开源 AI 之争、Anthropic 的15亿美元赔付、纽约社会主义者：驱逐租客 = 暴力？

It was a big week. The continuing number-one story in the world is Kimi K3. It sparked a debate about banning Chinese open-source models here in the United States, and it’s gone all the way to the White House last Friday.

China’s Moonshot AI released the Kimi K3 open-source model. Its performance is on par—not 6 months behind, not 12 months behind, but now on par—with models like Opus 4.8 and GPT-5.6, which in and of itself is extraordinary, but it’s about 50% cheaper. This has created a bit of a panic, similar to the DeepSeek moment that we had here back in early 2025.

The White House, David Sacks, has gotten involved. Michael Kratsios, friend of the show, said, quote, “We have information that Moonshot AI distilled Anthropic’s Claude for the development of its K3 model.”

Here’s how the Trump administration has reacted so far. On Monday, Axios reported the White House was considering banning Chinese open-source models. A couple of weeks ago, David—I think it was 3 weeks ago—I gave that to you as a hypothetical, and here we are. On Wednesday, Wired reported that Howard Lutnick from our Commerce Department, friend of the show, does not want to ban Chinese models.

So apparently, there’s palace intrigue. There might be different opinions inside Trump’s White House. Instead, they want to incentivize more U.S. frontier labs to develop better open-source models. Polymarket says there’s a 45% chance the U.S. government bans an open-source model in 2026. That was a brand-new market; it started at just 22% a couple of days ago. Sacks, you called it 2 months ago: our first victory fap of the episode.

1. The Open Source Ban Debate

I think where it’s all leading to is an effort to ban open-source models. There are a lot of breadcrumbs leading here. You look at a lot of the rhetoric around how models need to have guardrails and how, with open-source models, the guardrails can be removed and therefore they’re dangerous.

You see this rhetoric already in Anthropic’s blog posts. Any threat that they describe, they go out of their way to take that shot at open-source models. I think, again, they’re trying to create ideas or put predicate facts in the public record to justify an action later on. I think it’s just a matter of time before they feel like they’re in a position where maybe they can push for that type of ban directly.

All right, Sacks. What’s going on at the White House? What is the administration’s position here? Why are we getting multiple opinions? Is the White House testing and probing to figure out what its position is here, or is it just that this is a super-dynamic situation? What’s going on?

Well, look, I have it on good authority that there is no decision by the White House to ban open-source models, and I think they want that known. I think there’s an ongoing conversation happening around what to do about Chinese distillation, and we should talk about that, but no decision has been made.

The president listens to a chorus of voices. He wants to get advice from as many people as possible, and I’m confident that if everybody weighs in, the president will make the right decision, as he always has with these tech issues. I think his instincts have been absolutely impeccable on this, and he’s always supported a lighter-regulation, more-open approach. That’s why I think the U.S. is winning the AI race. I think that’s kind of where things stand.

Where do you stand? Where do you stand, Sacks? That’s what everybody wants to know.

Yeah. I think it’s important for me to make my opinion known in the spirit of contributing my voice, so the president hears all perspectives and then can make the best decision.

Look, I think it would be a tragic mistake if the government were to take action against the open-source ecosystem. That would do nothing but hurt America’s position in this AI race. It would backfire badly. The key point here is that, regardless of what you think about distillation, you cannot punish American developers for it. You can’t say that American companies and American developers can’t use Chinese contributions to the public domain. That’s just cutting off our nose to spite our face.

Yes.

Obviously, American companies have to be able to use everything that’s in the public domain because the rest of the world will be using those things.

Let me just say, I’ve said for a while that Anthropic is guilty of regulatory capture, of attempts to seek government protection. This is a company that is the fastest-growing tech company at scale in history. They started the year at $10 billion of ARR. They’re now over $70 billion of ARR. This is not a company that needs government protection.

No.

This is not a company that’s under threat from competitors, whether they’re Chinese or otherwise. And yet they have been very successful at trying to panic everybody into thinking that they need some sort of government protection.

The tell on this—the way that you know that this whole distillation thing is fake—is that if stopping distillation were their primary objective, Anthropic would push to ban Chinese access to American models, not American access to Chinese models.

Yes. They could, and that is achievable. They could block it.

They’re the ones in the best position to block it.

Yeah.

If industrial-scale distillation is a national-security threat, they’re the ones who need to stop it because that is the place where distillation occurs. You have to stop it at the source. Once you allow Chinese companies to distill, the horse is out of the barn.

Yeah.

The reality is that, in their lust for growth, Anthropic has done a very poor job at stopping distillation. They’re saying that distillation is occurring at industrial scale. If it’s industrial scale, it must be pretty obvious to see.

Yeah.

So why aren’t you stopping it?

The way it’s happening is through waves of accounts being created by students, rolled up, and sold on the dark web in those kinds of channels. You’ve got people in Manila, in the Philippines, I understand, and India, signing up for all those accounts and then sending them to the dark web and selling them, using IP addresses from America. That’s how it occurs.

Sure, but the more industrial-scale it is, the more obvious it is to see.

Absolutely.

And Chamath has been saying for a while, “Why don’t you KYC your customers?”

Exactly.

Well, they know that if they KYC their customers, it’ll slow their growth. So instead, what they’re saying is, “Hey, ban our competitors.”

Well, that’s ridiculous. They’re in the best position to stop the distillation. I think they’re negligent about doing that. Or, if they really think it’s that big a threat, they should use a few points of their 90% gross margins to do that.

What you don’t do is then say that American developers cannot use everything that’s in the public domain. It seems to me that this debate is all backwards. The question should be on Anthropic to explain why it’s doing such a bad job—

All right.

—not on the whole American open-source ecosystem to be punished for Anthropic’s failure.

All right. Friedberg, I have a really good question for you, but before we do that, Chamath, can you give me maybe a little bit of what you’re hearing on your sales calls for 80/90[?]?

You’re talking to enterprises. They’re hearing all these reports—whether it’s you, Dario, this pod, or other places—talking about, “Hey, open source is ready. This is the moment. Get control, AI sovereignty,” et cetera. They must be calling you up and saying, “Okay, we’re ready. How do we get these things on-prem? How do we do it?”

What’s happening on those calls you’re doing with enterprises? And then can you maybe give people an idea of what distillation is and why it’s so important here?

2. How Distillation Works

Let’s start with the second thing. Distillation is when you fire up a model, ask it a question, observe it, take its output, and use that in training your own model. Now multiply that behavior by tens of millions, and what you exfiltrate is essentially trillions of questions and answers.

Sacks is right. If you really care about distillation, you implement KYC. You force people to make an account not just with a username and a password, but with some form of identification, maybe with a bounded credit card. There are all kinds of steps that you can take that would frankly slow things down in terms of revenue traction but would solve the distillation problem on its face. So that isn’t really a thing. It’s a bit of a red herring.

The other thing on distillation is that everybody has at some point distilled. The question is, who is distilling from whom? It looks like that funny meme where there are 9 Spider-Men all pointing at each other. That’s what this is, because Anthropic has distilled from all of these publishers. They just paid a $1.5 billion fine. Apparently, OpenAI distilled from The New York Times. There’s still an ongoing lawsuit. The Chinese labs have distilled from Anthropic.

It’s wholesale stealing everywhere. Yeah.

Well, I don’t want to call it stealing because it’s not clear who actually owns the copyright in the first place, but here should be the important observation.

3. AI Models Become Commodities

These models are getting commoditized much faster than anybody thought. How do we know this? Because there is no meaningful, sustained advantage once a model publishes its performance criteria. What you see is that, literally within weeks, other models—some open, some closed, some open-weight—are able to match and, in some cases, exceed the performance.

So I think what's happening here is that a handful of American companies have realized, “Whoa, this value that we are seeing today may not be sustainable in a 5- and 10-year period.” When you go and present a business model to Wall Street, you need to have that certainty; otherwise, it impacts your valuation. I think a lot of what's happening right now, Jason, is a valuation-preservation game by the closed frontier labs.

Because if you actually understood how commoditized these things are becoming and the velocity at which it's happening, you see that the real business model is not in the foundational model anymore. It's at the application layer above, and it's in the infrastructure below, whether that's the cloud or whether that's chips. In the absence of regulatory intervention and the absence of the United States government stepping in to put its thumb on the scale, what will happen is that, as people learn about how value is changing, they're going to put more value in the application layer and more value in the infrastructure layer.

That is bad for closed frontier labs, especially when they're mispriced at 25 to 50 times the open alternative. This is an attempt to stop a competitor that is of the same quality but just much cheaper. Now, there's another important thing here, which is that if the United States government intervenes, it will tank the stock market.

Okay.

Period. Not debatable. Now, you can debate which companies get tanked, and we can probably play that scenario out. But, for example, if they said, “No more open source. American companies cannot use open source,” okay, let's take it from a stock perspective. Let's take an average normal company: Coca-Cola.

“Hey, Coca-Cola, you're trying to use AI to improve your business. You know what? You can only use these 2 options, and those things cost 50 to 100 times more than your other best alternative.” That'll eventually show up in your costs because AI is supposed to be this incredible thing that just solves every problem and does everything for you.

This incredibly important input into your cost model is now orders of magnitude—multiples—greater than that of your competitors outside the United States, simply because you're in the United States. So what would the capital markets do? They're going to say, “Wow, you have a crazy cost structure. This doesn't make sense. You're forced to absorb costs that aren't rational or market-driven.”

