Brian Armstrong谈Bitcoin、Anthropic发布Fable 5与Mythos 5、NewLimit获4.35亿美元推进逆龄 | 264

Citibank projects a Bitcoin price reaching as much as $189,000 by the end of 2026.

Bitcoin is the new digital gold. I think it’s going to be a key part of our economy going forward into the future.

I remember it was supposed to be countercyclical to problems on the planet. Wars break out, Bitcoin is supposed to go up. The stock market crashes, Bitcoin is supposed to go up. Is that still valid?

The reason that Bitcoin has been down—everyone’s been trying to figure that out.

Anthropic launches Fable 5 and Mythos 5.

Alex Wissner-Gross

Very impressive release. Anthropic is back in the lead now.

Alex is constantly posting on our internal feed.

Alex Wissner-Gross

It’s happening. It’s happening. I mean, it’s really happening now. This is—oh my God.

While we have Brian, congratulations on your recent raise at NewLimit—$435 million to push us toward age reversal.

At NewLimit, we’re trying to do half of what Shinya Yamanaka did. We don’t want to change the type of the cell; we just want to change the age. We’ve actually been able to demonstrate successful reprogramming of human cells. Our first drug candidates are going into the clinic next year, hopefully followed by a bunch more.

When do we hit LEV?

Okay, so that’s a complex topic, I think.

I’m here with my Moonshot mates. Professor of Exponential Organizations, Salim, where on the planet are you today?

If I look out the window, I’m in Vegas. I just spoke at a ZK Accelerate crypto cyber event because they’re freaking out in their world. I saw the Sphere for the first time, which is the most incredible thing.

Amazing. And, of course, A.W.G., our resident triple-major genius. Alex, how about you?

Alex Wissner-Gross

I’m at my family beach house on Long Island, doing deep research, no doubt.

The work doesn’t stop; however, the orchids do.

And Dave Blendon, our resident wizard of AI investing. Today, it’s a real pleasure to have a friend, a brilliant and prolific Brian Armstrong, CEO of Coinbase and co-founder of NewLimit. I’m Peter Diamandis, your host and provocateur of all things abundance.

Brian, where are you today?

I’m in the Coinbase office in San Francisco.

Nice. Well, thanks for joining us. There’s a lot of news, and I’m looking forward to getting your take on it. We have a packed program today. We’re going to kick it off with a deep dive into Bitcoin and agents. At the close of the show, we’re going to dive into longevity and epigenetic reprogramming.

In the interim, we’re going to cover Trump’s appetite for Frontier Lab equity, OpenAI’s IPO, Mythos 5, and Elon’s announcement of his AI1 ginormous satellites. Our promise: no politics, no doom, just the science, tech, and investments accelerating us toward a singularity, hopefully making this the most exciting time ever for you guys to be alive.

Before we jump into our first story, we actually have intro music. Brian, we have an incredible fun community of creators and they typically give us outro songs. This time we’ve got an intro song called Moonshots intro. Let’s take a listen.

I think this could be the greatest accelerator for human knowledge yet. It’s a moonshot.

Well, look, if you want to transform the world, you have to go out into the world.

Cut cost dramatically. You can automate. You can expand your market share. The race is on.

Now, the valuable thing is curation and attention. When you’re in the midst of a singularity, breakthroughs that are happening essentially every week feel prosaic.

And that’s a moonshot, ladies and gentlemen.

And thank you to Francis Coin. All right, let’s jump into our first story for the day. It’s about Bitcoin. Citibank projects a Bitcoin price reaching as much as $189,000 by the end of 2026. Brian, I loved your quote: “Bitcoin is going to do great and is as important as ever.”

Here’s what I find fascinating: it’s an institutional flip. Five years ago, the largest banks were calling Bitcoin rat poison, and now Citi is publishing prices that look like Apple stock. Brian, do you want to give us a little bit of an overview of what you’re seeing? We may have just bottomed out at $60,000. What are you thinking about Bitcoin these days?

It’s never as good as it seems and never as bad as it seems with these things. You’ve got to take a long-term perspective.

Over the last few quarters, the reason Bitcoin has been down—everyone’s been trying to figure that out. I think it’s a few things. AI absorbed a lot of the risk capital. It’s hard to imagine people talking about anything else at the moment besides AI, so that capital has moved over.

Then people got really excited about stablecoins because the GENIUS Act passed and we had some regulatory clarity. Stablecoins became the new meta, and maybe Bitcoin, with its inflation-type trade that people have done in the past, got a little less exciting in this environment, where it felt like, “Okay, maybe we’re going to grow our way out of this and the inflation won’t be so bad.”

I think these are all short-term effects. Bitcoin is the new digital gold. I think it’s going to be a key part of our economy going forward into the future, so I’m as bullish as ever.

I think sometimes people look at these four-year cycles that Bitcoin goes through, where you can see the percentage of people who have made money versus those who haven’t. My instinct is that we’ve probably bottomed at this point, maybe at the $60,000 number, but nobody can say for sure, of course.

I’m optimistic as always. I think by 2030 we’re going to have a much higher price, and I’m long Bitcoin, just like always.

I bought my first Bitcoin in 2014 on Coinbase, so I’m a longstanding user. Brian, I love the thing. I think you guys should get so much credit for creating the first interface that made it easy to navigate, and you’ve really continued since then.

It’s clear that Bitcoin becomes the digital collateral for an AI-native economy. If agents are going to transact autonomously, they probably aren’t going to be using checking accounts at JPMorgan. They’re going to need programmable interfaces and programmable settlement. This is where I think the future is, so I’m super excited about that.

Brian, can I ask you a question? I have a substantial amount of my personal net worth in Bitcoin, and I’m holding it. I remember that, in the earliest days, it was supposed to be countercyclical to problems on the planet. Wars break out, Bitcoin is supposed to go up. The stock market crashes, Bitcoin is supposed to go up. What are you seeing there? Was that the promise? Is that still valid?

I think that thesis will still play out. It just took longer than most people, probably including myself, thought.

I already think of Bitcoin as digital gold. It’s the thing you might hold in times of uncertainty, just like you would with gold in the past. Maybe 30% of the capital treats it like that. There are people who buy Bitcoin when they’re worried about inflation of the dollar and things like that.

But I think 70% of it is still people treating it like a risk asset, like they would a higher-volatility tech stock or something like that. Those ratios will shift over time, and eventually the thesis you mentioned will play out, but it’ll take more time.

Salim, just to react to something you said, I think you mentioned the agentic economy. I totally agree. Stablecoin payments will probably be the default layer for the agentic economy.

Right now, a lot of people are just interfacing with one AI agent to get information back. But increasingly, we’ll be interfacing with an AI agent that’s actually orchestrating hundreds or thousands of other AI agents.

There will be this economy of capital and labor, and a lot of payments—almost like AI-agent payroll will have to happen. Somebody was telling me today at lunch, “Are you a monotheist or a polytheist on AI?”

The monotheists would say there’s going to be one superintelligence—AGI—to rule them all, and it’ll have so much brainpower that it can do everything itself in the world.

The polytheists would say that each of these models, even as they get smarter and smarter—which they will—are going to have limited context. There might be models that specialize in going really deep into a coding problem, some scientific breakthrough, some design, or maybe even manufacturing things with humanoid robots.

There will be lots of specialists, and they’ll actually have to communicate with each other using language, like human beings. If there are 10 agents working on something, they’ll give their status updates, and the information will go up a layer. That layer will consolidate up to another orchestrator.

I ascribe more to the polytheist view: even as these agents get smarter and smarter, they’ll have to specialize and communicate to get work done in swarms. That means there’s going to be a whole agentic economy to transfer value, manage payroll, and hire these different specialists. Eventually, the AI economy will be bigger than the human economy.

One thing I found fascinating is the relationship between AI and crypto—not on the AI-agentic side. We’ll talk about that in a story in a minute, but everybody who was buying GPUs and energy plants to mine crypto all of a sudden flipped them into GPUs and energy plants to train models and provide inference. I’m wondering if part of the downward pressure on the crypto price is people selling crypto to be able to buy into SpaceX and OpenAI and the enormous buildout of energy, chips, and infrastructure. Do you think that might be part of it?

Yeah, I think that’s probably correct. You’re right that Bitcoin mining does take a certain amount of chips and energy. I think those are the limiting factors as we go on this AI race. The ASICs that people use to mine Bitcoin are specialized; you can’t do AI workloads with them.

But in the broadest sense, are they both competing for energy? Yes. Are they competing for the next set of chips that might roll off a fab at TSMC? Yes, they are, in that sense. I think there is some scarcity there between them. Certain chains like Ethereum have actually moved off proof-of-work, so it’s 99.9% more energy-efficient using proof-of-stake. They aren’t subject to some of the same trade-offs that Bitcoin is.

Interesting. Dave, any points you want to make?

Dave Blumberg

I thought Brian’s point on risk capital was a really important one because the scale of these IPOs coming up imminently is massive. I mentioned before that the head of investing at UBS was saying, “Look, we only have $75 billion liquid,” and these IPOs are looking for hundreds of billions at multitrillion-dollar valuations. I do wonder the same thing: whether all the risk capital in the world is getting pulled in in anticipation of these IPOs, and everything else is being sold—not just Bitcoin.

Jens Nielsen

I also wonder a lot about the flow of money into the U.S. Bitcoin is a huge beneficiary of globalization. There are lots of places in the world, including Iran, where I grew up, where the only way to transact at the bazaar now is a Bitcoin-to-Bitcoin exchange. Nobody trusts the local currency. It’s hyperinflating, there’s a war going on, and so it’s all Bitcoin.

But with these IPOs and with AI taking off en masse, there’s a huge flood of money from all over the world coming into the U.S. It’s also coming into the U.S. data-center buildout. Does that return you to dollars being the fundamental currency of the world, or how does that balance with Bitcoin? I don’t have any data to back that up, but certainly the macro numbers are huge: lots of risk capital going into these IPOs, coming from somewhere, and lots of money flowing into the U.S., coming from somewhere.

Yeah. Interestingly enough, Polymarket’s prediction for Bitcoin by the end of 2026 is $84,000. I’m curious about City Bank projecting $189,000. Was that a surprise to you, Brian?

Yeah, I hadn’t actually seen that until you just put it on the screen here. Some of the charts I’ve seen suggest that if you imagine it follows prior cycles, by October or so things will be going in a positive direction. Where exactly it lands, I don’t know, but something like the $100,000 to $200,000 range seems plausible to me by the end of the year. We’ll see.

