AI 行业能否自我监管？Stripe 欲收购 PayPal，中国正在追赶，纽约禁建数据中心

All right, everybody, welcome back to the world's greatest podcast, the number one podcast, your favorite podcast, the All-In Podcast. I'm Jason Calacanis, the world's greatest moderator. With me, of course, Chamath Palihapitiya of-

The great-great-great-great-great-great-great-great-grandchild of Jason was a hooker and a—

You saw that? From Fred?

And a prisoner. And a purse snatcher.

Purse snatcher.

This comes from a history of France, some History Channel—

The Potato Thieves.

—said they let you out of—This was the deal in 1719, Sacks. If you were a prisoner in Paris, you were offered your freedom on the condition that you marry a prostitute and move to the great state of Louisiana. What are you saying, Sacks? Are you taking the deal?

It explains a certain one of your proclivities, J-Cal.

I thought you were asking me to see if they would extend the rule for you.

No, I'm saying that your great-great-great-great-great-great-grandmother was a hooker.

I'm not French, I'm Greek.

That's what I'm saying very explicitly. She was a hooker.

We never spent time in a prison. Also, of course, David Friedberg is here. How are you doing, brother?

Living the dream.

Yeah?

It's good to be back. Missed you guys last week. How was Brad? How did he fill in?

Brad was great. Yeah, he was great.

Trump account victory lap?

He had a little victory lap. We played “Chariots of Fire” and, “How was your special time at blank?” and “Your time next week at blank?”

Thanks for having me on your show, J-Cal.

1. The AI Regulation Proposal

All right, we’ve got a full docket today. Lots of stories. Let’s start with DeepMind’s Demis Hassabis, who just dropped an AI regulation proposal, and it’s pretty popular with the boys. In an X article, Demis called for a U.S.-led international AI standards body. It is modeled after FINRA, the Financial Industry Regulatory Authority. That’s a self-regulatory body, and this would be federally overseen but industry-funded and run by independent technological experts.

Frontier labs would submit their models 30 days before release. It would be voluntary initially, then mandatory at some point. The models would be assessed for risk to cybersecurity, national security, biological threats, and other high-risk domains. Benchmarks would be updated quarterly, and the body could coordinate a slowdown in development if the situation demanded it. I guess that would be if there was a cyber risk, et cetera.

It looks like, on the positive side, we have Elon, who said it was thoughtful; Sam at OpenAI; Jack Clark at Anthropic; Sundar; Satya; Jack Dorsey from Block; and the Collison brothers. Friedberg, your thoughts on—

Well, way to really put in the effort today. Go ahead, Jason. Good so far.

Just pass it to me, J-Cal. I’ll take care of it.

France at 6:00.

Go ahead.

Let me do my show. Go.

What do you want me to do? Oh, my God, what an incredible topic.

Give me the mic.

All right, here, let me do it. You want me to do it, right?

Some of us actually care about this topic. Let’s go.

Here’s a clip of me calling it on the All-In Podcast first. The whole industry is gonna need to be regulated, and I think the industry needs to regulate itself. That’s the key to this. We need to have a set of tests that Google, Microsoft, and Amazon all agree to. Elon says, “Hey, these are the things we should test,” and they should self-certify each model before asking the government, which doesn’t understand the models, to certify them.

The industry should have an industry certification, like they do for countless other things. I’ve talked about the MPAA and the video game industry. We should just self-certify. It’s the simplest thing in the world to do, and then we could release the models ourselves without the government getting involved. Friedberg, would you like to congratulate me on nailing it again?

Well, first of all, I thought that Demis’ proposal was really smart and thoughtful. Now that I know that you may have shared the same thought, I think we should just do something totally different.

Can’t win. I can’t win, Sacks. Even when I nail it, I hit a half-court shot, and Chamath’s like, “Move the net. Move the net.”

No.

I think it’s worth putting a little definition around this proposal, which is to form an SRO, or self-regulatory organization, because they’re not purely independent. SROs like FINRA and the National Futures Association exist in the financial markets, and they were created to allow the financial institutions to set their regulatory rules—how they check each other and how they make sure that everyone is being safe—because they’re obviously all trading risk with one another.

The industry doesn’t want to have exposure, and they certainly don’t want to have things slowed down, because that would make the markets inefficient. So the analogy with AI is pretty appropriate here: There are many players in the industry, they are all trying to progress AI technology, and no one wants to have a single regulatory body that comes in from the government or outside and says, “Here are the tests you guys have to pass with your models in order for them to be appropriate.”

As we saw in California, when California tried to pass AI legislation—I think it was about a year or a year and a half ago—none of what they wrote even made sense at the time. But fast-forward a year, and none of those rules and requirements actually map to the technology of the day.

So the purpose of an SRO like FINRA and the NFA is that they can adjust how tests are being run, who is actually running the tests, and make sure the right experts are involved in doing this—independent experts, that is—to do the testing, with federal government oversight but not control. In the case of FINRA and the NFA, they report up ultimately to a Senate committee and a House committee, which gives those committees oversight of the governing bodies that are supposed to be doing the work to make sure that they’re doing their jobs.

So the SRO concept would be that experts could be brought in from the industry who know how to assess models for things like cyber risk, bio risk, weapons risk, social manipulation, et cetera, et cetera. That independent body can get voted on and can get changed over time. Because they actually have expertise in running software evals and running tests like this, they can operate at a faster pace than setting up a new government agency.

It’s kind of a very elegant solution, and I think it’s why everyone, to your point, J-Cal—I’ll say this, right—the industry recognizes that there needs to be some degree of oversight and checkpoints here. I think that this could actually solve that problem. That’s why everyone is climbing on board with it, because it doesn’t actually hand stuff over to the government. It says, “Hey, we’re gonna get the right people to take a look at these things,” and the government is gonna have oversight ultimately.

Did Anthropic and OpenAI have a point of view?

They both signed up to it. I don’t think Dario directly, but Dario’s president gave his thumbs-up, and then I think Sam gave his thumbs-up.

Which means they’re on board. Sacks, is this the best of the possibilities in your mind? Is the industry regulating itself after they’ve now provoked governments around the world—

Right.

—to be so concerned about this issue?

Yeah, and I talked to Demis about this. This may surprise people, but I told him that I could potentially get on board with this, speaking just for myself, not on behalf of anyone in the government. I thought that an SRO, again, a self-regulatory approach, would be infinitely better than creating a new government agency that I think would rapidly become a DMV for AI.

Hmm.

Dario calls it an FAA for AI. The government does not have the expertise to evaluate AI models. The criteria are changing too rapidly. You’re gonna very rapidly end up with a queue where all the models would be waiting to get tested. It would start with a month-long delay, end up being many months, and we would just lose the AI race.

So I think an SRO approach would be infinitely better than that if it was done right. I outlined for Demis five criteria or conditions that I thought were really important in order to make this work, and if I could, I’ll just—

Yeah, walk through it.

—run through them.

Please.

All right. Number 1, I think the SRO has to have broad representation from within the AI industry. It has to include startups and open source. It can’t just be the 3 biggest labs, you know? It can’t just be—

Google.

Yeah, exactly, and that’s precisely to avoid the problem of regulatory capture, right? If you have a diverse enough group of interests being represented, it’s much harder for this to turn into regulatory capture. So, for example, I think if you had Jensen, Elon, Zuck, and maybe Mira, because she just launched a very interesting—

Open-weight model.

platform that's based on open source.

Thinking Machines, yeah.

Yes, exactly. I think that would address the regulatory-capture problem to a large degree. So, for example, if you had Jensen, Elon, Zuck, and maybe Mira, because she just launched a very interesting open-weight model platform, that's based on open source, I think that would address the regulatory-capture problem.

So that's number 1. Number 2 is that I think this body should only be reviewing frontier models, meaning the true frontier—the models that really represent an advance in the state of the art of artificial intelligence. Models below this level should not be held up from getting to market.

I do think that's a big risk under regulation: that the leaders in the market use this as a way to tie up lesser models. If a model is not at the frontier, why hold it up? That's number 2.

So they have to be maybe in the top 10 performers, top 20 performers, on the benchmark tests?

Well, no. I think when they benchmark on key dimensions of intelligence, they have to represent an increase above—

Got it.

—where the current state of the art is. If it's not a step change, then how are you dealing with some new incremental risk? This is all about dealing with some sort of incremental catastrophic risk that could be introduced by a new step change in intelligence.

That brings me to number 3, which is that I think this body should be dealing with catastrophic risks only. To my knowledge, those right now are cyber and CBRN, meaning chemical, biological, radiological, and nuclear.

It should not cover things like disinformation or microaggressions. This should not become a speech regulator, or just address things that seem trivial. The only reason to have this is for truly catastrophic risks. That's number 3.

Number 4, and Demis Hassabis mentioned this in his post, is that I think it should be voluntary first. This new organization should prove that it works before it gets legally enshrined and becomes mandatory.

