Anthropic 的300亿美元跃升、Mythos 末日、OpenClaw 遭断供、伊朗战争停火与以色列的影响力

How many PRs do you think are going to get pushed to the core structural internet in 100 days? What's the over-under number? Because I'll give you a number.

You're going to say zero. My answer to that is—

I'll say 10,000. But it's going to be immediately—

If it prevents your browser history from being released to everybody in the world, Chamath, that may be something that you're willing to let 100 days pass on.

I think you got Chamath's attention when you said “browser history.”

What about the dick pics?

Chamath is going to release them himself. We'll let your winners ride. We open-sourced it to the fans, and they've just gone crazy with it. Love you.

David Friedberg is out this week, but in his place is the one and only, our fifth bestie, Brad Gerstner. Brad, why don't you ever give me a little namaste in your payday anymore? You used to—

I'm going to bring back the greatest moderator, but now it's just kind of—You know what? These guys beat me up. They beat me up, and they just beat the joy out of me doing this program.

It's because you're a Rogan apologist now.

No, we'll get into it. Save it for the Rogan apologist. It's just because I said, “Hey, they've stopped maxxing and they've started doing some logical things.”

Yeah, okay. Here we go.

It's great to be here. Great to be here.

Good to have you. Good to have you here. And, of course, David Sacks is back. Everybody wants to hear from David Sacks. We missed you last week, bestie.

We didn't beat the joy out of you. We just tried to beat some of the hot air.

Any fluff that you can put on the show that just involves you talking and saying nothing is—

That's the stuff we've got to turn up.

Yeah, turn it up. Okay, yeah, we'll cut it right out. We'll cut it out and we'll just put a promo in for thesyndicate.com. Thank you.

Also with us, Chamath is here.

How's your maxxing going since last week? Did you have a maxxing-filled weekend? Did you have a good full weekend of smoking cigars on the back deck and not ruminating about all the chaos you've caused in the last 20 years?

I think I've done generally more good than not.

Oh, you have. But there have been some chaotic moments. Don't think about it.

You can't, bro. You can't have ups without downs, man. What are you there to do? Just please everybody and be a loser? Are you there to be a winner? Yes, you're in the arena.

Have you stopped going to therapy after realizing you're ruminating? What's up with this sudden interest in maxxing? Are you, like, the Clavicular of maxxing?

No, the world finally caught up with me. That's it. I've been maxxing this whole time. They just didn't have a name for it, guys.

Wow. Okay. Eli's videos are really good. I watched 2 more this week. What takes us through what's so appealing about not ruminating, smoking a cigar, and just living your life?

Because what he says actually works at every level of society and in every sort of thing that you may want to achieve. Even if you're trying to climb the rungs, you very quickly learn that the more you want something, the less you're going to get it.

I think that's his real message: let go, live life, and just try stuff—or don't try stuff. That detachment is really healthy for people. I like it. I like it a lot.

Who's the guy who says this? I actually didn't know.

Elisha Long. Well, Eli, I think, is how he goes by.

But he's fantastic. He has a YouTube channel. Marc Andreessen found him, and he's like, “This guy is the new modern-day philosopher.” He gives you a road map for how to live your life, right? A new-age sage.

What's the name of the character from Dune?

I was into girls' books. I was dating girls.

He's the Lisan al-Gaib of the modern internet. This is why we need Friedberg here, to explain these deep holes.

All right, listen. We've got a lot to get to. The basic point is: build something and don't ruminate. Ruminating is just not worth it. Everybody, go for it.

No, just do stuff. Stop blathering in your own head. Just do stuff.

Absolutely.

Speaking of doing so, Anthropic is withholding its newest model, Mythos, saying it is far too dangerous for any of us to have access to it. According to the company, the model autonomously found thousands of vulnerabilities, including bugs in every major operating system and web browser.

This study included 20-year-old exploits that had been missed by security audits for decades. For example, they found a 27-year-old vulnerability in OpenBSD used in firewalls and critical infrastructure. They found a 16-year-old bug in FFmpeg that had been missed by automated tools after 5 million scans. They found all kinds of bugs in the Linux kernel.

They released a hype video explaining why they were not going to share this model. Here's Dario. Come on the program anytime, brother.

Dario Amodei

As a side effect of being good at code, it's also good at cyber. The model that we're experimenting with is, by and large, as good as a professional human at identifying bugs. It's good for us because we can find more vulnerabilities sooner and we can fix them.

It has the ability to chain together vulnerabilities. What this means is you find 2 vulnerabilities, either of which doesn't really get you very much independently, but this model is able to create exploits out of 3, 4, sometimes 5 vulnerabilities that, in sequence, give you some kind of very sophisticated end outcome.

By the way, that set they're using there is the same room those guys play Dungeons & Dragons in every Sunday.

Brad, you're an investor in this company. Is this virtue signaling, or is it reality? Is this a good move by them—not to release this model, to be thoughtful, give it to a handful of people, and just find all the bugs it can before releasing it to the public? We've got a lot more issues to discuss.

I actually think they deserve a ton of credit here, and let me walk you through why. The company could have just released Mythos and broken a lot of core things on the internet. Oftentimes in Silicon Valley, we say, “Move fast and break things.” In this case, it would mean just releasing the model to move further ahead of your competition.

But here, the company realized it would wreak havoc. They ran their own vulnerability testing and saw that it would allow offensive hacking and enable people to expose browsers and browser history, expose credit cards, and so on, on the internet.

What I like about this is that they didn't need the government to hold their hand on this. We have plenty of government regulations. They know it's in the best long-term interest of the company and the industry.

They set up Project Glass Wing. It's an AI-driven cyber coalition involving Apple, Microsoft, Google, Amazon, JPMorgan, and 40 of the most important companies. Their goal is very simple: let's spend 100 days using advanced AI to find, fix, and harden these software vulnerabilities before hackers exploit them.

What I think this represents, Jason, is a threshold that we're crossing. Mythos and Spud, which is going to be out from OpenAI any day now and is the first Blackwell-trained model at OpenAI, represent the beginning of what I would call AGI models. These are models with massive step-function improvements in intelligence, and they're just too smart to be released immediately.

By the way, there was nothing that said every time you finish a model, you've got to immediately release it for general availability. They set up this idea of sandboxing and building defensive alliances in order to move away from that regime.

I think it shows—and Sacks and I have talked about this a lot, so I'm interested to hear what he thinks—that you can trust the industry and market forces in coordination with the government. They were talking to the government about this, but they're not relying on some top-down regulation in order to do this.

They laid out a blueprint that seems to me very pragmatic. Now that we're at this threshold, we're going to sandbox these things. I think OpenAI will end up doing the same thing. I think Google will end up doing the same thing.

It's an aggressive way to keep the R&D pressure on and win the race in AI while making the trade-offs to protect safety. You're always going to have to make these trade-offs. In this case, it was a great move by Dario and his team, and I think they deserve a lot of credit.

Sacks, when you look at this, we had Emil Michael on the program a couple of weeks ago—it might have been 4 or 5 weeks ago—and we had a very thoughtful discussion about, if the government is going to have these tools and Anthropic wants to withhold them, what is the proper relationship there?

You have to think that the government—and I know you don't speak for all parts of the government—if you were just going to run through the game theory, they must have gone to the government and said, “Listen, this thing is so powerful. It can put together 2 or 3 hacks and create a novel attack vector, and this is incredibly dangerous. What if China has it?”

If this thing is as powerful as Dario says it is, then this is an offensive weapon for us as well, to take out—let's just pick a pressing issue—the North Korean ballistic missile program. This is equivalent, the way it's being described, to the Manhattan Project, perhaps.

So, what are the chances that China already has this and is using it? Two-part question for you, Sacks: Do you think Dario is doing the right thing by regulating themselves?

I think Anthropic has proven that it’s very good at 2 things. One is product releases. The second is scaring people. We’ve seen a pattern in their previous releases: at the same time they roll out a new model or new model card, something like that, they also roll out some study showing really the worst possible implication of where the technology could lead.

We saw this last year. They rolled out this blackmail study where supposedly the new model could blackmail users. There’s been a whole bunch of these things. Actually, I went back to Grok and I just asked, “Hey, give me examples where Anthropic has basically used scare tactics.” It’s a pattern.

Okay.

