tell you the safe way to use this,” is absurd. I think the Department of Defense was absolutely right to say, “We are not signing up for some bunch of guys in California to tell us what to do. We are the Department of Defense, empowered by the president and the Constitution to defend this country. Go away, little boy.”

So I think they were entirely right. If you wanted not to sacrifice your principles, and if you believe that you have really nuanced opinions on how AI should be used in war, then you probably shouldn't sell to the Department of Defense, because they correctly don't give a damn about your principles.

I feel weird saying this, because I actually love Dario's writing and his speaking. He's some of the clearest and best writing I've seen on almost anything. I happen to know, because he reads as a big believer in Richard Rhodes's The Making of the Atomic Bomb, a book they share, right?

When you read that book, the first conclusion you should come to is that General Groves didn't give a rat's ass about the scientists. He humored them, gave them what they needed, and let them not wear uniforms because it made them happy. But when it came to the crunch, they weren't even in the room when they decided to use the bomb.

Thinking that you're going to have the luxury of getting to be part of the decision is unrealistic, naive, and actually constitutionally wrong. I mean, I don't love the current administration, but they were elected by 80 million people.

And, as you remember, Tony Benn—the English politician, a Labour, very left-wing socialist whom I don't like in any dimension—had a great phrase for the civil servants. He turned to them and said, “And who voted for you? And who voted for you, guys at Anthropic? Because 80 million people voted for us, and the Constitution says we're here to defend the country.”

So I think drifting into this situation, given that that's the core organizing principle, is probably a mistake. It might work from a business and marketing perspective, because you've got a huge amount of outside lift from it. On that basis, it's been great, but this is the big but: you're now a little bit at the mercy of how the administration wants to handle this.

The thing that's in your favor is that they often have a short attention span and move on. The thing that's against you is that they're quite hard-nosed sometimes about grinding you down, and if they want to grind you down, they can.

I don't think they should. I think they would be wrong. I think they should say to Anthropic, “You're not in our business with us. That's fine. We'll keep our $200 million, but we won't grind you down, we won't declare you a supply-chain risk, and we won't try to destroy your business.”

But you've drifted into a situation where it's at least plausible that they can. I don't think that's a good idea.

If you were an Anthropic shareholder today, would you feel more or less confident today? You mentioned the lift. For context, they are now number 1 in the App Store, ahead of ChatGPT for the first time. They have a consumer brand that consumers actually care about at scale for the first time, and Dario has presented himself as a very thoughtful and progressive leader who cares about safety.

[Speaker?]

Same expected return, wider variance. By definition, the risk increased, and therefore you should be unhappy. Even despite the lift, you were doing really well a week ago. You were clearly the dominant company.

2. Why Salesforce Could be the Best Buy in Public Markets

You made entire stock indices move at your will, right? You write a blog post and every legal tech company goes down 20%. You had a good thing going, as they say in that meme, right? Now you've drifted into a situation where you've got this ongoing fight with the government, which has way more power, as you're suddenly discovering—the new DOD. I don't think you're happy about the increased risk in your deal.

I think we've entered an era where no one really cares as long as the numbers are there. No one cares. There are fewer ethical concerns; there are fewer concerns of any kind. If I just got my March 1 investor update in Anthropic and we had just crossed $16 billion, my email back would probably be, “Great job,” with 3 exclamation points and maybe an emoji or 2 if I'm on my phone. This is the world we live in.

We don't have the rights. I don't think almost none of the money in Anthropic has any rights whatsoever. With most startups, it no longer matters what I think as an investor. It literally doesn't matter. Only in the struggling ones does it matter what I think, and then you don't want to care.

Yeah. [Laughter.] Then I have my board partner. But you're a cynic. No, to be fair, I think the question is not, “Do they care about your opinion?” The question is, “How would you feel?” And I think you're right: if the numbers go up, you're fine. But you just have to accept the fact that, in return for a significant boost in popularity, you have a mild existential risk that you can probably beat in the courts. Nonetheless, if the people on the other side of the table decide they want to be vindictive, they can make things very hard for you.

Just one thing that I was thinking—I had to highlight it if we get to Block later, right?

Guest 2

We will. If we talk about the power of labor versus management—and I know we don't like these terms; they sound very Smithian and stuff—but it's true. Anthropic, OpenAI, and xAI are at the extreme end of the power of labor. People are leaving 8 figures of vesting stock at OpenAI because they want to work somewhere else after 13 months or 6 months. This is unprecedented. Even at the hottest startups, people used to stay.

On the other hand, for non-AI folks—for Block, we have the least power versus management ever. You're less than fungible. You're not even wanted, right? There's a subset of folks—and listen, Anthropic's probably at $15 billion—who are an important subset. It's very weird: labor has all the power, and Anthropic has done the best job of anybody.

A lot of the xAI founders are gone, too, right? This is like 2021 all over again. You've got to do whatever it takes to retain the team—everything. If they want you to take on the Fox News guy, Pete Hegseth, you do it. If they want you to tap dance on the roof of the company, you do it to keep the researchers.

What Maggie told us—I didn't even know this—when we were in London with Harry, at OpenAI, they don't even let the salespeople talk to the engineers, the researchers. They're not even allowed to attend meetings in the engineers' building. Their cards don't work in the building where the engineers are because they don't want to be bothered by the go-to-market team. They've got to go to the building next door. [Laughter.] That's protecting your talent at all costs, right? So that's why I think Dario had no choice. I don't think the team would tolerate anything else.

First of all, I love the labor-capital distinction. You said it sounds very—it's really more Marxian than anything else—and I'm going to go with it: capital versus labor. I think you're right. It's super interesting to see that, in a situation like Anthropic, capital is weak and labor is strong because those are the most talented people on the planet, with the skill that you need the most. And you're right, in many other organizations it's weak.

But I'm going to remind you: if we're going to do political economy—not politics, but political economy—classical political economy says there was a third actor. It's called the state. And I think what you're seeing very clearly now—I just want to say this very clearly—is that the state is more powerful than Anthropic. That's the lesson you have to learn.

You can have all the labor power in the world. It's interesting: if you listen to some of the folks talking about AI, even 2 or 3 years ago, they were hinting at this. They were asking, “Will the state intervene?” I think the state has extraordinary powers because it's the state. That state has, as Weber points out, the monopoly on violence within any geographical area.

In terms of appeasing your employee base, you've got to do what you've got to do. But drifting into a conflict with the Department of War over something where I think, logically, you're in the wrong because you're trying to interpose yourself over and above the people who are charged with doing this is just a bad place to be. I would want to tiptoe back out of there.

If all it costs you is a $200 million contract in the context of a $14 billion business, you should declare a win and move on. Because if the state decides—for all this talk about AI being so scary, let me tell you what we learned in the last week. AI might be scary in the theoretical sense: “Oh, it could be used for this; it could be used for that.” The state is scary in the real sense: we have laws, we have men with guns to enforce those laws, and we can take your company if we want to.

I bet you, even though you look at those laws, like the Defense Production Act, and you look at being declared a supply-chain risk and kind of go, “I think I could win that one in court,” it's still a sobering moment when you realize that the person in charge of the most powerful military on the planet thinks that you are a supply-chain risk, wants to invoke powers granted to him by Congress to do that, and you've picked on them. I would have preferred to skip that. And you've got David and the rest coming straight for you. Nonstop, right? It doesn't help.

When you're talking about the power that labor holds, what you're suggesting, given Sam sweeping in and taking it, is that he does not face the same pressure from his employee base.

Guest 2

No, he did. His team hated that he did that, right? He had to immediately say that he's going to unilaterally change the terms of the contract the next day. I mean, he's a very smart guy. He's signed a few contracts in his day. That was a numbnuts thing to say: “We're just going to change the clauses ourselves on ChatGPT so that they say what we want them to say.” Good God.

3. Was Sam Altman Wrong to Take the Deal

I agree with Jason there, Harry. I think what the last 4 days have shown is the power of labor at OpenAI, too. And it's ironic, because the first time we saw the power of labor at OpenAI was the power of labor to insist that Sam come back, back in 2023, when you had that magic moment when labor basically said, “We want to pick our boss,” and the owner said, “You're right. You should pick your boss.”