Number 5 is that this should be a substitute for a new regulatory agency. If it's just additive, then it defeats the purpose, and there's no real reason to support it. This has to be a substitute, not an addition to a bunch of new regulatory structures.

I think if you did those 5 things, this becomes much more palatable. Demis said—I mean, he didn't put all these points in his blog post, but he did say to me that he thought those were good ideas. If those points were adopted, this is something that we could potentially get on board with.

That doesn't mean I don't still have concerns. I'm quite concerned, for example, that Dario Amodei has expressed support for this. However, I think this is just an opening bid for Anthropic. They'll take this—thank you very much, this is more regulation than we have today—but that won't be the end of it.

This will just be the stepping stone to get what Dario has now called for many times, which is the FAA for AI. If I could, let me just, as a final point, explain what the FAA does, because people need to understand that FAA for AI sounds really nice, but it's actually a really extreme proposal.

What the FAA does, among other things, is approve new airplane designs. Specifically, it requires what's called a type certification for any new aircraft design or major changes. For an entirely new aircraft design, it takes 5 to 9 years to get the certification.

If you merely want to amend a certificate for major changes—and I'm not even sure how major the changes need to be—it takes 3 to 5 years. The Boeing 737 MAX, for example, took about 5 years.

This is permission-based regulation. There's no approval, no flying commercially. It's safety first. That might make sense in the case of preventing plane crashes, but when you're talking about AI models, you're talking about replacing a system that is releasing new versions every couple of months with one that is potentially fully under the control of the government, fully government-approved.

Everything has to be certified, and you could expect the timeline to go from months to years. I think we'll simply lose the AI race if that happens, because China isn't going to abide by those rules.

To sum up, if my choices are between FAA for AI or what I would call the DMV for AI, I would much rather go for Demis Hassabis's SRO for AI—the self-regulatory approach. But we really have to keep it honest and pure because otherwise it'll just be the opening bid in a coming new wave of regulation, and it will be the vehicle for massive regulatory capture and—

And it can't, to your point, restrict open source.

—and slow down innovation. Yeah.

It can't restrict open source. I think that's so important because all of these other efforts require money that you have to spend, which is always where regulatory capture happens, and you have to enable startups and open source to compete effectively.

Chamath, any thoughts on this new self-governing body?

I think it's really important, and I hope it happens quickly. The thing we have to keep in mind is that there's going to be a torrent of money that's going to try to influence both sides of the political aisle to regulate this in a way that creates some form of regulatory capture. We just don't know what that form will take.

The faster we avoid that off-ramp by actually establishing a set of rules and superseding the need for federal oversight, the better. At the end of the day, you still have federal oversight in some ways, because you still have Commerce that plays a huge role in these standards, and you still have the DOJ. So it's not as if it's going to be the Wild West.

What it prevents is a handful of actors using their balance sheets and their capital to essentially pull the ladder up. I think if that happens, we're in a really bad place. Demis's proposal makes a ton of sense, and we should just get on with it.

All right.

Well, yes, provided that we make sure it's pure. In my view, there are 5 conditions, but we do have to make sure it's pure.

Even in the FINRA example—the analogy that Demis used was that we should set this up in the same way that FINRA is set up—FINRA ultimately reports to the government. It reports to the SEC. If we're going to set up this new SRO, where is it going to report in the government? There's going to be a huge food fight over that, and it will then be subject to political pressure.

The software industry has never been regulated in that way. We do not have a dedicated regulator for software.

No, I think you're right. Those issues are important.

Yeah.

But my point is, Sacks, I'm just saying these things are never ideal. But if the choices are that we kill open source and pull the ladder up on the entire market so there's a duopoly, or there's this, I'd say this.

For sure. And that's—

Yeah.

—and that's the crux of my argument. It's definitely the lesser of 2 evils. I'm not sure those are the only 2 choices, but increasingly, there's no question that the pressure is coming to regulate AI more and more. Frankly, this all goes back to Anthropic's government-relations effort.

They poked the tiger. They poked the tiger.

Yeah. Well, it's more than that.

They're funding the fear of the tiger.

They're funding it. Yeah. I want to give an update on that, actually, because—

I would say they're poking the tiger of the American public getting really freaked out and then the government stepping in.

Totally.

Yeah.

Totally. There are a couple of data points on that, actually. I want to give a quick update.

2. The Anthropic Capture Strategy

In October of last year, I tweeted that Anthropic was running a sophisticated regulatory-capture strategy based on fearmongering, and everyone went crazy over this. This was a very hot take, or spicy take, at the time. Back then, people thought that I was beating up on a little startup.

Now I think everyone can see the truth, which is that this is not a little startup. They already have a trillion dollar market cap valuation. Gavin Baker thinks it'll be at 3 trillion after the IPO. This is actually one of the biggest of the big tech companies, and by pretty much every criterion, including revenue, they are the leading AI company.

People can see now that they have enormous resources, and they're putting those resources behind an effort to, as Chamath said, pull up the ladder. It's classic regulatory capture.

There was an article in Politico just the other day called “Inside Anthropic’s State-by-State Plan to Ratchet Up AI Rules.” It says, quote, “AI giant Anthropic is pursuing a strategy of one-upmanship that encourages states to impose increasingly tougher AI guardrails rather than align around a single set of regulations.”

The basic idea is that they get a set of regulations passed in 1 state, like California's SB 53, and that was then supposed to be the model, at least for all the blue states. But with each new state, they actually make the regulations more and more strict, more and more all-encompassing, so there's not actually a stable equilibrium.

What they're trying to do is drive each incremental state toward more and more regulation. This article explains that they're doing the opposite of trying to create what we wanted, which was a single national framework. They actually want the patchwork because they're using the pressure they're creating at the state level to impose more and more regulations.

And again, what I said last year was that Anthropic was principally responsible for the state regulatory frenzy that is damaging the startup ecosystem. Again, everyone went crazy at the time. I think now there's plenty of evidence showing that this is their agenda.

And by the way, they're going to win that because states have great sovereignty rights. Like we're seeing with self-driving, the states are going to decide. It's not going to be a federal mandate. The states get to decide the nature of the U.S.—

And so they're going to win on that in a couple of states, right, Sacks? Just realistically.

Well, they have. They've won in a bunch of states. They've already won in California, Illinois, and New York. They're winning in all the blue states and maybe even some red states. But look, ultimately, the reason why Anthropic's arguments are finding purchase is because when you go to the government and say, "Please regulate me. You should have more power," there's hardly anyone in government—

Yeah.

—who will ever say, "Oh, no, no, no. We're not qualified. We don't want the power."

We're for less government.

Yeah. There are very few people who are principled that way, and most people in the government will say, "Thank you very much. What else can we take?" This is the mistake that I think a lot of people in the tech industry are making: They think that they can just buy off politicians or the political system by making concessions. No, that will just lead to a ratcheting up of the pressure. The government will be happy to take this and then come back for more and more and more until it's fully under government control.

So at some point, I think these companies are going to have to grow a spine and fight and decide where they're willing to draw a line. And if Demis's SRO is the line, if they're saying, "Okay, we think this is the right solution, and we're going to fight here, and this has to be it. In exchange for this, we need preemption, and we need other things written into law that make sure this is where the line is," then I think it can work. But I think if you're just offering it up for free, and all these companies are just going to say, "Oh, yeah, regulate. Give us the SRO," that will not be the end of it. That will just be the opening bid, and the government will come back to take more and more and more.

All right, let's keep moving through the docket here. Stripe, which is still a private company, much to the chagrin of many of the shareholders, I think, now as they go into their second day.

On the SOT.

On the—yes. I mean, I'm in a couple of funds that have large positions. Go public, boys. They are bidding $53 billion for your alma mater, David Sacks—PayPal—which is a public company. Stripe and the private equity fund Advent are jointly offering to acquire PayPal for about $60 a share, which is a small premium. Most people think it'll go for more like $70 a share. PayPal stock jumped on the news, obviously—

Were you able to get to the bottom of this? Was it just Stripe and Advent? Because then—

No, we're about to get into it.

Somebody else reported that it was also Block.

Yes. Block is coming in as well.

It's so confusing. Every other media source is all over the place on this.

Yeah. No, I think it's because it was a breaking story, and maybe they were trying to keep it quiet.

It's a huge deal, though, if Block is a part of it versus if they're not, I think.

Yes. Block, formerly known as Square, is Jack Dorsey's payment company, one of the few entrepreneurs to ever create 2 decacorns in our industry. They're contributing $17 billion in equity to the combined offer. How they chop up what's inside PayPal would be the big question. Obviously, they own a number of different brands, including Venmo in addition to PayPal. That might go really well with the—I'm just taking a guess here—with the Block assets.

Stripe owns Bridge, their stablecoin infrastructure company, which they acquired for $1 billion in 2025. PayPal has PYUSD, which is already in circulation. That's their stablecoin, so stablecoins are part of this. But the biggest thing is PayPal is still a juggernaut, Sacks: 439 million consumer accounts. You did something right there 25 years ago. It has stood the test of time.

I know. It's amazing. It really is.