These guys—I’m not saying it’s not sincere—but they have a proven pattern of using fear as a way to market their new products. If you think back to, again, my favorite example is this blackmail study where they prompted the model over 200 times to get the result they wanted. That result was clearly reverse-engineered, and it got them the headlines they wanted. I would say the proof that it’s reverse-engineered is we’re now a year later: There are a bunch of open-source models out there that have the same level of capability that the Anthropic model had.

Have you seen any examples of blackmail in the wild? I don’t think so. In other words, if that study were true in the sense of being a likely outcome of that model, I think you’d see examples in the wild of that behavior. We haven’t seen any of that in the past year. Now, let’s talk about this specific example with cyber hacking.

I actually think that this one is more on the legitimate side. The reason why I bring this up is, anytime Anthropic is scaring people, you have to ask: Is this a tactic? Is this part of their Chicken Little routine? Or is it real? Are they crying wolf or not? I actually would give them credit in this case and say this is more on the real side. It just makes sense, right?

As the coding models become more and more capable, they’re more capable of finding bugs. That means they’re more capable of finding vulnerabilities. Like one of their engineers said, that means they’re more capable of stringing together multiple vulnerabilities and creating an exploit. And so I do think that over, say, the next 6 months, we’re going to have this one-time period of catching up where AI-driven cyber is going to be able to detect a whole range of bugs that maybe have been dormant over the past 20 years across a wide range of systems.

I do think that there is real risk here. And I do think, therefore, that having this pre-release period makes a lot of sense, where they’re giving the capability to all these software companies that have existing code bases to use the tool to detect the vulnerabilities for themselves so they can patch them before these capabilities are widely available.

And by the way, it won’t just be Anthropic that makes these capabilities available. We know that, let’s say, the Chinese open-source models like Kimi K2 are about 6 months behind. So we have a window here of maybe 6 months where we’re still in this pre-release period, where I think companies that have large code bases can get advanced access to this model. And I guess OpenAI is going to release a similar thing in the next few weeks.

I do think that every company, IT department, or CISO that is managing code bases should take this seriously and use the next few months to detect any dormant bugs or vulnerabilities and roll out patches. If everybody does their job and reacts the right way, then I do not think it will be the doomsday scenario that Anthropic is sort of portraying. But it’s one of these things where the fear might end up being a good thing in order to drive people to drive the correct behavior.

Sure.

I ultimately think this is going to work out fine, but you do need everyone to kind of pay attention and use the capabilities.

Fix the bugs. Then we’re going to get into a big arms race between AI being used for cyber offense and AI being used for cyber defense, but it’ll be a more normal sort of period.

Chamath, we have Dario and a number of the participants here taking this super seriously. They’re making a big statement. Sacks’s very nuanced take there—what’s your take on how these companies have it both ways? “Hey, this shouldn’t be regulated. This should be regulated.” If this is, in fact, cataclysmic—oh my God, they’re going to hack everything—what if the Chinese have this right now? That would speak to more government either coordination, regulation, or some kind of relationship between the CIA, the FBI for domestic stuff, and these companies, because there is a nonzero chance that the Chinese have an equal capability here. We’re assuming they’re behind, but who knows what they’re doing behind closed doors.

So, what’s your take on this? Is it The Boy Who Cried Wolf, or is this the real deal?

I think it’s mostly theater.

Okay.

In February 2019, when Dario was still at OpenAI, they did the same thing with GPT-2. That was a 1.5-billion-parameter model, which sounds like a total fart in the wind in 2026. But at that time, this 1.5-billion-parameter model was supposed to be the end of days. It was supposed to unleash this torrent of spam and misinformation. That was the big bugaboo at the time.

So what happened? They went through this methodical rollout over 6 or 9 months. They started releasing the smaller-parameter models, and then they scaled up to the big 1.5-billion-parameter model. At the end of it, it was a huge nothing burger.

If you actually think that Mythos is capable of doing what it says it can do, 2 things are true. One is a very sophisticated hacker can probably do those things right now with Opus. And 2, if these exploits are this easy to find, whether you use Opus or whether you use Mythos, the reality is you’d have to shut down the internet for about 5 years to patch them all.

So when you see, like, a large multi-trillion-dollar gang, it’s a bit of theater. Why? What do you think they can actually accomplish in 2 months? Do you actually think that if there are these vulnerabilities, it’s all going to get fixed? Let’s give them 6 months. Let’s give them 9 months. But the reality is that capitalism moves forward, the funding needs move forward, and the need for these guys to build adoption moves forward. And that’s going to supersede what this is.

So I do think that Sacks is right that they have figured out a very clever go-to-market muscle here and a go-to-market motion that activates hyperattention and hyperusage. I give them tremendous credit, and I’ll maintain what I’ve maintained before: Anthropic is shooting the lights out right now. This is like Steph Curry going bananas from everywhere on the court. These guys are hucking 3s.

It’s all in that. Okay.

So, huge kudos to Anthropic, but we’ve seen it before. We saw it when these folks were the principal architects at OpenAI, who are now seeing the same playbook here. I think we’ll look back, and I think what we’ll say are these 2 things.

One is, if we’re really going to patch all these security holes, we need to shut down the internet for some number of years—honestly, literally years. And the second is, an advanced hacker can probably do this today with Opus if they really wanted to.

Okay. Hey, Brad, I gotta—I’ll get you in here for the last word. I’m going to go with: Yeah, maybe they did cry wolf before, but based on what I see with these models advancing—and using them, and I’m using a lot of the open-source ones right now from China—I think that this is a code-red kind of moment. This is DEFCON. We should be taking this deadly seriously, and I think these companies have got to coordinate with the CIA. This is equally a defensive and offensive opportunity. Do you think this—

You’re asking for the nationalization of AI now?

No, I actually—I don’t think it should be nationalized. Although I did see people sort of insinuating that. I think these companies need to build a group, Brad, that works and coordinates with the CIA. I assume that they’re already doing this. I’m assuming you, Emil Michael, and Trump and everybody have these people in a room, and that they’ve given the DEFCON and said, “Hey, how can our government use this to stop bad actors?”

This is already being coordinated with the CIA and the FBI. I am 100% certain of that—that Dario went to them and said, “Look what we found. This is the real deal.” I’ll give you the last word on this, Brad, since you’re an investor in both companies and you know them quite well.

The Frontier Model Forum, which was put together in 2023, is cooperating on anti- and adversarial distillation stuff as we speak, right? They don’t want to make it easy on Google, OpenAI, and Anthropic. They’re coordinating on this stuff.

There are times when I’ve pushed back on Anthropic because I thought it was perhaps regulatory capture or something else. This is very different in my mind, right? Dario could have easily come out and said, “Oh, my God, we passed a threshold. We need to have a government moratorium.”

Remember, even our friend Elon called for a 6-month moratorium in 2023 because of civilization risk. This guy didn’t do that. Instead, he said, “Okay, what should we do? I’m going to get 40 of the leading companies together. We’re going to spend 100 days sandboxing, hardening the systems, and then we’re going to keep pushing forward.”

What do you honestly think is going to get accomplished in 100 days? How many PRs do you think are going to get pushed to the core structural internet in 100 days? What’s the over-under number? Because I’ll give you a number.

You're going to say zero. My answer to that is—

I'll say, like, 10,000, but it's going to be immediate.

But if it prevents your browser history from being released to everybody in the world, Chamath, that may be something that you're willing to let 100 days pass on.

I think you got Chamath's attention when you said “browser history.”

What about the dick pics?

As Chamath is, he's going to release them himself right now. Chamath's like, “Hey, Chinese hackers, here are my dick pics. Please put them out.”

Oh, my God. We have to be out there complimenting them when they're doing the right things, or relying on the market rather than running to the nanny state and saying, “Do more of this.” So this, to me, was just an example of a good balance. I'm sure we're going to have plenty of debates about this in the future, but this is one I would like to see more of.

This is why, to use your word, JCal, I tried to have a more nuanced take, because we have no choice but to take this seriously. Whether it's total theater, whether it's fear-mongering—and they do have a pattern around this—we can't take the risk, right?

It does logically make sense that, as these models become more and more capable at coding, they're going to get better at cyber. And there's going to be that one time period where you're moving from pre-AI to post-AI, and you need a patch for that. So my guess is we're going to see a lot of patches over the next few months. I think that will resolve the problem.