So was Sam wrong to do that deal, in your mind? You said it was naive of Dario to enter it. Do you think Sam should not have entered it as well?

Guest 2

It's the most entertaining of all possible outcomes. He's probably right on the merits. And, as often happens in life with a certain kind of person, it's only when you're absolutely right on the merits that it can bite you.

When you actually believe what you're saying—and I read what he's saying about how, in a democracy, the final arbiter should be the government, not individual people—I actually think he was right on the merits. Which, of course, means, pleasingly, it's the one thing that's going to bite him in the ass. If you see what I'm saying, ironically, the one time I think he was entirely correct in terms of how companies should interact with the Department of Defense, it's actually going to be the thing that bites him.

I think actually signing the contract on the merits wasn't awful, but you're right: his labor is not going to consent to that. In terms of opening Pandora's box in his company, he was wrong. I think he saw that, all of a sudden, there was a break, which happens to every founder. You see these moments in time when your competitor goes down for a week or screws up a key deal, and he did what most of us in B2B would do: he pounced on the deal.

He immediately shot the email to the CEO: “Grok was down all weekend. I can have you up on our infrastructure on Monday.” I'm actually going to argue something that's going to sound naive. I think it wasn't just a pounce while the other guy was weak; it was roughly an attempt to do the right thing. As I say, I think the odd thing is that he was right on the merits. It's like, we should be willing to sign deals like this because the government should decide. But his labor is not going to consent to that.

I think it's just everything is accelerating in the capabilities of these models.

Everything has accelerated since December, right? Everything’s gotten better. So, it’s just very hard to predict, for a military application, what these models will be able to do in 12 months. It could literally be beyond what we could conceive these LLMs could do. And I’m not saying that that is a bad thing because there are bad actors in the world, but even understanding where it’s all going to go, it’s hard to predict.

Rory O’Driscoll

Hard to predict. Yeah. To ask a political economy question, who should make those decisions on how to use this technology? A privately held company, or someone elected pursuant to the Constitution of the United States of America, with 80 million votes in Congress? Who should make those decisions if they are, quote-unquote, harming—?

I mean, I don’t want to spend all our time on this. It’s tough because there’s no good answer. And we know the last thing we want is Ro Khanna, or whoever his version is on the right, making a hyper-political decision. We don’t want that. We don’t want to politicize it, right?

No, I disagree. We do want political decisions. How do you think we make any decisions? It’s called a Constitution for a reason: we elect people.

I don’t think most of our political leaders even understand how AI works. I don’t think most people in general, even on X, understand how it works. They don’t understand.

Rory O’Driscoll

Hard-nosed common sense again. You like the common sense.

There’s logic and there’s rationality. You’re logically right, Rory, but Jason is rationally correct.

No, I disagree. Governments around the world are not sufficiently versed in artificial intelligence, its deployment, and its evolution to make sound decisions. I think we would all agree on that.

Rory O’Driscoll

I wouldn’t. Look, by definition, if you’re running the government on any individual thing, experts know more, right? Your job is not to be the expert. Your job is to be able to manage the experts, right? And I think it’s the same mistake, frankly, that folks made in COVID, where they didn’t defer to science.

No, it’s: listen to science and make informed judgments. Now, you might argue that the people making those judgments at various times have been mediocre or even bad at making judgments. But the way to solve that is to elect different people to make judgments.

You can say, “I don’t trust the US government, so I don’t want to deal with them.” I’d say that’s a totally rational thing to do, up until the moment when they invoke the Defense Production Act, right? Totally rational. But what you cannot do is say, “I’d like my $200 million, but I’d like to tell you what to do as well.”

Just one interesting thing on the deal structurally, putting that aside, and then we can move on. I don’t know exactly—it was being negotiated before it was signed, right? By the nature of any big deal. So, it was signed last summer. That means Anthropic was probably closer to a $1 billion run rate and still trying to break even, because it was $1 billion in January of 2025.

Signing a $200 million deal, plus $200 million from Palantir in pull-through, plus blocking your competitor from Palantir and others, was probably a pretty big deal at the time. That was probably one where he was honest that he swallowed some of his principles, like raising money from sovereign wealth countries. But it was so important to winning in the enterprise and building this cloud, it probably actually made more sense to take the risk at a billion-ish, right, than at $15 billion, where you’re like, “Oh, you know, everyone’s going to be like, ‘You throw away the $400 million with Palantir?’ Okay, $200 million, $200 million.”

It’s less of a hard, tougher decision at a billion-dollar run rate. Most of our portfolio companies, if they came in and said, “We can grow an extra 40% this year if we sign a contract we’d kind of rather not sign, but it’s totally legal. And instead, Rory, instead of growing 60% this year and having no exit, we can grow 110%,” I think most investors are going to say, “Take the deal. Swallow your pride,” right? And so, it was a bigger deal at the time than it is today, I think, right?

Rory O’Driscoll

Agreed.

But I actually genuinely, now having been harsh, I’m going to compliment them. I don’t think they did it—I don’t think this happened because they needed the money then and they don’t need the money now. I think, like often happens in life, a bunch of people drifted into a situation that nobody fully thought through. And then you end up at this point, even though you didn’t plan to be there.

I think the relationship happened initially via Palantir. They were selling to Palantir, as one would. Great customer to sell to. Palantir sells to the government, you get dragged in, and you probably sit there thinking, “Well, they’ll be reasonable. We’ll go and talk to them about how we operate.” I mean, The Wall Street Journal covered it today. Part of this, as is always the case, is personality clashes. You’re like, “Oh, this is the way I look at the world, and this guy on the other side of the table looks at the world totally differently.”

4. OpenAI's $110BN Mega Round: The Breakdown

Final thing on OpenAI and Anthropic, and then we will move off them. The ultimate dealmaker there is Sam Altman, who closes a $110 billion round, 4x the size of the largest IPO ever. I mean, Jesus, you’ve got to give the man credit. What an absolute machine.

Rory O’Driscoll

Totally. Anything other than plaudits? Is there any money left in the world to fund these businesses? The $200 million and $400 million in revenue seem relatively paltry compared to the $110 billion he just raised.

Yeah, I mean, there’s some meta. Obviously, they’re not public, and obviously the benefits are accruing to a relatively modest number of shareholders, right? All of that stuff. But it does make you wonder if we should even be talking about the IPO window or any of this crap. It doesn’t even matter when OpenAI can raise $110 billion.

We’re all worried some B2B company can’t IPO at $400 million, growing 40%, at $3 billion. What difference does it make? It’s so minuscule. I mean, it matters to the humans that work there, but it’s not even relevant to the economy when OpenAI does $110 billion. And they’re going to do another one before they IPO, probably before they IPO if they can, right? Maybe not. It’ll either be an IPO or another round like this, but it’s a force of nature, right?

Rory O’Driscoll

It’s clearly a force of nature, and it is astonishing. Any way you cut it, when you add up venture dollars raised year to date and invested year to date, you kind of go $30 billion on Anthropic, $110 billion—if you call it VC—it’s $140 billion. It’s more than US venture for all of last year. So, these are astonishing numbers, right?

I do wonder, and you’re right, it’s a very odd circumstance to have a private round be 4x the size of the largest IPO ever, right? And it does make you wonder how that IPO gets done, right? We’ve talked about this before, right? And it’s interesting—there are little nuances here.

Amazon’s, I think, $40 billion or $50 billion. Some of it’s up front, some of it’s contingent on either an IPO or AGI, which is just one of those weird things where some of the money only has to go in once the company either achieves AGI or goes public, right? Which, to me—and Jason, you mentioned this a while back—exudes a little bit of the IPO support coming in. In other words, is this really $15 billion now and $35 billion as kind of a placeholder for the book on the IPO or something?

There is always precedent for NVIDIA investing $100 billion and then kind of walking back to “up to $100 billion.” Is it that again? I’m genuinely confused.

No, it’s not quite that, because NVIDIA talked about investing $100 billion. Remember, I typed out—I printed out—the press release, and then they ultimately invested $30 billion. So, they folded back slightly.

Amazon came in for $50 billion, but of that $50 billion, it’s not just a press release. It’s actually a real commitment, but it’s a commitment subject to closing conditions that are either an IPO or AGI. So, that’s not money now, right? If you needed to pay your workers next week with that $35 billion, it wouldn’t be there.

