Anthropic 的 Fable 被美国政府禁用｜Wix 与 Adobe 跌至历史低点｜Mistral 正按200亿美元估值融资

Rory O'Driscoll

Anyone who has been blindly loyal to Elon has gotten stupidly rich. At face value, this is a Rubicon moment in the history of the AI industry. It’s the first time that the US has ostensibly regulated an AI model based on capabilities.

Good intentions bite you in the ass more than evil deeds.

This week, number 1 on the agenda: SpaceX completes the largest IPO in history. Then we hit on Anthropic launching Claude Fable on Monday, and the government banning it by Thursday. What does this mean for sovereignty?

Next, Salesforce acquires Fin for $3.6 billion, one of the nicest teams in tech. And then, finally, Adobe beats and raises, but the stock falls 6% as the CFO exits. This and much more in an incredible week of news.

Rory O'Driscoll

Outside of China, there are basically no good open-source models. Where are the US open-source models? There’s a partner here at Benchmark who says any liquidity for pre-AI SaaS companies is top-decile performance. Any liquidity at all.

1. SpaceX's Record-Breaking IPO & Elon's $1 Trillion Day

Ready to go? Okay, boys, I am so excited for this. We have a very special guest. Ev, thank you so much for joining us, man.

Rory O'Driscoll

Yeah, yeah, Harry. Thank you for having me.

Now, where else are we going to start? SpaceX, the largest IPO in history. It was a very successful IPO, and it’s held its price really well in the last 24 hours. It’s hit $2.7 trillion.

Elon Musk has actually earned in 24 hours what Warren Buffett took a lifetime to earn in terms of net worth, and he is now $1 trillion richer than the next person as a result of that increase. How do we evaluate the SpaceX IPO?

Rory O'Driscoll

Well, with jealousy. It was an amazing outcome, and you said it’s traded well. Arguably, it’s traded way more than well. It’s up 30% or 40%. I haven’t checked today because I just got off the course after playing, but it’s been an astonishing performer.

On the day, I was trying to decide: Is there a number above which Elon gets into trouble with Bill for leaving money on the table? I was watching it during the day. It went up to 30%, and I thought, “Bill’s going to be mad.” Then he dialed it perfectly. I think it ended at 19%, which is the perfect pop—the designed pop.

So he delivered. Not only has he changed the world and built the biggest, most valuable company in the world, pretty much without any price discovery—remember, we talked about this last week; he didn’t do any price discovery. He just told everyone the price he was going to take, went out and got it, and then landed the plane at a 19% day-one pop, which is kind of at the high end of perfect. I was like, “This is perfection.”

The last 2 or 3 days, it’s just traded up nicely since then, so it’s amazing. What do you say? It’s really hard to say, “I made $1.2 trillion, but I might have left $50 billion on the table. How do I feel?” I think he feels okay.

Yeah, I mean, that one piece about all these IPOs: We can’t help ourselves, because the company starts trading and you can do the math on everyone’s net worth immediately once it starts trading. But I do wish there were a moratorium on even talking about—or, even if you own the shares, looking at—a stock price when the shares are still locked up.

What I mean by that is that, in all of these cases, a traditional lockup means that the venture insiders, the CEO, the management team, and the employees typically can’t sell for 6 months. What often happens when you have so little of your shares trading—for example, right now, only about 4% of SpaceX’s total share count is being traded—is that it’s a tiny amount of float.

That allows for all sorts of weird things, like gamma squeezes, which we can talk about, and other things that make the share price insanely volatile over the first few months of trading until the lockup releases. Can we just stay on that? What is a gamma squeeze, and how does it impact SpaceX’s price?

2. What Is a Gamma Squeeze and Why It Matters for SpaceX Stock

Rory O'Driscoll

Yeah. So, a gamma squeeze is when people are allowed to trade options on a stock, which opened today. SpaceX options started trading today, so it’s gamma-squeeze day.

Ev

If I buy a call option, I’m buying the right to buy a stock at a certain price. There’s a market maker on the other side of that trade writing me the option, and what they need to do on their side in order to hedge their risk is buy some of the stock directly.

A gamma squeeze happens when you have a ton of people buying a ton of call options on a stock. You then have a bunch of counterparties—a bunch of market makers—who have to structurally buy the stock in order to hedge their risk. You create this cycle, this self-reinforcing loop, where the more call options people buy, the more forced buying there is from market makers, which then forces more buying from counterparties and encourages more retail investors to buy more call options because the stock is going up.

When only 4% of the float is trading—only 4% of the shares—there’s such a thin market that something like a gamma squeeze, where you have this forced-buying, reinforced loop, can happen very quickly. That’s not to say the stock is going to collapse or anything, but when a company starts trading, especially a really hot company, the price action is very fun to talk about because Elon can make Warren Buffett’s net worth in a day or whatever.

Nothing matters until the lockup is gone, because that’s the only time anyone can actually— it’s almost like a private market. It looks good, you love to look at it, and you put it in your little spreadsheet, but you can’t take it to the bank. It matters much less than the stock price 6 months from now. That is what we should be talking about, and that’s going to be the barometer by which we can grade investors and insiders.

There have been plenty of IPOs before where people do the victory lap on day 1 of the IPO, the stock goes down 60%, and then it’s a different situation once the lockup is up.

Have you regretted joining this show already? In 6 months’ time, will this stock price be above or below where it is today? I’ll give you the over or the under.

Rory O'Driscoll

Because of the retail mania around the stock in particular, I would personally probably take the under in 6 months. Not because I don’t think the company is going to be valued extremely well—I have a couple of funny stories on this—but because of the 4% float, call options coming online, and all these engineered things that are going to make the stock price go up over the next month, including some of this index inclusion, where there’s more forced buying.

There are just no shares available. There’s a lot of forced buying and a lot of retail activity, and it’s going to be popular for people to buy call options on this. I think it’s still going to be worth a ton, and I think it’s going to trade really well, but if I had to go over or under, I’d go under 6 months from now.

As you know, that’s where I come out when you asked me. But there’s just so much in that. First of all, I just, in passing, had a laugh at that—you’re exactly right there.

There’s so much room between $2.6 trillion and even the last private round at $400 billion. There is $1.8 trillion of value between this and the last private round, so there’s lots of room to adjust. But I do think you have it right.

The 2 comments are, first, my gut would be lower than higher 6–12 months from now. Second, I think you’re also right about the short-term comment: In the interim, there’s just a lot of news that will be strong, good news and strong technicals in the face of a low float, which means the short-term bet is much harder to call.

We can talk in a second about Crusoe, where I believe Benchmark has a stake. I believe they just announced that’s going to close. But the funny thing is, when I do the show, I always try to avoid having opinions that I don’t have the courage to act on.

So I started saying to myself, “Okay, Rory, today options became tradable for the first time, right? If I think it’s going to go down, you can make that bet, right?” You go out and price it, and you realize how hard it is to make money.

Roughly speaking, 6-month, at-the-money puts—the idiot version is that you have the right to sell the stock at the same price 6 months from now—are going to be expensive, depending on the estimated volatility. Volatility is what drives the pricing of options.

Stepping back a million miles, options are the coward’s way of shorting a stock. If you short the stock naked, you can lose all your money. If you buy a put option, you’re taking a bet, and the worst case is that you can lose the amount of the consideration. It’s the baby way of shorting, and I’m a baby.

I look at it and go, “Oh my God, it’s roughly 20%.” If you wanted to buy an at-the-money put, it would be, based on yesterday’s estimates—and I haven’t seen the updates today; options just started trading—20% of the share price.

If the share price is $200, it would be $40 just to buy the right to sell the stock back at par. That means if it goes down to $160, you only break even. It has to get right down to $120 to make a 2x.

And that's a 2x on a security where you can lose all your money, right? So I'm looking at it and going, “I might have these opinions about the SpaceX price going down. Do I have the guts to put $1 million on the line and say I believe it's going down and buy those options?” I decided I didn't have that just yet, right?

3. Elon's Long-Dated Call Option Playbook: From Tesla to SpaceX

Rory O'Driscoll

And I will say, I think the one thing that people very much underestimate is—I had a coming-of-age moment as a young value investor at the time. I think every value investor's gone through this. I was in this value-investing club in college, where you basically read Benjamin Graham and Howard Marks and think you're smarter than everyone else and buy things at 5 times P/E. After college, I was still on that kick, and with all my college savings, I shorted Tesla.

I feel like everyone has gone through this experience if they were a value investor at some point in their life, where it was like, “Oh my God, Tesla. Maybe it's not a fraud, but it's so overvalued. It's an auto OEM,” all these things. I lost all my college savings shorting Tesla. That was such an unbelievably valuable lesson for me in my life, and I was able to rectify it years later, which many people aren't. But I was able to rectify it when I went to Kleiner Perkins.

4. SaaS Winners vs Losers: The Good/Bad AI Framework

The very first investment I led in 2022 was actually SpaceX. When we were going through it, and when people—LPs or anyone else—had questions around, “How much upside is there?” I think it was at $120 billion or something. “How much upside is there here?” The way I talked about the investment was like, “Look, the numbers alone get you to a solid return, but it would be dumb to not incorporate what Elon can do.” He sells the market on these incredible long-dated call options, like these tech call options. He's made his whole career doing it.

At the time, whatever it was, 2019 or whatever, Tesla wasn't worth the $400 billion of its market cap, but he convinced everyone that, “Hey, I'm going to figure out full self-driving, and that's going to drive $1 trillion of value. It's going to take 8 to 10 years.” Then you get to that point and he's solved it. Tesla, on the numbers, still probably isn't worth $1 trillion, but now he has Optimus. That's another 10-year project that's going to create trillions of dollars of value if he figures it out. And guess what? He's probably going to figure it out.