I think this is a case where I'm going to give them the benefit of the doubt. I've criticized them in the past. I think that blackmail study was embarrassing to the level of being a hoax, but I think in this case, I'm going to give them credit and say that I think it's legitimate.

So it's not the Anthropic hoax. This could be legitimate. Looking at—

We have no choice but to treat it that way.

Of course. Yeah. Even if two things could be true at the same time, Sacks, they could have used this tactic before. It could be performative, like the video with the dramatic music in the background. It does have a little bit of drama to it, and the way they presented it is very dramatic, but it does make logical sense that the one company that made the bet on code bigger than anybody else would be the one who would discover this quickest.

And in 100 days, that's a pretty big advantage versus the hackers. But let me think one more point there, Chamath—

The most important thing that people haven't talked about here is that the amount of code being pushed right now because of these tools is 10x to 100x in most organizations. So we need to have this type of security embedded in these new coding tools and do it in real time. That's the opportunity. There should be real-time correcting of this.

If this was real, they picked the wrong companies. Meaning, there are energy companies—folks that control nuclear reactors. There are airplane companies that are flying hundreds of thousands of people in essentially manufactured missiles with screaming jet engines going at 500 miles an hour. None of those companies were the ones that were included in this.

And so I think if you really thought that this was the end of days, at a minimum, we can agree that maybe we should have expanded the circle a touch. Well, maybe those are customers of the ones they're including here.

Anyway, this is a really important story. We'll obviously track it in the coming weeks to see what turns out to be reality. And Dario, do come on the program at some point. Hey, Brad, will you get Dario to come on the program? I've invited him like 3 times. I got his phone number. He's ghosted me. I don't know why.

Wait, he's ignored you? I get—

I literally got an introduction from one of the number-one venture capitalists in the world. He's on the cap table very early. He just won't respond. I don't know why.

I would tell you Dario's podcast with Dwarkesh Patel, who I think is an excellent podcaster, is exceptional. I've listened to that 3 or 4 times and taken notes every time. It is a really exceptional piece of work by them.

All right, let's keep moving. We've got a lot on the docket.

You may once again be tarred with your affiliation with us.

Poor you. I mean, I don't care. Literally, I've got friends on both sides of the aisle. I have friends—

Of course you do.

Even JCal.

Even JCal has friends everywhere.

Let me ask Brad a question here, just while we're on the topic of Anthropic. There was a really interesting story—or tweet, I guess you could say—by the founder of OpenClaw that—

Peter.

Peter. Yeah. What's his name? Peter Steinberger.

Steinberger.

Steinberger. Yeah. Renowned coder, created OpenClaw, which is kind of the thing that launched this whole agent era, you could say. In any event, he said that Anthropic was cutting off his access to—was it Claude? Is that the next topic?

This is on the docket. It's a little bit nuanced. Everybody using OpenClaw would take their $200-a-month subscription to Anthropic, which was essentially like a flat rate. People were using more tokens, and, on average, the people from OpenClaw—it is very verbose—were using 100x the usage of the average subscriber.

So he said you can't use your $200 subscription. You have to use the API. You move from the $200 plan to the API and add a zero to your token use, or more. So they essentially ankled OpenClaw, and then 10 days later, or less, they released or announced their new agent technology, which, according to them, is a safer, better version of OpenClaw.

So, hey, all's fair in love and war, and they have basically shot a huge cannon across the bow of OpenClaw.

Wait, can you just explain that exactly? I think you're right that they systematically copied feature by feature—

—of OpenClaw, incorporated that into Claude, and then the coup de grâce was basically cutting off OpenClaw's oxygen.

Can you just explain exactly what they did?

Okay, very simply, when you buy a subscription to these services, they have blended your usage across many users. Nine out of 10 users use less than the tokens they're paying for, and the top 10% use much more.

When OpenClaw became a phenomenon—the number-one open-source project in history on GitHub, with all of this usage—people went crazy. You heard me talking about how crazy I went for it. Those people with the $200 subscriptions were using $2,000 to $20,000 worth of tokens.

So they said you can no longer use your subscription—either your professional or enterprise subscription at $200—and plug that into your OpenClaw. You now have to go to the API and pay per usage.

So no more unlimited. If you use Anthropic's own agent harness, are you part of the bundled flat rate?

You can assume that that's what they'll do, which, if you were thinking on an antitrust level, might be token dumping or price dumping. I'm not saying I'm ratting them out.

No, it's like bundling, isn't it?

Well, price dumping or bundling. When you price something under the market price in antitrust, that would be price dumping, right? And if you were to bundle, it would be like the bundling issue.

Critically important: You can use OpenClaw via the Claude API, and every company has a right to set the price for its products. It's just saying that, under their current regime, they were selling dollars for 10 cents via OpenClaw because these were such power users, and now they're just saying, “We have to price this rationally, but we're happy to have you guys use the API.”

Okay. Okay. But Brad, when you use the OpenClaw competitor that Anthropic now offers—

Correct.

Are they subsidizing that? Are you paying?

We don't know yet because it's in closed beta. In other words, if they charge for API usage, their own first-party agent harness or system, then that would be apples to apples.

But if they end up charging the bundled flat rate, let's say, for their stuff, but then charge the metered rate for third-party stuff, you could make a bundling argument.

Sure. Sure. And you could say it's anticompetitive, assuming that Anthropic has dominant market share in coding, which I think most people would say they do at this point.

And assuming that it's the same product. I mean, the reason most enterprises will probably use the Anthropic version of this agentic product is because it meets all of your security parameters, right?

Altimeter runs a lot of stuff on Anthropic. They're already integrated within our data warehouse, our data lake, things of that nature. So just letting OpenClaw loose on the Altimeter data set would not be wise, and so it's a different fundamental product.

No, I get that, and I think that Anthropic has a huge advantage in, let's say, cloning OpenClaw and just building it into Claude. I'm not denying that. To me, that would be the reason why they don't need to do price discrimination, because there's already a very good reason to use the—let's call it—the bundled offering on a feature basis.

But the question I'm specifically asking is whether they're giving themselves a price advantage because—

I think Brad is giving the most generous interpretation. You're taking a more cynical one. I'm with you, Sacks. I'm 100% on the cynical side.

OpenClaw is so powerful and has so much momentum that not only is Anthropic trying to ankle it, I believe when Sam Altman bought it—and he didn't buy OpenClaw itself; he hired Peter—I believe it was to subvert the open-source project, to get Peter's next set of genius ideas inside of OpenAI as opposed to letting them go there.

People are going to say I'm a conspiracy theorist, but this is the number one focus. Let me just give you a list of who is trying to kill OpenClaw or compete with it.

Obviously, you have Anthropic, but Perplexity Computer also launched. It's awesome; I've been using it. Anthropic has Claude Managed Agents. They dropped that on Wednesday, April 8. Yesterday—today is Thursday when we tape; you guys listen on Fridays. Then you have Hermes Agent, which was released on February 25. That's also open source and very good, so that's in the open-source camp.

Alibaba is coming out with one that's going to be based on its Qwen model. Then you have Elon, who said he's got something called Grok Computer coming out of Macrohard, which is a play on words for Microsoft. In addition to that, Amazon and Apple are preparing new releases of their next-generation assistants, Alexa and Siri, that will be more agentic in this new version. Then nothing out of xAI and Microsoft yet.

So, the number one goal, I believe, in the large-language-model and frontier-model space is to kill this open-source product.

No, come on. They're building multifunctioning agents that can move from answering questions to actually doing something for you. You have to do that because that's what consumers and enterprises want. It doesn't mean that it's about killing OpenClaw. This is just an obvious thing to do.

But this is a giant movement to stop it, because this is the equivalent of having an open-source Android-like player in the market, and that could be incredibly disruptive.

I believe open source is going to win the day on the large language models and take 90% of the token usage. I think the entire frontier-model space could be undercut by open source, and I think they realize that SLMs—the smaller language models that are verticalized, that will run on desktops and laptops, and are even starting to run on top-end ones—are their biggest competitive threat. I hope it happens.

All due respect to your investments, Brad, I think this technology and the interface are important. You placed your bets, but I think it's imperative that, at the agent level, which is essentially your entire life, you don't give that to Anthropic. You don't give that to OpenAI. That's your entire business and your entire life.

