20VC：Jensen Huang 调查：接下来会发生什么 | Sequoia 退出伦敦、关闭 Scout 项目并终止 Evergreen 基金 | 纳斯达克 24 小时内蒸发 2.5万亿美元 | Anthropic 瞄准 1.4万亿美元 IPO 估值

He needs a pass. This is too important to AI and too important to the US. China loses big time here. China’s going to lose. It’s going to be harder not only to sell GPUs to China, but also to host them across the border. It’s all going to get shut down.

The AI trade doesn’t crack on the supply side. It’s far better to start low and go up in an IPO than to start high and come back down. You’d much prefer to be these guys than OpenAI right now. They’re going to have the money, they’re going to have the float, and they’re going to be done with the private begging bowl bullshit. Lucky us, all the billionaires can come back home. No one in 2026 is going to say, “Yeah, I love me those tech bros.”

You have now arrived at your destination.

Boys, so many things happened this week. I’m really looking forward to hearing the different perspectives. We’re going to start with Sequoia. They made a lot of changes this week. They announced, first, that they’d be departing their London office, and that some of the team would be moving back while others moved on to different things. They also announced that they were ending the Scout program and bringing the early-stage and growth teams together into one team. There were some personnel changes around it, but those were the core structural elements. I think they’re really meaty in their own right, so I’d love to break them down.

1. Sequoia Reworks Its Model

You guys know him better than me, but I thought you missed the most interesting part of what I read: they’re winding down the Evergreen program. I don’t care about the other stuff. There’s partner drama; people are going to come and go. No question, the kids from London are going to drop by the office in the Bay Area, and no question it’s better to be local and close, like 20VC is. We can talk about all this stuff forever, right? But the idea, which sure feels like it’s back again, that you should hold these assets for decades—generational assets for decades—what the hell happened there? Why the hell are we winding down the Evergreen fund?

Notwithstanding a few bumps this week with Jensen, what if they’d held all their NVIDIA stock until today? That’d be a good one. What about all their Google stock? I mean, literally, what the hell happened to the Evergreen idea? What the hell happened? This is the best idea in venture ever. You have the dominant brand, right? My understanding, reading about it, is that the LPs weren’t allowed to leave this Evergreen structure. You’re pretty much stuck, right? So why do it? Once you’ve got a couple hundred million or a few billion, go Evergreen.

No. Actually, I thought the Evergreen structure—which, to remind listeners, basically incentivized or almost pushed LPs to allow Sequoia to continue holding long after a public offering—came from the correct analysis that no matter how well you do as a venture investor, the real money is made after the IPO. They’ve been something absurd like 25% of the NASDAQ. So by definition, if they’d held all those stocks, they’d own a big slug of 25% of the NASDAQ.

It’s true, arguably almost always true, that if you hold a diversified portfolio of really interesting tech assets, over the long term they compound well above any kind of S&P return. So the idea of saying, “Hey, why don’t we just hold them?” wasn’t illogical. That was the thinking behind doing the Evergreen fund. There were probably 3 problems with it, though. The first problem is they picked a horrible time to do it, because they implemented it in ’22 or ’23, when stocks actually turned down. Over 20 years, that doesn’t matter, but if you decide to make everyone hold stocks and then a year later they go down 30%, it hurts.

The second thing is LPs don’t like it, because they have different liquidity needs, and you want to be able to make your own choices. The last thing, which is super interesting, is that you should always think about the comp plan, right? The individual partners in any venture firm can decide to hold their stock forever. Once you’ve cleared carry, they get all the return they’re going to get from doing that. If I was lucky enough to be Sequoia, and lucky enough to do Cisco and take it public in, I want to say, ’86 at a $400 million market cap, I can hold my shares, distribute to my LPs, and they can make their own choices.

In retrospect, a lot of them would look back and go, “Oh, I wish I’d held my Cisco stock.” But now you’re basically trying to force LPs to take their own medicine. It’s like, “Damn it, you want to sell the stock, but I’m going to make you hold it.” Frankly, it was complex and trying too hard. Again, it is intellectually correct in the sense that if you hold a diversified portfolio of high-quality Sequoia-backed venture companies that go public, you’ll have lots of variability in that portfolio, but in the end it’ll be a compelling portfolio. So it wasn’t a crazy insight. It was just trying too hard, with too much complexity. I think getting rid of it was just a, “Let’s keep it simple here, people. We can still make our money.” So that made sense to me. Sorry, that was a bit longer than I thought.

No, maybe when you have a CFO running the place, it makes sense. It all makes sense on a spreadsheet. That’s what CFOs do: they make it all make sense on a spreadsheet, right? I was watching something Ho Nam from Altos Ventures said. He talks a lot about generational investments, but he said they set up a series of continuation funds to do this, so LPs could exit if they wanted and stay in if they didn’t. I assume that if you did that perfectly, other than some conflict around how those funds work, it lets folks go either way, right?

Yeah. But the odd thing is, you’ve already been fully paid for what you’re doing. So now, as I say, you’re basically giving the LPs their vitamins. You’re saying, “Damn it, you will hold these stocks because it’s good for you. I’ve already made my money.” We’re going to talk about NVIDIA in a while, and one of the things we’ll mention when we get there is that there’s a Sequoia partner on that board from 1993. I bet he’s held all his stock, and I bet he’s worth a couple of billion bucks. So you can do what you want to do individually, but I don’t know if every LP in Sequoia would really have wanted to hold that stock from the IPO in, what was it, ’95 or ’96, all the way to today. To me, that one was an easy give. Let’s stay with Sequoia on the other stuff, though.

You know, the other meta thing about Sequoia—I know it’s captain obvious, but I think for founders and others it’s a reminder: the institution may be durable; no particular partner may be durable.

The CFO who’s the CEO is gone. All the London partners are gone. A bunch of the U.S. partners are gone. Now, listen, this is an organization designed to make a buttload of money, and it is not a nonprofit. So it all makes sense, right? But founders sometimes think of their VCs as being more permanent than they are, right? I float in and out. But, like most things, the founder CEO has the longest lifespan. Jensen, for example. The founder CEO will probably have the longest lifespan in tech.

Actually, I might argue with that, but your salient point is right. Change is the only constant there, and I give them credit for making change.

Totally. I think London’s important. I think a lot of people use this moment—

Well, clearly that’s not an opinion shared by everyone, but keep going.

Well, I think it actually is. When you look at ElevenLabs, which is a phenomenal investment of Sequoia’s, and DolarApp, which is an phenomenal investment of Sequoia’s in London, and Odoo, a great investment in Europe too, and Bolt, I think, bluntly, what they’re saying is, “Europe does have great assets, but we feel we can sufficiently cover it from the States.” I don’t disagree with that.

One thing I’ll add—and there are obviously two sides to all this—is that I believe in being close to founders. I believe there’s value in a London office. We can talk about that. But I’ll say, I don’t get invited to all the things you do, Harry, but at every elite event or dinner I’m at, Mati’s there. Half the time, Anton’s there. It’s hard enough to get to the Bay Area from London, but getting there from Sweden is a pain in the ass. Still, I don’t need to go there to meet these guys. Forget about raising money—you want to be in the deal and close it quickly—but especially if I’m not doing the pre-seed or something, I don’t think I need to go there. They’re everywhere, and kudos to them, because it’s work. I would almost think ElevenLabs were a Bay Area company. If it weren’t for Anton’s accent, I genuinely would. They’re everywhere.