5. How Musk Solved His Compute Problem Twice

With SpaceX, he's done the exact same thing. At first it was rocket reusability, and then the second one was, “Starship's going to work.” Now he has orbital data centers, the Mars mission, the moon, the moon mass driver, all these things. He sets up these stories, and he honestly usually accomplishes what he's going to lay out. They're basically these long-dated call options where he can go to the market and shareholders and say, “Look, on the numbers, obviously you can't really look at the numbers of SpaceX. If you were just looking at the numbers in the P&L alone, it'd be very hard to even get to a $2 trillion valuation as a fair market value.” But he says, “I'm going to figure this massive thing out,” and that's where the next trillions of dollars of value are going to come from. Historically, he's had such a track record that the market's willing to believe him.

When we look at the assets that you mentioned, that big new story from today—that exercising the right to buy Crusoe at $6 billion is looking like an incredibly prescient deal—I think it's at $4 billion today, and it's going to be $6 billion by the end of the year. He's acquired one of the best teams with lock-up, and so he's got retention of them baked in.

Rory O'Driscoll

It was for a price equal to 1/3 of the volatility of the stock between yesterday and today. Exactly. It was a great deal when he did it; it's a better deal since then. Yeah, exactly right.

The odd thing is, he solved his “What am I going to do with this compute?” problem not once but twice. He solved it first of all with, “Oh, look, I have Crusoe,” right? I forward-bought Crusoe to fill the gap in Colossus. Then, obviously since then, but before the IPO, he announced the contract with Anthropic for $1.25 billion and the contract with Google for, I think, $700 million. So roughly $2 billion a month of compute, again filling the gap in the Colossus revenue stream and doubling his revenue.

So I agree. Between $24 billion a year in CoreWeave-like revenue from Google and Anthropic, and then the $6 billion run rate—or $6 billion year-end potential—from Crusoe, yes, his AI business is now, as a matter of fact, a statement of reality. As of the minute those contracts kick in in September, the revenue run rate of the quote-unquote AI business across Google, Anthropic, and Crusoe is larger than the revenue run rate across the entire SpaceX and Starlink business.

So, yeah, just great corporate execution. I think it's the move-fast company. Everyone else has spent 2 years thinking, “Yeah, we should probably do something in AI. Maybe we should build a data center.” He's like, “No, I've built 2 data centers. I didn't get the model working great, so I bought a company to fill it with coding, and then that wasn't enough, so I did 2 huge contracts.” Moving right along, people. It's your turn now. I mean, it's still the execution speed that's just so impressive.

100%. And you know, I love your comment on the long-dated call, because the really weird thing is contrasting it: The volatility in the stock is super high, and that's what the option traders are saying. But you contrast that with the implied predictability of the long-dated call options. They're basically saying they're highly predictable.

In other words, the stock is moving around like a crazy thing, but the valuation is effectively saying, “Hey, we, the equity holders, believe—as I've believed—that this guy has done these things. They're worth whatever.” I was going to say X, but that would be stupid because the thing itself is X, so it would be self-referential. But they're worth $400 billion today. He has ideas to be worth $1 trillion more, and I'm just going to give him 100% credit for it.

I agree that is the secret to the fundraising. If it ever stops, if the time to deliver goes too long, there's a big gap between fundamental value and anything you've got here. But right now, he gets the benefit of the doubt like no one else. And the weird thing, because I've been thinking—obviously, anyone in our business who's not thinking about it this week is a [__]—is the competitive advantage that gives him, because his cost of capital is low.

Rory O'Driscoll

That's right.

Because no one else could have built—I mean, if you got into your local bank and said, “I think AI is going to take off. I have a crackerjack engineering team. I really can build Colossus 1 and 2 in less than a year. I'd like $24 billion, and I don't have any contracts, but I reckon shit'll work,” you'd have got laughed out of there. He gets the chance to roll the dice and it comes up, right?

No one could have done that unless they had the cost of capital. And the cost of capital allows it; it's kind of a self-reinforcing cycle. It allows him to make these bets that no one else can. I mean, he's earned the right to play.

Rory O'Driscoll

I remember both for the X deal and when xAI was fundraising, there were a lot of people that were, again, on just the pure dimensions of the companies, finding it hard to look at both of those investments and think that they were going to be good deals.

The X takeout at whatever $45 billion or whatever it was, and then just generally xAI, felt like it was too little, too late to keep up. What people forgot in that moment is that anyone who has been blindly loyal to Elon—in terms of, any time he's asked for money, you could just give him money for whatever he wants you to invest in—they've all gotten stupidly rich. Every single person. There's like a small village of these people, by the way. There are all these funds that you've never heard of. To be clear, 95%—like, 95% of their cumulative invested capital—is in SpaceX and/or other Elon companies. They've all gotten stupidly rich.

And by the way, you could say that about Tesla shareholders, too. Anyone that's held Tesla for 10 years, anyone that's had anything to do with Elon and just trusted him with their money, has gotten unbelievable returns out of it. So the amount of surplus goodwill he has to burn down—I think SpaceX, I don't think it will, but I think it could go nowhere operationally for 5 years, and people would still be believers in him, because he's basically done it for shareholders every single time that they've gotten involved with him.

And to your point, Rory, he has a much lower cost of capital because he has an army of people, like a small village of people, that will blindly give him money to do whatever he wants because he's been such an unbelievable steward of capital to anyone who's given him money.

Yeah, I wrestle with that. You're right, and I wrestle, to be honest. I don't know if I could get my head around that belief statement in an investment memo, even though I objectively recognize—if it's true—that you're right. I know some of, as you said, the small village, and it may be a small number of people, but they're going to have very large houses in that village pretty soon.

So, they’re going to be an Incline Village, for all the obvious tax reasons. So, yes, they’re going to do it. I would wrestle with that kind of approach. The rational in me says that’s not the way capitalism should work; even the best person should have all these different deals.

But you’re right, it’s worked, and it does mean it’s a single point of failure in the sense that, if it goes wrong, it will go wrong for everything, because they are going to roll in Tesla and it’s all going to be one big happy family. It’s going to be awesome, right? But it does mean he’s kind of taking on this ever-larger burden of making more and more return for more and more people, and the whole mystique of it is tied up in that sentence: anyone who’s piled in has never lost money.

There’s a little part of me that goes, “Wow, that’s a tough way to live your life,” versus, “We do deals; some work, some don’t.” But it’s worked for him, and it’s probably going to. When something works that well for someone for 20 years, they don’t change it.

Will Tesla and X be together in 3 years’ time?

Rory O'Driscoll

Look, I don’t know. It seems to me that it’s the thing that’s happening. I thought the CEO of Berkshire Hathaway said it’s probably easier for Elon if they’re together, and I reckon if I was worth $1 trillion, I would probably solve for things on the basis of, “Oh, it’d be so much easier if I only had 1 board, not 2. I had 1 board to share all this.”

It seems to me that it’s a thing that could happen very comfortably. Again, back to what I’ve said, there ain’t going to be a ton of votes saying no, especially after you’ve made him $2.2 trillion.

I think if you look at the Polymarket and Kalshi odds on a Tesla–SpaceX merger—again, not that those markets are so big that they’re that indicative—I think the probability they have on there is something like 70% in the next 2 years or something like that. So, the free market seems to believe that it’s more likely than not, but I’ve no idea.

On the other hand, some poor fool bet that it was 95% likely that Spain would beat Cape Verde last night.

Rory O’Driscoll

That’s right. I saw this for an $85,000 gain.

So, yeah. So, close to you.

Rory O’Driscoll

Yeah, but Elon is more predictable than a Spanish soccer team, so keep rolling. Sorry, I cut you off. The coffee’s kicking in.

He’s not used to such caffeine. It’s okay. The second on the list is Anthropic. Anthropic launched Claude Fable on Monday. Unbelievable reviews. I actually had a founder on this morning who said it was the ultimate game changer for them. Unparalleled.

The U.S. government bans it by Thursday, igniting this kind of global sovereignty argument over how far governments should reach into private companies. How did we analyze this?

Rory O’Driscoll

I’ll start with trying to put together the facts, because you’ve got to start with that, right? Obviously, this is the frontier model to the Mythos model, which was just explained step by step by Dario. When they announced that, they said, “This is so dangerous for cybersecurity that we’re not going to make it available to everyone.” So, they kind of prefigured that this was dangerous, right?

6. Anthropic's Claude Fable Banned by the Government

Then, obviously, they announced this model, which is effectively a front end to that model and, quote-unquote, shouldn’t be able to do this kind of cybersecurity stuff, right? So, they prefigured that this was problematic.

And then what happened was—again, maybe sticking with the facts first of all—it looks like Amazon discovered a case whereby you could, in fact, interrogate the model and have it give you some cybersecurity input, right? So, they pick up the phone and ring the government.

This could go back to the fact—I’m sorry, I’m a bit incoherent here—but the 2 parties in question don’t really like each other, right? So, no one’s assuming good faith, right? I read Dario’s statement, I read David Sacks’s statement, I read all the other stuff, and you could literally say everyone, according to their rights, is correct, right? But there’s just 0 communication. That’s the big thing: there’s just 0 communication, 0 trust.

So, in terms of the actual facts on the ground, I see Dario’s point in terms of what he was saying: why that “jailbreak” of the model didn’t represent a meaningful threat, and it was coherent with why he thought Mythos in general was threatening. It was a very logical, coherent argument, as you’d expect from a world-class scientist who is a very, very smart man.