So then Coca-Cola has to get re-rated. But then you look at the people who are selling those tokens, and this is where Anthropic and OpenAI need to understand: if the government comes in and actually tells you that there's no open source, their valuation will crater. Why? Because all of that revenue is artificially being propped up.

Hmm.

It's not being driven by market demand, where you're being forced to compete; it's because of regulatory capture, where you now get an artificial constraint. But it only works in 1 market. So, anyway, all roads lead to market chaos—

Love it. Yeah.

—if anybody gets involved, so we should just not get involved.

What it sounds like is you're saying that American enterprises will pay a token tax if the government gives—

Yeah.

—Anthropic and OpenAI—

Yeah.

—a government-enforced duopoly—

Yeah.

—and enterprises are no longer free to use open source like the rest of the world. Do svidaniya. Yes.

Yeah, the markets will decapitalize.

We will put ourselves on an island. Yeah. We'll be on an island of overly expensive AI. You can have Coca-Cola or Pepsi.

Yeah. Well, it's not two—

It's beverage choices.

You have 2 beverage choices, but they cost 50 times more—

Yes.

—than Coke outside of America. This is the point.

Yes.

We already have Coke everywhere. So you can buy 50-cent Coke or $50 Coke. Why would you buy—

Yeah.

—$50 Coke when you can buy 50-cent Coke?

Yeah. All right, let me get Friedberg in here. Yeah.

Okay. I was going to say just 1 thing on this. This goes back to my point about these proposals to ban open source that are coming from Anthropic. They don't solve the distillation problem. If distillation's the problem, you have to stop it at the source. In other words, if you want to ban open source in America, the rest of the world will still be using Chinese open models. We want to solve that problem.

Yes, and that means they'll get the data, and they'll get the reinforcement learning, and then we lose the AI race, guaranteed. Friedberg, let's take it from a Graham Allison perspective and level up the discussion here. It would be quite provocative to ban the Chinese models. How does Xi Jinping respond to that? How does the CCP respond to that? This is a crazy chessboard. It would seem like a pretty escalatory move, and we'd be going up the ladder. Yeah, Friedberg?

4. Open Source Benefits Everyone

Yeah. Look, I don't think it's as relevant that the open-source model is published by China. There may be security risks, but those can be estimated and addressed. I think the 3 things that are worth highlighting on this issue—I’m very much aligned with Sacks and Chamath—and I think the 3 things are really around distillation.

I don't think distillation is just about AI. Distillation is a process whereby you look at the end product that someone else has produced and think about and learn about how to engineer your product. It is a common technique that is used across every product category in every industry. One carmaker will look at how the other carmaker's car operates, and they will use that to help them design a better car.

At Google, in the early days, we would submit millions of search queries to Yahoo and Microsoft search engines to see what the result sets were, and we would compare our results against their results as a way of improving our search engine rankings and our algorithm. It was a very common technique. It doesn't mean we were stealing their algorithm. We didn't go into their servers and steal their software. We looked at the output of their software and used that to improve our software.

It was called benchmarking, right? It was benchmarking.

You can call it—there have been a million terms. Exactly right. I don't think that this matters as much. I think the question around copyright infringement or IP infringement—

That's our second story. Yeah, we've got to—

And Sacks is right. There's a terms-of-service question here, but that's on the service providers to fix—the terms of service blocking people from doing this, if they so chose. But the copyright argument, the IP argument, is really about: Did they steal the software?

Right.

They didn't steal the software; they just looked at the output. That's not IP infringement. That's not copyright infringement—

Right.

—in the classical sense. So I do think, from a distillation IP argument perspective, it's the output, not the process, that matters. The question is: Are they taking copies of copyrighted software and using it, or are they looking at the output? I think output, not the process of engineering, is what you really need to assess here.

Can I insert something, Friedberg, and you can comment on it?

Yeah.

I think a lot of people in the policymaking community don't understand this distinction, but I think everybody in tech does, and I think it's a big part of why there's a disconnect on this. There's a huge difference between model weights and outputs, right? The weights are the file of numbers—the numerical parameters in a model.

That's the software code. That's the code.

That's the software. That's the code. And if Chinese companies were to steal—

That software—

—proprietary weights from Anthropic or OpenAI, that would be theft.

That's right.

Okay? But that's not what we're talking about here because no one's accused them of that. What we're talking about here is taking model outputs and then trying to learn from them. And—

Using other people's software to learn.

Yes, and it's exactly the situation you said with the Google searches or whatever. Here's the thing that's so hypocritical: OpenAI and Anthropic have both argued that they are free to train on all the world's output, regardless of whether the creator wants them to or not.

That's right.

That is their current position. That's like Chamath mentioned with The New York Times lawsuit.

We're going to get to that. Yeah.

The New York Times is suing OpenAI right now for going onto The New York Times' website in violation of The New York Times' terms of service, scraping all the information, and training on it. OpenAI's argument is, “Look, we're not stealing anything. We're taking the output of The New York Times, and we are deriving our own model weights.”

100%.

And that is exactly what these Chinese models are doing—

100%, yeah.

—they are taking the output of American models and then deriving their own—

And they're learning from it.

weights. They're learning from it.

Yeah. And so the accusation, though, just so we're clear here—

Yeah.

—is that there are ethical issues around the industrial-scale, covert breaking of terms of service. That's what the White House has been—

The terms of service, yes.

—talking about as well.

Yeah, 100%.

So just so people understand.

Yeah.

There is a bit of recognition of that in this.

Yeah, look.

Yeah.

Let me be clear that I'm not defending China in this. In fact, look at my bona fides. I was the first administration official to even talk about distillation. I did it in January 2025 when DeepSeek came out. I went on Laura Ingraham, and I think I was probably the first person in the government to even explain this concept publicly to people.

Moreover, I was a co-author of the Winning the AI Race report, in which the whole premise of it was that we want to win, we want to beat China. I'm definitely not someone in this camp that doesn't want the U.S. to win. I want the U.S. to win. The question is how, and if we shoot ourselves in the foot by banning open source—which is to say, not letting all of our American companies take advantage of open source when the rest of the world is able to—then that is a huge problem.

Now, I'm fine with Anthropic and OpenAI enforcing their terms of service. They need to do a better job and stop the distillation from occurring in the first place. If there are things that the government can do to help them, okay, but I'm not sure what those things are. What needs to happen is those companies need to do a better job enforcing their terms of service.

Yes, that's right.

Okay, now, Friedberg, you had—I think you said you had 3 points you were making. I think you made 1. I want to get the other 2 out of you.

Yeah, so the other one was just on the free-speech argument. What is an open-source model? I think the audience needs to understand this if you're not from the software industry. Open source is a downloadable package of software. You can think about it as downloading a book. You just got all the code. Once you get the code, you've got it on your computer. You don't have to be connected to the internet. You can just run it and use it.

I think part of the challenge that's going to be faced here if there is any attempt at restricting open source is how you actually enforce restrictions on it. You're basically telling people that once they've downloaded and gotten a copy of this free, publicly available software, they're not allowed to use it, and that becomes a real challenge. I don't think we have a lot of great precedent for that. It's going to be very ugly to try and stop open source.

I do think there's a question on copyright action, but if there is, there's a legal due process to go through to make that case and stop that open source from being available. My 3rd point is just that open source is better for the world. To Chamath's point, this is 100 times cheaper. That is better for the industry and for enterprise. The beneficiaries are going to be the economy and the consumer.

Fundamentally, if you look back on the internet, in the early days, Netscape made a proprietary browser and proprietary server software—the Netscape software. They went public, they were the first to do this, and it was super valuable and profitable. That company ended up getting crushed because of open source.

The Mozilla Foundation was formed to create an open-source web browser called Firefox, and then Google ended up hiring everyone and made Chromium, but it was still open source. The Apache Software Foundation set up the first HTTP server as an open-source product rather than having to pay Netscape, Microsoft, or Oracle for their server software. Anyone with a computer could download the Apache software, make a web server, be on the internet, and create a website.

What ended up happening is that the value accrued to the internet. It didn't accrue to the small number of software providers that controlled the gate and the portal of the internet.

Yes.

Basically, everything got open-sourced, and Google took off, eBay took off, Etsy and Amazon took off, along with the millions of small websites, the millions of small businesses, and everyone that benefited from an openly accessible, open-source internet. If the internet was closed and there were proprietary software gates and portals throughout the internet that everyone had to pay to get through, the internet would not have taken off, the economy wouldn't have grown, and all these jobs wouldn't have been created.

We had that, Friedberg.

Exactly.

It was called AOL and CompuServe. Literally, we had that.

And now, when we look at this analogy, the analogy here is that if open-source AI takes off, then all the worries that Bernie Sanders, Elizabeth Warren, and all the socialists are harping and barking about are not going to be the case anymore. You're not going to see all the value of AI accrue to 2 or 3 or 4 companies and their small group of billionaire shareholders.

What will happen is AI proliferates, and 1 million AI-integrated enterprises all over the world will benefit. Everyone will benefit. The economy will grow, jobs will be created, and AI becomes a power for good, for creating an open economy.

I know that some people who are listening to this in the government and who are on the other side are going to say, "But Chinese open source versus American open source." Let it all proliferate. Frankly, if the Chinese are violating copyright or stealing software, go after them for that. Put in place trade sanctions. Do all that you can do to stop that from happening.

Fundamentally, open-source AI will transform the global economy, and it will ensure that the economic value of AI will diffuse to everyone and not be held captive by a small number of controllers—

This is why—

—of controls.