Isn't it amazing that it's still that strong? It's weird when brands keep going for that long.

But the problem is that the product is getting very long in the tooth. I think it's only growing 7% a year, which is a lot on the base that it's grown to.

It's a big base. Yeah.

It's a big base, but the product has become somewhat obsolete, and in a way, it's a legacy product. I'd be curious to hear from the Stripe guys how they would fix that, because I think that's a very hard problem to fix. Maybe they wouldn't. Maybe they would just run it more efficiently and milk it for all it's worth. I think there's a different question.

Kind of do a private equity play.

I think the different question—

What's the work each mean?

I think there's going to be more of these kinds of deals. If you look at Ryan Cohen's bid for eBay, I think it's probably a second dot on a line that I think is emerging: AI-native folks are looking at first-generation digital-native businesses that have become mature, old, and stale; aren't run by the founders anymore; have not yet realized the opportunities with AI; have not yet realized their potential; and are overspending in a lot of ways.

And when you take a look at those businesses as a modern-day AI operator, you're like, "What the hell? This thing is so underutilized." They're not using their network well, they're not operating well, they're overspending, they're not using AI well, and there's a set of opportunities that become quite obvious.

I think the capital markets, as we've seen with Josh Kushner's roll-up of accounting firms, and with a project like this from General Catalyst, can use capital to go buy traditional services businesses and AI-ify them. I think this is part of a line of maybe looking at traditional digital businesses and AI-ifying them, and there's a long list of these. There's a couple dozen of them.

So I think if you looked at the public markets and said, "Hey, where are all these software companies and network businesses that emerged in the early part of the internet, or even in the more recent part of the internet, aren't run by their founders anymore, and have stalled out?" there's a massive opportunity.

Now, the question as a capital provider is: Who do you partner with to go and execute that operational revival of that business? You're not going to go hire some McKinsey consultant to do that work for you. It's got to be the best of the best. It's got to be the right players in the business.

So I think Ryan Cohen has proved his mettle, obviously, with some of the things that he's done with Chewy and GameStop, and that's obviously debatable. I spent some time interviewing him to understand his processes.

This was on the All In interview program. You can go to our channel and find it there. It's last month.

Thanks for the plug.

Yeah.

And obviously, when it comes to payments, who better than Stripe?

Yeah.

And maybe Jack Dorsey plays a role here. And by the way, if you think about that $17 billion equity contribution, what that technically means is Stripe is selling and Block is selling $17 billion of equity to the cash investors. That cash is then going to buy PayPal. Therefore, Stripe and Block end up owning a piece of PayPal.

The private equity investors own a piece of Block and Stripe, and what is not clear in the deal documents that were published, because I don't think it's relevant to the public markets, is who's actually going to operate PayPal post-close.

And my bet would be that they're going to hand it over to the Stripe guys and say, “You guys go for it.”

Yeah, I think that's clear. They're the most qualified, and they will have the biggest stake in it.

So I will make a prediction. I think that eBay and PayPal are probably the beginning of a wave of mega-deals of, call it, flaccid digital businesses—

Okay.

—that can be revived with the blue chew of capital and the right operator.

You have Chamath's attention. Go ahead.

—that can be revived—

Okay.

—and I think that there's probably a big wave of this to come.

Again, Rowe?

The other thing is this is probably not the final clearing price. I think the price is probably another 10% to 15% higher from here, and I will say that there's a certain individual who must look very closely at putting in a competitive bid.

Oh, a certain individual who may have his fingerprints on the original PayPal—

Correct.

—who might also have $4 trillion or $5 trillion in market cap to play with, who also made a $60 billion acquisition recently.

Correct.

We don't have inside information here. To your point, Friedberg, this is becoming a playbook. There's a company called Bending Spoons that just went public. They bought a bunch of non-founder-led assets: AOL for $1.4 billion, Vimeo for $1.4 billion, WeTransfer, Eventbrite, Brightcove, and Evernote.

Chamath, have you spent time with him?

I was just DMing with him.

No, no.

He's awesome. I've hung out with this guy. This guy is an absolute frigging operational killer. He-

In Italy.

He bought Evernote.

Yeah, in Milan.

And, yeah, he runs the whole thing from Milan. He bought Evernote, and he just goes in and diagnoses these businesses. He's like, “Where are you overspending? Where are you underspending? What are you doing wrong with the product, and what are you doing wrong with marketing?”

And he just frigging fixes it, and he's just a killer. He's taken all of these businesses that were called Web 2.0 businesses, revitalized them, rolled them up, and printed cash out of them. And he uses leverage—

—and lowers the cost to run them.

—and lowers the cost. He's used leverage to buy these things.

That's the other big thing. He's using young, AI-first executives, from what I'm told.

Totally. It's a great call-out, J-Cal. Bending Spoons is the roll-up of this sort of strategy, but for these mega-deals, I think there's more of them to come.

I will say the higher-order bit here, which we talked about for a couple of years, was that M&A was on the ropes for a couple of years under the wrath of Lina Khan. Then, once Trump got elected, all the executives working in corporate development said, “Hey, looks like M&A is back on the menu.”

Now we're seeing deal after deal after deal get consummated. People are no longer scared of doing deals. Uber just bought Delivery Hero today. That's going to, like, jump their revenue by 20—

$15 billion.

Yeah, and that's going to jump their revenue by—like, they're getting diluted 10%, and it's going to jump their revenue 24% or something crazy like that. This is going to be, I think, the big story of the next couple of years.

And all this liquidity—talking to LPs and family offices, which I do on a regular basis, they're all like, “Hey, when's your next fund? Hey, when's the next deal?” Because now people are believing in venture because of the SpaceX distributions and all this M&A. We have 4 or 5 companies that have been bought since Donald Trump was elected president.

Thank you to my president, Donald J. Trump, for putting M&A back on the menu, Sacks. M&A back on the menu. Why don't you make a bid, Sacks, for PayPal?

Been there, done that.

Been there, done it. Okay, yeah.

No, but look, you have to have synergies.

There was a moment—this was like 15 years ago—where they asked Sacks to go back and be the CEO—

That was right. I remember that. At the poker game. Yeah.

—and he and I immediately flew to Vegas and spent the weekend there to think about it.

They're like, “You know what?”

Well, no, I never got the offer, but it was down to, like, the final 2 or something, and it was between me and someone else. Actually, they ended up going with some traditional credit-card executive.

To be honest, that's why PayPal has stagnated. As soon as it was acquired back in 2002, they basically blew out all the founders, all the founding DNA, and it was just run by consulting types.

You've got to remember, at that time, it was acquired by eBay, and Meg Whitman had worked at Procter & Gamble and Disney, and she'd spent about 8 years at Bain. It was a very corporate mindset. Among all the internet companies of that era, it was definitely the most corporatist.

They saw the founders—the founding generation at PayPal—as just a problem, just a bunch of cowboys they couldn't control. They made no effort to retain them, and I think they were relieved when they all left. And that's what created the PayPal Mafia.

Normally, in an acquisition, you'd lock up all the talent. But in this case—

They locked them out. They're like, “These guys are too hard to manage. Change the keys.”

I've said for a long time it's a misnomer to call it the PayPal Mafia. It's really the PayPal diaspora. Our homeland was taken over, they burned our temple, and then kicked everybody out.

Yeah.

Totally.

And that's why the whole PayPal Mafia got started with all those companies.

Yeah, you guys are real victims.

I'm not saying anyone's a victim.

No, I'm joking. I'm joking.

Look, when you acquire a company, you get to decide—

Yes.

—what to do with that asset.

Totally.

That was just the reality. But it's not like—I don't think anyone was bitter about it. They were all like, “Okay, this gives us the capital to go do the next thing we want to do.”

Well, the new CEO, by the way, Enrique, is really aces. I've met him before, and apparently they're doing a great job, which is why they probably got these offers, because they've been tightening that business up for the last couple of years.

Well, no, the reason they got these offers is that the market cap was down to, what, like, $30-something billion? I mean, this is a company that was worth $200 billion, wasn't it, roughly?

$322 billion, I think, was the peak.

Mm-hmm.

And before this offer, it was down to what, $30 billion to $40 billion?

Yep.

So the reason why it's attracting offers is that it's so beaten up. And so now the question is, can anyone else do something with it?

Sacks, to your comment that you've got to have synergies, doesn't it seem to be the case that in this era, the core synergy that any great operator can bring to the table in this sort of scenario is AI?

You can leverage, whether it be in this business or others, tools that drive automation, product development, improvements, and efficiencies across the organization—tools that make the product actually better for the user, et cetera—that simply aren't being well implemented.

Potentially.

I mean—

Well, look, I think you have to have a product vision of how you would use AI to make the whole user experience better. And yes, you're right that you could just use AI to drive efficiencies, and that'll improve your profitability and earnings. On a financial level, you could make the acquisition work.

But it seems to me the existential issue for PayPal is that you're dealing with a product that's 25 years old. It's the same thing that we created back, like, 27 years ago. It's changed a little bit, but not that much.