At the same time, politically, I get why the other guy said, “Fuck this. After 90 minutes, I’m pulling the plug,” right? And it’s going to be really problematic. So, that’s kind of the big-picture comment.

Maybe drilling specifically on the cybersecurity comment, what they’ve said consistently about Mythos is that it’s not that any cybersecurity bug it finds is so terrifying that, “Oh my God, no one else could have done this and now you can.” It’s that the scale at which they can find bugs and problems is what’s terrifying. AI doesn’t get tired. It can find 1,000, and you can’t defend yourself against 1,000.

When this 1 thing happened, their comment was some version of, “Hey, look, yes, I get that this thing gives cyber advice, and the way the model was released, it wasn’t meant to do that, right?” To say it wasn’t meant to do that, technically, it was a foot fault, right?

But they were saying to Amazon, and they were ultimately saying to the government, “The real danger of Mythos is that they can spin up 1,000 or 10,000 different instances, all of them finding cybersecurity things that would overwhelm us, and that’s not what’s happening here. This fault wouldn’t have allowed that to happen.” So, they were technically correct.

However, if you run around telling everyone that you’ve got the scariest thing in the last 20 years, and it shouldn’t do any cyber, and you’re not going to let it do any cyber, and then it does cyber, and they don’t like you anyway, they’re going to pull the pin on you. So, that’s just what happened. After 90 minutes, they were like, “Fuck this. You’re not taking us seriously.”

You can see it. They’ll be very logical: “I’m a scientist. I’m a computer scientist. Let me explain why this doesn’t matter.” You’ve got a bunch of people saying, “I’m the chief of staff for the U.S.” They had an assistant chief of staff on the call, reporting to Susan Walsh. They’re like, “Dude, if you think I’m going to brief the President of the United States on multiple instances of cyber incidents, and he’s going to get that, you’re dreaming, right? We’re pulling the pin here because you have to stop it.”

They were like, “You stop it now, or we’re going to declare this export restriction thing, which means you just can’t export it.” It’s very restrictive. The terms of this are pretty restrictive. So, you can see how it all happened, right?

Both sides’ logic, in the context of where they’re coming from, made sense to each of them individually, but it’s just a communications mismatch. And it’s problematic for Anthropic because one of the big things here is the Export Control Reform Act, which is the thing upon which the ban was put.

Unlike the thing that happened with Hex-Rays, there’s no judicial review here. This is clearly within the powers of the administration to make this declaration, right? So, they’re outside, right? They don’t have a ton of leverage in court, right? So, it’s a tricky situation.

You think the hardest part of this story is just that there’s a lack of reliable narrators?

Rory O’Driscoll

Yes.

You know, like you mentioned, Rory, there are 2 parties that clearly don’t like each other. There have been a ton of conflicting reports on this.

Rory O’Driscoll

Yes.

Not that journalists get things 100% right most of the time anyway, but there’s been actually contradictory reporting on the way things went down and how it all went through.

I actually think what matters most, though, is what happens next. I think we’re going to know a lot more about the significance of all of this in a few months’ time, in 6 to 12 months’ time, than now, because on the face of it, this is a Rubicon moment in the history of the AI industry. It’s the first time that the U.S. has ostensibly regulated an AI model based on capabilities.

The first scuffle with Anthropic was due to a disagreement over basically the contract that they were signing and what the government could use the models for. They were not regulating it or restricting it based on capability, to non-U.S. citizens. They are now saying, at least on the face of it, that they are regulating an AI model based on the capabilities and what those capabilities could do in foreign-adversary hands.

I think that’s a huge deal, because what we’re going to see is whether they actually mean that or if this is again just another battle between Anthropic and the government. When OpenAI or Google get to Mythos–Fable-quality models very soon, within the next 3 to 6 months, we’re going to see if the government actually wants to regulate models based on these capabilities and if we’ve crossed this Rubicon where now the government is going to gate access to intelligence.

Guest

And I think the most interesting thing is if we extend the analogy even further. There have been some people on X who have talked about this as a possible future. Imagine if we get ASI. Imagine if we actually get things like recursive self-improvement kicking in and we actually have artificial superintelligence. We have this vision of geniuses in a data center. What happens when the government is gating access to superintelligence? Who gets access to that?

Do NATO allies—do just a select few allies of ours—get access, and do their citizens get access? If we think about how powerful a lot of people think AI is going to be, and the power of the government to either give or restrict access on a country-by-country basis or a citizen-by-citizen basis, it gets pretty freaky from a macro and geopolitical standpoint. Imagine if you have the economic implications of the United States and China having access to superintelligence and Greece not having it. That's just a random country, but imagine if—

What about Argentina or Brazil?

Avi Eyal

Yeah, yeah. Any of these—if you only have 2 countries that have superintelligence and everyone else is metered or gated, it makes a lot more sense why people are starting to take the idea of sovereign AI much more seriously. The only issue is that none of these sovereign AI plays have amounted to anything yet, and so it's sort of a—yeah. So I think, in general, it's a Rubicon moment, and I think we're going to know a lot more in 6 to 12 months than we do now about how important this moment in time was.

Agreed. And, yeah, I'm sure if I am the Mistral AI shareholders, I am lighting a candle in church at this one and going, “This is the best thing that happened.” But I want to go back to it. The funny thing about this thing is it's so illogical. What I love about some things in politics and crises is often people end up on surprisingly the wrong sides.

And if you look at it, Dario, the guy who's been saying, “This is dangerous. We all need to be careful,” he's been funding the PAC for regulation, right? Meanwhile, on the other side, you've had the administration—David Sacks, I might be doing better, hard to say, but excellent work trying to keep the government out of AI's hair, right? You've had a bunch of VC investors giving money to the “Leave AI Alone” PAC and kind of feeling like they're with the administration, and suddenly everything's mixed up.

Now Dario's in a very tough position of saying, “When I said it was really dangerous and we should regulate it, what I meant was it's really dangerous and we should regulate it, but not regulate me.” Oh, it's a tricky one. And then, on the other side of the table, all the, “Oh, leave it alone. We should deregulate this thing.” Suddenly, this administration, as I've said, is taking one of the strongest actions we've ever seen against the technology.

Right? So it's kind of everyone's—as they piece through what's happening, everyone should be looking, going, “Hang on, I thought I was on this side of the table and you were on that, and we've just switched,” right? Which, again, makes it hard.

7. Does the Export Restriction Act Even Make Sense?

Forgive my naivety. If you were internal within Anthropic's research teams, what would you do now?

Guest

Well, first of all, if you're internal to Anthropic's research teams and you're not a naturalized U.S. citizen, you step away from the keyboard, because otherwise you and your organization are going to be in breach of the Export Control Reform Act, and I'm willing to bet there's an attorney at Anthropic who knows exactly the penalties for that, right up and down the line. So, first of all—and let's get frank—a huge percentage of these teams are overseas, highly intelligent—

I think Andrej Karpathy is actually one of them.

Avi Eyal

Oh, wow. I mean, there's a reason my wife, on the 5th year, the day we were eligible, applied for our green cards, because at some point they'll turn on you. So, first of all, those guys step away. What do you do if you're an American and you're there?

The weird thing is, I really like what you said because you triggered the thought that if they don't treat the other labs the same, then Anthropic may have some kind of due-process claim. Because—I'm guessing here—but, yeah, if the Export Restriction Act—I mean, there's probably some pretty untrammeled, executive-administered power because it's a national-security thing, so it's delegated. But I do think if you don't treat everyone the same, it'll be problematic.

I'd laugh again. It could just mix things up. Does OpenAI go with, “We're as good as them, but then they don't want to get regulated”? Or do they go with, “We're not quite as dangerous as Anthropic, so you kick those guys in the nuts. We're the AI that's not quite clever enough to fuck up your cybersecurity, so we can sell to everyone”? It's just really zany.

But if they are comparable and they don't ban all of them, then they have a problem. You're right, because they're being illogical, but if they start doing it, it's a huge moment.

You're right. I'm just struggling here because you said that. Are we done, then? Because you said about 6 to 12 months. 6 to 12 months today. I mean, it's years in prior cycles.

Avi Eyal

That's right.

We were planning an Anthropic IPO.

Avi Eyal

Yeah, this is definitely a downer. I can tell you that there's someone updating the prospectus.

So, you know, Titanic on Goa goes 7 out of 10 holiday. Yeah, downer.

Guest

So, here's—this is a bit of a tangent, but I think it's related in that some of the—if you talk to some smart people in AI, they will tell you that, actually, with the right harness and the right amount of test-time compute, you can use open models to find all of the vulnerabilities that Fable can find. And so, actually, this is sort of a nothingburger because you can replicate this with non-Mythos-level models.

I think, again, the legibility of AI to the administration—I mean, the legibility of AI to a lot of people in Silicon Valley—it's just such a fast-moving field that if you're not inside one of the labs, it's just hard to keep up with everything that's happening all the time. Now think about DC and the average age of someone in the administration, or a lawmaker in Congress or the Senate. The legibility of these things is very hard.

So I do think that by saying, “Hey, we've created this god model that's very dangerous,” by pounding their chest, Anthropic has painted a target on its back. I would say the inverse of this whole thing, where you could say, “Oh, well, that's a negative for Anthropic and the labs and IPO prospects and everything like that,” the massive positive from this is this idea that with the right amount of test-time compute—basically the right amount of inference thrown at a given query into a model, or a harness with an open-source model—you can replicate these results.