I do think that at pre-seed, you want to be local, right? You could challenge that later. You’re there; you could challenge it. I genuinely believe that, even if it’s not captain obvious. But if Mati wants Sequoia, he’s there every month. You’re going to meet him. Sequoia is a good enough brand that you’re going to get a shot to put a term sheet in front of him. If you’re willing to pay up, I don’t know if it’s so crazy. But again, I’m murky on pre-seed. You’ve got to roll up your sleeves to do that job. That’s the Lord’s work: pre-seed.

Accepting all that, you go back to the decision: it was the right thing for them. I mean, I think—

For them.

Yeah.

Yeah, for them. Yeah.

You’re trading off maybe a little less coverage early on in the UK, but you’ll get the good ones. I think it’s what you said, Harry: it’s all about focus. We need to get everyone in a room. We need to get our shit together. I give them huge credit. They’re always good at reinventing themselves.

One thing I didn’t really get: I always thought the Scout program was relatively little lift for a very significant advantage. I was surprised to see that. I don’t know if you have anything to say about it, but that one just seemed strategically, huh.

Well, it did get cloned. I think it worked better when it was a secret. They were years ahead of everybody else. In the early years, I think it wasn’t to be spoken of if you were a scout. I don’t know if you had to sign an NDA or not, but it wasn’t spoken of, right? Maybe it’s that simple: it just got too flat.

To your point, we could talk about it forever, but I read this as not having the energy. I don’t want to run a Scout program with a small partnership. I don’t want to run Arc, a mediocre accelerator that isn’t really even Speedrun, let alone YC. I thought in the early days about running an accelerator for SaaStr, when our brand was the hottest thing in SaaS. I thought, “My God, I’d have to hire 15 people.” Then I thought, “I’d have to manage them.” My LPs would tell me to do this, and I’m like, “I’m going to jump off the roof if I have to recruit and manage 15 folks, all arguing.” It’s just too much. I don’t know if they have the time to do this and loop in six partners in London, in a different time zone, with different motivations and different economics. It’s too much if you want to have a tight partnership.

The other thing that I guess is part of it is, obviously—I'll share one other story. When Sam Blond, who’s now at Monaco, was at Founders Fund for a while, I didn’t understand Founders Fund. It seemed like too much capital for too few partners. I didn’t understand how so few people could deploy so much. But when it works, you need the ability to have conviction and write large checks into winners, right? Sequoia has some good ones. I know you can talk about what they missed out on, but if they have the conviction that every GP can easily—“easily,” in air quotes—deploy a billion or two regularly, then you don’t need all this other stuff. It’s a distraction. It’s exhausting to manage the folks in Arc.

I don’t disagree with you at all. Founders Fund deployed the growth fund in 11 months.

With how many partners and how many employees—how many investing professionals?

Honestly, like three—

Six?

…to five real ones. But they’re in Stripe, Anduril, and SpaceX with that fund, and no LP is complaining about the compressed deployment timeline there. Those are phenomenal companies, but you’re right. This is large amounts of money with very few people.

I didn’t even think it was possible. I’m like, how do you—how many people are here? Oh, no associates? How does this work? I don’t get it.

You have to be somewhat diversified early on, but widely concentrated later on. I think Founders Fund does a really nice job of both. They do it a lot. If you look at the early-stage stuff they do, they make a lot of little experiments. They do a lot of early deals that are small and kill them quickly, and then you ride on the growth fund. They concentrate down from, I think, 20 investments in the first fund to seven or eight. It’s exactly how you should do it.

Okay, moving swiftly on. We have Nvidia. Jensen Huang admits to having an affair with an alleged or suspected Chinese spy. The border, as a result, is opening an investigation into what happened and what the impact is. This wasn’t on my bingo card, I’m going to be honest, guys. Thoughts?

It obviously sucks on a bunch of different levels. Just to put it out there and then move on: it sucks on a personal level for him and his family. It’s all yuck, because you’ve got a personal drama going on and suddenly six billion people have an opinion on your married life after 40 years. That just sucks.

Moving on from that, because I’m trying to avoid commenting on it, it sucks from a company perspective. You’re right: this is the most consequential company in AI. I think Sean Maguire had a post basically saying, “Please don’t fuck it up just because the Chinese intelligence agency did some kind of kompromat.” In other words, let’s not lose the plot here. In one sense, I’m sympathetic to that. It would suck to lose Jensen, because I think a different CEO would run that company very differently and perhaps wouldn’t have the drive to make some of these deals happen.

But then you have to consider the third point: there’s a government investigation. To be very clear, it’s not an investigation into a relationship. It’s an investigation into compliance issues. It’s an investigation into whether there were breaches of security and whether there are issues with how the government handled it. Didn’t Nvidia lobby the government, the Justice Department, to back off on a test on MegaSpeed? There’ll be a whole bunch of stuff like that, and it’ll be messy. There’s at least some nonzero chance that the stuff that comes out is problematic, right? To repeat, that sucks because this is a world-class executive who’s taken this thing from ’93 all the way to today—the longest-tenured tech CEO, period, full stop. It’s hard to argue he’s not a success when you’ve had a $5 trillion market cap.

Listen, all politics aside, I think he needs a pass. This is too important to AI. It’s too important to the U.S. I think this is a point a lot of folks were making, one way or the other. Objectively speaking, he needs a pass, right? We briefly discussed on an older pod how Steve Jobs got a pass during the backdating stock option days. This isn’t the same, but we needed Jobs, and I think, at least for now, we need Jensen.

He’s got to retire someday. But it’s a bummer it was MegaSpeed in all of this. It’s a bummer that it’s not something less central to where these GPUs should be going, like whether it’s really okay to host inference and GPUs for China-based models in Singapore and Thailand.

It’s unfortunate, even beyond this spying thing, which I think—I don’t think he leaked secrets. Come on, it’s possible, but it’s moronic. It’s just a bummer it’s MegaSpeed, right? Nvidia has banned them now. Nvidia even said internally they’re not going to sell them anything.

Jason, I love you so much. You’re like, “Why can’t you just cheat with someone from SAP?”

Yeah. I mean—

Change your destination.

I don’t think it was intentional. It’s just a bummer that I think this is going to last much longer than it otherwise would. I know this cap table. Because it’s so tied up with an issue everyone’s been dancing around: Is it okay to sidestep selling GPUs to China by just putting them in hyperscalers or data centers a few miles away?

This made no sense. “We’ll let you train all your models. We’ll let you do whatever. Just please keep the chips a few miles over the border, guys.”

Two comments on that. First, do you need a pass? You’d like one, and you’d like this to go away. But to be clear, you’re right that there are all sorts of national security questions and interference questions. I don’t think this is as easy to make go away as some of the other things we talked about, like the backdating options with Steve Jobs. I think this is more problematic.