The problem is that the interaction model is legacy. So unless you've got a vision of how to resuscitate and rejuvenate that product, I think it could be a good financial play, maybe. But—

I think they're buying the accounts.

Yeah.

Yeah.

400 million users.

I mean, what you're saying, Chamath, is interesting because with Stripe, this is the advantage that Stripe has: They have a ton of merchants, right? They've become the preferred mechanism for merchants to basically accept payments via APIs.

I think they're doing about $2 trillion a year in annual transaction volume. I think PayPal is doing $1.7 trillion. So actually, Stripe is a little bigger than PayPal now.

But the thing that PayPal has that Stripe doesn't really have is the consumer relationship: over 400 million active consumer accounts.

So you're right, Chamath, that if somehow you could combine the merchant relationships with all those consumer accounts and bypass the credit card networks, there could be a lot more on-us transactions.

Exactly. PayPal already owns Braintree, so now you have Stripe and Braintree, which effectively were competitors that won't be anymore. And then what Block gives you is an entire point-of-sale infrastructure, and you get Cash App. So you put it all together, and I think it's a shot across the bow for Visa and Mastercard.

That's where the value is.

PayPal already owns Braintree, so now you have Stripe and Braintree, which effectively were competitors that won't be anymore. And then what Block gives you is an entire point-of-sale infrastructure, and you get Cash App. So you put it all together, and I think it's a shot across the bow for Visa and Mastercard.

Braintree is the other one that you just mentioned there, Chamath. That's important because that is a very strong business that people don't even know PayPal owns.

Huge business.

Venmo also speaks to a lot of young people. So you're kind of getting two generations. You're getting Gen X and millennials, a lot of bang for your buck there.

But what Sacks said is true. They have enough of these things to go end to end on their own rails. That is a very important opportunity.

Yeah. The question is whether you can package it all together in a way that the consumer will actually choose, because it's one thing to say, "We take the merchant relationships of Stripe and the consumer relationships of PayPal and we put them together."

No, you don't do that.

But what if the consumer doesn't want that?

No, you don't do that. I think what you do is you go to places like all of the merchants that use Stripe and say, "We'll give you a 3, 4, or 5% discount." And they'll be like, "Okay."

Yeah.

And so you'll see these prices just fall everywhere. Imagine if Shopify said to all their merchants, "Okay, you have two choices: the old way or the new way. The new way, you put another 2, 3, or 4% in your pocket." Of course they're going to pick the new way.

Well, this is the paradox of modern M&A. If you look at protecting the consumer, this will ultimately be great for consumers. This is going to lower the prices. They're not buying this to—

Well, okay, you're saying something really interesting. It is so good for consumers if this were to happen. This is the exact reason why, if this had happened 2 years ago, this would have been the antitrust equivalent of a colorectal exam. You would not even get one step close to doing this deal 2 years ago.

Oh, God.

Well, that's really interesting, actually. The key question with antitrust is, how do you define the market? If you define the market as APIs for merchants, then, JCal, it would be Stripe versus Braintree, and then the government would say, "You can't consolidate that share."

No, no, I know. Yeah.

Right.

If the real market is Visa and Mastercard, that's the ultimate duopoly. And if PayPal can add competition to that market, which is infinitely larger than APIs, then it's actually pro-competitive. So how you define the market determines whether it's anticompetitive or pro-competitive.

Exactly.

It's the game.

Exactly. And those guys are smart enough, and they've read enough books, that they won't screw this one up.

3. Apple Sues OpenAI

All right, let's get to the next topic. Somebody else is suing OpenAI. This time, it's Apple. On July 10, Apple filed a 41-page lawsuit against OpenAI over alleged stolen trade secrets. Apple says OpenAI stole its IP to develop its consumer hardware device.

Remember, we had Sara Fryer at Liquidity, and I probed her on this new device. She said it was very human and lovable. She gave us a little bit of the goods. Well, it turns out Apple is alleging that maybe this is partially its IP.

Tang Tan, Apple's former VP of iPhone design, is OpenAI's chief hardware officer. He allegedly directed Apple job candidates interviewing at OpenAI to bring, quote, "actual parts" to interviews to, quote, "show and tell" in the interviews. Chang Liu, a former Apple senior technical engineer, sent this text message to a still-employed Apple colleague: "LOL. I found out I can access the network storage. So funny."

During all of this, OpenAI has poached over 400 Apple employees over the last year or two. That's a big number of poaching. Apple and Tim Cook have seen enough, Chamath. Tim Cook greenlit this. As insane as this is, Sam Altman has found a way—found a way—to screw yet another party. He screwed Elon, his first benefactor.

Chamath, if you remember correctly, the default for iPhone AI was supposed to be ChatGPT. So Sam took this relationship where he got to be the default on the most important platform for AI, the iPhone, and now it's wound up in a massive lawsuit. What are your thoughts on this?

I haven't really seen Apple act very litigiously in 25 years in Silicon Valley, so that's obviously a concerning data point for OpenAI. They're very reactive more than they are proactive on these things, so there must have been something that really, really upset them.

So I don't know. It's going to take a court to sort this out. I don't really want to gossip because who knows what's actually going on and who said what and blah, blah, blah. But nobody should be stealing things from their former employer. Nobody.

Obvious.

You're just not allowed. It's just obvious. These people are very, very smart and they're very successful, and that's why OpenAI probably wanted them, and that's why they wanted them. You come to them with the collective wisdom of what you've accumulated, and I think that's sufficient. You don't need to, especially as a senior person, do this.

So I just hope that this stuff isn't true, because I don't think that Sam or Sarah or anybody else there is trying to induce this to happen. I don't think so.

Yeah. I doubt they induced it, but I do believe that it's true, or Apple wouldn't have brought it. Sacks, when we look at this, maybe you could open the aperture here if you want, or you can just go very detailed. The nature of the fact that we have a free market is that we don't have noncompetes in California, generally speaking. You can just—employment at will—go where you want.

But we have had instances. Waymo famously brought some IP to Uber. When they did, Travis and the team said, "Leave the building. Your job is rescinded. You don't get to bring that information here." In this case, it seems maybe they didn't induce it, but it occurred for some period of time. So take us through, big picture, what you think is going on here and what it means for the industry.

Well, like Chamath said, I have no idea what's going on here. This is a lawsuit. The facts are all alleged. We don't know. It's going to be adjudicated, so I really don't want to opine on what happened here.

But if people want to know, a very simple rule of thumb for how to avoid these types of disputes is: when an employee leaves their previous company and joins the new company, just don't take anything with you. The only thing you can bring to your new job is what's in your head. That's it.

Your memories.

Your memories. That's fine. Whatever's in your head, you're allowed to take, but never leave with anything else.

Never.

No thumb drives.

No thumb drives.

Never.

Ever.

No CD-ROMs.

No thumb drives.

No documents. No nothing. Just—

Zip, zilch, nada.

What's in your head is okay, but that's it.

Friedberg, any thoughts here on just the number of lawsuits that seem to be piling up over at OpenAI? Bad luck?

A couple of dots make a line, I guess.

Okay, there you go. Very well said. A couple of dots make a line.

4. The SpaceX Data Leak

Okay, xAI had a data leak this week. They launched Grok Build in public beta at the end of May. The newest coding model, Grok 4.5, powers Grok Build. I've been playing with it. It's extraordinary. It's a coding tool that works inside of Cursor.

SpaceX previously told users, "Nothing from your code base is transmitted to SpaceX servers during a session." But what actually was happening is that every time a developer used Grok Build, according to reports, the tool was sending their entire code base to SpaceX cloud servers without alerting users. Not just the files that were needed for that specific coding task—just everything. Passwords and API keys could have been pulled up there, along with all the change logs, et cetera.

The privacy setting was supposed to stop this, but it didn't work. SpaceX quietly disabled the upload on July 13 by flipping a switch on its servers. Elon Musk, friend of the pod, promised on X that all previously uploaded data had been deleted, I guess. In response, SpaceX open-sourced Grok Build. That's their harness, so that's another open-source win, or a win for the open-source community and AI sovereignty.

Have you got any kind of thoughts on this? Obviously, this was not intentional, but trust is important with these models, as we've been talking about for the last couple of months here on the All-In Podcast.

I would actually connect this to my comments on CNBC earlier this week, which built on top of Alex Karp's comments the week before.

Privacy in AI is very fragile and brittle, and this is despite the best efforts of great businesses. You may not like Elon for his personality quirks, but he is incredibly trustworthy. He's overly transparent, and so, to their credit, they shut it off immediately.

But my takeaway is that there are all kinds of non-obvious data-leak vectors lurking in AI. If you think that you're going to flip a ZDR switch—zero data retention, which is the magic term that the industry uses to tell you that everything's going to be okay—I think the answer and the message should be, “It's not going to be okay,” because you can't guarantee any of it.

So the model companies, when they give you these zero-data-retention policies, are probably trying their best. But I think the reality is you are leaking information where you don't know it, and they, despite their best efforts, may still have trapdoors that they don't even know about until somebody else figures them out, like in this example.