Sorry, the balloons are blowing up here, guys. Oh, my God. That just exceeded test-time compute, too. Yeah, yeah, it happens. Okay, keep going. With the right amount of test-time compute, we can keep this thing on the track. I'm interested. Keep going.

Avi Eyal

So, with the—Noam Brown tweeted about this recently, where he said, “The way we think about these benchmark cards, these scorecards where it's like this model has larger numbers than the last model, aka good”—he's like, “That's all wrong to think about because the models actually perform very, very differently if you just continue to throw more compute at it at test time, at inference time.”

8. Why Benchmarks Are the Wrong Way to Think About Models

And so the really, really good thing for, I think, all the frontier labs and just the inference market in general—companies like ours, like Fireworks, that run inference platforms for their enterprise customers on open models—all of these companies that have to do that are in the token path—is that it's very clear that if you just throw more test-time compute at any frontier-level model, you continue to get results.

And it's like, you almost—they haven't found where the wall is, where you stop getting better results the more test-time compute you throw at it. And so we've had all these step-function increases in how much compute AI models use. We went from very simple autoregressive next-token completion to these agentic models that do chain-of-thought. Agents use an order of magnitude more inference.

And now you have this whole thing around, “Well, if you want to have Mythos-like performance, just spend a lot more compute—do a lot more inference, spend a lot more tokens.” And so I think the other side of this is that even though we've had such an insane growth in the amount of tokens processed for the industry over the last 3 years, we actually might see another kink in the curve as more test-time compute creates more and more of these unbelievable results for capabilities and things that you can do with models, whether or not they're Fable or Mythos or whatever OpenAI comes out with, but open-source models as well.

But what you're saying on that is—and, yes, I've seen the same statements—which is that, yes, Mythos can get there quickly, right? And it actually is the defense that Dario is offering: that an open-source model, just given enough time, will find many of the same issues, right? Because what it means logically, then, is if we can't have Mythos at all for national-security reasons, then we can't have open-source models provide those capabilities if they have more than X compute.

Avi Eyal

You're right. What it points to is that this is going to be an unsustainable medium-term position, right? It's hard to imagine, unless the U.S. government really wants to start regulating AI up and down the board, that this is going to sound like it was a needed decision one Friday afternoon. It's going to be a very hard decision to implement with any degree of coherence across 12 to 24 months.

And for what it's worth, I think that's because the original premise, which is that this is dangerous, is in and of itself incoherent. There are risks, but the idea that—I mean, I think this is where the kind of quasi-wolfism, overwrought quasi-wolfism, is biting everyone in the ass.

That was going to be my question: to what extent is it a model capability question versus a miscommunication question from your marketing?

Avi Eyal

Well, it is a model capability question. They can do the thing that they've said, and it's also true that the open-source stuff can do it. A non-frontier model can get to the same place.

I love this expression: the Russian army—quantity has a quality all its own, right? The ability to find 1,000 bugs in 2 hours is more terrifying than the ability to find 1,000 bugs in 1,000 hours, right?

So there was an issue that had to be addressed here. It wasn't just marketing, and oddly enough, in a weird kind of way, they tried to address it, but I'm docking credit, right? One thing in politics I often notice is that good intentions bite you in the ass more than evil deeds. Machiavelli explained it years ago, right?

They were trying to do the right thing, and then they got caught in this buzzsaw of, “We've warned that this is dangerous, but now that you have an instance of it, people are coming at me.” And the real truth is they're trying to say it's dangerous, but we're the good guys, so trust us.

I think the stepping-back comment is this: private citizens don't get to run around and say, “This is really dangerous, this could be awful, this could cause world damage. Oh, but by the way, we're the arbiters of the decision-making.” You've made yourself part of the political process because you claim to have invented the most dangerous thing since the atomic bomb. You are ipso facto political, and you'd better get good at politics really fucking quick.

Guest

Zooming out, I think the most important thing in all of this is that we step back from Anthropic and realize that the models are in place now, whether they're from Anthropic or elsewhere, where they can autonomously find and chain multiple vulnerabilities together and orchestrate an attack autonomously.

If they get good enough that they can take down parts of digital infrastructure that run the US economy or Western economies, then there is some concern about a nation-state adversary, like North Korea, China, or Russia, having those capabilities. So it's a real conversation that we're going to have to have as Western democracies very, very soon, and we probably should have already had it.

Anthropic right now is the poster boy for it because of the relationship that they have with the administration and the things that they've said, but this is just true for AI now. This isn't an AI discussion; this isn't an Anthropic discussion anymore.

9. Will Anthropic IPO This Year?

Yeah, yeah. So, before we move on, should Anthropic go public this year? I mean, statistically, more than 50%, yes, because in a world where this doesn't get solved for 4 or 5 months, there are so many other problems that our heads are going to hurt, right? So you have to say that, probably, in a world where it does get solved, then yes, they should go public.

Still statistically, but obviously, again, I'm just a simple Bayesian. If your prior before was 90% plus—which it should be, because we've just seen a stellar reception to a company that, even though it was called SpaceX, turns out that the vast majority of the market is in AI, at least as they frame it now—you've just seen that have a stellar reception.

You'd be pretty much of an idiot to see SpaceX trading at $2.4 trillion and say, as the board of Anthropic, “Let's hold on for a better market.” Of course. So your prior should be that they're going public with 90% certainty. This thing is a significant wrinkle, and they're going to have to work through it. So it lowers the probability, but still way more than 50%, because they'd be crazy not to.

Guest

I think they will. I think they will. I think one last final anecdote, and we can move on, is—I mean, obviously, they're already on this incredible trajectory that's been reported on publicly. We had a founder in our portfolio who was active in Claude who, after a day of using it, said, “With Fable, they're not going to hit $100 billion of ARR this year. They're going to hit $150 to $200 billion.”

That was just an anecdote from their personal use because they thought that the model was so unbelievably powerful, and they were going to spend that much more money on it. But I think, with or without Claude, we'll see. I think it's about as good a market as any for them to go out, and I think the results and the numbers are just going to be eye-watering.

I'm sorry, Avi. While you were explaining how America has changed model capabilities, Rory was struggling with European inventions of bottle caps or water bottles. [Laughter.]

Rory O'Driscoll

I'm aware of the European nanny state. By the way, it's really sweet that, as an English person, you call it Europe, given you're not in Europe anymore. But I know what you meant.

Well, speaking of Europe, and to the point that you said about sovereignty, Mistral and others are rubbing their hands with glee at the potential. Mistral is in talks to raise $3 billion at a $20 billion valuation. They've actually been incredible in terms of their foundation model. They've scaled to over half a billion dollars, with some of the biggest enterprises in Europe. Will we have many more sovereign models like Mistral? How do you read this?

Rory O'Driscoll

You'll have a push for it, right? And look, even if—I mean, this is going to sound awful—but even if the European alternative isn't as good as the US alternative, there will be cases where it's not quite good enough, but you don't have the sovereignty risk. So, yes.

This is the whole European economy, isn't it?

Rory O'Driscoll

I was getting up for doing the swing. I was going to build up to it, but you just took the words—

Come on, dude. You took the words right out of my mouth.

Rory O'Driscoll

Absolutely. No, I mean, there are areas where they're genuinely better, but there are a lot of—you know, Europe is still trying to do some kind of GPS alternative. The more we exert our sovereignty in the United States and do this kind of thing, the more important it will be to decouple from it, right?

Would you prefer to have the second-best model that you have access to, or the best model that gets cut off once every 6 months on a random basis? I think you're seeing it in defense procurement. Yes, there's still stuff that you can only get from the United States, but more and more, if you can get it from either place, you'll do it.

And I think Europe, to the extent it perceives this as important at the margin—I mean, it'll be the classic European thing. They'll care enough to fund it, but we won't be able to care enough to find the $100 billion to compete with it, right? So, good for Mistral. They're in a good place. I find it plausible that that would continue even if it's not the best model.

Do you think Greece is going to come out with a model soon?

Guest

They should. Well, I don't know if this is real or not, but I saw on the timeline that Rio de Janeiro—the city of Rio in Brazil—apparently post-trained or fine-tuned an open-source model and created their own city model, and it was near-frontier or something. It might have been fake, but that's what I saw.

I think, again, everyone—it's sort of like what we've seen in defense, where, especially in Western Europe, you've seen so many startups pop up to respond to the demand from Western European countries to have their own defense supply chain and manufacturing and all these things. I'm sure we'll see the exact same thing in AI, where sovereignty will become much more important.

The thing that's really struck me is just how hard it is to actually build these models and have them actually be good. If you think about it, outside of China, there are basically no good open-source models. NVIDIA has one that's pretty solid now with NeMo Tron, but where are the US open-source models? We don't even have any in the US now.

And so I do think the talent, the capital, and the focus over a long period of time to actually be best in class at pre-training, at mid-training, and at post-training—doing everything that you've done, or doing everything that you need to do, in order to create a really, really good model—is an extremely scarce skill set. I don't think it's this fungible thing where everyone's going to be able to do it.

And so I don't know if that leaves people where maybe the answer for a lot of countries is to just take an open-weights model from China or elsewhere and post-train or fine-tune their version and have that be what they start from.

But I also don't think that it's just this trivial thing that every country or every region can have a sovereign player, because most of the sovereign players have fallen far behind. Mistral itself has fallen very far behind on the actual model side, and they've done a very good job becoming this inference platform and the FD model and all these things. But outside of China, there's been no great model producers besides people at the frontier.

