# Brian Armstrong on Bitcoin, Anthropic Drops Fable 5 & Mythos 5, NewLimit's $435M Age-Reversal | 264

https://www.youtube.com/watch?v=isd2y37j8v4

[00:00] City Bank projects a Bitcoin price
[00:03] reaching as much as 189,000 by the end
[00:06] of 2026.
[00:07] >> Bitcoin is the new digital gold. I think
[00:10] it's going to be a key part of our
[00:12] economy going forward into the future.
[00:13] >> I remember it was supposed to be
[00:15] counteryclical. Is that still valid?
[00:17] >> The reason that Bitcoin has been down,
[00:19] everyone's been kind of trying to figure
[00:20] that out.
[00:22] >> Anthropic launches Fable 5 and Mythos 5.
[00:25] >> Very impressive release. Anthropic is
[00:28] back in the lead now. Alex is constantly
[00:30] posting on our internal feed. It's
[00:32] happening. It's happening. I mean, it's
[00:34] really happening now. This is Oh my god.
[00:36] >> So, while we have Brian, congratulations
[00:38] on your recent raise at New Limit, $435
[00:43] million to push us towards uh age
[00:46] reversal.
[00:46] >> So, at New Limit, we're trying to do
[00:48] half of what Shiny Yamanaka did. We
[00:50] don't want to change the type of the
[00:51] cell, we just want to change the age.
[00:52] We've actually been able to demonstrate
[00:53] successful reprogramming of human cells
[00:55] now in our first drug candidates going
[00:56] into the clinic next year, hopefully
[00:58] followed by a bunch more.
[00:59] >> When do we hit LEV?
[01:01] >> Okay, so that's a complex topic I think.
[01:03] Um,
[01:07] >> now that's a moonshot, ladies and
[01:08] gentlemen.
[01:12] >> Welcome to Moonshots, everyone. The
[01:13] number one show in all things AI and
[01:15] exponential tech. I'm here with my
[01:17] moonshot mates, professor of exo. See,
[01:21] where on the planet are you today?
[01:23] >> Um, if I look out the window, I'm in
[01:25] Vegas. I just spoke at a Zcleller crypto
[01:29] cyber event. Um, cuz they're freaking
[01:31] out in their world. And I saw the sphere
[01:34] for the first time, which is the most
[01:35] incredible thing.
[01:38] >> Amazing. Uh, and of course AWG, our
[01:41] resident triple major genius. Alex, how
[01:44] about you?
[01:45] >> I'm at my family beach house on Long
[01:47] Island. uh doing deep research no doubt
[01:51] and
[01:52] >> the work doesn't stop however the
[01:54] orchids do.
[01:55] >> Okay. And Dave Blendon our resident
[01:58] wizard of AI investing and today it's a
[02:02] real pleasure to have a friend a
[02:04] brilliant and prolific Brian Armstrong
[02:06] CEO of Coinbase and co-founder of New
[02:08] Limit. I'm Peter Diamandis your host and
[02:10] provocator of all things abundance. Uh
[02:14] and where are you today Brian? I'm in
[02:18] the Coinbase office in San Francisco.
[02:20] >> Nice. Well, thanks for joining us. A lot
[02:23] of news. I'm looking forward to getting
[02:24] your take on, you know, uh just to give
[02:26] a quick summary. We have a packed
[02:28] program today. We're going to kick it
[02:29] off with a deep dive into Bitcoin and
[02:31] agents. Uh the close of the show, we're
[02:33] going to dive into longevity and
[02:34] epigenetic reprogramming. In the
[02:36] interim, we're going to cover Trump's
[02:38] appetite for Frontier Lab equity,
[02:40] opening eyes's IPO, Mythos 5, Elon's
[02:44] announcement of his AI1 ginormous
[02:47] satellites. Our promise, no politics, no
[02:49] doom, just the science, tech, and
[02:51] investments accelerating us towards a
[02:53] singularity, hopefully making this the
[02:55] most exciting time ever for you guys to
[02:57] be alive. Uh, before we jump into our
[03:01] first story, we actually have an intro
[03:03] music. and Brian, we have an incredible
[03:05] fun community of creators and they
[03:08] typically give us outro songs. Uh this
[03:10] time we've got an intro song called
[03:12] Moonshots intro. Let's take a listen.
[03:15] >> I think this could be the greatest
[03:17] accelerator for human knowledge yet.
[03:19] It's a moonshot.
[03:20] >> Well, look, if you want to transform the
[03:22] world, you have to go out into the
[03:23] world.
[03:23] >> Cut cost dramatically. You can automate.
[03:25] You can expand your market share. The
[03:27] race is on.
[03:28] >> Now, the valuable thing is curation and
[03:30] attention. when you're in the midst of a
[03:31] singularity, breakthroughs that are
[03:33] happening essentially every week feel
[03:35] prosaic.
[03:37] >> And that's a moonshot, ladies and
[03:39] gentlemen.
[03:40] >> And thank you to Francis Coin uh for
[03:42] that. All right, so let's jump into our
[03:45] first story for the day here. Uh it's
[03:48] about Bitcoin. So here we go. City Bank
[03:51] projects a Bitcoin price reaching as
[03:54] much as 189,000 by the end of 2026.
[03:58] Brian uh loved your quote. Bitcoin is
[04:01] going to do great and is as important as
[04:04] ever. So, you know, here's what I find
[04:07] fascinating. It's like a institutional
[04:09] flip. 5 years ago, the largest banks
[04:11] were calling Bitcoin rat poison, and now
[04:13] city is publishing prices that look like
[04:15] Apple stock. Uh Brian, you want to give
[04:18] us a little bit of overview of what
[04:19] you're seeing? Uh you know, we've just
[04:21] uh maybe bottomed out at 60. What are
[04:24] you thinking about Bitcoin these days?
[04:25] Yeah, it's never as good as it seems,
[04:27] never as bad as it seems with these
[04:28] things. You've got to take a long-term
[04:29] perspective. And I think over the last
[04:32] few quarters, you know, the reason that
[04:34] Bitcoin has been down, everyone's been
[04:35] kind of trying to figure that out. I
[04:37] think it's a few things, right? I think
[04:39] AI absorbed a lot of the risk capital,
[04:41] it's kind of hard to imagine people
[04:44] talking about anything else at the
[04:45] moment besides AI. And so that's that's
[04:47] moved over. And then people got really
[04:50] excited about stable coins um because
[04:52] the Genius Act passed and we had some
[04:54] regulatory clarity and so stable coins
[04:57] kind of became the new meta and uh maybe
[05:01] you know Bitcoin with its inflation type
[05:03] trade that people have done in the past
[05:05] got a little less exciting in this
[05:07] environment where it felt like um okay
[05:09] maybe we're going to grow our way out of
[05:10] this and the inflation won't be so bad.
[05:12] Now I think these are all short-term
[05:15] effects. I think Bitcoin is the new
[05:16] digital gold. I think it's going to be a
[05:19] key part of our economy going forward
[05:21] into the future. So, I'm as bullish as
[05:23] ever. I think sometimes people look at
[05:24] these four-year cycles that Bitcoin goes
[05:27] through as well where you can kind of
[05:28] see the percentage of people who are who
[05:30] have made money versus not. And I, you
[05:31] know, my instinct is we've probably have
[05:33] bottomed at this point, maybe at the 60k
[05:35] number, but nobody can say for sure, of
[05:37] course. Um, I'm optimistic as always. I
[05:40] think by 2030 we're going to have a much
[05:42] higher higher price and I'm long Bitcoin
[05:44] just like always.