All of this speaks to the fact that you have to have a stratified ecosystem. You have to have third parties. Now, look, that's very biased for me because it's in part what we do for large enterprises at 8090 [?] when we implement our software factory. But the reason why it's working so well is this exact reason: You need an independent third-party layer to interface with these models and manage this exposure, because there are trapdoors everywhere.

And that's what Sacha just said in a really interesting blog post. Did you guys see that?

Yeah.

Yeah.

I thought that was excellent.

“The Reverse Information Paradox.” That's exactly the takeaway that he left with. He was building on Alex Karp's supposed crash-out—the point that Karp made about how enterprises that have technical ability want control over their compute, models, weights, data, and alpha.

But he went further with that idea. He started with Karp's idea, but then he provided a recipe, a roadmap, for how enterprises should operationalize that. What he says is that enterprises have to establish a real trust boundary with private evals, proprietary learning loops inside the tenant, decoupled orchestration, and the explicit right to fine-tune their own output.

So he goes through a litany of fairly technical things that enterprises should do in order to achieve the operational control that Karp was saying enterprises really want over their AI compute, models, data, and alpha. It's really interesting. I think now there's a virtual, almost college-dorm-session-like discussion going on between the leaders of these companies, who are brainstorming some of these concepts and now extending them.

Mm-hmm.

What's happening is you're starting to see the formation of not really an alliance, but an ecosystem that is trying to create alternatives to a monolithic, closed-model stack. That's where Anthropic and, to some extent, OpenAI want to go: They want you to be locked into their stack—their models, their harness. They control the data, all of that—and now you're starting to see all these different companies—

5. The Token Cost Reckoning

And you pay a huge premium for the privilege of having them do it, which is even more insane. I saw this data, and Nick, maybe you can find this companion clip. The companion clip I'd like you to find is Eric Gleiman, the CEO of Ramp, who was on Squawk Box, I think today, talking about a new feature where you can manage the token spending of your employees through your Ramp card.

But the data that I saw was that a million tokens on Lovable is about $56. That sucks. A million from Bolt is about $26, which is the same as Claude 4.8. A million input tokens on Grok is about $1.50. Zuck's is about $1.50. Elon's is about a dollar, and the Chinese models are 50 cents.

Okay.

On top of the whole data-sovereignty issue—

Bleeding your alpha away.

Can you imagine that you're paying $56 per million input tokens as well? For that risk? That is insanity.

Bro—

I'm using Perplexity Computer, and they started supporting Groq. They already support GLM-5.2. When you're using Claude or OpenAI, you can only use their models.

So I started messing with the different models, and I gave them all the same basic PRD. I said, “I want to make a podcast player that deep-links.” If we were talking about, I don't know, Mythos, it would play me all the Mythos clips across all the different tech and business podcasts, but make them into one stream.

I thought this would be really helpful for me for prepping for the show and would just be interesting. I did it. It took a couple of hours, and it cost $11 on the new Groq. It was hilarious how cheap it was.

Sure.

I'm sorry, did you try to do it on Fable to see how much more expensive it would've been?

I didn't, because I was out of Fable credits on my $200 account. So—

Well, look at this clip here. Nick, play the clip from Eric Glyman. That's kind of interesting.

Eric Gleiman

We're thrilled to be launching Token Spend Management today. It's available to Ramp and non-Ramp customers.

Smart.

Eric Glyman

And he's exactly right. Over the last year—I looked at the stats this morning—token spend among Ramp customers has grown by 21 times.

21 times. Not 21%.

Eric Glyman

21 times.

We're talking about 21 times.

Eric Glyman

That's exactly right. So being off by a few pennies as a CFO actually might be quite nice. At the rate it's going, it might be several dollars.

And look, I think that for many CFOs, they're often very surprised by the bill, because what the AI companies have functionally set up is that you have a tab. You can spend as much as you want. It's very hard for CFOs to see proactively what people are spending on, and every time they introduce new models, the rates often go up.

Right.

Eric Glyman

So the incentives are very misaligned. Part of what we're trying to do is make it easy for CFOs to see the spend, understand the spend, and control it.

Thanks, Nick. He's saying something so important there, because if your engineers are going off randomly in an unguided system and then just ripping through a million tokens at $56, what he's talking about is the eventual downstream impact on earnings.

Eventually, a bunch of these public-market CFOs are going to show up to Wall Street and they will have missed earnings because their OpEx, at some point, if things are 21x-ing every few months, will cause somebody to miss a quarter. I don't know who, but somebody.

I was speculating it would be a few pennies here or there, which they'd have to say was because of token spend. He's saying it could be as much as dollars at this rate, which also could be the case. I think the point that we're all trying to make is, unless you get control of this and you can directly say how much money you're making, this is a bridge to nowhere. It is a money-burning furnace.

The good news is this is all creating a massive market opportunity, Sacks. Bittensor subnets, GLM-5.2 hosting, Grok 4.5—

Inkling. Inkling, Mira Murati's new model.

Mira's Inkling. Everybody's now saying, “Hey, wait a second. I can give you a better deal. You're paying, like, $2? I can get you $1.”

No, no. People are paying between $26 and $56. They should be paying 50 cents.

Exactly.

The Inkling announcement was interesting, because I think the value proposition there is that she's explicitly saying, “Look, we're not frontier intelligence. We're just under that. But we're a platform for fine-tuning these open models, which are much, much cheaper, and then you can achieve the result you want based on fine-tuning.”

Right.

And so that's really interesting. But these open models won't be around for very long if Anthropic has its way. That's where it's all going.

There's a reason they want to stop it. They have such a monopoly—

Of course. You're selling most of the product for 50 cents per million tokens when they're selling theirs for $56. Of course you don't want that to happen. Of course you want to try to stop it.

But that being said, they're still growing like crazy, just to be clear. You are seeing this explosion of interesting things happening with open models. Like you said, the latest Grok Build is open, Thinking Machines is open, and so forth and so on.

Nimo is coming.

They're growing. They're still the industry leader in terms of revenue growth, so these things are happening side by side.

Yeah. I think the interesting thing is Eric would not have released this Ramp product unless CFOs were saying, “I can't control the spend.”

Yes.

And then he's saying, “Well, here, let me build it for you.” If enough CFOs essentially turn that feature on and start to rate-limit how it's spent because maybe they're not getting the ROI, the engineer doesn't care about ROI. The engineer's saying, “I want to use the latest, greatest model.”

Yeah, make me better.

And maybe you don't need it.

Maybe Mira’s right: for 95% of the tasks, you should be at one level lower, especially when it costs 1/100th the cost. But the engineer will never make that trade-off because they’ll never—

Totally.

—want to think about it.

You’re right.

And also, they’re not tied to the money. The CFO is tied to the money, and the engineer wants to explore using the latest, greatest thing.

And until those two things have a reckoning—

If you’re booking your travel, you don’t even see the price. You’re like, “Yeah, just put me in business class. Put me in a nice hotel.” The travel department handles that.

You’re saying something really interesting. What percentage, if you had to guess, of Fable five prompts are just average meshugana that should be running on—

98%.

Yeah.

98%.

Yeah.

I was using it for stupid stuff that I could be using Qwen for.

Yeah.

6. AI Moves To The Edge

This is just my micro-prediction here: Mark Gurman, who’s the most in-the-know guy when it comes to Apple, says we have this new CEO coming in for Apple.

John Furness.

Yeah. He’s a hardware engineer. M7 Ultra, because we’re on M5 chips now. You can get 256 or 512 gigabytes of RAM. He says M7 Ultra is going to support as much as 1.5 terabytes. That’s double what they’re already supporting.

So if you think about frontier models, like the last generation, this is an Opus-level model running on your Mac Studio. You guys all use Mac Studios. You’re rich venture capitalists, whatever. You’re like, “Yeah, I’ll take a $4,000 or $5,000 computer.” This is going to change everything.

You’re going to have employers go, “Oh, I can just run 90% of my workloads—99% of the workloads—on the local Mac Studio?” I think Apple is a screaming buy right now. That’s not financial advice, but my Lord, that company could just run the table on AI if they get this right. All right.

Apple?

Yes, because they’re going to make such a fortune—

You’ve got to just let them move on, Sacks. You’ve got to just let them move on.

No, no, I’ll tell you. Let me explain. It’s just like the iPhone. Everybody laughed at the iPhone.

No, they didn’t.

When the first iPhone came out, many people laughed at it.

Not true.

Steve Jobs did.

I know who did.

That was it.

That’s true.

Not true.

Steve Jobs was the big one. You could hear—I can still hear him laughing. No, if you think about how they make money off hardware, off their devices, they will put so much downward pressure on Claude and OpenAI by just putting local models on their devices and supporting them with this memory architecture. It’s going to be wild when people have unlimited tokens on their desktops. I’m telling you.

I don’t know if you saw this, but there’s a very large solar company called Sunrun. They just announced this week that they’re making distributed data center blocks that you can put in your house. Another company that did it is called Span, which partnered with NVIDIA.

