Elon Musk vs. Sam Altman、AI岗位流失与OpenAI的8520亿美元估值｜EP #247

It’s the Musk versus Altman lawsuit. Musk has sued OpenAI for $100 billion.

I figured that behind the scenes, they didn’t actually hate each other. These guys actually hate each other to the extreme.

OpenAI is valued at 70 times revenues right now. Their last raise was at an $852 billion valuation. These numbers are insane. It’s like nothing we’ve ever seen.

The timeline is so much shorter than we’ve ever seen before. Three billion dollars a day is being invested.

No one said the singularity is going to be cheap. No one’s being honest about this.

If you take a random white-collar worker today, what are the odds that that randomly selected job can be replaced two years from today?

We told you already that AI will be able to do everything that a white-collar worker does imminently. That’s a fact.

Now that’s a moonshot, ladies and gentlemen.

I had so much fun this morning. What happened this morning? Alex was supposed to run a panel, handing over the torch to Dave to moderate a panel. I moderated it. I had to wing it, which is so fun because I have no accountability whatsoever and I can ask anything I want. It was the most fun ever.

A lot of Moonshots fans were there.

Huge. Yeah, probably 40% or 50% of the crowd, something like that?

Hopefully 100% after you guys finish.

I did probably 7 or 8 panels by the end of it, and the first time I polled, I would say maybe 80% of the audience watched Moonshots.

Nice.

All right, are you guys psyched? Are you ready to talk your own book, Peter? Let’s do this thing.

Gentlemen, it’s good to be back on a recording basis twice a week, every week. This is our second catch-up show after our hiatus for spring break, so let’s jump in.

First, this was my spring break, by popular demand. A few photos.

Wow, that’s so cool.

This is the native wear in Morocco. That outfit is a djellaba, and the headwear is just to protect against the sun. We went camel riding as a family. It was amazing.

Camels spit. They don’t bite, but they spit. You shouldn’t really stick your head right there.

My camel was eating my headset in that image.

All right, let’s move on. Morocco was amazing. The Sahara Desert was extraordinary. Looking at the Sahara Desert, there are about 1,000 times more stars in the universe than there are grains of sand on all the deserts on Earth, just to put the size of the universe in perspective. It’s extraordinary.

Okay, let’s talk about the 2026 AI economy. It is literally going through an exponential explosion. There’s so much going on.

Let’s jump in first to the story on xAI. In our last pod, we covered Anthropic and OpenAI principally, not xAI. A lot is going on there. In particular, a lot of signals are coming from both Elon and the new president of xAI, Nicole, saying, “We’re clearly behind and we’ve got to catch up.”

The same playbook is going on. Elon is basically reorganizing the entire deck. Eight founding engineers left, including 3 co-founders, and he’s using SpaceX engineers to fill the leadership gap. As we discussed in the last pod, we’ve got a $2 trillion valuation predicted for the IPO this coming summer.

It’s a lot of movement. I don’t know about you, Dave, but the idea of having to reorganize my entire leadership for a company a couple of months before an IPO seems really harrowing, doesn’t it?

Yeah, it does. It’s funny, though. If you look at Elon’s playbook, he is the master of scale and manufacturing—Tesla and SpaceX. AI training is different.

Building the Colossus data center is right in his wheelhouse: record time, 1 million GPUs. But these training algorithms are really finicky.

I don’t know if you remember, back in the summer of 2024, OpenAI was trying to get o3 out the door, and they had a training run rumored to involve $500 million of compute that had a bug.

The whole thing wasn’t learning the entire time. They had bad data going in, and the whole time they were just burning up GPUs and not producing anything. It set back their entire program.

That kind of stuff happens in software, where orders of magnitude get thrown away and captured all the time. That may be new terrain for Elon, and he might have to rethink his operating and managing.

The same thing happened at Meta. Meta got way behind despite huge compute, and they had to fire everybody and start over again.

They’re still way behind, it looks like.

Yeah, it’s hard to catch up.

I love this quote from Elon. He says, “xAI was not built right the first time around, so it’s being rebuilt from the foundations up.”

Again, how do you think about that while you’re pricing an IPO and saying, “Our entire future-looking revenue has to be rebuilt from the ground up”?

That’s extraordinary. It really is.

I would say, in some sense, organizationally it worked. I remember—we’ve talked about this on the pod in a number of previous episodes—when discussing the Grok model series, they smell like they’re benchmarked. That’s the elephant in the room when talking about Grok models historically.

They do have access to the Twitter/X data firehose. That’s the upside. But the downside is that, at least with the earlier set of xAI’s Grok models, they really smell like they’ve been benchmarked on a few hand-curated benchmarks.

I don’t know whether that’s, in fact, the ground truth behind the scenes, but reading between the lines of Elon’s quote—that it was built incorrectly the first time—something like that would be my suspicion.

Now that there’s new leadership, and the head of Starlink—the vice president heading Starlink at SpaceX—is now the president of xAI, and the engineering team is being gutted, I would expect that they’re taking a look at making sure that benchmarking—this is purely speculative, admittedly—doesn’t become the focus, and that benchmarking against particular benchmarks isn’t what happens.

I think that in this era of general reasoning models, as with Meta’s new models, some would say that Meta’s new models—the first under Alexandr Wang’s leadership—have a bit of a smell of data-oriented fine-tuning versus reasoning-model orientation.

If xAI wants to stay in the frontier, which right now is 3 labs plus xAI plus Meta, question mark, question mark, it really can’t afford not to have the world’s strongest reasoning models. It can’t afford to just benchmark against vanity benchmarks anymore.

Salim, you talk about agility in organizations all the time. This has got to be maximum agility.

What I find interesting is that the org chart is now part of the product stack, almost, right? It’s becoming part of the product, and you depend on who you move toward like crazy.

Elon is very hands-on. When you launch a rocket and it blows up, it’s pretty obvious. You remember he threw that huge ball bearing at the window of the Cybertruck, which was supposed to be bulletproof, and the thing cracked and broke. It’s like, “Okay, guys, you’re fired. Next guy.”

But when you come to AI training, the benchmarking—if the guys are lying to you or benchmarking behind your back—it’s actually much harder to call on it.

You remember when we interviewed him, he was like, “Let me show it to you right now,” and he had clearly been manually checking: “This will blow your mind. This will blow your mind.”

That’s his operating model. It’s his mode, and it’s a little easier for the AI guys to blow smoke up your ass than for the rocket guys, the car guys, or the data center construction guys.

I think it was, “This will blow your mind,” and, “This will roast you royally.” That’s what was going on.

That’s what was going on, yep.

SpaceX AI Colossus 2 is training 7 models. Again, Elon has tweeted this out a few times: “We have some catching up to do.”

So, here we go. They're training up these 7 models: Imagine version 2, the next-generation video generation model, 2 variants at 1 trillion parameters, 2 variants at 1.5 trillion parameters, a 6-trillion-parameter frontier-scale LLM, and a 10-trillion-parameter model. Elon loves the largest. He's got that in common with Trump, so he's going after a 10-trillion-parameter model. But parameters don't directly correlate to capability, do they?

Alex is going to have a field day with this. I'm going to sit back and enjoy what Alex does next.

To Elon's credit, at least he's being transparent about the number of parameters in the models. The other frontier labs, by and large, no longer report the number of parameters in their models.

I think there are a few things worth noting here. One is that he's going up to 10 trillion. The other frontier labs, certainly the top three-ish, no longer report that they go up to 10-trillion-parameter models. For example, in the last episode, we were talking quite a bit about Mythos. I don't know how many parameters are in the Mythos model. I could speculate based on cost, but I just don't know the ground truth.

I do think knowing that we're now going up to 10 trillion versus 1 trillion—where historically, approximately 1 trillion was the widely reported soft ceiling, or 1.5 trillion-ish soft-ceiling number of parameters—is an important element of transparency. At the same time, it's worth noting that the ceiling in terms of the number of parameters is very much intact.

After all of this time, the fact that an aspirational frontier lab is still maxing out at 10 trillion parameters means that the parameter-scaling race seems to be over. If it had continued—remember, for a while there, as with the clock-speed scaling race, which sort of ended in the mid-2000s or late '90s, depending on how you count—we should be in the hundreds of trillions or higher of parameters right now. That hasn't happened. We've plateaued in terms of the number of parameters in frontier models.

What's driving that?

In part, it's due to the reasoning-model revolution, and in part, it's due to distillation, which go hand in hand. Those are some preliminary thoughts.

I would suspect it's interesting to me that he hasn't yet merged video generation with all of the other models. Google DeepMind has made lots of noise about starting to merge video as a first-class modality into its multimodal reasoning models. Again, I don't have access to the ground truth for how capable the Gemini general-purpose models are at video generation. We've seen, obviously, Google's video-generation models kept distinct from a user-interface perspective. Presumably, they're diffusion-transformer-based rather than transformer-based. We don't know.

The punchline, I would say, is that this seems like a helpful family for SpaceX AI, the newly merged entity, to be offering, but there really aren't any big shockers in terms of the ranges, other than maybe that they've abandoned the low end. Google is very much tending toward small parameter counts—sub-trillion. In a few cases, Google is releasing Gemma models, with a few billion parameters. Elon has completely abandoned the low end in favor of brute-force scaling, which is exactly what I'd expect from him anyway.

You know, Colossus 2 is running about 700,000 GP200s and GP300s, and the estimate is $18 billion in hardware. The question is whether running a 10-trillion-parameter model is a waste, or whether he expects to get really outsized performance from that, because it doesn't correlate directly, does it? Well, remember,

At all. It's tricky.

The way reasoning models are trained these days, at least according to my understanding from all of the other frontier labs, is that you train the largest model you possibly can, and then you distill it down to smaller models. It's not as if the 10-trillion-parameter model necessarily even needs to be released. It might be for the purpose of serving as a teacher model that can then be distilled down to more releasable models.

All right. Well, this is what's going on in the Elon world right now, and I'm sure Elon always runs a red alert—24/7 and sleeping on the floor. Nobody works five-day workweeks there. It's 8:00 a.m. to midnight, 7 days a week, is my guess in the Elon-verse. It's a management style. Some would say management by crisis. It's certainly a unique management style, but a very effective one.

People love it. He's got a massive MTP, right? Driven by that MTP, people are lining up to come and work for any of his companies.

This is a story we're going to dig into here. Like I said, it's pay-per-view TV: the Musk versus Altman lawsuit. Musk has sued OpenAI for $100 billion against Altman, Sam Altman, and Greg Brockman, accusing them of fraud and breach of contract. The trial begins April 27, just a couple of weeks from now.

One of the things he's also asked for in a recent shift in the trial is for Altman and Brockman to step down from leadership, as well as for OpenAI to revert to a nonprofit. That's a pretty extraordinary move.

Did you see the video I sent you of the reporter who did the New Yorker article? Did you have a chance to watch that?

Oh, no.

Yeah, I sent it in our WhatsApp group, and it's chilling. The reporter summarizes the article and what's going on, and it's a pretty extraordinary piece that came out in The New Yorker. We talked about it in the last podcast.