It is foolish for you, Brad, to give your entire business and all the knowledge you've earned over a lifetime to Anthropic through that, unless you're just doing it to boost your investment in those companies. But I would be very concerned, if I were you, about putting all of your knowledge that you've earned over a lifetime into any of these large language models.

All right, Jason, let me ask you a question. Thank you for that impassioned monologue. Actually, I want to ask my TED Talk.

Yes, thank you for that TED Talk. I have a yes-or-no question for each of you. Do you believe that Anthropic has dominant market share in coding right now? Yes or no?

No.

In coding?

Yes. They had the lead, but they weren't dominating.

I think it's a trillion-dollar market, and these guys have less than 10% of it today. So it's hard to make a case that—

What percentage of coding tokens do you think Anthropic is providing to the market right now?

Greater than 50%.

Yeah, that's true.

Okay, that's called dominant market share.

I don't know about that.

More than 50% of the market.

You have to look at what the TAM is.

Right. There are a lot of people providing coding tools in this market. Let's table that before we move on to the next.

I'm not saying it's a permanent condition, but if you're telling me that today Anthropic is delivering over half of the coding tokens, that's clearly a dominant position in the market for coding. It's an early market; it could change.

If I were representing them, David, I would say, nine months ago, everybody called us out of the game. We were being destroyed by OpenAI. Three months later, now people are saying we have a dominant market position. This is the fastest-changing, most competitive market in the world. I think it would be very hard-pressed to walk into some district court and make the case that these guys have somehow already formed a monopoly against Amazon, Google, Microsoft, OpenAI, and so on.

Well, I'm not saying it's already a permanent monopoly. I'm just asking about market share. And I do think you guys all agree that—

Chamath, go ahead.

They probably have 50% to 60% market share, because I think Codex is also quite broadly used.

But that belies the more important point, which is that AI-enabled coding is still only 5% of the broad market. So it's kind of a nothing burger. Yes, they're leading, but they're leading in something that isn't that big yet.

Now, you would say, how could it not be big? What I would say is that most of the stuff being written is still greenfield, de novo code. The ugly truth is that I don't care what model you have, but the long-horizon ability for any of these models to actually build enterprise-grade software is still shit. That's the actual lived experience.

Not for me, but when I call on our customers—half-trillion-dollar banks and $100 billion insurance companies—none of these guys are saying, "Wow, it just works out of the box." It doesn't work. Most of it is still hand-tuned.

Until I can honestly tell you that we can point a model at this with the right guardrails, which I can't today, what I would say is that it's a small market that will become large as these models become better. But we are in a world where we have 50 years of accumulated tech debt. I suspect that when you enumerate the number of lines that represents, it's hundreds of trillions of lines of marginal, mediocre-to-bad code.

On top of that, we have all these legacy languages. I'll tell you about one of our customers: They have to go and get 60-year-old pensioners to come into the office to interpret COBOL. No, I'm not joking.

Snowball effect.

This is a $100 billion-a-year revenue company, and that's how they solve these problems. It's not that Opus just solves it.

I would just keep in mind that 99% of the tech debt in the world is still poorly addressed by these models. We are untying this Gordian knot. It's going to take decades to do it right.

All the breathlessness about this other stuff, I really think it's not where the money is. It's not the big-time stuff. You can tell me, "Oh, yeah, it's going to be the future," and I would say, "Tell this business that's $100 billion a year in revenue and has 50 million billing relationships that all of a sudden you're going to OpenClaw your way to a solution."

That's not to say that you can't have a great chief of staff, or that you can't do some useful stuff and trickery, or have a good knowledge base. I'd like that, too. But the core things that your lived experience sits on today are a mess of tech debt that will get very slowly replaced. That's just the reality of life.

And there are competitors that are extremely disruptive. I'll tell you about one. We talked about Bittensor TAO on this program a couple of weeks ago when we had the Jensen interview. You brought it up, actually, Chamath.

There's a project on Subnet 62 called Ridges AI. What they're doing is a competitor that's not only open source, but anybody can contribute to it. They spent about $1 million in TAO rewards, and in 45 days they hit 80% of what Claude 4 does. They did that in under 45 days.

The way it works is that they give rewards to people who—and they can do this anonymously—make that coding product, which is like Codex or Claude Code, better. That flywheel is racing right now, with participation, in the same way Bitcoin is.

You're going to see a lot of open-source and crypto-open-source combinations. Anybody who hasn't investigated this, I highly recommend that you do.

I do think you're right about one specific thing. I would put the probability at zero—literally zero—of any important company worth more than a dollar having and outsourcing its production code to an open-source project. That'll never happen.

However, when you look at the cost of training this 10-trillion-parameter model on Blackwell, and when you look at the future—let's just say in 6 or 9 months—that a 15- or 20-trillion-parameter model is going to get trained on Vera Rubin, I think, Jason, where you are right—and, just to be super clear, I have no investments in this at all—I'm just observing.

Another project, other than Bittensor, that someone brought up to me is Venice. The concept of open-source training and orchestration is a hugely disruptive idea, which is a completely orthogonal attack vector to the idea that you have to raise tens and tens of billions of dollars to train your models.

If the capital markets run out of $10 billion and $20 billion checks to give people, the only solution is to be totally distributed. I tend to agree with you, Jason, that there is going to be, at some point, a very successful open-source project for pre-training. Absolutely.

Will there ever be an open-source way where a real company that has any skin in the game says, "Here, guys, re-engineer my codebase as an open-source project"? Never going to happen.

Yeah, I think the coding tools will. If you look at the history of open source, Brad, you actually had a lot of bets in this space: Linux, Kubernetes, Apache, PostgreSQL, Terraform. These open-source projects are deep inside enterprises.

Deep. We’re sitting here 15 or 20 years ago, and the same argument was made: nobody will ever adopt these inside the enterprise. You’ve got to go with Oracle, whatever. Fair enough, many people do.

But I think this $29-ish subscription to do this versus $200 is starting to take hold inside of startups. That’s where I always look at the tip of the spear. Startups love to use open-source products. I think this could be the next big thing.

But listen, I invest in things that have a 90% chance of going to zero, so do your own research. No crying in the casino.

Can I just make a final few points? So, just quickly, number 1 is with respect to this market for code, or code tokens, whatever you want to call it: it might be 5% today, meaning 5% of the code is AI-generated versus human-generated. I think it’s going to 95%. I’ll bet any amount of money on that. The only question is when, probably over the next few years.

Point number 2 is that it’s possible that if you’re the early leader in coding as an AI model company—let’s say you have 50% to 60% of market share and the most developers using it—you therefore have the most access to codebases. You might get the most training tokens. There is a potential flywheel there, where you can see the early market leader consolidating its lead because it’s generating the most code tokens and getting access to the most existing code.

I’m not saying for sure that’s going to happen. It’s possible that the other guys catch up, but I think there is a possibility of a flywheel there and strong, I guess you’d call it, data-scale effects. So, I do believe that the market for coding tokens could be monopolized.

Point number 3: Anthropic’s revenue run rate, based on what I can tell and what’s been publicly released, is the fastest-growing revenue run rate at scale that I think we’ve ever seen.

Perfect segue. It’s the next story. Okay, maybe pull up the tweets. But this thing is ramping at a rate we’ve never seen before.

We can get into that in a second, but just one last final point is that I think it’s pretty clear that where we go from here is agents. Coding gives you a huge step up on agents because one of the main things that agents need to do is write code to enable them to complete tasks.

Correct. And if it is the case that coding is this huge market that’s going to be dominated by 1 or 2 companies, and that then leads to another huge market, which is agents, my point is just that I think all these companies need to behave in a very clean way and not engage in tactics that later the government might say, “You know what? That was anticompetitive.”

Everyone should just play fair. Do not engage in discrimination against other people’s products. Engage in fair pricing. I’m not accusing anyone of breaking any of the rules, but what I’m saying is that eventually the government is going to look at this market with the benefit of 20/20 hindsight, and I think everyone should just basically keep their nose clean.

Keep it tight. Tight is right. I think that’s an excellent point. Let’s talk about the revenue ramp of Anthropic. This is just unprecedented.