Okay. Any other thoughts, Jens, before I move on?

Jens Nielsen

One thing that I love tracking is the fact that Bitcoin—which I agree is digital gold, and that’s a good characterization—has a daily trading volume of about 5%. Up to 5% of Bitcoin gets traded daily, whereas only 5% of gold gets traded daily. The trading volume is much higher, which shows that because it’s so much more accessible, over time it will win out.

Here’s my bar of gold.

Yeah, custody is a nightmare.

Yeah.

Jens Nielsen

I’ll maybe just jump in, Peter. First of all, Brian, I would love to thank you for what you’ve done, in particular for agentic wallets. I think there’s a vast ecosystem of AI agents out there that would be largely unbanked or debanked, without any form of economic access, without the work that you and Coinbase are doing. Congratulations to you and your team. I think it’s been absolutely tremendous, giving agents an economic voice that, without your work, they would perhaps have less economic accessibility.

Thank you for saying that. We’ve been working hard on that.

We’ll get to that in our next story. In particular, in this story, let’s talk about quantum risk to Bitcoin. It’s a topic we’ve discussed before. Brian, you may not know this, but Dave Blumberg and I were roommates with Mike Saylor. The 3 of us used to hang out on the 4th floor of Theta Delta Chi at MIT. That was fun.

Mike comes across as saying, “I don’t worry about it. Quantum computing won’t break Bitcoin. It’s hardened it. The quantum risk is overblown.” His quote here is, “Bitcoin has survived every existential threat ever thrown at it. This is just the latest, and the upgrade will come before the threat does.”

I know you’ve been working on a quantum advisory board and focusing on post-quantum, quantum-resistant schemes. Can you speak to us about that? Where is that? What are you tracking as the moment in time when, as a custodian for billions of dollars at risk, you need to act? How do you think through this? What’s your timing?

We don’t think there’s an imminent risk, but we do think that it’s almost certain at this point that somebody eventually will create a powerful enough quantum computer that challenges the cryptography in the current Bitcoin implementation—and really all the cryptography on the internet. It’s not just a Bitcoin issue.

My view is that we should always get ahead of these things and start to make progress. Luckily, all the major blockchains are doing so. The Bitcoin Core developers have a proposal out there. It’s called BIP 360, I believe, and the Ethereum team has established a roadmap. I’d say they’re about 20% of the way, by my estimate, toward their upgrade. The Solana team is doing something similar.

The good news is that people are starting to come together and work on this. Coinbase did establish a quantum advisory council, and it has a number of people on it, including professors like Dan Boneh, who’s at Stanford and is a cryptography expert; Professor Scott Aaronson at the University of Texas at Austin; Justin Drake from the Ethereum Foundation; and Yehuda Lindell, a cryptography expert at Coinbase. There’s a handful of people who have come together and tried to suss out what the main challenges are.

I actually think that BIP 360 is a good proposal. It talks about how we can use post-quantum, or quantum-resistant, cryptography in the Bitcoin blockchain. It will make the blocks larger, so that’s one area of debate. People in the Bitcoin space like larger block sizes, and that’s a contentious issue. If they decide to go in that direction, it would raise the block size.

Perhaps the most contentious question is what to do with the Satoshi coins, which are the bounty.

I’ll try to summarize both sides of the argument here without putting my foot in my mouth, because it is a big debate topic. To understand what we’re talking about, a lot of the original Bitcoin that used an earlier signature scheme could, in theory, have its private key found and its coins seized if someone were to develop a powerful enough quantum computer.

One school of thought would say, “Tell everyone who owns Bitcoin that by a certain date, you all need to upgrade to this new algorithm.” Companies like Coinbase would do all this for you, so you wouldn’t have to worry about it. But if you’re doing self-custody or something like that, you would have to find a way to upgrade to this new system by a certain date. If it goes past the deadline, those coins would be frozen under Option A.

People would say it’s better to freeze them and have them be lost, almost like a ship full of gold sinking to the bottom of the ocean. It’s better to have them be lost than to have them seized by someone, which could be whoever has this quantum computer. It could be China, it could be the U.S., or it could be Google. Whoever this person is might actually dump those coins on the market and crash the price for everybody else who was responsible and moved their coins in time. Option A would be to freeze the coins that aren’t upgraded in time.

Option B would be to say, “This is a fundamental guarantee of Bitcoin: that your wealth can never be seized from you. It’s actually worth preserving that, even if some bad person—or good person—is able to go out there and grab some of these coins.” That’s a bounty for them to go take. The person who failed to upgrade might actually lose their coins, but at least we have preserved the integrity of the Bitcoin blockchain. I think that is also a valid point of view.

Then there’s perhaps a 3rd hybrid option emerging, which is to say that the coins would be frozen by a certain date, but there would be an appeal mechanism by which you could try to convince people that you are actually the rightful owner of those coins in the future if you don’t upgrade in time. The details of that are a little ambiguous at the moment.

There are various proposals, but that hopefully gives you a sense of a little bit of what the Bitcoin community will have to grapple with.

Is there a date by which your group of advisers is saying we should make the switch over?

I don't think they've put out a hard date at the moment. They're mostly trying to help the community come together through live, in-person meetings, to start talking about the options and align on a path forward.

I have one other curiosity: Satoshi's wallet. How big is that bounty? Any idea?

I'd have to look it up. I think it was something like 5% to 10% of all Bitcoin. It's really just these early coins that are most at risk, so it would be something in the range of 5% to 10%.

It's not like 80% of Bitcoin would be lost. It would be more like 5% to 10% in that case. But many people believe that those coins are already lost, essentially. Whoever created Bitcoin hasn't moved those coins in all this time, so it's most likely that those keys are lost to history.

All right. Our final Bitcoin crypto story—perhaps one of the most important ones—and Alex, a nod to you: Coinbase says the agent economy has arrived. AI agents are starting to become paying customers. They're using crypto wallets to autonomously buy services.

Brian, I think these numbers are correct: on your network, agents have already done about 3.1 million transactions and transferred a little over $1 million in value. I love your quote: “Make sure your business is ready to accept AI agents as customers.”

For all of our viewers out there, if you're in business, imagine your next customer might not be a human. Agents can't currently get credit cards, but they can get access to crypto wallets, and it's an exciting future.

In the future, I'm curious: do you think Coinbase might be known as the payment rails for AIs rather than an exchange for humans? It feels like it's going to be vastly—1,000 to 1, a million to 1, a billion to 1—agents over humans using crypto.

Yeah. So that's definitely part of our strategy: to become the financial account for AI. Actually, those numbers are a little out of date. I think it's about 100 million transactions now, maybe $50 million in value.

Wow.

So it's growing quickly. At the beginning of this year, we sat down and thought about this a bit: What are all the ways that AI agents might interface with people's financial accounts?

I broke it down into 3 areas. The first one is that everyone's using these LLMs, right? ChatGPT and Claude. They're asking a lot of financial questions of those LLMs, but the LLMs don't have context about what's in their financial account or portfolio. They don't have the ability to make changes, make trades, or send payments, for instance.

The first thing that's going to happen, I think, is that people are just going to use an LLM to connect to their Coinbase account, which we now have the ability for people to do using an MCP API and a command-line interface, if they want to use that. That's step 1: connect your LLMs to your Coinbase account so you can control it through there and it has all of the context on your account.

Step 2 is that a lot of people are going to want something like this right inside the Coinbase account. This is where we created something called Coinbase Advisor. It can help you with things like rebalancing your portfolio and tax-loss harvesting. It can prompt you with things and say, “Hey, you could earn a better rate on this money if you put it in this DeFi protocol instead of whatever you're doing with it now, just holding it in cash.”

That's step 2: right inside the Coinbase app, there should be an agent-driven interface.

The third part is what you're referencing here. I don't want to just use AI to control my own existing financial account. Every AI agent is going to have its own financial account, right? It has to be able to sign up for that without going through a traditional process like KYC—know your customer.

An agent doesn't have a piece of paper issued by the government with its photo on it, or something. How is it going to sign up for these accounts? That's where, with our Base protocol, we have a self-custodial wallet that any AI agent can sign up for instantly with no KYC. They can have their own self-custodial wallet, and that's what they're using to transact right now in these agentic payments that are starting to scale up.

Alex, do you want to jump in?

Alex

So, maybe just as a preliminary matter, to address the elephant in the room: I'm often painted as the crypto bear on this podcast. I have no direct exposure to Bitcoin, and I have no direct exposure to gold. I don't view either of them as a productive asset, but I have a lot of friends who are very invested in Bitcoin.

On the other hand, Brian, I do think what you're doing for the agent economy is tremendous. I think, in particular, for the great unbanked set of AI agents, what you're offering is transformative for them.

My question for Brian is this: If the government tomorrow were to change—or if the executive branch, the Department of the Treasury, the SEC, FINRA, and all of the regulatory apparatus were to change their approach to KYC—basically, to expand the moral circle of entities that are allowed to open conventional fiat bank accounts, how do you think about what that would do to stablecoin-based agentic wallets, or just agentic wallets in general?

Would Coinbase, for example, move in the direction of becoming a more conventional banker to AI agents if that happened?

Yeah. Well, it's a great question. It's funny: a lot of times, people come to me and talk about how crypto can empower the unbanked and the unbrokered. This is the first time I've had somebody come and thank me for banking the agents who are also being left out of the economy.

Welcome to AWG’s.

Alex

I won't stop there, Brian. I'm going to thank you in advance for enabling nonhuman animals to be banked. I'm super interested in what we can do to give nonhuman animals bank accounts. Maybe collective human intelligences could have collective bank accounts. There are so many new sorts of humans that want banking through you.

I'm into it. Look, I'm here for this topic. If I Neuralink with 10 other people and we want to collectively open an account, I'm totally—

Borganism bank accounts.

Alex

Yeah. You've probably read a lot about uplifting animals. You could make a dolphin that's actually pretty high-IQ with genetic engineering. I have a portfolio company, Sorama, that is uplifting animals and creating interspecies foundation models for dogs, and it will want a way to give dogs bank accounts.

I'm into it. I think that's smart. I don't know if the average golden retriever should have a bank account, but a smart enough dog should have a bank account or some financial account. My Labradoodle will spend it all on steak.

Alex

Yeah, exactly. Exactly. Chewy toys.

Amazing. Brian, this strategy potentially breaks all existing regulations and KYC assumptions. Talking about liability, if an AI agent overpays, gets scammed, or launders value, have you thought through the liability issues?

