Canva 下调增长预期｜Google 人才出走｜Revolut CEO 的500亿美元激励方案｜Musk 的550亿美元 Terrafab

There's going to be a lot of people paying the bill in '26 and '27 for a certain amount of hesitancy in '23 and '24. The only way you prove that you're not dying is by growing. It doesn't matter what you think, big guy; you're in this journey for the next 12 months. Buckle up, because liquidity will only come at the end of the journey.

It must be extraordinarily validating, if you're Jeff Dean, to leave as a non-CEO of a $2 or $3 trillion market-cap public company and have the stock go down by a couple of hundred billion dollars. Google's efforts so far are B+, A-. They're not A+.

This is such an entitled podcast. Oh, poor Anthropic engineer only made $35 million. Go out to the goddamn Panhandle—no one's making $50 grand.

When they say it's not about the money, it's about the money.

Any investment I've made that is not run by a founder is a zero. It's going to be a zero in this age.

This is 20VC with me, Harry Stebbings, and it's my favorite show of the week. Rory O’Driscoll, Jason Lemkin, and me shooting the shit on the biggest news in tech. So what happened this week that we discuss? Canva cuts 2026 growth by a third, ouch, as AI serving costs blow up. Next, Jeff Dean leaves Google after 27 years. It gets worse. Demis Hassabis, the OG of AI, then steps back from Google DeepMind also. God, poor Sundar, that is one bad day at the office. And then Elon Musk, as always, comes out with one of the most ambitious projects with Terrafab, where we unpack the jobs that come from it, the first real installment of $16.8 billion, and just what it would mean for him in terms of structurally not having to beg at the tower of TSMC.

Boys, it is another week. We have Jason joining us from Hawaii. What is the hula? You know, the Hawaiian dance?

The hula? It's the hotspot. HULA means “hotspot of AI.” That's what it means. It's an acronym: “hotspot of the AI.”

Something like that.

Right?

1. Canva Faces AI Pressure

Boys, we're going to start with something other than OpenAI or Anthropic today. We're going to start with Canva, baby. We had Cliff on the show before. Now, Canva cuts 2026 growth by a third as AI serving costs blow up. So, Rory, for those who may have missed this story on Canva, what should they know here?

Sure, yeah. Let's start with the facts and then come to the question. The facts are that Canva, a large privately held company in the creative-suite space, discloses its revenue even though it's private. They were at $3 billion in GAAP revenue last year. Going into this year, they're growing at 30%, and the CEO, Melanie Perkins, disclosed midyear that they're probably going to be growing 20% by the end of this year.

So, as you say, that's a one-third slowdown in the growth rate, but it's still a healthy 20%. The other half of what she said was interesting, which was that they're obviously adding a ton of AI features. Those features cost real money, and part of the reason she claimed for the slowdown in growth was that it was just too expensive to effectively be subsidizing users with cheap AI when, in fact, they're incurring significant costs.

So there was an implication there, which I'm actually going to tease out later. I'm not sure I fully buy it, which was, “My growth rate slowed, but if I was willing to lose more money, it mightn't have slowed by as much.” So there's an implied statement on elasticity there.

But the big picture—and I want to zoom out and ask Jason a question—is that there are 3 massive creative-software companies. There's Adobe, which does $23 billion, growing at 12%, trading at 3 or 4 times revenue. There's Figma, which is also public, doing $1.4 billion and growing at 40%, the fastest. And in the middle, there's Canva, still private, at around $3.6 billion and growing at 20%.

The big question for all 3 of them—and this is why I want to put it back to Jason, who's much more of a user; I've used them, but not as much as Jason has—is: Is AI going to be a feature they can incorporate, or is it a new, new thing that makes them obsolete? To me, that's the meta-question.

It's not about 30% growth versus 20% growth because of a little bit of gross-margin compression. If that was the only issue, we could talk about that. That's a second-order business-model issue. The real question for all these companies is: Are you at 30% on the way to 20% on the way to 10%? Because there's a whole new set of companies doing this, and I know you guys are in Higgsfield. Or is this something you can incorporate, survive, and continue to grow? I think that's the question, and Jason, I'd love to get your thoughts.

I don't know, man. I found the Canva stuff depressing because Canva seemed to me like one that AI was obviously maiming. With every ChatGPT release, every Higgsfield release, and everything else, you can do more and more of its functionality in core AI, right? This is what we fear as investors: that you can do what our investments do in ChatGPT or Claude, right? That's the ultimate fear.

Yet somehow, Canva seemed to be defying that gravity. It seemed like it was a non-issue, despite Emilian and me having churned. We churned. This is a tough one. We churned both Canva and Notion, not because they're not great apps; we just no longer had any need for them in the agentic era. Canva and Notion did nothing wrong at all. Nothing, and we didn't need them.

But until this, it seemed somehow I was wrong. They were defying gravity. But 30% to 20% in 1 year? I love Rory's optimism that they're going to bounce off and get back to 30% or 40% next year when they find a few extra tokens under the covers. I think it's terrifying. I was hoping they would defy gravity, but it doesn't look like it.

I'm defending myself on the optimism charge because no one ever accused me of optimism. I actually didn't say that. What I said was, “That's the implication that they're saying.” I don't know the answer.

I'm going to advance the bull case to some extent, just to articulate it more. Again, what the CEO is saying is, “Look, we can't envelop it in AI because we were using frontier models, and they're just costing us a shit ton.” The first shoe that's clearly going to drop is that they're not going to continue to spend a whole bunch of money with Anthropic or OpenAI—probably OpenAI, given its images, of course, Rory. But they now bought their own model and are building their own in-house, image-focused model, which makes absolute sense. So let's assume they do that.

Well, why didn't they do that last quarter?

Agreed.

And agreed. Let’s just say they’re even a year or 2 late, because I think I want to push back to the big question. The interesting question is if—let’s just say they get it done, and in 6 months their in-house model, which is 80% or 90% cheaper, is just as good at images as the frontier models. Then the question still remains, to your point, Jason: do you think they can cram—I hate being vague—enough AI functionality into that product that you would have retained? Or do you, as a pretty active user, say, “No, I just prefer to go to an AI-native product from day 1”? Because that’s the big question.

You know what’s scary? Something that sounds nerdy is becoming mainstream. Our agents never even suggest these products. That’s the danger. And it’s not just agents. We talk about AIO and GEO. What do Claude and ChatGPT say?

But it’s worse than that. As we become agentic, our agents—we can’t choose everything ourselves. We built our own ad server and ad-generating network that builds our own creative and our own collateral and serves it to the SaaStr community. It’s all built on our agent. It never occurred to the agent to use Canva for this. It never once occurred to it. And so, even with some sort of open-weight parity, at a meta level, I think what’s scary is that the most exposed part of the market is the prosumer market.

Yep.

Everyone is ChatGPT-fluent, and if it works in ChatGPT or Claude, you’re just going to use it. And if you believe these Gartner numbers—and I’m out at a big sales source event, and they have all the data—it may be that less than 10% of the enterprise has even deployed an agentic application successfully. I actually believe that, okay? Despite what we’re seeing in the... Because all the hot enterprise AI companies are still serving early adopters and outliers to a large extent.

On the prosumer side, everyone’s used ChatGPT, and so there’s no going back. My related concern is, if you compare it to Figma, well, Figma missed the quarter in a sense. Figma traded down 20%, but they burned the tokens. Dylan was clear: “Our gross margins are going to be significantly impaired going forward because our agentic products are being used.” It’s not identical, but they took the hit, right? And they’re public. It’s more painful to take the hit when you’re public than when you’re private. I really don’t think Blackbird and friends are going to beat Canva up if the bottom line is missed slightly. That’s an internal decision, right?

You’re right, and there’s a lot buried in it. I just want to unpack it again for folks. One is, you made a distinction. Let’s talk about the enterprise versus prosumer distinction, because you’re exactly right: Figma is much more an enterprise product. It’s kind of large groups of people building software and coordinating. So even if you automate creativity, you’ve still got bureaucracy and corporate processes that you make money off in terms of managing workflow.

But you’re exactly right. Canva is the prosumer: “Shit, I want to generate a flyer, I want to generate a cheap website, I want to generate some kind of content,” and that’s exactly where AI is the most accessible. Because you can just go on and type in, “Generate me a flyer that says this,” and there it is. So you’re right, they are more exposed in that sense.

I guess that’s the Fortnification that Jason often talks about in terms of the shrinking TAM: when you have a dinner invite that you can do on ChatGPT and bundle it into your consumer subscription versus using an additional tool.

The nerdier version was something Amjad said about Airtable, not Canva. In response to the CEO of Replit’s quote about Airtable, he said, “No criticism, but the era of no-code is over.”

No-code was a bunch of tools where, without developers or AI, we could build stuff. Airtable was a no-code database disguised as a spreadsheet, right? It was a wonderful product before AI. Notion is a no-code database disguised as a word processor, and Canva was a no-code way to design stuff. It was a breathtakingly disruptive product. I didn’t need a designer anymore. I didn’t need to know how to do HTML or anything.

But the era of no-code—of things that we humans can do without engineering resources—is slowly winding down, and if it’s in ChatGPT, man...

2. Canva Weighs Going Public

Is it a blessing or a curse that they didn’t go public already, then?