You can imagine a series of facts starting to emerge. You also have the dynamic of the government knowing about this early: Marco Rubio having a report, telling Trump, then not passing the information on. There’s a whole bunch of messy stuff here. So I don’t think you can assume a pass. You can hope for one, but you can’t assume it.

The second thing is that the really odd, amazingly foolish thing is this: My sense, just having listened to a bunch of the interviews Jensen’s done, is that without any lobbying from any Chinese agent or entrapment agent, he wants to sell GPUs to anyone who’ll buy them. It was all a mess. I mean, he—

Yeah, this is the part that’s going to backfire. There are so many clips of him pushing hard to sell these chips to China.

But my point is, I don’t think he was doing that because someone was lobbying him. I think he passionately believed that, correctly from Nvidia’s perspective. His job is to run Nvidia, and President Trump’s job, until there’s another president, is to run the country. From Nvidia’s perspective, they need to sell chips to anyone who’ll buy them. They don’t want another alternative provider based out of China eroding their market share.

My point is this: You didn’t get anything by entrapping him, because for entirely logical reasons, he’s been pushing that agenda pretty consistently.

No, China loses big time here. China’s going to lose. It’s going to be harder not only to sell GPUs to China, but also just to host them across the border. It’s all going to get shut down. It was already stalled in Congress. I think it passed the House and not the Senate, right? This is not what China wants. This is bad for China.

The only other thing I might add for founders and others, to get away from the salacious stuff: I was thinking about this before. In my whole career, I’ve been involved in 2 incidents where boards did investigations with CEOs—1 female, 1 male. Nothing exactly like this, but it’s similar in some ways. It certainly wasn’t MegaSpeed and transshipped GPUs.

The obvious insight is that you kind of get who you hire for the investigation. This won’t be true of Nvidia, because it’s under such scrutiny, but elsewhere, you should consider who investigates it, who you hire, and whether they’re biased in favor of an outcome, as objective as they may be. They’re paid, whether they’re a law firm or an investigation agency. Be thoughtful when you have to do something like this. Understand that the output may already be decided when the input begins.

The only comment I’ll make is that I understand what you’re saying: If you’re asked to run this investigation, you can set its terms and choose whom you hire as an attorney. To some extent, that pre-biases it. That’s arguably true.

But on the other hand, I think I saw someone make this point: When you’re doing an investigation that you know will be second-guessed by another investigation, probably by Congress and the Department of Justice, the scrutiny level goes way up.

One comment: This is an astonishingly long-tenured board. Four of the 9 board members are 2 of the original VCs, Mark Stevens and Tench Coxe, the original audit chair from ’93, Brooks Seawell, and the CEO of Synopsys. These are people who’ve been in this chair for 30 years and are clearly very loyal to Jensen. I’m sure they see this as something they need like a hole in the head. But people tend to be professional, and the CEO of Cooley, the law firm, is on the board and is in fact leading the investigation. Professional people, especially when they know scrutiny is coming, aren’t going to just close their eyes.

So, unfortunately, this thing is going to run, and they’re going to do an investigation knowing their homework will be checked by someone else who won’t be as friendly, who might be in Congress.

For sure.

2. What Drove the Market Selloff

I’m going to pull this back to the numbers that really matter: the Nasdaq 100 loses $2.5 trillion in a day. I had hedge fund manager friends calling me in a tone and depressive state I haven’t seen in years. That much value eroded in such a short amount of time. Can you help me understand what happened here? What caused this? What’s going on?

You can start by saying it’s clear proof of the old expression, “No shit.” We’re recording this on Tuesday, and this will be released on Thursday. The crash you’re referring to happened last Thursday. As a reminder, on Tuesday night it became pretty obvious the Dems were going to win, and win big. Then, on Wednesday, the stock popped. So the first response was to go up. A day later there was some, arguably, economic news, and then some angst about data centers. On Thursday, the stock popped back down.

A lot of this is fairly reactive. It’s some version of, “Oh my God, we can’t build data centers. All these people aren’t going to let us build data centers, so maybe we won’t sell as many chips.” But I don’t know how deeply thought through it was.

So you don’t see this as the AI trade cracking and showing its first signs?

I’m going to say this again, and I’m going to say it every week: The AI trade doesn’t crack on the supply side. It doesn’t crack because data centers can or can’t be built. It doesn’t crack because Nvidia loses market share to AMD. It doesn’t crack for any of those other reasons.

If the demand is there—and that demand will be evidenced in ARR for OpenAI and Anthropic—everything else will be roughly fine. There might be some companies that get ahead of themselves, like CoreWeave, but it’ll all be fine.

Looking at last Thursday and trying to extrapolate some kind of informed view—“Oh, I’ve suddenly realized we’re only going to sell $300 billion of tokens instead of $500 billion. Let the stock go down”—is just grasping at straws. I think a whole bunch of aggressive pods in hedge fund land got nervous and got blown up because they were over their skis. Remember, the good folk at Citadel and places like that basically take your checkbook away at a 5% drawdown and fire you at 7.5%. It doesn’t take a lot to have a bunch of these pods blow up. That’s all that happened last week, I think. No great thought.

Jason, do you agree with that?

I’d say 2 things. First, we’ve discussed it before: I think it makes sense for the markets to be nervous about this. There are a lot of assumptions, a lot of capital to be deployed, and a lot of debt required.

3. Anthropic Heads for an IPO

As we’re recording this, Anthropic is finally getting ready to IPO, and it’s a little murky to me what happened, but it looks like they couldn’t fully fill out the round. The round is $100 billion. They needed about a $20 billion private placement to fill it out, I think. Some of it was Google’s pre-commitment repackaged as part of this round, and they had to lower the size of the round and the valuation.

So the market is a little jittery about this, even though I think Anthropic also said they’re at an $80 billion run rate right now. You don’t get much better than that, but they weren’t oversubscribed at $100 billion—or so it appears.

No. Remember, pushing back on that: they’re not at the oversubscribed stage. We’re not in that stage of the process yet.

It hasn’t fully priced. You’re right.

I actually think they played it perfectly. Given the dynamics of what’s going on, they put a recommended price of $70 to $78 a share, which is relatively low. As a reminder, the last private round was $54 a share, and we’ll talk about the market cap in a second. So it’s up from the last private round on a per-share basis, but feels modestly priced. The whole plan, I’m sure, is to generate interest and then, if it’s successful, walk it up. It is far better in an IPO to start low and go up than to start high and come back. Once you start coming back, everyone loses confidence, and you’ve blown it. So I thought it was fairly clever.

From a valuation perspective, it’s literally up 20% to 30% on a per-share basis. The reason people are talking about $1.4 trillion versus $1 trillion is that when you do a valuation on the private side, you include all the options, so the same price per share looks like a bigger pre-money valuation. When you value it on a public basis, they only account for vested options. That’s why the newspapers say it’s roughly $1 trillion today, but about $1.3 trillion or $1.4 trillion if you counted all the granted but not vested options. To me, it all looks smart. They’re beating the numbers, they’re pricing it low, and the plan, I assume, is that people pile in and maybe they can raise the filing range, or at the very least get it done at the high end and declare victory. I think it’s perfect.