—the 1% that controls open source, if they have it—

We have to acknowledge—

—the oligarchs are going to get their clocks rung, and we don't need to have a wealth tax. Chamath, you were going to add to this.

Yeah. We have to acknowledge that it's incredible how fast the value captured in this segment of the market has basically evaporated. I've never seen it in my 25 years in Silicon Valley where a sector of the economy can absorb hundreds and hundreds of billions of dollars—

And then you think that there's going to be economic pricing power many decades into the future, and it effectively evaporates in months. Months.

It took off in months. Remember? Remember how big Anthropic was?

It took off in months, and it's evaporated in months.

5. Closed Models Still Win

No. This is an area where I disagree with you guys.

“Evaporate” is a strong term. It does look like it could slow down or plateau—

Go ahead, Sacks. Make your argument, and I'll give you my argument.

Yeah.

But let me pull up the Anthropic revenue chart before you go there, Sacks. This will help mitigate it here and educate the audience. As you can see here, we got a little bit of a stall in Anthropic's revenue in the last couple of months, and it seems—and this is 3rd-party tracking, so it's not perfect data—but we do see that this is a clear headwind.

Do you have the 2nd chart I had? I talked on the pod just last week or the week before when we were having the discussion about open source. I think it was Brad Gerstner from Altimeter who was saying, "Hey, tokens are still growing." There are routers that track OpenRouter. The better an open-source model does, and the easier it is to implement on your own servers and take it in-house, et cetera, those are dark tokens. They're not recorded. You're not going to see them show up on a revenue chart anywhere because they are essentially free. You just need to have servers and energy to do them.

Here you see that now well over 50% of these are coming from Chinese models.

By the way, Sacks, I want to be clear before you give the counter. I'm not saying that these companies won't make money. That's not what I'm saying. But what I am saying is that markets are very savvy in looking through current earnings and asking a very specific question: What does this revenue look like 10 years from now? Does it go up? Does it go down? Is there more competition, or is there less competition? Is it effectively monopolistic, or is it more of a commodity?

If it's a commodity, how many people can price this good? At what price is the market-clearing price of that good in 10 years? All I'm saying is that normally those variables get exposed and those cards get turned over relatively slowly. And so you have 5- to 10-year cycles to transition from being an exclusive provider of a good to effectively a commodity provider of a good.

All I'm observing is that it's so unique that only technology could create a market where that cycle could get compressed into a few years. Because it is very hard, if you're an allocator of money, to sit there and look at this data and not wonder to yourself why it's not a commodity in 5 to 7 to 10 years. And when they get to that conclusion—which every capital allocator will, because it'll be pretty negligent not to—it's very hard to assign huge future premiums.

And where the real money is going, by the way—and we saw it in Google's earnings, which I'm sure we'll talk about—

That's a third story.

It's going to the cloud. It's going to the infrastructure.

Yes.

So, by the way, there's another cohort of people that don't want to see the end of open source because they want to serve the cheapest models possible because—

Sacks.

They know that's where all the margin capture is.

Where is this going, Sacks? Is it both open? Are we just going to see a proliferation? As far as I'm concerned, there is an unlimited appetite for on-demand intelligence, and there will continue to be. I don't think there's an upper bound for how much intelligence you can tap as long as it continues to get better.

The thesis would be that it's a commodity and the prices keep going down, but consumption keeps going up. What are your thoughts?

Yeah, look—

Then we're going to go to our second story, which is the IP story. The third story is going to be the markets, and Google specifically, and Tesla and SpaceX.

Look, there's been a lot of violent agreement on this show so far, so let me just make the counterargument. I think that both open source and closed source will be big winners in this. I think the market is huge, and they each serve their purpose.

What happened with the introduction of Kimi K3 is that there was a little bit of a panic in which everybody said, "Oh my God, all the Chinese models have caught up, and they're giving them away for free, and they're going to destroy our leading American frontier labs." Anthropic and OpenAI are running around saying, "Listen, we can't continue to invest billions of dollars if Chinese companies can just steal our weights." Right? So that's the argument they're making that government officials are responding to.

The truth of the matter is that when Kimi K3 first launched, I was concerned about it. I thought, "Oh, has China caught up? Are they now able to produce a much cheaper frontier model?" Then the details started coming out. Ben Thompson, on his blog, went through some of the cost numbers, and it turns out that Kimi K3 is not that much cheaper to run. There's no significant cost advantage to it.

So that's point number 1. Point number 2 is that China has not caught up. It's true that Kimi K3 scored really well on the Arena battleground for frontend coding and web development, but that's just 1 test. That's just 1 dimension. There are areas where it scores really well, but there are lots of other areas where it doesn't score that well. So it is not a clear advance or a clear catch-up to the leading American models.

Moreover, you still have stuff in the labs by Anthropic and OpenAI that is way ahead of this, and the reports are that Sam is going to Washington next week to talk about GPT-6.0, which is blowing the doors off. So I don't believe that China has really caught up.

Sam's going to go see Daddy. What is he going to ask for?

I think there are incredible unreleased models in the pipeline, and we need to let our horses run here and not slow them down with a bunch of unnecessary hoops. If we do that, I think we're going to be just fine. I don't believe that China has caught up. I still think we're 6 months ahead.

The final point on this is that if you look at revenue, which is the test of real usage in the real world, Anthropic and OpenAI are blowing the doors off. They are by far the fastest-growing tech companies at scale that we've ever seen.

Jason, you showed this chart that supposedly shows a hiccup in Anthropic. Their internal forecast was to 10X this year, from $10 billion to $100 billion. We're in the middle of the year, and they're already over $70 billion in ARR.

They're doing great.

They're easily going to get to $100 billion. We don't really know what this little blip is here. I think one thing it might be is that if you superimpose the OpenAI chart on this, they have reaccelerated over the past month. So I think that if you were to add OpenAI and—

Yeah, Codex is excellent. Codex is excellent.

Codex is excellent, and I think they're taking a little bit of share. Sam is out there tweeting, "We've got our mojo back." They're taking their forecasts up. I think they were expecting to end the year at $60 billion of ARR, and I think they're forecasting more like $75 billion of exit ARR.

My guess is that if you were to superimpose OpenAI and Anthropic and look at them together, and you were basically just to say, let's call it the frontier-model duopoly in the U.S., you do not see any slowdown.

Yeah.

You don't see any blip. They're taking their forecasts up. The reality is that if distillation is going on, it's been a thing for—again, I pointed it out back in January of 2025 with DeepSeek. So it's been a thing this entire time that they've been growing exponentially.

I just don't believe that these guys are actually suffering in any way. I don't think they need government protection. I think they're still growing exponentially. This is a little bit of a case of—what do you call it in basketball when a player flops? You know, that you do—

A flop. You're flopping.

A foul.

Yeah, you're trying to draw a foul.

A foul flop or whatever.

Yeah, it's a flop. Yeah.

Or you basically act super dramatic after a foul—

Oh—

—in order to draw the charge.

Okay, you're foul-baiting when you're trying to—

Yes.

—get a foul, and then you're flopping, which is just about exaggerating—

Yes.

—like LeBron does. Yeah.

And that's exactly—yes. It's a LeBron—

LeBron. LeBron.

It's a LeBron flop, you know, where—

Yeah, flip-flop.

These guys are trying to draw the foul. They're trying to get the government to intervene. Now, why are they doing this? Because they're in the middle of roadshows right now.

And Chamath, to your point, I do think they get the legitimate question: "Why won't you be commoditized over time by open source?" By far the best response to that would be that if they can lure the government into giving them a government-protected duopoly, that would be incredible.

The best answer is what you gave, and Anthropic and OpenAI should own this because they're good at it, which is that they're going to go up the stack to the application layer.

Well, they already have.

They've done—

Yeah.

I know, but they've done it in this way, which is a little ham-handed in some cases, but they're excellent at it. These end-user apps are really good, and they should just own that.

That should be their answer to Wall Street, which is, "Guys, we have the best model. We will eventually go up the stack." If I were them, I'd actually practice the following answer: "There may be a version of a model that I don't release and just keep for myself, and I'll just use it in my own applications." How about that?

Yeah. Well, that would be super—

That's the real answer to their question.

That's anticompetitive, yeah.

No, it's not.

Yeah.

They're allowed to build a model and not release it—

Well, it would create a chance—

—and use it for themselves.

Hold on. Hold on. Let me answer that. It would be anticompetitive in the eyes of their customers. It might not be in the government's eyes. But if you are using them as a customer, and you're Lovable, which I talk to, and you're paying a ton of money, or you're ElevenLabs and you're paying them a ton of money, and they say, "Hey, we've got our latest and greatest. You can't use it because we're going to compete with your company," you would stop using it and go to open source.

I'll tell you why, Sacks: I think you're wrong on this issue. I think open source is having its moment. I work with startups. They're all moving off of these models, and they're using open source at much cheaper rates because a lot of the jobs don't need the latest models. They can use the last generation's models, and people are moving them local. They're hosting them themselves.

I think there's a nonzero chance this is going to derail Anthropic and OpenAI's IPOs and their—

No, they're going to be fine.

—midterm future.

They're going to be fine—

They're going to be fine.

—as long as everybody stays in their lane.

I don't want to be rude. Let me—no, no, no. Hold on. Let me finish, guys. I'm making my point. Shut the fuck up for 30 seconds.