It depends on who you’re asking. But a blessing and a curse for whom, right? I was thinking about this a lot because I knew this question would come up. Because if you think about it, if you’re the founders, doing this publicly is just marginally—arguably marginally—more painful. So maybe you’re happy to be doing this in private, right?

What you’re really saying is this. Let me translate, because we never say this explicitly. When you say, “Should you have gone public early?” what you’re really saying is, “Oh my God, if I’m the venture guy who did this thing at $100 million, like Blackbird or a couple of them—I think Felicis was in early and Matrix was in super early—and then even the guys who came in at $1 billion, you must be like, ‘Oh my God, if we’d accessed that $50 billion valuation in 2021, I would be so gone now.’”

And that’s really what you’re talking about. So, to be clear, some of the CEOs are a little unsympathetic to this line of conversation, because this is their life’s work. Because really, when you say, “Should they have gone public early?” what you’re saying is, “Boy, I wish that the fast money had gotten out while the CEO and the management team would still be there, just in a different forum.”

Good point. Getting out early is a critical question for VCs, right? How the hell do I take something... We can talk about IPOs and M&A, but our shares are illiquid when we buy them. We put them in and we hope, we pray, we face east that we’ll ever get any cash out.

If I’m an employee at Canva or a founder, and the founders have already given away 90% of their shares, they’ve already given it away. They’re on the mission of their life. Maybe in Sydney, in Australia, it’s easier to retain your talent than in the Bay Area, right? Maybe they’re not going to quit and go to OpenAI the next day if you don’t go public and make them a million dollars effectively a year in stock.

Maybe it is better to hide. You’ve built an iconic company that isn’t going anywhere. I mean, I’m just playing devil’s advocate, but Rory’s point is, if I were the founders—if the 3 of us were the founders—I might want to run this thing like Basecamp, right? 37signals. Guys, let’s just hunker down. We’ll increase profit sharing. Probably the VCs wouldn’t let you get away with it, but if I hadn’t raised a ton on my cap table, it might not bother me so much.

Yes. And I think one of the totally separate threads is that public markets have to be more attractive, not just for people exiting like the VCs, but also for founders to be able to reignite and reopen the window. And if it’s rational to say, in certain circumstances, it’s easier to be private, then that just probably weighs into the calculus when you decide, as Canva could have done, whether you should have gone public in 2021.

Again, to repeat, for the most important people in the company, who are the founders, who own effectively the company from an entrepreneurial-oomph perspective, regardless of the cap table, I don’t know if that matters. I don’t know if public versus private matters nearly as much as this being the platform-shift challenge of all time. And assuming you do get almost-free AI via your standalone model, you’ve got to figure out, to Jason’s point, a product in the next 12 months that’s as easy and as accessible to your user base, because you know the segment of the market you’re addressing.

Yeah.

It has to be as easy for that user base as ChatGPT is to generate the products they generate with it. That’s your challenge.

My point is, I don’t even think it’s about easy. I just think it’s about the bundling of consumer real estate—where they spend time. I just interviewed the president of Uber. What is his single biggest fear? It’s actually the disaggregation of UI, or the removal of UI, where you say, “I want a car,” and ChatGPT automatically routes you to Lyft, Uber, or another provider based on price.

That’s what our agents already do. They just routed us around Canva.

And so my point there, Rory, is that ease doesn’t actually matter.

No, I know. You’re exactly right. Look, to be very clear, going back to something Jason—

Yeah.

If the AI models become the universal—well, and ChatGPT, let’s say, which is more consumer-focused—if that becomes the universal interface for functionality, then you’re exactly right, and all model choices are back-end choices. Then you never even get the chance.

I don’t know if it does. I could be wrong, and in fact, this is probably an area where I’m still trying to triangulate. I don’t know which things we do separately in the West, because it’s interesting: China, obviously, in mobile has a single super app for everything, and you do everything through WeChat.

But just to pick on 2, what I’d call high-cognition tasks that are very different—actually, polar opposites—one is creativity, building something creative for a consumer, and the second is doing your taxes. There is a credible argument in both cases that ChatGPT can suck those revenues up, which is why Intuit was down. I’m not sure I fully believe it, but it’s why Intuit is down, and it’s what Canva’s wrestling with.

So it's a super interesting time here. As yet, I'm willing to bet—I could be wrong on this—that there's still a role for a company like Higgsfield, which is an aggregator across models. That goes against what you're saying, Howie. That's a company that's saying, “Hey, Mr. Consumer, you could do this directly on one of the video-generation companies, but I'll aggregate the models. I'll give you a slightly better UI. I'll help you with billing,” and maybe there is a business on top of the models.

I don't know if Jason would agree with me. I'd say they're serving 2 different markets. I think the ChatGPT cannibalization of Canva is me and my partner doing a dinner invite with ChatGPT that we would have done in Canva, and then Higgsfield is actually a business that uses video as a more primary method of delivering their message. And so it's a slightly more prosumer, professional business, I would argue. Jason, I don't know if you agree with me.

I think it is. I think you could argue it, for sure. Almost all that growth—$700 million in revenue to date—is from this video creation, complex video creation, where you're creating functionality out of the models that's very complicated to harness, right? It's a harness that allows you to do something that's very complicated with the models.

Their original model, which was just to aggregate models to make short videos, is cash-flow positive, to Rory's point, but it's not an exciting business, right? They stumbled into the bigger one. The tough question is whether that whole business, which will shortly be a billion in revenue, could have been Canva's if they'd aggressively got into it, just like I firmly believe a big chunk of Replit and Lovable could have been Figma's if they'd done it.

It's easy to take shots when it's so hard to run your core business, but I think we're starting to see the outcomes of how hard it is to run your core business in the age of AI. You've added all the AI stuff. Figma's added great agentic features. Canva's a little slow, but they've added them. Even that's just not enough, and you're seeing gravity weigh you down. It's a tough job today, right?

I think that's actually a super interesting meta point, Jason. You're right, because I tend to be an incrementalist, but there are times when the world opens up and there's a crevice between the before and the after. If you make that jump, you have to make it quickly, and there comes a time when the gap is too big, right? That's what you're saying here, right?

I've come to the conclusion that this could be one of those times, and to some extent, we're going to see a lot of people paying the bill in '26 and '27 for a certain amount of hesitancy in '23 and '24.

3. Incumbents Race To Adapt

Can you give me an example of a Figma- or Canva-generation company that has gone, “Hell, I see this coming. I'm going to move fast, and to hell with it”? They've done it and done it well.

The boring one is obviously Intercom. We've talked about it a lot, and we were investors. I'm sure that after the deal closed, Eoghan would be the first to say, “That was a journey, and a wild journey, and a hard journey,” and he earned every dime. Let's put it that way. So I give him credit for that.

But I'm just trying to think here. It's a hard one, isn't it? The interesting thing is, to some extent—and we'll talk about Atlassian in a second—one of the big questions is how much of your business is going to change? There are some businesses that, just by virtue of the software process they automate, aren't going to change as much.

I think accounting, for example—we have real interest in the next generation of accounting companies—but it's a fairly slow-moving market compared with, say, individual prosumer creative tools. So, to some extent, the speed at which you have to move is in part a function of the kind of business you have. I'm just thinking aloud.

Actually, I will give you some, because they're right on the nose: some of the coding tools. I think Windsurf and, come on, Cursor were doing something else right at the start. But because they were super small, they pivoted in '22 really fast. I don't have a good example of a 2017 or 2018 coding company making that pivot.

Replit's an example. It was frigging in the wilderness for 6 years until it added the models, right? It was a super-nerdy web IDE.

I think, to answer Harry's question, it's a tough question to address as investors and employees. The ones that have accelerated—the pre-agentic ones—are the ones that were in position to catch the wave. The Datadogs, Cloudflares, and Palo Alto Networks of last week. These were guys that were already positioned for it.

I didn't think Twilio would benefit from this. Jeff Lawson saw it when he was on this pod, right? He's like, “Agents are going to need more. They're going to need more voice and more text.” So Twilio, which was the hipster's API for voice and data when we all met Jeff, became your granddad's tool, but it was still well-positioned for the wave.

He was holding the boogie board just right, and the wave came in, and he's flying, and frigging ChatGPT is just tumbling poor Canva side over side. I think in the enterprise it's happening, it's just slow. It's happening quarter by quarter, slowly.

But I want to make a distinction here, because I think the Datadog example is an interesting one. I think you have to distinguish between whether it's easier to survive if you're well-positioned and don't require a business-model change, versus if you're not well-positioned.

The challenge facing the CEO of Datadog versus, say, the CEO of Canva is very different. If you think about Datadog, they sell observability. They sell it to infrastructure vendors. Nothing in their model has changed, except there is now an infrastructure vendor who needs to buy 100 times more Datadog than anyone else has ever bought. So all they had to do was show up and sell more, and that's true for all those guys.

We were lucky enough to be in JFrog. You can see that Cloudflare, Datadog, and all the infrastructure providers—it's not like they're inventing a new thing. They're just saying, “This is the greatest infrastructure boom in history. I sell infrastructure. Time to make out like a bandit.” There are some tweaks at the margin on the products, I agree, but fundamentally, that's it.

There is one—Captain Obvious, and just so I don't get flamed in the comments too much—that did it. I think there are 2 important reasons. Obviously, Palantir did it. Palantir went from 18% growth to 98% growth, right?