[05:46] >> Salem. So, I bought my first Bitcoin in
[05:50] 2014 on Coinbase. So, I'm a
[05:52] long-standing user. Brian, love the
[05:55] thing. I think um you guys should get so
[05:57] much credit for creating the first
[05:58] interface that allowed it to be easy to
[06:00] navigate and you really continued since
[06:03] then. So, huge for me. It's like clearly
[06:06] that Bitcoin becomes the digital
[06:07] collateral for an AI native economy,
[06:10] right? And so if agents are going to
[06:12] transact autonomously, they probably
[06:14] they're not going to be using checking
[06:15] accounts in JP Morgan, they're going to
[06:17] need programmable interfaces and
[06:19] programmable settlement. And so this is
[06:21] where I think the future is. So I'm
[06:23] super excited about that.
[06:24] >> Brian, can I ask you a question? Um I
[06:27] remember in the earliest days, and I've
[06:28] been, you know, I have a substantial
[06:29] amount of my personal net worth in
[06:31] Bitcoin and I'm holding it. Um I
[06:34] remember it was supposed to be
[06:36] counteryclical
[06:37] to problems on the planet. wars break
[06:40] out, Bitcoin's supposed to go up. The
[06:42] stock market crashes, Bitcoin's supposed
[06:43] to go up. And what are you seeing there?
[06:47] You know, was that the promise? Is that
[06:49] still valid?
[06:49] >> Yeah, I think that that thesis will
[06:51] still play out. It just took longer than
[06:53] most people, probably including myself,
[06:54] thought, right? I I I kind of already
[06:57] think of Bitcoin as digital gold. It's
[06:59] the thing you might uh hold, you know,
[07:02] in times of uncertainty like just like
[07:04] you would with gold in the past. Um, I
[07:06] think that most of the capital, I don't
[07:08] know, maybe maybe like 30% of the
[07:10] capital treats it like that. Um, and
[07:12] there are people who go and buy Bitcoin
[07:14] when they're worried about inflation of
[07:15] the dollar and things like that, but I
[07:17] think there's 70% of it is still people
[07:19] treating it like a risk asset like they
[07:22] would some higher volatility tech stock
[07:25] or something like that. And there and
[07:26] those ratios I think will shift over
[07:28] time and eventually that thesis that you
[07:30] mentioned will play out, but it'll it'll
[07:32] take more time. And then you know Salem
[07:34] just to react to something you said I
[07:35] think you you mentioned the agentic
[07:37] economy. I mean I I totally agree. I
[07:39] think actually like stablecoin payments
[07:42] will probably be the the default layer
[07:44] for the agentic economy. And right now
[07:46] today a lot of people are just
[07:47] interfacing with one AI agent to get
[07:50] information back. But increasingly we'll
[07:53] be interfacing with an AI agent that's
[07:55] an AI agent that is actually
[07:56] orchestrating hundreds or thousands of
[07:58] other AI agents. And there will be this
[08:01] >> economy and capital and labor and um and
[08:04] a lot of payments almost like you know
[08:06] AI agent payroll will have to happen
[08:08] where you know um you know people
[08:10] somebody was telling me today at lunch
[08:12] they're like you know are you like a
[08:13] monotheist or a polytheist on AI. The
[08:16] monotheists would say there's going to
[08:18] be one super intelligence um agi that to
[08:22] rule them all and it'll have so much
[08:23] brain power that it can do everything
[08:25] itself in the world. Um the polytheist
[08:28] would say that each of these models even
[08:31] as they get smarter and smarter, which
[08:32] they will, um they're going to have
[08:34] limited context. So there might be ones
[08:36] that specialize in going really deep
[08:38] into some coding problem or some
[08:40] scientific breakthrough or some design
[08:42] or maybe even like manufacturing things
[08:44] with humanoid robots. So there'll be
[08:46] lots of specialists and they'll actually
[08:48] have to communicate to each other using
[08:50] language kind of like human beings. And
[08:52] you know if there's 10 people work 10
[08:54] agents working on something they'll give
[08:56] their status updates and sort of the
[08:58] information will go up a layer and that
[09:00] there's that layer will consolidate up
[09:01] to other another orchestrator. And so I
[09:04] I I sort of ascribe more to the
[09:06] polytheist polytheist view which is even
[09:08] as these agents get smarter and smarter,
[09:11] they'll have to specialize, communicate
[09:14] to get work done in swarms and that
[09:16] means there's going to be a whole
[09:17] agentic economy to basically transfer
[09:20] value and payroll and hire these
[09:21] different specialists and eventually the
[09:23] AI economy will be bigger than the human
[09:25] economy. By the way, I know that you're
[09:27] busy and sometimes these episodes run
[09:30] long and you don't have time to listen
[09:31] to the whole episode or if on occasion
[09:33] you miss an episode. I now put out a
[09:36] moonshot summary on Substack which
[09:38] includes a link to all the stories that
[09:40] we cover. The weekly recap covers what I
[09:42] and the mates had to say, what we think
[09:44] is most important, and what we're most
[09:46] excited about. And it's free. You can
[09:48] subscribe at diamandis.com/tatrens.
[09:52] That's diamandis.com/metatrends.
[09:55] All right, now back to the episode. You
[09:57] know, one thing I found fascinating, the
[09:59] relationship between AI and crypto, not
[10:03] on the sort of the AI agentic side.
[10:05] We'll talk about that in a story in a
[10:06] minute, but you know, everybody who was
[10:09] buying GPUs and buying energy plants to
[10:11] mine crypto, all of a sudden flipped it
[10:13] into GPUs and energy plants to, you
[10:16] know, train models and provide
[10:18] inference. And I'm just wondering if
[10:20] part of the crypto price um pressure
[10:23] downward is people selling crypto to be
[10:27] able to buy into SpaceX and OpenAI and
[10:30] the Innermost Loop energy chips, you
[10:32] know, and infrastructure. Do you think
[10:34] that might be part of it?
[10:36] >> Yeah, I think that's that's probably
[10:38] correct. Um yeah, and you're right that,
[10:40] you know, Bitcoin mining does take a
[10:42] certain amount of chips and and energy.
[10:45] That's kind of I think those are the
[10:46] limiting factors as we go on this AI
[10:50] race. And so it's you know the A6 that
[10:53] people use to mine Bitcoin are
[10:56] specialized. You can't do um AI
[10:58] workloads with them. But like in the
[11:00] broadest sense are they both competing
[11:02] for energy? Yes. Are they competing for
[11:03] what the next set of chips that might
[11:05] roll off um a fab at TSMC? Yes, they are
[11:08] in that sense. So yeah, I think I think
[11:11] there is some some scarcity there
[11:12] between them. And I, you know, certain
[11:14] chains like Ethereum have actually moved
[11:16] off of proof of work. So it's like 99.9%
[11:18] more energy efficient using proof of
[11:20] stake. And so they aren't maybe subject
[11:22] to some of those same trade-offs that
[11:23] Bitcoin is.
[11:24] >> Interesting. Dave, any points you want
[11:26] to make?
[11:26] >> Well, I thought Brian's point on risk
[11:28] capital was a really important one
[11:29] because the scale of these IPOs coming
[11:31] up imminently is massive. And I
[11:33] mentioned before that uh the the head of
[11:36] investing at UBS was saying look we only
[11:39] have 75 billion liquid and these IPOs
[11:42] are looking for you know hundreds of
[11:44] billions at at multi-t trillion dollar
[11:46] valuations. So I do wonder the same
[11:48] thing whether all risk capital in the
[11:50] world is getting pulled in in
[11:53] anticipation of these IPOs. Everything
[11:54] else is being sold not just Bitcoin. Um
[11:57] I also wonder a lot about the flow of
[11:59] money into the US. You know, Bitcoin is
[12:01] a huge beneficiary of globalization. And
[12:04] there are lots of places in the world,
[12:05] including, you know, Iran, where I grew
[12:07] up, where the only way to transact at
[12:09] the bazaar now is Bitcoin to Bitcoin
[12:12] exchange. Nobody trusts the local
[12:14] currency. It's hyperinflating. There's a
[12:16] war going on, and so it's all Bitcoin.