When does this go public? You’ve got to believe that the next round for OpenAI, Anthropic, and SpaceX are all public offerings. Articulate who writes the next check privately and what the investment thesis is. I don’t think there’s more capital there. I could be wrong, but it’s hard to imagine another round after this. So, by a process of elimination, at some point you’ve got to go public. The next round probably is a public round.

An important point, and an asterisk that we have a double asterisk to: it’s a $110 billion round, but $50 billion from Amazon, of which only $15 billion is up front, right? The rest is AGI or IPO.

But Amazon’s own free cash flow has fallen to $11 billion a year because it’s spending so much. So, OpenAI literally had to go to Amazon, and they had to go to everybody on the planet—NVIDIA and so on—and they’re exhausting all of them. Amazon doesn’t really have the free cash flow to fund its commitment, right?

5. Who Has a Bigger Valuation Premium: Sam Altman or Elon Musk

So, it’s not that Amazon could do $150 billion or $300 billion. We’re reaching a limit here where the folks that did this round literally cannot do any more, right? And this is probably why Jensen Huang walked the $100 billion back. There’s a theoretical idea where it makes sense, but the world has changed, and that’s a lot of free cash flow even for NVIDIA—$100 billion, right? So, we’re reaching their limits.

I know this is an absolutely stupid suggestion. You can both fire me for it. Is there a Sam Altman premium? We’ve spoken about the Elon premium before. If Sam were to say, “I don’t want to do this anymore,” is there a Sam Altman premium?

I don’t think in the same way. His genius has been deal-doing, and having the oomph and the chutzpah to raise $200 billion-plus for a privately held company.

Guest 3

He’s obviously technical. He has a computer science degree, from memory. I wouldn’t swear to that, but Elon’s genius at the margin has been the ability to pull off amazingly complex engineering projects sequentially. So, I don’t think there’s the same premium there.

You could even, to make it a little more pointed against Elon just for a second, say that OpenAI trades at $700 billion pre-this-round, but that’s because it’s hyper-growing. I can’t remember the exact growth—from $3 billion or $4 billion in GAAP revenue to $12 billion last year, to a $20 billion ARR run rate. It’s 30 or 40 times revenues for hypergrowth.

Astonishingly, I continue to point out that Tesla continues to trade at 10, 12, 13 times revenue for declining revenue. The Elon premium is way higher relative to the performance of the underlying asset. You asked who’s got the better premium, Elon or Sam, and actually, the easiest way to ask that question is to say to yourself, “Remove the person with the premium.”

Tesla trades at $1 trillion today. I think if Elon died tomorrow, it would trade at $200 billion. OpenAI trades at $800 billion today. I think if Sam Altman died tomorrow morning, it would trade at $600 billion. They’d say, “We better get someone else to run the thing.”

Bret Taylor would be in charge, right? Bret Taylor would be great, and he’d figure it all out. They’d build a company and go public. If Elon goes, who’s going to make the robots? Who’s going to make the robotaxi? If you don’t make one of those 2 things, you’ve got a car company with a declining car product line, flat revenue growth, and a tough market for electric vehicles. You’d be down $800 billion. I think that’s fair.

Yeah. If Sam left, you buy Sierra for $80 billion, 10% of the value, and you roll it in.

Rory O’Driscoll

You buy out the shareholders—whatever the math works, whoever gets what—you just do it overnight for $80 billion, and your problem is solved. Possibly the company’s stronger afterward, other than the fundraising. It might be better.

Guest 3

Yep, agreed. If you have one of the greatest technical CEOs of our lifetimes in charge of it, instead of a very brilliant but ultimately nontechnical founder, it’s tough to beat Bret Taylor’s background. He’s already on the board, right?

But you know what, fundamentally, Jason, I thought you were right. Fundamentally, you’d be like, “We’re just going to build a company now. We’re going to build ChatGPT. We’re not going to have distractions. We’re just going to be fine.” You’re exactly right.

I’m going to push you both before we move on. When does it go out, and what price does it go out at?

Guest 3

I think there’s now a subtle, unspoken rush to the money taking place between those 3 mega-cap companies. Everyone else is kind of twiddling their thumbs on the sidelines until these deals get done. I personally would say the sooner, the better for everyone. I’d love them to go public as quickly as they can and get it over with.

The question isn’t when they want to go public, because I think if they could all wave a magic wand, by the end of this year, all 3 of those companies would wave the wand and say, “Let’s be public. It’ll be safer.” I’m going to say October, and I’m going to say it’s going to go out at $1.5 trillion.

For OpenAI?

Guest 3

Mhm. I think you can trade that, probably, on some of these tokenized bets. I mean, which probably should be something we should do. It’s a logical bet.

They might actually seek, initially, a lower price to not stretch it for an IPO and then see what happens. The SpaceX IPO price that Elon is making up—$1.75 trillion—I mean, it’s a crazy number, but it’s not 2 times the last round, right? So, it may make sense to have a modest step-up, but that’s really just aesthetics. The market will settle.

I do think, thinking on it live now—and listen, I could be wrong on this one—I do think there’s a risk that they’ve exhausted some pools of capital from the round-trippers: the SoftBanks, the NVIDIAs, and now Amazon. A lot of that money—the fact that so much of Amazon’s money isn’t in the bank today, to me, it doesn’t count.

Maybe it sets the IPO up, to Rory’s point. That’s great: to have $35 billion of your IPO pre-sold does make the IPO easier. Let’s be clear, having half your book sold is a gift walking in. But if they’ve exhausted the money, then, yes, they should go public in October. I think it’s right.

Just a comment on the valuations. I’m not in the “Oh, it’ll be fine at $1.5 trillion” camp. Those are such huge numbers that they’re very much predicated on the overall market continuing to be extremely strong.

The S&P 500, despite all the noise within the system and all the SaaS apocalypse, is plus or minus 2% from an all-time high and trading at a 20-year high in terms of any kind of trailing P/E basis. This is selling securities at a time when people are hot to trot and want to buy securities. So, getting it done in today’s market—who the hell knows how it gets done? It could easily get done at an incredible price.

But it is worth pausing and looking at the vast distance between the valuation on any kind of fundamentals basis and the amount of leaning into the future you have to do to get to those kinds of valuations. It’s slightly different between them because OpenAI is growing fast—really fast. Provided it continues to grow really fast for 2 or 3 more years, it all works.

You can pay 40 or 50 times revenues. The fact that I even said that sentence at scale is how that one happens. If there’s an appetite to pay 40 or 50 times revenues at that point in time, you can do it.

SpaceX is even harder because you’re basically having to underwrite a first slug of next-generation technical risk. Admittedly, you’re underwriting it with the person who, in the last 20 years, has proven most able to deliver that technical risk. But you’ve got the whole Starship risk, then you’ve got the next-generation Starlink 2.0 direct-to-cellular risk, and then you’ve got the data centers in space risk.

You need all that time because the existing business is $18 billion, growing at 20%. So, it’s not obvious that it’s worth 100 times trailing revenues for 20% growth. But if that were a SaaS company doing $18 billion, growing at 20% with modest profitability, Jason would be sneering and saying 5 times.

Guest 3

I wouldn’t be sneering. I’d be saying 5 times.

Yeah, you would be. I’d have my head in my hands, but I would not be sneering. I’d be commiserating. I’d be like, “Could I just go back to December 2025, please? Could I just roll back time just a heartbeat, just to the end of last year?”

6. Why We Got the SaaS Apocalypse Wrong?

The eternal, “Oh my God, if I’d known then what I know now, I’d have sold everything.” But, yeah, you’re right, Jason, and it’s a big gap between that and $1.7 trillion. That’s all about what you said: if it was a public company doing 18% to 20% growth.

Jason, before we started recording, you said something and I wrote it down. You said you thought every public company would miss its numbers. Can you just—

Guest 3

They’re just—I think all the public companies—I don’t mean to be negative. I think almost all the public software companies are just going to do worse and worse for the rest of this year.