That's a good one.

Unprecedented in our lifetimes, right? It may be the one of N, or N equals 1. Maybe it was well-positioned. The thing was, it really leveraged a combination of outcome-based deals and pricing, and true FDEs that no one else had.

We talk about FDEs all the time with our portfolio companies. It is fair to say they're really solution architects or sales engineers with an FDE T-shirt. Palantir had people who, for a decade and a half, were out there deploying massive business change in the field for their customers. So when their customers needed AI, they had the people to do it, right?

Then Alex Karp did the crazy thing that the VCs talk about, but it's hard for companies to do: He did outcome-based deals. “Give me $2 billion. I want a $2 billion contract, but I'm going to save you $8 billion, or I'm going to give you $6 billion more in revenue on the commercial side.” No one does that outcome-based deal. They talk about it, but they just pretend that they do it. No one puts a $2 billion deal on the line for an outcome-based resolution.

They had both of these things. Canva's on the prosumer side, but in the enterprise, it's very hard to change to true outcome-based pricing and to have a suite of FDEs that can deploy AI. I really wonder if the average SAP SE is as literate in the models as they are at Palantir.

First of all, I totally agree, because we talked about infrastructure, and now we've moved on to apps. In the apps, I think Palantir is an excellent and possibly unique example of someone who's done that.

It's very interesting because their existing model wasn't that threatened, but it was growing slowly and was very government-centric. If you read the book, the Palantir, the Karp biography, huge credit to them. In '22 and '23, they saw the LLMs and grokked them immediately. They said, going back to the thing about making your bets in '23 that come good in '26, “We're going to put all our wood behind this. We're going to build the enterprise version of the product.” I can't remember the code name for it now. “We're going to make this bet.”

You're right. It turned out that the combination of AI knowledge and FDEs was exactly what enterprises needed. I agree. That's an example of someone like Replit who had to rethink everything and pull it off. Someone like Canva still has to rethink everything and has a lot of pressure on it. Palantir could have chugged along at 20% and been roughly fine with the government, but instead, they grabbed the moment.

I'd give them the positive Grab the Moment Award. You with me? It's one where I would argue that, unlike Datadog or somewhere where all you had to do was do the same thing, you have to give the CEO and team credit. They said, “If we turn the crank even slightly on our offering, it will work for a whole slew more customers than we've had.”

So I agree. I think that's a good example of grabbing the moment, and I think Replit's a good example of doing the even harder thing, which is, “Shit, I gotta do something else, but if I do it, I'll win.” But there's not many. There's not many.

4. LPs Reprice Private Winners

Can I just do a quickfire? I don't want to take it back too much, but a lot of LPs listen to the show, and they have Canva in their books, and they're going, “What do I do with that?” How should they think about that, given what we've just said?

Well, what do you think it's worth? I'd say it's probably worth $12 billion right now. 20% growth at $4 billion ARR in the current public markets and not decelerating. There's some sort of Rule of 40 number that's better, but I'd say it's worth about $12 billion.

You might be right, but I'm going to push. What's interesting is I can find you companies that are in 20–25% GAAP revenue growth, free-cash-flow positive, trading significantly above that because there's no existential question. One of the big things that's happening is that Datadog, Cloudflare, JFrog—all those guys—are in the mid-20s growth, with 20%-plus operating margins, trading at 15 to 17 times NTM. But I think the difference is there's no existential question here.

Which is why maybe I'd answer the question in the following way: if it's 20% and the existential risk is there, then Jason's right. It could be $12 billion or even less. If they can transcend that risk, you're still a top stock buy. You're going to be grounded by reality now because you're not selling a Brave New World, but it's kind of $12 billion and up.

Listen, I hope you're right. Again, I don't want to be negative. I want Canva to defy gravity. But when every single person on Wall Street uses ChatGPT, why do you think people won't perceive it as having existential risk? I think existential risk is both reality and perception. If monday.com and HubSpot do, I don't see why they won't see the exact same thing for Canva.

I'll tell you exactly why. Because you're right: in the short term, they will. But you can't control that thought. You can't control what the 27-year-old on Wall Street thinks. Let's take the example of Atlassian. We had Mike Cannon-Brookes on as well, right? They killed it last quarter. The only way you prove that you're not dying is by growing, to Jason's point.

The thing that's pleasing about life is if you pull off the important thing, the hard thing itself, then the markets will follow. You're right. Right now, it's a very tough time because when the existential risk is posed, the only way out is to prove it. So right now, I think there would be a wide variety of perceptions on valuation for something like a prosumer company like Canva, and it would be hard to peg value and hard to get liquidity at scale, by the way.

One of the other differences, at the margin, between a private company and a public company is that when the window shuts in private and the appetite dies, it's very hard to get anything done. So the real answer to you, LP, is it doesn't matter what you think, big guy. You're in this journey for the next 12 months. Buckle up. Liquidity will only come at the end of the journey.

It's also potentially important—I don't want to over-kibitz; I'm not exactly a public-company PR expert—but I do think it's important to get ahead of the narrative Rory's describing. I do think that once everyone starts saying that ChatGPT is killing Canva because you can make poster-sized images for free with your subscription, it's just the dumbest meme.

I think the 3 of us can probably agree that one of the dumbest AI memes was that everyone would vibe-code their own CRM. Even though Harry's had guests that do it, this makes no sense for 99.9% of the world, okay? You can't maintain it. You can't build the integrations. It's more complicated. Most of the folks that say that have never used a CRM.

The idea that everyone would vibe-code their own CRM is so visceral. And the damn 20VC show has been part of it, bringing in all these guests who are trying to hide slowing growth by talking about how they built their own CRM. But it has taken hold, right? The shorts have jumped on it, and the haters have jumped on it, and so be it, right? But if I were Canva, I'd be worried that this would become a meme.

I think there was a quote, something like, “The worst thing in the world to fight is a bad idea whose time has come.” And you're right. We're seeing that in the SaaSpocalypse now, and the only way out is through, which is that the companies that have produced revenue growth have seen upticks. Overall, World Cloud's up 50% since the bottom of the SaaSpocalypse. But the people who've struggled are still struggling. So you're right: you will have to prove it.

Rory, what do you think the answer to LPs is? Obviously, I think after Airtable and this Canva quarter, it's probably time to be a little extra skeptical of marks, just being realistic. These are 2 events that I think have quietly hit old marks. Even you should have marked them down last year. These are events that are difficult to hide. They're difficult to say, “My guy's gonna turn it around,” okay? After these events, I think they do kind of shake the ground a little bit.

They definitely do. And it's funny because, just taking the Airtable example, we all—I used to mentally say to myself, and I'm sure you did—had Airtable and Notion in the same bucket. And then, I mean, I don't know if the Sacra numbers are correct, but Notion is apparently $800 million, growing at 70–80%.

That gets back to the same comment: when I think about valuation, stepping back, at a minimum, you have to look very objectively at the actual growth rates and be brutally honest as you think about valuation relative to that growth and that projected forward growth. You have to ground yourself in those facts as step 1, and then the second-order question is, do you grade up or down for existential versus lift?

But at a minimum, yeah, it's no longer acceptable to say, “Once upon a time, we raised at $42 billion, therefore we're holding for $42 billion,” right? It's, “We're doing $1 billion, we're growing at 30%. That is this multiple. We're doing $1 billion, we're growing at 10%; that justifies that multiple.” I totally agree, Jason.

For me, I thought one of the tweets of the week was from Dave Samuel—I think that's his name—from Freestyle, who mentioned that the blended exit price from Airtable was actually $6 billion, and the importance of selling along the way and being very thoughtful about selling in the good times.

Totally. And it's always true when it goes down, and it's never true when it goes up.

One of my great friends is a multibillionaire, and he told me, “You know what, Harry? I never regret making millions of dollars,” and I say this through my G650. And actually, you know what? I've sold stuff now, and, yeah, I've lost on upside, but I'm happy that I locked in some wins.

Yeah, but here's the thing. We can move on. I get all that, the Freestyle argument. It sounds great on Twitter, and it's mathematically true, right? But if you want to have an outlier fund—I don't know, man—that math really only works if you've got 6 or 8 of them in the fund. Listen, maybe at a bigger scale, the math's different.

But for a smaller fund, I'm lucky to have 3 fund returners, okay? That's hard. And if I start taking early exits on those and I don't have a 10x fund returner, okay? My LPs want these frigging 5x, 6x, 8x, 10x funds. The math gets kind of tricky if you sell too much early. I don't care what X says. You've got to keep doubling down.

I literally just did this analysis across my whole lifetime of all the things I've been involved with—personal, angel, and venture. Who should have sold, and who should have sold earlier and taken the secondary? For me, it broke roughly 50/50.

Statistics would say it probably breaks 70%: you should have sold; 30% or less you should have held. I mean, I can tell you what the facts are. But the next sentence is the key: the holders compound forever, and the ones you sell don't compound at all from then on.

So it's the Hendrik Bessembinder research from the guy at Arizona State, showing that less than 1% of all the companies ever give 90% of the capital gains in the public markets. It's the same in the private. It's just the nature of power laws. Most of the time you won't regret trimming, but on the few that you regret trimming, it turns out to be most of the value.