Guys, what does it go out at? I know you don’t like predictions, Rory.

Well, not things that can’t be predicted. I think it’ll be very successful. I think it’s a great company.

So you think it’ll go at $1.1 trillion and pop to $1.4 trillion?

Wouldn’t be surprised. In today’s market—I mean, yesterday’s market, today… Yes, in a normal market, if nothing crazy happened, I wouldn’t be surprised. Totally.

Look, it’s not my gig, but for what it’s worth, I think it’s going to trade flat.

Why?

Rory’s right in noting my point that it doesn’t feel like the round got filled out. I’m just using “round” for fun. It hasn’t priced yet. They can go high into the range. They can even raise it again, I guess, if they want. I think there’s still time. Maybe I’m wrong. Retail’s hard to predict. Don’t get me wrong: I’ve been a big proponent of this, but I feel like the market is fatigued here. They couldn’t raise the full $100 billion from it directly in the IPO. OpenAI fatigued its market. It was only able to raise $10 billion. We’re just getting fatigued. This is my view, so I could be wrong.

I just got my email from Robinhood saying I can finally be an Anthropic shareholder. Hooray. Very exciting. But I think we’re fatigued. I think we’re fatigued.

I actually just spoke to one of the largest allocators, a trillion-dollar fund, and they were like, “Listen, I’m not worried about this $80 billion. I’m worried about the next $100 billion, because that layer is very thin.”

Well, that’s really the same point, right? It does seem to have been fatigued.

How do you mean, the next $100 billion?

All of these companies are going to need more and more money, so to speak, but especially Anthropic and OpenAI. I just don’t think there’s the depth of capital to continuously fund them.

Again, it’s odd that I’m being the bull here. I think this is the first chance to buy a pure-play, next-generation AI company other than just picks and shovels, so I think there’ll be wild interest. On the capital thing, if they continue to grow like a weed, they’ll need a lot more capital. But if the growth slows down, remember, this is a company that will have $200 billion of cash on the balance sheet—more cash on the balance sheet than any other company in America. I think it’ll be fine. I think it will, as I say, trade up well.

Going back to my first point, it’s a great company. Strategically, all I can say for sure is that if you think this is a tough discussion, you’d much prefer to be these guys than OpenAI right now.

Why?

Because they’re going to get their IPO done in the next month. They’re going to have the money, they’re going to have the float, and they’re going to be done with the private begging bowl bullshit. I’m willing to bet that six to nine months from now, that CFO will have a convertible on file. He’ll raise, you know, $50 billion. They’ll do all the tricks you can do when you’re a big-ass public company. They will access public capital at scale.

Meanwhile, OpenAI is smaller, starting to reaccelerate, doing great stuff, but still unprofitable, still private, and still trying to raise $11 billion plus retail. Tough place to be. I would much prefer to be public. I think the exhale you’ll have if you’re on the Anthropic board when you go, “Ah, we’re public. We’ve got $200 billion. We’re kind of out of operating and compute, break-even depending on the quarter. We’re done. We’re golden,” versus, “I’m still private”—it’s such a better position to be in.

Yeah, Rory’s right. I assume if they can, they’ll do a convertible or a secondary or some sort of offering, one of the myriad variants they’ll be able to do at that scale, within four to five months. They’ll put another $40 billion, $50 billion, $60 billion on the balance sheet, and they’ll be able to. The float will be there, and that’ll hoover up that last bit of demand. It’s a good point. It’s just going to be easier for them to do it. Not easy, but easier.

Finally, to Rory’s point, it’s going to be a gift because we’ll finally get quarterly real numbers. They’re not going to disclose everything, but we’re going to learn a lot more. We’ve had to sit around trying to figure all this crap out from CoreWeave, Nvidia, and Nebius. Now we’re going to have Anthropic disclosing at least far more than we’ve ever seen before every quarter. It’s a gift to all of us.

4. OpenAI and Anthropic Compete

We mentioned OpenAI and being public. OpenAI did come out and say—this is Greg Brockman. I think it’s interesting, too, that Greg’s further in the spotlight. Greg is the one saying a lot of these things now, which I think is interesting; it wasn’t the case before. Annualized revenue grew 70% in Q3. Enterprise doubled and became a majority of revenue, which I thought was interesting. And obviously Codex is crushing and really handing it to Claude Code in a lot of ways. Any note on them closing $11 billion at $1.25 trillion, or would you rather be Anthropic?

No, I mean, again, I don’t want to be snide with any company. If they really did reaccelerate to 70%, that’s stunning.

And for that quarter, it looks like faster growth than Anthropic. What’s interesting now is the competitive two-way race on these products. It’ll be interesting to see token pricing. We’re going to discover the demand elasticity of token pricing, because I think these guys are going to push down on pricing. I think you’re already seeing some of that.

Then we’ll discover the demand elasticity of tokens. Does a 10% or 20% reduction in price result in a 21% increase in usage, such that revenue goes up? Or does it result in a 15% increase in usage, in which case net revenue goes down? That’s what we’re going to find out, because the basis of computation will be pretty clear: compute, capital, and token pricing.

I think it’s just interesting: if they’re growing at 70% now, what was it earlier in the year, at the nadir? 30% or something?

Yeah, it was 18%. It was a miserable quarter.

It’s just a reminder to me. I know I’m Captain Obvious again, but when Dev was on the show from MongoDB, what he said is there’s no loyalty with developers. I just think we can’t predict anything. We can predict that Anthropic and OpenAI will have more revenue 12 months from now than they have today, and we can predict it will be radically more. But running a very simple eval between Astra and Fable, or Astra and Opus, is truly trivial.

The entire time we’ve done this show, it’s been ping-pong between these guys. They both grow, and it’s going to continue to be ping-pong. No one else can acquire the compute they have, at least in the U.S., but it’s hard to predict what the ratio will be 12 months—or 100 days—from now.

They also came out and said that Dots, their agent product, is very expensive and compute-heavy. Does that raise the question of whether this agent proposition and consumer product is about to be challenged before it’s really able to hit prime time?

I do think it’s interesting. Maybe I added that question. You and Rory kind of jumped all over me. Rory said they’ll figure it out; great startups always do. And I believe that if you didn’t need billions and billions of compute when none was available, that would be true.

Harry said he’d do Instinct at $30 billion. It was a deal to die for. I’m waiting to see. These products are so good, but they’re just so expensive to run. I misunderstood. I thought back when we started talking about this that the cost to run Dots or Grok or Muse or Instinct was like four to five bucks.

That’s what it costs to run Lovable or Replit, okay? It’s VMs—non-persistent VMs that spool up and spool down for your website. That’s what it costs those guys, and it’s a good business, right? If Lovable and Replit can charge you $25 to $30, it’s a good business, even for Wix, right? At $200 million for Base44, I was wrong. It’s up to $20 or $30 a month to run these things.

The learning from GroqBot and Dots is that we want them to do even more. My Muse is running all day long for $20 a month. I know Rory thinks it doesn’t matter for Meta, and maybe it doesn’t, but everyone else is compute-constrained, right? If they gave up on video, I don’t know how long they can stay. We’ll just have to see.