Workday was down, what, 20%—the largest drop ever after a strong quarter—because they said growth would drop back to the teens, right? Even though they crushed their quarter. You had Eran from monday.com on the other 20VC this week, who was great, but the public markets think that they’re going to keep re-rating growth lower and lower each quarter.

I just don’t see any positives for the existing group of B2B companies. Rory’s point is that we just need more good ones to raise up the median and the numbers, and maybe it’s that simple. But I actually think we got the SaaS apocalypse all wrong.

I think it’s not vibe coding that’s killing us. It’s just that everybody has lost the way to growth. Because of revenue retention, because so many of these numbers are backward-looking, I think almost everyone is worse than it looks.

It’s funny: I wrote it up this week. You could expand the basket very widely, but for the stocks that I follow, the most successful one is DigitalOcean. DigitalOcean is up 28% this year. DigitalOcean had all these years to get to $1 billion in cloud revenue.

In some ways, it’s very impressive to get to $1 billion, right? But on the other hand, it’s a pretty good business to only be doing $1 billion, right? You could be critical, but you need to be radically accelerating growth and profitability at the same time.

You’re looking at these public companies and you’re like, “Who the hell is going to radically reaccelerate growth? Who?” That’s why, when I had to come up with my 4 to bet on, I could only find 3. I’m still going to do 4, but I could only find 3 because I can make the case that this one’s undervalued and that the markets don’t see the inherent value in it.

But the markets are saying, “You guys have got to reaccelerate,” and they’re all going to decelerate. I just think almost all of these public B2B companies are in worse shape than they look. While I wish we didn’t have this SaaS apocalypse, man, I just think it’s worse than vibe coding.

I think vibe coding is one of those minor threats to software companies in our lifetimes, and I vibe-code constantly.

It’s a minor threat. And that’s why they all have to cut half their teams. To our conversation last week, that’s what I said: it’s so pressing. What are you going to do when, for the next 2 quarters, you keep having to drop your earnings estimate and write it down? You may even make the quarter, but you keep having to lower your earnings estimate. At some point over the course of the year, you’re just going to have to cut half your team, because otherwise you can’t make the math work.

I want to talk about the team, but before that, I just want to put 1 caveat out there. You do have to take price into account, right? Let me tell you what I mean by that. It had a very strong quarter, a modest delta in terms of future guidance versus expectations, and dropped 25%. Why? The answer is that it was trading at 40 times next year’s EBITDA.

When you’re trading at a super-high price, 8 or 9 times GAAP revenue, and you go from 30% growth to even—I think the guidance was 23% or 24%; clearly, the assumption would be that 24% or 25% would be the actual result—when you’re trading at 40 times EBITDA, it doesn’t take a lot to knock you off your pedestal, right?

Conversely, a whole bunch of these things are now trading at 7, 8, and 9 times next year’s EBITDA, where even 8%, 10%, or 12% growth can probably make it, at the very least, highly unlikely to decline by 30%, and arguably there’s more upside than downside, right? So I do think: why will they reaccelerate? I just don’t believe they’ll reaccelerate, Jason.

Guest 3

I think there are 2 separate things, and this gets to your Block comment. You’ve got to do 1 of 2 things. I do agree you’re right: you’ve got to do 1 of 2 things. If you reaccelerate enough, then they’ll cut you slack on EBITDA and profitability, provided you have some outperformance or convergence on profits. If you don’t reaccelerate, then you’re right: all you’re selling is a profitability story, and then we’re going to segue straight to Block.

The only way to do that is to look at your cost structure and say, “If my revenue growth is 10%, I can subtract 10 from 40, and I come up with 30% free cash flow required.” So you’re right, you start running the math.

This is brutal because the amount of spend that has gone to cloud and LLMs is like nothing we’ve ever seen before. If you’re a public software company, you’ve captured none of that. It’s a disaster. It’s a Titanic. It’s worse than it looks, because you can’t figure out any way to take all of that spend and turn it into an agent or an LLM that your 10,000, 50,000, or 100,000 customers want to pay you more for. I mean, it is the biggest fail and own goal in our history, and it’s going to get worse.

There are so many public folks who are rolling out a beta or saying, “We’ve added a cloud connector this week,” and it’s a disaster at this point in 2026. You can see it when a lot of these public company CEOs speak. They don’t have the conviction. I was on the earnings call with Marc Benioff. He’s got the confidence. He believes there will be a lift from Agentforce. He’s growing already. Some of it is inorganic, but growth is up and he believes he has a performant agent. I was there in the building—the dude believes, okay?

Guest 4

I hear you, but I can smell the lack of belief in other public-company CEOs. I know Harry can, too, when he interviews them. You can just smell it out of the pores, and as a human being I have empathy, but good God, that’s why I think it’s all worse because of retention. It’s all worse than it looks.

It’s all worse than it looks because you can hide under your multiyear contracts, annual contracts, and pricing increases, but you’ve had so much time. These 3 kids from Stanford figured out how to use Opus. Why couldn’t you? It’s out there. It’s not even that expensive. Why can’t you figure out how to use Opus? Why can’t you put your 20 best engineers and just freaking clone Harvey, ElevenLabs, or Lovable? How hard is it to clone those apps, guys? You all deserve to be fired.

Hang on. I actually thought Owen at Intercom had a really great piece just this week on what it took to take his existing business, gut it, and build Fin on top of it. That’s what I meant when I said, “Jason, it’s hard.” Not from a technical perspective—that’s part of it—but just from the fortitude to make all those changes, to put all your bet on the new thing.

There was probably a year of feeling ridiculous because you’re talking about the small thing and the big thing, and so many customers are saying, “Why don’t you focus on the big thing?” He described very well what it took to be bloody-minded enough to make that happen. You’re right, Jason. I do think you are correct. That’s the kind of attitude that the opportunity is going to require. I’m interested to hear your take on Agentforce either this week or next week or whenever, but that’s the kind of bloody-mindedness it’s going to take to push through, I think, more than ever.

Guest 3

I just don’t see it. I just don’t see it across startups, and I don’t see it in the public companies. I don’t see it. And then, now listen, Eoghan had to come back as CEO. There’s complexity there, too, right? But he’s been brutal that this is a founder market, and it’s hard. Yes, it was harder for them because they had an institutional base he had to partially retain, but also partially abandon. A lot of complexity, but woe is me.

His points were all really good, but you know what? Maybe he got there a quarter or 2 earlier than others because it wasn’t CX and support, okay? So he could see the change earlier than everyone. It wasn’t just going all in; it was also that he was in 1 of the 2 categories that changed the most quickly, right? But what about the rest of the world? You’ve had time.

You can go out there and say on Twitter, “LLMs are fungible, and systems of record should benefit because we own what matters, and LLMs are commodity.” I mean, give me a break. Where’s your hunt? Show me $100 million. Show me $200 million. Show me $500 million. You’ve had 16 months now.

I think Tobi Lütke at Shopify is one of the most brilliant CEOs and technical minds. I think Gustav Söderström at Spotify is, too. I’m genuinely naive here. Are you telling me they’ve missed a beat as CEOs and—

Guest 4

No. Tobi is dragging his company mercilessly into the agentic era, rolling out its agentic commerce. I don’t know how successful it will be, but it’ll be front and center of Shopify, right? It’s becoming front and center. But he feels to me roughly at the pace of Salesforce. He probably should be faster, right? Because it’s a more agile organization, but they’re dragging their companies with real products into the future.

I think everyone else is already starting to fall into the event horizon, into a state of terminal decline, because their teams don’t want to do the work and they don’t have the ideas. They don’t have the ideas. “Oh, Jason, I rolled out AI SDRs.” Great, but you’re a software company. Where’s your agent? Where’s your agent that goes out and does all the work for your customer base?

They’re just headed to the event horizon, and they’re trying to put on the jets, but it’s getting tougher each day to recover.

Well, Roy, we’re going to do a Christmas party, the 3 of us. Jason’s the mover and shaker.

Guest 4

Yeah, he’s going to be the fluffer-upper.

Guest 3

Listen, if you’re really in the game, I will be your biggest supporter. I will use your product. I will support your product. I will say it’s great. That’s why, honestly, for a long time, with no benefit to me, I was so supportive of Agentforce, because we started using it and it works.