I'll never forget having Jake Saper from Emergence, who I like a lot, and I like Emergence a lot. Brilliant firm. But they sold Salesforce reasonably early in the arc of the Salesforce value-accumulation journey. And if everything else didn't matter and there was just a hold on that decision, it would dwarf all the other outcomes.

We can continue in this vein and discuss Atlassian and HubSpot, or should we talk about Jeff Dean's Google talent changes?

I mean, I'm not an expert on the Jeff Dean talent changes, but clearly, this is the time to go off and build, man.

Let me leave the comfy coop where I’m making 9 figures a year to just talk about AI in a comfortable conference room with a mug and go out and do it, man.

5. Google Loses AI Legends

So let me provide some context. Google had some talent loss. Jeff Dean, one of the godfathers of much of AI, has left after 27 years, taking 3 legends with him. I’m going to pronounce their names wrong, so I’m going to leave it there.

And then Demis Hassabis is also stepping back, or whatever elegant marketing message we want to put around it. He’s moving into a chairman role. Power is centralizing back, really, to Silicon Valley as well with that. That was the big news from Google, and obviously shares tanked as a result.

I actually think Jason made the best point. First of all, it must be extraordinarily validating if you’re Jeff Dean to leave as a non-CEO of a $2 or $3 trillion market-cap public company and have the stock go down by a couple of hundred billion dollars. If you want to increase your sense of self-importance and self-worth, that was a good moment. That was what therapists call validation at a high level. So let’s move on from that.

I actually think Jason’s genuine comment here is the correct one. We can analyze what it means for Google, but think about it. If you’re Jeff Dean, you’ve done 27 years at Google. You’ve made gazillions of dollars. The mission at Google, like it or not, is to allocate a lot of the compute to the Google Cloud business to just be a hyperscaler. That’s boring as shit to you.

Allocate more of the remaining compute to build a competitive frontier model, mainly focused on the big things in consumer, because that’s what they care about, and coding, because that’s what they care about. At this point, after 27 years, that’s fairly boring to you. And get some time to do a little bit of medical discovery and scientific discovery that’s really exciting to you. But because of the relative size of those businesses, that’s always going to be in third place. That’s option A.

Option B: you can go raise all the money you want. I mean, it was really sweet that they even built a PowerPoint. I doubt they needed to, right? I think, “Hi, Jeff Dean, I’m raising money,” would have sufficed. And go away and do exactly what he says, which is use AI to, quote, “investigate advanced scientific questions.” What are you going to do with your late 50s in life? It’s totally natural.

Yeah, I didn’t buy that. I’m not being rude, sorry. If you look at the resources he had available to him at Google, with the data that he had available to him at Google, he could go into Sergey and Larry’s office and say, “Hey, I want X.” And they would say, “Jeff, you have whatever you want to do.”

Well, I don’t think that’s what’s happening. With all due respect, I don’t think that’s what’s happening. I think what’s happening is every piece of compute that you give to Google Cloud turns into 30% operating margins in a day, because they can sell it to Anthropic. Every bit of compute that you give to building Gemini might turn into a decent coding model if they get their shit together, and maybe you can get some Anthropic-like revenues or some ChatGPT-like consumer revenues.

Every bit of compute that you give to drug discovery, materials discovery, or physics discovery turns into a long-shot, 5- or 7-year moonshot that maybe will indulge at the 10% level, but it’s not going to be the core thing they do. So if you’re a senior executive in those companies, you’re probably expected to do your day job. What was the old thing? 20% kind of fun time. Eighty percent of the time, you’re meant to deal with boring shit.

So I think at some level there’s a desire to focus full-time on that. It’s hard for you to understand when you’re younger. As you get older, you start saying to yourself, “Is this it? I optimized ads. This is all I want in life? Maybe I want to be my own boss. Maybe I want to just focus on scientific discovery.”

If Rory’s right, and it makes sense to me, I just don’t know. If, in a sense, they have access to all the resources, but the team’s somewhat deprioritized because of where the cash flow is, I can just tell you—not to go back in time—but when I was an SVP at Adobe at the number-three business unit, it sucked.

I’d be with the other 50 VPs, and we wouldn’t even get to talk about what we were working on. We were only doing $800M at the time, but the number-three BU is invisible. It was just an afterthought.

So if that’s the vibe today, and if maybe for the first time ever I could take my team, my whole team, and get all the capital I really need to do what I want to do, I would leave if I was number three. Listen, I’m not quite smart enough to know, but if Rory’s right, that may completely explain it.

If you’re the number-one priority at Google, you’re going to stay, right? Because it’s easy, and it’s a pretty bucolic environment there—or at least it used to be. I used to sell to Google all the time. I was there every 3 or 4 weeks—the most bucolic campus ever. But maybe today it’s a little more stressful.

If you’ve been deprioritized, as important as you and your team are, and you can raise $1B or so, I’d check out, man. I’d go do it, right? And the VCs aren’t going to put the traditional annoying pressure on you in the first 24 months.

That’s another interesting point worth making: the appetite from venture to finance moonshot-type, AI-will-solve-science bets has never been higher. The proof is not yet in, as a reminder, and a lot has to happen to make these bets work. But if you’re a scientist at your core, and if you’re a believer in knowledge and discovery, this is a once-in-a-lifetime chance to make that unhedged bet with no corporate BS to deal with.

Even 2 or 3 years ago, you would not have got that kind of money to make that bet. These kinds of science-based neo-labs are really a phenomenon of the last 2 years.

And maybe just 2 small things. First, I think Vinod’s leading the round, or co-leading the round, or something. So he’s just redoing their playbook again. Granted, OpenAI hasn’t gone public, but he’s already had a little bit of a win here, right? That guy Vinod.

And then, going back to Harry’s question on LPs, I think this week it’s a little murky. I’m trying to read the news from Hawaii. I’m not quite in the SF Bay. But if Anthropic really is going to IPO—this time, for real—in the next 60 days, it’s just going to tangibilize all of this once again, right?

Hopefully for the better, maybe slightly for the worse, but it will make these bets seem more and more like the present, and Canva and friends more and more like a distant memory—a distant memory of a bygone era of software.

Two questions for you. How significant is it losing this many high-caliber people this quickly? We haven’t mentioned Demis Hassabis. Obviously, Demis, founder of DeepMind, led London AI efforts—visionary, genius. I’ve been fortunate to interview him. How significant is it, honestly, to Google?

On one hand, obviously, early on this business had been very individual-centric, so losing these 2 talented people from their full-time roles, and the 3 people who left with Dean, is obviously really significant.

On the other hand, just to put it out there, whatever Google was doing wasn’t quite working. We’ve gone through the “Google is dead” phase 12 months ago. Then we went through the “Google is amazing” phase 6 months ago, and now we’re kind of going to the middle.

Google’s doing a good job in cloud, selling compute to Anthropic. They’re doing a good job selling the picks and shovels of TPUs to Anthropic. They’ve kind of got a model out there, but they haven’t made any impact whatsoever in coding, which is the mother lode that’s feeding the Anthropic beast right now.

So you could look at it and say, “Google’s efforts so far are B-plus, A-minus. They’re not A-plus.” I don’t think anyone wanted anyone to leave, to be really clear, but it’s probably been unsatisfactory.

You can imagine: you’re the CEO, and you’re saying to your 2 most talented human beings, one of whom has a Nobel Prize for Medicine, just to remind you, “Why aren’t we building a better coding model?” And they’re sitting there thinking, “Why haven’t we cured Alzheimer’s?” At the end of the day, that’s a really boring E-staff meeting because we’re just talking past each other.

In a perfect world—I mean, this is oddly what they got rid of 2 years ago—if everyone had been self-actualized, they’d have put someone like the guy who’s running it, a tactical executive, in charge of grinding this out, and maybe given these guys more running room earlier to do fun things.

Because I think that’s been the dynamic all along. If you read Cade Metz’s book about the DeepMind acquisition, all along it’s been, “How much corporate shit do I have to do? Because what I really want to do is get a Nobel Prize.” And who am I to argue? We’ll remember the Nobel Prize long after we forget the Google Q2 earnings, right?

So he’s entirely right to want to do it. But unfortunately, Google has to make Q2 earnings, and if you’re the CEO there, you need an executive who’s willing to drive what it takes to get a comparable chat model out there to compete with ChatGPT and a comparable coding model out there to compete with Anthropic, neither of which you’ve done now.

So if someone comes in and says, “I've launched this initiative, and I think in 5 years' time we'll bring out a simply amazing drug, and it will cure cancer, cure Alzheimer's,” whatever, your brutal comment is, “If we do that at the expense of a coding model and a chat competitor, that's a mistake, and our stock will go down by 50%.” Because the largest drug company in the world is valued at $1 trillion, and Google's valued at $3 trillion. So the corporate imperative is to get someone who wants to do those things, not someone who wants to save the world.

It's super difficult today. The best AI engineers, the best AI researchers, really want to work on what they want to work on, and they don't want to work on stupid things. They don't want to work on obvious things. They really only want to work on stuff at the absolute cutting edge that is extremely intellectually interesting to them. They don't want to work on anything else. They just don't want to do it, and they don't have to anymore.