I don’t know. Compute is going up. GPU prices are going up. There is no deflation in GPUs, right? So I don’t know if it’s sustainable. We’ll see. I just think—I’m sorry to ramble—I think ultimately these products are less interesting if they’re not fully persistent. If you can’t run an agent as often as you want, 24/7, doing your work, it’s just not that interesting. It’s just an automated workflow. It’s great, but it’s not that interesting.

Yeah, I hear you. You’re right: costs go down. And you’re right, short-term compute prices go up because of supply and demand. Long-term compute prices go down because of Moore’s Law. In the end, it’ll get cheaper. It’ll be fine. The question is, are you 2 years away or 5?

Maybe 5. Yeah.

Yeah. My intuition is the boys at Meta can carry the cost.

Yeah, yeah. It could be the one winner, right?

5. Three Deals on the Table

Investment committee time. Honestly, the team edited Rory’s statement last week, where he said, “Harry, that was so much better. That was so much more intelligently posed,” just for us to play in the office, which is nice. The way that we structure investment committees has changed within O’Driscoll Lamkin Ventures.

We have 3 deals on the table today. We have Zipline, raising $1.5 billion at $21.5 billion, 3 times its January price. Option one. Option two, Fal doubles to a $16 billion price on more than $800 million of ARR. And we have Revolut, the fintech juggernaut from London. The price is $115 billion in a secondary round, up 53% from its last round. Now, Mr. Lamkin, of the deals on the table—

Yeah. Listen, I like all 3 of these deals, and if we really don’t care about having a concentrated strategy—and I think the fund needs to—I’d be good doing all 3 of these. I think these are all great deals.

I think Zipline went long. I didn’t get Zipline years ago. This has been forever. I didn’t really get doing drones in Africa. I know people love the founders and the company. It never made sense to me, right? But fast-forward to where we are today with autonomous everything, autonomous delivery, autonomous Waymos, autonomous cybercabs. Zipline appears poised to be a winner. It seems rich, but I like it better than investing in Waymo. The unit economics are better, and I think it’s clearly a winner. I think we should make this bet.

Revolut appears unassailable, right? Unlike the others, the margins are more attractive, and it seems to be going from strength to strength. It’s the simplest one I could see doubling or tripling. But Fal—it’s just for now. I know Rory’s right: compute will become a commodity in 3 to 5 years. You just can’t get compute today. So Fal, Fireworks, and others—I have long-term concerns about these businesses. Fireworks was hacked twice this month. It’s brutal.

But this is unstoppable. In venture, sometimes we just have to be tourists. We have to be part of a wave. If you make me pick, I’m going to pick Fal because of the underlying trends for the next 2 to 3 years. I know some of the guys at Altimeter and others. I think we can flip this at $40 billion in the worst case, so we have more than our pref. So I’m going to do this one. But unlike all the other weeks, I’m good with all 3. I just don’t think we get the concentration for the fund that we want.

I remember Peter Thiel and my other good buddies at Founders Fund said, “We’ve got to pick.” But this is the toughest week. These are good ones. The reason is they all have the right trends going for them. Maybe Fal, with the 6% gross margins, might be an issue, but—

Is that a fact or just a snide comment?

I don’t know what it is. It’s going to be pretty low. These are good ones. That’s what I really believe. But obviously Fal might be the least durable of them, right? I could flip it around and say, if we really want to go long in our evergreen fund that we just set up—we’re going a little counter to the other funds; we’re going evergreen—we can’t do Fal because it’s just unpredictable. But Zipline and Revolut are enduring properties.

Oh, I’m starting to feel really uncomfortable about where this market is at. I want security and safety, and I think Revolut is the best asset by far here, with by far the best founder of our generation. Give me Revolut at $115 billion every day. Let me compound for 10 years, and I’m calling it a day.

Harry just wants to put his money in preferred stock of a bank.

Yeah, I—

And I respect that. Funny enough, they’re super interesting. To some extent, I—these were very different. They were all interesting and high-quality assets, but very different.

Let’s just go to Zipline. The biggest attraction of Zipline is that it’s N-of-1. People know what they’re doing, but they started off doing drone delivery of medicine in Africa. The big business now is drone delivery in the United States. They have a contract with Walmart. They’re doing a ton in Dallas. They’re rolling out in other markets. That’s the real positive here, and there’s a huge amount of logistics to make all that happen.

The scary thing is, they’ve done 3 million deliveries in the US. You probably only get $5 or $10 a delivery, right? You’ve got to do a lot of deliveries to make this work. This is an amazing company. I haven’t seen the numbers, but my guess is you don’t get there on a revenue multiple. You get there on the idea that this is the next infrastructure play. If people are willing to have drones drop stuff everywhere, you can make out like a bandit. It’s a big story and a plausible story, but at $21 billion, you’re really leaning into that Walmart contract. So that’s amazing, but scary.

Yeah, that was sort of the Waymo comp. That’s why I brought it up. They make sense if you ignore a bunch of things. It’s hard to argue with the market position, and the technology’s exciting, right? But the math may not pan out. How does the math pencil out to get our 3X?

Agreed. And then Fal is almost the opposite. It’s not an N-of-1. They have a differentiated position in video delivery versus more generic inference-based Together AI and Fireworks and all those. But in that broad market of people doing that, the argument is what you said—I think Jason, you said, “The trend is your friend.”

Right now, the total third-party inference market, excluding the foundation models, is $5 billion, plus or minus, and growing like a weed. So you probably get a couple of good years of growth here. But you’d be scared that the dynamics of that industry, which is wildly attractive now, could change at scale.

You’d be tempted to do that one just because it’s worked so well. We looked at these inference providers early on and were incorrect about the long-term structure of this industry. It turns out you might be right on that in the end, but when demand is trebling or quadrupling every year, no one gives a shit about the long-term implications. At $1 billion, $2 billion, $3 billion, they’ve been great. At $16 billion, you might be closer to the end than the beginning.

So you’re driven back to Harry’s choice, which is maybe I’ll just put it in Revolut, who are busy destroying—or taking away the business of—every European retail bank and taking their young customers. The attraction of that, to your point, Harry, about the next 20 years, is that when you look at the demographic of their customers, Revolut, even though we call it a bank, is mainly about money movement. It doesn’t do a ton of lending. But over the next 10 or 20 years, all those customers who are 19 to 25 today are just going to get older. They’re going to want mortgages. They’re well-positioned to be the biggest bank in Europe.

They are obliterating European incumbents. We will see market share for Revolut unlike we’ve ever seen for any historical incumbent, and we’ll see a much broader product suite. The depth of the market and the product suite they have means I really do think there’s at least a 4X from here, with the downside that we have key man risk on Nik. Do not let him leave the building. I want him nowhere near roads, cars, or any moving objects.

Jason, you mentioned Fireworks and the security incident there. Do you want to touch on that? Or we can touch on the security M&A, because CrowdStrike and Palo Alto have been making moves.