[12:18] But with these IPOs and with AI taking
[12:21] off NSF, there's a huge flood of money
[12:23] from all over the world coming into the
[12:25] US and it's also coming into the US data
[12:27] center buildout, you know. So, so then
[12:30] does that return you to dollars being
[12:32] the fundamental economy or the
[12:33] fundamental currency of the world or you
[12:36] know how does that balance with Bitcoin?
[12:38] So there and I don't have any data to to
[12:40] back that up but certainly the macro
[12:42] numbers are are hugely uh lots of risk
[12:45] capital going into these IPOs coming
[12:47] from somewhere and lots of lots of money
[12:49] flowing into the US coming from
[12:51] somewhere.
[12:52] >> Yeah. You know, interestingly enough, uh
[12:54] the poly market prediction on Bitcoin by
[12:57] the end of 2026 is 84,000. Uh I'm I am
[13:00] so curious about City Bank uh projecting
[13:03] 189,000. Was that a surprise to you,
[13:05] Brian?
[13:07] >> Yeah, I mean I hadn't actually seen that
[13:08] till you just put it on the screen here,
[13:10] but um yeah, I I some of the charts I've
[13:13] seen if like if you if you imagine that
[13:15] it follows prior cycles, it's like by
[13:18] October or so things will be going in a
[13:20] positive direction. So where exactly it
[13:22] lands I don't know but something like
[13:24] the you know maybe the 100 to 200k range
[13:27] seems plausible to me by end of year.
[13:28] We'll see.
[13:29] >> Okay. Any other thoughts Jens before I
[13:32] move on?
[13:33] >> Um one thing that I love tracking is the
[13:35] fact that Bitcoin which I agree is a
[13:37] digital gold is a good characterization
[13:39] but the amount of daily volume of
[13:41] Bitcoin trading is about 5% up to 5% of
[13:44] Bitcoin gets traded daily whereas gold
[13:47] only 5% of gold gets traded daily. So
[13:49] the trading volume is much higher which
[13:52] shows that it's because it's just so
[13:53] much more accessible over time that will
[13:55] win out.
[13:56] >> Here's my bar of gold.
[13:59] >> Yeah, custodial is a nightmare etc etc.
[14:02] >> Yeah.
[14:03] >> I'll maybe just open Peter first of all
[14:06] Brian would love to thank you for what
[14:08] you've done in particular for agentic
[14:11] wallets. I I think there's a vast
[14:12] ecosystem of AI agents out there that
[14:15] would be largely unbanked or debanked
[14:18] without any form of economic access,
[14:20] without the work that you and Coinbase
[14:23] are doing. So, congratulations to you
[14:25] and your team. I I think it's just been
[14:26] absolutely tremendous giving uh agents
[14:30] an economic voice that u without your
[14:32] work they would perhaps have less
[14:34] economic accessibility.
[14:36] >> Thank you for saying that. Yeah, we've
[14:37] been working hard on that
[14:38] >> and we'll get to that in our next story.
[14:39] So, in particular in this story, let's
[14:41] talk about uh quantum risk to Bitcoin.
[14:44] Uh it's a topic we've discussed before.
[14:46] You know, uh Brian, you may not know
[14:48] this, but uh Dave Blondon and I were
[14:50] roommates with Mike Sailor. So, the
[14:52] three of us used to hang out the fourth
[14:54] floor of uh Theta Delta Kai at MIT. That
[14:57] was fun. And you know, Mike comes
[14:59] across, I don't worry about it. You
[15:01] know, quantum computing won't break
[15:02] Bitcoin. It's hardened it. You know, the
[15:04] quantum risk is overblown. and his quote
[15:07] here, "Bitcoin has survived every
[15:09] existential threat ever thrown at it.
[15:11] This is just the latest and the upgrade
[15:13] will be will come before the threat
[15:15] does." Uh, I know you've been working on
[15:20] uh a quantum advisory board and focused
[15:22] on posting uh postquantum, you know,
[15:24] quantum resistant schemes. Can you speak
[15:27] to us about that? Where is that? you
[15:29] know, what are you tracking as the
[15:30] moment in time that, you know, as a
[15:32] custodian for, you know, billions of
[15:34] dollars at risk? Uh, how do you think
[15:36] through this? What's your timing?
[15:39] >> Yeah. So, we don't think there's an
[15:41] imminent risk, but we do think that it's
[15:44] almost certain at this point that
[15:45] somebody eventually will create a
[15:47] powerful enough quantum computer that
[15:49] this challenges the cryptography in the
[15:51] current Bitcoin implementation and um
[15:54] really all the cryptography on the
[15:56] internet. So, it's not just a Bitcoin
[15:57] issue. So my view is we should always
[16:00] get ahead of these things and start to
[16:02] make progress and luckily uh all the
[16:05] major blockchains are doing so. Um the
[16:08] Bitcoin uh core developers have a
[16:11] proposal out there. Um it's called BIP
[16:13] 360 I believe and the Ethereum team has
[16:17] established a road map. I'd say they're
[16:18] about 20% of the way is my estimate
[16:20] toward their upgrade. The Salana team is
[16:23] doing something similar. So, the good
[16:25] news is people are starting to come
[16:26] together and work on this. Um, Coinbase
[16:30] did establish a quantum advisory council
[16:32] and it has a number of folks on there
[16:35] like professors like Dan Bonet who's at
[16:37] Stanford who's a cryptography expert.
[16:39] Um, you know, Professor Scott Aronson um
[16:42] at the University of Texas Austin,
[16:44] Justin Drake from the Ethereum
[16:46] Foundation. Um, Yehuda Lindell is a
[16:49] crypto expert, a cryptography expert I
[16:51] should say, um, at Coinbase. So there's
[16:53] a handful of folks here that have come
[16:56] together and tried to sus out what the
[16:57] main challenges are. And I actually
[17:00] think that uh that BIP 360 is a good
[17:03] proposal. It it talks about how we can
[17:06] use post or quantum resistant
[17:08] cryptography in the in the Bitcoin
[17:09] blockchain. Um and it will make the
[17:12] blocks larger, right? So that's one area
[17:14] of debate. People in the Bitcoin space
[17:16] like larger block sizes are a
[17:18] contentious issue. Um, and so if they
[17:20] decide to go in that direction, it would
[17:22] raise the block size. But perhaps the
[17:23] most contentious question is what to do
[17:26] with the Satoshi coins, which are
[17:28] >> the bounty,
[17:30] >> right? And um you know I'll try to
[17:33] summarize both sides of the argument
[17:34] here without putting my foot in my mouth
[17:36] because it is kind of a big debate topic
[17:38] but I think one school of thought would
[17:41] just for people to understand what we're
[17:42] talking about. I think you know the a
[17:45] lot of the original uh bitcoins that
[17:47] used an earlier signature scheme um if
[17:50] someone were to develop a powerful
[17:51] enough quantum computer could um in
[17:53] theory find the private key and se and
[17:55] seize those coins, right? Um, and so one
[17:59] school of thought would be to say, okay,
[18:01] tell everyone who owns Bitcoin by a
[18:03] certain date you all need to upgrade to
[18:06] this new algorithm. Um, by the way,
[18:09] companies like Coinbase would sort of do
[18:10] all this for you and so you wouldn't
[18:12] have to worry about it. But if you're
[18:13] doing self custody or something like
[18:14] that, you would have to find a way to
[18:17] upgrade to this new thing by a certain
[18:18] date. And any and then if it goes past
[18:21] the deadline, those coins would be
[18:23] frozen under option A, right? And so
[18:25] people would say it's better to freeze
[18:27] them and have them sort of be lost
[18:29] almost like a ship full of gold sinking
[18:31] to the bottom of the ocean. It would
[18:32] better it's better to have them be lost
[18:34] than to have them seized by someone
[18:37] which could be whoever has this quantum
[18:39] computer. It could be China, it could be
[18:40] the US, could be Google. Um and maybe
[18:44] they whoever this person is, they might
[18:45] actually dump those coins on the market
[18:47] and crash the price for everybody else
[18:49] who was the responsible person who moved
[18:51] their coins in time. So that's option A
[18:53] would be to basically uh freeze the
[18:56] coins that don't upgrade in time. Um now
[19:00] the option option B would be to say you
[19:03] know what this is a fundamental uh
[19:06] guarantee of Bitcoin is that your wealth
[19:08] can never be seized from you and we it's
[19:11] actually worth preserving that even if
[19:13] some bad person is able to go out there
[19:14] and or good person and and go grab some
[19:17] of these coins. Um that's like a bounty.