I don’t think there are that many people who use 5 or 10 different GTM agents, including Agentforce. We might be an N of 1, and I can tell you the flaws, the issues, and the challenges, but at least they’re in the game. There are armies of FTEs and armies of folks deploying custom agents, and it will get better.

I will tell you, it is still changing. Agentforce is constantly changing. It is not static, but like a real AI company, it gets better every month or 2. It keeps getting better and better and better, like our AI tools, rather than being stuck in beta. So I’m not saying Salesforce is going to turn into 50% growth next year, but I want to see this or better from anyone in software. You’ve got to be doing as good as or better than Salesforce. I think you’re hopeless.

I just want to reiterate my position in the boring middle here, which is that I think the straw man—that these public-company CEOs are ignoring it totally—isn’t really true. That fear isn’t really true. I believe that most of these SaaS companies don’t evaporate or go away.

Provided they get with the program, depending on the industry and market, there’s a market where you’ve got plenty of time, where AI is not going to be widely disruptive from a top-line perspective. I’ll talk about opex next. That goes all the way to markets where it’s obviously disruptive. I think it’s well within the capabilities of highly compensated executives who have their ass on the line to figure this out.

Most of the companies that already have scale should not be banished from this, which is a different statement than saying they will ever regain their 20- or 30-times EBITDA multiples, still less—God forgive us—20-times revenue multiples, which they got in 2021.

Guest 3

Here's the thing that I don't understand about the SaaS apocalypse, but this is what I think makes these companies terminal. What happened through December 2025 is that the public markets basically said, “Listen, we're actually going to give you a growth premium for ever-lower growth.” The top tier in 2025 was if you grew 30% or higher, you were, I think, valued at an average of 25 times revenues.

This held even as, each year through 2024, growth slowed. For whatever reason, the markets would say, “Listen, it's cool. We'll ratchet down what we expect from public software companies.” We'd ratchet down, and then at some point in January, the market said, “No longer. No longer will we ascribe high growth premiums.” It came out of the blue, and this is why everyone is still stunned now, because this deceleration with multiples staying strong if you were in the top quartile worked for 4 years: 2022, 2023, 2024, and 2025.

Facts, not the spin, because you're right—zooming out 20 years.

Guest 3

I see them stunned. I think Harry sees them stunned, too. They just didn't look at the chart, right? If you look at the chart, it's really simple.

No, that's your stock price fall. But no, hang on—stop. First of all, I agree with you on the conclusion. I know where you were at, right? There were 15 or 16 years of the bucket growing at an average of 30%, valued at an average of 6 times revenues. Then there were 2 years of COVID, with the bucket growing at an average of 40%, valued at an average of 20 times revenues.

You're exactly right. What happened is we decelerated not just back to 30%, but down to 15%. Then the market said something stupid like, “Well, we're back to 6 times revenues. It's okay, like it used to be.” You're right, Jason: once upon a time, you were at 6 times revenues and growing at 30%. Now you're at 6 times revenues and growing at 15%.

You sat there for 2 years where, when you do a chart that just shows revenue multiples over time, it's a stupid chart. It looked like things had gone back to normal, but you're right, Jason: the growth had gone out the bottom. Then all that happened last year is people's eyes were opened, and they were like, “It's not a temporary growth decline because of XYZ. It's now permanent because of AI. Oh my God, what were we thinking? You should be at 4 times.”

7. Block Lays Off 40% of Team: AI or Overhiring

That's the movie. It's now permanent because of AI. That's a really interesting statement. We obviously saw, following Jason's statement—I don't know if Jack was listening. I'd like to think that he had heeded Jason's guidance and instructions and announced, obviously, that he was laying off 40%.

Is it truly that, or is it a very bloated organization that needed resizing? Jack has continuously overstaffed his companies, and it was just a resizing with a mask of AI. Jason, you should go first because you called this last week. You didn't name the company, but you called the catastrophe.

Guest 3

Well, look, I think what people missed on Block in particular—I know that their profitability is way up, but I think their top line is only growing around 3%, something like that, right? Everyone missed this. Everyone got suckered in by the press release that said gross profit was up 27%, and that's true. But you only lead with that when you're not growing. When you're growing, you lead with revenue growth.

What are you going to do? I think this story is simpler than it looks. Jack, they're growing 3%, and he's given up on a return to growth. This is my point. Most public software companies will return to growth during the course of this year. They will give up on returning to growth. The CEOs will give up. This is my point: they will give up on returning to growth, and he gave up.

What's your next play? You've got to get more profitable, right? There's a limit where more than 50% operating margins aren't worth it unless you're an IP-licensing shop. But it's the “get it done because I got to get to my next state,” which is hyper-profitable.

My only point is that people reacted to this like this was a high-growth tech company. It has been many years, or at least years, since it's been a high-growth tech company. 3% is pretty— I mean, a SaaS company can at least raise prices 4%. At least we have that in SaaS. We just go out to our customers and say, “Due to the incredible features we've added this year, including changing the colors of our buttons, we're raising prices 6% this year.”

Totally. So, I think there's a lot in that. First of all, I big-picture agree. You have a mid-20s-billion-dollar-revenue company, growing 10% a year ago and down to almost no growth last year. You've given up on the revenue growth, so there's only one button to press if you want to make the stock go up.

A couple of comments, though. One is on employee count. You're right, it had ballooned. There had been a couple of years where it grew at 50%. You've got to calibrate that because, interestingly, if you look at revenue per employee or gross margin per employee compared to SaaS companies, it actually wasn't that far off.

But of course, it's not a SaaS company. It's a financial services company. When you look at it on that basis, where you have all these interchange and pass-through costs, relative to best-in-class financial services companies, there was clearly fat to cut there. And, Jason, you're exactly right—that's why they did it.

Talking about AI, there are 2 separate uses of the word AI, right? Are you talking about AI on your top line or just AI on your OpEx? It's worth pointing out here that this is not an AI top-line story.

Salesforce is an AI top-line story. AI is either going to increase your top line if you build and deliver Agentforce, or it's going to reduce your top line if someone else delivers Agentforce and you don't. It's an AI top-line story.

This is not an AI top-line story. At the margin, maybe a little. I'm not going to say no, but Jason will correct me, like he did entirely correctly on YouTube. But it's Cash App, at the margin, maybe Buy Now, Pay Later, or Square. AI is not going to fundamentally change the top-line offering here.

So, this is not an AI-positive or AI-negative transformation story. All they're saying when they use the word AI is, “Dude, we don't make or sell AI to our customers, but maybe we can use AI to cut some OpEx.” Is that true or not? I doubt it's true at the 40% level.

When you look at it on that basis, it's not a growth story. It's not an AI growth story. It's a “we need to make the profits go up, and therefore cut all these expenses and plan or hope that some combination of people working harder and AI can allow us to service the customers at a much lower headcount.”

Guest 3

Entirely. Yeah, and get the stock up 20%.

The other thing, Jason, you write is that it is implicitly an abandonment of revenue growth as the plan out of here, at least in the short term.

Guest 3

Yeah, I'm not saying that this doesn't give folks cover to make similar cuts. I'm also not saying AI is exaggerated as the root cause, but this isn't quite the play people think it is.

When growth has decelerated from 10% to 3%, you're in a tough spot. I just think where it may inform other CEOs is that folks in the mid to low teens may adopt this. They may throw in the towel not at 3% growth, but at 12% or 13% growth.

If, by the end of this year, they're like, “You know what? This AI is such a force of nature, but I haven't found a path to acceleration by the end of this year,” you may see people copy it just for that reason, because they're out of ideas.

I want to ask you a question, Jason, and I'm deliberately going to do it in a way to separate 2 things: AI on the top line and AI in OpEx. Let's assume there's no AI impact on the top line. It's some financial services business that just has no AI revenue lift, right?

Let's assume they were efficient beforehand, which I think Block wasn't, right? You could definitely argue there was some fat. What's your mental model? If I have 100 people doing X, doing whatever revenue this year, how many people per year do you think a well-run company should be trimming based on AI? Do you have a mental model for that? I doubt it's 40 in 1 year. What's your mental model?