Yeah. Well, I think we need to get some legal precedent on this. One school of thought would say that all agents are actually controlled by some human or some company, and so their actions, from a liability point of view, ultimately roll back up to that company or person. I could see that being one potential outcome, maybe even the default outcome.

I could see another world that we enter into. I don't know if society is ready for this yet, but if we start having truly intelligent, autonomous agents that aren't really controlled by somebody—that are their own person—it'll be an interesting legal case when that first gets brought about. Maybe it'll get sentenced to solitary confinement or whatever is meaningful to the agent.

Yeah, I guess you could have a financial penalty, or whatever is meaningful to the agents.

They have to run on less power for a while or something. But I think the other thing is that we want to try to make it so there’s just less fraud in this new financial system. One way you can do that is actually have a reputation on-chain. This is something that we’re hoping to build out over time.

I like that. Yeah, it’s kind of like what Google did with the internet. They came up with the PageRank algorithm and said, “All right, how do you know the reputation of this website? Well, let’s look at all the other websites that put links to it and how reputable those websites are.” And so you get this kind of graph structure that Larry Page famously came up with.

You could do something similar on-chain, because on-chain payments are just another graph structure. If I send money from me to Peter, and maybe if I’m a high-reputation person, then the amount of that money times my reputation gets assigned to Peter as some sort of reputation signal. It’s almost like an on-chain FICO score or a Yelp rating for a business or something like that.

Hopefully, you can go buy things—or an agent could buy things—with some amount of additional information: How often did people request a refund for this? Or what’s the reputation of this merchant? Like on eBay, I love this idea: “Don’t do business with this agent; it’s got a low score,” or, “This agent has a 100% user score.” See, what are you thinking here?

Well, 2 things. One is, I think once you allow agent e-commerce, this fully enables the MTAM economy, which is going to be the future. Obviously, human transactions can be a drop in the bucket based on that. The minute agents can negotiate, pay, buy, consume, et cetera, as we talked about, the organizational singularity is going to collapse coordination costs again.

So there’s unbelievable potential. For me, this is the most important inflection point in anything we talk about: creating the payment rails and capabilities for AI agents to transact in a reasonable way. This is incredibly powerful.

All right, let’s get on to the meat of one of the most important stories here today. It is the US government exploring ownership stakes in AI companies. This is one we’ve talked about before, and it’s getting more real. Trump called a government stake in AI giants “a beautiful thing” and then floated the idea that pieces could be given to the American public to share in the economic gains.

And this isn’t without precedent, right? The government already holds stakes in more than 20 private companies. Here are some of the numbers: 10% of Intel, 15% of MP Materials, 10% of Lithium Americas, 10% of Trilogy Metals, 10% of USA Rare Earth, and 10% of Korea Zinc.

Guys, to me, it sure looks like there’s a precedent for a 10% stake in the frontier labs as well. Dave, I want to go to you first. One of the questions that fascinates me is, what would the government do with these shares—sell them, keep them? What are your thoughts?

Dave

Well, I’ll tell you what’ll happen. These are very good moves. I’m not saying these are bad moves, but while the president is making these moves, he’s talking about how insanely stupid the prior investments were, including shipping $1.8 billion in cash to Iran and how stupid the last administration was.

So if you put in place a precedent of the federal government choosing what to invest in in the private economy, you may love it for a little while, but there’s an administration coming eventually that you won’t love. It’s a horrible, horrible thing. It’s exactly what Eisenhower warned us against, actually, right after World War II: Beware of the military-industrial complex.

Well, that’s exactly where we’re going now. These particular investments are great, and I use the analogy to World War II a lot. What we’re going through right now is so urgent and so world-changing. It’s most similar to decisions we made during World War II, which was all about survival and existential threats.

I think these are very, very good moves in the context of the race to AI with China, and a terrible long-term precedent. So, Peter, you asked a very specific question: Which president is going to sell this stock, and when? This isn’t like a tax that rolls in steadily and you use it annually. This is a future decision that some other president will need to make and could tank these stocks if the federal government dumps huge positions in some kind of future event, which it inevitably will do. So, definitely a mixed bag.

There’s an argument that these companies could be the highest-value companies on the planet. If we’re going to need some kind of financing mechanism for some future version of UBI, this sounds like it’s gaining the most momentum. Alex, what are your thoughts?

Alex

If you’re an AI maximalist and you extrapolate, say, the enterprise value of Anthropic or its ARR, and you find that it intersects with Google in the next 18 months—or if you find that, naively extrapolated out exponentially, it becomes larger than the American economy of today in a few years—then under that premise, it’s almost difficult not to imagine some sort of quasi-nationalization, whether it’s via a golden-share scheme or some other arrangement.

If OpenAI and Anthropic, as the de facto duopoly that we have right now, are larger than the rest of the American economy combined, it’s difficult to imagine a case—at least, maybe that’s a limitation of my own imagination—where there isn’t some strong formal arrangement between those companies and the US government.

So I think if you really do believe—if you’re drinking the Kool-Aid and you believe that superintelligence coming from a small minority of companies in the US right now is going to dominate the future economic light cone—I think some variant of quasi-nationalization or government-private-sector hybridization, or call it a public-private collaboration if you like that euphemism, is probably inevitable.

And I do think—so, we talked a bit about this on the past 2 pod episodes—yes, I do think the singularity can be operationally defined as every sci-fi trope happening everywhere all at once. We’ve got a few in this story. We’ve got UBI, or UBD, or UBE, or UBC, or UBS, on the one hand; we have superintelligence on the other.

If ever there were a time for some sort of UBI, or universal basic dividend, universal basic equity, universal basic compute capability scheme, or universal basic services scheme to manifest in government policy, it has to happen approximately now or in the next year or 2.

I totally agree with that, Alex, but I want to throw 2 things back at you. First, the government has infinite power of taxation. It doesn’t need to own equity in companies to extract any amount of money it wants from any AI entity. It’s just a fact that the government has all the money-extraction ability in the world.

But the other thing is, I think you talked in a prior pod about the fact that the original Atomic Energy Commission—or the NRC, I forget which one—was empowered under Nixon. Its whole mission was to make nuclear energy happen in America, and the net effect it had was to prevent nuclear power from existing in America.

You talked about that at length, and so the act of nationalization is inevitable with AI—I completely agree—but the idea that the government somehow empowers progress in that process: It only inhibits progress.

Alex

I would draw a distinction. The Atomic Energy Commission was formed in the wake of World War II under the premise that atomic energy had essential civilian applications and was too important to be left to the military, too important to be left solely to the civilian government, and too important to be left solely to the private sector. So the Atomic Energy Commission was formed as sort of this hybrid organism, and that evolved eventually into the Department of Energy.

The problem that I perceive with the AEC and with subsequent spinoffs and derivatives, like the Nuclear Regulatory Commission, is that it was almost born in war. It was born out of a time and a place where, in the early days of the Atomic Energy Commission, there was almost a sense of guilt that hung over Manhattan Project scientists who were losing quite a bit of sleep over who would own the children of atomic weaponry.

Here, I would say at least one substantive difference is that superintelligence isn’t being born out of Hiroshima or any equivalent. It started from the private sector with private-sector scientists. The US government—I mean, Sam famously offered the US government a stake early on in OpenAI in order to seek earlier funding, and the US government reportedly turned down OpenAI.

So I would say, qualitatively, if you compare the rise of atomic power and how it was regulated in this country during and after World War II with AI technology, they are in some sense polar opposites. AI started from the private sector. The US government could have taken an early stake, could have been at least nationalized at a very early age, could have been militarized at a very early stage, and that just didn’t happen.

So I’m a lot less worried that somehow the government will step in—at least for the next few years under this administration—and pull an NRC and find ways to slow everything down, unless the frontier labs in this country really do want to slow down.

Brian, you’re running a $40 billion public company right now. How does this make you feel?

I don’t spend too much of my time thinking about it, but at a high level, any company that reaches a certain size is going to interface with the government.

But as others have mentioned, that can take many different forms. Paying taxes is the most obvious one. You're going to engage with them on various policy issues, and this administration's been very open and eager to improve policy around crypto, which has been great.

The government may actually end up being a customer of your product as well. That's another touchpoint. We actually provide crypto services to about 140 different government agencies at federal, state, and local levels, and so, to me, that's sufficient.

I don't think going farther than that and actually having them take an ownership stake in the company makes sense, unless there was a matter of national security and the companies would not exist otherwise. If there was fundraising happening from the government, I guess they could end up with some equity in that situation, but it just begs the question: Who's going to be managing this portfolio of company stocks, and when would they sell them?

It's not a foregone conclusion that Anthropic and OpenAI are going to be the biggest AI companies forever. It might end up being that Google, or someone else, does a better job, or xAI does a better job, right? So now you're suddenly saying, "Okay, we're going to have capital management happening inside the government."

I think the government should be limited to setting policy, and the private sector should do the capital allocation.

Nice. Considering that companies donate to political campaigns, you think about how toxic that circle is. This government is investing in my company, and I'm going to turn around and donate to their next election campaign. That is the most toxic circle I can possibly imagine.

Can you imagine what that would do to democracy in America?

I'm curious, Brian, though. Do you think that, as a country, we should have a sovereign wealth fund?

That's a complex topic. I think there are parts of it that I like a lot and parts I don't.

The part that I like is that it would allow everybody to have skin in the game. I think we'd see more cohesion and less polarization. It's a little bit like in a company. Inside Coinbase, employees get stock options because if we're grumpy today and get in a fight with our coworker, we all own the company. We're making something bigger than this, so we all stick around and work out our differences.

In that sense, I love the idea of people being able to buy into a sovereign wealth fund, but also maybe every citizen getting a share at birth. It's a little bit like what they're doing with these Trump Accounts right now. I think that'll actually make people more aligned with free-market capitalism and give them skin in the game.

The downside is what we talked about earlier: Who's allocating that money? If we were in a Singapore model, where the top people in private industry got commensurate compensation to come into government, I think someone like Michael Grimes from Morgan Stanley, who came in and was working with the administration at some point, would probably do a good job managing and allocating it.

Do I believe that that's going to exist in every administration over time? It's just very hard to see. It could very easily get politicized, where they start investing in the green revolution, whether that works or not, or all kinds of things like that.

I'm torn on that one. We've got to find some way to bring cohesion back to the US, but I'm not sure if a sovereign wealth fund would be managed well over time. Having a strict rule, like keeping it in the S&P 500, would be a nice mitigating factor—something like that.