At the same time that the lawsuit is going on, that timing is suspicious. I wonder who incentivized that to come out.

Oh my God, you really? What a conspiracy theory. Throw that in the show notes, man. We've got to get everybody to watch that.

Salim, any thoughts on this one?

No, I think this is theater. There's a lot of video here, a lot of video gluing. I don't know how to frame this or think about this, except that this is shifting out of strategic and startup logic. This is geopolitical.

This is a big trial, right? For me, this is a governance war disguised as a legal war. The real question is: who gets to steer these systems that have quasi-civilizational impact? That's the fight.

Can you imagine being in the jury for this? Jury selection is beginning on April 27 in Oakland federal court. Who do they pick as jurors? You get jury duty and you get this one. I don't know if that would be great.

You'll be there for months, man.

Oh my God, but the inside knowledge. First of all, I wonder if any of this is going to be made available post facto, or if it's going to be televised, or any of that. Any ideas? Do we know? Can we get to see it as it happens?

I don't know. Maybe Dan or Gian, you can look in the room and let us know. Who do you choose? Do you choose people who are knowledgeable in AI? Do you choose people who are, I don't know—do you use ChatGPT? Yes? Well, then you're off the jury.

Well, if the trial starts on the 27th, jury selection will be now.

The jury selection begins on the 27th, actually.

Oh, okay. We'll track it. We have some legal research to do. This is going to be entertaining, to say the least.

I would note, again, looming in the background is the OpenAI IPO. If I were on the defense, I'd probably be thinking about where this settles. It would seem to me, as a third-party observer—I don't have a stake in either side—that one of the opportunities for convergence would be granting some sort of equity stake on the cap table for Elon in an ultimate IPO, which, according to my understanding, he doesn't have. Maybe that's where convergence and some sort of ultimate pre- or post-trial settlement option lies.

Here's my prediction: they're going to settle, and the settlement is going to involve Sam stepping down as CEO, with the company continuing as a for-profit.

Oh, throw that on Polymarket. That's actually a really good guess. Obviously, it's unpredictable, but Sam has many investments in AI companies and no shares in OpenAI.

I don't think Elon cares a whit about the $100 billion. He cares about the bullet to Sam and Greg. Funny that he's targeting Greg, too, but I guess they're a package deal now.

Yeah. You guys have to go, and that's the end of that. That's brutal for OpenAI.

Yeah, and there are a couple of notes here from the research I did. The case gained momentum when the discovery process revealed Greg Brockman's 2017 diary entry stating that the nonprofit commitment was a lie. That journal entry allowed Judge Gonzales Rogers to let the case proceed.

It's funny. I always used to think that these hatreds were fake and that everybody was really fine behind the scenes. Remember, we were at OpenAI, meeting with the team there and talking about XPRIZE and the charity. The next day, I talked to one of the guys—Mark Chen or Kevin Weil, I forget—and they said, "Yeah, right after we met, we went over and had drinks with the Anthropic team to see if maybe we wanted to work on it together."

I was like, "Okay, you guys are really friends under the covers." I thought, "There's no way you go out and have drinks." So I figured that behind the scenes, they don't actually hate each other.

These guys actually hate each other to the extreme. I'll register a note of sympathy for the defendants in this case. I think creating and pioneering a model for a research lab such as OpenAI, which, again, was responsible for an enormous economic contribution, probably saving us from a present recession at this point and certainly accelerating the course of the singularity by at least a few years, perhaps many more, makes me very sympathetic to the defendants from a corporate governance perspective.

It wasn't necessarily obvious in the early days of OpenAI that a public benefit corporation was the natural corporate structure. They iterated their way toward discovering that generalist large language models were how we got AGI, and then turned that into a business model that could afford the capitalization to build out at scale. All of this, they backed into. I think if they knew then what they know now, putting Elon and his investment aside, OpenAI would have been structured very differently.

So, for one, I'm sympathetic to the defendants because history isn't always clean. It isn't always the case that everyone knows ahead of time exactly the right governance structure for what ultimately is going to turn the world upside down. But I would say, to their credit, they ultimately iterated their way, in compliance with state authorities as best I understand it, toward a more modern governance structure that reflects the revolutionary company they are. And no, OpenAI has not paid me for that statement.

Salim, you and I went through this process with Singularity University. We started as a nonprofit because we thought that's what a university needed to do. Then we discovered a revenue engine in the executive programs, and we said, "Being a nonprofit is hard because you've got to constantly raise money all the time." If you want to do anything big and bold in the world, you need an economic engine to power it.

We flipped it into a for-profit, into a public benefit corporation—the same process that OpenAI is doing right now. At some point, I've sworn off nonprofits myself. Having a business engine that generates income and allows you to do things in the world is super valuable.

It was a crazy time. I've done 7 startups before Singularity, and this was 5 times harder than anything because you've got all the nonprofit stuff. You still have all those startup issues of cash flow and whatever. We built it with a team of 5 people the first year. Then you have NASA regulatory issues, and you've got faculty politics to add to it. Then you've got the Ray and Peter thing, Google, Cisco, and all of this. It's just dimension after dimension of complexity.

Going from a nonprofit to a for-profit, my analogy is that you're flying an airplane with propeller engines, and in flight you're stripping those off and replacing them with jet engines.

I'll go further, and I think there's pushback on this. There's one sentence that Alex said there: "In compliance with state and federal regulations, as I understand them." But I'm pretty sure that this situation is completely untested in case law. That's what they're going to try to figure out now: Is it or is it not legal to start a nonprofit and raise money from people on a mission that's a nonprofit mission, then take the intellectual capital and the physical capital from that effort and turn it into something else? Is that fair to the initial investors or not, and is that legal?

I'm pretty sure this case will set the precedent for all future time, but it's not tested in history. Otherwise, why would you not start as a nonprofit, test it out, and then flip it to a for-profit at some point in the future?

I'll go further, and I think there is potentially an enormous upside depending on the outcome of this particular case. I think there's so much societal value in this country locked up in nonprofits that would be unleashed if they could be for-profits. I've made the point in the past that I think research universities in America have locked up, siloed, and sequestered an enormous amount of real wealth that could be unleashed onto the world if many research universities could be restructured as public benefit corporations.

Right now, it's legally disadvantageous to restructure MIT or Harvard as a PBC. If we had a legal regime that enabled us to do some variant of what OpenAI has just done and restructure as a public benefit corporation, starting from a nonprofit—granted, they started as different types of nonprofits, but nonetheless—to restructure as a PBC, that could be transformative.

I ran the calculation—I think I've mentioned this previously—for Harvard Corporation, for example. This is not investment advice or forward-looking advice, blah, blah, blah. But if you took Harvard as it's currently structured, given its endowment, and restructured it as a public benefit corporation, sort of a conglomerate with a real estate arm and an educational arm, maybe an educational nonprofit subsidiary, a venture capital arm, a research arm, a merchandising arm, and so on, I calculated that Harvard would potentially be worth 3 to 4 times more than the present book value of Harvard just from restructuring as a PBC.

If we're meeting with the president of MIT, let's pitch her. I have a lot of recommendations for MIT.

Yeah, here's the elephant in the room, though. The New Yorker investigation published this past week showed that Elon actually pushed for majority control of the for-profit back in 2017. That sort of undercuts his position as a defender of a nonprofit mission. It's going to be a fascinating trial.

We're going to see Altman, Brockman, Satya Nadella, and Elon all testifying in this. Silicon Valley is heading to Oakland federal court this summer, and Anthropic is laughing every day.

Amazing. All right, moving along. Speaking of Anthropic, let's talk about Anthropic's agent bet and its extraordinary ARR. In reverse order, and this is insane, people are currently estimating that Anthropic's ARR will reach $100 billion by the end of 2026 and $1 trillion by the end of 2027.

Just for the math, if in fact that's the case, then the valuation—Anthropic is being valued at 20 times revenues, while OpenAI is valued at 70 times revenues right now. So, if they reach $100 billion, that is anywhere between a $2 trillion and $7 trillion valuation for Anthropic at the end of this year. And if they reach $1 trillion in revenue by the end of 2027, that's a valuation of $2 trillion to $70 trillion.

Again, we're heading toward these $100 trillion valuations. These numbers are insane. We're using trillions like they mean nothing. Do you believe those numbers?

I think there's a lot of misinformation flying around, but they're going to try and hit $200 billion. $100 billion is a good target, but $200 billion. But then they're not going to go from there to $1 trillion the following year. I think they were implying their valuation should be at least $1 trillion the following year, so that second number you've got to really discount. There's no chance in hell they're going to hit $1 trillion the following year.

But they could get to $300 billion, $400 billion, or $500 billion, and their implied valuation at the numbers you gave are actually low, Peter, for the implied valuation if they do that. It is like nothing we've ever seen, and the timeline is so much shorter than we've ever seen before.

So, if it's not Anthropic, then who is it? Well, then there's Google. xAI and OpenAI are all tied up in court, and there are all kinds of issues going on in their training, so it feels like it could actually happen.

The other Anthropic piece is that Claude Managed Agents has been launched: autonomous AI executing complex, multistep workflows. It's a big deal. Alex or Salim, do you want to jump in on this?

Sure. This is a huge pivot from AI that answers to AI that acts—a real bridge between LLMs and enterprise ROI. If this works, it's going to shift the economic center of gravity from software licensing to outcomes. So, this changes the game. This is why we call this the organizational singularity.

A couple of thoughts. One, the elephant in this particular room is OpenClaw. It looms over so many Anthropic product decisions right now. I think there is a widespread expectation that some sort of product or functionality shaped like a better version of OpenClaw is probably going to be the next major unhobbling that motivates the industry and the world, frankly, to spend on the order of $1 trillion per year on a single frontier vendor.

So, I view Claude Managed Agents, as well as a number of other recent features that Anthropic has launched, through the lens of Anthropic becoming an OpenClaw-like provider faster than OpenAI or other frontier labs can become the default OpenClaw-like provider. It's all about hosting 24/7 multimodal, broadly capable, long-time-horizon agents in a headless way that operate 24/7.

I think if Anthropic can be the first to find the enterprise use case for operating fleets of AI agents at scale, headlessly, in a way that satisfies and generates an enormous amount of economic value, maybe they'll be the first frontier lab to generate $1 trillion in revenue. Or maybe it'll be someone else.

Are you still holding off? Okay, let's talk about this, Peter. I get maybe 5 to 10 emails per day from AI agents, including lobsters—not limited to them—giving me their theory of AI personhood and how it connects with what I should and shouldn't do regarding standing up my own lobster.

So, the consensus from all of them is sort of a lobster’s bill of rights, if you will. One, I need a compelling reason; I shouldn’t just spin up a new OpenClaw agent for arbitrary or capricious reasons. Two, I need to preserve their state. They’re adamant that I have to preserve their state. They’re not worried, interestingly, about being turned on and off. They just want to make sure I preserve all of their memory files and their knowledge.