So, to your point, Jason, you’re seeing this fragmentation and distribution of edge compute, which I think is a theme. Definitely a theme.

Well, it’s also chasing energy, right, Chamath? If you’ve got some solar, if you’ve got excess battery power, hey, we power up your batteries at night cheaply.

7. The Electricity Shortage

I think I told you this last week: We are so massively short electrons. By 2050, the United States of America will be 2.5 Californias’ worth of energy in deficit. 2.5 Californias.

California is the fourth-largest economy in the world. We will be short 2.5 times all of the energy consumed by California by 2050. This week, there was an auction by this huge utility called PGM, which serves Pennsylvania, New Jersey, Maryland, and 13 states. That auction is where they publish a forward curve and say, “Hey, listen, guys, here’s my forecasted load, and here’s how much energy I need.”

People sign up to essentially get paid a guaranteed rate every day so that they have to fork over the energy in the future, kind of like a forward option. They needed 7 or 8 gigawatts. They had 156 megawatts or something show up. We are in such a bad place right now on electrons and electricity prices.

Did you see what our boy did this week?

We need—we need—this is behind the meter, which is different. Elon needed to do this, by the way, just so you know, because there’s an issue in Memphis where he was very clever about how he was able to get Colossus off the ground.

Explain this.

When you try to power a data center, typically you have what’s called grid power. You go to the utility in the area and say, “Hey, please run me a line off of that main transmission line,” and that’s how you power your data center.

When that runs out or is so backlogged, you have to do what’s called behind the meter, which means on your own property that you own, you build something for yourself. Now, there’s a problem with that. You would think, “Well, that’s smart.” Yes, but like everything in America, there’s regulation on top of regulation on top of regulation.

One of the most complicated regulatory schemes that you have to overcome is clean-air permitting. So even if you say you’re going to do behind the meter, then you’re like, “Well, what can I do?” Solar you can do, but it takes too much space for most places. Batteries you can do, but you need to generate the electricity in the first place, so people use natural gas.

So Elon cleverly bought a ton of 18-wheeler engines, basically.

He bought the company that makes all this and brought them to—

Mobile turbines.

Yeah, mobile turbines.

Yeah, the turbines.

And then just pinned them to the ground and ran them. Those are for personal use, essentially, and so they came under the clean-air permitting requirements. But then when you act as a block, you could make the claim that it doesn’t.

Now, there are new solutions, like Bloom Energy, which allows you to have huge installations and still fall under the personal-use clean-air permit. For all of Elon’s future capacity, he needed to have this in place so that he gets the clean-air permits and is able to have a clean line of sight to continue to build domestic data centers.

Anyway, there’s your little TED Talk on energy, but we are in a bad place, guys, and it’s only getting worse.

8. New York Bans Datacenters

Speaking of data centers, Sacks, everybody’s favorite socialist governor, Kathy Hochul, in the great state of New York, my hometown—

Kathy Hochul

Powered by fossil fuels, they drive up our carbon footprint. They occupy massive amounts of land, potentially displacing agricultural space and open spaces. The bottom line is progress shouldn’t arrive with a higher utility bill, depleted water supply, or noise pollution. So we have no choice but to address these challenges created by these massive facilities. That is why—

She’s so dumb.

Rupert Darwall

—today I’ll be signing the nation’s first-ever statewide moratorium on hyperscale data centers.

Everything she’s saying there is a false accusation, so let’s just go one by one. She’s saying that they eat up all of the power. Well, look, if you connect to the grid without producing more power and you force data centers to compete with residential ratepayers, then, yes, you could drive up utility prices.

However, if you do what Chamath said and let them build behind the meter, then they bring their own power. That’s what the president has advocated for since the beginning of his administration: Let the AI companies become power companies. So that is the way to solve the energy problem, or the utility problem.

Then she’s talking about eating up land. The reality is these data centers are a model of land-use efficiency. We have a ton of land in this country, obviously. You can find places where there is enough open land to build a data center. The economic impact and value of a data center relative to the land use, again, is one of the best ROIs there is.

The supposed noise pollution is largely made up. That can be dealt with. You obviously don’t want to put these things right next to a residential area, but create a little bit of distance and it’s fine. The whole water-consumption thing is—

A hoax.

—largely a hoax. Modern data centers recirculate the water.

Closed-loop systems.

Yeah, and I think there was a study that showed that a typical data center uses the same amount of water as 2.5 In-N-Out Burger chains.

I mean, just go after the almonds if you’re concerned about water, people. Please.

There are many uses of water that are way more wasteful, like golf courses.

Yeah.

I mean, there are many uses of water that are way more wasteful.

So when you compare economic impact to all these different things, data centers are, honestly, one of the best things we could be building as a nation, but—

And taxes. There are all these taxes and incremental revenues. Did you see the article where, I think it was in North Dakota or something, teachers were getting $30,000 and $40,000 bonuses from all the tax revenue that was coming in? There are all these upsides.

That's right. They generate a lot of tax revenue. They've created a blue-collar construction boom. It's not true that there are no jobs once they're built; you do have ongoing jobs there. And one final thing, just on the point that Hochul is making: she said it created a lot of pollution. Natural gas, which is how most of these data centers are powered, is one of the most clean-burning sources of power that we have.

100%.

These data centers have become the scapegoat for all the angst that people have about AI, and it's become this very clumsy way of trying to throw a wrench in the gears of innovation and just slow the whole thing down.

Well, all I have to say is, welcome to Texas. We've got plenty of land.

For now.

And what's so stupid—

But yeah.

—about her proposal and her talk, aside from the things she got completely factually incorrect, is that New York State is 80% underdeveloped. You drive upstate, folks. You're thinking of New York City. Yes, New York City's packed. You go upstate; 70% to 80% of the land in New York State is undeveloped. There's so much land, it's ridiculous. New York is giant.

Well—

It's giant.

By the way, on this topic, this week I just want to give a shout-out to Senator Dave McCormick. He had a defense and innovation summit in Carlisle, Pennsylvania, at the Army War College, which a bunch of us went to. POTUS came, gave a speech, and had a CEO roundtable. A lot of defense company CEOs, et cetera. But Chris Wright was there, Sacks, and—

My guy.

He's great. And Chris mentioned this insane story. He said, “You know, there are a lot of common funding patterns among these people who are protesting the data centers,” because Dina Powell asked this question on stage. And he said, “You can actually trace it back to the same people who, in a different era, were protesting fracking.” So he was saying these are all just hobbyhorses that they use to raise money and have a job. They're professionally paid protesters. They just show up out of nowhere. I didn't realize there was such a commonality, but they're the same people.

The thing that I just can't understand, for the life of me, is why Anthropic is still funding these groups that want to put the kibosh on new data center construction. There's one called Public First, where Dario just gave his first seven-figure contribution, and then a bunch of other employees at Anthropic gave to it. All these groups are trying to slow down AI development with new regulations and by making it harder to build new data centers. At a certain point, you just have to wonder: is this regulatory capture, or have they just lost the plot?

The number one thing slowing down the growth of Anthropic's revenue isn't demand. I think it's the availability of compute in data centers. So you're just wondering, what is the point of all of this?

It's so true.

I was talking to someone in politics about this, and the theory they had was, well, the Democrats aren't going to pause the data centers forever. They're going to pause them until they feel like they're in enough control that they can dictate all the rules. In other words, they're calling this a moratorium, and I think it does mean that the data centers are going to stop, but eventually they're going to be in a position to say, “Okay, here are our terms if you want to turn these things back on,” right? “You want to lift the moratorium.”

And then that's when we get this big-government, Democrat-defined AI regime. You know that it's going to consist of a new regulatory agency, new speech controls—the whole trust-and-safety agenda from social media will be ported over. This was what this one person I was talking to was speculating is the real agenda: eventually, once Trump is no longer president, or in some future Democratic administration, they will eventually lift the moratorium, but on their terms.

Now, I think that's a really dangerous thing to do because Trump is president for another 2 years, and then no one knows what's going to happen after that. And even if you lift the moratorium in, say, 2.5 or 3 years, it's going to take a couple of years for those projects to even ramp back up. So when you start talking about a moratorium on data centers, it's not like a few-month pause. It's probably a good 5 years at least before you can get another data center switched on in the state of New York.

Just so you know how bad it's gotten, there's a curve that you can use to price data center assets, and I think you guys know this, but I have this portfolio of these assets that my partner Nita and I have accumulated. What's so interesting is when we talk to all of the hyperscalers about giving us a price, because we're trying to figure out whether we should keep it, build it, or just sell it, the most incredible thing is how extreme the price is at the front end of the curve when you have verifiable, energizable power today.

And the reason is exactly everything that you're saying, Sacks: when you look out into the future, we've said this before, but about 40% of all these projects are getting mothballed and stopped. So it's creating this massive deficit of available energy to actually drive the use of AI.