Which is why you've got to admire the brilliance of the team at Anthropic, that they have simultaneously managed to—not convince themselves, because that sounds judgy—they feel they're on this mission, as a public benefit corporation, a mission to bring AGI to the world, all that good stuff, while simultaneously making every single correct, rational financial move over the last 2 years, including, to your point, going public first, which I think they will as soon as possible, because they'd be insane not to.

The 1 thing we can stipulate is those guys are not insane. They are right on it. They will go public because this is peak brass-ring moment, and you could argue the trends in '27 are tougher. You've pulled ahead of ChatGPT, OpenAI, comfortably, to a point where it's embarrassing. It's never going to be better.

There's just been a trillion-dollar IPO that, all in all, went okay. It's back to its offering price. You should go. You should go now. You should go fast. You should be done. Just put a nail in the Anthropic pear, right?

6. AI Talent Resets Compensation

If I'm the founder of an early-stage company, do I just accept that I'm going to have B-tier or C-tier AI talent? And I don't mean that denigratively, rudely, or horribly, but they're at Anthropic and OpenAI. I mean, Google can't freaking keep—

I think it's the wrong framing. If you think about it, when you were building a software company in the age of the PC, you had fourth-tier chip talent because you weren't building a chip, right? The point is, if you're an AI company and you feel the need to build a frontier model, then yes, you've put yourself in direct competition with someone, and if you don't have the good people, you're toast.

What you've got to do is make the model a complement and have A-tier talent at UI, A-tier talent at AI implementation, and A-tier talent at the things where you have your competitive advantage. Even the very best companies that are taking open-weight models and fine-tuning them should be experts at fine-tuning and experts in their data domain, but they're probably not going to be as good at building an LLM from scratch as the guys who've been doing it for the last 2 years. But that's okay.

One tough thing, though, that is happening for sure is that, when we started this show, there started to be 2 tracks on compensation. I have to break my salary bands for my AI guys because they're worth so much to Anthropic and OpenAI. Now we're seeing 3 bands of compensation. We're seeing the regular human beings, then we're seeing the AI guys, and then we're seeing the 1 to 5 superstars that we're talking about, right?

I have to find a 7-figure package for them as an early-stage startup because they're going to get it. We have to provide them everything: the outsize equity and the outsize cash. Especially when you talk about folks that are mature—the $100 million-and-up ARR guys, $200 million and up—they all have this god tier of compensation.

If you have the revenue, it's sort of fine, right? You can afford it. If you're doing $200 million in revenue, you can have 4 god-tier employees. It's not going to break your model, but it is something that folks have had to accept. There are a bunch of CEOs I work with informally, whom I'm not an investor in, that I work with at that scale, and they've all created god tiers.

They're like, “I got 4 guys. They are the core of my next-generation product, okay? They're all making 7 figures. They all have equity stakes 10 times what an employee at this late stage would have.” It's the best investment I've made—this god tier. But it's tough on the rest of the team, right? It's not the way we used to do this kumbaya-style when you've got this god tier.

It's tough, but you're not going to pull off a Palantir and Intercom without a god tier. You need a Skunk Works and you need a god tier, or it ain't going to work. The siren call of the Anthropic comp is too high. OpenAI just did a secondary of $7 billion, didn't it, this week? Something like that. That sounds pretty good to me, guys.

Markets and prices are all about incentives and signaling, right? It's a way of sending real information. There's no doubt that I saw the analysis that someone who got $1 million in stock in Anthropic in '23, it's worth $51 million now. That's a signal that just ripples through the hiring environment across the entire ecosystem.

Now, I would remind everyone that that's what's known as a once-in-a-lifetime change. I don't think the person today will be getting 50 times their return. Whatever, it does have an impact of distorting what everyone thinks is possible, and we are naturally attracted to the narrative around the outliers. That's not the norm. But yes, that is the California Gold Rush part of the story, and you're going to see it even more written about when the pricing happens.

The only other thing I would add is: if you're not one of those companies, how do you compete with them? How do you compete with that? I do think you might have to have a god-mode compensation package.

The only thing is, a lot of those jobs that you're offered do kind of suck. They're not all Jeff Dean and his buddies sitting around on a whiteboard designing the future of Fable, the 7.2. A lot of these jobs, for folks who aren't quite at that tier, are not that great.

That's always been the job of a founder: to find those pirates and romantics at the edge that could get it. Back when I was a founder, our test was always, “Did you get an offer at Google?” If someone got an offer at Google back in the day, you knew that they were top 10%—top 5%.

You could do the same test today: “Did you get an offer at Anthropic or OpenAI? And what was the offer?” “Well, that flashing thing in Claude they want me to work on—the red or orange thing. I'm color-blind. I'm going to make $1 million a year getting that pulse just right.” Or, “I have to work on watermarking for my first 18 months.”

You can find the folks that say, “Yeah, accounting software would be more fun than that. I'd like to do LLMs for accounting.” You've got to interview everyone on planet Earth, and you will find someone that doesn't want that job, right? That's the job. But you might have to pay them a lot more than you had to 24 months ago.

7. AI Infrastructure Faces Pushback

Talent is 1 bottleneck. Another bottleneck—and I did think this was a really interesting news story that came out this week—was the backlash going federal when it comes to new data center creation. Rep. Ro Khanna said he will introduce a data center bill of rights that will give local communities the right to say no to AI data centers. Yes, you go, Rep. Ro Khanna. That's exactly what Xi Jinping is going to say.

The hero of Silicon Valley.

That's what Xi Jinping is going to say: “Stop your DeepSeek moonshot. We don't want that in this rural community.” What a freaking joke, hey?

I have talked with some folks who truly are experts at this. Most folks think that this is pretty lame, this NIMBYism, right? It's even crazy that Texas is in on the NIMBYism. Elon pointed out that the Terafab has already created 3,000 jobs, but it's only at 10% capacity. It could be 30,000, to kind of get folks to see the other side of it.

But I think folks who are close to it think that there will be enough counties and jurisdictions that want these data centers, and that as we push through this and as government gets on the other side, it will ultimately work itself out. This will not be one of the great issues of our time, even though it seems ridiculous today that you don't want these.

There are only so many people working at these data centers. But this one may work itself out. At least that's what the folks I've talked to who know it more closely than I do believe.

Do you not think this will be a material blocker in our speed of deployment?

Well, we have 50 states, and I don't know how many counties we have. It's a lot of counties. There will be some with water and power that want this business, is the meta point, and this backlash isn't going to last. These are not all destroying our water tables. These are real jobs—3,000 jobs for a lot of poor communities.

I mean, this is such an entitled podcast. Oh, a poor Anthropic engineer only made $35 million. I mean, go out to the goddamn Panhandle—no one's making $50,000, right? These are not enough jobs, right? But these are real jobs with real money, and they're going to last years and years and years.

They bring a limited amount of real economic benefit. But it is a bummer for places where we should be building these data centers, for sure. It's not a net positive; it's just a question of whether it will work itself out in the US, where we do have a nice set of competition here—regulatory competition between states and counties.

Yeah. The irony of Ro Khanna, the Silicon Valley congressman turning into the Marxist wolf in sheep's clothing, is pleasing to me, as I would not tend to be on that side of the voting aisle. Just watching all the Dems get suckered into thinking he's a moderate has been worth the price of admission as he starts advocating the billionaire's tax, so there you go. But, yeah, who am I to judge?

You mean a millionaire's tax, Rory. I think you mistook it.

Yeah, yeah, that's exactly right.

Right. Yeah.

So, that's funny in and of itself. But once you're elected congressman, your next step is up, and the truth is, the way to electoral success probably lies through a fair amount of populism for the next period of time.

On the data centers, I saw a good piece, I think in The Atlantic or something, really just talking to people in the areas wrestling with these issues, and it was a very interesting point. It was much less even, “AI is awful,” and much more, “I don't know what I'm getting here. It's all very opaque. What's the deal?”

And I think if tech wants to get this stuff done, there are 2 risks here, and they're almost opposite each other, because if you want to get local support, you have to figure out what's the package that moves it for them. It definitely isn't, “Oh, by the way, you're going to get a 25% increase in your electricity costs.” You're seeing that now.

I think the smart people are saying, at a bare minimum, if you want to get a data center in here, you have to find a way of making sure people aren't going to pay for the electricity, and there's probably going to be some kind of dividend. If you tell people that you're going to get this job business, there's not going to be an increase in electricity, and there's going to be a $5,000–$10,000 distribution per person in the township, you'll probably go, “Yeah, we should look at this.”

The other thing is, do some of these statewide laws just make that impossible to do? Because the truth is, at the moment, there is a fair amount of “we hate tech bros” out there. And as we said a million times, it turns out if you spend 3 years saying AI's going to kill you all, you shouldn't be surprised we hate it.

I think there is some wood to chop technically. But I do agree, Jason, it's a great point. One of the best things we have over here, unlike, I say, the UK, which is one of the most centralized states in Europe in terms of central authority, and Andy Burnham's trying to change that, is that we've got 50 states. If North Dakota hates this shit but South Dakota likes it, then South Dakota can put something in place and it can happen.

Dude, the UK is the size of South Dakota.

Well, yes. And my point is that we've got diversity here. So hopefully it won't be a huge block. I mean, right now the practical point is it's actually the availability of power rather than just pure data center blocks. But there's definitely a whole series of things slowing things down between power availability, compute availability, and then political willingness to turn this stuff on.