Guest 3

Well, look, just one thing first, then I'll answer the question directly. I do think that even if you don't think that a company like Block is threatened by AI, here's the thing I know even just from my own portfolio: folks are building agentic applications that do what Block and others do, and therefore are displacing them, because that's where the energy is.

There are a lot of products at Block. But if you go back to some of the more simplistic products, like Square, as agentic products process point-of-sale and other transactions and add more value, you can lose even if you say, “The atomic version of my product doesn't really need AI.” It doesn't. But when these agents are doing everything in your industry, I think the agents are capturing so much value that there is limited budget in every space, and they're going to take budget away from you.

So, I don't think even folks that seem immune are actually immune, because the new entrants will find ways to tap into that budget and add more value. In terms of cutting the team, honestly, I worry. I think everyone thinks that they don't need 40% of their team today.

Every single CEO I talk to doesn't think they need 40% of their team, and so I think the Block number will become the default: “I don't need 40%.” It's a gift. I got into a bit of trouble because I tweeted over the weekend to 3 CEOs between 500 and 1,000 people, and all 3 said to me that they were cutting by a minimum of 20%. I think this does create more acceptability around layoffs.

Guest 3

It's the Overton window argument.

Yeah, I was stunned to learn that, in recent memory, this was the largest percentage headcount cut. I saw a bunch of data on that, where there are a few 15% or 20% cuts that were bigger because when someone like Amazon cuts 15% or 20%, it can be 30,000 or 40,000 people. I think this was 10,000 to 6,000.

Guest 3

10,000 to 6,000, exactly. I've been looking at the before-and-after headcount numbers because, obviously, in venture-backed deals, there are occasional horrible moments when you cut by 50% or 60%, but that's typically when you have 40 people and you lose product-market fit. This is the biggest percentage change for a publicly held tech company in the last 20 years.

It's just an astonishing statistic. So I think, Jason, to your point, it definitely expands the Overton window of what's doable if it works. Let me give you a simple example for fun. I was helping a very hot AI company think about growing their GTM team. We get called into this stuff all the time now, right? They're going from 10 people in sales last year to 250 this year. Let me just ask you: how many of them do you think are going to be great?

Guest 3

20.

20 out of the 250?

Guest 3

Yeah, maybe. Maybe a few more.

More. 30%.

Guest 3

A few years down the road, if you have to cut, this is going to be so easy. I know the human impacts are horrible, but I can't think of any company I've ever worked with north of 500 people where 80% of the people were great. Managers of managers always lower the bar. First-time managers always lower the bar.

It always becomes headcount-driven and hierarchy-driven. Forty percent of these people—we just have no idea what they're doing. Even in AI companies, they have no idea what they're doing. They don't even know, in a team of 500 or 800 people, what 20 people are doing in marketing. They don't know.

It's funny because whenever you're this extreme, I always want to argue with you because you're so extreme on it, but sometimes I find myself realizing that you're correct. I was just reflecting on this, going back to capital versus labor. In the 1950s and 1960s, the biggest companies were 300,000 or 400,000 employees, like General Motors. Walmart had a million at one point in time.

The amazing thing now is that Apple has 160,000 employees. Google has 190,000. The trend says you're right, because even before AI, the valuable companies have been the high-IP, low-headcount companies. You look at the trend today and the revenue per employee at a company like Cursor is already about the same as Apple, maybe even a little higher. Apple's about $2.6 million in revenue per employee, and Google is $2.1 million. These guys are already at that level at a much smaller scale.

I don't know if it's as extreme as you say, but the trend is definitely proving you right: more and more valuable companies with less and less people. I don't think it trends to 1. I don't think it trends to a billion-dollar single-person company, but the direction of travel is on your side of the table. I do think we'll see an acceleration of layoffs, with founders appreciating this kind of acceptability around doing large layoffs in the wake of this. That's what I hear.

Guest 3

I hate the word “acceptable.” I think the more common, harsher word is “necessity.” Acceptability sounds like it's a social thing. I wouldn't want to get rid of people's careers just because it suddenly became acceptable. If you're subscale, growing sub-10% and kicking off 10% free cash flow, if you don't take the actions, someone will.

Remember, price clears all markets. I like to repeat myself over and over again. If you're only doing less than 10% revenue growth and 10% free cash flow, you'll be trading at 4 or 5 times, and perhaps 3 or 4 times, and someone will know how to come in and buy you and make those changes. Then, a month later, they'll be worth 8 times.

I think you do have a choice. I think Harry's point is an important one. It doesn't even matter at some level that that's the past, but the Block thing frees CEOs to make what they believe is the right decision for the company without overanalyzing the horrific human impacts. They are now free to sit in the board meeting next week—these meetings will probably all happen in March—and say, “Guys, let's just talk about Block calmly. I'm not saying we should do this, but given that we missed the quarter again for the fourth quarter in a row, what could we do like Block?”

I think the fact that it's 40% allows them to cut deeper. Forty percent is a big number, and I think it allows them to go further.

Guest 3

Yeah, but it's also that these companies are in the past. I think future companies are just not going to hire this way. The weird thing to think about is that there's a chance we look back at 20 years of brute-forcing revenue growth with human-led sales and marketing as an anomaly in the era of software.

In the old days of software, when I was a kid, software was wildly profitable. Then, for 2 decades, B2B software grew like nobody's business, but it was never really profitable. We may look back and see it as just jaw-dropping—the efficiency of AI leaders—and we may just not go back to that old era.

We may also say, “Listen, we may double or triple sales quotas for traditional software companies and just let the chips fall where they may. I'm not going to do the $400,000 quotas anymore that you wanted, or the $500,000. I'm done.”

8. Cursor Hits $2BN in ARR… so not Dead?

Why don't we talk about one of the future companies that we mentioned on the revenue front? I'm confused, chaps. I speak to everyone about their product teams and their engineering teams, and everyone has moved off Cursor and moved to Claude Code. I tweeted about this the other day, and thousands of people commented on it. Everyone says, universally, “Cursor's dead. Claude Code rules all.”

Then Cursor announces its movement from $1 billion to $2 billion in ARR in 3 months. There are rumors of a secondary round being done at $50 billion. My favorite saying, which Rory often said about Benchmark, is, “Reports of my death have been greatly exaggerated,” and I thought of that here.

Guest 3

One comment: maybe Rory has a deeper insight. Second, I will say it's funny because this is clearly one of those situations where VCs judging the world based on their portfolio companies and the real world are out of sync.

Literally right before all this came out and blew up, I had 2 board meetings of my fastest-growing portfolio companies. At one of them, they all joked about how people weren't using Cursor. One said, “Mostly, it's only grandpas at the company that still use it.” The CTO stood up, who's pretty young, and said, “Grandpa over here, I still enjoy it.” He said, “There are a couple of us that still use Cursor.”

Then I was with another company that has blown up all of its AI, and there were 2 people who used Cursor. You heard these stories all the time, and you just assumed that everyone had mainly moved to Claude Code directly, right? Product teams had, too. You just assumed that's the way the rest of the world worked.

That's why Cursor was smart to slip this news release to Bloomberg: they'd happened to double in the last 90 days. Our portfolios are not representative. This is as much as I know, other than that we all made the same jump to the conclusion based on our 40 portfolio companies.

If 60% of their revenue is enterprise, the answer is probably right there, right? This is a trusted tool with a lot of things, including guardrails, that can manage agentic swarms and others. They're pointing out how banks use them and how other conservative enterprises are rolling them out. I don't know this to be true, but if Cursor is the more conservative choice, that's not the worst place in the world to be, especially if you can pick your model. Maybe that's the answer.

It's not my 5 fastest-growing investments. That's not where they're crushing it today versus 6 months ago. But it's so funny that we got it wrong. Numbers don't lie.

The feedback that I got from the tweet was very simple from insiders.

Rory O’Driscoll

It was twofold. One, you have enterprise crushing it, exactly as he said. And, 2, it’s actually just that annual subscriptions and their cycles aren’t complete yet. So there will be a wave of unsubscriptions, and that churn will be real on the consumer side.

But that can’t—Rory’s the expert here—but even if that’s true, and it’s a really interesting topic in general, we should maybe talk about it. You don’t go from 1 to 2 billion in 90 days if you’re having massive churn, like deferred churn at the end of the contract. That’s a different issue that’s going to come down the road.