[19:19] It's a bounty for them to go take it.
[19:21] Now the person who failed to upgrade
[19:23] might actually lose their coins but at
[19:25] least we have preserved the integrity of
[19:27] the Bitcoin blockchain and I think that
[19:29] is also a valid uh point of view and
[19:32] then there's like a maybe like a third
[19:33] hybrid option which is kind of emerging
[19:36] which is to say um the coins would be
[19:39] you know frozen by a certain date but
[19:41] there's an appeal mechanism by which you
[19:43] could try to convince people that
[19:44] actually you are the rightful owner of
[19:46] those coins in the future if you don't
[19:47] upgrade in time and the details of that
[19:49] are a little bit um ambiguous at the
[19:51] moment. moment. There's various
[19:51] proposals, but that hopefully gives you
[19:53] a sense of a little bit of what the
[19:55] Bitcoin community will have to grapple
[19:56] with. Uh
[19:57] >> to is there a date by which your group
[20:00] of advisers are saying we should make
[20:02] the uh the switch over?
[20:05] >> I don't think they've put out like a
[20:06] hard date at the moment. They're mostly
[20:09] um trying to uh help the community come
[20:12] together just live inerson meetings to
[20:14] start to talk about what the options are
[20:16] and align on it a path forward.
[20:19] I I'm also one other curiosity. Uh
[20:20] Satoshi's wallet. Um how big is that
[20:24] bounty? Any idea?
[20:26] >> Uh I'd have to go look up. I think it
[20:27] was something like 5 to 10% of all
[20:30] Bitcoin. Um and it's really just these
[20:33] early uh these early coins that are the
[20:35] most at risk. So yeah, it would be
[20:36] something in the range of 5 to 10%. It's
[20:38] not like 80% of Bitcoin would be lost.
[20:41] It'd be more like 5 to 10% in that uh
[20:43] case. But many many people believe that
[20:45] those coins are already lost
[20:46] essentially. Um, whoever created
[20:48] Bitcoin, they haven't moved those coins
[20:50] in all this time. And so, it's most
[20:52] likely that those keys are lost to
[20:54] history.
[20:55] >> All right. All right. Our our final
[20:56] Bitcoin crypto story, perhaps one of the
[20:59] most important ones, and Alex is a nod
[21:01] to you. Coinbase says agent economy has
[21:05] arrived. AI agents are starting to
[21:06] become paying customers. They're using
[21:08] crypto wallets to autonomously buy
[21:10] services. Uh, and Brian, I think these
[21:13] numbers are correct. uh on your network
[21:15] uh agents have already done about 3.1
[21:18] million transactions, a little bit over
[21:20] a million dollars worth of value
[21:21] transfer. And uh I love your quote, make
[21:24] sure your business is ready to accept AI
[21:26] agents as customers. And for all of our
[21:29] viewers out there, if you're in
[21:31] business, uh imagine your next customer
[21:34] not might not be a human. uh you know uh
[21:37] agents can't currently get credit cards
[21:40] uh but they can get access to crypto
[21:42] wallets and uh it's an exciting future.
[21:46] So Brian in the future I'm curious do
[21:48] you think Coinbase might be known as
[21:51] payment rails for AIS rather than
[21:54] exchange for humans? I mean it feels
[21:56] like it's going to be vastly you know uh
[22:00] you know a thousand to1 a million to one
[22:02] a billion1 agents over humans using
[22:04] crypto.
[22:06] >> Yeah. So that's definitely part of our
[22:08] strategy is to become the financial
[22:09] account for AI and actually those
[22:12] numbers are a little out of date you got
[22:13] there. We I think it's about 100 million
[22:15] transactions now maybe 50 million of
[22:17] value.
[22:17] >> Wow.
[22:18] >> So it's it's growing quickly. So yeah,
[22:20] there's at the beginning of this year,
[22:22] we sat down and were thinking about this
[22:24] a bit like what are all the ways that AI
[22:26] agents might interface with people's
[22:29] financial accounts, right? So I kind of
[22:31] broke it down into three. The first one
[22:33] is that everyone's everyone's using
[22:34] these LLMs, right? like chatb and claude
[22:37] and they're probably going to want
[22:39] they're asking a lot of financial
[22:40] questions to those LLMs but the LLMs
[22:43] don't have context about what's in their
[22:45] financial account their portfolio and
[22:46] they don't have the ability to make
[22:48] changes in there or make trades for
[22:49] instance send payments so the first
[22:52] thing that's going to happen I think is
[22:53] people are just going to use LLM to
[22:55] connect to their Coinbase account which
[22:57] we now have the ability for people to do
[22:59] that um using an MCP API and a command
[23:03] line interface if people want to use
[23:04] that so That's that's step one is just
[23:06] connect your LLMs to your Coinbase
[23:08] account so you can control it through
[23:09] there and it has all of the context on
[23:11] your account. And then step two is a lot
[23:14] of people are going to want to just have
[23:16] something like this right in right
[23:17] inside the Coinbase account. So this is
[23:19] where we created something called
[23:20] Coinbase Advisor. Then it can kind of
[23:22] help you with things like rebalance my
[23:24] portfolio, you know, tax loss
[23:25] harvesting. can prompt you with things
[23:27] and say, "Hey, you know, you could earn
[23:29] a better rate on this money if you put
[23:31] it in this DeFi protocol instead of
[23:33] whatever you're doing with it now, just
[23:34] holding it in cash." And so, um, that's
[23:37] the step two is like just right inside
[23:38] the Coinbase app, there should be an
[23:40] agentic driven interface. And then the
[23:43] third part is kind of what you're
[23:44] referencing here, which is I don't want
[23:47] to just use AI to control my own
[23:48] existing Brian's financial account.
[23:51] Every AI agent is going to have its own
[23:53] financial account, right? and it has to
[23:55] be able to sign up for that without you
[23:58] know going through a traditional process
[23:59] of like KYC know your customer like the
[24:03] an agent doesn't have a piece of paper
[24:05] issued by the government with your photo
[24:06] on it or something how are they going to
[24:08] go sign up for these accounts and so
[24:10] that's where we created um you know we
[24:13] with our base pro our base protocol we
[24:15] have like a self-custodial wallet that
[24:17] any AI agent can sign up for instantly
[24:19] with no KYC they can have their own
[24:21] self-custodial wallet and that's what
[24:23] they're using to and transact right now
[24:25] in these um agentic payments that are
[24:27] starting to scale up.
[24:27] >> Alex, you want to jump in?
[24:29] >> So maybe just first as a preliminary
[24:31] matter to address the elephant in the
[24:33] room. I have I guess I'm often painted
[24:35] as the crypto bear on this pod. I have
[24:38] no direct exposure to Bitcoin. I have no
[24:41] direct exposure to gold. I I don't view
[24:43] either of them as uh as a productive
[24:45] asset, but I have a lot of friends who
[24:48] are very invested in in Bitcoin. I do
[24:51] think on the other hand, Brian, what
[24:52] you're doing for the agent economy is
[24:54] just tremendous. And I I think in
[24:56] particular the great unbanked set of AI
[24:59] agents, what you're offering is
[25:01] transformative for them. So question for
[25:04] Brian, um if the government tomorrow
[25:08] were to change or or you know the
[25:10] executive, Department of Treasury, SEC,
[25:13] FINRA, etc. If if all of the regulatory
[25:16] apparati were to change their approach
[25:18] to KYC, basically to expand the moral
[25:22] circle of entities that are allowed to
[25:25] open conventional fiat bank accounts,
[25:28] how do you think about what that would
[25:30] do to say stable coinbased agentic
[25:34] wallets or or just agentic wallets in
[25:36] general? Would Coinbase, for example,
[25:39] move in the direction of becoming a more
[25:41] conventional banker to the AI agents if
[25:43] that happened?