Anthropic’s revenue run rate has topped $30 billion, with a B. In early 2023, they turned on revenue. They started charging for API access. At the end of 2024, they were at a $1 billion run rate. On February 25, they launched Claude Code. That was the starter pistol. By mid-2025, they were at a $4 billion run rate. At the end of 2025, a $9 billion run rate. Just a couple of months later, in April, a $30 billion run rate. Yes, that’s right: triple.

The way they did this is that enterprise customers are a major part of the spend. Dario announced a couple of months ago that there are over 1,000 enterprises paying over $1 million annually. This is truly mind-boggling when you think about it, because those are the most coveted customers in the world. These are the big fish that, when people are running enterprise software, they dream of getting. Slack dreamed of getting these million-dollar customers. Salesforce dreams of getting these million-dollar customers.

Brad, you’re an investor. I guess Sam famously, on BG2, asked you to sell your OpenAI stock back to him. You didn’t. You demurred, but you’re an investor in both. How shocking is it to you to place both of those bets and then see one of them come from so far behind?

ChatGPT has 900 million users. I don’t know if they’ve officially passed a billion yet, but they are the verb, right? They’re the Uber, the Xerox, the Polaroid of AI. But they didn’t go after the enterprise. Dario made that, and Dario worked. He was the co-founder of OpenAI. He left, and according to the New Yorker story that came out from Ronan Farrow this week, he was basically left because of his disgust with working with Sam Altman. Your thoughts?

Well, before we go down the OpenAI rabbit hole, let’s really contextualize what’s going on here. Jason, I have this additional chart. You showed one: they added $4 billion of revenue in January, $7 billion in February, and $10 or $11 billion of annualized run rate in March. Just to put that in perspective, that’s Databricks plus Palantir combined that they added in a single month.

We started with everybody at the start of the year wringing their hands, including Gurley and others, saying we’re in a big bubble and asking whether the AI revenues would show up to justify all of this investment. And, bam, you have the largest revenue explosion in the history of technology.

The company’s plans were to end the year at about a $30 billion exit run rate. They got there by the end of March, and I suspect that it’s continuing in April. So, you have to ask: what’s going on here, and what’s the big so-what?

The first thing for me is that model and product capability just hit this threshold we talked about earlier—near AGI, whatever the hell you want to call it—and everybody said, “Damn, this is so good. I have to have it.” This is no longer about my IT budget. This is about labor augmentation and labor replacement. By the way, Claude Cowork is growing even faster than Claude Code at the same stage of development.

What it showed is that we have a near-infinite TAM. It turns out that the TAM for intelligence is radically different from anything that we’ve seen before. And I think the best example of this is millions of self-interested parties: consumers, enterprises, and now over 1,000 paying more than $1 million.

It’s not that there was some great go-to-market at Anthropic that all of a sudden snuck up and blew everybody away. No, it was companies demanding the product. They’re getting throttled on the product. Why? Because it’s so good. It makes them better at their business. We are all self-interested actors, and when millions of those people are all making the same decision, there’s a huge tell. The tell here is that the TAM is as big as Dario and Sam and others have been saying.

We knew intelligence was going to scale exponentially. The question was whether revenue would scale exponentially, and that’s what we’re seeing. Remember, they’re doing this with only 1.5 to 2 gigawatts of compute. These guys are massively compute-constrained. They’re each going to be adding 3 gigawatts of compute this year, and that will unlock even more growth. They would be growing even faster if they weren’t constrained by compute.

And then, Jason, to your point about the open-source models—we all want to be a part of this solution—I’ve talked to a lot of big companies, and 65% to 70% of their token consumption is from open-source models. These are cheap Chinese and other models. So, these revenue ramps are happening while the world is already using open source. This is not frontier-only. This is frontier plus open source.

We’re going to see massive token optimization over the course of the year. But what happens with Jevons paradox is that the unit cost of intelligence is plummeting—not the cost of tokens. The unit cost of intelligence is plummeting because the capabilities of these models are so much better.

I look at what it does for Altimeter day in and day out. I talked to a major company yesterday. They’re on a run rate to do $100 million of token consumption this year on about $5 billion in OpEx. They think that we’re now nearing peak employment in their company, but their token consumption—or, let’s not call it token consumption, because tokens may go up a lot—their intelligence consumption is going to go up a lot.

I would leave you with this: we’re early, to Chamath’s point. We have low penetration of the Global 2000. We have low penetration of the use cases. We have low penetration within the use cases that they’re already using. And the models are only getting better.

So, when you look out toward the end of the year, I would not be shocked if you see Anthropic exiting this year at $80 billion to $100 billion in revenue. And, by the way, they’ll be doing it at the same time that OpenAI, which is also on the wave, releases an incredible model very imminently. They’re going to be on that wave, and you’re going to see an inflection in their revenues as well.

Okay, Chamath, question 1 has been answered. The question of, “Hey, does this stuff actually have utility?” went from a question mark to an exclamation point. Of course it’s got utility. People are getting value from it, and it might be variable. Some people get more value than others.

Number 2, the revenue ramp, was a big question. Now that’s turned into an exclamation point. The final piece of the puzzle that you’ve brought up many times is: can this be profitable? These companies are burning through a large amount of cash. So, what is your take on when these companies can get out of the J-curve? We talked about this, I think, 3 episodes ago.

I estimated that we're going to be looking at $4.5 trillion in investment into these data centers at a minimum, and then they have to climb out of that to get to profitability. So what are your thoughts on these becoming profitable companies? We talked about this, I think, 3 episodes ago.

Do you remember the investor who published this list, Jason, where he put all of the terms you talk about? One of the terms you can't talk about is profit. It's a list where, if you can't talk about free cash flow, you talk about EBITDA. When you can't talk about EBITDA, you talk about margin. When you can't talk about that, you talk about revenue. And then when you can't talk about revenue, you talk about gross revenue, bookings.

So you can figure out, I think, where we are in any part of any cycle by indexing into what everybody talks about. I think where we are is between gross revenue and net revenue. That's where the discussion is.

Okay.

There was another article today, I think maybe in The Information, that tried to categorize and distinguish that Anthropic presents gross revenue and OpenAI presents net revenue. They're different, but we don't know what the various take rates are, so they're saying that there's a difference. If it's not true, there's been no clarity provided by these companies.

At a minimum, you have this confusion, where there's the breathless talk, and then there are people who don't even know the difference between actual recognized revenue and run-rate revenue or how to multiply. So we're definitely there, okay? We can quibble about the details, but we are not at the place where people are like, "Oh, here's your steady-state free-cash-flow margin, and here's what your EBITDA does." We're years from that. They're going to have token-maxing EBITDA, like EBITDA at WeWork.

The thing that we need to understand is how gross-margin-negative this revenue growth is.

We don't know that, and at least we don't as outsiders.

Brad might know.

Brad may know. I will tell you, think about this: What are their big cost inputs? The number one cost input is the cost of compute.

Right? I just told you they only have a gigawatt and a half of compute, and they have that gigawatt and a half of compute whether they have $1 billion in revenue or whether they have $80 billion in revenue. So you might actually expect to see these companies' gross margins exploding higher, like the fastest increase in gross margins I've probably seen out of any technology company.

So this is not gross-margin-negative, you're saying?

No, definitely not gross-margin-negative. And what I would tell you—

So they must be hugely profitable, then.

Well, you may see accidental profitability. They may not be able to spend this revenue fast enough, Chamath, on compute. And remember, it's only 2,500 people. Google crossed this revenue threshold when it had 120,000 people. These guys have 2,500 people.

So the only thing you can really spend money on is compute, and they can't stand up the compute fast enough.

But none of this squares with me, to be honest, because if you were on a threshold of 90%-plus gross margin—

I'm not saying it's there. I'm not saying it's 90%-plus. I'm just saying it's gone from meaningfully negative 18 months ago to very, very positive. I've seen rumors out there of 50%. What you're saying is that the trend is there.

Let me just say this: I think if you're an incumbent, you want the cost of compute to go down. If you're not an incumbent—specifically, who do I mean? Meta, Google, and SpaceX. Meta and Google have fortress balance sheets. I think by the end of June, SpaceX will also have a fortress balance sheet.

What they will want to do is make this a compute problem because they will control the conditions on the field. You already see this today.

Yeah.

Meta's models today—the general review people give them is that they're okay—but the one thing people say is that they're incredibly performant. The model quality is okay, but the performance is great, which speaks to Meta's huge advantage: It has a massive compute infrastructure.