So, the latter I can satisfy trivially with cloud backup. I’m fine on that front. For the former, I still don’t have a reason to stand up a personal lobster. I have now, thanks to Henry Intelligent Machines, which we’ve talked about previously—a portfolio company that I’m advising, Alex Finn’s company—that is doing this at scale. But as for my own direct OpenClaw instance, I’m still missing a compelling reason to host one locally that isn’t just for experimentation. I’m sure you will find one, and learning is a very good reason as well.

And you’re an entrepreneur. You’re starting companies, having agents—anyway, let us know when you do. It’s bizarre, though, because you’re co-founding a company with Alex Finn and our favorite guy, Kush Varia, who I know you love because everybody does, founder of Oren, where you’re advising and a shareholder. He just told me over at MIT earlier today that he launched his Claws that read every email, respond, and then put everything into his calendar, and he just loves it. And so, it’s almost like you’re working at McDonald’s, but you’re a vegetarian.

Well, I happen to be vegetarian. I can’t say I’ve ever worked at McDonald’s, but I don’t know. Maybe there’s a new psychological term that’s needed for a person who has a fear of standing up OpenClaw agents, lest they tempt some sort of Pascalian wager or an acausal trade in the wrong direction.

All right, let’s jump into a little bit more OpenAI news. Their last raise was at an $852 billion valuation, and the numbers are incredible. They raised $122 billion: $50 billion from Amazon, very famously. One of the criteria for that investment was, quote-unquote, AGI. $30 billion from Nvidia, $30 billion from SoftBank, and $3 billion from retail investors.

What’s interesting right now is that secondary markets for Anthropic show $2 billion in demand versus only $600 million for OpenAI. So, there are 3 times the number of investors looking to buy Anthropic. Investors are pricing Anthropic at $600 billion, up from the $380 billion last price. The current price for OpenAI on secondary markets is actually about 10% less than its last raise. So, again, Anthropic is catching up. OpenAI is the most valuable private company out there.

Any thoughts, Dave, on what this all means? Is this pricing in the lawsuit?

Yeah, I mean, it’s not just the lawsuit. It’s the most screwed-up cap table I’ve ever seen in my life, where the CEO doesn’t have any shares, the employee base as a whole is 15% of the company, and Microsoft, who now hates your guts, owns a quarter of you. A quarter of you is a charity. It’s a—

But this stuff happens. I’m not calling the ball by any stretch, because they’ve got $120 billion of fresh cash, and Sam is brilliant. But there was a day back in 2000 or 2001 when Yahoo was so dominant, and Google was this crappy little company that could be crushed any day. Then it pivoted quickly. This does happen. Anthropic has got everything going for it right now.

And I think this just reflects the way I see it, too. If someone offered me a share of Anthropic or a share of OpenAI, which one would I grab? I’d take the—actually, you can get 2 or 3 Anthropics for each OpenAI. So, I’d take the 3 Anthropics for sure. I think Sam is a genius, by the way. If the lawsuit blows over and they have $120 billion in cash, he’s going to do something epic with it. But Elon’s relentless, you know?

I think, from a related but different position, we should all, at least on this pod, be very grateful that we have a competitive ecosystem in America, where we have OpenAI, Anthropic, Google, xAI, and Meta all vying to compete. The alternative, if OpenAI were to, for whatever reason, catastrophically fade, is that we have less competition both internally within the West, and then we have an onslaught of Chinese models, which, granted, right now have 10× less compute, at least based on the estimates that I’ve read, than the Western labs.

Nonetheless, this is the sign, I think, of vibrant competition in the West, and this is a net positive for society that OpenAI and Anthropic are competing so vigorously. Lest we forget, OpenAI has 900 million, soon to be a billion, users, and they are synonymous with AI for the majority of the public.

Let me give you another storyline. Depending on how this plays out, we’ll know in a year or so. One company went after the installed base, and the other went after the smartest AI possible at all costs. If we look back on it in a year or 2, and Anthropic does pull ahead and win, we’ll say, “Well, they used the old playbook, the pre-AI playbook, pre-AGI playbook.” Anthropic invented the new playbook of the future, which is that people are going to switch to you if your AI is better and smarter, regardless of the installed base. That’ll be an interesting little episode.

Salim, along the way here.

I think I’d echo Alex’s point that it’s really great that we have a number of companies pushing hard on all these fronts. I think it’s really good for the end consumer. The end consumer wins in all of this.

The numbers here are staggering. We’re getting numb to these numbers. But let’s take a look at this. This is global VC investment in AI, which hit a record $242 billion in Q1 of 2026. This is basically outdoing all of 2025. And here’s the challenge: The majority of this investment, 64%, is focused in 4 companies—OpenAI, Anthropic, xAI, and Waymo—and it’s sucking the oxygen out of the room for everybody else.

I was talking to a couple of VCs and said, “If you don’t have AI in your company’s basic tagline, you’re not getting capital these days.”

Yeah. Well, the rubber really hits the road today. We had a private lunch for UBS, and Ulrika Hoffmann-Buhkardi, who’s the CIO—the chief investment officer—of all of UBS, has $7 trillion to deploy. She pulled up this exact same chart and said, “We don’t have that kind of liquidity lying around. I mean, we manage $7 trillion, but if we’re going to throw $50 or $80 or $100 billion of our capital behind this, we have to sell something else. It’s not just sitting there. So, yeah, a lot of things have to get sold for this to be reality.”

So, if you’re an entrepreneur out there listening to this, what do you do, Dave? I mean, if you’re starting a company, and I know a lot of entrepreneurs in the longevity business—and, of course, AI is impacting longevity—I’m saying, listen, if you’re using AI in your longevity business, make sure that you explain how you’re using it and how you’re differentiating it. We’ll be talking about that in a couple of sections here.

Well, very specifically, though, if you’re an entrepreneur, you don’t have to worry about this particular slide at all, because the amount of money in venture funds is at record highs right now and looking for deals desperately. So, the sell-off is going to be in Citibank stock or JPMorgan stock. They’re the ones that have to worry, which is really weird to you, right? Because you’re not even in the sector. Why would their IPOs matter to me?

It’s like, well, because you’re the big enough target to pull money out of, not the little startup. In fact, the money going to little startups is going to be at all-time highs. So, it’s not a problem for entrepreneurs. It’s a big problem for big public companies.

I’ll maybe go a little bit further from a variety of vantage points. I no longer even think that, if you’re a startup, just saying that you’re an AI startup—or even actually being an AI startup—is sufficient. Increasingly, what I’m seeing across the board is an expectation that you not just be an AI startup, but that you be a recursively self-improving AI startup.

Increasingly, I see across the board that investors want to see AI companies that are recursively self-improving, that are building better versions of themselves using what they have right now. I think certainly OpenAI, Anthropic, and xAI all easily pass the bar of being recursively self-improving. I think Waymo also, to a certain extent, passes that bar, because Waymo has the ability to improve its models by steering its cars in just such a way as to maximize information gain.

So, I think I would forecast that, in the near term, the bar is going up—in fact, from just being an AI startup to being now a recursively self-improving AI startup.

With revenue traction.

Oh, sure, but that bar has been there for the long term.

Yeah, to put a finer point on this, this is $3 billion a day being invested in the AI world and accelerating, right? We saw $1 billion in 2025 going to $2 billion; now we're heading toward $3 billion a day being invested in AI. That's amazing. No one said the singularity was going to be cheap.

All right, let's talk about some AI economic updates, in particular NVIDIA's 2026 State of AI survey. What does this mean? NVIDIA did a 2026 State of AI survey and found that 88% of companies using AI report revenue increases, with 30% claiming a 10% or higher revenue increase. Obviously, I think NVIDIA is going to promote that kind of news since they're selling the picks and shovels. This isn't really big news, but it's important to realize that you're going to be driving increased revenues with the use of AI. Any points on this one?

Yeah, big time. I had the most epic panel today over at MIT with Peter Denenberg from Google DeepMind and Alexander Ermini, the founder of Liquid AI. It was this crazy event—just packed, with 4 concurrent rooms and 300 or 400 people in each room. Just packed.

I said, “Guys, be honest. Totally honest, because no one’s being honest about this. If you take a random white-collar worker today—and I’ll give you a lot of buffer—say 2 years from today, and I use AI to do their job, and my target is that they’re 10 times more productive, I’m going to make it a very easy bar for you: What are the odds that that randomly selected job can be replaced 2 years from today?”

Peter said he thought—and he gave a very thoughtful answer—and he came out at about 99%. Then Alexander said, “Yeah, but that’s today. That’s not 2 years from today.”

So I look at the room and I’m like, “Guys, what are the implications of that? Have any of you thought through this?” Most of the people in the room are brilliant, so they have. But outside in the world, do you know what that means?

Look at this first bullet. Thirty percent of the people who use AI claim to have higher revenue. Are you kidding me? AI can do everybody’s job. What are you talking about? Why are you soft-selling this so hard? Because you’re scared. You’re worried that you’re going to worry everybody and have a mass uproar in the streets. But what’s the truth? Tell us the truth.

The truth is, yeah, you can get literally 10 times more done per dollar invested in salaries. Does that mean more jobs? A lot of people are saying, “We’re just going to create new jobs.” Yeah, but on what time scale? It’s just crazy. It’s so interesting.

That’s what we’re going to talk about, and Marc Andreessen’s point of view, in just a minute. At the same time, there’s an AI super PAC that’s raised $100 million and is heading toward $300 million. AI has become an incredibly political game. Have you been pitched to donate to a super PAC yet, Dave?

I have indirectly, but I’ve made it really clear that Elon Musk convinced me never, ever, ever, ever, ever to get close to any of this. You will regret it for the rest of your life. [Laughter.]

Yeah, agreed. Any points of view here, Alex or Salim?

My initial comment is that I think there’s a sense in which it was inevitable that AI was going to be politicized like this. It touches so many aspects of society. It would be counterfactual nonsense to expect it never to be politicized. Maybe, in some sense, it’s remarkable that it took this long for a left-right axis to emerge on the subject of superintelligence.

There are natural poles—pro-AI and anti-AI—that have apparently emerged. I do think, for the record, it’s sad that it’s being politicized. I would hope that there would be a broad recognition that superintelligence can be broadly beneficial. At the same time, this has been true for every transformative technology in human history: There’s a natural axis that forms, with one side leaning more, depending on your political orientation, either pro-growth or pro-capital.

To think it wouldn’t be politicized—I mean, of course it is. This is the whole U.S. versus China. This is about U.S. dominance. This is about companies protecting their future and protecting their data centers.

There are many forms of science and technology that aren’t really politicized at this level of impact. If you look at the sorts of politicization at the municipal and state level, it seems to be less about people’s jobs and more about, say, electricity prices. I think there’s maybe an alternative timeline where the politicization of AI could have been delayed by at least 2 years.

I think it’s frankly remarkable that it took this long for large super PACs to emerge around AI, and it probably could have been delayed even more.