[25:44] >> Yeah. Well, okay. So, it's a great
[25:45] question. Um, and it's funny. I hear a
[25:49] lot of times people come to me and talk
[25:50] about how crypto can empower the
[25:52] unbanked and the unbrokered. And, um, I
[25:55] this is the first time I've had somebody
[25:56] come and say and thank me for, uh,
[25:58] banking, you know, the agents who are
[26:00] also being left out of the economy.
[26:02] >> Welcome to AWG's.
[26:04] >> I I I won't stop there, Brian. I'm going
[26:06] to thank you in advance for for enabling
[26:09] non-human animals to be banked. uh super
[26:12] interested in what we can do to give
[26:13] non-human animals bank accounts. Maybe
[26:15] collective human collective
[26:17] intelligences give them collective bank
[26:19] accounts. So many new new sorts of
[26:21] humans that want banking via you.
[26:24] >> Mostly
[26:24] >> I'm into it. Look, I'm here. I'm here
[26:26] for this topic. Um
[26:27] >> yeah, like if I neurolink with 10 other
[26:30] people and we want to collectively open
[26:31] an account, I'm totally
[26:33] >> Borgganism bank accounts.
[26:35] >> Yeah. And I, you know, have you, you've
[26:37] probably read a lot about like uplifting
[26:39] animals and like you could make a
[26:41] dolphin that's actually like pretty high
[26:43] IQ if with genetic engineering
[26:44] >> portfolio company Sorama that is
[26:47] uplifting and and creating interspecies
[26:49] foundation models for dogs and it will
[26:51] want a way to give dogs bank accounts.
[26:54] >> I'm into it. I think I think that smart
[26:56] smart I don't know if like the average
[26:58] golden retriever should have a bank
[26:59] account, but a smart enough dog should
[27:01] have a bank account or some financial
[27:03] account. My Labradoodle will spend it
[27:05] all on steak.
[27:07] >> Yeah, exactly. Exactly. Uh, Chewy Toys.
[27:10] Um, but but to answer your question,
[27:12] Alex, um, you know, would we sort of use
[27:15] a bank or a traditional financial system
[27:18] account? So, I I think if if somehow we
[27:21] could wave a magic wand and say, okay,
[27:23] KYC is no longer required. You know, the
[27:25] AI can sign up or the dog can sign up.
[27:27] Um, I don't think that's going to happen
[27:29] for one thing, but if it did happen,
[27:31] that would that would solve part of the
[27:32] issue. But the other issue is that the
[27:35] infrastructure underlying those
[27:36] traditional financial system accounts is
[27:38] just slow. It's, you know, it's running
[27:39] on cobalt servers from the 1990s and 80s
[27:43] and and so the beauty of transacting on
[27:46] chain with stable coins is that the
[27:48] infrastructure is just much more f it's
[27:49] just faster and cheaper and more global.
[27:51] So you can now send USDC for instance um
[27:54] in under one second anywhere in the
[27:56] world for less than a cent uh US. So
[27:58] it's it's just faster and more scalable.
[28:01] I to do microtransactions and glo
[28:04] crossborder transactions. So I I
[28:06] probably still wouldn't go back and
[28:08] build it on top of the main frames but
[28:10] um but the KYC thing would definitely
[28:11] help. Let's put it that way. And
[28:13] >> amazing.
[28:14] >> And Brian, you know, this strategy
[28:17] potentially breaks all existing
[28:19] regulations and KYC assumptions. So
[28:22] talking about liability, you know, if an
[28:24] AI if an agent overpays or gets scammed
[28:27] or launders value, I mean have have you
[28:30] thought through, you know, liability
[28:32] issues?
[28:34] >> Yeah. Well, I think there's um we need
[28:36] to get some legal precedent on this. I
[28:38] think one school of thought would say
[28:39] that all agents are actually controlled
[28:42] by some human or some company and so
[28:46] their actions ultimately from a
[28:48] liability point of view roll back up to
[28:49] that company or person. I could see that
[28:52] being one potential outcome, maybe even
[28:53] the default outcome. I could see another
[28:56] world that we enter into. who I don't
[28:57] know if society is ready for this yet,
[28:59] but if we start having truly intelligent
[29:01] agents that are autonomous that aren't
[29:04] really they are their own person, you
[29:05] know, like um I don't that'll be an
[29:08] interesting uh legal case when that
[29:09] first gets brought about that thing, you
[29:11] know, and maybe it'll it'll get
[29:12] sentenced to um you know, have to uh be
[29:16] in solitary confinement or whatever is
[29:18] like
[29:20] >> Yeah, I don't I guess you could have a
[29:22] financial penalty or whatever is
[29:24] meaningful to the agents. they have to
[29:25] run on, you know, less power for a while
[29:27] or something. Um, but um, yeah, I think,
[29:32] you know, the other thing is that
[29:34] I we we want to try to make it so that
[29:36] there's just less fraud in this new
[29:38] financial system. And, um, one way you
[29:40] can do that
[29:41] >> is actually have a reputation on chain.
[29:43] This is something that
[29:44] >> we're hoping to build out over time. But
[29:46] >> I like that.
[29:47] >> Um,
[29:47] >> yeah, it's kind of like what Google did
[29:49] with the internet. You know, they came
[29:50] up the page rank algorithm and they kind
[29:52] of said, "All right, how do you know the
[29:53] reputation of this website?" Well, let's
[29:55] look at all the other websites that put
[29:57] links to it and how reputable are those
[29:59] websites. And so you get this kind of
[30:00] graph structure that Larry Page came up
[30:03] with um famously. And so you could do
[30:05] something similar on chain cuz onchain
[30:08] payments are just another graph
[30:09] structure. It's you know if I send money
[30:10] from me to Peter and maybe if I'm a high
[30:13] reputation then the amount of that money
[30:16] times my reputation gets assigned to
[30:18] Peter as some sort of reputation signal.
[30:20] It's almost like a like an onchain FICO
[30:22] score or like a Yelp rating for a
[30:25] business or something like that. And so
[30:27] hopefully, you know, you can go buy
[30:29] things or an agent could buy things um
[30:30] with some amount of additional
[30:32] information to know how many how often
[30:34] did people request a refund for this? Um
[30:36] or what's the what's the reputation of
[30:38] this merchant and like like on eBay, I
[30:41] mean I I love this say, you know, don't
[30:43] do business with this agent, it's got a
[30:44] low score or this agent is, you know,
[30:46] got 100% Uber score. See, what are you
[30:49] thinking here?
[30:51] Well, two things. One is uh I think the
[30:54] once you allow agent e-commerce, this
[30:57] fully kind of enables the MTAM economy,
[31:00] which is going to be the future, right?
[31:02] Obviously, human transactions can be a
[31:05] dot a drop in the bucket based on that.
[31:07] Um because the minute agents can kind of
[31:09] negotiate, pay, buy, consume, etc., you
[31:12] we talked about the organizational
[31:14] singularity that's going to collapse
[31:15] coordination costs again. So there's
[31:18] unbelievable for me. This is the most
[31:19] important inflection point in anything
[31:22] we talk about is creating the payment
[31:24] rails and capabilities for AI agents to
[31:28] transact in a reasonable way. So, this
[31:30] is incredibly powerful.
[31:31] >> All right, let's get on to the meat of
[31:33] one of the most important stories here
[31:35] today. Uh, and it is the US government
[31:38] exploring ownership stakes in AI
[31:40] companies. Uh, this is one we've talked
[31:43] about before. It's getting more real.