I believe that this is going to derail their IPOs. I'm taking it from the top.

It's going to derail their IPOs. There's going to be headwinds against it because I think they're going to have massive margin compression. I believe they're spending so much money that I think they're going to get caught in a trap. I think this could be a trap for them. They overspend, they don't have the same profitability, and it doesn't pencil out.

Startups are the future. Startups are what the enterprise eventually copies. I think you're wrong. Sacks, go ahead.

Well, I'll come back, but let Chamath respond—

Okay, sure. Go ahead.

Because he's raising his hand.

I think you're wrong, Sacks.

Okay, you're totally wrong.

I think they've caught up on 95% of the jobs. This is a crazy habit. I'm not saying that the IPO is wrong.

No, they haven't.

It's a non-zero chance that they're going to get slowed down.

I think the answer is slightly different. It's not that 95% of the bleeding-edge tasks can be done by everybody. It's that 95% of the tasks can be done by many different models. That's the actual answer, and that's okay.

It's also okay for Anthropic to say, "You know what? I have this next-generation class of model. I'm going to instantiate, I don't know, a life sciences program or a cybersecurity business." That's where they can capture, as Sacks was saying before, over time, trillions and trillions of enterprise value, because I do think they have excellent models, excellent engineers, and momentum.

But if you're going to build a business model that tries to ascribe a lot of terminal value to this layer, I think that's a mathematical mistake. That's it. You can't do it.

That's exactly right, Chamath. I agree.

Open source can be a specific headwind to these companies, specifically because all of their best customers—and I talk to them—whether it's Lovable or ElevenLabs or the startups that were spending hundreds of thousands of dollars with them every quarter just 6 months ago, have all moved en masse to GLM-5.2 and are making their own models. The cat's out of the bag. They're going to start losing a lot of customers to open source, and Google, AWS, and Elon Web Services are going to host them.

How do I know this? Every time I do a job, I'm using Perplexity Computer—not a paid partnership or anything; it just happens to be the best harness that I found—and I start with Grok, Nemotron, GLM-5.2. I also put it into Claude, and the results are as good or better. In other words, I can't even tell the difference between these.

Let's superimpose the OpenAI numbers on top of the Anthropic numbers, because I think it supports the point I'm trying to make here.

Yeah, and these are estimates, by the way. This is not—

Right.

Literally from the companies. Just want to make sure people know.

So, according to this company that—I mean, look, who knows how they derive this? We don't know that they're totally true. They're basically showing that OpenAI's run rate, which is ARR, rose from $33 billion in May to $41.3 billion in July, so they're seeing acceleration. Like I said, Sam is out there saying they got their mojo back. They're going to have their best 12 months forward-looking ever.

And look, Anthropic is still growing really fast. We've talked about this on a previous show. It's not physically possible to grow 10× year over year forever. You'll run out of compute. You'll run out of energy. You'll run out of everything.

People.

Yeah, there's just no way.

To do things.

But Anthropic, I mean, look—

Economy.

If anybody had said Anthropic would be at over 70 billion at the midpoint of the year, back in January, when they were at 10 billion, you would've said, "This is the fastest-growing tech company of all time." The idea that they're at risk of getting their entire franchise destroyed—it's not in the data yet, is what I'm trying to say.

Moreover, Chamath, to your point, it's not only about the model. It's also about the harness, the connectors, and the enterprise agreements. There are a lot of things here that you need in order to grow a business—

Yeah.

—like this. And, you know, this idea that they need to race to get government protection because of a competitive risk that might happen in the future, I think is just kind of unseemly and gross. This is literally the most successful tech company of all time, and they're racing to the government to basically say, "You need to protect us against our competitors."

It's a great argument.

Not just our Chinese competitors—

No, no, our potential future competitors.

Our American competitors.

Our potential future competitors. Yeah, they should hire Lina Khan. Okay.

Frankly, it's gross. Let me give a couple of examples, because I know people don't have a lot of sympathy for Chinese companies. That's fine. I'm not defending Chinese companies. I'm defending American developers who need to be able to use everything in the public domain.

Let me give you an example. Mira Murati's new company, Thinking Machines Lab, currently has the best American open-source model. You know how it was trained? It was bootstrapped. It was distilled from a Chinese model, Kimi K2.5.

Now, if you say that Chinese model is based on IP theft, what is Thinking Machines? It's a derivative work from a model that Anthropic is trying to taint as IP theft. By the way, there's been no evidence of this. There's no evidentiary process. They're simply trying to paint with a very broad brush here and say that now the model is tainted.

Let me give you another example. Cursor rolled out its new product, Composer 2. They were able to post-train that model using Kimi K2.5 on their own proprietary coding data. So think about this: They started with a Chinese open-source model, then they used their own data, and they came up with a new derivative product.

This is the way open source works. You take things that are in the public domain, you fork them, and you make them your own. By the way, once it's in the public domain, it's not a Chinese model anymore. It is open weights that are freely available to anyone. It's a file, okay?

You take that, you fork it, and you run it on your own hardware in an American data center. No packets are going back to China. No data is going back to China. Nothing is going back to China. An American company has taken open-source contributions in the public domain, made them its own, and then developed its own model.

If you say that American companies can't do that, or that somehow it's tainted with IP theft, you are basically going to put a dagger through the heart of the entire American open-source ecosystem. And that is exactly what Anthropic wants, because they do not want to have the competition.

Friedberg, maybe you can close this out here, and then I'll put my final—

I'll just zoom out for a second, and I'll say—

—stamp on it.

Think about the strategy as well for China. If you think about the global economy of the last 50 years, the US has accrued so much value by being at the core of the knowledge economy and effectively a services economy. In that sense, through the development of intellectual property—of IP, of knowledge—and then the conversion of one bit to another bit, we've been able to derive trillions of dollars in GDP. Meanwhile, we outsourced manufacturing and created a sleeping giant in China where they have this incredible manufacturing capacity.

At the end of the day, if you think about the course of human technology evolution and human prosperity, it's largely driven by our capacity to convert molecules from one form to another and use the least amount of energy possible to do that. That's it—all of technology ultimately leads to that simple equation: molecule conversion. Making that beautiful couch behind you at the lowest cost possible, making materials, making semiconductors, making all this stuff. Everything in our world is driven by molecule conversion.

So, at the end of the day, the knowledge economy and the services economy get compressed, much like AI—and open-source AI in particular—effectively flattens that value. All of that value is now open source, it's free, and it's simply a function of turning on a switch and running it. The value of the US and the Western economy has been largely degraded. What's left is the value of the molecule economy: the ability to convert molecules and use energy to do that.

When you look at the juxtaposition of China versus the United States today, we have 1 terawatt of electricity production capacity in the US, and they're on their way to having 8. We have about 10 billion square feet of manufacturing capacity. They have 200 billion square feet of manufacturing capacity. So they have 20× the manufacturing capacity, 8× the electricity production, plus all of their other sources of energy.

I think that's the long game for China over a 1-, 2-, or 3-decade process: by compressing the knowledge economy and the services economy, commoditizing it completely, they are left holding all the value in the global economy because they can make stuff, and they can make it cheaper than anyone because they have the most electricity production.

That’s a very simple rubric for how I look at the long game that they’re trying to play here.

You believe they’re trying to commoditize this very important space, just like they did for—

Global knowledge—

—cables and cars, et cetera.

Global knowledge and global services. The creation and movement of bits gets commoditized, and what’s left over is the creation of electricity and the creation of molecules, both of which they have this very difficult-to-surmount advantage that’s going to make them the core dependency for the world. That’s what I think is kind of the long game here.

And just so you know, this data comes from reports in places like The Information, other sources, or leaked numbers, and they try to make charts based on it.

Yeah, and by the way, I have my own sources too. I’ve talked to investors in these companies, and I’m just telling you that both Anthropic and OpenAI are taking their estimates and forecasts up right now.

So, look, I think in the future it may be the case that open source takes share. Fine, it’s because the market’s so big, and there is always a market for open because open is more controllable and more customizable. You can own it. You get the data sovereignty.

Sovereignty.

You get the sovereignty, but it’s also more work. So there are different use cases and different parts of the market. And the reality—

That’s actually very interesting. The more-work part there, Sacks, is super interesting. Three to 6 months ago, it was so much work to stand these up, and now there are so many intermediaries building the harnesses that default to it that that’s getting worked out. But that has always been the issue with open source, for sure: the amount of work it takes to implement.

Man, everyone’s talking their books.

I’m not. I’m not. I’m not.

Actually, I’d say that includes the Anthropic and OpenAI investors. It’s amazing how many of these investors—

Yeah, Brad was on the show 2 weeks ago. Were you on the episode? Brad was like, “Let me tell you why this is gonna…” He’s holding on to this “white knight” freedom. He’s like, “Ah.”

I actually give Brad a lot of credit because I do think that he’s objective about public policy, or as objective as you can be given that he does own all these companies. But look, let me just tell you that I see a lot of folks who are suddenly China hawks and saying, “We need to stop China. We need to stop them.” It’s like, okay.

Oh, yeah, now they’re China hawks—

Yeah.

—after they wanted to sell their chips there.

Like, how about disclosing first whether you’re on the cap table of Anthropic, okay?

Absolutely. Are you guys on any of these cap tables?

No, I’m not.

Not directly. No, not directly—that’s how I would put it. Indirectly, maybe I’ve got a little access. You know what it’s like, Chamath? It’s like when you got that great flush or straight and you’re like, “Please don’t pair the board.” Brad’s like, “Don’t pair the board. Please don’t pair the board.” All right, here we go.