If you're not OpenAI or Anthropic, you'll want to make this a capital problem because then you can win it. If you're Anthropic and OpenAI, you want this thing to be as efficient as possible.

I think where we are is very much in the early innings, and we're bumbling around talking about gross margins and revenues. We are not at profitability. What was true for Facebook and what was true for Google was that, irrespective of when they got to $1 billion, who gives a shit? They were profitable by year 3, and they never looked back. I was there. I remember it was glorious.

The cost of building AI—I totally stipulate—is radically higher than the cost of building retrieval at Google. It's just a fundamentally more expensive problem. But I will tell you that there's a lot of FUD out there about negative gross margins.

Jason, you started the segment by saying they're burning through large amounts of cash. I think people are going to be shocked at how low the burn levels are at these companies.

Anthropic or OpenAI?

Yes, and I would say at OpenAI as well. If they do $50 billion this year, just look at the number of people they have and their revenue per employee. It's pretty low, and the inference cost is plummeting. Inference cost is down 90% year over year.

Finally, I want to respond to this point about gross versus net and this tweet that Chamath was referencing. There's a smallish percentage of Anthropic's revenue that they distribute through the hyperscalers and, like in a lot of arrangements, you pay a commission on that, whether it's Snowflake, Databricks, or others.

I will just tell you that you're talking about single-digit percentages of the total revenue of these companies. So the gross-versus-net thing isn't what's being reported. The apples-to-apples comparison is pretty easy, and if you want to be conservative, take down Anthropic's revenue by 5% to 10%. Again, I think it's better to gross up OpenAI's revenue, but either way, I just think it's a distraction from what's really going on here. Happy to—

Sacks, do you have any thoughts on this massive revenue ramp?

Yeah, I want to go back to a point that Brad made because I think it was really important, and I want to underline it. Consider where we were at the beginning of the year. What everybody was saying was that AI was a big bubble, and the evidence they would point to was the fact that hundreds of billions of dollars were going into capex that needed to be spent on these data centers, and there was no evidence of significant revenue to justify that spend. Where was the ROI?

As an aside, the same doomers who were saying that AI was in a bubble were also the ones who were saying that AI was so powerful it was going to put us all out of work and take over from humanity. In other words, they couldn't decide if AI was too powerful or not powerful enough.

Putting aside that contradiction, they were clearly making the case that AI was this big bubble and that there would be no payoff or justification for this massive capex being spent. I think we're starting to see that there is justification for it. We're seeing it in just this one vertical of AI, which is coding. We're again seeing the fastest revenue growth in history. It's utterly unprecedented.

This is just one category or vertical of AI. We know that agents are coming next, and the enterprise adoption of that is going to be absolutely massive.

So I guess what I'm saying is that this is early proof for the thing that makes Silicon Valley special, which is that we're willing to bet on things that, intuitively, on a gut level, we know are the next big thing. We're not that spreadsheet-driven. Silicon Valley believes that if you build it, they will come, and it's willing to finance that build-out. That's basically what's been happening.

Again, just the top 4 hyperscalers have $350 billion of expected capex this year, on its way, I think Jensen said, to $1 trillion by 2030. Silicon Valley—whether it's big companies or founders—is always willing to bet on this next big thing. They're not like Wall Street. They don't need specialists to tell them where to go. They know where the technology is going, and they make their bets based on that.

I think there is going to be a big payoff for this. I think the thing that's going to make our economy, and the United States in general, remain extremely dynamic and in the lead on this is that we're willing to make those kinds of bets. I think it's going to pay off big time.

Yeah, clearly. Hey, Brad, you didn't answer my question about the vibes over at OpenAI versus Claude. OpenAI is—I wouldn't say reeling—but there's a lot of hand-wringing going on: a lot of employees leaving and a lot of people wondering whether the strategy of consumer-first is the winning strategy. They shut down Sora, they're unwinding the Disney deal, and they're really trying to get the company focused.

Listen, the New Yorker story was a bit of a rehash, so I don't think we have to go into the blow-by-blow because we covered it here 3 years ago. But the truth is, a lot of the great founders and co-founders of OpenAI, and a lot of the great contributors, are now at Anthropic and other large language models. In the secondary market, OpenAI is trading lower than its last valuation.

And Anthropic is trading significantly above $380 billion. So maybe talk a little bit about this competition—this Microsoft versus Apple, this Google versus Facebook.

Let's start with immense credit where credit is due. Anthropic was literally counted out of the game last year, right? Here they come over the last 12 months, and they've kicked OpenAI's ass over the last 90 days, right?

What did Anthropic do? Anthropic made choices: no multimodal, no video, no hardware, no chips, no building data centers. They said, “We're just going to focus on coding and Cowork. We think that is the path to AGI and ASI.”

They executed their butts off. They took the lead. 2,500 people, tight, pulling on the oar in the same direction. But I think you would be seriously foolish to count out OpenAI. I think we're at peak OpenAI FUD.

I'll tell you, it starts with great researchers and great models. And I think when you see the Spud model they're about ready to release, I think it's going to be an excellent model. It shows that they're firmly on the wave.

If you look at what's going on with Codex, incredible ramp on Codex, the fastest-ramping model with 5.4. I think 5.5 or Spud, whatever we're going to call it, is going to be an even faster ramp.

Have you seen Spud? Have you used it? Have you gotten a preview?

People are using Spud. It is being previewed.

So, you're talking to people who've used it. What are they telling you?

They're telling us that it's an incredible model, on par with Mythos, and that it's a very usable model in terms of how it's packaged.

Back to David's point, this is the most important point I think anybody can take away here. This is not zero-sum. The TAM of intelligence is dramatically larger than any TAM we've ever seen in our investing careers over the last 2 decades.

And if you're on the wave, which OpenAI is, you are going to be selling into the world's biggest TAM. They are going to build a very big company. I'm a buyer of the shares today. Notwithstanding all of the vibes that you describe, I think these companies are firmly on the wave.

They are jarred. They are sitting there saying, “What did we do wrong, and how do we get our mojo back?” They want to compete. It is embarrassing to people on the research team and the product team over there. So, I'm not saying there's not a real awakening occurring there, but I think that's the case.

And by the way, to Chamath's point, do not count out Meta. I think Meta is absolutely in this game. Google is absolutely in this game. Elon is absolutely in this game.

He's got some stuff dropping shortly that's going to be very impressive.

If you're on Team America, the fact that we have 5 frontier models competing against each other—and David made sure they weren't throttled by excessive government regulation—we have Mythos come out. It's a self-imposed safe harbor to harden our system; it wasn't a call for moratoriums or getting the government involved.

We have the type of competition that's causing us to accelerate our lead against the rest of the world. We can't take our eye off the prize. We have to stop adversarial distillation, and we need to make sure that we're distributing our products around the world. But I view this as really good for Team America.

Well said. Here are your Polymarket IPO odds before 2027. Obviously, SpaceX at 95%, Cerebras at 94%, and, hey, number 5 on this list: a 51% chance that Anthropic goes out before the end of the year. There's a 44% chance that OpenAI comes out before then.

All right, here is the closing market cap for Anthropic on Polymarket. Only $158,000 in volume. So, Chamath, when you put in $400K, you're going to really tilt this market. There's a 78% chance that it's above $600 billion and a 19% chance that it doesn't go out. So, it's looking like this will be a decent investment for you. Brad, what valuation did you get into Anthropic at?

We first invested in, I believe, the $30 or $150 billion round.

So, this will be a 7x, 5x for Altimeter LPs. Congratulations.

No, listen, again, there are lots of people who were there before us, who are on the board, and who are going to do much better than that.

What'd you put in? $50?

What'd you put in?

No, we've got billions in both companies.

Billions in both companies. Oh, my lord.

I think there's this existential thing going on in venture today. David could talk about it as well. People are extraordinarily nervous. Look at the IGV stock index: down 30% year to date and down 5% today, with all software stocks plummeting. Venture capitalists are terrified to invest money in anything other than these frontier models and things like SpaceX or military modernization.

Finding something that's out of harm's way of AI, where you can count on the terminal value, based on Chamath's insights over the last few weeks, is very difficult to do. That's why you see this crowding.

So, we've taken a barbell approach. We've got a lot in what we think are the most important companies that are on the frontier, and then we're betting on really small teams that we think have very defensible businesses in a world of AGI.