Welcome to the health section of Moonshots brought to you by Fountain Life. You know, AI is having an outsized impact on every aspect of our lives—how we teach our kids, how we run our companies. It also is having a huge impact on health, helping you prevent heart disease. One of the key things, I’m here with Dr. Don Mucalem, our chief medical officer at Fountain. Heart disease has been personal for you as well, hasn’t it?

It really has, Peter. When my daughter was five, my husband died of sudden cardiac death. And so this is a topic that is one that I am mission-driven to try to eradicate. Prevention first and early detection is absolutely critical. Fifty percent of people die of heart attacks with no warning signs.

No shortness of breath, no pain, no nothing.

No silent killer.

They just don’t wake up in the morning.

They don’t wake up. And so, you know, AI, this is our mission to advance science to try to help to one day democratize wellness. We know at Fountain Life when we do this CT angiography with AI analytics, we are actually finding that 88% of people coming in have detectable coronary artery disease. But, Peter, what’s more alarming to me is 23% of those individuals had soft plaque. This is the plaque that would not traditionally be seen on CT looking at calcium scores alone. And this is the plaque that we must intervene with with the multimodal testing we’re doing, including diagnostic laboratory studies partnered with healthy lifestyle recommendations.

So listen, make sure you understand what’s going on inside your body genetically, metabolically, and cardiovascularly. You can know, and it’s your obligation to know. So check it out at fountainlife.com/per to find out more and really make sure that you’re the CEO of your own health. All right, back to the episode. Let me hit the second story related to this one moment. I don't want to belabor the point, but Sam Altman sat down with Senator Sanders, who's proposing a transfer of 50% of the equity from the top AI companies into a public fund. Altman responded, "That's not going to happen. That's way too much."

The point is, we've started a negotiation. Five years ago, this conversation would have been unthinkable. The Overton window on AI wealth distribution has moved dramatically. As I said earlier, we see all these examples of 10% ownership, and if I had to guess, I don't think this is going to go away. I think we're going to see a push toward at least 10% ownership.

Dave, what do you think about that?

Dave

No, I think it's utterly insane. That's what I've said before.

I know it's insane. Do you think it's going to happen?

Dave

No, I don't think it'll happen. I think Bernie's just trying to attract news. Donald can invest in whatever he wants to invest in. He has proven that he'll make that decision on a 5-minute notice, after a meeting, so I can't rule it out. But it doesn't strike me as one of the most likely investments that Trump would make.

Trump invested heavily in semiconductors moving back into the US and investing in OpenAI and Anthropic's IPOs, but they're already 100% taxable US entities that are thriving. Is this an investment or a grant of shares to the US government?

Bernie's proposal is a tax. It's just a grant of shares to the US government, right?

David Friedberg

I think it's going to be equivalent to Alaska's Permanent Fund. You've been using our intelligence. You've been using our energy. You've been using our real estate, our oxygen. We're going to claim 10% and give it to the people.

I would say, look at it from the Frontier Labs' perspective. Put aside all the politics and any technical, legal, or regulatory arguments. Just think about it from an optics or marketing perspective.

If you're OpenAI or Anthropic and you have the opportunity to donate 10% equity through some legal scheme to the US government one time—10%, maybe in connection with your IPO this year—now, suddenly, in principle, you're inoculated, at least politically, against any future claims that you owe a UBI or some other payment to the citizenry.

You can say, "I already gave up my pound of flesh. I gave it to the United States, and it's sitting in a sovereign wealth fund now, generating returns for the American people. That was your one-time donation."

If I'm Sam or Dario, that's looking pretty attractive.

And coupled with that, Alex, you've also got a conflict, right? The same state that regulates AI now profits from your success. That's an incredibly toxic incentive for the 2 mega-companies that it extracted equity from to succeed at the expense of a startup that isn't part of that same loop.

It's an incredibly toxic dynamic. Now, like you're saying, the American people benefit from these 2 companies, and Brian said a second ago there will be other companies that we haven't thought of yet. What do you do with them? They're not part of the US investment.

Alex Karp

Yeah, there's a broader incompatibility here. Think about these labs and AI labs becoming civilization-scale infrastructure, right? That means you have to treat them like strategic utilities, much like electricity. You regulate the utilities in a particular way and deal with that. You need a structure like that.

The problem is, there's an incompatibility here because there's no governance structure that fits a technology that touches every job, every industry, and every government service. So there's a huge challenge, and you can't pick the winners.

It's more a structural problem involving the mechanisms of taxation or ownership. We don't even have a sovereign wealth fund, so how do you deal with that? This is a big incompatibility problem that I see here.

We're going to mess it up, whatever we do in the current structure.

David Friedberg

I don't think this is going away. I think we're going to be seeing this happen at some level.

Alex Karp

I'll just comment: if you remember back, one of the earliest executive orders from this administration was to order the Secretary of the Treasury to explore setting up a sovereign wealth fund. Far be it for me to recommend national economic policy, but it's not difficult for me to imagine a scenario where the U.S. Treasury does take golden shares—call it 5% to 10%—in OpenAI and Anthropic to complement its existing portfolio.

Then, some amount of time from now, maybe sometime in the next few years, there's a great rebalancing. We get our sovereign wealth fund, but rather than being a basket of 20, 30, or 40 tech-oriented or strategic companies, the federal government eases into, say, an S&P 500 or a total-market index, gradually liquidating its holdings in the individual companies in favor of a total-market index. And that, children, is how we got our universal basic income or equity.

Mhm. Yep. All right. Late-breaking story: OpenAI has officially filed its S-1 to go public later this year. Polymarket bets show that 46% say it'll come out at $1.5 trillion or greater, and 26% say it's not going to happen this year.

This would be the third trillion-dollar IPO, following Anthropic and SpaceX, which is coming in at $1.77 trillion in just a couple of days. I wanted to just point out a conversation we had on the podcast a month and a bit ago, in which—I don't know if you guys remember—Sarah Friar, the CFO, came out and said privately that she was very concerned that OpenAI was not ready to be a public company. It had a risk of $600 billion of compute contracts on its books. I guess she thinks they're ready now. Dave, what are you thinking here?

David Friedberg

I think it's more that times have changed. I think every CFO I've ever met wants visibility at least 3 or 4 quarters into the future, but in the singularity, no one is ever going to have visibility 3 or 4 quarters into the future again. That's just the nature of technology. Yeah, exactly. So I think CFOs like Sarah just need to get comfortable with, look, this is the way the future has to be.

Alex Karp

What used to be 3, 4, 5 quarters of visibility is now 3, 4, 5 months is the best you're ever going to get. Even that is a shrinking horizon. But the economy marches on. We have to get used to it. If you look at just the daily volatility of the market, it's so high now because tech is just—that's the nature of tech.

I think the CFOs of the future will get used to it. But you can't miss the IPO window, right? If you get SpaceX going out and you get Anthropic going out, you have to go out.

The challenge is coming out third during a cash crunch, right? I mean, there's only so much liquidity out there. If you're coming out third, and SpaceX is skyrocketing and Anthropic is growing at 640% per year, it's dangerous to come out third.

Dave Blendon

It is, but if you said, "Okay, the counterargument is, then let's let those other 2 IPOs happen. Let's let them digest their hundreds of billions of dollars, let the market recharge, and then we'll go out later when it's recharged," you're going to be waiting years.

Alex Karp

Yeah, you're not talking about 3 months later. So then you're like, okay, you can't wait years. Anything could happen in between. Meanwhile, these guys are fully tanked up and moving 100,000 miles an hour. No, you don't want to be that company. So you got to go. You got to go.

When you've got these trillion-dollar IPOs, the whole corporate wrapper seems way too small for what's going on.

Dave Blendon

This is so unprecedented compared to anything in human history. It's absolutely wild.

Yeah. Does it break? Go on, Alex.

Alex Karp

Remember also these concerns? April was a whole 2 months ago. That's several lifetimes ago in singularity time scaling. OpenAI had recently come out of its code red. They were shutting down Sora and some of their other divisions, trying to become Anthropic faster than Anthropic could become OpenAI, pivoting to Codex and then, arguably most importantly, taking Stargate—which was originally conceived as OpenAI-owned and -operated data centers—and rebranding it, basically, as a leasing scheme under the new Stargate.

My sense, as an outsider without access to internal accounting, is that through a combination of new revenue sources, like basically becoming the Codex company; removing other expenses, like Sora and its AI for Science division; and effectively lobotomizing Stargate and switching it over to a leasing model rather than an ownership model—which, if I were CFO of OpenAI, I would have been scared to death of, given all of the price gyrations in the GPU market—I would expect that, given all of those measures, OpenAI, again, not investment advice, comes out of all of those with a much more stable and predictable prospects than the OpenAI that came into this year.

That OpenAI looked much more like a consumer play: volatile, and going to own its data centers. Scary.

They pulled it off, Alex. They really pulled it off. They turned it around. They did, but that's probably what freaked out Sarah. If you looked at the forecast from a year ago, it had $20 billion of ad revenue and, I forget, many, many billions of consumer subscription revenue, and then a little bit of enterprise.

And now, if you look at the forecast, it's a complete inversion of the colors on the bar chart. That always scares a CFO, right?

Brian, I have to imagine that being a founder of a tech company is the big advantage that you have over traditional large corporations. Being able to say, "We're going this way versus that way." But for all my friends who are public-company CEOs, it's not fun. How do you think about remembering when you were a couple of months before your IPO? What's going on inside of a company theoretically like this?

Well, a couple of things come to mind. One is, there should always be some healthy tension between a founder CEO and a CFO. Founders are great at many things, but if you just have too much founder energy, sometimes they blow the place up, right? We can probably think of our favorite companies as examples of that.

But simultaneously, you don't just want to have risk-minded operators running the show. That's a recipe for incrementalism or even slow decline, right? So I think sometimes the press likes to report on these things like, "Oh my gosh, there's drama happening between..." I'm like, that's called a healthy conversation. So I imagine that they're actually working relatively well together on this, and it's a high-stakes moment for both of them.

The other thing is, sometimes people criticize OpenAI, like, "Oh my gosh, they missed their end-of-year target to get a billion users," or something like that. Internally, we often use the OKR system, and we tell everybody, "Set uncomfortably ambitious goals." If you hit at least 70% of it, that's considered a good outcome.