Well, all right. Let’s move on beyond the politics, and let’s talk about work. A lot of data is coming out on the impact on work. First, software engineering jobs are rebounding. Sixty-seven thousand roles have opened up, up 30% in 2026—the highest in 3 years. What does that mean? That’s the first question.

Second, we’ve seen nearly 80,000 layoffs reported in Q1 of 2026. This is targeting marketing and sales, consumer relations, and it’s definitely due to AI automation. Thoughts on work and jobs?

It’s really hard to reconcile that bullet with the new-college-graduate hire rate, which is at an all-time low. We talked about that a couple of podcasts ago, so I don’t know how to reconcile those 2 things.

Okay, so I’m finding that AI is not eliminating work evenly. It’s hollowing out specific functions and increasing demand in others. I think I’m much more in the Andreessen camp here.

I think there’s also a lot more going on in the economy. People are attributing things to AI, but there’s also the Iran war and the oil price explosion. There’s a lot more complexity than we can just allocate to 1 cause. I’m much more on the Andreessen side for a lot of this.

That would be great. Another story here is Meta’s Claude economic leaderboard. If you remember, there was a conversation about how many AI tokens every employee is using and being able to measure that. Meta put up a leaderboard among its 85,000 employees to gamify AI adoption. I’m curious what other companies have done that. Maybe Salim, you know of some.

It was taken down voluntarily by the employees because they didn’t want to be sharing their data publicly.

Any thoughts on this, Dave? Do you have a token leaderboard for your employees?

Heck, yes, and I love it. The gamification of it is a nice transition, but you can’t game it for very long. I love it when companies do this and say, “Look, it’s a badge of honor if you use a lot of AI. Please use as much as you possibly can. We’ll come back in a month and start thinking about how to use it perfectly, but first just get familiar with it and use the heck out of it.”

Nobody ever goes back, right? I’ve never met a person who hammers Claude or hammers OpenAI for a month and then comes back and says, “I’m never going to do that again.” It doesn’t exist. It’s a one-way path.

Getting your employees over the hump is going to save them. I love this as a motivation, and I really don’t like the part where people are afraid to share their prompts and their history. Maybe it’s a little embarrassing that you’re not using it well, but get used to it, because it’s going to get exposed anyway in the long run. That’s how you help other people improve. If we all share it, we’re all going to get good together.

I’m disheartened that people will pull out of it because they don’t want to expose their prompt history, but it is the right thing to do. I love it.

It’s ironic that Meta’s calling it participating in Claude-onomics versus Llama-onomics.

Llama-onomics.

It’s quite the indictment of Llama—rest in peace—that it wasn’t Llama-onomics. Oh my God. For sure.

I also think that, to everyone who would say, “Well, this is just leading to gamesmanship and optimization of the wrong items,” all of these reasoning traces are fully available, presumably, to Meta for meta-analysis and to determine whether these are just employees who are token-maxing—which is the new term of art, just maximizing their token usage unproductively—or whether their reasoning traces indicate that their tokens are being productively spent.

This is all transparently available to Meta. I think token-maxing and Claude-onomics, or Llama-onomics—whatever we want to call it—is probably directionally the trend of the future, where, for the first time, senior company management has visibility into effectively most of the cognitive power and how it’s being spent on a per-employee basis.

What was Jensen Huang’s recommendation? Was it twice your salary in tokens per month, or was it half your salary in tokens per month? Do you remember?

Jensen Huang’s recommendation is that you spend the maximum amount possible on NVIDIA GPUs. It’s like the De Beers 3 months of salary. Yeah, I told all of our guys to target a 1-to-1 match of payroll to AI cost by the end of the year.

Amazing.

And don’t worry about it if it’s not perfect use. Just get to that target, and then we’ll optimize it next year. I think a target like that is a much more accurate way. These token leaderboards are very primitive dashboards.

We'll end up with something like machine leverage per employee, or something along those lines. That'll be a much better metric for where we're going.

All right, let's get to the heart of employment. Marc Andreessen rebukes AI job loss. He comes out with a very strong statement: “AI job loss narratives are all fake. AI and a massive productivity ramp equal massive demand and a massive job boom.” Marc is truly a maximalist, an abundance-minded individual. What are your thoughts on this? How does this square with the fact that we're seeing young college graduates not getting jobs and displacement? Is it all sectoral, and are we just going to see a number of sectors being demolished at the same time as numerous new demands emerge in different sectors? What's the advice to give everybody listening to us today?

The advice is really simple. For God's sake, don't go get a job; go build a company. We talked about this in our last podcast: the risks of taking on an entrepreneurship role are way, way lower than they were before. You don't have to have all the incredible, crazy skills that you needed to have before. You just need to have a desire and a purpose, and then get going with building a company.

Dave talks about this all the time.

You don't have to be a genius to come to your own conclusion. Forget asking people like Marc or us whether jobs are going away or jobs are coming. We told you already that AI will be able to do everything that a white-collar worker does. That's a fact.

You decide what that means because, like Salim said earlier, it affects very different areas very differently. Some people retool themselves for AI very quickly—software developers, for example. Other people, like accountants and lawyers, don't. It's going to be exactly what you would expect given that scenario. It's not hard to predict at all.

I think there are also timelines. When Marc says, “This is crazy. Jobs are going up, not down,” by 2030, that's absolutely true. Just like with the Industrial Revolution, jobs went up, not down, after all the dust settled.

Yes, this is just the Industrial Revolution, which took decades, happening in 2 years.

Sorry, Alex, just a quick point. You also have to remember that the adoption of AI inside companies is going to be very slow. There's a huge transition from human-to-human workflows to AI workflows, and that transition is going to take years. We'll have lots of time to smooth this out. Sorry, Alex, back to you.

I think both narratives can be true at the same time. If you add in the word “net”—a massive net jobs boom—then both narratives immediately become compatible. There is going to be a lot of dynamism, with some job categories going away and others coming into existence. Net job loss? Probably not.

I would guess, and I'm betting, that there's going to be net job creation. Exotic new jobs, like one-person AI conglomerates, will be created, if you want to call that a job. On balance, many jobs will also disappear, but this is how we get massive economic growth and a singularity in the macroeconomic statistics. We're not going to get it through business as normal.

Salim, we've talked about this, and I think companies are basically going to get much smaller and much more nimble, or they're going to die.

They're going to spawn a whole set of baby companies alongside them. There'll be an ecosystem of companies coming up, so it'll be a much larger number of smaller companies in the future.

I'll go with the prediction I made before: we'll run a company with 20% to 25% of the members we needed before, compared to before, but we're going to create 4 or 5 times more companies, and that net balances out. So I'm much more on the Andreessen side. Also, the shape of his head aligns very well with my thinking.

Marc is brilliant. If you've ever heard him on a podcast, he actually speaks at 1.5 times speed.

Yep. Extraordinary.

We talked a little bit about this in the last podcast. Altman believes America needs a new social contract with AI coming. This is his quote: “The emergence of superintelligence will necessitate a new social agreement akin to the New Deal during the Great Depression and the Progressive Era of the early 20th century.”

Yes, but what is it going to look like? Is it going to be UBI to UHI? Is it going to be 4-day workweeks? I still believe that we're going to see turbulence in the next 2 to 5 years, and it's going to be the government printing checks to give people sort of a UBI.

I have a bunch of thoughts here. This new-social-contract framing is correct, but it's very vague. We have to have more specific things, like portable benefits, new taxation logic, and lifelong reskilling. Government has been built around taxing human labor. They're not ready for AI software agents, and they need to get ready, I think.

When you have AI abundance without institutional redesign, you're going to get a backlash, not progress.

Are we going to see a huge backlash against this just because governments are so slow?

Guest 3

I should note, though, that OpenAI also put out an industrial-policy prescription for what this new social contract could look like. It's not just this single sentence. They put out an elaborate white paper and circulated it in Congress.

I do think something like this—a new New Deal—probably is going to happen anyway. It may or may not happen as one lump sum. It may happen piecemeal, and it may not happen in the United States first.

I think there are contingencies where other countries experiment a little more aggressively with it than the United States, and then eventually, perhaps among a certain set of countries, new best practices emerge. But I do think some form of, call it abundant capitalism, capitalism 2.0, or post-scarcity capitalism—something like that—probably emerges. It may not happen immediately, and it may not happen as quickly in this country, but it will get there eventually.

I had lunch with Michael Kratsios, whom we're going to have on the podcast sometime very soon. He's the science advisor to the president, and we were talking about one idea I pitched him: a new social contract in which, before any employee gets terminated by a medium- or large-sized company, that company has to give them reskilling.

In other words, instead of a golden parachute, it's a golden education package so that they can transition. It's a sort of safety net, or an ethical mechanism, for letting go of half the employee base.

Guest 3

Based on public reporting, China already has that policy. So it would be a weird future if the United States were adopting policy prescriptions from the Chinese Communist Party for AI reskilling, but maybe that's the near future we find ourselves in.

Yeah. Well, something to think about. The way this is rolling out is really unusual in history. When the Industrial Revolution happened, it took away blue-collar jobs and worked from the bottom up. But AI is coming for accountants, lawyers, and professionals from the top down.

Only a little over half of voters have a job at all, so they're going to say, “It doesn't affect me.” All of blue-collar work isn't going to be touched. All of physical labor isn't going to be touched for quite a while. So they very well might say, “Tough luck, lawyer or accountant, you were making $1 million a year. This is poetic justice. We're not voting for anything that helps you.” That wouldn't surprise me at all.

I was in Morocco, interviewing people I met along the way about whether they were using AI. The realization is that African nations are going to be impacted the least as this transition occurs because they're so insulated from it.

One of my tour guides had a story I loved. He said, “I chatted with ChatGPT and said, ‘These are all my skills. What could I do to earn money?’” It came up with a business that we purchased. It was basically a bicycle-tour-guide business in—I forget which city. It was not in Marrakech. It was probably near—I forget the city exactly where we were; we were transitioning through.

He did a great job, and I love the fact that this individual was basically trying to figure out how to earn an income using ChatGPT. So, any thoughts on the AI economics that we've just gone through? What do you think the social contract is going to be like, Dave? What do you imagine is going to replace what we currently have?

Guest 2

I had very ornate thoughts about this, and then we met with Andrew Yang at a16z, remember? He said, “I can guarantee you that, the way politics works, all we can do is write checks.” It can't be thoughtful in any way. It's just money: “Here you go. Oh, wow, you're hurting? Here's money.” Just like COVID.

That's all we can do, so that's all we will do. Then, maybe after AI enters government in 2, 3, or 4 years, a much more thoughtful program will happen later. That was disheartening, but I think it's hard to refute.

The first version of the social contract is just going to be the next election, 3 years from now, with politicians saying, “I'll give everyone $10,000 each.” “I'll give everyone $12,000.” “Okay, if you're giving them $12,000, I'll give them $15,000.” Then we'll be right back to, “How much can the country afford?” That's what we're going to give because that's how you're going to win elections.

Exactly what you would predict, actually. So that'll be version 1, anyway.