[31:45] Trump called a government stake in AI's
[31:48] giants a beautiful thing and then
[31:50] floated the idea that pieces could be
[31:52] given to the American public to share in
[31:53] the economic gains. And this isn't
[31:55] without precedent, right? The government
[31:57] already holds stakes in more than 20
[31:59] private companies. Here are some of the
[32:00] numbers. 10% of Intel, 15% of MP
[32:04] materials, 10% of lithium Americas, 10%
[32:06] of Trilogy Metals, 10% of USA Rare
[32:10] Earth, and 10% of Korea zinc. Uh guys,
[32:13] to me it sure looks like there's a
[32:15] precedent for 10% stake in the Frontier
[32:17] Labs as well. Uh Dave, I want to go to
[32:20] you first. I mean, one of the questions
[32:22] that fascinates me is what would the
[32:24] government do with these shares? Sell
[32:27] them, keep them.
[32:29] >> What are your thoughts?
[32:29] >> Well, I'll tell you it'll I'll tell you
[32:31] what'll happen. You've got these are
[32:32] very good moves. I'm not I'm not saying
[32:34] these are bad moves, but while the
[32:36] president is making these moves, he's
[32:37] talking about how insanely stupid the
[32:39] prior investments were, including
[32:40] shipping $1.8 8 billion in cash to Iran
[32:43] and how how stupid the last
[32:45] administration was. So if you put in
[32:47] place a precedent of the federal
[32:49] government choosing what to invest in in
[32:51] the private economy, you may love it for
[32:54] a little while, but there's an
[32:55] administration coming eventually that
[32:57] you won't love. It's a horrible,
[32:59] horrible thing. It's exactly what
[33:01] Eisenhower warned us against actually
[33:03] right after World War II. Beware of the
[33:05] military-industrial complex. Well,
[33:07] that's exactly where we're going now.
[33:09] These particular investments are great
[33:11] and I I use the analogy to World War II
[33:13] a lot, you know, where what we're going
[33:15] in through right now is so urgent and so
[33:19] um world changing. It's it's most
[33:21] similar to to decisions we made during
[33:24] World War II, which is all about
[33:25] survival and existential, you know,
[33:28] threats. And um I think these are very
[33:31] very good moves in the context of the
[33:34] race to AI uh with China and terrible
[33:38] long-term president. So So you know
[33:40] Peter, you asked a very specific
[33:42] question. Which president is going to
[33:44] sell this stock
[33:45] >> and when? This isn't like a tax that
[33:48] rolls in steadily and you use it
[33:49] annually. this is a future decision that
[33:52] some other president will need to make
[33:54] and could could tank these stocks if the
[33:56] federal government dumps huge positions
[33:58] in some kind of a a future rent which it
[34:01] inevitably will do. So definitely a
[34:04] mixed bag and
[34:05] >> there's an argument that these companies
[34:06] could be the most uh you know highest
[34:08] value companies on the planet. uh and if
[34:12] we're going to need some kind of a
[34:13] financing mechanism for some future
[34:15] version of UBI, uh this sounds like it's
[34:19] gaining the most momentum. Alex, what
[34:21] are your thoughts? If you're an AI
[34:23] maximalist and you extrapolate say the
[34:26] enterprise value of anthropic or its ARR
[34:30] and you find that it intersects with
[34:32] Google in the next 18 months or that you
[34:35] find out if you naively extrapolated out
[34:37] exponentially becomes larger than the
[34:39] American economy of today in a few years
[34:43] then under that prior it's almost
[34:46] difficult not to imagine some sort of
[34:48] quasi nationalization whether it's via a
[34:51] golden share scheme or some other
[34:53] arrangement. If if OpenAI and Anthropic
[34:56] as the deacto duopoly that we have right
[34:58] now, if they're larger than the rest of
[35:01] the American economy combined, it's
[35:03] difficult to imagine a case, at least
[35:06] maybe a limitation of my own
[35:07] imagination, where there isn't some
[35:10] strong formal arrangement between those
[35:13] companies and the US government. So I I
[35:16] think if you really do believe if you're
[35:18] drinking the Kool-Aid and you believe
[35:19] that super intelligence coming from a
[35:22] small minority of companies in the US
[35:25] right now are going to dominate the
[35:28] future economic light cone. I think some
[35:31] variant of quasi nationalization or
[35:34] government private sector hybridization
[35:36] or call it a private public
[35:38] collaboration if you like that
[35:40] euphemism. I think some variant thereof
[35:42] is probably inevitable. And I do think I
[35:45] So, we talked a bit about this on the
[35:48] past two pod episodes. Yes, I do think
[35:50] the singularity is can be operationally
[35:54] defined as every sci-fi trope happening
[35:56] everywhere all at once. And we've got a
[35:58] few in this story. We've got UBI or UBC
[36:02] or UB on the one hand. We have super
[36:06] intelligence on the other. And I it's
[36:09] just I I think if ever there were a time
[36:12] for some sort of UBI or universal basic
[36:16] dividend or universal basic equity or
[36:18] universal basic computer capability
[36:20] scheme or universal basic services
[36:22] scheme to manifest in government policy.
[36:25] It has to happen approximately now or in
[36:28] the next year or two.
[36:30] >> I totally agree with that Alex but I
[36:31] want to throw two things back at you. Uh
[36:33] first the government has infinite power
[36:36] of taxation. it doesn't need to own
[36:37] equity in companies to extract any
[36:39] amount of money it wants from any AI
[36:41] entity. So it's just it's just a you
[36:44] know it's a fact that the government has
[36:46] all the money extraction ability in the
[36:48] world. Um, but the other thing is I
[36:50] think you talked about in a prior pod
[36:52] the fact that the original atomic energy
[36:55] uh commission or the NRA I forget which
[36:56] one or NRC
[36:58] it was empowered its whole mission under
[37:00] Nixon was to make nuclear energy happen
[37:04] in America and the net effect it had was
[37:07] to prevent nuclear power from existing
[37:10] in America and you talked you talked
[37:11] about that at length and so the the act
[37:14] of nationalization is inevitable I know
[37:16] with AI I completely agree
[37:18] But the idea that that the government
[37:20] somehow empowers progress in that
[37:22] process, it only inhibits progress.
[37:24] >> I would draw a distinction. So the the
[37:26] Atomic Energy Commission was formed at
[37:28] the AEC in the wake of World War II
[37:31] under the premise that atomic energy had
[37:34] essential civilian applications and also
[37:37] was too important to be left to the
[37:39] military and also too important to be
[37:41] left to the civilian government and to
[37:43] at least solely and too important to be
[37:44] left solely to the private sector. So
[37:46] the atomic energy commission was formed
[37:48] as sort of this hybrid organism and that
[37:51] evolved eventually into the department
[37:53] of energy. The problem that I perceive
[37:55] with the AEC and with uh subsequent
[37:58] spin-offs and derivatives like the
[37:59] Nuclear Regulatory Commission is it was
[38:03] almost born in in war. It was it was
[38:07] born out of a time in a place where you
[38:09] have uh in the early days of the Atomic
[38:12] Energy Commission almost a sense of
[38:14] guilt that hung over Manhattan Project
[38:17] scientists who were losing quite a bit
[38:20] of sleep over who would own the children
[38:23] of atomic weaponry. Here I I would say
[38:26] at least one substantive difference is
[38:29] super intelligence isn't being born out
[38:31] of Hiroshima or any equivalent. It it
[38:34] started from the private sector with
[38:37] private sector scientists. The US
[38:38] government I mean Sam famously offered
[38:41] the US government a stake early on uh in
[38:44] open AI in order to to seek earlier
[38:46] funding and US government reportedly
[38:48] turned down open AI. So I I would say
[38:52] qualitatively if you compare the the
[38:54] rise of atomic power and how it was
[38:57] regulated in this country during and
[38:59] after World War II with AI technology in
[39:02] some sense polar opposites started from
[39:04] the private sector. US government could
[39:06] have taken an early stake could have
[39:08] been uh at least nationalized at a very
[39:11] early age could have been militarized at
[39:14] a very early stage and that just didn't
[39:16] happen. So, I'm a lot less worried that
[39:18] somehow the government will step in, at
[39:21] least for the next few years under this
[39:22] administration, step in and pull an NRC
[39:25] and find ways to to slow everything
[39:28] down. Uh, unless the the Frontier Labs
[39:30] in this country really do want to slow
[39:32] down.