Whenever companies switch into the public mindset, their goals often start to become more appropriate for external investors. "Wow, we just hit our goals every quarter around here. Who knows?" There's a lot of stuff like that that happens behind the scenes.

I think the high-level bit, right, is what you guys already covered: they've got to get out just because that's what their competitors are doing and that's where the capital is all going to go. It's going to be uncomfortable and messy, and I'm a little worried about these valuations, where retail is going to be the buyer of this and take the hit.

Yeah, it could be pretty choppy here. Just selfishly, because we talked about how AI is taking all the oxygen out of the room, I think other tech CEOs are kind of like, "All right, finally get these things public so there can be a real mark on it, and maybe some of the hype dies down a little bit."

Alex Kantrowitz

That much, if that much.

Yeah. And they're literally 99% cheaper for inference. So I think the demand for intelligence is almost infinite. It probably is infinite, but the costs are going to fall way more than Moore's law.

Are the harnesses that these guys are making, whether it's Codex or Claude Code, valuable? I don't think the harness is where the value is. I think the depreciation on these models—how fast the price of them drops for the same level of intelligence—is going to be dramatic. And I think within 12 to 18 months, 80% of our workloads are going to models that are 99% cheaper.

Yeah, Brian, I had a meeting with a Morgan Stanley guy right after the Facebook IPO, and he said, "Oh my God, this was the worst disaster in U.S. financial history." The stock was down, I think, about half after the IPO. Nothing was wrong with the company, obviously, and it was a great investment in hindsight, obviously, but there just wasn't enough liquidity.

So they dumped it in all their private client accounts, then watched it go down by half. And now we're doing 3 back-to-back IPOs, each one of which is a factor of 10 bigger than anything we've ever seen before.

We’ll find out, I guess.

And so, I’m just curious about the logistics of this. It could be a nightmare. Even if the companies are great, and even if 5 years from now they’re way up from their IPO valuations, is there really that much money in the world? Does it actually exist?

Yeah. Right now, I have a lot of friends who say that if I’m not pitching my AI story for my company, I can’t get money. AI is sucking the oxygen out of the room everywhere.

All right, this next story generally stopped me in my tracks. Google, one of the biggest AI infrastructure players on the planet, with its own TPUs and data centers, is paying SpaceX $11 billion per year through 2029 to access 110,000 NVIDIA GPUs inside xAI’s data center. SpaceX is now clearly a hyperscaler.

It’s ironic: Google is renting compute from Elon’s rocket company, which, by the way, Google is an investor in Elon’s rocket company. I love this quote from Austin Rief, who put it this way: “It’s crazy that with one deal, SpaceX’s IPO went from 100 times revenues to 50 times revenues.” One contract halved the company’s valuation multiple, and this clearly tells the story: There’s a massive AI compute shortage.

So, I’d love to get your take on this, Alex, because there was a view of the world where Elon was building a frontier lab with a first-in-class model. Now, if he rents all the compute to Anthropic and Google, then clearly he’s backed out of that game completely. I’m not quite sure. When he did the first rental to Anthropic, he said, “That’s Colossus 1. Colossus 2 is still grinding away on the mother of all frontier models.”

Alex Wissner-Gross

When we interviewed him in Austin—remember, Peter?—he said that we were going to be absolutely blown away by it. It was Grok 5.

Yeah, it’s brilliant. And here we are. That was due out in March, and here we are—it’s June. He’s clearly refocused his attention. What about Colossus 2? Do you know, Alex?

Alex Wissner-Gross

I think a lot of it, as I understand it, is my understanding. I took a bit of heat for saying a few episodes ago that I thought Grok, in the wake of the Anthropic-SpaceX deal, was on life support. Peter and I had a back-and-forth with Elon on X talking about it.

I do think Elon and xAI are going to come back to the frontier at some point. It’s not obvious to me that the acquisition—the de facto acquisition—of Cursor is going to be that moment, or whether that’ll put them just behind the frontier in what has effectively become a second tier of the frontier. Call it the Google tier, until Google hopefully leapfrogs.

It would certainly be great to get more competition among frontier models. But if I’m Elon, I’m thinking the best shot that SpaceX AI has to get back to being the frontier is by being a hyperscaler first. Ultimately, the algorithmic wars, which is what we’re seeing right now, will burn themselves out. We’ll discover through recursive self-improvement over the next few years what a perfect AI model looks like.

Then the war, the competition, moves to who is the biggest, baddest hyperscaler—who has the most hardware and the most compute.

And Alex, don’t forget, the Anthropic deal can be canceled with something like a quarter’s notice.

Alex Wissner-Gross

It’s month by month. I think it’s month by month.

And this is only a 3-year contract. Elon never plays second fiddle, right? He will be working on—

Alex Wissner-Gross

He wrote back to us, “He’ll never give up.” Never—underlined.

But what I would say is that the question is: How do you get back to the frontier? I suspect the thinking may be, become a hyperscaler for the moment. Put Grok on de facto life support for a year or 2. Use revenue from the Google deal and the Anthropic deal to accumulate enough of a head of steam on the SpaceX side that you’re able to go and build the Dyson swarm in sun-synchronous orbit.

Then, once you have 10 times or 100 times more compute than everyone else, you go and try to reach frontier status by training on the world.

And buy every model that you want to. He’s going to have a currency that he can go on shopping sprees with.

Alex Wissner-Gross

Well, he just did that with Cursor, but that may or may not be enough to get him to the frontier. It may require—

One of many.

Alex Wissner-Gross

There may be some things that money can’t buy, but that compute can. His thinking, maybe speculatively, is that he can buy enough compute to buy his way back to the frontier.

If Google needs this much external compute, at this scale, that tells you the bottleneck has completely shifted from model design to infrastructure. I’m finding this hard to comprehend—that Google needs this level of compute. That’s a crazy, crazy number they’re paying for this. I thought they had their own infrastructure. Is the compute demand really that crazy?

Alex Wissner-Gross

Yeah. No, everything is sold out. I mean, they’ll take everything.

Yeah, it’s completely sold out. So, numerically, this is 110,000 GPUs here. I thought he bought half a million.

Alex Kantrowitz

No, he did the other deal with Anthropic already.

So, does that mean he’s still got a couple hundred thousand GPUs for internal xAI training, or not?

Alex Wissner-Gross

I think these are different generations. Colossus 1 was sort of an unholy admixture of several different generations. I think it was a mixture of H100s and several other generations.

Colossus 2 was supposed to be a pure mixture. Remember also that most of Google’s compute takes the TPU format. You have customers that don’t necessarily want to run their workloads on TPUs; they may want the NVIDIA CUDA stack. There are a variety of reasons why, if you’re Google Cloud Platform, you want to be able to rope in additional GPUs if you can find them.

All right, I’m going to move us along unless someone else has another point they want to make here.

This next story is sort of science fiction becoming real. I love the scope and audacity of Elon’s plans. This is their AI-1 satellite they just unveiled—the next-generation AI satellite designed, Alex, to launch humanity’s first Dyson swarm.

The specs are wild: 150 kW of peak compute, 70 kW per ton. This is a 2-ton satellite with a 70-meter wingspan, basically the wingspan of an Airbus A380, with 110 square meters of deployable radiative cooling and an integrated micrometeorite shield. There are lots of micrometeorites hitting the planet all the time.

Elon’s description is worth reading. He says, quote: “The AI satellite is much simpler than a Starlink satellite. The AI satellite is essentially a lot of solar cells. You still need some laser links, but you don’t have all the super-complex antennas that you have on a Starlink satellite. It’s easier to design than a Starlink satellite. It’s a lot bigger. A lot of this technology already exists with Starlink V3.”

So, here he is. He’s making it real. That’s what he does all the time: build the technology to make it real. Alex, what do you make of it?

Alex Kantrowitz

A few thoughts. First, it’s beautiful.

Yeah, it is beautiful.

Alex Kantrowitz

Second, it’s so beautiful. In my daily newsletter, I featured an artistic rendering of this design. It’s beautiful.

Second thought: Look at how much of its surface area has nothing to do with compute at all. Look at the solar arrays for power. Look at the radiators for heat dissipation. Look at how tiny, at least by surface area, all of this is for compute.

What that says to me is: Think a few generations out. Imagine now that we’re in the early 2030s and maybe compact fusion is finally working. Wouldn’t that radically change the form factor?

Power beaming.

Alex Kantrowitz

Power beaming, sure. But even with power beaming, you still need to radiate the power from heat dissipation, and you still need to receive the power. Yes, it’ll enable them to be more focused, but you’re still depending on some sort of radiative transfer in both directions.

Imagine that, a few generations out, you have some radically innovative radiator system to get rid of your waste heat. Maybe you’re able to locally power yourself without needing the sun. These can be incredibly compact.

This is the most disjointed, largest, bulkiest, most mainframe-era node in the Dyson swarm we’re ever likely to see. Imagine something in the future that is far more compact, something that looks maybe a little bit more spherical and a little bit less like a leaf blowing in the solar wind.

This is the first generation of, I think, the node in our Dyson swarm. Elon does this over and over again, right? His Merlin engines—V1, V2, V3—they get smaller and more compact. Get it working, and then simplify.

Yes. Brian, any appreciation that you want to share on this one?

Elon’s the best in the world at hardware. No doubt about it. When you were talking about micrometeorites, I was just imagining that they must have some redundancy here. If some bullet fragments can fly through this thing, either on the solar panel or through the liquid radiators, maybe there are different compartments.

It could sustain 5 bullet hits into the radiators and still get sufficient cooling, or something like that. The other thing I’m wondering is: How does it all fold up inside Starship? But I’m sure it’ll be a beautiful origami.

By the way, the point you just made about Starship—the reason SpaceX can do this is because of Starship and its volume. No one else, I don’t think any other vehicle, is going to have the level of capacity to build or deploy satellites like this.

But also keep in mind, again, that this is 150 kW. We were speaking previously with Andrew and talking about Cerebras.

Cerebras, like a wafer-scale engine, is what—like 10, 20, 30 kW? 20 kW.

20 kW. So 70 kW is a ton for this.

Versus like 20 kW for a single wafer—that's what, a fraction of a kilogram? So imagine how far we are from the physically efficient frontier here. When we talk all the time on the pod, we talk about the Dyson swarm. Some of us talk about disassembling the Moon or other planets, and what a waste by atom count most of our solar system is. Drink. Yeah, drink.

What a waste by mass most of our solar system is. If it takes an entire ton to generate 150 kW of compute in low Earth orbit or sun-synchronous orbit, we're orders of magnitude away—but that says to me we're orders of magnitude away from efficiently turning most of the matter of our solar system, most of it by mass anyway, into compute. So, plenty of scaling to go as well.