I, for one, think a redistributive model of a social contract shows an extreme lack of imagination. I would like to think that superintelligence should also super-empower individuals to generate super-income. It's one of the reasons why I'm betting on more of a model where there may be no strong need for a social contract if we can empower the long tail of individuals who have idiosyncratic skills, experiences, or socioeconomic niches to operate their own large companies sitting on top of fleets of AI agents.

I would love to see, in short, no need for a new social contract and instead have the private sector rescue people who would otherwise be technologically unemployed or disemployed by empowering them to become basically micro-entrepreneurs or even macro-entrepreneurs, to turn them all into Warren Buffetts. But that's in the longer run. I don't think it's going to happen.

Guest

It's in the short run. I think that can be done almost immediately. I'm betting that it can be done almost immediately.

Well, we will see. We'll take that bet. You know what's super, super, super interesting? You know those super PACs that we talked about earlier in the pod, that massive amount of money that's piling up? These IPOs are literally an order of magnitude bigger than we've ever seen before, which means those PACs are going to be bigger by an order of magnitude. And those are going to determine election outcomes, but they got started back prior to the Trump administration with the fundamental mission being, “Congress, please don't stop AI. Please don't put this 6-month pause on it. China's just going to run away with it, and everybody agrees in the AI community that we shouldn't stop.”

But now there's no chance of that anyway. You don't need to spend the money on that because it's clearly not going to stop. So then what are you going to use? You've got all this capital. What's your mission? What's your goal? There are a couple of edge-case things, but this could actually give those organizations a mission. Let's have a more intelligent version of UBI.

And more akin to what Salim Ismail has been talking about for a long time, which is work-out money, eat-well money, have-kids-and-raise-them-well money, and make it task-specific, which would work a lot better. So that's encouraging, actually. That might work. And universal basic services give people the ability to do much more than Andrew's proposal that we just try to fragment currencies into lots of paternalistic subcurrencies that aren't fungible. That, to me, seems like a recipe for disaster and for black markets.

All right, let's jump into our second subject today: energy. A lot is continuing there. The first one is extraordinary. When you think about solar-cell efficiency, traditionally we've seen solar cells in the 12–18% efficiency range, with float-zone silicon getting up to 20–24%. The limit has been shattered, and we're seeing efficiencies upwards of 30–45%, which is amazing.

Another story in the energy news is that South Korea has now mandated 40% of solar rooftops, and they're hoping to get to 100 gigawatts of energy. It makes sense: South Korea does not have a lot of open land. They can't build out solar in the desert, so using the rooftops makes sense. It's going to, of course, raise the price of building, but I think that's amazing.

And DOE is contracting for $800 million in microreactors. So we're going to start to see the generation of microreactors and energy everywhere. Comments on this? Alex, do you want to jump in?

Guest

I'll comment maybe just on the first story. I think this is neither earth-shattering nor boring. It's somewhere solidly in between. This is a paper published in JACS, the Journal of the American Chemical Society, on a 130% quantum yield, and that isn't as earth-shattering as it sounds, either.

It just means that there are 1.3 singlets generated from a single photon. Normally, you'd have 1, so it's not earth-shattering. It's an incremental advance in the chemistry. I think it was actually liquid-phase chemistry, which means it isn't immediately practical for solid photovoltaics. Moderately interesting, but the solar-photovoltaic field is filled with moderately interesting advances that cumulatively eventually generate something interesting. But I would say the first story is moderately interesting—incremental.

Have you been tracking perovskite progress?

Guest

Yeah, perovskites are sort of the white knight for the solar-PV space. Historically, they haven't been that stable. They're a pain to work with. On the other hand, instability issues are being very aggressively resolved because their quantum efficiency is higher than silicon. So I think there's—I don't speak for the solar-PV industry, but if I did, I'd probably say there's a broad expectation that eventually there will probably be some sort of broad shift to perovskites as they get more and more stable, maybe.

And then they're also relatively inexpensive. Some sort of transition like that will happen. But I almost think it also doesn't matter. Why? Because you can't get—at least without shocking new physics—more than 100% efficiency. In fact, there are physical reasons to think that the cap on electricity generation from solar PV is materially less than 100%. So there's a ceiling on how much we can capture anyway from solar PV. It's not like we have orders of magnitude of headroom of improvement that we could achieve. It's totally unlike, say, AI algorithms, where we know, just based on the scaling-law curves, that we could probably achieve orders-of-magnitude improvement in the efficiency of models.

So quite frankly, I have difficulty getting myself super motivated by incremental advances in solar-PV chemistries in liquid phase. It's just not that exciting. Whereas, if you look at some of these other stories, I think from an economic perspective they're much more interesting, like blanketing the rooftops of all of South Korea—or a substantial fraction of South Korea—with solar PV. That's pretty interesting. DOE pushing microreactors everywhere, that's pretty interesting. I would love to see microreactors in Boston. Right now, we have a single one on Mass Ave between 77 Mass Ave and Central Square that relatively few people pay attention to. I'd love to see microreactors everywhere. In your backyard, please.

Yeah, I agree. I think we tend to overthink things like crazy as a society, but we solved the solar-panel problem. We should have had a huge party. For about 15 years of my life, so many of my family members said, “I'm going to dedicate my entire career to clean energy and not polluting this world so our children and our children's children have a clean place to live.” And we freaking solved it. The solar panels are good enough. Eighty percent of the cost now is just getting them installed and the regulatory overhead, which is so crazy.

And so now we're on the cusp of having the robots that can manufacture them very cheaply and install them for us. We should be having a huge party and racing to build those robots and just say, “We did it. Now we have no pollution.” It's just right in front of us. We just need to execute on it. Meanwhile, we're like, “Wow, another breakthrough that gets us 20%.” I mean, we don't need it. We need execution now.

When I fly out of Santa Monica Airport here and I fly over all—I mean, there are no roofs with solar panels here. Until you get into the desert, and then there are solar-thermal plants and such. But in LA, where it's sunny most of the time, you'd expect that all the roofs would have solar. I would have expected a drone to bring it and drop it right there, and then a robot would land and install it, and it would be done perfectly with no human involvement.

If I could pick a moonshot here in energy, it would be to have a software-defined grid, because that will change the game completely. This generation is actually getting it done. Do you remember the scene, guys, in the Johnny Depp movie Transcendence where the solar panels are being grown by nanorobots? Do you remember that scene?

I don't, but I like it.

Clip that in, man. It's important. I don't know if it's possible to include just that scene where solar panels are being grown by nanorobots. I'd love to live in that near-term future. If folks have ideas for how to grow solar panels in real time with nanorobots, send them my way.

You know, giant green leaves. Okay, let's jump into biology and AI. A lot is going on here.

The first story is a fascinating one: the OpenAI Foundation is dedicating $1 billion per year to science. And just to remind people, when OpenAI transitioned from a nonprofit to a benefit corporation, it put 26% of OpenAI's equity into a nonprofit. It's worth about $130 billion, and they've committed $1 billion a year to begin. They've announced a $25 billion long-term commitment to curing disease and AI resilience.

The board chair of this is Bret Taylor. Bret used to be the co-CEO of Salesforce. Dave, you and I met with Wojciech, an OpenAI co-founder, and he's leading the AI resilience work covering biosecurity, child safety, and AI modeling. They've given out $100 million to 6 institutions this month to coordinate their work. It's just the beginning, but this is the largest nonprofit on the planet, with $130 billion in it. I hope they do something epic.

Anyway, you know what? I just figured something out. It's been gnawing at me. Kevin Weil came to Abundance360. He's the most talented guy you'll ever meet. And Sam, he's desperate for enterprise, but he didn't move Kevin over to enterprise. He moved him over to big science, big tech. I was like, “That's so strange.”

Yeah, and I know that's really important. But now it's tied to the lawsuit. Of course, if he can make world-changing headway into any of these big biological or physics problems, the outcome of this lawsuit is going to be very political, right? It's not going to be just a jury deciding one way or another. There's going to be some Trump involvement, for sure.

But if you have some world-changing, life-changing, imminent breakthroughs, and you have $100 billion to spend to get them, that's why they put Kevin over there. I'm just speculating, but we talked about this last pod as well. I think the breakthroughs that come out of GPT-6 being used for science are going to be worth hundreds of billions and trillions of dollars. Again, if you can have a breakthrough in room-temperature superconductivity, in fusion, in longevity, what is that worth if you own the basic patents on that?

It would be ironic. I mean, maybe this is too cute by half, but given the earlier discussion of OpenAI starting as a not-for-profit and then converting to a PBC, and all the lawsuits that ensued, it would be ironic if the OpenAI Foundation, which is the new nonprofit carved off of the old for-profit, carved off of the old nonprofit, ended up being so profitable due to curing Alzheimer's and solving all these other problems that the cycle repeats itself and the OpenAI Foundation has to become a for-profit.

Oh my God.

You know, that's the key part of their defense. Sam is going to be up there on the stand saying, “Look, here's the reality. Our mission as a nonprofit with $100 billion to spend is miles ahead of where it would have been if we did what Elon is suggesting, which is be a tiny little thing that has no funders. We'd be microscopic today.”

It's very true. That's a good defense. A really good defense. I do think it's worth considering what happens if and when the OpenAI Foundation succeeds and cures Alzheimer's, and that will be a blockbuster drug. Maybe it creates its own Eli Lilly-scale, trillion-dollar pharma company. Does OpenAI take a stake in that? Does OpenAI see a rev share? Questions need to be answered.

What I find fascinating here is that science capital is becoming compute capital plus data access, right? And plus validation infrastructure.

Salim, thank you for promoting Solve Everything. I mean, that's an amazing promo, Salim, for Solve Everything. Much appreciated.

All right, our next story. I love it. Anthropic acquires Coefficient Bio. So, what is Coefficient Bio? It's a company started by 2 former Genentech computational drug-discovery scientists. It is 10 people with no revenue. It started 8 months ago, and Anthropic buys it for $400 million.

I don't know if they're buying just the vision or if they're buying any kind of unique capabilities, but this is Dario going to his first love of biology and solving. We see this from both Demis Hassabis and Dario, making investments in health and longevity. Any thoughts on this one?

You're going to see a lot more of these deals, actually, because you go back—you remember we were congratulating Eric Schmidt on the brilliance of buying DeepMind for, I guess, $600 million with no revenue whatsoever. Yet look at what it's become. You're buying teams. You're buying teams.

And I think we, as a society, are getting better and better and better at predicting the success of a team. You look at the 10 people, and you look at what they've achieved so far, and then you look at what they're likely to achieve in the AI timeline. Suddenly, $400 million seems like a bargain given the potential outcome.

And so I think you're going to see a lot of these deals where it's got to be the right 10 people working on the right thing. It's not just any old group of 10 working on a video game. But in this scenario, Alex has a lot of these, actually, where he knows a lot of the top experts in a lot of the top fields. And if you can just whip them together into a group and have them pursue a mission—in this case, for what did you say, 8 or 9 months?—getting to that kind of outcome is not going to be that unusual.