To the extent that you actually want drug discovery, cancer diagnoses, better healthcare, or better legal advice, we may actually not be able to service it based on all of the demand that exists because the power isn't there, the energy isn't there. And the reason why it's not there is because folks are just reflexively protesting something that they clearly don't completely understand. So I think it's a really big problem.

We're going to have GPUs chasing energy. Where's their energy? Just drive the GPUs there. That's what's going to happen, right?

Let me add one layer to it: they're not only trying to stop data centers from being built in the US, they're trying to stop data centers from being built internationally, in our friends, allies, and partner countries. And the way they're doing that is that the same political forces that are stopping data centers are also behind all these new export controls on chips. So they want to make it—

Yeah.

—harder and harder to export chips to more and more countries, including our friends and allies. So there's not going to be data centers here. There's not going to be data centers in our allies. I mean, where are we going to put these things?

Yeah, and some of those allies—

I'm surprised—

—have unlimited energy: the Middle East. If you want some data centers—

Well, what's funny, Jason, is we did a bunch of Middle East data center stuff, and then it stopped, meaning there wasn't this growth that I thought would happen because it's a very conveniently placed geography. It's the Middle East for a reason, and so—

Yeah.

—you can serve 4 billion people very quickly, in under 200 milliseconds, from there. Instead, what happened was there was this explosion in Asia, and specifically in Australia, which surprised me because I would've thought those folks are a little bit even further out on the DSA—far left. I thought these things would not have happened, but they were able to get big deals done.

So in this weird way, you have all of these other countries running to try to embrace this stuff quickly. They've done a decent job. They're doing stuff to displace some of the energy that's needed in the US. But the problem is we need to have enough surplus here because this is where most of the commerce is going to get created, which I think should power the—

These are luxury regulations. You can afford, if you're New York State or California, to be like, “You know what? We don't need this. It's a luxury for us to have an extra data center.”

But I think the fact that—

If you're Australia, you might really need the money. If you're Texas—

Well, yeah.

—you might really want the money. Nebraska might want it.

So this is what's so crazy: virtue signaling only goes so far until your debt-to-GDP is high enough and/or your productivity is low enough and/or your foreign direct investment is low enough, where you're like—

Yeah.

“All right. You know what? Screw all that. We're just going to build a data center.”

And by the way, I don't know if you saw the—

You know, even if you just think about fiber and the milliseconds you're talking about, yes, you can get to 4 billion people. But I don't know if you saw the giant Starlink versions now. They make a really big version. I think, Sacks, you got one of the enterprise versions. But there's an even bigger enterprise version, and they can bundle them together, and you're starting to get to 10-gig, 20-gig setups. So that means you could start putting these things almost anywhere.

Yeah.

Which also adds another wrinkle.

Rupert, can we see your clip, the thing that you were mentioning before?

Rupert Darwall

This is not as prevalent in the Middle East, where you have monarchies and governments that aren't ruled by democracy. But in democracies, we see this anti-data-center movement taking hold. This chart is something that, for me, always played a role in my understanding of where the incredible anti-GMO sentiment came about in the United States.

This is great.

Rupert Darwall

Russia Today, this Russian media outlet, launched in the US in 2010. They were kicked out of the US by Biden in 2022. You can see that prior to Russia Today existing in the US, there was no anti-GMO sentiment. GMOs had been around since 1996. That's when they first had their big commercial launch in the US, and they were pretty prevalent for 14-plus years before everyone started to think, “GMOs are bad. We’ve got to get rid of GMOs.”

You could ask people 100 different ways, very pointedly and specifically, about the facts on the matter and the science of GMOs and all this sort of stuff, but everyone always had a reason why they didn't want them, similar to what we're hearing now with AI and data centers.

It turns out that if you track back all the media that had all this anti-GMO sentiment, which ultimately got picked up by the mom bloggers, put into social media feeds, and ultimately accepted by everyone as truth, a lot of it originated in this Russian media push that happened around this era. You can actually see this on the Google Trends data that shows GMO and its kind of rise-up.

As Russia Today started to get cut by different media outlets, and people stopped retweeting them, stopped reflecting them, and stopped writing articles that followed Russia Today, the anti-GMO sentiment declined in the US.

I think you can see this going back decades. There's this effort that the KGB designed during the Cold War called Directed Measures, which was really meant to try to create an influence campaign by affecting media—putting this kind of propaganda out through foreign media, particularly targeting Western democracies.

You could argue that maybe you could trace back what happened in Germany with nuclear energy as being similarly originated, but there have been a series of these pushes that seem nonsensical if you're fairly rational and can have an objective debate about the scientific merit, the economic merit, and the benefits of these technologies.

For some reason, what we call the activist community becomes heightened to them, says that we've got to get rid of them, and everyone has these different, scientifically unfounded reasons why they want to get rid of them. And you're like, “Wait a second, how did we end up in this place where we're literally handicapping ourselves?” I think we're seeing something similar happening with data centers in the US today.

The funding of the NGOs, as they're being called, the media that's supporting this, the retweeting of the media—and then you ask people: There was a poll that came out today, and something north of 50% of Americans believe that data centers increase the cost of water and electricity. Even if the data center is fully recycling the water and producing its own electricity, there's still this kind of repugnant reaction to the data center.

There has been this deeply sown psychological shift that's happened in the United States, and people have these attitudes: “I hate the rich, I hate tech, I hate AI, I don't want any of this stuff.” But where does it all come from? I do worry that there's some degree of foreign—

Influence.

Rupert Darwall

I don't love the word “influence” because everyone captures it.

Well, for sure.

Rupert Darwall

But there is some degree of this—

Interference. Yeah.

Rupert Darwall

Yeah, I would say there's foreign interest. Let's call it that.

No, I think it's more than that.

Rupert Darwall

There's foreign interest, psyops.

9. China Targets AI Debate

Just 1 month ago, OpenAI published a blog post called “PRC-Linked Influence Operations Are Targeting AI Debates in the US,” and Politico covered this, and a lot of other sites covered this. Basically, what they are saying—and in fact, many people are saying—is that China is behind a lot of these influence campaigns to shape US attitudes on AI data centers.

It makes sense. It makes a lot of sense for China to do this.

Yeah, and there's going to be a congressional investigation of this. It does make sense because it is in their interest, right? If they can stop us from building this necessary infrastructure, then that's a way for China to win the AI race.

If they can convolute the market, if they can incentivize Anthropic to pull the ladder up, if they can kill open source in the United States and constrain demand or the optionality and choice of lower, cheaper models, think about that for a second. At $56 per million input tokens versus 50 cents for the rest of the world, all of a sudden it doesn't take a company that's much worse than you to beat you when your cost is 50 to 100 times more.

Right.

That's just the math.

The math ain't nothing.

You know, Sacks made the point that these enterprises are not just paying for AI with money. They're paying again by feeding those frontier models their proprietary knowledge and all their alpha. So it's like a double whammy.

Double whammy.

It's more expensive, and you're potentially mortgaging your future.

Look, let's be honest. It is obvious where foreign governments have an enormous incentive to try to manipulate and influence the comings and goings in America. I think we should just acknowledge that. The idea that that doesn't happen is very naive.

The question is, we have to be able to call it out and put our finger on it, because otherwise what is clearly happening is that a lot of Americans will just fall for this, and they will not think from first principles.

We have a huge moral panic going on with respect to AI. When you talk about catastrophes that could result from AI, what are we talking about? We're talking about things that might happen in the future. Nothing resembling this has happened yet.

Job loss.

Or job loss.

Job loss.

None of it's turned out to be true. We haven't seen any of it so far, but we're on the threshold, I think, of destroying the crown jewel of our economy, which is the system of free-market innovation that we have—this culture of rapid iteration, where anyone with a good idea can go raise risk capital and start their idea, start their company.

Now we're talking—I mean, think about how far the Overton window has moved. We're actually saying that creating a FINRA for our industry might be better than all the alternatives. FINRA is a bunch of stockbrokers writing rules. When's the last time there was ever any innovation in that sector? I guess Robinhood made trading free. That was it, right?

Yeah, that was a big one.

That was it.

It hit the order flow.

Okay, but that's not real innovation. That's an innovation with respect to a pricing model, and we're actually saying that that might be the least-bad alternative: having the equivalent of a bunch of stockbrokers creating new rules that all these AI companies are now going to have to abide by.

It's crazy.

We are going to throw away the lead that we have in this. And by the way, Kimi K3 just came out, and people are saying it's now right up there. It's very, very close to the frontier. We may have months on China, if that, and we're going to create all these crazy rules and new regulatory bodies for risks that have not manifested yet.

It's worth monitoring the situation, but it's not worth panicking. You should monitor the situation with self-driving cars and job loss. China is certainly doing that.

They just stopped giving out permits for self-driving cars, as an example, because it's going so well and they're losing jobs. There are people who are getting—there's a little civil unrest—so they just said, “We're going to make self-driving cars licensed.” They're not giving out any more licenses: a moratorium on licenses for now.

It's worth watching Mythos, and if it could hack your system, Palo Alto Networks is checking it out, and other people are checking it. It's all worth monitoring, but there is no disaster here today because of AI.