Yeah, and I think there's the question of people beyond OpenAI and Anthropic. That's one dimension, and then the other dimension is open source, closed source. And it's worth pointing out: many of the open-source vendors are morphing more towards a closed-source model. I mean, obviously, Facebook—Meta—did that to some extent. I think some of the Chinese vendors are starting to do that.

As I said, it's extraordinarily expensive and extraordinarily difficult to do, and if you can't command a return at the end of it, what's it all for? It can be done, but it's a brutal struggle. I don't know if the critical mass of knowledge for a compute task like that exists in Europe in a way that, obviously, aerospace was a core competency in Europe.

I agree. I think you'll have a whole bunch of people flailing around and trying. It'll make NVIDIA ever so happy because everyone will feel the need to have their own chips. But fast-forward 5 years, unless the US is ludicrously obnoxious in its sovereignty acts, it should be a small oligopoly, like cloud, of 2 to 3 players in the US, with some kind of affiliations in Europe. That is the likely structure of the industry, just given the costs required to play.

I have a direct question, and the latest Benchmark fund is going to be, I think, one of the best Benchmark funds in history. Truly astonishing. But I had a GP from USV on the show earlier, and I said, “How do you evaluate not being in one of the model providers as USV?” Can I ask you the same? Benchmark is one of the best firms. Do you sit and think, “It's not our game; they're too large”? How do you guys sit around the table and reflect on not being early in the model providers?

Guest

No, it sucks. It's terrible.

It's a complete and utter failure on our part, and I think we'd all say that. We had dinner with some of the leadership of another—one of the other best funds in the Valley—that also have a large position now, but they weren't early in the model providers either. You can't be in a situation where you have a chance to make a 30x on invested capital, as you know. If you want to claim that you're one of the best firms in the Valley and you have a chance to make a 30x on invested capital in 4 or 5 years and you don't do that, that's always a failure. So that's always a failure, no matter the way you cut it.

It sucks. It's great that we did a bunch of other amazing investments. These were all obviously before my time, but the Sierras, Firework, Lagoras, Recores, Langchains, and Hagens of the world in that fund, and many others as well. The funds obviously look awesome, but no, it stings, especially when all your friends are sending you their implied look-through ownership of Anthropic, OpenAI, and SpaceX. Those are eye-watering numbers, and it definitely stinks.

Harry, my friend, I've told you before: attribution is everything in venture. Those great deals were all you, okay? The amount of Accel GPs I've met that did the Facebook deal—we could have a series of podcast episodes.

We're moving, speaking of big wins, to a fantastic outcome and a team that I love, I know you love, Rory. Salesforce acquires Fin, formerly known as Intercom, for $3.6 billion. I just think that Owen and Des, the 2 that I know from the founding team, are incredible. They've been pounding the pavement at SaaS firms around the world for years. They've built an amazing company. Rory, you're an investor in the company. I'd love your thoughts.

Rory O'Driscoll

Yeah, exactly what you said. I think it's a smart deal—a smart deal for Salesforce. These guys became the canonical example of an old-school SaaS company that made the transition and pulled it off, right? There's not a lot of that going on. Whatever's in the water, that's what Salesforce needs to do, right? So I think it's just smart on that basis alone.

You have to transition your products, and you have to transition your model. Moving from seat-based to outcomes-based, which to me was putting your ass on the line in terms of your performance, is what made it real. They were starting to get—you know, you get paid per intervention: 99 cents per intervention. So you could literally measure value and deliver value. And it forced the company to be very close to the customer.

I think if you step back, the whole history of software has been ever-increasing alignment with the end-customer need. If you step back 20 years, in client-server software, you literally gave the customer a license key and they gave you $1 million, and you didn't even care if they deployed it. Then SaaS came along and you gave the customer $100 a month per seat, and then you had to care if they deployed it, because if they didn't deploy it, they didn't pay you. So it pushed you closer to the customer.

10. Adobe Beats, Raises, and Still Crashes as AI Fears Intensify

The next generation, which is where AI has taken us and where a lot of these SaaS companies haven't gone, is: don't just deploy the damn seats. I want the business outcome of resolutions of customer requests, and I will pay you per request, per resolution, right? And it forces a whole load of things. It's obvious: whatever it is, a buck a resolution. But to make that happen, you've got to be in the resolution path. In other words, what percentage of the time is the software set up to solve the answer versus maybe the customer decides, “I'm not going to have the software do that”? You've got to be there as the vendor, pushing that: “Hey, I can solve your problems.” And then you've got to get the answer right. If you do those things, you get money.

I think what Intercom did was align around how value was created. You saw it in Owen's numbers that he blogged about, just how it is really increasing GAAP revenue. We all know, if you run the math—we'll talk about Wix later—it's not easy, but it's plausible to 2x, 3x, and 5x as a startup. But to take something going at $300 million and growing at 7% and increment it up to something at $400 million growing at 25% is literally like pushing a rock uphill.

They pushed that rock uphill for 2 or 3 years and made it happen. I think it was a smart deal for Salesforce to buy it, because that's what they need to do. They need to have Agentforce be just as meaningful if they're going to reignite growth. So, yeah, good deal. Couldn't be happier for those guys. Great win for our firm. I was actually in Ireland when it happened.

We're a small country. You don't have a whole ton of $3.6 billion outcomes. Good for them.

Yeah, congrats to you, Rory. Huge one. I think Owen and the team have laid out the golden path for pre-AI SaaS companies in terms of what they need to do—burning the boats to get to a great outcome. There's a partner here at Benchmark that says any liquidity for pre-AI SaaS companies is top-decile performance. Any liquidity at all.

What Owen did and what the whole team at Intercom, or Fin, did is take a situation where the equity of the company was essentially worthless, because no one's going to buy $300 million growing 7%. It's just not a viable SaaS asset with no AI story. That's just a zombie company. They transformed it into hard $3.6 billion in cash or Salesforce stock. I don't know if it's cash or stock.

If any boardroom for a pre-AI SaaS company isn't going into its next board meeting and having this as topic number 1—“Is this relevant for us? How can we follow a similar path?”—it's a failure of the board.

Rory O'Driscoll

Totally agree. And I think they'd be the first to say it. By the way, it's hard, because they're competing against one of your very best companies, Sierra. They're competing against startups with a very low cost of capital and a great new architecture, right?

While it's a success, they'd be the first to say, “Congratulations. You've made yourself not a dead SaaS company anymore, and you've earned the right to punch against more impressive people who are in this AI-first marketplace.” So, yeah, that's what it takes to win. And it's brutal.

I like that top-decile comment. You're exactly right. It's very true. It's like being able to do that, and it's brutal.

If I am a pre-AI SaaS company founder listening, is there anything other than “Go in, burn the boats” that I should take from this?

Rory O'Driscoll

Be realistic, first of all, about what you can do. I really liked what you said, Harry: customer support was a space where you really could obviously add AI, add value with AI, and it was pretty obvious what you had to do.

There are other areas where I think sometimes they're trying to do a little artificial—I'm going to say it, almost heretical—thing here, like almost trying to do AI for the sake of it, versus thinking about what delivers value for your customer, right? It may be that in some of the financial accounting plays, it's going to be a longer journey with AI, and you've just maybe got to set your stall on that basis and maybe focus more on profitability and steady growth.

Guest

It has to be situation-dependent, right? Rather than saying, “Burn your boats.” It’s an easy answer to say, “Burn your boats and do everything,” but you should burn your boats only after you’ve checked where the boats are. The original cliché is from Alexander the Great’s invasion of Persia: Do you want to stay in Persia or fight the Persians?

That was more topical than I intended. Sorry, everybody. I’m going to talk about Alexander the Great here, people. Don’t revoke my citizenship. If you want to stay and fight, then you burn the boats. If you want to do something else, then you think differently.

So, be very realistic about what AI means for your particular asset and what AI can and can’t do. Then, once you do it, I think Alex is right: execute violently toward the new thing, because you don’t want to be the guy in the middle.

Yeah, it is true that I say that, but the most annoying board member ever is also going to go into a company where this isn’t relevant at all and say, “Why can’t you do what Intercom did? Or why can’t you do what Fin did?” That’s also completely unfair to a management team when it’s just not realistic.

There’s been a lot of VC advice around, “Oh, if you’re a pre-AI SaaS company, you just need to triple your growth rate or get margins to 30%.”

Yeah, I’ll get right on that.

Alex Taussig

That’s really hard. I would love to grow 3× faster or increase our margins by 30%, but these things are very hard.

Again, all of this comes with the caveat that it’s going to be different for every company. The amazing thing that I think Fin proved is that there is a market. It’s not like—some people said, “Oh, well, the only thing that these incumbents are going to buy are these new-age startups building AI companies if they want to buy AI companies.”

I forget when Intercom was founded. I think it was 2012, or—

2012. Yeah.

Alex Taussig

So, you’re buying a 14-year-old company, and it’s an AI play for them. It’s not that incumbents aren’t going to buy older companies to bolster their AI efforts, because it turns out that in AI, you can teach old dogs new tricks.

Nice. More money in the bank for Chamath as well. I’m glad to see that his liquidity is coming this year with Grok.

Alex Taussig

Was that a Mamoon-ism?

Yes, it was. I looked at the cap table. It got done by Bessemer. I went to Hawaii for Christmas, and I only did the deal with Bessemer. Sad face, but good for him. No, it was Mamoon at the AI, then when he was at Social, and then Bessemer did the deal.