I fucking love open source. I love this open source AI stuff.

I love it.

I think it’s so awesome—the value, the—

So punk rock.

There’s so much to be done with it. It’s just—

Let’s go.

—exciting and awesome. And, yeah, Chamath’s point is exactly right. Most of the models can do 95% of the tasks.

Yeah.

And if that’s the case, then it’s not like everyone needs to scramble to get the best open source model. You just need open source to do 95% of what you want to do with AI, and then for the other 5%, you get something specialized or high-value, or you pay a premium.

And by the way, if you’re a big enterprise and you need to have wrappers and support and all these other tools for your employees, buy Anthropic or OpenAI or Groq’s tools or Gemini. There are plenty of options.

6. Anthropic Settles Copyright Case

All right, Anthropic copyrights. Here we go. Anthropic settled its AI copyright lawsuit for $1.5 billion on Monday—the largest copyright settlement in the history of the United States of America. It’s the first major AI copyright lawsuit to settle, and there are many more in the pipeline.

Anthropic downloaded 7 million books from pirated websites to train Claude. That alone may or may not be a crime. This has been adjudicated a little bit in the courts. They had ruled previously that training AI on copyrighted books is legal under fair use, but there are going to be some future cases.

So this is a settlement. They didn’t go to the mat. Lawyers are getting $101 million; authors get $3,000 a book. 500,000 books were covered in it thus far. Ninety-one percent of the covered authors have claimed their share. Tons of other ones are on the way.

And here’s your second victory fap of the episode: content providers as a group need to get together and fight for their rights in unison.

You are nuts.

Fight for their right to party?

No, fight for their right to get paid and to survive. I want you to go to ChatGPT and say, “As a group, either give us these terms or don’t index us.” They are interfering with their ability to leverage their own content. It is profoundly unfair, and those magazines and newspapers need to—

You’re gonna get steamrolled is my prediction, JCal.

What’s that?

You’re gonna get steamrolled.

It’s possible. YouTube is a great example. That’s what’s gonna happen here. There’ll be a settlement where they are going to be able to claim their content.

There will be no moaning.

I will bet anything. I will bet any amount against your premonition here, JCal. This is like the opposite of No Prono Bet.

Okay, let’s make a bet. I’m going to go with, for my biggest winner of 2024, training-data owners like The New York Times, Reddit, X, Twitter, YouTube, et cetera.

I think what we learned in 2023 was that the language models are starting to hit parity very quickly and that the real value is going to be in the training data. The models may even become commodities, and open source may win the day. So then I think the winner is folks who have the training data.

You know, the best thing about this is watching Jason’s reaction to Jason, where he’s sitting there—

Oh, my gosh. Oh, my God.

—I just love it.

Could you do a picture of me just being like, “Ooh, go JCal”?

Oh, my God.

7. Fair Use Remains Unsettled

Well, actually, this settlement—I don’t think quite proves exactly what you want it to prove, JCal.

Good.

Do you want me to explain?

No, I’m sure.

Can I make a nuance here?

Sure.

Okay.

Of course, of course.

So, okay, look, and obviously I’m not a huge fan of Anthropic because I think they’re potentially destroying the whole ecosystem for their own purposes of regulatory capture. But let’s just be very clear about what—

Come on the show anytime, Dario.

Yeah. Let’s just be very clear about what this judgment was and was not.

Okay.

What Anthropic did is they pirated all these books from LibGen and trained on them, and the reason why they got in trouble is because they basically took stolen books. They didn’t even pay for 1 copy of them, but if they had paid for just 1 copy of each book, they could not have been nailed for piracy.

They would have been potentially under fair use, which I understand, JCal, is still being litigated in the courts, but that would have been their defense.

So the reason they got nailed with this $1.5 billion judgment is that they wouldn't even buy 1 copy. It is still Anthropic's position, and it's OpenAI's position, that they should be able to train on all these books under fair use if they buy 1 copy, and that issue has not been resolved yet.

Now, you should be able to see the total hypocrisy of their point of view relative to the previous issue, which is that they believe they should be able to train on every creator's output in the world—as long as, I guess, they bought 1 copy of it—against the will of those creators, whether those creators like it or not. They believe it is fair use to train their models and derive their own weights based on fair use. However, they say that the 1 type of content that you should never be able to train on is their output. That is currently their position. It's completely hypocritical.

Actually, if you go back to Anthropic's blog post in February, where they define this concept of industrial-scale distillation attacks—for the 1st time, they coined that expression—you know, I worry that people in the government and policymakers don't understand that this is all part of an Anthropic op. No one used the terms “distillation” and “attack” together until Anthropic wrote that blog post. Distillation was simply an industry-standard practice, but then Anthropic coined this idea of industrial-scale distillation attacks.

In any event, if you go to that blog post, search for the words “IP theft.” It's not in there. Anthropic did not claim—even though they were trying to coin this new concept and brand this idea of industrial-scale distillation attacks—they did not have the chutzpah to claim that it was IP theft.

The cojones.

The cojones, the hypocrisy, the chutzpah to claim that it was IP theft. Why? Because they maintain that it is their right to train their models on all the world's output, even if the creators don't want them to.

IP for me, but not for thee.

Exactly. So, JCal, I don't even want to get into whether you're right or not on the fair-use question. Maybe you are right. I don't know, okay? But my point is about the hypocrisy, and they themselves never claimed that this was IP theft by the Chinese companies. What they tried to claim was that it was a national-security threat, because what would happen is these Chinese companies would distill off them, create their own models, and those models would not have guardrails.

Yeah.

They were making a different kind of argument.

Distraction.

That argument never found purchase with policymakers because I think that they could see that, yeah, look, guardrails are important, but it never really found purchase until Anthropic started claiming, “Oh, this is IP theft.” But they have not been willing to make that argument publicly because they know that it would poison all of their fair-use lawsuits that are happening.

Chamath, like you mentioned, The New York Times is currently suing OpenAI for basically an industrial-scale distillation attack. OpenAI went on The New York Times website, used scrapers, slurped up all of their information at a scale that no human could achieve, and then they used that as training data and reverse-engineered the model weights, effectively. So my point is that even Anthropic and OpenAI won't publicly admit that what China is doing is IP theft because they are doing it themselves. It's totally hypocritical, and I don't think policymakers should be making arguments that these companies themselves won't make because they know that they will lose all these court cases.

Friedberg, any thoughts here? Obviously, this is still being litigated, as we've discussed. There are 150 major cases. The New York Times is 1 of them, along with the music industry.

Let me ask you a question, JCal. Let's say you wrote a book—

Oh, a question for me? You have a question for me?

Let's say your book—let's call it Genuflecting: The Great American Novel, the great American novel.

Well, it could be Angel in 11 different languages. Sure.

Yeah. Whatever. Okay, so you write this book.

Thank you, HarperCollins.

And you opt not to submit it to AI because there's a restriction. You keep it closed. No one can read your book.

Yeah, like we did for Google Search. You can't be on Google Search.

No one—yeah, no one can read your book. You're not letting anyone read your book.

You have to pay for it.

You want to pay for your book if you want to read it.

Yes. $10.

And someone pays $30 for Genuflecting: The Great American Novel, and they read it, and then they write a review. They publish their review on the internet. Now, on the internet, the reviewer talks about your book and describes your book, gets web-crawled by an AI engine, and the AI engine learns from that—learns about your book. Now there's some commentary made about your book when someone asks a question about your book in the AI engine. Do you feel like your copyright was violated in that sense?

So the book—the AI never ingested your book. It ingested metadata about your book, reviews about your book, and third-party analysis about your book, all of which was on the open internet. It didn't just copy that stuff, but it used it to learn about your book.

I got it.

So you—

So you're saying these 1,500 reviews for Angel: How to Invest in Technology Startups—Timeless Advice From an Angel Investor Who Turned $100,000 into $100 Million. These reviews would then be the basis of the AI. So I guess I would have to be okay with this eloquently brash blueprint for angel investing, that verified review.

Yes.

Yes, I would be fine with that 5-star review being in there. The 1- and 2-star reviews, I don't want in there.

Okay, so now you have no knowledge of the process.

Did you plant that 5-star review, JCal?

Absolutely.

Did JCal plant that?

Absolutely.

JCal, how many of those 1,498 reviews did you not write? 3?

Yeah. No.

Okay, so—

Actually, I'll tell you the secret. When you guys actually get asked to write a book—

Uh-huh.

—or any of you have the capability of completing a book.

Yeah, because I'm 97 years old, living in the 19th century.

What AI agent did you use to—what bot did you use to post that?

AI didn't exist in 2017.

Now we know where he pointed that stupid open-source AI slop cannon that he's built.

Yeah, exactly.

Point it at the review section.

Yeah.

Yes, I verified the purchase, too.

Slop cannon.

Now, Friedberg, you make a great point. You make a—oh, there it is.

There it is.

Genuflecting

The Great American Novel. Practical Wisdom for Mastering Ambition, Building Resilience, and Winning at Life. Work of innovation.

Another master of virtue-signaling.

Yeah, another master of virtue-signaling.

Incredible. Incredible.

That is beautiful.

Incredible.

That is beautiful. But, JCal, I mean, this is my point: Knowledge can't be contained; it's diffused. The form of copyright is very clear. The case law on copyright is very clear. I cannot lift text out of your book, reprint it, and claim it as my own. That is a violation of copyright.