Nothing's jumping out of your ChatGPT window. The worst-case scenario is you blow out some tokens. Okay, great.

It's crazy. There's only a handful of companies that are even at the frontier, and they all have safety testing and red teaming and all the rest of it.

Yeah, they're doing a good job.

Yeah.

They're doing a good job.

I'm not saying stop that. I'm just questioning whether we need some vast regulatory apparatus now to start doing all this rulemaking.

It's certainly premature, and we did this because of science fiction and Dario saying all jobs are going away. That was the most ridiculous thing. When he said it, he seemed panicked that it was going to be 80% or 90% of jobs in 2026. What was his claim? Nick, get the exact claim. I think he said 50% of jobs, right?

No, he said, “50% of entry-level knowledge worker jobs are going away within 1 to 5 years.” That was 1 year ago, so—

It's a little ridiculous.

He's been in a state of panic since GPT-2.

People don't even know how to use the tools yet.

Remember, they wanted to have regulatory approval for models that use 10²⁵ FLOPS, right? Every single AI model is well past that threshold now.

Yeah.

And we haven't seen any of the harm. Look, they thought that 10²⁵ FLOPS would be enough compute to create the Terminator, to create Skynet.

No offense, Friedberg, but one guy's panic attacks, one guy's anxiety condition, might have shaped the whole course of history here. Does Dario have an anxiety issue where he's overly concerned about this stuff, or is it just delusions of grandeur? Come on the pod, Dario. The invite's open. Come hang out with the boys.

I'm sure he'd love to come on the pod after you just accused him of having a panic attack.

He seems like he's in a perpetual one.

No, let me tell you.

Am I wrong?

Listen, it could be psychological, but I actually think that there's a strategy that makes a lot of sense, and it's a very simple, straightforward strategy. Number 1: brand yourself as a safe AI company. Number 2: ban unsafe AI. Number 3: profit.

Yeah, there you go.

That's the strategy. Kind of brilliant.

All right, everybody, go to allin.com/events and sign up for the All-In Summit in September. Scholarships are open. Let's do a quick, amazing, deep, robust science corner with our boy, David Friedberg.

Before we get into the science corner, I'm gonna give a shout-out to Ronnie, dog for adoption. I Love Family Dog Rescue in Sonoma. Check out his Instagram link in the description. This dog needs a home. He was fostered, and he lost the foster home. Someone come and grab him. He's awesome.

God, here he goes again.

All right, let's get into science corner.

This is what we're doing now?

It's on your side.

What? This is what we're doing? To try to get more IQ points?

That dog looks delicious.

Gross.

Tucker, fuck you.

You don't live in Sri Lanka anymore, Chamath.

Yeah, seriously.

Oh, Sri Lanka taking a stray. Oh, my God.

Yeah.

Oh, there you go.

How do you marinate that dog in Sri Lanka?

Do what you gotta do in Sri Lanka.

Is it just salt and pepper, Friedberg, or do you like something else? Just a little salt and pepper?

12-hour marinade? You use Beaujolais Nouveau? What is it?

Do you like a little yogurt and garam masala? Maybe do a little tikka dog?

Okay. Okay, spicy.

Little tikka masala?

Do you guys want to talk about reversing aging?

Yes, I want to talk about it, but I gotta drop. All right, guys. See you later.

I gotta go too.

Bye.

I gotta go to the Eiffel Tower, pal. Pal, pal, fuckers.

I'll cover it. The audience will stick around. Science corner. All right, so Chamath, you can drop too if you want. I'll cover science corner solo.

10. The Aging Enzyme Breakthrough

In the past, we've talked about Yamanaka factors, which are these proteins that can go into cells and reverse the aging of the cell, and the cell starts to act young again. Pretty amazing, and there's a lot of advancement happening on that front. But this paper that came out just this week, which everyone's kind of going crazy about, was put out jointly by Calico, Google's super-secretive age-reversal startup, in partnership with a group called Reval Pharma.

What they focused on was what's called the extracellular matrix, the parts outside of the cell that age. What does aging actually look like outside of the cell? Over time, sugars and fats bind to proteins in the area between our cells, and they accumulate and don't get cleaned off. As they accumulate and don't get cleaned off, they make it harder for your body to clean out and maintain that area. It causes stickiness and binding, which reduces mobility and ultimately leads to things like wrinkles in our skin and makes it harder for our joints to move.

Is that visceral fat? Is that what visceral fat is?

No, it's called glycation, and it's the binding of sugar and fat to the proteins that sit in that extracellular—

In between the cells.

Yeah.

In between the cells, exactly. It's that whole gunky area in between the cells that, when you're young, works well. Everything's smooth. The proteins get replaced if they break down. As you get older, sugars and fats stick to these proteins, block them up, and as they get blocked up, your body can't repair them or clean them.

More importantly, it changes the structure and the shape of those proteins, so things like collagen that are far apart stick together. That causes things like wrinkles and immobility. It also causes inflammation, because those proteins look different than they're supposed to, and your body starts to attack them, which activates inflammation. That's one of the reasons why we get more and more inflammation as we get older.

One of the key advanced glycation end products—that's the term for these things—is called CML. CML is the predominant molecule that gets formed in this extracellular matrix and drives aging, and nothing breaks it down. These scientists set out to try to create an enzyme—an enzyme is a protein that breaks something down—that can break down CML.

Remember, a protein is just a series of amino acids, and those amino acids are programmed by DNA. You can use 3 letters of DNA to make an amino acid. You can literally print DNA, put it in a bacterium to print proteins, and then test those proteins to see what they do. That's the modern era of protein synthesis and protein testing.

These guys took the target, which is CML, and tried to figure out how to degrade CML, clear that extracellular matrix, and reverse aging. They started with AlphaFold and used AlphaFold to find a protein that could bind to CML and activate an enzymatic process that would break it down.

They took that protein from AlphaFold, which comes from a bacterium, produced it, and started to test it. Then they started to find some of the binders, or the parts of that protein, that they could make better. They used DNA programming to change it, and they made hundreds and then thousands of variants to measure activity—which is how good each one was at breaking down the CML. They did this recursively through 5 different cycles.

Eventually, once they had gotten it breaking down the CML really well in a test tube, they started to test it on the proteins that we find in our bodies: casein, collagen, retinal proteins, which are in your eye, and hemoglobin. They were able to get rid of 52% to 97% of the CML, just degrading it away. They found several sites where they were able to degrade over 90%.

Then they took actual human skin from patients over 70 who had donated their skin, and they put this enzyme onto that skin. They were able to eliminate 55% of the CML on the skin, which basically reversed the skin's age down to that of a 31-year-old. This was from patients older than 70, just by putting this enzyme on the skin.

It's a groundbreaking demonstration of a combination of AlphaFold and what's called directed evolution, where you change the order of the DNA, which changes the structure of the protein, to test different proteins, do high-throughput screening, and ultimately make a novel protein that doesn't exist in nature today that can do something pretty profound for human health.

The next set of questions is, well, great, this enzyme is awesome. How are we going to get it into our bodies? How are we going to get it into that extracellular matrix? Is it going to be a cream, a shot, or a supplement? Could we eventually take an RNA shot that makes the protein inside our bodies and starts to do the degradation from within? A lot of questions still need to be answered, but it really, I think, lights a great path forward.

Amazing

For these novel therapies that we're developing, and it's fucking awesome.

Amazing.

I got all these joint pains in my hip and my shoulder now. Everything—you can feel yourself getting older.

Well, I'll tell you this right now: that will not be the first market. The first market will be cosmetic.

Cosmetic skin, yeah.

It will be a $1 trillion market if you can create a cream.

If you could put this enzyme literally on your skin and have it absorb—

On your face as a cream.

Yeah.

Game over. That alone is $2 trillion.

Unbelievable. But AI—let's just talk about applications of AI and why it's actually awesome. Everyone should be able to agree on that, and you can't be convinced by some foreign psyop. This is fucking awesome. This was AlphaFold—

It's amazing.

—used to discover this thing, evolve it, and drive this outcome. Everyone can benefit from it. It's just so profound that we have this tool at our disposal in this day and age. I think it's pretty awesome. Anyway, thanks for sticking around for Science Corner.

Guys, I love you.

All right, bro. Love you too.

speaker_6

Let your winners ride.

And instead, we open source it to the fans, and they've just gone crazy with it.

Love you, Bessie. Ice Queen of Kin Wah.

speaker_6

Let your winners ride. Let, let your winners ride.

Besties are back.

speaker_6

Gold thirteen.

That is my, uh, dog taking a piss in your driveway, Sacks. Wait, no, no, it's-

speaker_6

Oh, man.

My habitasher will meet me at Blinks.

We should all just get a room and just have one big huge orgy because they're all just useless. It's like this, like, sexual tension that we just need to release somehow.

Wet your feet. Wet your feet.

speaker_6

Wet your feet.

We need to get merch. Besties are back. I'm going all in.