Now, the main man himself, always lacking in ambition, Mr. Elon Musk, unveiled Terafab, which we touched on there: $16.8 billion. I think it's going to be the most expensive build-out of a real estate project, I think I read. As we said in terms of jobs, extraordinary in terms of how many jobs will be delivered—different numbers, but between 2,000 and 3,000. Really, it's him saying that he wants to sidestep TSMC's queue and obviously build out his own fab capabilities. How did we think about this news, both in terms of the strategic decision and the scale of it?

Look, this is someone with boundless ambition, plus success at delivering on this boundless ambition, plus access to capital at an unprecedented rate. So he's probably going to try and do all these things. It feels wildly ambitious. You have to build the gas turbines, build the fab, build the robots, and build everything. It's vertically integrated on every level, but he has a piece of vertical integration that's been superb, which is satellite launching integrated into Starlink. It's been superbly vertically integrated, right? You can join the dots in the past and say it all makes sense.

I continue to think the scale of ambition—if there's any slowdown in the AI spend, then the all-in bet is the one that slows down the most, the fastest—and this is the all-in bet. So watch this space. But right now he's got the capital and he wants to do it.

I think that's right. I just think at the end of the day, beyond all that, he's just saying, “Listen, there's a decade of supply chain limitations that's going to damage my ambitions. I've got to do it, right? I have to do it.”

It's not just that the investment is unprecedented; it's that you can't get RAM, you can't get chips, you can't get what you want. We've had limited issues in the past, but I don't think we've ever looked forward and said, “For a decade, I'm not going to be able to get what I need on a cost-effective basis. I can't even get TSMC on the phone because Jensen Huang's out there all the time.”

I could be wrong, but I have a limited amount of experience here from the old days. I don't think it's ever been like this, where you could— It almost feels like infinite time before I can get the capacity I need at any tolerable price.

Yes, which is why it's super interesting to tie in something. Intel is part of the Terafab consortium, and I just saw it today. I did not know this. Intel completed an equity round, which I read somewhere—and I haven't verified it—and it was the first time they raised equity since, like, they went public in '79.

In other words, they've been profitable from cash flow and returning capital like a real company is meant to through the '80s, the '90s, the 2000s, and the 2010s. And now, with the AI CapEx boom, plus obviously their deteriorating performance, they said, “It's time to access the capital markets again.”

8. Founders Demand More Control

While we're on Elon Musk, Elon Musk did have a very unusual incentive package, which obviously expanded with the expansion of the company's valuation. Revolut announced an incentive package for the CEO—or it was leaked—whereby it basically ratcheted up with different prices of the company. He'd get another, I think, 5% to 7% at $200 billion, and then ultimately, at $500 billion, he would have circa 39% to 40%. Is this the new norm, and should every CEO be asking for ratcheted incentive packages alongside valuation bumps?

It's not going to be the new norm, and if it is, logically stock prices should go down by 10% or 15%. Because what you're basically saying—I mean, I've read what's available about the package. Multi-year packages with incentives around market cap—in other words, significantly beyond the normal CEO comp—have been a thing.

It obviously worked for Elon in the 2018 Tesla package, and obviously, after a lot of toing and froing, he got another package just recently finally approved after they moved to Texas. So they're not, quote, “the norm,” but they are put in place for reasons we'll come back to, for a small number, typically founders who've fully vested in all their shares and who want to be incentivized again by boards who feel that they have to be incentivized again.

By definition, that's not, quote, “the norm.” 90% of public companies aren't run by the founder, and frankly, the number of people willing to run a public company for $10 million or $20 million a year turns out to be remarkably high, right? So no, they're not the norm, Harry. Are they the norm for founders? We're seeing some of them.

Most of the time, I think, especially if they're badly designed and focused on stock price only, they often fail. We saw a whole bunch of them in 2021 that got unwound in 2023 and 2024 because they weren't based on operating performance. They were based just on, “Hey, if the stock is at $200, we'll give you more shares.”

And then what happens is the CEO executes brilliantly, but the market is down, so he doesn't get his shares, and he comes back and he says, “Look, I've done my job,” right? Forgetting that he would not have made the same argument on the other side—he or she, right? So the record is fairly mixed.

But at the same time, I'm going to acknowledge something. For that special thoroughbred CEO who thrives on challenge, you can put them in place if the incentives are right, and maybe you do get extraordinary performance in return for extraordinary comp. So it's not utterly crazy.

They're a very high-cost, focused instrument, and I think boards have to be fairly careful when doing them. I'm not, for example, a fan of the purely stock-price-based ones. And to be fair to Elon's '25 package—we went over this before—it was a great package because it had requirements: you have to do Mars, you have to do Optimus, you have to do lots of cars.

At that point, give him the damn money, people. Now, an interesting comment I saw in the Wall Street Journal today: There is a clause that says if there's an M&A above a certain value, you do get a lot. You might see some acceleration of that package.

I haven't read the detail. But it would be interesting if SpaceX and Tesla merge: does he hit the big ka-ching on Tesla as well?

That's what it implied, that he might hit it just by merging the companies, right?

Totally.

Before the details, Rory, just one question to you, or to you and Harry. I don't mean to interrupt, but when I read this, I thought it was more about control than just money. I also just read a story that the CEO of Revolut tried to get out of paying a $20 million broker fee on a $400 million yacht he bought, so clearly he enjoys the good life, right? As well as working hard, right?

This is not a CEO who does not care about money. But to me, and I think Elon was very clear on this, “I need to control these companies or I'm walking,” is what he said when the first one failed, right? So if Nik owns 40% of Revolut, he controls it, especially with, I'm sure, a supermajority board and all that. It's his company. That's what he wants.

The money, I'm sure, is part of it, but this is not going from 2% to 6%. This is going to 40% ownership. That's a lot.

If he made that argument, to me it's about control. If I were the chair of the compensation committee, I'd say, “You're exactly right, Nik. So I'll tell you what: we'll give you 3 votes on each of your existing shares. Now you don't need control, and you don't need any more shares.” And he would come back an hour later and say, “I also want the money,” right?

No, you're right. But I will say, I think we've all learned that there are limits to supermajority shares. We all have learned that there are other sources of pressure, whether they're VCs, shareholder activists, or other issues. There are levels here of control.

You can control 99 out of 100 board seats, but if you own 6%, you may get pushed out of your company. It is entirely possible unless you go to the mat on it.

I could talk for hours on that, but I do think Zuckerberg would be an example of someone whose control is— You can have ironclad control as a public company if you want to.

No, but you can be pushed so hard that you walk. With 40%, goodbye, guys. I would just end the Zoom with Wall Street if they didn't like what I was doing at Revolut. I would just push the button and say, “Goodbye, guys. Go short me. I'll see you later.”

You say that, but actually it turns out your problem is that, yes, you can ignore people, but they can also choose not to buy your stock. A super-interesting thing happened today on the control issue. This is going to sound unrelated, but humor me.

If you read Zuckerberg's philosophy on AI, remember, this is a person who controls his board absolutely. You can show up, you can tell me what you think, but in the end, I control, right? It was super interesting, and he's pretty much had that sole control.

He said, as part of how they think about governance, that he didn't want personal control over the decision to release new models. It should be a board-level decision. I will admit, I'm like, “Hmm.” And that, to me, was an example of, yes, Jason, you can have control over everything, but then you own everything.

And at some point, even if, say, you own 20% of your company but you have 10× voting control, you can't make them buy the other 80%, so you can't keep your stock up, and you own every problem. This might be a very smart man saying, “I'm not sure I want to be the sole person releasing these models.”

It was super interesting. It was the first piece of control that Zuckerberg has given up in 20 years. So I did note that in passing, right? Control is interesting, even when you have it. I've actually changed my opinion.

Because going public is so shitty, because of all the problems with going public, I've actually come to the conclusion that giving founders more control over their life's work—which is what it is—is an acceptable price to pay to incent them to go public, right? So I've actually changed my opinion on that.

I actually think, even though some of these control things are weird, and I do think they probably shouldn't be in passive indexes as much, there should be some discount for that. I've come to the conclusion that weird control terms are an acceptable part, because otherwise everyone just does what the Collisons do and stays private. They're like, “I don't need your shit.”

But let's go back to the money, because in the words of, I think, Senator Dale Bumpers in the Clinton impeachment trial, when they say it's not about the money, it's about the money.

I think it's about the incentive. I don't think it's about the money. I think your point was so good, Rory. I don't think I've heard it expressed enough that way.

Going public sucks so bad today. Look at the public company CEOs we've had on this show, or Harry has. It sucks to be public today, okay? It was fun during lockdown when you could grow 90% without a new feature. It's not fun.

And I can't imagine, having been a founder twice, a more helpless feeling than being a public company CEO. I'd want to quit. I would just hate it, right? Having control and equity has to somehow tie to it, or it's a partial fiction. That would make it worth it.

I might leave the keys on the table if I had no control of the company I founded. I'd already made plenty of money. I was diluted to nothing. I had a board that didn't understand my product telling me how to run my company. I might just leave the keys for you guys.

Agreed. And I say that not because I like it, to be clear. I say that because I'm just looking at people staying private. I mean, I think the real solution will be when the private capital markets evaporate or deteriorate, and then they will go public because they have to, but that's by the by.