I mean, that is growth that is just massive market pull—epic market pull—to grow that quickly, right? Just to sign the deals takes so much energy.

Rory O’Driscoll

Yeah, exactly. And I think the lesson is: never underestimate big markets and momentum. If you come into last year with the momentum they had—or the middle of last year—even if things are slowing at the margin, even if at the marginal startup you’re losing to Claude Code, you’ve built your name and you’ve built your brand.

Every enterprise is trying to adopt Cursor. The adoption cycle there isn’t, “I tried Cursor yesterday, I’m going to try Claude Code tomorrow.” It’s like, corporate’s got to approve, purchase orders have got to be raised, we’ve got to have security review it, legal’s reviewed the contract, and we’ve signed a deal.

You’re not going to run in the next day and say, “We should switch to Claude now.” It’s like, “No, we’re going to use this for the year. We’ll do an eval midyear and think about it next year.”

I mean, Jason knows better than me. The bigger idea here is that when you’re in a great big market with a massive trend of adoption over the course of your 2 or 3 years, if you’ve got your share of that, you’re going to do just fine. At the margin, are you probably losing share relative to Claude, at least in the last 3 months since the last set of models? Probably, but it still means a big outcome.

We talked about this 2 or 3 months back. It’s clear that they’re going to be 1 of the 2 or 3 players at scale. And scale gives you a lot here.

Guest 3

I would add: you’re right, Harry. Numbers don’t lie, but I haven’t seen the numbers, right? I haven’t seen the revenue, and I haven’t seen the gross margin. But you can easily see how someone with an early lead would be able to maintain that lead even when you’ve got a super-credible competitor who can bundle their product into an adjacency, taking some of that space, because the market’s expanding enough to cover a multitude of sins.

The knife fight—maybe this is the real sentence, Harry—the knife fight doesn’t start until the TAM is 60% or 70% saturated, which, by the way, happened to SaaS in 2021. That’s when the knife fight started. Suddenly, 1 in 2 of every new customer is a switch. Now it’s time to kill the other guy.

Until then, let’s all go one big happy family. We all have a win rate of 60%. We don’t compete against each other. All those happy things that you hear in the boardroom when the market is so big that we’re not overlapping. Once things get saturated, then it gets uglier.

Yeah, I think there was a tweet or a quote from someone at Cursor. I might be wrong about who said it, but I think it fits in; it’s an interesting point: Barclays had just rolled them out. That was the first agentic coding product that was approved at Barclays.

Banks are a big deal because they have the biggest budgets, but they’re conservative at the same time. It’s an odd dichotomy, right? They’re super aggressive and super conservative. We might have gotten Cursor wrong.

I mean, Cursor is more enterprise. You can run it without data retention. You have full SSO, role-based access controls, and all these things that you’ve just got to build in the enterprise—the right audit logs, the right hooks. And to a CISO or CIO, as long as the product works, that could be 10 times more important than what someone in the other building wants.

Guest 3

And I think the other thing that’s happening right now—good God—is that it’s been in all of these products for a while, but in Claude 4 Opus, these swarms of agents got really good. These swarms of autonomous agents.

Michael Truell from Cursor wrote a post—a tweet article—which seemed kind of weird, talking about how the whole future of Cursor is safely managing these autonomous agents. The early days were clicking Tab to finish a line of code, which was what Codeium and Windsurf got going with.

When you have swarms of autonomous agents, good God, they’re not totally safe. As much as we all love our OpenClaws, they’re not totally safe. The more of them you have that are autonomous, and the more they can touch your data, they’re not safe.

So if Cursor can do both—if it can unlock the power of a swarm of autonomous agents running in real time, running concurrently, and make it safer—I’ve got to think 95% of CISOs want that one. In a lot of conservative organizations, the cranky CTO doesn’t get to make all the decisions like at our startups.

At our startups, the CTO decides what the team’s going to use, right? But, good God, at a bank, what if you repeatedly leak everybody’s confidential data? I mean, it’s exhausting to have to go through that. This is not a joke.

Every agent will leak data if it’s allowed to. Every single agent will. That’s how goal-seeking works. Being the most enterprise leader in these mature spaces, you can win, because 99% of the world doesn’t look like us.

By the way, that “doesn’t look like us”—Jason, sold.

Rory O’Driscoll

I mean, all you have to do is look at Azure’s market share in cloud. I’ve sat in millions of board meetings—not millions, but tens of board meetings—where we’ve talked about, “We’re running on AWS. Should we run on Azure? Our customers want to run on Azure.”

The value proposition of Azure for 20 years was, “It’s not nearly as good, but it’s getting there, and Microsoft has great relationships with every CTO, so it’ll be fine.” You’re exactly right. There’s a large amount of software selling that can be done on the basis of safety and integration and selling all those boring bits.

What you’re dealing with is a massive market where everyone’s making adoption decisions in an extraordinarily short period of time. Let me repeat what I said earlier: they’re both going to crank as fast as they can.

As people make initial decisions and more and more of that market has an initial buy decision made, you’re going to start running into each other. Is that going to be later? The competition’s just going to intensify.

If Cursor went from $1 to $2 billion, I think in the same period this year I saw some Anthropic statement about, “We’ve added, what, $2 billion in ARR year to date?” So both of them are exploding. This is a huge market—the biggest freaking market—and those are the 2 most viable players in it.

Guest 3

Look, it’s going back to last week. It’s the Fortnite effect. Claude Code will keep doing more, and Cursor has to keep ahead. That’s the job.

Apparently, 60% to 70% of Cursor model calls are still using Anthropic’s API. So even of that $2 billion, a lot is flowing back through to Mom and Dad. This is the point Michael Cannon-Brookes made the other day.

And Claude, even if they may not care about Cursor as a team, will inherently build all these features as an enterprise play. So the bubble will shrink, and then Cursor just has to do more. But what we’ve learned this week is they’ve kept ahead. They have kept ahead of the Fortnite storm.

Every CEO out there crying and saying how hard it is—if they haven’t kept ahead, that Fortnite storm’s going to shrink to a pixel, and they’re going to die.

Jason, I think you’re right. That is the sound, but the correct response to that is congratulations. In the great race for what looks like a $50 billion to $100 billion minimum coding prize, you have conquered Level 3, which means that a whole bunch of people who are stuck on a back-end Level 2 now get to go to Level 4 and play again, all right?

It turns out it’s hard to make $100 billion, but they deserve all the credit for being the early winner, cranking, and grabbing the story.

I mean, again, back to the swarms of agents, it’s just worth stating. This is obvious, but just to say it: if you look at the Cursor story, he spelled it out in that kind of post, which I thought was very good.

If your story is not changing significantly every 6 to 9 months, you are probably falling behind. In fact, not probably—you are falling behind. You were autocomplete, then you were an IDE, now you’re agents, now you’re swarms of agents.

These models are moving so quickly, and the technology is moving so quickly, that if you think you can rest on your laurels, every 6 months you’ve got to reinvent. The prize for winning is to reinvent the company from scratch and the product from scratch every 6 to 9 months. Congratulations, it’s a fun game. But the prizes are great.

The one thing I’d still love to know about Cursor—the growth is incredible—is this: without being very 2025 about it, the Claude API is expensive. I’ve measured our API calls. It’s so much cheaper to run something inside Claude. My God, I’m on Claude Max.

Amelia runs out of credits using Claude Desktop and everything. Or she runs out of time; she doesn’t even run out of credits. But what I get for $100 a month, and then I track what some of our API calls cost, and for a good prompt, a good call, it can be well over $1. Maybe for code it could be more. That adds up really quickly.

Whether you think it is subsidized, whether you think Anthropic is subsidizing its own native use or not, their margins are strong; it doesn’t really matter. So that is still the more—like, “Hooray, you hit the top line”—but when you’re still paying probably rack rates, I can’t imagine Anthropic is so desperate for Cursor’s business that they’re giving them a massive discount, right? Harry may find that out through his network and his input, but if they’re paying close to rack rates, which I think they are, because they also say you can bring your key with you. If it’s cheaper, bring your own API key with you. It says there’s not much of a discount. That remains: you’re at Anthropic’s whim because they’re already subsidizing their own product.