6. Inference Security and M&A

Well, I think the Fireworks thing is interesting from a security perspective. It’s niche, but Fireworks has had 2 major breaches this month, or in the last 30 days. It’s pretty bad. Two things that are interesting to me.

One is that everyone’s trying to steal API keys for inference. They’re all trying to steal Anthropic and OpenAI keys, and they have massive swarms of bots. Everyone’s been hit by this, right? Railway was hit by this, and Northflank, the other competitors, have been hit by this. The genesis of this is trying to steal any Anthropic or OpenAI tokens you can.

As an individual user on Ultra, on Claude, my spending was set at $100,000 by default, and I didn’t even know. If someone stole my key because it was on Fireworks and Eval, they could have charged $100,000 in minutes. That’s what they do. These crime syndicates steal the keys, scrub them, and use them to get cheap inference. There’s so much demand for it, as we’re talking about, right? They can spend that $100,000 in seconds. It’s gone, and that’s what happened. Fireworks was hit twice.

Everybody’s been hit. Any way you can get inside someone’s infrastructure and take their keys, people are going to do it. It’s the most lucrative criminal trade you can do right now, right? This is better than selling Rolexes you steal off someone’s wrist in London. This is the crime of the year, and it’s only going to get worse.

I know multiple other folks who have been hit similarly, and you’re going to be hit. It’s exhausting, and it’s just beginning. The flip side is, going back to Fireworks, this is terrible, right? But no one can leave. You can’t replace the capacity. Where’s Harvey going to go? Where are Stebbings, O’Driscoll, and Lemkin’s startups going to go? You just can’t. All these guys have seven-figure and up contracts. There’s nowhere to go.

But everyone’s going to get more and more sophisticated at stealing your API keys. Can Harvey trust them now? Harvey runs on it; they’ve trained their models and done all this. But what are we going to do? It’s very interesting and very complicated if even Fireworks couldn’t do anything about it.

That was point one. Was there a second point?

Oh, just that it’s going to happen to everybody, man. Listen, Fireworks isn’t that old a company, so maybe, as embarrassing as it is, their SecOps and DevOps teams are a little bit immature, right? I guarantee you they’re better than 99% of SaaS companies that just laid off their whole SecOps team last week as they’re trying to get fit, right?

Everybody’s going to get hit, and everyone’s data’s going to get stolen. It’s great for the big guys, the big security companies, and folks figuring this out, because everyone got hit in Black October, or whatever we called it, right? Everyone. It’s only going to get worse with these swarms. Everyone said they got hit by swarms of bots—swarms of them. It’s something we just haven’t really seen before.

Fireworks is… maybe they didn’t disclose everything that happened, but they were pretty clear that these swarms are difficult to mitigate. It’s terrible. This and outages were the things I… That’s why I wanted to sell my last company. It’s so stressful. Wiz said nine of the leading AI developers disclosed a swarming incident in October alone. Nine.

Wow.

Unit 42 said bills are $1 million before detection. If it’s $100K by default for basically a prosumer key, it’s pretty easy to steal a million pretty darn quickly, right?

Guys, which M&A is the most interesting, do we think? CrowdStrike and Palo Alto, Dialpad and Ada, Fox and Roku, or the potential of Silver Lake and PIF taking out Unity?

I think you have to break them up. You highlighted two categories of M&A that are very classic: a bigger company acquiring an adjacent product. That’s CrowdStrike and Palo Alto. They’re all doing small deals, around $400 million, for perfectly great products in agent security, which totally makes sense. The end-to-end platform—agent security is the number one issue.

CrowdStrike and Palo Alto are consummate acquirers, and they know their job is, if the new security issue is agent security, to make acquisitions and deal with the problem. So that’s incredibly obvious. One of my colleagues looked at this space, and there are typically three or four companies in it. The trick is to buy the hot one before CrowdStrike or Palo Alto intercepts it fairly quickly. So that made sense.

Then it’s the same thing with coding agents: Meta and OpenAI buying adjacent products all about the software development life cycle, which is all being redone from zero. I think those two groups of acquisitions—the cyber and the coding agents—are very classic incumbent buys of products to meet the needs of an evolving technology market. It all made total sense. It’s great to see, because there’s been a dearth of venture M&A, and it’s nice to see these guys coming in and making good offers.

When you get Dialpad buying Ada for $400 million in stock at a $21 valuation, if you’re the Ada investors, are you like, “Yay”?

For disclosure, we also saw Dialpad in the past, so I have an investment in Dialpad. We actually had a competitor to Ada, which we sold to Zoom three or four years ago, so I have some form in the space.

I think if you’re Ada, you’re a very good pre-LLM chatbot, a services chatbot, right? The problem is that you had a prior generation of technology. Maybe you’ve replatformed to an LLM-first technology, but you probably lost the wind out of your sails. Your ability to compete now with, at some level, Sierra, maybe Decagon, is just hard. So I think getting acquired as part of a bigger company makes sense.

7. Legacy Companies Face AI

You know what depresses me a little bit, related to this? We can talk about the deal set. There was a Marc Andreessen clip this week, and it really resonated with me. A year and a half ago, when folks started talking about AI-native startups, it kind of triggered me. I don’t really get triggered, but it got to me. What magical sauce do they have using GPT-3? These kids are so brilliant because they grew up in the age of AI. I think a smart CTO or two can figure this stuff out, and indeed they can.

But as we’re rolling into 2027, guys, with companies like Ada, I’m so disappointed that these non-AI companies can’t get their acts together. They can’t run their companies differently from 2021, 2022, and 2023. They can’t move faster. Their teams are stuck shipping at the same rate. Hooray, bolt-on LLMs, but look at poor Canva, right? Too slow. Everyone’s too slow.

This isn’t a Bending Spoons deal, but if you can’t accelerate your team, you’ve got to find some sort of soft landing. I added this to the list because everyone said this was an amazing exit. I don’t view it as an exit. If there’s no liquidity, I don’t think it’s technically an exit. It may be the right thing for both sets of investors, right? Dialpad gets another player, probably with $50 million of revenue that’s somewhat enduring, and some folks.

But I’m bummed that people can’t get their act together. It’s really true how much faster “AI-native startups” are than they were four or five years ago. I just did an investment this week, and you could pick at it a little bit. I love it, but the rate they’re pushing out software is maybe the fastest I’ve seen from anyone.

Culturally, for your portfolio, 12/31 might be the time you’ve got to give up. You had two years, and you get these email updates: “We’re going to put together an AI-native team internally next year.” That was your 2025 plan, guys. All these CEOs and startups have the right ideas. They all get it. They’re just too slow, and I’m about… I’ve kind of given up, right?

So is this an exit for the Ada investors? I don’t think it is, but it’s probably the best option that’s out there, right? Craig’s a great CEO. Dialpad will find its way to an exit. They’re probably at $600 million or something like that. It’s still going to be brutal in today’s IPO world, but at least they’ll find a way to an exit.

But no one can move. They just can’t move like the kids these days, and kids could be older. It’s actually not even an age thing, but it is truly a generational thing.