But my reading of your book, my reading of the reviews of your book, the diffusion of the knowledge that arises from your book—that is ultimately going to lead to some abstract transformation of knowledge into a new output that someone might read. I think it is very unlikely that we will find ourselves in a place where the idea that knowledge is transferred digitally, processed digitally, and turned into other content is going to end up violating copyright in that sense.

I understand your position. There are workarounds. Obviously, we've always had CliffsNotes, right? If a book became good enough, somebody could write the CliffsNotes of it. You can't stop that. There's a 4-part test for this. We've talked about this for 3 years here on the pod.

What I'd say is American companies should take 10% of their revenue if they're building these models, do splashy cash, and do settlements. That's exactly what's happening. If you're a copyright owner, you should study what the music industry does. They are rabid dogs, and they will fight tooth and nail and keep you in the courts until you submit and make a settlement and agree that you're licensing it. Then that gives them that case law and that settlement to go to the next person, and the next person, and the next person, and that's why they've been able to successfully defend it.

And then, if you're competing with me, that becomes the issue. So if you said, “Hey, what's his book about, and what do people think about it? What are the best parts of it?” And that comes from the reviews—okay, fine, fair enough. The problem is—and I'll share with you, there have been some other lawsuits here that are making their way through the courts—what's going to be the problem is the application layer that you talked about, Chamath. As they go into the application layer and they use this, there's a big court case here, and obviously you know about some of these.

But there are now other cases. There are music cases, and there’s a New York Times case. The one that’s kind of interesting is Thomson Reuters versus Ross. This is a final judgment on AI training copyright. There’s a company called Westlaw—they’re like LexisNexis—and people have been trying to claim that they can train on the outputs of something like Westlaw.

When you’re in the same business as me, that has a special place in copyright law because you’re infringing on my ability to use my copyright. Your argument, I think, Friedberg, holds up: it wouldn’t stop somebody from buying the book. But the second you are actually competing with me, directly, that’s when these things have problems, and that’s why I think the music industry is going to win, and some other places are going to win. But listen, this is—

Let me ask you a question, JCal.

These lawsuits are brand-new territory, and IP law does not actually have the nuance yet.

I agree.

So we’re going to, as a society, have to make a decision here on what is fair. I suggest, just like the self-regulatory group that we talked about last week, all the AI companies should get together, take 10% of your revenue, put it in a pool, and keep paying the people and getting permission from them so you can get updates on the content. So you can get the next book, the next New York Times story, the next Reuters story.

10% isn’t going to satisfy the rabid dogs. Let me tell you, they’re going to go for 100%. Now, JCal, here’s the question I want to ask you.

Yes, please.

Given that the fair use doctrine is not a decided matter yet, given that Anthropic and OpenAI are embroiled in huge lawsuits against very well-financed content creators and those communities, and the outcome is indeterminate, and there are billions of dollars at stake, maybe even their entire product is at stake, do you think that they have potentially made a fatal mistake by arguing that distilling content against the wishes of its creator is IP theft? Do you see what I’m saying?

Yes.

Is this potentially a fatal mistake? Here’s what they could have done.

Well, you would bring that to the Supreme Court and say, “Hey, listen, you already said it.”

What Anthropic could have done is they could have said, “Listen, these Chinese companies are creating fake accounts, and they’re using proxies to basically use our product in violation of our terms of service.”

Sure.

Now, using those model outputs is not IP theft because it’s fair use. However, it’s a deceptive business practice for these guys to lie about who they are when they set up accounts at scale. So they’re engaged in a deceptive business practice, and we’re going to do everything we can to stop that, but we’d like the government’s help in stopping that, too. However, we’re not saying anything about IP theft.

Yeah.

Wouldn’t that be the more nuanced approach?

Of course. Yeah.

Because I think they’re on the verge of being hoisted on their own petard here.

Yeah, I mean, it’s a self-own. I think the kids call it a self-own, right? Basically, it’s just a classic self-own. And in most of these cases, settlements happen. So again, if you just look at the music industry, the newspapers, the magazines, some of those folks, and book authors, they’ve tended to be very meek.

There’s a new trend happening now, Friedberg. A lot of content providers are saying to Google, “Take us out of the index,” because Google has one bot, and that bot does the Google crawl, and that bot also does the AI crawl. What the industry is saying is, “Hey, split that up. I want to be in Google, but I don’t want to be indexed in AI.” So people are now saying, “Hey, we’ll take you out of the index,” which Rupert Murdoch got right.

If all the newspapers said collectively, “Do not index us, Google. We’re no-index,” that would have made Google come to the table and give them a royalty and give them some money for being indexed, I believe.

They did that, dude. There was a deal that happened, but didn’t it go the way you’re describing?

Well, because they didn’t have a united front. Now I think The New York Times learned that.

No, it’s because these guys wanted Google’s user base, so they ended up doing a deal where they had—

Yeah.

—this paywall exclusion rule, whereas you could show a certain number of free articles. There was a whole negotiated settlement.

Well, I’m talking even long before that, when the first index happened. But here’s what I think: We are on the cusp of some type of settlement getting done here. I think that would be good for America to take a leadership position in that. Because you do want to keep getting that, but all of these content companies should be meeting with each other—the music industry, The New York Times, all of them—to stop their content from getting used without their permission and from competing with them. That’s my belief.

Well, you only get a settlement when both sides can agree, and it seems to me that if you’re one of these content creator lobbies and you see that Anthropic has just told the government that training on a creator’s output without their consent is IP theft—

Yeah, they’re on your side.

They have now basically confessed to their entire product being stolen. It seems to me: Why wouldn’t the content creators now assert that they’re entitled to own 100% of Anthropic’s revenue? It seems to me that this could be a bridge too far. They’re so good at regulatory capture. They’re so good at making these arguments and getting the government involved to create new regulations to protect them, but I wonder if this was just a little bit too cute.

Again, if they had just positioned it slightly differently, if they said, “Look, these Chinese companies are creating fake accounts. That’s a deceptive business practice. We’re not saying this is IP theft,” right? But instead they said IP theft, and now the whole startup community is activated. You saw that—Garry Tan and 200 startups wrote that letter. Why? Because they know that if this IP theft thing sticks, all derivative works of Chinese models are tainted now, too. So that means the whole startup ecosystem is now at risk.

I just wonder if these guys have just gone, “It’s all a bridge too far.” They’d be better off rolling it back.

Yeah.

If you go to Washington, and you lay down with the dogs, don’t be surprised if you wake up with the fleas. They decided to engage in this. So they may have poked the tiger. I think it’s pretty accurate.

Oh, by the way, the publishing schedule for 2027 was released. We have some new books coming. You heard mine, Genuflecting, coming to you now. Ferrari on My Wrist: How to Win at Life and Afford a $250,000 Watch from David Sacks.

Oh my God.

This is pre-Ozempic Sacks. We’re going to need to get that done. Here it is: The Fight to Save America: Destroying Socialism, Shrinking the Debt, and Winning AI by David Friedberg.

I like that book. I’m in.

Yeah, it’s a pretty good one.

That’s a good book. I love that.

By the way, this is our new All-In—here it is: Chamath Palihapitiya: A Sexual Italian Summer. Woo! It’s a romance novel. This is the only person who decided to do fiction. This is fiction.

Oh, fiction.

It’s fiction. It’s fiction, Chamath.

I got Chamath’s nonfiction novel also here.

Oh, you have his nonfiction as well. Here it is. Here it is: Enterprise Sales. Chamath Dystopia: How My Software Startup Fucked My Italian Summer.

There you go.

That’s great. It’s great.

Oh my God.

Oh my God. Wow.

It’s really good.

That’s coming from All-In—

The IT guy in Milan, you know.

Absolutely. This is coming from All-In Books. It’s our new publishing label coming in 2027. We’ll also have the Brad Gerstner, Bill Gurley, Elon Musk, and other titles coming. So we’ll have those on future episodes. More titles coming.

8. Google And Tesla Spend Big

Breaking topic here: Google and Tesla shared their results today. There was a lot of talk about capital expenditures. Google blew the doors off—

Blew the doors off. I mean—

It was insane—

Outrageous.

Outrageous. They were down like 7% to 10%. Google Cloud growing—

Because of their cash flow numbers.

82%—

Because of their cash flow numbers.

—year over year. Yeah, because of the cash flow numbers. And now it’s on a $100 billion run rate. That’s but one business. Tesla’s CapEx surged 140% year over year, and they expect $25 billion in CapEx. Google’s CapEx forecast is $195 billion to $205 billion this year, so next year will be even higher.

Tesla was down 14%, and Google was down 7% at our taping. Who knows? But both reported negative free cash flow. In other words, the amount of cash in the bank went down instead of up, and for Google, that was the first time ever. IPO update: SpaceX is down 30% from its day-one closing price, now trading at $1.5 trillion. A lot of pressure on the stock.

We'll talk about that as well. Obviously, they went public at $2 trillion, got a big pop—the Elon pop—and there could be more downward pressure, or it could have found a bottom. You never know with these things. This is unprecedented territory. We've never had an IPO this big, but there are some lockups.

Here's the chart. A bunch of lockups are happening at different staged intervals. So, Chamath, let's talk a little bit. Here's your SpaceX.

I guess CapEx, Chamath, is being built out obviously for AI, and there is the case of OpenAI spending on CapEx in order to provide their service. But then there's also Google, which is making these CapEx investments to resell as part of Google Cloud—an incredible product—and, on the other side, they're using it for their own infrastructure. Their business is just growing like crazy, whether it's YouTube, Google Cloud, or even Search, which is still growing.