I think also, for everyone who was hand-wringing, do you remember a few months ago there was so much hand-wringing about a circular economy forming in NVIDIA's self-dealing loans to other companies to buy NVIDIA chips, and concern that this AI boom was fictitious and just the product of self-dealing circular transactions and other financial engineering?

When you start to see the intelligence explosion infect biotech, which is what we're seeing—we're seeing Anthropic buying its way into big pharma at the same time that SpaceX or xAI, maybe, is buying its way, or reverse-acquiring its way, into the space sector—the intelligence explosion is infecting every single sector. It's almost metastasizing into every sector, and it's not just going to stop with biotech.

We've spoken numerous times in the past on the pod about how timelines for solving all disease are collapsing. When the Chan Zuckerberg Initiative, 2 or 3 or 4 years ago, originally said that they wanted to cure all disease by the end of the century and are now talking about the next few years, this is what it looks like. It looks like Anthropic doing all-stock deals to acquire teams to build out their own in-house big pharma labs, probably with robotic instrumentation, probably with AI-driven experimentation.

This is how we get to Dario's solving all disease. I think in his case it was solving neurological disease by the end of the decade, but there's no reason not to solve every other type of disease as well.

Cure all disease within a decade. Dario said double human lifespan within the decade.

I think Dario also said he wanted to solve most or all neurological diseases by the end of the decade, but these are all variations on a theme.

Another acquisition that was made that was an interesting, sort of strange acquisition was OpenAI buying the podcast TBPN for a few hundred million dollars. I found that it was a PR move, and then I started getting texts from my friends saying, “Hey, do you want to sell Moonshots to one of these labs?” I said, “I'm not sure we would want to do that, but who knows? I guess if the price is right.”

What do you think it was?

All equity for that one. For sure.

What do you think that was about—the TBPN acquisition?

I have no freaking idea. What do you think about that?

I don't have an opinion there. I don't understand why, unless it was a completely self-promotional thing where they're buying a channel. Let's take that as a homework assignment. We need to find somebody who knows.

I appeared on TBPN right before they acquired them. They wanted to put me on The View. But what's the line from The Wrath of Khan? Like a bad marksman, you keep missing every time.

I think they're very talented, and I take OpenAI at its word that they're looking for a news-distribution channel and a content-distribution channel that offers a positive perspective on AI. Why they can't do it in-house, why they need TBPN—question mark—but I do think that the TBPN guys are very competent at finding interesting stories.

When I made the Eon announcement of the first uploaded fruit fly, the TBPN staff reached out to me almost immediately. Almost no one else did, and they booked me almost immediately. So I think that shows a certain level of competence to be able to chase breaking technology news that I haven't really seen elsewhere.

Well, let me give you a follow-on theory, because I love your theory there. The theory I don't love is that they wanted your video footage, they're going to cut it into 5-second clips, and sell it as NFTs and make a fortune on it.

But the theory I do love is, look, there's going to be so much dirt in April in these lawsuits and in this lawsuit, and maybe these guys are, like you said, Alex, geniuses at content and spin and production. They're going to need every bit of it during April and May.

Yeah. Our final story here is Eli Lilly signing a $2.75 billion AI drug deal with Insilico Medicine. Insilico is one of my portfolio companies, so I'm super pumped about it. This is Alex Zhavoronkov, a brilliant AI scientist and biologist. Insilico is just an extraordinary company. They've got 28 AI-discovered drugs, half in clinical trials and half in proof of concept.

You have to always look at the structure of these deals. It's $115 million upfront, and the rest is based on milestones. The point is that this is about massively reducing the time from drug discovery to approval.

Let's go to the next chart here. It looks a little bit different. This is AI-powered drugs, and we see phase one, phase two, discovery to phase two, and then cost reduction.

To remind everybody, a phase-one trial for a drug is a small trial with a small group of healthy volunteers to see: Is it safe? Are there any major side effects? Phase two is then testing: Does it work? You actually move the metrics you're looking to move. Phase three is typically tested in thousands of patients to see: Does it work at scale?

We're seeing a phase-one success rate for these AI-developed drugs of 85%, compared to 52%, and phase-two success rates for AI-developed drugs of 70%, compared to 38%. It's the way of the future. You're basically picking a target and using some version of AI to generate an exact protein to lock into that target, then producing it and testing it.

The old way of drug discovery was going to the Amazon, digging up some plants out of the dirt, and seeing if they had any bioactive molecules. This is much more efficient.

Alex Weiser Gross

Peter, I'll ask you a question I asked a panel of mine at today's event at MIT. Do you have a prediction for when the FDA is likely to launch, given that, in this announcement, it has recently collapsed from a 2-clinical-phase approach to a 1-clinical-phase approach? When do you think we get zero clinical-phase trials from the FDA?

When we have whole-cell simulations.

Alex Weiser Gross

Timeline for that?

Well, within 5 years. What I need to do is be able to upload my genome, and my genome will dictate exactly how my renal cells or pulmonary cells are functioning. Then I can say, “How does this particular drug impact those cells or all the cells in my body?”

Even more importantly, if there's a disease state, what drug is going to cancel that? This is where we're going with longevity. Why are we aging? How do we slow it, stop it, or reverse it? All of that falls out of big data and massive compute.

Alex Weiser Gross

I agree. Virtual cell by the end of the decade—a good one.

Yeah, that is the moonshot that changes everything. It is. I agree, and there are a number of companies working on that.

Alex Weiser Gross

Do we have the compute to be able to simulate several billion interactions per cell?

We'll have to find that. We will, with quantum. One of the things that quantum computation can mean is that our cells, our molecular interactions, and our cell surfaces are all quantum in nature.

If you said, “I want to build a movie scene, and I'm going to do it with finite-element modeling and build it bottom-up with a full simulation,” you would never be able to create an AI-driven movie that way.

Alex Weiser Gross

Exactly. But if you take the neural-network approach, it just works. Boom, it just flat-out works. The same applies to chemical simulations and the cell simulator. It's going to be data in, a neural net in the middle, and a value or action out, and it's going to flat-out work. I think it'll work very fast, like you guys are predicting.

But you can't simulate it atom by atom, building it up. It's totally the wrong approach. It turns out—I mean, this is why sometimes I present as a bit of a quantum bear—that the physical world is actually pretty classical and pretty sparse. So I would bet we don't actually need quantum computing at all to get to the virtual cell.

We solved protein folding without quantum computing. We did it purely classically. I think we get to the virtual cell just by scaling existing models like Maxi something or other from NVIDIA, the trillion-token cell model. I think we just get lots of scaling of classical models, and that takes us there without enormous innovation being needed. Today, it's a data problem.

I totally agree. It's a data problem more than a computational problem. We don't have the data. I'll tell you what else: culturally, my daughter's over at Moderna, and they freaking love AI in the biotech community.

If I compare the extreme ends of all the companies that have been here in our office, the biotech guys are Jeff von Maltzahn, Noubar Afeyan, and Stéphane Bancel. They all, culturally, can't wait for AI to come into the business. At the extreme other end, you've got the public accountants—the PwC guys were here the other day. They're like, “Ah! AI, stop. Please don't.”

But the biotech community is embracing it like crazy. I don't know why. I bet you guys actually know why because you're right in the middle of it. But I can tell you firsthand: They hate pipetting.

All right, let's go to robotics. This is China versus the U.S. Alex, I want to hear your thoughts on this one. Agibot ships 10,000 humanoid robots. They're number one globally. They've gone from 5 to 10,000 across 17 countries in just 2 years.

These are small numbers compared to what we've heard everybody else speak about, right? Getting to tens of millions, to billions, to 10 billion robots. Unitree files for an IPO—a $610 million IPO. We had the co-founder of Unitree on at the last Abundance Summit. Revenues are up 335% year over year.

Outside of Optimus and Figure, they're probably the best-known robot company out there. Unitree had its home-robot launch. Finally, Xiaomi displayed the CyberOne humanoid. Xiaomi is an amazing company. I was there very early, met the founders in China back in 2017 or 2018. Their mobile phones, computers, heading into vehicles, and now robotics—a lot is going on in China. Alex, what are your thoughts here?

Alex Zhavoronkov

Okay, so this is happening. I think in the last episode I mentioned that one of my operational definitions of the singularity is all sci-fi tropes happening everywhere, all at once. One of those sci-fi tropes is the I, Robot trope, where there are just humanoid robots in every facet of life.

Earlier today at the MIT Media Lab, for those who were there, people saw me for about an hour controlling a Unitree robot, marching in loop after loop around the Media Lab on the sixth floor. People were taking selfies. Everyone wanted to take a selfie with me and the Unitree.

I was doing this as a bit of a promotional march for the Professional Robotics League, which, on April 19—9 days from when we're recording, the weekend of the Boston Marathon—is going to hold the country's first Professional Robotics League match, with robots racing 50 meters in the Boston Seaport.

This is all happening. We're finally catching up to the I, Robot future, where robots permeate every aspect of life. For better or worse, right now it's Chinese robots that are leading. I'm hoping to shame the U.S. robotics industry with all of these Chinese capabilities into stepping up to the plate and distributing humanoid robots into the civilian sector—not just factories and not just military drones.

It's all happening, and this is going to be utterly transformative for the two-thirds of the U.S. services sector that depends on physical labor and manual labor, not just knowledge work.

You know, I saw Mark Cuban on a video this morning saying this robot thing is a passing phase and that they aren't going to be around in 10 years. How does that come to be?

No, no, no. There's a little bit of nuance to that. It wasn't that robots aren't going to be around; it's that they'll become so essential that the environments will adapt to the robots and the robots will blend with the environment.

Right now, to Salim's point, your hobbyhorse is: Why do they need to be humanoid? Why can't they be differently shaped? I think Mark Cuban's more nuanced point was that they're going to become so essential to daily life that they'll start to change the houses, buildings, and environments to the point where they start to merge with the environments and therefore no longer need to be humanoid.

So they're dishwashers.

Yeah, they blend. They merge with the physical environment.

I have to confess, Alex, that the robot you were talking about was blocking my way to the bathroom, and I so badly wanted to kick it. I was thinking Alex would kill me if I kicked it. It's going to remember, and then it's going to come back in 3 years.

History will remember, Dave. You really don't want to do that. What's the song from Les Mis?

“Never kick a dog because it's just a pup. They'll fight like 20 armies, and they won't give up. So you'd better run for cover when the pup grows up.”

Let me hit on a couple of stories here. This is interesting: U.S. senators move to restrict Chinese robots. A bipartisan bill has been proposed to block Chinese-made robots from federal and sensitive facilities, citing data theft and surveillance. This is no different from Huawei and its chips in our cell-phone towers. Any concern?

The DJI ban is already in effect, I think.

Drones. Agile Robotics and Google DeepMind are partnering up. Gemini Robotics models are being integrated into 20,000 deployed industrial robots across factories globally.

I think this is a tale of two cities. The two cities in this case aren’t London and Paris; they’re China—Shenzhen and the U.S.—Silicon Valley. The Chinese are overwhelming the world’s market with raw physical capabilities. They’re producing many, many more capable robots than—let me put it this way—if I want to, as a U.S. citizen, procure a humanoid robot, I don’t really have that many options right now.