So I agree with you, Jason, on that. But on the other hand, let me take the other side of it now. On these kinds of deals, the thing you look at is the participation rate, which is how much of the total value creation is going to the CEO, right?

In other words, the way this deal was announced—and to be clear, the Revolut deal has not been put in place yet—it was something like: for his existing stake, if he gets it to $200 billion, he gets to 30%. And then if you get from $200 billion to $500 billion in value, which is $300 billion in delta cap, he gets an extra 10% of the company, which would be $50 billion.

Which would mean that for $300 billion in value creation, he's getting 16% of that, which would be abnormally high, to be clear. Right? I think—

16%?

16%. Abnormally high, right?

That's less than our carry checks. Those are 20%. This sounds low if we're doing—

But I can tell you, yes, but if you're getting it on private—

You think I deserve 20% of what my portfolio does after a certain point? Nik's only getting—poor Nik's only getting 16%.

I disagree. I think the market—I mean—

You think you're working harder for your portfolio than Nik is working for Revolut?

No. I don't think it's about working harder. I think the nurses in the fucking emergency room are working harder than both of us, Jason, right?

I could not agree with you more. I could not agree with you more.

So, right, let's go for it. The question is, 2.5×: taking something from $200 billion to $500 billion gives you $50 billion, right? Do you think you could get a Jamie Dimon-level CEO for $10 billion? I mean, he's only made $1 billion taking—

It's too— I think it's too hard. Let me tie it back to a different point, and you can challenge me on this all you want. Harry asked what LPs should do looking at this, right? I'll tell you what I'm doing.

Any investment I've made that is not run by a founder is a zero. It's going to be a zero in this age. It's going to be a zero. I look across—now, we have different portfolios, but the ones I have that are not run by founders, whether they're at $20 million or $200 million, they're all going to be zeros.

And so, if the price of me not having a zero is getting Nik to 40%, I wish I were a shareholder. If I'm Balderton, whoever—this is my best name—and that's the price, I'm going to pay it in a heartbeat, because I do not believe Jamie Dimon's lieutenant, with his starched shirt, blue-and-white collars, and cufflinks, can run Revolut.

It's not that mature. The space is not that stable. I don't buy it. And I'm not a banking expert, but I don't buy that it's possible. I believe he will run that company into the ground, just like every non-founder has run my portfolio companies into the ground.

I'm going to get killed if I don't say one thing here, which is our former guest, Nikesh.

Yeah, but when did he join? Two years ago?

Yes.

No, I just know my portfolio will be zeros without the founders. I'm not saying there aren't examples out there you can find. I just know for me, to the LPs, they're zeros. They're all going to be zeros, no matter how much ARR they have. They're going to be zeros.

And this is the question. That's totally true at $1 million, $10 million, $100 million, maybe $1 billion. The question is, I think Revolut's doing $5 billion—ah, I used to know it—$5 billion in revenue and $1 billion or $2 billion in profit, right? It's an extraordinarily big and very impressive company.

At some point, it becomes not true. Or maybe the better statement is this, and this is the interesting one, because—and I know this sounds really negative—but remember, going back to interests, it's a corporate governance question. Having just come out in favor of founder control, all the things I said earlier, and I stand by them, you still need a dynamic to protect the other shareholders.

Because if you take the logic to the extreme, I saw Nik at Revolut make a comment. It's a very interesting comment that I think is bullshit. He said—and it doesn't sound like bullshit, but it is when you think about it—quote, “I deserve more because the investors, after they give capital, they do nothing else.” The second half of that statement is true. After investors give you capital, they do nothing else. That's the world of capital. But just because that's true doesn't mean the founder can... I mean, what's the limit then? Or, to put it another way, fast-forward 30 years: what's the limit?

I think the world has changed. I don't think most founders care anymore, and so I think you have to adjust. I think Nik is saying what half the class at most accelerators agrees with: “I'm just going to raise it 50, and if it doesn't work, I'll just do whatever.” It's going to happen.

No, but you're not addressing the issue, Jason. What you're saying there is the cost to run a company from $200 billion to $500 billion is 10% dilution. Is the cost from $500 billion to $1 trillion another 10%? Is the cost from $1 trillion to $2 trillion another...? I'm just trying to get a sense of it. Because if it is—if it is—then 2 things are true. One is, you should pay less for that stock because you're going to get way more dilution.

Well, PitchBook had an article this week saying how massively returns are being compressed on outcomes north of $500 million to $1 billion. Outcomes are being massively compressed by unprecedented dilution and high entry prices. So this is just the world we have to live in.

As a seed investor, I've only been doing this for so long, but I've been doing it for a while. When I started, my model was, I'm actually paying twice my entry price. That's how I model. Now it's 4X. I'm going to suffer 75% dilution, and that really means my entry price is 4X what it looks like on paper. That $50 million post-money valuation you want is really effectively $200 million if we hit it, right?

I could complain about it just like the Nik thing, but Nik is going to do it. The baby Elons are going to get these packages, and it doesn't really matter what I think, or any of us think, because enough investors are going to go along with it that they're going to get these packages.

But the more important point you made is, how elite will the companies be that this is reserved for? We could debate whether Nik deserves this, but this is a generational company, right? The question is, do sub-generational companies get these packages, and how does it impact things? But, yeah, our dilution—I think all of our dilution is under-modeled. PitchBook said this week that all of our dilution is under-modeled.

Yes. And look, I might hate the role I've adopted in the last 10 minutes because I'm generally the softie on the comp committee. I love writing big comp checks for successful equity packages. But at some point, you do have to have some kind of linkage, and at some point someone has to sit there and say, “Okay, what are we getting for this?”

And you're right. For what it's worth, on this particular one, I'd probably do it, but I'd insist on non-market comp and non-market stock-performance metrics. If you build the biggest bank in Europe operationally, not just on stock-price value, then I would totally say you're worth the $50 billion.

That's why I'm saying a lot of it's in the details. It's really boring, but a lot of it's in the details. Let me be very clear here: if you're going to give someone $50 billion, $50 fucking billion, you owe it to yourself to spend more time thinking about what you're getting for your $50 billion than, “Hey, I'll give you $50 billion if it's valued at $500 billion in a few years.” You probably need to think about it a little more carefully.

I think that's what you should do. In my limited experience with my portfolio, these mini-Elon packages are basically all focused on 10X what the last guy paid. That's what all the late-stage investors do. Whatever I paid—$20 million—I just want $200 million for you to get it. I don't care whether there's a little dilution. I want to make 10X post-dilution, and then you get your piece, right? So it's what you aim at, but I think it's what a lot of late-stage investors want.

Yes. Again, go back. I actually think you're right. In which case, given the last round was at $100 billion or $200 billion, that's my comment here: 10X would be a trillion. In other words, the participation rate is just a little high. These are fine packages. This one looked a little—if the numbers bandied about are real, it's probably, ooh, that's a lot. Maybe you could pay a little less and get a little more, but it's a thing.

The one thing I will say is, I've interviewed 1,000 founders—Sam Altman and Demis included—and I've never interviewed anyone like Nik.

Look, remember last week you asked which stock I like? Well, I think it's an amazing stock with amazing potential in market cap, and I just want to be sure I get operational performance before I pony up the $50 billion.

I know it's Captain Obvious, but Revolut's all green, right? Everything seems—at least externally—to be all green. I'm sure there's stuff under the surface that's struggling. If you do these packages too late, it's too late.

I think that's right.

And that also means you have to overpay and pay up because you have to do these packages at the right time, right? You try to do this when the company's growing 4%, sure, but you've missed your window there.

Yes. I actually think that this is the second package that that particular CEO got, but yes, I agree.

9. Whatnot Finds Life Beyond AI

Team, where do we want to go from here? We have Whatnot, which obviously raised a very large round: $545 million at $20 billion. We've got DeepSeek raising $8 billion at a reported $74 billion. ByteDance bans distillation of U.S. models, which I thought was interesting.

I think we should talk about Whatnot, if for no other reason than it's such a relief that there's more to life than AI. There's shopping.

There's more to life than AI shopping, Rory?

Yeah, I think it's a great story. For background, Whatnot raised about half a billion at a $20 billion valuation, and it's a live-shopping company—the internet equivalent of QVC. The minute I heard that story, my response was, “That'll work.” You know why? If you look at QVC, if you look at the Home Shopping Network, these were the equivalents, in the pre-internet days, on TV: live sales, right? People enjoy that shit.

Someone explained Whatnot to me a couple of years back, and I was like, “Oh my God, it's a great idea.” It's not my space, not what I do, but that's going to work. You're going to have people live-selling shit. It's going to be a little bit of retail, a little bit of commerce. It's going to work.

The big 3 of this space have been QVC. Interestingly enough, by the way, that's now bankrupt, probably because all those people died and now they're replaced by Whatnot people, right? And then eBay—you know, we forget it, but eBay is the other quirky way to sell shit from the 1990s—and that's got a $40 billion or $50 billion market cap.

So what's interesting here is something where you look at it and go, “That's going to work,” and fast-forward 2 years and it's $20 billion. Now, it's growing 2X year over year. You would measure GMV, which was about $8 billion last year and is going to $16 billion this year, and then they get a 12% take. It's a great business. That's all. So I'm just like, “Yeah, go team.”