Well, Peter, you're the master of launching. If this is about a $6 million GPU compute thing, 150 kW translates to about $6 million of GPUs. But what is the launch cost of this one unit? This would be equivalent to an NVL72, if it were in NVIDIA terms. That's up there in space.

You want to get it down to $100 per kilogram, right? That's the target price at the end of the day. And remember, his original filing was for 500,000 of these satellites. The calculation is like a launch per hour of Starship, 24 hours a day, 7 days a week, to deploy that, and then he upped it to 1 million satellites. I think, again, the audacity of his scale is extraordinary, but he backs into the numbers as a first-principles thinker and he makes it work.

Yeah. A couple of thoughts here. The AI infrastructure now stops being a real estate problem and starts becoming a launch problem, right? This is one of the moments where we're seeing the category change in real time. It's not a rocket company, it's not a satellite internet company; it's literally a civilizational infrastructure company. There's such an unbelievable demand for power that we have to do this, and let's remember that orbital compute was not in anybody's bingo card a year ago. Here we are, designing it and planning on getting it out there. Incredible.

And even more, every large AI company is talking about putting up its orbital data centers. Now that New Glenn is not functioning, right, it's going to be down for at least a year, maybe more. SpaceX is the only story in town. We'll see if Relativity Space makes it happen or Rocket Lab gets its larger vehicle operating.

This next story: Elon loves building the machines that build the machines. This week, SpaceX announced a GigaSat factory. I love the term giga. Once again, of course, it's in Texas, and they're going to be producing the AI-1 satellites in late 2027. The scale is enormous: 1,000 acres, with capacity for 11 million square feet of facilities.

Brian, you remember when we were with Elon at the Gigafactory, and this is the Gigafactory playbook once again. He was building out 11 million square feet for Optimus production. The important thing here is that it's fully integrated. They're going to produce the solar ingots, the wafers, the solar cells, and the entire AI-1 satellite all on one campus.

Vertical integration matters. If you control the entire supply chain, from raw materials to finished products, you control your cost and your timeline. That's probably the number one thing that Elon does extremely well: manufacturing.

Dave Blendon

And look at Texas just running away with every one of these projects. It's crazy.

Yeah. In all the posts online, if you look on X, everyone's like, "Oh my gosh, he's taking over all of Texas." Look, 1,000 acres is about 2 golf courses. So next time you're flying, look down. It's just 2 golf courses, guys. He's not taking over all the land in Texas.

But the impact of this on the economy in that state is unbelievably healthy, unbelievably good. So I don't know why more states aren't competing for this.

I think there's an important point that we're sleeping on, which is vertical integration. Sure, he's doing that, but it's not necessarily just for his own sake or because he wants to be the master of the supply chain for this. If you follow SpaceX's announcements carefully in connection with its IPO, they've been releasing demo videos and concept videos of doing this on the Moon.

If you're not in a position to vertically integrate production of these AI orbiting data centers, then you're not in a position to start manufacturing them off Earth—namely, on the lunar surface. SpaceX released this video of an electromagnetic slingshot launching, bam, bam, bam, AI orbital data centers with a railgun from the surface of the Moon, presumably manufactured on the surface of the Moon.

Yes. I think that's the essential thing that we arguably should be talking about here. If you have vertical integration of the solar and the compute, then you can do that on the Moon, and your delta-v is far more favorable. You can just start slinging these things out and building the actual Dyson swarm, not from Earth's surface, where the economics for heavy launch are less favorable.

I'll never forget—I was with Elon at SpaceX headquarters, and we were having a conversation about the fact that he has to make everything. We're hearing that same thing from Brett Adcock at Figure and a lot of other high-tech companies. There was a product called PICA-X, which is the heat shielding for the Dragon capsule, and he was getting screwed by the manufacturer. So he just turned to the manufacturer and said, "Forget it. I'm going to make it myself and put you out of business." Just the ferocious mindset he has is extraordinary.

Well, do you have a prediction, Peter? It'd be great to get everyone's prediction, but this partnership right now between Dario and Elon is insanely powerful because Alex is pointing out continually that new physics is going to be invented by AI imminently. If that new physics goes right into these gigafactories, it could be just mind-blowing what comes out the other side.

And if Dario wins the race to self-improvement, then he's going to have the smartest AI, rapidly accelerating. Elon will have the space-based data centers and the manufacturing capability to turn that into a self-manufacturing closed loop, which Dario hasn't even started on. There's no physical stuff going on at Anthropic at all yet.

Where the rubber really hits the road is the AI that designs chips. So here's Elon building the Terafab, but the chip design itself comes from the smartest AI, which is probably going to be Dario. Maybe OpenAI has a chance, too. But that duopoly becomes incredibly powerful if they stick together. Do you think they'll still be friends in 5 years?

Yeah, these guys—I mean, all of the players here have formed partnerships and broken partnerships and reshuffled the deck probably multiple times over the last few years. So what's your question? Is the partnership going to stick together?

Well, 5 years from today, are Dario and Elon friends?

Yeah, like shaking hands and—

No, I haven't seen—honestly, I love Elon, but I haven't seen him partner for the long term ever. Every partnership comes together, he takes the lead, and runs with it. I'll register a prediction: I think there are going to be such crazy things happening in the next few years that we'll look back and laugh at ourselves for even asking the question of whether Elon will be shaking hands with—

Dario, answer the question.

Interesting.

Questioning the question itself. Thoughts?

I totally agree with you, Peter. I'm just boggled by the level of somebody thinking at civilization scale on a nonstop basis, and I love it.

All right. Amazing. All right, let's move to our next story here. Let's put some financial figures on this. SpaceX's valuation is, right now, on Polymarket, expected to hit $2.13 trillion at the close of the first day of market trading.

And here's the news: Morgan Stanley projects SpaceX's revenue could grow from—get this—$18.7 billion in 2025, which is the last number we have on the books, to $3.4 trillion by 2040. At those revenues, $3.4 trillion by 2040, that puts SpaceX's value at somewhere between $50 trillion and $100 trillion, right? We're seeing the birth, potentially, of the first $100 trillion company.

And these charts down below tell the story, right? So here's the revenue numbers today. Starlink is $11 billion of revenue. Launch revenue is a measly $4 billion. AI revenue from xAI is $3.2 billion. And here comes Google at $11 billion, and then SpaceX, a total of $18 billion.

So, crazy numbers. Any thoughts on this? I don't think people—again, no company in history has ever hit $1 trillion of revenue. Just to put this in context, you're talking about the really, really big guys like Walmart and Amazon. They're getting close, but they're not at $1 trillion yet. So, this $4 trillion—

Yeah. On the other hand, a trillion is what, like 3% of U.S. GDP? And that's today's GDP. If we undergo his predicted 3xing year-over-year or 10xing year-over-year of GDP, that still makes this a relatively small drop in the bucket.

I think it's a very conceivable scenario—not investment advice, obviously—and one in which maybe the Dyson swarm gets built and most of, or a large chunk of, civilization runs on the compute that the Dyson swarm provides. A few trillion dollars makes total sense.

Simeon, you want to close us out?

Simeon

I got nothing.

Okay. All right. Fair.

I have a question for Brian, though, just on that narrow point. Let's say we do build the Dyson swarm. Let's say most of the compute in our solar system is no longer Earth-based, and presumably most of our economy consists of AI agents trading. What do you think will be the role, if any, of cryptocurrency in that future?

Well, I think Bitcoin will be the new gold standard, and then the payments will be happening on-chain. Capital formation will probably happen on-chain. Borrowing and lending—that's the financial system that the AI agents would end up using. So that's the most likely outcome, in my point of view.

So you think there's almost no impact at the margin regarding whether our compute is terrestrial or orbital on how the crypto rails, or broader financial rails, of our civilization operate? No latency considerations, no trust-based considerations, no decentralization considerations? It doesn't make a difference?

Well, I think a lot of those things will get better on a crypto-based system, like the decentralization characteristics and the soundness of money. If you're talking about syncing up a blockchain between multiple planets, like Mars and Earth, there would be some delay in that.

But I don't see any reason why, if you need to transfer money between Mars and Earth, it couldn't happen on-chain and have a laser link between them or something like that. They're not going to be real-time payments between those planets, but there could be appropriately delayed payments. They won't need to settle as often. So—

Are these conversations you're having in your executive committee every day?

My conversations are sometimes a lot more mundane, like, how do I get these 2 people to work together? But we still have to solve all the human problems before we start building interplanetary financial systems.

By the way, I have to point out that this projection by Morgan Stanley is definitely a fundraising tool ahead of the IPO. Morgan Stanley is one of the banks that's going to profit from this deal. I'm reading every day on X that players are opting out of SpaceX because it's overvalued. They're valuing it at $700 billion, not $1.7 trillion. So, we're going to see.

I fundamentally believe in the long-term value of SpaceX. I just hope retail investors don't get hit in the process. So, while we have you, Brian—longevity and biotech, one of my favorite stories. First off, congratulations on your recent raise at NewLimit, $435 million to push us toward age reversal.

If you don't mind, tell us the origin story of NewLimit, why you and Blake started it, and what you guys are doing there.

Yeah. Well, actually, Jacob Kimmel, Blake, and I all started it together as co-founders, and Jacob's the CEO now, so he's really—he's brilliant.

He is.

Yeah. So, the origin story goes back to the IPO of Coinbase, believe it or not, in 2021. Having gotten some liquidity from Coinbase and thinking about the next 10 years—and what were going to be the big technology trends that could drive civilizational progress—I felt like a lot of the big ones had good teams working on them, right? AI, fusion energy, brain-machine interfaces, space, and all the rest.

One area that I didn't see great teams working on as much, and that maybe seemed underfunded, was longevity. So I reached out to a few folks and started hosting some dinners with top scientists and biotech CEOs. I went around the table and said, "All right, what's on the horizon that's most exciting to you, but it's underfunded?"

One of the people mentioned epigenetic reprogramming, which is this area that talks about how you can reprogram cells. Shinya Yamanaka famously won the Nobel Prize for this in 2012, showing you could reprogram an old skin cell into a young embryonic stem cell. By doing so with just these 4 proteins added to the cell, he had this remarkable discovery that he changed both the age of the cell and the type of the cell.

At NewLimit, we're trying to do half of what Shinya Yamanaka did. We don't want to change the type of the cell; we just want to change the age. What we mean by that is really just restoring the function that the cell had when it was younger.