11. Wix Slashes Guidance & Cuts 1,000 Jobs

Speaking of turnarounds, that was a turnaround that’s been very successful. Speaking of something potentially in need of a change, you mentioned Wix. Wix slashes 2026 guidance, cutting 20% of its staff—1,000 employees—and cuts its outlook by $50 million and revenue by $25 million. What the fuck do we do here? I really like the Wix team. They’re really good people. What do we do here?

Ed Sim

It’s hard. Let me just make the pro case. For some reason, this is one of the 2 or 3 corpses—and not corpses, that’s not fair—this is one of the 2 or 3 businesses we seem to love to dissect, especially when Jason’s here, because he’s really good at the Replit-versus-Lovable-versus-Wix discussion. So, I feel we give Wix more air cover than more grief than perhaps they deserve.

I think the 2 things here—I mean, there are some things that are relevant here. One is, they did do the right thing: They made an acquisition. The acquisition is growing nicely. It’s kind of that next-generation website builder. It was at $100 million, but of course Lovable and Replit are at $400–500 million.

I’ve just decided to walk away. He’s not interested in Wix. I’ve given you a fan, man. It’s a good idea because it’s pretty toasty in here.

Ed Sim

The interesting thing is that they did a buyback. I think the acquisition was a great idea. The buyback was obviously a bad idea because the stock is now, I think, well below half that price and trading at 1× revenue.

To Ed’s point, if you don’t have a compelling AI story, it’s just really hard. They’re going through the same journey as Intercom. I don’t think it’s done, and at 1× revenue, I actually made a mental note to think about what I would have to do to buy at that price.

The bear case is that the website-building market they created is going to become a subset of the Lovable web player: “Just build me a website.” They just don’t have any relevance there because they’re not far enough along. If they can create any kind of leverage, then at 1× revenue, that’s actually a very cheap stock.

Obviously, it’s hard to say things are great when you bought your stock back and now it’s at half the price. It’s been a tough period. I wouldn’t be giving up, but to Ed’s point earlier, there’s clearly a relevant AI story. Not only is there a relevant AI story, there probably isn’t a story without AI. They are definitely in the “burn the boats” camp, because they’re stuck in inertia and have to figure this out. There should be a doable answer.

Alex Taussig

I think the SaaS market in general obviously took a huge bath over the last year. Until recently, every SaaS stock was getting cut in half, cut by 60%, or cut by 40%. It seemed very indiscriminate.

The nice thing that’s happened for the SaaS market in the public markets for software over the last 3–4 months is that there’s now at least a filtering mechanism. It seems like there are clear scales. You have Palo Alto Networks, CrowdStrike, Cloudflare, Datadog, Palantir, and all these companies trading above 15× NTM revenue again. The premium stuff is still being priced at a premium, which is nice, because for a while it seemed like, “Wow, I guess this category is just dead.”

We’re actually going through this with a founder and trying to collate what public markets want to see: What are the good and bad attributes of any of these companies that impact their multiple? I’ll go through a few of them.

On the good side, all these companies that are trading well have a usage-based component that scales in relation to tokens or correlated AI. Datadog and Snowflake are good examples of this. The second is that there’s a clear AI tailwind for the use case. Cybersecurity is a big one there.

The third good one is that you can actually leverage AI to accelerate share gains. Maybe you’re not the market leader, or you’re a faster-growing, smaller company. If you’re a newer company that has better AI than an incumbent, you can actually accelerate your share gains from that.

On the bad side of the T-chart, the first one obviously is perceived business-model exposure. You’re a per-seat model in an economy going down this AI, outcome-oriented, or token-oriented business-model path.

The second is that you have an easily replicable product or easily replicable value with coding agents or something else like it. This is obviously where Wix gets really dinged. Intuit as well: I think a ton of Intuit’s profit comes from TurboTax, and everyone’s scared now that TurboTax will become extremely easily replicable in the future.

The third bad thing is that if you’re already the market leader and only have share to lose, that’s a really tough place to be. If you’re already the incumbent, there’s not really share to gain; there’s only share to lose. AI is a great way for startups to take share from you.

The fourth is if your product’s just lame. Ultimately, enterprises have an IT budget. There’s a pie that equates to 100%. Right now, on average, that pie is 10% AI spend and 90% other IT software. People probably want to make it more weighted toward AI, maybe even 50/50. So, that 40%—from 90% down to 50%—has got to come from somewhere.

If you just have a lame product that was already on the fringes anyway, you’re going to get dinged and your retention is going to go way down.

If we take Wix through this, everyone has some mix of these good and bad factors. What the market is doing is weighing whether the bad is worse than the good, or the good is better than the bad.

Right now for Wix, there’s a massive replicability problem. They were an incumbent that is perceived to have a lot of share to lose to AI players that are coming up in the market. Some of their product does have business-model exposure, and the product is, at least relative to the product experiences of Lovable and others, sort of lame.

On the good side, they did the Base44 acquisition, they have a usage-based component, and there’s an AI tailwind for what they’re doing. But the net of the scale is that people are giving way more credence to the bad than to the good.

Guest

And I think Figma also has Figma Make. They’re also in this market; they have this usage-based component now, and people aren’t giving them credit for that because they think the bad outweighs the good right now.

You’re essentially getting the core business for free there. When you look at Base44 at $150 million of ARR and Lovable being priced at $10 billion, you’re pricing Base44—

Guest

I agree, but if you step back on that, what I outlined was your pros and cons of a business, right? And you didn’t mention the stock price once. You basically said, “These are good things to have; these are bad things to have,” right?

Then what happens is the stock market is taking all that and saying, “Okay, you’ve got 6 good things and only 2 bad things; I’m going to give you 10x. And Wix, you’ve got 2 good things and 6 bad things; I’m going to give you 1x.” What it means is that at 1x versus 10x, at a shitty enough price, you can be a value investor.

We’ll talk about Adobe in a second. Maybe at 1x, Wix is now priced to the point where, logically, the expected return at that point should be equivalent to the 10x, right? What you’re saying is exactly that: if you’re on the bad side of the T-chart, the only forcing function left is price. Price has its wicked way.

I’m sorry, if you’re a founder, price has its wicked way. I’d be lining up the debt providers and the financiers to take this [expletive] private.

Guest

I told you, if I was running Wix, I’d be going private. I agree more. If I was running Wix, instead of having wasted that money—actually, maybe the buyback, because if you’re going to break your P&L for the next 5 years and they’re willing to give you the company at 1x revenue now, well, screw it.

Or Michael Dell, find your Silver Lake and take the company private.

Bill Gurley

That’s exactly it. That was the sweetheart deal of all sweetheart deals, which is why he’s top 10 on the billionaire list. Exactly. If I’m going to grind through, if the market doesn’t like me, well, I like me. All right, and I fancy my chances.

So you’re right. Actually, maybe we need to go do a PE roll-up of Wix with the boys and just call them and say, you know, because at 1x—

Don’t give Harry ideas. I see him writing down a note.

I’m excited to have Avishai on the show in the next few weeks.

Guest

Yeah, good. I mean, yeah, again—

Yes, yeah, that’s true.

Bill Gurley

Yeah, you should be. Look, genuinely, I have a ton of empathy and respect for you. You build this thing—such is capitalism—but you build this thing, you’re doing a couple of billion dollars in revenue, the architectural crank turns, and suddenly you’re, “Oh my God, I’ve got to do it again,” right? And I have a ton of respect for anyone who says, “Damn it, I’ll strap in and do it again.”

What could be worse? I could be Weebly. I mean, that’s—

Guest

Well, I think the genuine comment on that is, if you’re going to be in this situation, I actually think it’s interesting: being private, late-stage, with a lot of venture and a high valuation. I’m not commenting on Weebly in particular, but you wait—that’s even tougher, because at least these guys have capital and they’re profitable. But, yeah, it’s a hole if you’re—

Or like Squarespace. Sorry, I’m not picking on Squarespace. You haven’t got the Base44 acquisition and just have the legacy business.

Bill Gurley

Yeah, it’ll be hard. They had a takeout. They were private. I can’t remember who bought them, one of the PE guys.

Permira or someone. Yeah, yeah.

Bill Gurley

No, that’s tough. There’s no price at which you want to have 6 times revenue and debt.

[Laughter]

And that’s also why I laugh whenever you hear, “Where are the PE firms? Why aren’t they buying all these things?” I’m like, they are. They bought them all in 2021 and 2022, and now they’re dealing with companies that are probably worth 30% of what they bought them for.

Guest

Agreed. You bought—again, I’m just going to say this—just because you were willing to pay 10x in 2021 doesn’t mean you’d be wrong to pay 1.5x in 2026. But you have to have a really strong stomach to say, “I know this hurts like hell, guys, but what we really need to do is…” Well, maybe, you know, but yeah.

And just the opportunity cost. You’re going to have to grind it out.

Guest

PE guys are good at grinding. That’s what they do. Yeah.

Dude, I’m a poor guy.

Guest

They love the pain. Remember, we’ve got a value investor at Benchmark. That alone makes me happy. He knows how painful it is, which is why he switched.

Well, if you’re a value investor and you were—

Bill Gurley

No.

And you were saying about the bad, on the kind of negative side: Adobe beats and raises—

Guest

Mm-hmm.

But the stock falls 6%, and the CFO announces he’s leaving.

Bill Gurley

I think you should reverse that order.

What?

Bill Gurley

The CFO announces he’s leaving.

[Laughter]

Let me give you a clue. When you have a turnaround story and a complex story, and the first sentence is, “The CFO is leaving,” no one was on that call for the second sentence. They were pressing the sell button, right? Yeah, I mean, but yes. Keep going. Sorry, I interrupted.