If I were Orlando Bravo, Rory, I’d be paying $50 million a year to Mr. Jason Lemkin to come in and talk to our CEOs about how to make our very monolithic, old, stale organizations AI-first. Make every CEO see what Jason sees, because I think that’d be the most valuable $50 million any large PE fund will spend.

I thought you were going to say you’d have him talk to all the companies you want to acquire and make them so depressed about their prospects on a standalone basis that they’ll sell to Orlando for three times run rate.

Very possibly. That would be great, too, for Orlando.

You know what the problem is, too? I’ve been thinking about this a little bit. Sorry. You wouldn’t have to go down this rabbit hole for too long.

Then you might think, okay, so that's true, right? The pre-AI-native companies just can’t get there fast enough. But why don’t the AI guys just buy them? Why don’t you just buy them, buy the contracts, buy the customers, and have your agents buy them out? Actually, the problem is the AI-native guys are just growing too quickly. It’s not worth it.

It’s never worth doing.

It’s not worth it.

It’s never worth doing, no.

It’s not worth it. You both have to be growing within a band for this to be worth it, right?

It’s not worth it because the truth is, you buy someone else’s old contracts, and then you incur all their problems. The customer’s never grateful to you for solving them. They’re just mad at you. So I agree. The point is, it’s what you said, Jason: for all these older customers, it really is march or die. You either get with the program or you’re toast. It’s not even interesting to say it anymore. It was interesting two years ago when you started saying it. Now it’s just a given, so let’s move on.

Okay. It’s like shape up or ship out. Well, Cloudflare’s kind of done that. They shaped up, baby. They had 20% layoffs earlier in the year, and they’re crushing it, up 39%. Then you’ve got Freshworks crossing a billion with 30% fewer employees. God. Efficiency at its finest, huh?

Palantir is at $2 million per employee. It’s very enterprise, but it’s $2 million per employee. Growing 100%, growing 98%.

So we just didn’t need the people.

I think AI may be the excuse, and we didn’t need the people. AI opened our eyes to the reality of it, when it was always staring us in the face in 2021, working 20 hours from home and doing our four side hustles. It was always true; AI made it clear.

I think that’s the real truth, yes. To some extent, there’s probably AI leverage, but I think the real truth is, especially when you’ve got that kind of growth, if you’ve got a giant secular demand pull, you don’t need to do all the marginal things you were doing in 2021 to make revenue happen. You just have to staff the machine and take the business as it comes, right? Therefore, you can have compelling efficiency. I agree. It’s eye-opening how efficient some of these companies have been able to be. That makes sense. If you have $2 million and $5 million contracts like Palantir, and revenue doubles, you probably can’t hire headcount fast enough to catch up. You literally can’t stop your margins from going up, which is a wonderful place to be.

The one interesting thing about Cloudflare—and we don’t have to spend too much time on it—and even Twilio, right? These are not brand-new companies. We were just talking about AI-native versus non-native. Cloudflare may be a more agile company, still founder-run, than Twilio, which isn’t. I mean, who knows what it’d be like if Jeff was still there? When he was on our show, I think it’d be even better. Putting that aside, sometimes you wonder how big a deal agents are. Not AI, not LLMs, but agents. In some ways, I run everything on agents, but it still seems niche. But this is what’s fueling Cloudflare’s growth from 20-some-odd percent to 39%, at almost $4 billion in revenue. It’s agents, and it’s agentic spend at Twilio, too.

In some ways, for 2027, all I want to talk about is agents. This is what’s driving it. We had LLMs, we had inference. Those are good ones. I know we just did Fable. I feel good about the deal. It was a good 2026 deal. But all I want to talk about is how agents are changing buying, building, sharing, and learning, because agents are going to do the majority of everything in our space. Cloudflare is a public beneficiary of it. Going forward for the fund, if it’s not an agentic beneficiary, I don’t want to hear about it anymore.

Is that why Atlassian falls after it warns on seat and subscription growth? Is that it? If it’s not agentic, I don’t want to hear about it anymore.

Yeah. Listen, we all love Mike, but I just didn’t love his idea about how critical the front end and the UX are on 20VC. Agents don’t care. They don’t care about your UX and your UI, and they don’t care about any of this stuff. Humans do, but they’re a declining percentage. The percentage of everybody’s user base that’s human is declining. Even in vertical B2B, it’s declining.

Yeah. You talk about Atlassian: they had the “April of last year, the world is ending” moment. Then it exploded in the back half of the year. It was probably a relief rally, and now it’s a little bit like, okay, you’re not dying immediately, but the new trend is not your friend, because the new trend is agents, not engineers. That means the seat license thing is brought into question. But it’s still up on the year.

Jason, you said something about Canva earlier. In the challenge bucket as well is Figma falling below $10 billion for the first time in a long time. I think today it’s about $10.2 billion. Any thoughts there?

Well, my meta thought is—and again, I don’t want to talk about politics—we have a new Congress, and we’re going to have a new chair of the banking committee and others. We haven’t had a change in the executive branch. Even with that, I’m worried antitrust gets harder.

I know maybe Dylan’s happier running the company. He may well be, right? You know him, Harry. I don’t know him. But it would eat at me—not every day, because you’ve got to move on in life—that we’ve now fallen into the single-digit billions. When I had $20 billion plus a big package for him and others, plus instant liquidity for the investors, in 2022, this is just a tough one on the lifetime IRR. Do you really think you’ll get back to $40 billion or $50 billion? Do you really think agents are going to be using Figma? Even though the numbers are great for Figma, the trends are not ideal.

I worry antitrust is going to get harder and that we had this brief window where it was good. If it gets harder, it’s just a bummer for everyone in the venture ecosystem. It’s a bummer.

Agreed. At the same time, just to commemorate success, you’re dealing with a 40%-plus-or-minus revenue grower, with very strong free cash flow, even though they’re still operating positive. The trends are good, and the whole hit was because they’re spending more on AI, so that’s going to have more gross margin impact.

Yes, I agree. It’s a thing they tell you when you have these M&A discussions and someone says the old, boring statement—but it’s boring because it’s true: “Hey, you’re getting an offer of X now. If you turn down X, you could work for the next three years, do an amazing job”—because Dylan has done an amazing job—“grow your damn company 40% a year, get it to free cash flow positive, get the damn thing public, and at the end of that, you’re going to be worth half what you are today.” No one ever said capitalism was fair or easy. That often is the argument for taking that preemptive offer, which obviously they did, except the government wouldn’t let them. But you’re right, Jason. It’s going to psychically hurt. I don’t know well enough to know how it gnaws at him, but yes, I’m doing all this to get back to half of what I would have had for free in 2022.

It’s also really hard when you go into these companies and visit them. You’ll see it really is hard to retain people. It creates a malaise among the top performers. You can fix it with RSUs, additional grants, and targeted stuff, but there’s just this malaise among people. Even with Figma, we can talk about whether seat businesses are stable, but almost 40% growth at its scale—you can’t argue with it. It’s S-tier. But with the stock so far down, this malaise sets in. It sets in.

As founders, you can hunker down: “I’m still worth a billion,” or whatever, $500 million. What the hell does it matter? It’s not liquid anyway. This is my life’s work. But that malaise is hard to pull out of, and it’s hard to attract new people. So you double down on the existing folks, and I don’t have an answer to this one. When the malaise sets in, it’s tough.