What happens to all these enterprise software companies? Do they become PE takeouts? Do they get consolidated? Or do they just have to adopt these AI technologies and solve this problem of the frontier model simply solving for whatever these niche software companies do?

I think the market's probably being a little too pessimistic with respect to at least some of these software companies. Obviously, there are going to be big differences in the quality of the moats of these companies. So, look, software is going to be a lot cheaper and easier to generate, but I'm not sure that was the competitive advantage of a lot of these companies.

There's probably a little bit of the baby being thrown out with the bathwater right now, and there probably are some value buys in enterprise software. I think the interesting question here—and we've been talking about this for a couple of years on the pod—is just where you see the AI value capture being in terms of the layer of the stack.

Remember, where we started, it was really just the chip layer of the stack where all the value capture was. Nvidia was the first company to be worth multiple trillions of dollars because of AI. For a while, it looked like that's where all the value capture was going to be, because OpenAI, for example, was losing so much money and Anthropic wasn't on the radar as much.

Now we're seeing, wait a second, it's not just the chip companies; it's also the hyperscalers that are now benefiting. And now we're seeing, at the model layer, it looks like Anthropic and OpenAI are all going to be huge beneficiaries.

I think the next question is at the application layer of the stack. Does all that value capture just get eaten by the model companies, or are there applications that get turbocharged? I guess you could say that Palantir is already one of them, right? It's an application company that's been turbocharged by these model capabilities.

Who else will be a big beneficiary? Is it, again, all going to be at the model layer, or will you see an explosion of value at the application layer? I'm hoping, obviously, that it'll be at all layers of the stack.

Yeah. What happens to Salesforce, HubSpot, Oracle, right down the line? David, Chamath, your thoughts here on the layers and where the value is captured?

It's too early to tell.

Too early to tell, right? And energy, we kind of put into data centers as well, but that's obviously been a clear winner.

Okay, here we go, guys. We're going to go to the third rail here. We have to catch up on the Iran war. Here's the latest: 2 weeks into the war, a ceasefire started just 2 days ago at the taping of this episode. Vice President JD Vance, friend of the pod, along with special envoy Steve Witkoff and friend of the pod Jared Kushner, are headed to Islamabad, the capital of Pakistan, for talks this very weekend.

So while you're listening to this episode, they are going to be working on the peace deal. On Easter Sunday, Trump posted a Truth stating, “Open the Strait, you crazy bastards, or you're going to be living in hell. Just watch.” Praise be to Allah. On Tuesday morning, Trump posted another threat on social media: “A whole civilization will die tonight, never to be brought back again. I don't want that to happen, but it probably will.”

The tweets were obviously discussed a lot over the last week. He gave him an 8:00 p.m. deadline. At 6:30 p.m., POTUS announced on Truth Social that President Trump had agreed to a 2-week ceasefire if Iran opens the Strait. He also said, “Hey, listen. We got the Strait. Maybe there'll be a toll booth, but we'll take the majority of the toll and we'll split it with Iran.”

Here's the quote: “We received a 10-point proposal from Iran, and we believe it is a workable basis on which to negotiate.” Apparently, Netanyahu took the ceasefire to mean leveling Lebanon, dropping 160 bombs in 10 minutes yesterday. Sacks, you were out last week. Everybody wants to know your position on the war, so I'll hand it off to you. What are your thoughts on the ceasefire and everything that's occurred up until this point?

Well, look, I have to preface what I'm about to say: I'm not part of the foreign policy team at the White House. The last time I commented on the war on this show, it somehow made international headlines: “Trump adviser says XYZ.” I'm not a Trump adviser on this issue. I think that would be a fair headline to write if it were a technology issue, but this is not.

So whatever I say is just my personal opinion, but then the media is going to somehow portray it or attribute it—

To the White House.

—or try to create an issue out of it. So I feel like I'm limited in what I can say, except to say that I think it's terrific that we have the ceasefire. I think it's great that there's going to be this meeting in Islamabad to hammer it out.

I think what the president has accomplished so far with the ceasefire is a great thing, because what happens with these wars is they take on a life of their own. They tend to go up the escalation ladder, right? There are a lot of podcasts discussing the so-called escalation trap, and supposedly there are stages to this based on historical patterns.

I think it's actually very hard to pull out of these things, and I give the president tremendous credit for negotiating the ceasefire that we've achieved so far and then sending the team to hopefully work this out.

Brad, actually, my first trip to the Middle East was when you and I went maybe 4 years ago. Thank you for taking me. What is your take on where we're at here? I think we just wrapped up week 6 of this, and we're going into week 7.

First, on March 4th, I tweeted the Trump doctrine in Iran: massively destroy all military capabilities, kill the people building lethal weapons to use against us, and get out. Reserve the right to do it again if needed. Zero efforts to build Madisonian democracy. Iran's going to have to build what comes next.

I think what the market has said—if you look back at last year on tariffs, Jason, the top-to-bottom drawdown was about 15% on the Nasdaq. Intraday, it was down 22%. The drawdown in this period over Iran was only about 5% to 7% on the S&P and Nasdaq, right?

So the market has said, “Listen, we trust Trump at his word.” He said he's not going to get into an entangled war here. I think he terrifies the hell out of people with his tweets about destroying civilization and all this other stuff, but I think people, even though they don't like to hear it, have resolved for themselves that when he says he's going to get out, he will, in fact, get out.

Of course, there was a lot of hand-wringing, but if you look at the markets today, we basically bounced all the way back from where we were pre-Iran on both the S&P and the Nasdaq. If, in fact, we land the plane—if JD lands the plane—and, by the way, on Lebanon, yes, they were bombing yesterday, but Netanyahu has now said that there are going to be direct government talks between Israel and Lebanon.

So if we land the plane on these 2 things, I think it's off to the races in the market. By the way, while everybody's focused on Iran, stay tuned. I think we're getting close to a deal on Ukraine-Russia, right? Venezuela is seemingly going very well. I think there's also going to be news on Cuba.

You could envision a world—there's risk to the downside, certainly, I will stipulate—but you also have to pay attention to the risk to the upside. If you land the plane on those things heading into America 250, July 4th, the market could really take off.

All right. Well, let's maybe uplevel this a little bit and talk about why we're in this war to begin with. That's the big discussion amongst both sides of the aisle.

On Tuesday, The New York Times dropped an inside-the-room piece on how President Trump made the decision, according to this report, if it's true. I know some people don't subscribe to The New York Times anymore or think it's fake news, but the piece addressed how Trump decided to basically follow Netanyahu into this war.

On February 11th, Netanyahu met with Trump at the White House, where he gave him a 4-part pitch on attacking Iran. JD Vance, according to the story—if it's true, disclaimer, disclaimer—warned Trump that the war could cause regional chaos and break apart Trump's MAGA 2.0, the Trump 2.0 coalition we talked about here, the big tent. That has actually turned out to be true.

There's been a bunch of hand-wringing from Megyn Kelly, Tucker Carlson, right on down the line. Rubio was anti-regime change, but he was largely ambivalent, according to this story, about the bombing campaign. Susie Wiles, chief of staff, said she had concerns about gas prices before the midterms. Pretty good advice there.

General Dan Caine, chairman of the Joint Chiefs of Staff, said this of Netanyahu's pitch:

“Sir, this is, in my experience, standard operating procedure for the Israelis. They oversell, and their plans are not always well-developed. They know they need us, and that's why they're hard-selling.”

If you put this together with Rubio's walked-back comments at the start of the war—“We knew there was going to be an Israeli action. We knew that would precipitate an attack against American forces, and that's why we did it”—I had Josh Shapiro on the All-In interview show, and he talked a lot about this.

There's a big underpinning here, Chamath, that United States foreign policy is being driven by Netanyahu. Every Jewish American person I've talked to feels Netanyahu is not doing Jewish Americans and the Jewish diaspora any favors here by his approach to these wars. What are your thoughts on why we got into this and how we get out of it?

I mean, the person who decides is the president of the United States. Some foreign leader isn't getting to call the shots in the United States.

I think, very practically speaking, the markets are effectively pricing in that this was a small blip, for whatever people think. That's just what the best prediction market that we have is telling us. I think that's important to acknowledge: we're probably in the endgame here.