No, no, no, you’re absolutely right in terms of the chronology of that. But the CFO is leaving for Marvell. They have a lot of the characteristics that you mentioned on the negative side in terms of seat base and large, dominant share of market. It could be perceived as a slightly lame product with regard to a lot of the generative AI that we’re seeing. Is it just, to your point, that it takes a lot of negatives and not a huge amount of the positives?

Guest

It’s funny. Knowing that we would talk probably about the SaaS apocalypse, I looked at Adobe and I was like, “This thing trades for 8 times LTM free cash flow.”

Yes.

Bill Gurley

I’m like, “Oh my God.”

Deep in his heart, that’s the value investor.

Bill Gurley

Yeah.

Totally. It’s stunning.

Bill Gurley

It’s absolutely stunning. At the same time, again, we don’t need to go down the list again, but they basically have every single one of the bad attributes. They’re already the 80% share winner in all of the markets they compete in. They only have share to lose. The product is getting increasingly replicable, and it’s a terrible business model—it has been amazing for so, so long and is now the wrong one for the given moment, for what the market wants.

And then just very few of the upsides: no real usage-based components to the business. It’s hard to argue that their products get any AI upside that they can capture because they don’t have the talent internally.

12. Why Hedge Funds Are Dumping SaaS for Semiconductors

I think the other big vector on this that is somewhat under-discussed is that if you talk to hedge fund managers right now—if you talk to a TMT hedge fund manager—what a lot of them will say, I think Brad from Altimeter has said this publicly on podcasts, is, “Look, I can go buy Adobe for 8 times free cash flow with all of these problems that we don’t know how they’re going to resolve, or I can go buy Nvidia for 16 times earnings.”

Nvidia is the poster child of every single tailwind we were talking about in AI, and it has 80% share itself of the most important piece of the compute tech stack. What I’ve heard from a lot of public managers is that SaaS maybe is oversold, maybe it’s too cheap, but why does it matter? It’s just too hard.

You can buy these memory stocks for very cheap multiples. You can buy—and, again, all these guys and gals in the hedge fund industry, they’re graded on how they do versus the index. If the SOX Index this year is up like 45% to 50%, whereas the SaaS index is probably down 20% to 25%, even if you’re an expert in SaaS and you got the good ones, you’ve underperformed the semis index. Their job isn’t to be smart on SaaS; their job is to make money, and they can just invest in semis right now and play that—

Yeah, let me give you that number for 5 years, because I don’t have it with me—I’m on the road today—but when I do this show, I have my ETF list. The 5-year return from the WisdomTree Cloud Computing Fund, which is the definitive ETF for SaaS, is down 30%, and the 5-year return from semis is up 2.7x. It’s just been a great trade every time.

There was 1 month, just recently, where the SaaS thing bounced off the bottom and outperformed for a month. Even over the last 2 weeks, it’s down—it’s kind of a relative underperformer. And with the trend—“the trend is your friend” is the first thing that momentum traders learn, right?—

Bill Gurley

I just pulled it up really quick, just since you mentioned it. The Bessemer-Nasdaq Emerging Cloud Index over the last 5 years is down 44%, and the iShares Semiconductor ETF is up 325%.

Yeah.

Bill Gurley

And so it’s just like, you know, go short SaaS and go long semis, and you’ve made better money than any other hedge fund manager in the world, besides Leopold.

There’s a point there as to why the trade makes sense.

Bill Gurley

It is, but I think that it has to be a different thing for each of them. You ask yourself, in each case, what breaks that trend? And it’s a different thing for each. What breaks the trend on the semis—

Yeah, the catalyst on the semiconductor capex trend is a flattening out in capex. If that happens, then all bets are off, and those things are going to go down so fast and hard. So that’s a call you can make. That’s one call you can make.

Bill Gurley

The thing that breaks the trend on the software side, I don’t think there was a single thing. As I’ve said, there’s a bunch of sorting going on, because this is the winnowing of the weak, right? It’s never pretty. The guys who get through the gates are the winners, right? And, you know, Datadog, I’m kind of on the positive side. I watched them outperform.

For something like Adobe, I actually think there has to be an institutional catalyst before there can be a pricing catalyst, right? In other words—

What is that?

Bill Gurley

I think you need someone like Owen running it. I’ll tell you, I’m now going to vent. The one thing—I don’t like to pick on amazing companies. I mean, I was around when one of my partners years ago was early at Adobe, and she recounted, to your point, that they had 90% market share and were stuck just below $1 billion for 4 years, right? Turns out when you sold everyone PDF, you get stuck, right?

My vent on that company is they have milked their users for so long that it just feels like a piece of financial engineering. Interacting with the product, the logins are always crap. I don’t even understand the licensing model. I don’t know what I can use and not use. It’s just constant.

My sense is they’ve extracted every piece of value, right? It’s like being PE-owned without ever being PE-owned. Someone’s going to have to go in there and rethink what it takes to make their users love them. Until they do that, I would have maybe at 10× or 8×—it’s cheap, and, shocker, it might go to 10×. So you might get a one-off 20% pop. But to Ed’s point, it’s just a lot easier to own Nvidia and have it go up 30%, right? Because there’s no such—

Are you calling your shot as Owen as the next Salesforce CEO? Is that where you’re going?

Bill Gurley

In my experience, having sold companies to Adobe and Salesforce and many things like that, I’ve sold companies in the past where I’ve thought, “That person could easily run the acquired company.” Most times, it doesn’t happen. It’s just too hard. You’ve been your own boss for too long.

Brett Taylor could comfortably have run Salesforce. But it turns out there’s someone running Salesforce who appears to like running Salesforce and has done it well enough to keep running Salesforce. So, no. For that reason alone, no.

I just know Benioff. There’s no fucking way he’s leaving unless he’s 80 or 90 and in a coffin. This guy’s loving life more than ever in the AI world.

Guest

True.

But, yeah, as a comment, I don’t know if you see this here, but we’re seeing it.

Guest

One of the smarter things we’ve seen some of our good companies do is make some small acquisitions and bring in founder teams. One of the best ways to effect a little bit of a cultural change can be picking up some of these founder-led, early AI companies, and you should be doing that.

I’m thinking of one of my companies—a really well-run company. It’s an AI company, but kind of pre-generative AI. They’ve done a magnificent job of hiring and doing 2 or 3 small acquisitions. You fast-forward a year, and each of those guys is running a $20 million BU, and you’re like, “Wow, that works.”

Going back to the Adobe comment, I don’t know how they do it. But they need to—it’s not more of the same. If they hire more financial engineering, you need to shoot it in the head, right? They need to hire someone who says, “I know where this thing needs to go.”

I think this is where—this will be my last comment on this—but this is where this cycle becomes really insidious to these incumbents.

Bill Gurley

Yes.

Because I think in 2021, when your currency as stock is worth so much, you can do a lot of these really ambitious product-acquisition things. So you can have the Square-Cash App deal, and it’s fine because even though the stock goes down later, it’s like, “Well, at least you paid for expensive stock with this expensive stock.”

The really tough thing is when the pair trade—where all of the AI valuations are going thermonuclear up and your valuation’s getting cut by 60%. Your currency—you can’t buy anything. You want to know what would make Adobe’s stock price go down another 30%? If they paid $15 billion—if they paid a quarter of their equity—for some AI company that public shareholders might not even want to buy.

All of the really amazing AI companies—they should have robbed the cradle 2 years ago. They should have overpaid when those companies were still seed or Series A companies and gotten an AI product suite. They failed to do any of that.

Now every single good AI company that they could acquire is too big for them to acquire, and they can’t do it. So then it’s like, well, we don’t have the talent internally. We can’t build a good AI product portfolio because we don’t have the talent. The talent is very scarce. And now we can’t acquire the talent because all those teams are too expensive.

I’m not—you can’t pay $10 million for a company that’s at $200 million in ARR because it’ll just nuke your stock price more.

Guest

Which is why, going back to first principles, the people running the company have to be good enough and close enough to the metal to know themselves what to do, at least well enough to have a product vision, hire people, and, as I said, maybe make smaller acquisitions.

If you’re trying to do it from a McKinsey management perspective, you’re doomed, because I love what you said: you’ve got no buttons left to press. You had buttons to press in 2021. Now you don’t have buttons, because the kinds of things that you can do—obviously, if Adobe bought Runway or Higgsfield, or pick your beloved thing, great—but I think you’re right. They’d have a shit fit.

The only people who own this damn stock now own it because it’s trading at 8× cash flow. And if you tell them, “I’ve taken you at 8× cash flow, and now we’re trading at 47× cash flow because we’ve just spent cash flow and bought this loss-making thing,” they’re going to have a conniption. So you just—you’re right. You can’t do that. It’s a tough place to be.

You’ve got to fix it. It’s why, again, I want the Wix guys to make it, because they strike me as knowing what they have to do. They did the small acquisition. They’ve just got to grind it through.

How the hell is Adobe not a phenomenal acquisition machine? That’s the mystery that’s amazing.

Any that I have missed? Any that you think we should discuss that we haven’t?

13. Standard Bots' $200M Raise

Guest

Well, there are lots. I’ll pick just 1 random one. I think I saw Standard Bots, which is a company that raised $20 million. I do a lot in robotics, and I just like those guys. I spent a lot of time talking with them. I had something else going on, and they wrote a really good piece that Packy McCormick published, right?

It was very clearly calling a shot against humanoids. It was basically saying that, for a lot of these use cases, the humanoid is a mistake, because the humanoid is putting a whole bunch of money into legs, which maybe you don’t need for most industrial manufacturing.