The one thing I would say, Rory—I’m not an expert in the public markets. I’m not even an amateur. But I truly believe if you look at all the most impacted companies, they are seat-based ones that the markets do not believe will benefit from agents and AI. I believe Figma, unless it radically changes, is going to be in the doldrums, relatively speaking. The ARR multiple is still around seven or eight. But it’s going to be in the doldrums because people believe seat models are in trouble at Monday and HubSpot. The markets believe these companies aren’t going to die—maybe “terminal” is the wrong word—but that more and more will be agent-driven. They’ll bid up Cloudflare and Twilio, the old Twilio, and I don’t see any way the seat-based guys are coming back.

But Jason, okay, so you’re the new CEO of Figma. What do you do differently, then? I hear all that. What do you do differently?

Well, first of all, I think I knew, like, 10 months ago, when the world changed.

I’m not close enough to know Figma today. I will say this: agentic design—the ability to go into an agent, into Claude, into Replit, into whatever, and say, “I want a perfectly designed end-to-end experience for my app”—does not exist today. Design mode in these apps is a joke. I use them all day. I use Replit’s Canvas design today. It’s a joke. Claude Design is really neat, and it’s built in. It’s still kind of a joke.

Lovable and Replit are both at a billion. They are great tools for product people, but for design, they are a joke. So it’s not too late to do that, but you would have to just own it. I would go all in on agentic design. We will deliver—you know, I’ll pick Notion because it’s famous for it—we will build your app agentically. Every page, every element of it, will be as good as Notion or better, using our AI and our agents. It’s going to cost us a lot of basis points. It’s going to cost us a lot of gross margin, and we’re going to have to cut a bunch of things for our enterprise customers, which is where all our growth is.

But if we don’t do this, we’ll be like Adobe. We’ll have our role, our workflow role, but we’ll be part of the past because agents will bypass us. We talked a little bit when Claude Design launched. I generally think—and I use all these products every day, don’t get me wrong, thank you for them existing—they are a joke from a design perspective. They’re a C-minus. So there’s still a chance, but you can’t do it at the Canva pace.

I actually agree. We’ve looked at the creative market, and I’m instinctively not creative. But one of the things we did was work with 2 or 3 graphic designers, literally watched them do their work, and had them do the same task in Canva, Figma, and some of the modern video editing tools.

What you saw, Jason, is that when they use one of these modern ones that’s very agentic-based, it all just goes quicker. They type in what they want, and it shows the node display. Then you ask them, “Why? What was the experience like in each of these 3 or 4 different products?” You get some version of, “It just happened easier. I was able to launch and just do it seamlessly with the agent.” And you didn’t get that experience in Figma Make and Figma—whatever the other one is, right? There are 2 different ones, depending on if it’s graphic or coding-type stuff. But yes, you have to start with the product and really make it for that customer, really make it super usable.

Tim, is there anything I’ve missed? The wealth tax we can talk about.

8. California Rejects the Wealth Tax

Lucky us, all the billionaires can come back home, right? It’s always fun to watch California. The country swung Democrat like we haven’t seen since 1974. It was a real, “We’re done with these people. We’re voting for the other guys,” top-to-bottom swing. We might have a Democratic senator from Mississippi. It’s an amazing swing.

At the same time, California, the most prototypically Democratic state in the country, at some stage said, “Yeah, we’re doing all that, guys, but we’re not doing this. We’re not doing this wealth tax thing,” which is really interesting.

What’s the takeaway from that, then? I’m a Brit.

I think, one, there’s generally a suspicion of propositions. We talked about this 6 months ago: there’s generally a no bias in propositions because I think Californians recognize it’s an incredibly stupid way to make legislation. So there is kind of a default no.

On top of that, I think the no vote won the argument. They made good noise. It’s impressive because no one in 2026 is going to say, “Yeah, I love me those tech bros. I love me those tech billionaires.” But give credit to the Google founders, give credit to Stripe. They were able to make the case: “Look, this is just not good. You’ll lose these people. They’ll go to another state. You’ll lose their income tax. You’ll end up net down.”

I always said I think the voters are smarter than we ever give them credit for, and they basically said, “Yeah, we’re going to get rid of the old lot. We’re going to vote, make a clear signal we’re done with what we have in Washington, but we’re just not going to screw things up totally here, people.” That one was a no. It didn’t pass. And then the one that banned them from doing it again—which was, I think, Prop 42 or whatever one, the one after that—actually passed. The one that says, “Not only are we not doing a wealth tax, but you can’t try and do it again.”

So I think it was a very emphatic mid-50s vote, but it was a clear, “Please don’t do this.” A big loss for the side that wanted it. I thought it was good.

The only thing I might add is, Sergey and friends did outspend the unions 10 to 1. I do think that impacts propositions, if nothing else.

Especially propositions about money.

Exactly. In fact, I think this was very important. Listen, I don’t network the way Harry does. I don’t have his network. But I spoke to 2 quiet billionaire founders, not CEOs, who were both planning to leave. Not the Zucks—they were already planning to leave when this happened.

It’s not just about this. It almost ties back to the beginning of this conversation with Sequoia and London. It’s a bummer if the best people leave California. It’s even a modest bummer that Elon, politics aside, left and went to Texas. It makes sense for his current businesses. It was going to happen anyway. It’s a modest bummer when more and more folks just leave.

It’s not just billionaires. These are not passive wealth managers managing strings of laundromats and FedExes. These are folks building. You don’t want them leaving California. You just don’t. I genuinely believe if this passed, it would have sent a clear message that California is toxic to the most successful entrepreneurs. Parker said it, from Rippling to others, and I think it would have had a domino effect. Not as big as they said on Twitter, but I knew folks were already planning to leave.

It’s just not good for the center of AI and wealth. Anthropic’s going to generate, on average, $41 million for each employee in the IPO. The average employee—and that’s the average—is getting $41 million. You want to kill that? You want that to go to Miami? You really don’t. You really don’t. $41 million for the average employee. More wealth created in this IPO—far more wealth—than any other IPO that’s ever been created for employees. The founders are down to 1% each, 1-point-something percent. Crazy, right? Crazy. $216 billion going to current employees and $160 billion in the pool.

Wow. That’s real money.

Yeah, it’s real money. Nothing like that. You don’t want them leaving. I think Anthropic’s a hot enough company that it needed to be here, but maybe they could’ve moved. Maybe they still could’ve attracted some talent to the beach. It’s possible.

But I think even if the company doesn’t move, people can move when they’re making their tax decisions, so we agree. There’s nothing more to say. It collapsed. As I say, it’s super interesting to see it collapse in the context of an otherwise massive blue swing, which I would argue, as I said, shows people were awake and voting for a clear centrist majority. So all good.

This $41 million average for Anthropic, right? The founders are each selling $264 million, which I don’t think is a big deal, right? But it’s pretty crazy.

It is. Good for them.

Wow. All right, boys. Rock and roll. What a busy week, huh? Thank you so much for joining me.

No problem. See you soon.