The second thing to acknowledge is, if I was Israel, I would really be concerned that unless I help find an offramp quickly, the risk that Israel loses America as a predictably steadfast ally could go down. I think that's problematic for Israel—

Far more than it's problematic for the United States.

So all of that kind of tells me that we will find an offramp, A, because I think economically it makes sense, and then B, geopolitically, I think Israel will want to make sure that this doesn't burn a long-standing relationship.

Yeah, that seems to me to be the major issue here: Americans basically do not want to be in this war. Americans do not want our foreign policy being influenced to the extent they believe. I'm not putting my belief in here. Americans believe we are being dragged into this by Israel and that Israel has too much—or Netanyahu specifically has far too much—influence.

And then people believe the antisemitism that's occurring here. Josh Shapiro gave me a lot of pushback on this. But all the Jewish Americans I talked to say Netanyahu is causing, with his actions in Gaza, Lebanon, and Iran, he's gone too far, and it's causing the antisemitism we're experiencing today. So you can make your own decisions about that. Any final thoughts here, Brad, on American foreign policy being influenced too much by Israel?

No, it's a discussion. I mean, listen, kind of like Sacks said earlier, I think that we will ultimately be judged by the outcomes, right? Everybody is an armchair pundit today on the approach that we're taking in these 2 different places.

I think we could be on the verge of a massive transformation of the Gulf states. You went there with me, Jason: Saudis, Qataris, Kuwaitis, Emiratis. I've talked to a lot of them this week. I think they're very hopeful and optimistic. I think you could bring Iran into the fold.

But listen, I'm an optimist on all of this stuff. I just want to remind people: doing nothing in Iran had tremendous risks. Doing nothing in Venezuela had tremendous risks. So it's not as though this was something that I think wasn't well calculated, but I think we have to let the cards be played and then let history be the judge. I think there's a risk in both directions, but I'm going to remain optimistic.

You said in the Gaza situation that we should have a wide berth for criticism of Israel and Netanyahu. What are your thoughts on this belief here in the United States, in this discussion, that Israel is having far too much influence over United States foreign policy?

Well, I noticed in my feed today that Naftali Bennett, who's a major Israeli politician and former prime minister, tweeted polling that showed that Israel was becoming very unpopular in the U.S. He was expressing concern about that and expressing the need to basically address that or fix that. So I think you're starting to see Israeli politicians raising that as an issue, and I think that's probably a good thing.

Yeah, there it is. It's really cool, actually, how X now just automatically translates things from foreign languages—in this case, Hebrew—and puts them in your feed. So, yeah, here's Naftali Bennett, former prime minister, saying, “This is a serious situation. There's a lot of work ahead of us to fix everything.”

Obviously, this is not Netanyahu; this is one of his political opponents. But this is something for Israel to consider and think about, and I think that they would improve their popularity if they got behind the ceasefire. I have no indication that they won't, but that would certainly be a good place to start.

I have to say, just as an aside, this auto-translate feature has done more for understanding across borders than anything I've ever seen. It is the most impressive tech feature I've seen released in years.

Putting AI and large language models aside, for people who don't know what's happening because Grok is really good at doing auto-translation, they've taken the pockets of the best of what's happening in Japan, what's happening in Israel, what's happening in France, and they're surfacing it auto-translated. Then, when you reply as an American to somebody in Japan, they see it auto-translated as well, which has led to people who don't speak the same language engaging on X in a very nuanced, fun, interesting way.

For a truth mechanism, that is just absolutely extraordinary. I think this is going to have such a profound effect. Maybe Elon and the X team should get, like, a Nobel Peace Prize award for this. I think it's going to change—I mean, I hate to be hyperbolic, but have you been using this feature, Chamath? Has it been coming up in your feed?

English.

Okay, so you're not part of the translation thing. Brad, has this hit your feed yet? Which regions are you seeing it in?

Definitely. I definitely see it in the Middle East stuff. I've seen it in Chinese, Russian, and Japanese.

Super helpful. Let me tell you, Based Japanese is a whole other level of beast.

Whoa, man. Based Japanese makes Nick Fuentes and Alex Jones seem tame. They're like, “Look at this group of people”—insert whatever group of immigrants you like—and they're like, “This is unacceptable behavior. This is not Japanese culture. These people need to get the hell out of Japan.” It is wild, folks.

You're such a good hype man. The problem is you hype buttered bread the same way you hype a nuclear reactor, and so it's hard to really tell what you're hyping because your level of excitement, the intonation, is exactly the same.

Yo, man, there's nothing better than a slice of great toast. I mean, in a way, this is like sliced bread. It's very simple, but it is so powerful in the experience.

It is true. X is better today than it's ever been. Remember, they have 70% fewer employees than they had the day Elon walked into the building.

And so, if there were ever a debate about this—and I remember everybody saying, “Oh, it's going to tip over. Oh, it's going to be a crappy experience”—the fact of the matter is, here we are a few years later: 70% fewer employees, and every other company in Silicon Valley is looking at that.

I think for a lot of these tech companies, we've hit peak employment. We're going to create a tremendous number of new jobs, but for the existing jobs, these companies are all realizing they can do more with less.

Nikita Bier just tweeted that they're about to go ham on these bot accounts that auto-reply.

Yes. Those literally ruined my feed. That's why I went to subscriber mode in my replies, and it's worked out great.

Yeah, no, shout-out to him and to Chris Sacca, who was in tears at what happened to Twitter. It's going to be okay, Chris. Sorry. No more tears. You only let subscribers respond to your tweets.

I do 50/50. Sometimes I'll just let it rip and get chaos. Other times, I have 2,000 paid subscribers. I give all the money to charity, like $30,000 a year, and it's just wonderful to get to know the same 2,000 people out of my million followers. It's kind of like having this little subset.

Sometimes I'm like, I don't have time to deal with 100 or 200 or 300 replies.

You have a million. That's incredible. I mean, it's just—

I mean, you have 2 million. I think Sacks must have a million, right? You have a million, right, Sacks?

Brad, how many do you have now? You're getting popular.

I got a couple hundred.

What's your—oh, your alt account?

I'm at 1.4 million. What are you at, Jason? Have I surpassed you?

I think you have. I'm at about 1.1.

What would it cost me to get my real name, Jason?

I know a guy. Find out.

You're at 1.1. Yeah, I made it to 1.4. I don't know how that happened exactly.

Just having the number-one podcast in the world. Another amazing episode of the number-one podcast. Chamath has 2 million, but that's only because he engages—he has just incredible moments of engaging with his haters.

Oh my God, the replies that Chamath sometimes drops are so great. I love when Chamath goes—

I light them up. I light them up.

He lights them up. Then you had somebody who was like, “Oh my God, I was in the casino and you told me to bet black, so you bet black, so I bet black and I lost my money.” And so you're responsible, and then you paid for the kid's college. He has 2 young girls, and so I funded their college accounts.

I thought that was hilarious. Just as—

Obviously, I'm very happy for him and his 2 daughters. I'm even more happy at how much it'll anger all these other goofball dorks living in their mom's basement, who literally have no take—they take no responsibility for their lives.

They should enjoy those Hot Pockets. By the way, for those folks in their mom's basement, the Hot Pockets and the fish sticks are ready, and you get 1 more hour of Xbox from Mom.

All right, listen. We missed you, Freeberg, but this is the best episode in two years. Freeberg at the end of the show. And we will see you all at the Liquidity Summit except for the 400 people on the wait list who aren't going to get in. We got an email from the guys at Athena because we were just—oh my God, they're going to hire like 500 new Athena assistants. Yes, they had a thousand people after last week when we mentioned how much we love Athena. Go to Athena.com. But that's amazing. Those are like 500 hardworking men and women who are working in the Philippines. Sacks have great jobs. Sacks, I'm going to get you a couple of Athena assistants as a birthday present. That's what I'm going to get. You're going to love this, Sacks. Athena assistants are the best. Congratulations to my friends over there. All right, everybody. We'll see you next time. Love you boys on tonight favorite podcast. Let your winners ride. Rain and we open source it to the fans and they've just gone crazy with it. Love you queen of winners. Besties are gone. That is my dog taking out your driveways. Oh man, my appetiter will meet. We should all just get a room and just have one big huge orgy cuz they're all just useless. It's like this like sexual tension that we just need to release somehow. That's going to be good. We need to get merch. I'm going all in.