On the one hand, you have the humanoids, which are putting a lot of money into robotic foundation models. You’re putting a lot of money into a lot of enabling technology that maybe you don’t need for the task at hand. In fact, you’re over-scoping it.

On the other extreme, you have the old-school robotic-arm manufacturers who are doing all of the current robotic work. Most software vendors are using one of those old ones. They’re basically saying that somewhere in the middle, you can build this next-generation kind of robotic arm.

The analogy they use is that you have an integrated hardware-software stack like Apple. They’re going to do the same thing for a robotic arm that can be integrated, and they think they can take a lot of revenue that way.

What I liked about it is that it’s a pragmatic play. I think the case they articulated against humanoids in the short term is more correct than not: it’s overkill for many industrial practices. What you really want is a thing that can see, pick things out, flexibly be trained very quickly, and then pick things up and do discrete tasks pretty efficiently.

So, I think it’s a good play. It’s a US-based arm manufacturer, and we don’t have many of those at scale. I like that deal. I like that team, and I wish them luck.

Does Benchmark have a robotics investment?

Guest

We do. My partner Eric Vishria led the Series A of Sunday Robotics, which is, I would say, pseudo-humanoid. It doesn’t have legs. Sunday AI, I think, is the URL, but it has sort of a platform that can help it go up and down.

And then it almost looks like Ness from Super Smash Bros. It has a little hat and these long arms. So, it’s pseudo-pseudo-humanoid for the home.

But Rory is totally correct. I think the 2 opposing views on robotics writ large, at least on the vector of humanoid or non-humanoid, are pretty clear. The pro-humanoid argument is that the world is human-shaped: everything—our homes, our factories, our workplaces—is shaped for humans. We’ve built them so that humans can use them, and therefore humanoid robots are going to be able to most naturally interact with the environment because the environment is designed to be interacted with by humans.

The pro-non-humanoid argument is a lot of what Rory articulated, which is, look, robots are expensive. These parts are expensive. Actuators, which are the engines that provide torque for a robot, are expensive. For a lot of these use cases, depending on what the use case is, it might be gratuitous or vain to have expensive legs running around when you can just wheel the thing up to a warehouse where you’re doing pick and pack for logistics—where you take the packages and sort them. You might as well just have the arm that’s doing that.

I think, in general, we as a team have talked about robotics a lot. A thought experiment I love asking people is: in the year 2060, taking a really far-out view to allow the supply chain to catch up, what would the ratio be of digital agents to humans? And what would the ratio be of physical agents, like robots, to humans?

I think there’s some argument that by 2060, in some places like the US, there’s obviously no ceiling to the digital-agent ratio. Maybe it’s 10,000 to 1, or maybe even more, depending on how the agent landscape evolves. But by 2060, you might have a 1-to-1 ratio of useful robots to humans, or it could be much greater, depending on how fast we go.

Obviously, it takes longer to scale robotics AI than it does ChatGPT because a robot isn’t accessed through a website or an API. You have to buy it, put it together, and put it to use. But in terms of the things that are still in the first inning and are going to become trillion-dollar companies, we think robotics is one of the prime candidates, for sure.

It’s interesting because, look, I’m on the board of Locus Robotics. We have 15,000 robots in the field. We do kind of $180 million a year, right? But it’s stunning how long it all takes. I remind people that the total number of robots in the world today, in total, is about 3 million. And there are a billion people doing real work.

Those robots have been around for—arms have been around for 20 or 30 years. So, in 20 or 30 years, we’ve replaced less than 1% of the humans doing manual work. It’s a long journey. When you get to the front line, where these companies are actually trying to do the work, you see frankly how good humans are: how flexible they are. For anything other than very repetitive tasks, the human is better.

The funny thing is, I think, in software industries, the buyer is typically in an industry that itself has high gross margins. So, they’re a bit loosey-goosey on the ROI. You’re a high-value knowledge worker; we’re paying you $300,000. Fuck it, we can give you a $20,000 piece of software.

Let me tell you, when you’re running a warehouse with 400 employees, each of whom is getting paid minimum wage for pick and place, you know to the penny how much labor costs. If the robot doesn’t cost half that, you’re not going to switch. So, it just takes longer to adapt.

I do agree. I love the long-term trend. That’s why I have 4 or 5 different robot deals. I love the long-term trend, but I’ve been sobered. I’ve internalized that it’s a long journey. I don’t think there would be a takeoff—I could be wrong on this. That’s why I’m interested in Standard Bots. Those guys made a very compelling case that this could be the sweet spot. I’m watching it like a hawk.

Where you go from what we have now, which is steady adoption, to some kind of takeoff—I mean, Tesla’s Optimus is that bet. If Optimus takes off, for example, that could be it. But what you see when you get out on the factory floor is finicky little stuff that you wouldn’t think takes time, and you go, “Ooh, that was the issue. I didn’t have that in my investment memo.”

I have 1 robot company now where the biggest impediment to getting a very large order is something to do with the fact that when the poly bags aren’t flat, the label reader can’t read the barcode. So, the whole thing goes pear-shaped because you have to have people smooth them out. And if you need people to smooth them out, you don’t need the robot. I’m like, “Wow, I did not have that in my memo: the whole poly bag problem. Who knew?”

The poly bag problem. But it’s—so, I love the trend and I love the deals. I thought Standard Bots was super, and I hope those guys make it. I really liked it. It was a well-designed US arm.

I mean, the big manufacturers of arms today, going back to sovereignty, are German, Japanese, and Chinese. I think it’s a good thing if we can make a hardware arm in the US that’s cost-effective. It probably won’t be as cheap as the Chinese robot, but it will probably have way better software. That’s a credible bet.

Someone did a translation of Unitree. There’s a humanoid company, Unitree, that’s going public in China, and it’s doing $500 million. It’s a real company. Most of the humanoids are still being used for demonstration purposes, but there is a trend there where you go, “You’ve got to keep an eye on that.” Because that could be where you see it take off and I’m wrong. So, I’m just watching that space.

Guest

There’s a blog post I really love. It’s a little mini-essay called “Reality Has a Surprising Amount of Detail.” It’s ostensibly about making a set of stairs, but the whole point is that in the real world, in the physical world, stuff is just really complex. You think, “Oh, I’m just going to nail the stairs up there,” but then you decompose all the steps, all the complexity, and all the edge cases. Reality just has a surprising amount of detail. It’s a very complex place.

I think that’s what makes us so excited about this moment in AI for robotics, because for the first time you can actually have an edge LLM on a robot, even though the LLMs have to be much smaller than what we’d have for a frontier digital LLM because they have to be on the robot itself.

If you think about talking with—

Things.

Bill Gurley

You can do things. They can be very versatile. They can learn from very few examples, like, “Oh, I actually need to smooth out this label before it goes.” It’s easy to begin teaching robots these things—to be dynamic and versatile. Maybe not as versatile as a human for a little while, but probably approaching that asymptote, which makes these things generally useful. Historically, they’ve been very useful for an extremely brittle, narrow scope of responsibilities.

Totally. No, I would have, but we also put some money into 1X recently, and I was just wowed by the demo. It’s like, “Wow, you can—”

Again, I have bigger robotics companies in the field, and I know they’re exactly right. The good news is they can do stuff, and they can do it at speed and scale. The bad news is it’s fairly brittle. In other words, if the process drifts even slightly, there’s a little more programming.

The next generation of AI software, which is LLM-based, makes the bots way more flexible. Instead of telling them what to do at the individual level—“Move your hand here”—it’s like, “Put this thing in that thing,” and they can figure it out.

The way you see that manifest itself in a demo is that you do 3 of them in a row, and then on the next one, you push it and move it to a different place. The machine stops and thinks, like the raptors in Jurassic Park, and then it goes, “I see it.” It reaches over there and picks it up, and then you’re like, “That’s a—you know, that’s when you see the brain working, and that is the future.” Robot companies are pretty good.

Guest

I went to Sunday, and in the basement, all the robots were folding laundry on beds. There were a ton of these things, all full of laundry, and the Sunday employees were purposely messing with them. They’d be in the middle of folding jeans, and the employees would rip the jeans out, crumple them, and throw them.

The robot would sit there and be like— I almost started feeling bad for the robots. I’m like, “Let them finish the laundry.” I started to feel bad because they have these cute faces. I was like, “Gosh, let them finish folding the laundry.” But they’re very patient.

That’s the other beautiful thing about robots: they don’t get mad, they don’t skip work, they don’t do it. They just sit there and keep folding, and that’s going to be—

In 90 minutes of hanging out with us, we’ve turned him into Jason Lemkin.

Jason Lemkin

Yeah, yeah, yeah.

Yeah, yeah, yeah, because they don't complain.

Guest

They don't complain.

Jason just wants no people.

Jason Lemkin

Yeah, yeah, yeah.

We just want robots. Dude, I cannot thank you enough. It's so great to have you join us. You've been a fantastic guest—really, you've been a star.

Guest

Thank you. Good to see you, man.

You were there. I wish I was there celebrating Rory's in-person appearance, but maybe next time.

Guest

No, no, no. We're going to celebrate. We're going to watch—

We're going to celebrate, dude. Bring the tequila.

Guest

You're so wrong. I've got to go to a board meeting in the morning.

Oh, so much.

Guest

Tell me about it, but then tomorrow we'll watch England play Croatia and pray you win.

Dude, thank you so much.

Guest

Take care. That's me.

So much for that, dude.

Guest

You're a rockstar.