So I looked at this and I said, "Well, this seems like a really good use of capital." Instead of just giving dividends, building out this infrastructure sounds like an investment in the future. It seems like a buy signal to me, but the market is obviously disappointed. Why is the market disappointed? Is it a buy signal for you? Does it make you more excited about management and what Sundar and Sergey are doing over there, or does it make you concerned?

I'm more bullish.

Me too.

Do you know what Google's 25-year average return on invested capital has been since going public? Take a guess, Nik.

That's right. 17%.

No.

21%.

No.

23%.

No.

29%.

Keep going.

What? Is it The Price Is Right?

35%.

32%.

Jesus Christ.

When you are a machine—a group of people and a business model—that compounds money at 32% over a 20-year average, you give these guys the benefit of the doubt. These are not people who are flying fast and loose. They're methodically investing in their edge. I go back to the first conversation: They're going to get massively rewarded.

There was a lot of Twitter chatter or X chatter about Gemini usage: Was it real or not, and was their revenue growth really coming from AI-enabled workflows? It's all malarkey. Google has an incredible search experience. They seem to be navigating this transition to use AI, and it's been done very well. They have an incredible cloud business and an incredible silicon business.

The best thing that can happen to them is that 500 different models proliferate and they support all of them, because they'll make so much money at the silicon layer and as the cloud provider. They'll find a bunch of apps, including YouTube and other things, to make money from because they use AI to target ads better or to help make better content, et cetera, et cetera.

Fragmentation's good for them.

Oh, it's great for them. It's a compounding machine. I think the reaction, by the way, is because—I saw a tweet from Ryan Petersen. I don't know if it's true, but he said Google will be spending 20% of this year's military budget in CapEx. So maybe what people are reacting to is just the scale of the investment they haven't seen.

They're free-cash-flow negative for the first time since going public, so that obviously takes people by surprise. But they're in a huge investment period, and I think it'll pay dividends even if they spend, as Friedberg's first guess was, half the number. So even if they did half the number, they'd still be overachieving massively by 100.

Still yum-yum. Friedberg, if you look at Apple, I think they bought back half their stock. They've given hundreds of billions of dollars back in profits. It seems to me that giving all this money back—buying back your shares or giving tons of dividends—is great, but it's also great that tech companies are now saying, "Wait, we have something to invest in."

The next big thing is on-demand intelligence, and there is no upper bound for intelligence. I don't think anybody—I certainly don't see anybody—saying anytime in the next 10 years, "I got enough intelligence. I've solved all the problems in the world." I think they're going to keep wanting it.

What do you think about this incredible change, with basically $100 billion or $200 billion, depending on the company, just going into CapEx?

Right.

This seems like a savvy move, right? All of us think this is a good thing.

If you want to bet against Google's deployment of capital into infrastructure because you'd rather have them give you cash for your shares today, you shouldn't own the stock. Someone else will buy it.

I think Google wins in a lot of different ways. There's just so much to Google. There's the consumer business, which is a lot of stuff. There's also YouTube. There's also GCP. There's also this portfolio of other bets, which, by the way, includes 10% of SpaceX and a good chunk of Anthropic that they own, and so on.

And Waymo's worth $120 billion.

I think they just took a $10 billion write-up on Anthropic in the quarter. So they had a $10 billion mark-to-market on Anthropic in just one quarter, and they own a piece of all these businesses. There's a lot to like about Google.

But just on GCP, I think there's probably no better-suited enterprise layer than GCP to take advantage of capturing value with AI in that enterprise setting.

Why? Why is that?

I think you're just so much better because you have so much of your enterprise data—all your email, your Drive, and a lot of information that you'd want AI to have knowledge of and access to in order to improve workplace productivity. And they're model-agnostic. You can run any model you want, and you can run any workflow you want, and you don't have to be tied in.

A lot of other cloud service providers and cloud SaaS companies are kind of model-dependent and have to run in a certain way. With Google, you can better tune your system to how you want to tune it. What's the worst, worst, worst-case scenario? The worst, worst, worst-case scenario is that they have the lowest-cost infrastructure in the world to run other people's models as a service, like Elon did with Grok and Colossus.

Exactly. Elon Web Services seems to be doing pretty great.

And look at the return Elon is making on the Colossus install. So I think if Google, in the worst-case scenario, has none of its application-layer stuff work, none of its network effects work, and doesn't have any good models, it still has the world's best infrastructure to print cash on for years if you believe in AI.

If you want to bet on AI, I think the best public-market stock to own is Google. And, by the way, you also get YouTube, the consumer business, and everything else. The multiple is kind of ridiculous right now.

We talked about this on a previous issue. Chamath, you and I were talking about what Apple should do next, and I think we both came to the conclusion: Why not have Apple Web Services? They have such great relationships with developers, they have the App Store, and they have this deep developer relationship.

Well, it's not so easy because you have to—

Of course.

To build a cloud service provider, you have to build some critical infrastructure. It's taken Amazon, let's call it, 17 years to perfect it. It's taken Google, let's call it, 12 or 13 years to mostly catch up.

But the minute that you sign up to be a web provider and a cloud service provider, what you're really signing up for is five nines of reliability and uptime, and that is just extremely expensive. Getting to the first 2 nines—99% uptime—if you're hosting something for a pharma company or a defense company, you can probably do it relatively cheaply. Getting to the third nine, 99.9%, probably costs you in the billions. Getting to the fourth nine costs tens of billions, but getting to that fifth nine costs hundreds of billions, and that takes a real investment and real technical skill. There's only 3 games in town.

Yeah, I think Tim Cook's not the guy to do it, but this new CEO might be, since he's an engineer. But they've returned in the last decade—

They could buy Anthropic.

Approximately $900 billion: $755 billion in buybacks and $140 billion, Sacks, in dividends. Just let that sink in: $900 billion. If they had invested that in Anthropic, SpaceX, and other things, they could have found some good uses for that money.

But Apple could have been more ambitious if they had spent half that money, instead of on buybacks and dividends, on creating new products and actually releasing their car, maybe buying some interesting companies. I think they would be a better company.

Who's to say, sorry, but who's to say that the investors who got that money didn't find a good use for it?

Yeah, so on a societal level. But I just think Apple could have been more ambitious if they had spent half that money, instead of on buybacks and dividends, on creating new products and actually releasing their car, maybe buying some interesting companies. I think they would be a better company.

I think it was very much a do-no-harm capital-allocation strategy, which worked for the stock. It's going to be really interesting to see if John Ternus flips the script.

I think he does. I think he's going to be engineer guy, and, yeah. All right, enough on the markets. The markets are going to do what markets do. We will talk about Iran and other stuff like that when there's more news for venture capitalists to comment on.

Right now, it's just on and on. I know everybody keeps asking. I don't think there's much for us to say on it. I do think there's a lot for us to say in Socialism Corner, our new recurring theme here, Friedberg. Every week, there's more news coming out of Socialism Corner.

Why don't you show my videos on socialism going back 6 years?

Let's add to the Friedberg socialism rants. Earlier this month, New York City dictator/mayor Zohran Mamdani hosted a rent rip-off hearing at New York City's Tenement Museum. After the hearing, he introduced a Rent Rip-Off Report. If passed, it bars landlords from charging application fees for credit checks. It's going to let landlords require a credit check or the 40x rent-to-income standard, but not both. It legally recognizes tenant unions and more. He's obviously frozen the rent for a year.

At the hearing, an activist wore a COVID mask and referred to evictions as “the violence of evictions.” Here's your 20-second clip.

Guest

The Mamdani administration is emboldening us so that we no longer tolerate the violence of evictions as a matter of business as usual.

What were we just watching? Is that the guy from Fat Albert who had the hat like that? Sacks, do you remember him? Can we pull that guy up? You remember the guy from Fat Albert who had the hat? What's his name? Oh my God.

Yeah, is COVID still happening? I thought COVID was over. Anyway.

No, she just got a booster.

That was a super-duper mask.

Yeah, I know.

That wasn't a little cloth one. That was one of these...

Whoa.

Those big ones.

There it is. What is going on?

We're going to give Friedberg a chance to do a rant? Because I could do a rant on this if you want.

No, let's give Friedberg his rant. This is Friedberg rare meat. Friedberg, what are your thoughts?

9. Private Property Protects Liberty

In 1787, John Quincy Adams published a work called A Defense of the Constitutions of Governments of the United States of America, and in that work he had a comment: “The moment the idea is admitted into society that property is not as sacred as the laws of God, and that there is not a force of law and public justice to protect it, anarchy and tyranny commence.” Then, in 1791, he made the statement publicly: “Property must be secured or liberty cannot exist,” in an essay series.

Fundamental to the foundation of the United States of America was this idea of private property rights. Because if you think about where everyone who came to America was coming from, there were these tyrannical governments, monarchies or whatever, where some overlord or some cabal could decide at any point to take the things that you have. You had no private property rights as an individual.

They could come in and say, “That farm is my farm. You're actually a serf. I'm the lord. That thing is my thing. You have a right to use it because I vest you that right to use it. I am the all-powerful, tyrannical overseer of these lands.” And it was that stasis that drove so many to come to the United States and say, “We want a place where individuals, one person, can say, ‘I own something, and no one can take it from me.’”