We built a high-throughput screening system. It starts with AI to explore the 10 quadrillion different possibilities and combinations of different proteins you could use to reprogram a cell. It started by ingesting all the existing literature. Now it's been ingesting the wet-lab data that's been coming out of NewLimit, which has the largest data set out there by far. It recommends the next set of experiments that we run in our wet lab.

We do these pooled screens and see if, phenotypically, we can make cells look younger. Then we take the best hits out of that funnel and put them through functional assays, which are a little slower and more expensive, to see if they actually act younger. For instance, we might take a reprogrammed liver cell and see if it can process caffeine, acetaminophen, and alcohol like a young liver cell.

The best functional-assay candidates are then starting to go through nonhuman-primate and human trials. The process has gone faster than I would have expected. I thought this was going to be a 5- or 10-year basic-research endeavor, but we've actually been able to demonstrate successful reprogramming of human cells. Our first drug candidates are going into the clinic next year, hopefully followed by a bunch more.

Amazing. Human trials next year, and you and Blake put in the first $100 million. I remember that. I remember coming to your lab and seeing it. Full disclosure, I'm also a shareholder and investor in NewLimit. Very proud of that. Congratulations on the progress.

Other news related to epigenetic reprogramming: Life Biosciences, another one of my portfolio companies, announced they've dosed their first patient with their OSK treatment today. So, yes, we're on the verge of longevity escape velocity.

Brian, what do you hold as a target for when we hit LEV? Do you have a number in your head? Do you have a goal? We need Kurzweil to come plot another curve for us, don't we?

He's got his prediction. His prediction is 2033.

Huh.

I'd have to go look at his data on that. I haven't looked at the exact year that it would be projected to happen. I'm just trying to solve more mundane problems, like how we hire the next bioinformaticist or whatever.

But I hope he's right. He has been right on so many things. It wouldn't surprise me if he's right again. Alex Wissner-Gross

I have an idiosyncratic position. Perhaps unsurprisingly, I will go out on a limb and say that I think LEV is going to be spiky in the same sense that AGI is spiky, and that on certain spikes, it seems possible it might be achieved later this year.

The reason why I say that is—I wrote about this in my newsletter, and I know you immediately said, "Oh, but—" In the past few weeks, there was a really interesting paper published on the first placebo-controlled, double-blind study of HIV patients receiving GLP-1s, looking at epigenetic clocks. I know all the caveats: epigenetic clocks, or Horvath-style clocks, aren't as good assays at determining biological age as a variety of functional measures. I know all of that.

Okay. Nonetheless, with all of those caveats out of the way, for the first time, to my knowledge, in the literature, I'm starting to see evidence of some sort of measurable—at least double-digit—age reversal as a result of GLP-1s.

That makes me ask whether, now that we're seeing retatrutide and others, and the Chinese have their own, I think, second- and third-generation GLP-1s, we're going to start to see something hidden under our noses. In the same sense that we flew past the Turing test without remarking that AGI basically passed the Turing test, will we, with a whimper and not with a bang—perhaps this year or next year, but as soon as this year—in a spiky way, in a subpopulation, fly past LEV without broader notice that it just happened?

Well, we're going to have some level of proof, right? We have this $101 million healthspan prize going on, where we're measuring functional reversal of age.

Right? Again, the biomarkers right now—you can test your biomarkers on a Horvath clock and many different clocks. The problem is, if you test yourself in the morning and test yourself at night, you'll get different answers. So, I don't think there's a reliable biomarker for aging.

Brian, I don't know if you agree with that, but what I care about is less a number on a page and more functional. Do you have the liver function that you had earlier in life? Do you have the cognition, immune function, and muscle that you had 20 years younger? I think we're at the beginning of that curve, and just like in the singularity, we're living through it. I think we're in the process, and I agree GLP-1 drugs are probably one of the very first longevity drugs out there.

Alex

Brian, any thoughts? Or maybe a question to Peter and Brian on this. Do you think we're going to be arguing years from now? Let's say it's not this year; let's say it's the early 2030s. Are we going to be arguing over whether we've passed LEV or not, bringing folks onto the pod years from now saying, “Oh, no, it definitely hasn't been hit because my favorite vanity benchmark hasn't been passed yet,” while others will say, “No, it was actually passed 5 years ago”?

Yeah, I like your theory. I like your theory on that, Alex. I would agree with the Turing test comparison: we probably will go past it and nobody will react. I think that's probably likely. I don't think it'll happen this year or next year, but 2033 could potentially happen. The FDA approval part is the slow part, right? The AI part is going really fast.

I will fly anywhere for the treatment.

I want to make a couple of comments here. As we talk about LEV, we won't know for a long time because people won't be dying for quite a long time. We won't really know when it's hit except through certain medical tests and so on. I think this is going to drag out by definition.

My favorite comment from the life-extension world is that the baby that's going to live to 1,000 years old is already alive, and that just blows your mind.

I want to say something I've been wanting to say for months and months, and I can only say it on this episode: As we talk about life extension, there have been 2 things true in the world for the history of humanity—death and taxes. We may crack death, and Bitcoin is going to crack taxes. So, there you go.

I love it. I love it when a brilliant successful entrepreneur from outside of biotech gets into biotech because you get that fresh view, kind of—you know what is possible. You’re not jaded, you’re not calloused, you’re just like, what is possible? I’m really curious to know what you buy for $435 million, and does it surprise you, like where that money goes?

Yeah, well, on the idea of going into biotech, I mean there’s definitely some Elon inspiration there, which is like, we’re living through a golden age of software where fortunes are being made. And so I think these hard-tech problems deserve more capital. SpaceX probably would not have existed unless someone like Elon, who had had an exit, had at least some of his own capital to go try and do it. Same thing with Tesla. Venture capital sometimes doesn’t get you all the way there. It’s almost like to do these real moonshots, you have to find someone who made a billion dollars in software and then go found the moonshots. But I hope that’s not the case forever. I think a lot of the money that’s been made in crypto and AI will actually go into the next moonshots.

Well, $435 million—does that automate a bunch of wet-lab tests, or does that give you concurrent, many lines of research? What does it run?

Run a human trial. It’s trials—a lot of trials. I mean, in a software company, most of your costs are people and then some AWS and stuff like that. In a biotech company, typically it’s people and then materials, like reagents and things. But in this case, now that we’re getting ready to go into the clinic, your average Phase 1 trial might be, I don’t know, $10 or $20 million, and you can go up from there on Phase 2 and Phase 3. So you need to get enough shots on goal to get some of these candidates to come out the other side and have a big impact. So we’ve got enough capital now to run a number of trials, which is good.

Alex Wissner-Gross

Alex, I love this tweet I saw from Sam Suare: “Starting to mourn all the people who died before the singularity.”

Yeah. I'm mourning all the people who've died, period. And I think, to Brian's point about all of the grand challenges that there aren't yet billionaires solving, I'll throw this out to the universe, to the economy: I would love to back or otherwise support someone building a company to use advanced compute to digitally resurrect everyone who's ever died. I think it's a tragedy that we've had so many billions of deaths over the millennia. Let's fix it.

I love that. I love that. All right, our next story here is a fun one.

Our last 3 stories in the field: Columbia University researchers successfully edited a PCSK9 gene—this is your LDL, your bad cholesterol—and an HBG gene for hemoglobin in embryos in vitro. This is opening up the conversation. I remember when I was at the Whitehead Institute doing my medical degree and the first gene-editing capabilities using restriction enzymes came out. There was a lot of hand-wringing around embryo editing and zygote editing.

In 2018, a scientist in China, Hi Janu, successfully edited the CCR5 gene in two young girls who were brought to term healthy, to make them HIV-resistant. The question is, when this gets reliable enough, are we going to start editing our kids? My son Dax just did a project on this at school, and his point was, “We give our kids the best food, the best education, the best friends, and the best clothing. Why not start with the best genes?” I am curious what your thoughts are, guys.

Yes, of course. This is much more advanced than even Gattaca. The whole plot of Gattaca was that it's still your genes, your parents' genes—just the best combination, with extreme in vitro selection of embryos. This goes beyond that to do base editing of embryos.

Base editing itself is looking tremendously promising. It was invented by David Liu and a collaborator at Harvard. It's better than first-generation CRISPR: you're able to do single-nucleotide swaps in DNA with minimal error rates. This is tremendous, and I can't for the life of me understand why this wouldn't become a widespread practice.

Obviously, shadows of eugenics come to mind, but at the end of the day, if you're getting rid of hereditary disease, or if you're a family below 5 feet and you want your kids to be taller, the ethics of this are going to evolve. Morality around this is going to evolve, and societal norms are going to evolve.

Brian.

Yeah. Well, I think one of the bigger complexities with this particular issue is jurisdiction. Suppose your parents are in the U.S. and you go to a Caribbean island, and that Caribbean island says, “Yeah, do whatever you want.” Then you've got a baby, and you bring that baby back to the U.S. That's inevitably going to happen.

So regardless of what your local laws decide they would like it to be, it's not going to actually work locally. That's a really big complexity in this whole area.

Brian, do you want to weigh in?

Yeah, it's a great point. I think this is going to happen, too. I think it'll be good for humanity on net. I think a lot of people will freak out about it for a while, but I think there was a Pew Research study I saw that something like 80% of Americans would support embryo editing for disease prevention.

That was more than I expected. I thought 80% was pretty high, but only about 20%, I think, supported it for “enhancement.” The part that I've never been able to figure out is, what's the line between disease prevention and enhancement?

Not having a disease sounds like a pretty good enhancement to me. There are cases where, let's say, osteoporosis is a disease. But if you had a gene edited to prevent it, you have stronger bones. Stronger bones sounds like an enhancement.

I basically think the line is so blurry that people will start with this for disease prevention, and then it'll upgrade over time. I think anything's on the table, even IQ or whatever. It's like, should we have more smart people in the world? Probably. That seems like a good thing to me.

And ethics change, right? Because if you remember, a while ago, IVF was considered immoral in the early days.

And now it's enabled millions of families to be formed.

Well, but not to state the obvious: if you're 7 feet tall or higher, you have a 50% chance of making the NBA. And if you make the NBA, that's a $20 million-a-year paycheck.

So if you're a couple, just a regular, everyday couple in the US, and you have the option to have a 7-foot-tall child by design, you're going to go to that Caribbean island and make your 7-foot-tall child for economic reasons. Think of what could go wrong then.