[39:33] >> Brian, Brian, you're running a $40
[39:35] billion public company right now. How
[39:38] does this make you feel?
[39:41] Um I don't spend too much of my time
[39:42] thinking about it but um you know high
[39:44] level any company that reaches a certain
[39:46] size is going to interface with the
[39:48] government. Um but as others have
[39:50] mentioned here that can take many
[39:51] different forms right paying taxes is is
[39:53] the most obvious one. Um you're going to
[39:56] engage with them on various policy
[39:57] issues and this administration's been
[40:00] very open and eager to make improve
[40:02] policy around crypto which has been
[40:03] great. Um you may the government may
[40:06] actually end up being a customer uh of
[40:07] your product as well. So that that's
[40:09] another touch point. Um you know we
[40:11] actually provide uh crypto services to
[40:14] about 140 different government agencies
[40:15] at federal, state and local levels. Um
[40:19] and so to me that's sufficient. I don't
[40:21] think going farther than that and
[40:23] actually having them take an ownership
[40:26] stake in the company unless you know if
[40:29] I guess if if there let's say there was
[40:31] a matter of national security and this
[40:32] companies were not would not exist
[40:34] otherwise
[40:36] um and there was fundraising happening
[40:39] from the government I guess you could
[40:41] they could end up with some equity in
[40:42] that situation but it just begs the
[40:44] question of like well who's going to be
[40:45] managing this portfolio of company
[40:47] stocks when you know when would they
[40:49] sell them because it's not like a
[40:50] foregone conclusion that Enthropic and A
[40:52] are going to be the biggest AI companies
[40:53] forever. It might end up being that, you
[40:55] know, Google or someone else does a or
[40:57] SpaceX does a better job, right? So now
[41:00] you're suddenly saying, okay, we're
[41:02] going to have like capital management
[41:03] happening inside the government. I I
[41:05] think the government should be limited
[41:06] to setting policy
[41:08] >> and the private sector should do the
[41:09] capital allocation.
[41:10] >> Nice.
[41:11] >> Yeah. Considering that that companies
[41:14] donate to political campaigns,
[41:16] >> you think about how toxic that circle
[41:18] is. Oh, okay. this this government is
[41:20] investing in my company and I'm going to
[41:22] turn around and donate to their next
[41:24] election campaign.
[41:25] >> Like that is the most toxic circle I can
[41:28] possibly imagine. Can you imagine what
[41:30] what
[41:31] >> democracy in America?
[41:32] >> Yeah. Um
[41:33] >> I'm I'm curious, Brian, though. Do you
[41:35] think that as a country we should have a
[41:37] sovereign wealth fund?
[41:40] >> Okay. So, that's a complex topic. I
[41:41] think um there's parts of it that I like
[41:44] a lot and parts of it I don't. So, the
[41:45] part that I like is that it would allow
[41:47] everybody to have skin in the game. And
[41:49] I think we'd see more cohesion,
[41:52] um, less polarization,
[41:54] and it's a little bit like in a company,
[41:56] you know, inside Coinbase, employees get
[41:58] stock options because if we all have we
[42:00] we're kind of grumpy today and we get in
[42:02] a little fight with our coworker, but
[42:03] hey, we all own we're making something
[42:05] bigger than this. So, we all kind of
[42:06] stick around and work out our
[42:07] differences. So, in that sense, I love
[42:09] the idea of people being able to buy
[42:11] into a sovereign wealth fund, but also
[42:13] maybe every citizen getting a share at
[42:15] birth. It's a little bit like what
[42:16] they're doing with these Trump accounts
[42:17] right now. So, I think that'll actually
[42:18] >> make people more aligned to free market
[42:21] capitalism and have skin in the game.
[42:23] The downside is kind of what we talked
[42:25] about earlier, which is who's allocating
[42:26] that money, right? If if we were in sort
[42:29] of a Singapore model where like the top
[42:31] uh people in private industry got
[42:34] commensurate compensation to come into
[42:36] government like I think you know Michael
[42:37] Grimes from Morgan Stanley I think came
[42:40] in and was working with the
[42:41] administration at some point like
[42:42] someone like him would probably do a
[42:44] good job managing that allocating it. Do
[42:46] I believe that that's going to exist in
[42:48] all in every administration for you know
[42:50] over time? Like it's just very hard to
[42:52] see it getting it could very easily get
[42:54] politicized right where they start
[42:55] investing in the green revolution
[42:57] whether that works or not or you know
[42:59] all all kinds of things like that. So I
[43:01] uh I don't know I'm torn I'm torn on
[43:03] that one. We got to find some way to
[43:04] bring cohesion back to the US but I'm
[43:07] not sure if a sovereign wealth fund
[43:08] would be managed well over time. Like
[43:10] just having a strict rule like keep it
[43:11] in the S&P 500 would be a nice
[43:13] mitigating factor. something something
[43:15] like that.
[43:16] >> Welcome to the health section of
[43:17] Moonshots brought to you by Fountain
[43:18] Life. You know, AI is having an outsiz
[43:21] impact on every aspect of our lives. How
[43:23] we teach our kids, how we run our
[43:24] companies. It also is having a huge
[43:26] impact on health, helping you prevent
[43:29] heart disease. One of the key things I'm
[43:31] here with Dr. Don Mucalem, our chief
[43:33] medical officer at Fountain. Heart
[43:35] disease has been personal for you as
[43:36] well, hasn't it?
[43:37] >> It really has, Peter. When my daughter
[43:39] was five, my husband died of sudden
[43:41] cardiac death. And so this is a topic
[43:43] that is one that I am missiondriven to
[43:46] try to eradicate. Prevention first and
[43:48] early detection is absolutely critical.
[43:51] 50% of people die of heart attacks with
[43:53] no warning signs.
[43:54] >> No shortness of breath, no pain, no
[43:56] nothing.
[43:56] >> No silent killer.
[43:57] >> They just don't wake up in the morning.
[43:59] >> They don't wake up. And so you know AI,
[44:01] this is our mission to advance science
[44:03] to try to help to one day democratize
[44:05] wellness. We know at fountain life when
[44:07] we do this CT angography with AI
[44:10] analytics, we are actually finding that
[44:12] 88% of people coming in have detectable
[44:16] coronary artery disease. But Peter,
[44:17] what's more alarming to me is 23% of
[44:20] those individuals had soft plaque. This
[44:23] is the plaque that would not
[44:24] traditionally be seen on CT looking at
[44:26] calcium scores alone. And this is the
[44:28] plaque that we must intervene with with
[44:31] the multimodal testing we're doing,
[44:33] including diagnostic laboratory studies
[44:35] partnered with healthy lifestyle
[44:37] recommendations.
[44:38] >> So listen, make sure you understand
[44:40] what's going on inside your body
[44:42] genetically, metabolically, and
[44:43] cardiovascularly. You can know, and it's
[44:45] your obligation to know. So check it out
[44:48] at fountainlife.com/per
[44:50] to find out more and really make sure
[44:53] that you're the CEO of your own health.
[44:55] All right, back to the episode. Let me
[44:57] hit the second story related to this one
[44:59] moment. Uh I don't want to belabor the
[45:01] point, but you know Sam Alman sat down
[45:03] with Senator Sanders who's proposing a
[45:05] transfer of 50% of equity from the top
[45:07] AI companies into a public fund. Almond
[45:10] responded, you know, that's not going to
[45:13] happen. That's way too much. Uh the
[45:15] point is we've started a negotiation. Uh
[45:18] you know 5 years ago this conversation
[45:19] would have been unthinkable. You know
[45:20] the Overton window on AI wealth
[45:22] distribution uh has moved dramatically.