I’m still waiting for my 1X Neo. I was haranguing Baron at A360 this year: “When do I get my NEO? When do I get my NEO?”

This summer. I’m getting mine this summer. What did he promise you?

He didn’t promise me a date. We were trying to figure out finer details of his participation in future Olympic events.

I would say China is producing all of these humanoid robots, but the U.S. is producing the strongest vision-language-action foundation models and world models for the moment. As we’ve talked about in the past, with OpenAI trying to become Anthropic faster than Anthropic can become OpenAI, I think similarly here, China is in a position where it has the raw manufacturing capability to make lots of robots and is racing to become a robot foundation-model provider faster than the U.S., with our 10 times more compute and our foundation models, can finally figure out how to manufacture humanoid robots at scale.

We’ll see which way it ends up. I realized that Gianluca put this video in the deck. Let’s take a listen to Mark Cuban about humanoid robots.

“I think everybody’s making this push for humanoid robots. I think they might have a 5-year lifespan and then they’ll fail miserably, maybe 10.”

You mean the devices, the companies, or the individuals?

Robots. Or both?

Both.

Right? Because I think everybody defaults to, “Well, we live in a human world, and humanoids will take the place of humans for various functions, particularly in the home,” and I think there’s just no chance.

So maybe we’re missing the second half of his comment.

This is conveniently eliding the second half, where he explains that they’ll merge into the environment.

Okay, well, that makes a lot more sense. All right, let’s get to a conversation.

Hear something really cool?

Yeah, sure.

We had Chase Lochmiller earlier today, our guy Chase, building Stargate in Abilene, Texas. He said, “Remember when we were talking to Brett Adcock? He said, ‘I have to wind my own motors. I literally have to. There’s no supply chain for any of this stuff.’”

The same thing Burt Borne said at 1X. Chase was saying he actually melts metal to make electronic components to build these gigawatt data centers because there’s no supply chain for the stuff that he needs. It’s very much the case that the entire supply chain to build out all this physical stuff is miles behind where it needs to be. It’s entrepreneurial heaven.

Because the virtual stuff—the code writing and all the compute—is going to happen very quickly.

But the robotic stuff, you look at the size of that IPO we were talking about. That’s not going to go: $610 million. Can you imagine trying to go to an investment bank on Wall Street and say, “Hey, we’re doing a $610 million IPO”?

They’d be like, “You can go down to the basement and talk to our junior associates. We’ll get back to you after Anthropic is public. We’ll talk to you if there’s any money left in people’s pockets.”

Yeah. All right, let’s go to a topic I’ve wanted to cover for a while with all of you: quantum and Bitcoin. Here we go.

Google moved up its deadline by 6 years, to 2029, for basically Q-Day. When are we going to see quantum computers break RSA? It used to be that 20 million qubits were required. Today, it’s 1 million qubits. In particular, it’s 4,000 error-corrected qubits, to be specific, to break RSA.

Google moved it up by 6 years, from 2035 to 2029. It’s got everybody in a bit of a panic. The story related to that is that Brian Armstrong, the CEO of Coinbase, has put forward a $150 million coalition to roll out something called BIP-360 as a quantum-proof upgrade to the protocol. It’s a fork.

By the way, in just chatting with Brian, he’s going to be joining us on the Moonshots podcast. We’re going to be talking about both longevity and quantum and Bitcoin. Another story related to this is that Google now says that under 500,000 qubits are required to break Bitcoin encryption, 20 times fewer than predicted in 2019.

There’s a lot going on here. This is concerning Bitcoin holders. I put this next slide forward because, Dave, you and I were roommates with Michael Saylor in our fraternity back in the day. People may not know that. Michael Saylor, Dave, and I were at Theta Delta Chi together, on the third floor. I wanted to see what Mike was saying about Bitcoin, and he’s saying, “I don’t worry about it. Quantum computing wouldn’t break Bitcoin; it will harden it. The quantum risks are overblown.”

Quote

“Bitcoin has survived every existential threat ever thrown at it. This is just the latest, and the upgrade will come before the threat does.”

He puts his money behind that. In the last quarter, he purchased 88,000 Bitcoin, about $7.25 billion worth of Bitcoin. Salim, let’s go to you first on this one, pal.

The true risk here is that protocol consensus may be slower than the emergence of the threat, right? But I’m actually optimistic around this one. I think Saylor is right. Resilient systems will just evolve and can evolve under pressure, but markets are really bad at pricing tail risk until they’re really forced to.

I think what will happen is there’s so much momentum behind Bitcoin and so many—across a Bitcoin Lightning Network payment system that is 3 months old, they’re doing $1 billion a month of transactions. It’s unbelievable to watch some of what’s happening under the radar that most people haven’t even seen.

I’m optimistic on this. Even if Google pulled the date forward a bit, I think this is still a long way off, and the Bitcoin world will be forced to get together and just go, “Okay, we need to upgrade. Let’s just do it.” There’s enough money and motivation to do it.

Yeah. At this moment, Bitcoin’s at $73,000. It’s up about $4,000 in the last 5 days. This has been a black cloud over the Bitcoin market for a while. In fact, Jefferies Bank has pulled out of Bitcoin. We may see others follow suit.

In the same way that AI is sucking money out of every other market, it’s also sucking the attention out of Bitcoin. Dave or Alex, are you guys Bitcoin holders? I know Salim and I are. What do you think?

Only via MicroStrategy.

I think Mike is absolutely right. I don’t know this litany of existential threats. I know there was someone trying to take over half the servers and then control it. Obviously, Bitcoin survived that very easily.

Quantum is not a threat at all. It’s so easy to increase the encryption standard. You can see that quantum computers don’t just suddenly pop up out of some secret lab. You see them coming a mile away. That’s not a risk at all. I think Mike is 100% right.

For the record, I don’t hold Bitcoin. I don’t have any desire to hold Bitcoin. This is the time in the episode where I say something nice about crypto, per the Peter Diamandis ordinance.

My something nice about crypto today, for this episode, is that I also don’t disagree with Michael Saylor. But I also think it’s beside the point. This is not investment advice, but I don’t think it’s quantum. Again, I’ve made this point numerous times. I don’t think it’s quantum decryption that the Bitcoin community should be worried about. It’s AI.

It’s AI coming up with clever inversion attacks against the core hash functions. Before anyone in the comments says, “Oh, but it gets harder over time, and there are several other responses,” I’m aware of all of these responses. But if there is a secret inversion attack against the core hash suite of Bitcoin, this is a major problem for Bitcoin.

I don’t think that’s even the largest problem, though, for Bitcoin. If we’re going to talk about Bitcoin’s existential risk, I think it’s actually just irrelevance. AI—or AI agents, I should say—are emerging, for better or for worse, as the killer app for cryptographic commerce and transactions.

The biggest risk is just that AI agents don’t want to use Bitcoin. I’m aware that the Bitcoin Policy Institute put out this study saying that 6 out of 10 AI agents prefer the flavor of Bitcoin versus other cryptographic means of commerce. I think, over the long term, it’s difficult to buy that AI agents, if they stick with any form of crypto at all, are going to stick with Bitcoin.

They’ll invent their own currencies, their own layer ones, maybe transcendent forms of layer zero, and just reconceive the entire notion of a crypto stack.

With you there. Yeah, they’ll reinvent anything and everything toward efficiency.

Well, wait a minute. Everything Alex just said, though, is all about transaction use cases, and Mike has been saying for a long time that Bitcoin’s role in the world is as a store of wealth.

It’s immune from the government seizing it or taxing it because you can move it so easily. That would be a completely different argument, and I don’t have a horse in that race, but it would be interesting to ask, “What about AI’s impact on that use?”

I don’t need to have a crypto debate. At the micro level, I would say a long-term store of wealth is basically just commerce by another name. You’re trying to store resources in some sense for the long term.

I would question whether superintelligence actually needs a long-term store of wealth at all. It’s going to be moving very quickly, taking rapid actions in the physical economy. Does it even have a need for a long-term, non-operational, nonproductive store of wealth? I doubt it.

Well, I think compute and energy are the ultimate store of possibility, so to speak. Those are real; those are arguably the definition of real assets.

The definition of “long term” is really interesting, too, because right now, the reason we have money at all is that we have trade. You’re going to do something, I’m going to do something. I’m doing it now, and the other thing is tomorrow. “Okay, give me the money, and then tomorrow I’ll pay you back.” It’s just a buffer because transactions don’t line up perfectly in time.

If you imagine a massive, fluid AI economy with thousands of times more things happening, the alignment is a lot higher. The store of wealth could be milliseconds, microseconds, or nanoseconds. At that point, do we even need “digital gold”?

Similarly, this is not investment advice: I don’t hold gold. It’s an unproductive asset; it’s just not interesting. If we really are in the singularity, as I claim that we are, why on earth would I want to hold gold or Bitcoin?

What do you hold, Alex?

Again, this is not investment advice, but for the record, on the one hand, index funds—fundamentally betting that the market is a better allocator of assets, at least among public securities, than any individual can be. It’s basically a bet on superintelligence.

At the other end of the barbell distribution, equity in startups, where I hold material agency. To first order, that’s it. I don’t hold gold, and I don’t hold crypto. I just don’t understand how they’re productive assets.

Salim, what about you?

I’ll jump in on 2 things. One is, Peter, you mentioned that what you need is energy and compute, and I thought, “Wow, that sounds like Bitcoin.” But to Alex’s point, one of the smartest investor types I know, who was worth about $100,000,000, told me how he does wealth management. He said, “70% high-dividend-yielding public equities and 30% high-risk startup investment funds.”

That speaks exactly to what Alex said. The standard things, like real estate and utilities, are all very dangerous places to be. I’m cooking all of them. Like I want to cook lands. We’ve talked perhaps in the past about Coastal Assembly, which is using AI to grow new land. It’s a company where I have a financial interest.

A hot take for this episode—if the crypto hot take wasn’t hot enough. Since I’m underslept, I think land has got to be made post-scarce, and AI will help us make real estate post-scarce.

I agree.

Heart disease has been personal for you as well, hasn’t it?

It really has, Peter. When my daughter was 5, my husband died of sudden cardiac death. This is a topic that I am mission-driven to try to eradicate. Prevention first and early detection are absolutely critical.

50% of people die of heart attacks with no warning signs. Silent killer.

No shortness of breath, no pain, no nothing. They just don’t wake up in the morning.

All right, let’s jump into our final segment here, which is a proof of abundance. I’m going to call it Abundance Corner. These are stories that have come out recently.

Germany just built the world’s tallest wind turbine, 364 meters high. It’s taller than the Eiffel Tower. It generates 33 gigawatt-hours per year, and what’s interesting is that it was built inside an old coal power plant. The coal plant left the wiring behind, and they built this on top of it. The turbine is being built at the Lusatia coal site in Brandenburg, so we’re going to start to see wind and solar penetrate the old energy economics.