It is useful. Listen, I'm not an expert on Whatnot. I could speak more to Shopify, which blew out its quarter too, right? It's roughly related.

But I do think it's worth studying what isn't being destroyed by AI, right? What's going to happen with online shopping, online commerce? What's going to happen with restaurants? What's going to happen with cars? And there will be many good opportunities in spaces that aren't going to be destroyed by folks creating a poster in ChatGPT for free, and we should just study it more.

There's gold in the things that aren't going to be destroyed by AI as well as the things that are being decimated by AI. That, to me, is the only interesting part, right?

I totally agree, Jason, because look, Revolut's another example. I agree. There are 2 compelling large businesses catering for real, universal human needs—finance and shopping—that are building huge outcomes. I agree AI is most of what's going on, but not all.

But as an AI story, it could be worth 50X GMV. That's the only miss. Let's say they could pretend the GMV was revenue. What's 50X times $16 billion, Rory? Help me with the math. What non-LLM would be the next trillion-dollar AI startup?

Pleasingly enough, 50X times $16 billion is roughly Anthropic's market cap today.

Yeah, that's what I'm saying. We need another trillion-dollar AI startup. This one's a bargain at the Iconiq investment committee. We're getting this for $20 billion?

Yeah.

I do think there's plenty of that out there.

I think whether it’s literally GMV or not, there are plenty of folks getting benefits of revenue that make no sense. It’s not even just lying or cynical. I think investors, to some extent, don’t care as long as the growth’s there.

10. Software Winners Need Fundamentals

Final one, if I want to shepherd us: Shopify, which we mentioned, blew out the quarter. Atlassian blew out the quarter—the biggest jump since 2015 for Atlassian. Crushed it. Any takeaways from some of the big results that came out?

Yeah, just as I said, if you produce, you’ll get rewarded, right? I was delighted because when you pinned me to the wall a few months back and said, “Name names. What stock would you buy?” my first answer was the best one, which is just buy World Cloud, and it’s up 50%. But then you pushed me and I named some names. One of them was Atlassian, and honestly, 2 months ago, I felt like an idiot. I thought they’d pull it off, but it wasn’t there. Then, obviously, they nailed it, they got the growth, and the stock jumped.

If you look overall, some people like Datadog were down a little, and you lumped them together on the agenda, but those are different stories. Datadog’s story was just that everything’s amazing, but our biggest customer—and everybody knows it’s OpenAI, although no one says it—suddenly realized they maybe don’t need to spend $150 million and are spending less. So growth was down a little. But that’s because Datadog was trading at 18 times forward revenues, and now it’s 15. That’s one phenomenon of the AI-adjacent winners, which is very different from what Atlassian was going through, which was existential shit. They were trading at 3 times, and suddenly we nail the quarter and it’s an easy pop to 5 times, something like that. Those are different movies at the same time.

What the Atlassian story says to the Salesforces, the HubSpots, and your Canva, if it were public, is this: If you get it back on track with the fundamentals, the stock will follow. But if you don’t, and some of the others that you mentioned didn’t, then you’re stuck in 2x and 3x land forevermore until you get, as Jason said last week, Bending Spoonsed.

Yeah. I still think these are hard companies to run, to your earlier point. I mean, yes, Atlassian. But Atlassian also did something that Canva did, too, which I always find a bad sign—a sign of stress. Not Michael—Mike’s great—but they got rid of most of the free Loom seats, and this is what you do in times of stress.

The other thing that Canva did, because the revenue’s down, is push a lot of features into the higher-paid editions. It’s not the end of the world. Loom is not the breakout success of Atlassian, right? It is not. But getting rid of collaborative free seats, which is how we all grew up using Zoom, right? We could share and work on these together. That’s a sign of just how hard—even if you beat the quarter, guys, like Shopify or Atlassian, man, they’re leaving nothing on the table. These are not easy beats. This is not Anton at Lovable turning it around and adding $100 million last week without realizing why. Even the beats are hard.

The Loom one sounds minor, but whenever I see the base getting overly monetized or harvested, if nothing else, it’s a sign of stress in the organization, because no founder really wants to do that.

You’ve said that consistently, and I’ve totally come to the conclusion you’re right. When you talk to people one level in at some of these big software companies, and they’re doing a 7% or 8% quarter, and then you talk to a director of sales that you know, you suddenly discover it’s, “We’re jamming them on price. We’re jamming them on overages,” and you realize it’s just not sustainable. So I do hear you on that one. Overall, I thought it was a great quarter, but, yeah, sorry to lose your free Loom seat.

I still just worry whether the agents need all these products, but I hope so. I don’t want to be negative on Atlassian; I’m a superfan of Mike, we all are. I want it to win. But I also worry a little bit that it’s a Canva story coming, that it seems to be defining some trends. Now, you can say Atlassian’s very enterprise, right? Let’s not look at the developer side of things. Let’s not look at how we used to use Jira and other tools.

But our agents really don’t need these seats, and a lot of their revenue is still developer-focused, which is where I think the seat is under assault—a permanent, permanent assault. The seat is.

11. HubSpot Faces New Competition

I’ve got a provocative question for you. HubSpot today is sitting at $10 billion. How long will it be until HubSpot is bought by Bending Spoons?

I’m not going to dunk on HubSpot. My Series C investment in HubSpot—we did it at $47 million pre, so we’re still up.

A Series C at $47 million pre?

No, it was $70 million. I’m wrong; it was $70 million. I was thinking of Box.

It was probably hard to get done, too.

Yeah, it was hard to get done. I’m sure Brian, if he could’ve got one of the glamour people who turned him down and then did him later, he’d have taken them off of us.

Well, I wouldn’t go that far, but yeah.

Oh, I would.

You’re a podcaster, Rory. You stand up for yourself.

I’ll tell you why I don’t think they’re going to get Bending Spoonsed, for what it’s worth. I don’t know, right? I think there are so many things in the Airtable story that are scarier than they sounded, but I think one of them is that they only got 1 offer, and Bending Spoons is going to look at everything and it’s got to be perfect. Maybe they will buy them. First of all, it’s a lot—that would be a lot for Bending Spoons to bite off—but they could do it, right? I guess you can always line up the financing.

HubSpot, assuming they would sell—let’s just assume they would even sell, right?—and there are a lot of fiduciary questions here, there should be offers at $12 billion if it’s at $10 billion today. There should be, but I don’t believe there are. I don’t believe there are many.

I will tell you, at a meta level, if we want to break it down, there is an issue here, and it’s a structural issue in the world today, in the AI world. Just like if you’re Canva, the prosumer folks are threatened by ChatGPT. If it can be done in ChatGPT, even accidentally, you’re hyper-threatened. The SMBs—the HubSpots, Mondays, and others—they’re not really threatened by doing it yourself. That is a short if.

What they are threatened by is the fact that low-end competitors in SMB are really good. The low-end CRM competitors are exploding. The revenue growth from Monaco[?], Lightfield[?], Orasesell[?], and others is like nothing we’ve seen before.

You know, my first venture investment was Pipedrive. It would’ve taken 40 years to get competitive with Salesforce, right? It was just slow. That was the number 1 simple-to-use CRM. It exited for $1 billion—my first investment.

The problem across my portfolio is that you used to walk into a board meeting and the competition would be the guys bigger than you, right? Here’s what the big guys are doing. Now, if you walk into a board meeting and it’s SMB, they’re all guys that weren’t on the slide 24 months ago, and they’re really good, and their agents are good, and their LLMs are good.

The tough hand HubSpot has is that it spent the last 5 years beating Salesforce at the low end, right? It’s a CRM company now. It’s not a marketing automation company, and now the low end is so good. The number of founders who want to compete even in niche categories—they didn’t used to want to. This is the bear case for everyone in this SMB space, because there are just too many good competitors.

I actually think you’re exactly right. Genuinely, exactly right on that. It’s very well articulated. I remember the years of, “We’re doing CRM; we’re competing with Salesforce.” You can build really excellent software really quickly with a different twist using AI.

It’s why I was telling this to someone who runs a big PE shop in tech: If I owned one of these companies as a PE owner, I would just be at every Y Combinator Demo Day. I would be like, “You need to infuse some of that DNA quickly while you still have breath, and figure out what you can build.”

Are you fucking kidding me? With the loyalty that they have today, you think they’re going to stick? Let’s give a load of young people from YC a big chunk of money and say, “Hey, come work at this shit heap.” And they’re going to be like, “Yeah.”

Obviously, if you think it’s a shit heap, no. But all the PE companies, respectfully, are shit heaps.

You know why, actually, this strategy won’t work? You want to know the serious reason why it won’t work? Rory’s right, Harry. You know why it won’t work? Because all the hot startups have this model. They’re all picking off everybody.

I mean, I think one of my investments, Owner, has acqui-hired like 20 companies, and they get to go work for a reasonably hot company, right? So how are you going to compete with that when Rippling has hoovered up 30, Owner’s hoovered up 20, and Revolut’s hoovered up 10? That strategy worked 3 years ago. It’s too late.

Everyone is just sitting there hitting refresh, hoping these smart YC companies fail so they can hoover them up in an acqui-hire. I’m not kidding. It is a core strategy of many leaders.

Boys, thank you so much. That was awesome.