Private property rights are the foundations of liberty in America. This idea that you can then claim acts of violence, that you can then claim circumstances of extraordinary, extravagant wealth and say to that individual, “I now have a right—the government now has a right—to take your private property,” ultimately leads to this tyrannical form.

It starts out as being anarchic because—and remember, anarchy is a temporary state. It's always in between one state and another. All anarchies end up in tyranny. Groups of people fight each other. They're all stealing from each other. Everyone just goes and takes and gets what they want, and eventually people coalesce. They form groups, and those groups become the more powerful groups. The powerful groups end up winning, and they become the tyranny over the mass. That is why all anarchies eventually evolve into tyranny. So anarchy and tyranny are one and the same.

Fundamentally, what's going on with these socialist principles is that we are taking your private property, and you no longer have rights to your private property. Whether you are a landlord or whether you are a wealthy person that we've deemed to have too much wealth, we now will have the right to come in and take your property, control it, and take it from you.

It always starts with this framing of moralistic intent: We are good; you are bad for the following reasons. You have committed violence against the people that live in your building. You have taken too much wealth, and none of us have wealth. We have a right to go and take your wealth from you. It is always started from this point of view that you are bad.

The first framing is that the private property owner is evil, and that the private property owner has committed an act of injustice against those who don't have the private property. That is the justification for taking away their private property rights. It is the beginning of this transition toward a tyrannical system, which will ultimately be what I call this kind of great American Politburo, or whatever socialist framework gets set up by the cabal of the socialists and what they're trying to put together.

All of these little acts, while seeming ridiculous and insane and inappropriate, in aggregate are the same thing. They're a transition away from private property rights, which is the foundation of the United States of America. That's why I think we should all be so shocked.

To John Adams' point, we need to vehemently defend those rights. As soon as those rights start to slip away, even in the tiniest way, it is a cascading effect, and everything will become tyrannical. It will be very ugly in the United States of America.

Sacks, what are your thoughts here on Comrade Mamdani?

Yeah, I just want to add a layer to this idea. These DSA socialists say that evictions are violence, and they basically want to stop them. I think Friedberg's making the point that this deprives the landlord of their property, and that's true. But I think we also have to stop and consider what this means for the other residents in these buildings.

First of all, if the landlords aren't making income because they've got a bunch of delinquent tenants in the building, they can't pay for upkeep and maintenance. These buildings become more dilapidated, and that affects the other tenants.

But also, I have a friend who manages these apartment buildings, and he makes the point that it's often these squatters, these delinquent tenants who should be evicted but can't, who make the worst neighbors. When you think about the problems in an apartment complex where you've got people creating noise at night—maybe they're playing loud music—or you have people punching walls, there are disgusting smells coming from apartments, they're misusing common areas, or you have drunken and disorderly behavior, this is all the kind of stuff that happens in these rent-controlled apartment buildings.

You have to remember that there are longstanding residents, a lot of old people, who can't afford to find a new place. They depend on the rent control. When you can't evict that unruly tenant, yes, it affects the landlord, but it affects the neighbors even more because now they're stuck in a downward spiral.

I think this is a problem with the progressive mindset across the board. These DSA types are always highly educated and often affluent, and they can afford to have luxury beliefs about public spaces because they never use them, right? They don't use the bus or the subway—

Yes.

—or parks. They don't use parks. And so when they get taken over by homeless drug addicts, they always defend the addicts as opposed to the middle-class and working-class people.

An easy thing to do if you don't live in the Tenderloin.

Right, exactly. It falls the hardest on the working class because they actually need these amenities. They need the parks for their kids, or they need to use the subway, right? It's really a problem when you get people shooting up or doing drugs or defecating in a subway. It's disgusting.

Or you're walking your kids to school. The person in Marin County who has these luxury beliefs, or the person on the Upper East Side—they just don't have to deal with it. They're abstracted from this.

Right.

Yeah.

And it's no different with these rent-controlled apartments. I think that's the important point here: If, over a period of several years, you can't evict anyone, no matter how problematic they are, you effectively turn these apartment buildings into the equivalent of housing projects. That really affects decent people of modest income who have nowhere else to go, and their quality of life suffers.

And look, the private-equity wives in their gated communities will still feel good about themselves because they prevented these evictions, but it's the people in the building who are going to suffer the most.

10. Supply Solves The Housing Crisis

Well, at least—and there, Chamath, these people are not just thinking from first principles. If you want to solve the housing problem, anybody with any basic understanding of economics would just say, “Increase the supply and the price will go down.” It actually doesn’t matter which supply you add. It doesn’t matter if it’s luxury units, multifamily, or single-family.

As long as there’s more housing, and there’s transportation to get to it and to move people in and out, it’ll be fine. I’m sitting here in Tokyo. They figured this out a long time ago: just build up and put more units in. They figured it out in Texas, Florida, and Nevada. The only places that can’t seem to figure this out are New York, L.A., and San Francisco, which just happen to be liberal elite enclaves. It’s not hard on a conceptual basis. Just allow people to build more units and different types of units, and then the price will go down. Chamath, any thoughts here?

The data from Austin shows that once you relax the permitting constraint, you’ll get more units built. For every unit that comes online, it literally drives down the rent. So if you want low rent, you need to have more units. If you want more units, you need to permit more aggressively. That’s it. It’s just a decision.

With the same political will it would take Mamdani to pass this law, he could actually pass some permitting reform, and it would do a lot more good. The thing about private property—the reason I asked Friedberg to explain it—is that I really believe what he’s saying is important. It’s funny to me, this idea that, at the limit, let’s just say you’re driving your car and all of a sudden somebody jumps in it and says, “Now I’m here. You can’t kick me out.” Or you go outside and leave your door open to get a FedEx package or an Amazon box, and somebody runs in and sits on your couch. Now all of a sudden, you can’t kick them out.

If you force the owners of physical property to not be able to credit-check and differentiate whom they rent their apartments to, what’s going to happen is rents will go up even more. So I suspect that what they should do is pass this law and observe the outcome. Then you can do a pretty scientific A/B comparison between New York City and Austin, and you’ll know what works and what doesn’t work.

Yeah.

I mean, here’s the problem: the socialists never learn. We already have tons of studies.

When Milei got rid of rent control in Argentina, rents went down.

Yeah.

Yeah, I mean, this is the problem, Chamath. If the socialists ever learned from their failed experiments, you wouldn’t have Chicago. We don’t need New York to go down the tubes to know this isn’t going to work, because it’s happened in so many other places already.

Right.

But somehow, it just never seems to stop.

They never seem to learn, so they should run the experiment and learn. What’s crazy is that you’ll be learning, and it’ll be a failure on the grandest stage possible. You’re talking about the biggest, most complicated city—

New York City, yeah.

—my gosh.

You guys. Yeah, I mean, Sacks, what happens if a landlord cannot raise the rent reasonably? They have no incentive to invest in new units, and they have no incentive to upgrade existing units.

There’s this trap in New York City specifically. They have made these regulations for apartments such that if you do a renovation, it has to meet certain codes, and there are a ton of codes. That means it’s incredibly expensive. So now you have ghost apartments in New York.

Yeah, so now the housing stock becomes dilapidated. And look, they’re doing something even worse now, I think, which is banning landlords from doing credit and background checks on potential tenants. They say that you can’t look at their income, so you can’t vet whether they can actually pay the rent. So they’re banning eviction—

Who wants to be a landlord? No more landlords.

Right. So they’re banning eviction, and then they’re preventing you from doing the diligence to see if this is even a tenant who will pay the rent.

We’ve lost the script here.

So what are you supposed to do?

The interesting question is, if you were forced to live under these rules as the landlord, what would you do? The obvious answer is you’d set the rent 3 or 4 times higher, force people to sign up for a multimonth prepayment, and slowly ease those conditions until you find a market-clearing price. That’s the only way to do it.

So rents will not go down. Rents will go up. So—

Yeah.

—run the experiment and let’s just observe what happens.

That’s a really good point, Chamath. Let’s say that X% of tenants are going to become delinquent, right? And by the way, what’s their incentive to pay when they know they can’t be evicted?

Zero.

So actually, a pretty significant percentage of people could just decide, “I’m going to make rent optional.” Now the landlord has to absorb those losses.

Exactly.

And that means they have to pass on a higher rent to everybody else.

You’re going to set the rent 3 times higher, and you’re going to slowly meander it down. Like I said, you’re going to have to wire in the first full year of rent.

Okay, good luck. How is this more affordable?

It’s even worse, Sacks. Not only do landlords keep the apartments dilapidated, in some cases it’s better for them to just leave apartments—the housing stock—empty. If somebody leaves, they’re forced to renovate it, and it costs more—hundreds of thousands in renovations. So they say, “You know what? I’ll just leave it empty for now.” And so you have 50,000 ghost apartments, according to reports, in New York City.

Well, that’s an Airbnb problem.

The ghost apartments—

That’s like a—

No, they banned Airbnb in New York. You cannot get an Airbnb.

Yeah.

Yeah. Oh, really?

Yes.

That’s really interesting. Yeah, look, if you’re a landlord, one thing you would do is just sell and go to another jurisdiction. Another thing you could do is just wait this out.

Yeah.

Because it’s not profitable to run an apartment building. You can’t raise your rents, you can’t evict people, and you can’t diligence the tenants. So maybe you just leave the building empty and wait this out. That’s assuming you don’t have too much debt on it, right?

More ghost apartments, yeah.

And then you have ghost apartments.

Yeah.