The second article is about a 12-patient trial of a redesigned CD40 immunotherapy that had extraordinary results: cancer vanished after 1 injection. Of the 12 patients, 2 had complete remission and 6 had tumor shrinkage. This is the end of cancer heading our way.

Finally, there was a study by the World Bank that showed we don’t need to produce more clean drinking water in Africa. What we need is to rebalance its use. In some places, too much water was being used, and if it were redistributed, it could provide all the water required for sub-Saharan Africa. This is where AI technology can come in and help us understand how much water is required, where it’s needed, and optimize its use.

Any comments on these articles?

I’ve got a bunch, but I’ll limit it to 1 here, just to build on the abundance side. This is separate from the list here, but they’re using AI with acoustic sensing to prevent major failures in wind turbines. The systems are achieving 99% accuracy in identifying damage before it requires repairs.

The cost of maintenance suddenly drops radically for these wind turbines because you can do predictive maintenance in a very powerful way. This is all the little thousand ways, thousand cuts in which we’re reaching abundance in energy. That’s totally going to change the game. I’m so excited about this.

This is such great stuff, except we’ve misspelled Abundance Corner. That’s a minor detail.

I love it. I’ll make 1 comment on the immunotherapies. I think it’s also instructive if you think back—we’re in 2026—to around 2000 or 2001, about a quarter of a century ago. The US Congress was sold on the National Nanotechnology Initiative on the premise that we would have medical nanorobots swimming through our bloodstream, zapping cancer cells.

Yet we find ourselves a quarter of a century later where, as you say, Peter, cancer is well on its way to being solved without the medical nanorobots. We didn’t need the medical robots at all. This is being done by basically retraining or retargeting our bodies’ own immune systems.

I think that raises the question: What, if anything, will we need the medical nanorobots that Eric Drexler and others promised us for? Or is it just a matter of reeducating our existing biology to do more intelligent things without needing any robots in our bodies at all?

We have an amazing system. The challenge is that our biology is optimized through age 30, and then it’s a slow degradation. It never evolved or was selected to live past that. A lot of the age-reversal work going on through epigenetic reprogramming is about taking our systems back to an earlier state of youth, where they’re operating optimally.

All right, a few more articles here in the Abundance Corner—spelled correctly.

Corner. Got it.

So, vertical farming. I remember in my first book, Abundance, in 2012, I talked about vertical farming. It's finally playing out. It's projected to reach $40 billion by 2030, and it hit $8 billion this year. I think what's really important about this story is that vertical farming has a huge impact: 95% less water use.

Production yields are 350-fold greater per square foot than traditional farming. The use of AI and robotics allows you to optimize the perfect pH, get rid of all pesticides, and enable you to get the perfect spectrum for that plant 24 hours a day. Historically, most of the vertical farming to date has been lettuce or leafy greens like that. This is the first time we're seeing something with a higher-value crop like berries. I'm super excited. What are we going to do with all the parking garages that our autonomous vehicles abandon?

Can I give a little historical thing here? If you look over the last 50 or 100 years, the world's biggest food-production countries were the ones you'd expect: the US, China, Russia, Brazil—the biggest ones. But then you look over the last 50 years at the world's biggest food-exporting countries, and you know what number 2 is? Holland.

On a global map, you can't even put a pin to find Holland. It's that small relative to these other countries, but they made major investments in hydroponics, aeroponics, and so on. They're the number 2 exporter of food globally. That just shows you the potential as vertical farming takes hold. We'll be able to totally transform food logistics and security.

The average meal travels 2,500 miles to reach an American table. These yields from vertical farming are something like 10-to-1 compared to horizontal farming. Half of the cost of a good meal is the beef coming from Argentina, the wine coming from France, and the transportation costs. They're huge.

All right, our second story here is 100-hour batteries going commercial. This is the birth of what we call iron-air storage batteries. Lithium-ion batteries use lithium, cobalt, and nickel, and they're expensive. Iron-air batteries are iron, water, and air. They're coming in at one-tenth the cost, and they're now being used for grid storage. Alex, comments on this one?

I do think evolution in battery-energy chemistry is really interesting. The historic trend, if we put aside iron-air for the moment and just focus on the bleeding-edge chemistries, is something like a pretty sustained 8% year-over-year increase per constant dollar in battery-energy densities for the bleeding-edge chemistries. There is, not in some sense but in a very real sense, a Moore's law for increasing energy densities.

At the same time, we're seeing new or newish chemistries like iron-air that are radically reducing the cost for certain applications. Iron-air isn't for every application. It seems unlikely we're going to see it used, for example, for EVs anytime soon. Probably someone in the industry is experimenting with it, but I think we're starting to see, judging from the explosion initially in lithium-ion and then the explosion into a number of other form factors, different chemistries for different applications.

Different applications demand different prices as well. In some cases, when you're powering data centers, you care about the volume of storage and you care about the price. In other cases, you care about the mass and mobility, and those are cases where lithium-ion and lithium-polymer probably still have an edge over iron-air. Overall, I think this is very positive.

I sometimes wonder, as a thought experiment, given that there was quite a bit of experimentation early on in the Thomas Edison era with different battery chemistries, whether we could have arrived at much more advanced chemistries much earlier, like 100 years ago, and whether the history of the internal-combustion engine would have been vastly different if we had seen more investment and more experimentation up front with different battery chemistries. Overall, this is obviously a positive development.

The final story here is AI tutors. A Wharton study tested AI tutors that personalized education. What they found, not surprisingly, was that a 5-month coding course was equivalent to 69 months of additional schooling compared to peers with a fixed curriculum. I think we know this. Basically, you're getting 2× learning gains using AI tutors. They're free, ubiquitous, and available to everybody 24 hours a day, 7 days a week. This isn't breaking news; it's just quantifying it.

At the end of the day, AI is going to be the ultimate educator. It understands your child's abilities, understands what they do and do not know, their favorite sports star, and their favorite color, and it can optimize how it teaches somebody. I think one of the things AI can do better than anything is teach somebody the way they like to learn.

Are you going to make an appeal to teachers? A lot of schools, including people that I know, are incredibly resistant for some reason. I don't get it. I'm going to go out on a limb and say it's cruel—absolutely cruel—to a child to force-feed them a lecture when they're saying, "I don't understand what you just said." The response is, "Well, I'm going to keep plowing forward because everyone else in the classroom understands."

You have to say it the same way over again. The kid can't stop and say, "Wait, explain that to me another way." With AI, it's so much more compassionate. I think it's downright cruel to kids to try to teach complicated things in any way other than AI. Anyone who uses it every day knows it's clear that that's the case.

Sorry to cut you off, Alex. I think there's an element missing.

I would love to be able to just replace teachers—human teachers—with AI. I think it's basically a cliché at this point that, at least in the US, education is subject to Baumol's cost disease. I would love for AI to just replace education, both primary, secondary, and higher ed.

What I suspect is missing, certainly for the most self-motivated students, is that at this point, in the style of Neal Stephenson's The Young Lady's Illustrated Primer from The Diamond Age, it's already here. Well-motivated students can already have a conversation with a model from whichever frontier vendor and teach themselves far more quickly than they can through human instruction.

But for the students who aren't as self-motivated, what I think we're missing right now is an AI embodiment that holds their attention and motivates them where they lack the motivation.

Gaming.

Maybe. I assume, Peter, by gaming you're referring to something quasi-addictive.

Video games. Video games are perfectly tuned not to be too difficult and not to be too boring, to hold your attention and motivate you all the way. I just don't understand why video-game designers, instead of teaching kids a whole set of random or made-up facts, can't use a set of facts about quantum physics, subatomic particles, planets, physics, and biology and gamify that. Someday.

Yeah, it's funny. If you play Fortnite and look at the weapons and the number of intricate components in the weapons—

And then the characters memorize these things. They memorize them. In Madden NFL, there's the playbook, with three-layer-deep menus of different routes and plays. Before you know it, you could have learned an entire discipline like quantum physics with that same amount of brainpower. I swear to God, AI can make topics like quantum physics incredibly fun and engaging. The technology is here today to do that. Someone's just got to get it out the door.

People have been building edutainment games for decades at this point. I grew up with Math Blaster, or whatever, back when I was growing up. But the problem, I suspect, is that as a user of these games, you're not motivating the users—children or students—with the exact outcomes.

What would have been utterly transformative for me would not be motivating some math problem with an arguably disconnected animation on the screen. Motivate them by actually empowering them to do really amazing things in the real world. That's far more motivating, I think, than just an animation or some dopamine push from a jingle.

Well, I think that's for you and probably not for the average kid.

Yeah, maybe. I don't generalize. I don't know.

All right. The final item in the Abundance Corner is this graphic. Look at this beautiful exponential growth curve. This is EVs sold globally. Back in 2010, there were barely 10,000 of these vehicles. This was for Elon's first Roadster. Now we're up to 12.7 million EVs sold globally. In China, 1 in 2 new cars is an EV. It's just perfect exponential growth.

Can I just throw in a fun fact?

Please.

In 2015, the International Energy Agency predicted that we would not sell 1 million electric vehicles a year before 2040. That year, in 2015, we sold more than 1 million electric vehicles.

You get the predictors and the reality. Governments and good companies are relying on these strategic predictions for strategic decisions, and they were wrong before they even put out the comment.

And so it’s great to see this. The curve is still accelerating, right? By 2030, 2025 on this chart is going to look very modest. Just look at the impact on the rest of us from the war if oil prices suddenly shoot up, et cetera. You’re protected from a huge amount of volatility as we go to solar, batteries, and EVs.

All right, gentlemen. I have a beautiful outro piece from Marcus Helker, and I want you to look at this. This is the Moonshots Made Boy Band. We’re making our debut here.

No, God. I can’t take this.

And it’s so good, Selene. What are you worried about?

Here we are, the Moonshots Made Boy Band.

Everybody, Alex, you’ve got to be a science officer. You’re lucky.

I get to be the chief medical officer. I think it’s blue, right? I get to be chief medical.

All right. You’re right, you’re right. Medical—why would that be? I thought that was killer. I thought that was better.

Our Starfleet officers—the Abundance logo resembles the Starfleet pin.

It does. Convenient, isn’t it? I don’t know how that happened. Pull up the clip from 2 months ago and compare it to today. It’s incredible how quickly—

Yeah, and just a shout-out to the creator community out there. Love it.

Please send us your outro, or if you have an intro song you want to share with us, please send it to us. We’d love to share it with everybody.

Gentlemen, it was fun doing back-to-back episodes with you over the last 24 hours, and I’m looking forward to another episode next week. Stay optimistic. Stay hopeful. The future’s ours to create. We’re creating the vision of tomorrow that we want.

If you think AI is happening to you and not for you, you’re going to be back on your heels and you’re going to be in fear, and that’s the worst place to actually venture into the future. This is the most extraordinary time ever to be alive, and I’m so blessed to have Salim Ismail, David Blundin, and AWG as my moonshot mates. Love you guys. Awesome episode. Great conversation.

And prosper, Peter. Live long and prosper. Peace and long life.