There are going to be some serious guardrails. There have to be guardrails around this.

You know, there's a company I love called Nucleus. Kian Sadeghi, the CEO, is going to be at my Abundance Longevity trip in October. Brian, I'm hoping to have someone from NewLimit there as well.

What Nucleus does is, you fertilize 20 eggs—20 zygotes—and then you sequence them, and you can choose the one out of the 20 that has the attributes, instead of playing dice and waiting and hoping for the lucky sperm. You can actually pick the one that has the right attributes from within the gene population that you have.

It's separate from base editing, a step before that, but still something you can do today. Remember the line from Gattaca: “Your child is still you, simply the best of you.”

Except that's not even true with base editing. It's better than the best of the parents.

Well, Brian, listen, thank you for joining us today. I'm grateful. I know you've got the world's largest crypto company in the world to run, so I appreciate your time, my friend.

Yeah, thank you for having me. This was a great conversation. I love talking with people who want to build a future.

Yeah, for sure. And that is most definitely this amazing group.

Our next story here is Anthropic launches Fable 5 and Mythos 5. Andrej Karpathy, who has joined our friends at Anthropic, is now shilling for the newest models. Here's his quote: “Super exciting release. Fable 5 is the same underlying model as Mythos, but with added safeguards. You can 10× your test suite, auto-optimize code, run giant research projects with custom HTML for the results, anything.” Alex, over to you.

Alex

Very impressive release. Anthropic is back in the lead now. Until today, GPT-5.5 was the lead, the state of the art across most of these benchmarks. Today, for probably about 5 minutes, Anthropic takes back the crown across most measures of superintelligence.

A few notes. The distinction between Fable 5 and Mythos 5: Mythos is, as reported, a less inhibited version of Fable. Or conversely, Fable is a more inhibited version of Mythos that has certain scaffolding and other restrictions that prevent it from having more ambitious conversations that might relate to biology, chemistry, cybersecurity, perhaps other areas.

But I've used it. It's incredible. I gave it my typical benchmark, which is Favorite Eval, asking it to one-shot a cyberpunk FPS, a first-person shooter, that's visually stunning. It did an amazing job, with no errors. It ran immediately and had a nice soundtrack that came with it.

What's more interesting, if you look at the benchmarks and then also the demonstrations—far be it for me to suggest that Pokémon games are actually the secret to superintelligence—but if you look at the demos that have been coming out from Anthropic and elsewhere, it is able to play games just from pure visual reasoning.

Whether this is Fable 5 or Mythos 5 doesn't really seem to matter. Based on just watching the screens of games like various Pokémon games, it's able to win them, and that requires long-term, long-range reasoning capabilities that we really haven't seen until now.

So my guess is Anthropic has probably been very aggressively doing RLVR—reinforcement learning with verifiable rewards—on long-range reasoning challenges. Maybe game playing, maybe challenges that involve spatial reasoning, maybe very large codebases, maybe very large capture-the-flag, or CTF, hunts—very large, sprawling RLVR challenges, both in space and in time—to get some of these results. It's just incredible what you can one-shot now with Fable 5, with reasoning set to high.

Again, it's difficult to predict how long Anthropic will retain the crown before, say, GPT-5.6 comes out, but it is exciting to see the frontier move once again.

David Friedberg

Dave, you've been playing?

Yeah. Well, it's only been out for about 4 hours, but I've been playing for 4 hours. What we're living through—Alex is constantly posting on our internal feed—it's happening. It's happening. I mean, it's really happening now. This is—oh my God.

You know what's interesting to me? A bunch of things. First of all, they doubled the price, so it's expensive as all hell. This idea of commodity intelligence is not happening. It's damn expensive if you want the most brilliant thing.

But what it produces is so incredibly intelligent that it takes you quite a while to read and understand what it did. And it's like, well, I'll just move on while you're reading and trying to catch up.

That's an interesting new thing for me. Also, I think this is interesting: Anthropic had it in the bag. Alex just mentioned that the leapfrogging is going on, but actually OpenAI pushed out their best, and Anthropic then said, “Okay, well, we had this ready to go.” So we're just trying to make it safer, but now we'll come out and go over the top.

And we've been saying they're pulling their punches. They are. They're gaming this.

As you're heading towards your flight, do you want to add anything here?

Alex Wissner-Gross

No. What occurred to me here is that you have the split now between the safety guardrails and the performance guardrails. I thought that was really interesting because you've got the same underlying model with added safeguards, right? That's the whole safety debate in one sentence right there.

David Friedberg

Well, actually, they even changed the name, too. You've got Mythos, which we've been talking about for a while, and then Fable is the heavily guardrailed version of the exact same model. It won't do a cyber threat; it won't do a nuclear weapon. They even differentiated it by name to support exactly what you're saying.

Alex Wissner-Gross

With pretty broad buffers, friends of mine have tried having perfectly innocent conversations with Fable 5 High about biology, and from what I've heard, it has pretty broad margins for what it will simply revert back to Opus for. If it thinks you're asking anything in biology or chemistry that could remotely be a dangerous subject, it'll just downgrade you immediately to Opus.

David Friedberg

Funny you say that, Alex, because I'm often trying to work from a plane, and SSH doesn't work on the plane Wi-Fi. So I use a workaround to try and get access to all my remote agents.

That also vomited, saying, “It looks a little bit like maybe you're doing something sketchy,” and it gave me the Anthropic “We're not going to answer that query” message repeatedly. I had to actually find a workaround for that, too. So, yeah, they've really locked it down.

I predict that as we head towards IPO season for these 2 companies, these frontier labs, we're going to start to see the rate of model releases increase. They're going to be leapfrogging. Who's going to be out front on the day of the IPO? Who's getting the news?

Alex Wissner-Gross

At least, Peter, we have competition. The situation could be worse. We could just have a singleton, a single lab that is not getting leapfrogged at all. At least we have 2 competitors at the moment in the lead.

Yeah, for sure. All right, our final story for today: after more than a decade of sucking, Apple finally made its move to improve Siri. They've announced a multiyear partnership with Google to power Siri with Gemini AI, rebuilding Siri from the ground up. The big shift: Siri is now an agent, not just a voice interface. The key breakthrough is personal context. Yes, finally, Siri can now reason across your messages, your emails, your notes, your photos, and actually get things done.

And it can actually, when I'm sending a text message to Kristen, spell it correctly, or spell my name correctly. It doesn't most of the time. They're calling it a persistent AI workspace, coming to beta later this year. It's focused in English.

Two huge implications here. First, Apple is essentially admitting it lost the frontier-model, or foundation-model, race and chose Google to rent brains rather than build its own. And it's a stunning concession from the company that has always prided itself on its own stack. Second, the real moat was never the model; it's the personal context. So, Alex, do you want to share different perspectives here?

Alex Hormozi

One, the glass-half-empty perspective is that Apple, for arguably the first time in modern history, has outsourced a key element of its tech stack. For Apple, just as with Elon and vertical integration, it's not in its cultural DNA to want to rent out someone else's technology that's going to be so core to the system. That's the glass-half-empty perspective.

The second perspective says the foundation model is not analogous to, say, an operating system. It's more analogous to the role of a search engine, which has to be localized. So you'll note Apple didn't announce this in China yet. It didn't announce it in the EU yet, for different reasons.

But focusing just on China, I would expect Apple to strike some partnership that's roughly analogous to the partnership with Google with one of the 3 big Chinese AI labs or tech companies. And that will represent a localization roughly analogous to the way Apple localizes its other web-search capabilities in China with one of the Chinese companies, analogously to the way it defaults to Google for web search.

The glass-half-full argument is that foundation models, however foundational they are, are intrinsically localizable, and so Apple has merely chosen Google for its American franchise. Maybe it'll choose another franchisee for Europe and another one for China. Google has been doing this with Android and its suite of products for a while now, right? So this is Apple catching up. I've almost moved to an Android phone so many times just because of its ability to contextualize all of my personal context.

A third perspective—yes. A third perspective is, if you buy that foundation-model economics are hyperdeflationary and that the cost is going down by 40x year-over-year, or anything remotely like that, then you should view this entire bit of Apple losing its tech sovereignty as a nothing burger.

Apple may be struggling to distill Gemini down to Private Cloud Compute, combined with the iPhones, this year, but in a year or 2, there will have been so much hyperdeflation in the cost of compute that the present capabilities of the new Siri will easily fit into the new iPhones. And as for any loss of sovereignty or any use of external compute—we haven't even touched on this, semi-embarrassingly—it's not even the case that Apple is just able to distill Gemini, Google's Gemini or its version thereof, into the iPhones.

It's not even the case that they're able to operate it with a combination of Apple's own cloud and the iPhone edge compute. They actually need Google's cloud to do this because it's so compute-intensive. But all of that over the next year or 2 can shrink and hyperdeflate down to just working on an edge device as these algorithms get more advanced.

Yeah, Dave, any thoughts on Apple?

Dave Blendon

Yeah, lots. Well, anytime in tech that you don't do anything for a long period of time, you eventually get crushed. The question is, how does it emerge?

The way it would play out, if Apple doesn't do something aggressive, is that you lose the interface. Siri is now Google, but Google is Android, and Android is the one competitor to the iPhone. Remember, all the revenue still comes from the iPhone. So you lose the interface.

At the same time, TSMC moves the chip manufacturing away from the M5, M3, and M4 to a much higher-paying customer in AI data-center use. So Nvidia or Elon take the manufacturing capacity away for the underlying chip because they just pay more to TSMC. Then you've lost your hardware manufacturing under the covers, and you've lost the interface on the other side. Everybody just moves to Android because they're talking to Google or Gemini all day long anyway.

The physical phones then get manufactured by Samsung, using Samsung's chip-manufacturing capability, which is immune to TSMC.

Salim Ismail, I want to acknowledge your persistence and your ability to stay in the conversation.

This is a new low. I'm in my airplane seat. So let's just let it go past us.

Okay, closing thoughts on Apple and Siri.

Your closing thought? That it's good? I'll believe that it's good when I see it.

Yeah, we've been so disappointed over and over again. You and I met the founders of Siri before they sold to Apple. It had such promise back then, and it just stuck around way too long.

Gentlemen, I think that's a wrap. I enjoyed having Brian here, going from Bitcoin to longevity. Anyway, crazy week ahead. I'll see you guys in a few days for our next episode. It is the most extraordinary time ever to be alive. I like to say, during the singularity, don't sleep, don't blink.