Totally. No matter what anybody says, we’re not that price-sensitive today. It was super interesting. I’ve got to give a shout-out to Tomasz Tunguz, who we had on here a few months back. He did an interesting post just today, basically looking at the pricing of top-of-the-line frontier models versus 6-month-old open-source Chinese models, and the 10x, 15x, 20x difference in cost per token.

It was a good piece. You’re right: right now, people are paying full rack rate for the new stuff because the new stuff is so great. His point was that it’s only 6 months from something being so new that it’s amazing to, “Oh my God, you can get it open source for 1/20 of the price.” It really brings home the pace of change and the need to be on top of it.

I think OpenClaw confused things, but the power of running swarms of agents autonomously to build software means the rate of change is going to radically accelerate this year. I know you didn’t literally mean it, but the way you described Tomasz’s thing suggested change might be slowing down, right? And we might get a benefit from these super models. I know you didn’t suggest it, but there was a hint of it. I think it’s going to radically accelerate this year. Radically.

9. How to Pick Winners in AI?

Can you help me understand why, Jason? Because I actually don’t understand. What changed in December?

So, listen, I’ve been on Replit and Lovable since June. When I started, you couldn’t finish anything. You’d ask Replit or Lovable to build an app, and you’d get a lot of buttons that didn’t work. Nothing worked. They’d be fake buttons. Everything was fake, right?

Then it got better and better, and by December, especially with Opus 45, all of a sudden it could finish things. It wasn’t just that you could have it code for hours—whatever, that doesn’t really matter for a lot of use cases. It could actually finish apps that worked. Really worked, right?

Then, on our team, Amelia took it over and built an entire AI VP of marketing that is better than anything on the market, and she finished it without any drama. It took work, but she just completed it. It’s in production every day, doing all of our marketing—not just execution, but it has all the ideas. It’s like HAL. It wakes up every morning and tells us what to do with data. It’s a little freaky.

One of the many reasons was these multiple agents running in a swarm and working on the software together: an architect, a security expert, a back-end expert, and a database expert. Replit just did one thing early—which maybe Cursor did, too: they built a framework so you could use these multiple agents together efficiently before you could do it in Claude, right? You could do it earlier.

Now that someone who’s just very smart can birth an AI VP of marketing that literally works in 6 months, the applications we’re able to build—our jaws will just drop. We’ll just drop. I don’t even know what to do with Demo Day anymore. We literally got a sponsor that just raised $50 million, and their website was all built in Lovable. We were just laughing about it. But maybe that’s not funny anymore. Maybe these apps are so good.

What do you mean you don’t know what to do for Demo Day? I think—I guess what you’re saying is, what’s-his-face from AngelList said software is uninvestable. It should have been on the show. It should have been a topic. I agree, and I’m struggling with it.

What I mean is, you can show up to a Demo Day and build anything you want now in any of these coding tools. It’s really good—so much better than anything anyone would have shown up to a Demo Day with 24 months ago. Twenty-four months ago, half the folks would show up without a product. It was barely working. You could make fun of it.

Now you go, and there’s no excuse not to have something that’s great.

Hey, I could take what we built in a minute, show up to YC, and say, “We built this AI VP of marketing. It’s managed 1 million marketing transactions. It would blow your mind.” I’m raising an $80 million pre, post, post.

Just to be precise, Jason, because the sentence “software is uninvestable”—if software was uninvestable, why bother making the demo at all? It’s not investable. Clearly, what you’re saying is you don’t invest on a demo anymore, because a demo simply means that 24 or 48 hours earlier, you kicked off a coding project and got it done. So there’s no information in the demo anymore, right?

I think, for most cases, I don’t think there’s any signal or information in a demo anymore.

Yeah, agreed. So I think, at the early stage, it does not—that’s what you mean by Demo Day. The software is investable. Obviously, it turns out Microsoft’s stock is holding up pretty well, so software is investable. “Software is uninvestable” is one of those hyperbole statements, right?

Well, what I mean is, listen, one of the reasons I was modestly successful when I started investing is I was a founder who turned into an investor. One of the superpowers you get when you make that transition is that you know what’s better than you. You know the CEOs who are better than you, and you also know who’s better at building software.

So, when I would meet a founder who, week for week, pound for pound, had shipped better software than I did, I knew they had a decent chance. If the demo blew me away and the CTO blew me away, even if they had a couple thousand in revenue, I’d be like, “These kids can’t lose. They can’t lose.”

Jason, can you name 1 founder—1 founder—who is able to create, architect, and maintain agents in the way that you have done? I’m not blowing smoke up your ass, but I think this is where Jason’s like, “Oh, it’s going to accelerate so fast, going to accelerate so fast.” I don’t know 1 founder who has built an AI VP of marketing in the way that Jason has done.

I do think I’m ahead of most of the—listen, there are obviously hyper-advanced folks just building agents who are so far ahead of me I can’t even comprehend it. They mock me on X, and I deserve it. But I would say, for your average AI-focused B2B founder, I’m ahead of them on this stuff, right? Because I’ve been doing it for 6 months longer. They’ll eclipse me later this year, but they don’t get it.

I’m sure you’re getting investor updates in March that talk about how stunned people are with how much more productive their teams are, right? We’re all getting these, and they’re just stunned. “I know we’ve been talking about this, but now it’s finally happening.” It’s not just a few more pull requests and this and that. Their jaws are literally starting to drop in March because of what they built in February and late January with these tools now, because they can finish stuff. They can finish stuff.

When that 10x’s this year, here’s the thing: the rate at which your product has to accelerate will be like nothing we’ve ever seen before. Nothing we’ve ever seen before, right? That’s the real version of software that’s not investable. The real version is that there’s going to be infinitely more software. So your ability to carve out this little discrete—

Software alone will not be a competitive advantage. There will still be extremely large demand for software, and as a result of that, there will be very big software companies. But they will have to do something more than, “Hey, I wrote a bit of code for this particular vertical and this particular use case.” That’s not going to get you there.

It’s going to be some other element of competitive advantage around network, around distribution, around moat, around vertical knowledge. I think that’s the true version of the statement.

Yeah, I think it—listen, at the later stage, investing in ElevenLabs today is a very coherent bet, right? It just makes this idea that you could judge—or at least, if you had the right background, that you could judge early-stage experience better than many and write a check very—

Guest 5

They weren't huge investments, but they weren't tiny. Recently, I had 2 AI investments that did $1 million or more in their first week in market because the demand was so strong. One of them, literally 50 days before, they were showing me a demo. I'm like, “Guys, I love you, but this—I know all the vendors in the market. I put it back in the oven for a little while. I don't think it's competitive.”

And then you roll out and dominate the market. There are a lot of reasons, but that level of pace, we just didn't see before. It's going to accelerate. So how the hell do you pick favorites? How the hell do you pick favorites unless they had a—what's his name?—David Frankel. What's his name, Harry, that did—

David Frankel, yeah.

Guest 5

Yeah, he said he had, as one of his never-too-braggy ones, “Oh, I was the only pre-seed investor in Sonos, which is at $300 million in 2 years,” right? And he's like, “Why did I invest in it?” It's like, “Well, PhD from this, master's in that, understood the whole space. I met him. Everyone said he was the smartest guy in the world.” I mean, that signal still exists in 2025.

Do you want to hear a joke?

Guest 5

What's the joke?

He tried to bring me into that round.

Guest 5

Well, you should have done it.

Should have done it. Oh, no.

Guest 5

But he probably saw that you couldn't see quickly, though. That's a level of depth, right, that you couldn't see, right?

[snorts]

Roy, are you going to have your bets ready for next week?

Guest 5

I will have them for next week, Harry. I will have them.

I'll give you an allowance because it's 10:00 in London and—

Guest 5

No, I appreciate that, though, and I genuinely will. But I'm still up on my WCLD. I'm up about 6%, I think—6% or 7%. I was up 6% at one point in time.

I'm not sure, which is the WisdomTree index of all SaaS.

Guest 5

I just—

Yes, as the idiot version of that. So we're remembering.

Guest 5

Now, I haven't looked at it today, and it's been pretty grim. So who knows? But yes, I will have my act together for next week, because it is late in the day there in London.