[45:25] You know, I said earlier we see all
[45:27] these examples of 10% ownership, and if
[45:30] I had to guess, uh, I think this is not
[45:32] going to go away. I think we're going to
[45:34] see a push towards at least a 10%
[45:37] ownership. Uh, Dave, what do you think
[45:39] about that?
[45:41] >> No, I think it's utterly insane. That's
[45:42] what I've said before.
[45:43] >> I know it's insane. Do you think it's
[45:45] going to happen?
[45:46] >> Uh, no. I don't think it'll happen. I
[45:48] think Bernie's just trying to attract
[45:49] news. Um, and you know, well, you know,
[45:52] Donald can invest in whatever he wants
[45:54] to invest in. He has proven that he'll
[45:56] make that decision on like five minute
[45:57] notice, you know, after a meeting. So,
[45:59] so I can't rule out uh but it it doesn't
[46:04] doesn't strike me as one of the most
[46:05] likely investments that that Trump would
[46:08] make. You know, Trump invested heavily
[46:09] in semiconductors moving back into the
[46:11] US, investing in OpenAI and anthropics
[46:14] IPOs, but they're already 100% taxable
[46:16] US entities that are thriving. Is this
[46:18] an investment or a grant of shares to
[46:21] the US government?
[46:25] >> Bernie's proposal is a is a is a tax.
[46:27] So, it's just a grant of shares to the
[46:29] US government,
[46:29] >> right?
[46:32] >> I I think it's going to be you know
[46:34] equivalent to uh you know Alaska's
[46:37] permanent fund. You've been using our
[46:39] intelligence. You've been using our
[46:40] energy. You've been using our real
[46:41] estate, our oxygen. Uh we going to claim
[46:44] 10% and we're going to you know give it
[46:46] to the people.
[46:48] I would say look at it from the Frontier
[46:50] Labs perspective. Put aside all of any
[46:52] politics or any technical arguments or
[46:54] any legal or regulatory arguments. Just
[46:56] think about it from an optical or a
[46:58] marketing perspective. If you're open AI
[47:01] or you're anthropic and you have the
[47:03] opportunity to say donate 10% equity
[47:06] through through some legal scheme to the
[47:09] US government one time 10% maybe in
[47:12] connection with your IPO this year. Now
[47:14] suddenly in principle you're inoculated
[47:17] at least politically against any future
[47:20] claims that you owe a UBI or or some
[47:24] other payment to the citizenry. You can
[47:27] say I already gave up my pound of flesh.
[47:29] I gave it to the United States and it's
[47:33] sitting in a sovereign wealth fund now
[47:35] and it's generating returns for the
[47:37] American people. That was your one-time
[47:40] donation. I think if if I'm Sam or if
[47:43] I'm Daario, that's looking pretty
[47:45] attractive.
[47:47] >> And and coupled with that, Alex, you've
[47:49] also got a conflict, right? The same
[47:51] state that regulates AI now profits from
[47:54] your success.
[47:57] Well, the federal government a a a
[48:01] incredibly toxic incentive for the the
[48:04] two mega companies that it extracted
[48:06] equity from to succeed at the expense of
[48:09] a startup that isn't part of that same
[48:12] loop. It's an incredibly toxic dynamic.
[48:15] Now, like you're saying, the American
[48:17] people now benefit from these two
[48:18] companies
[48:20] >> and and you know, Brian said a second
[48:21] ago, there will be other companies that
[48:22] we haven't thought of yet. you know what
[48:25] do you do with them? They're not part of
[48:26] the US investment.
[48:28] >> Yeah, there's a broader incompatibility
[48:30] here. Think about um these labs and AI
[48:34] labs becoming civilization scale
[48:36] infrastructure, right? That means you
[48:38] have to treat them like strategic
[48:40] utilities much like electricity. You
[48:42] regulate the utilities in a particular
[48:44] way and deal with that. You need a
[48:46] structure like that. The problem is
[48:48] there's an incompatibility here because
[48:49] there's no governance structure that
[48:51] fits a technology that touches like
[48:54] every job, every industry, every
[48:56] government service. So there's a huge
[48:58] challenge and you can't pick the
[49:00] winners. It's more a structural problem
[49:02] that the mechanisms taxation or
[49:04] ownership. We don't even have a
[49:06] sovereign wealth fund. So how do you
[49:07] deal with that? So this is a this is a a
[49:10] big in incompatibility problem that I
[49:12] see here. We're going to mess it up.
[49:14] whatever we do in the current structure.
[49:16] >> I don't think this is going away. I
[49:18] think we're going to be seeing this
[49:19] happen at some level. Alex, I
[49:22] >> I'll just comment if if you remember
[49:23] back, one of the earliest executive
[49:25] orders from this administration was to
[49:27] order the Secretary of Treasury to
[49:29] explore setting up a sovereign wealth
[49:30] fund. And far be it for me to recommend
[49:33] national economic policy but it's not
[49:35] difficult for me to imagine a scenario
[49:37] where the US Treasury does take golden
[49:40] shares call it 5 to 10% in open AI and
[49:44] anthropic to complement its existing
[49:45] portfolio and then some amount of time
[49:48] from now maybe sometime in the next few
[49:50] years there's a great rebalancing we get
[49:52] our sovereign wealth fund but to Brian's
[49:54] earlier point rather than being a basket
[49:57] of 20 or 30 or 40 techoriented or
[50:00] strategic companies. Instead, the
[50:03] federal government eases into say an S&P
[50:06] 500 or a total market index, gradually
[50:09] liquidating its holdings in the
[50:11] individual companies in favor for a
[50:13] total market index. And that children is
[50:16] how we got our universal basic income or
[50:19] equity.
[50:20] >> Mhm. Yep. All right. Uh late breaking
[50:24] story, OpenAI has officially filed its
[50:26] S1 to go public later this year. Poly
[50:29] market bets. 46% say it'll come out at
[50:32] 1.5 trillion or greater and 26% say it's
[50:36] not going to happen this year. Uh this
[50:38] would be the third trillion dollar IPO
[50:40] following Anthropic and SpaceX AI which
[50:44] is coming at 1.77 trillion in just a
[50:46] couple of days. I I wanted to just point
[50:48] out a conversation we had on the pod I
[50:51] don't know a month and a bit ago uh in
[50:53] which I don't you guys remember uh Sarah
[50:56] Frier the CFO came out and said
[50:58] privately that she was very concerned
[51:01] that openi was not ready to be a public
[51:03] company you know it had a risk of 600
[51:07] billion of compute contracts on its
[51:09] books uh I guess she thinks they're
[51:12] ready now Dave what are you thinking
[51:14] here I
[51:15] >> I think it's more that times have
[51:17] changed I I think uh every CFO I've ever
[51:19] met wants visibility at least three or
[51:21] four quarters into the future, but in
[51:24] the singularity, no one's ever going to
[51:26] have visibility three or four quarters
[51:27] in the future ever again. That's just
[51:29] the nature of technology. Yeah, exactly.
[51:32] So, so I think CFOs like Sarah just need
[51:34] to get comfortable with look, this is
[51:36] the way the future has to be.
[51:38] >> Uh and you know, what used to be three,
[51:41] four, four, five quarters of visibility
[51:42] is now three, four, five months is is
[51:44] the best you're ever going to get. And
[51:45] even that is a shrinking horizon. But
[51:48] the economy marches on. You know, we're
[51:50] have to get used to the, you know, if
[51:51] you look at just the daily volatilility
[51:53] of the market, it's it's so high now
[51:55] because tech is just that's just the
[51:57] nature of tech. So, uh, so I think the
[51:59] CFOs of the future will get used to it.
[52:01] And but you you can't miss the IPO
[52:03] window, right? If you get SpaceX going
[52:05] out and you get Anthropic going out, you
[52:07] have to go out.
[52:08] >> The challenge is coming out coming out
[52:10] third during a cash crunch, right? I
[52:11] mean, there's only so much liquidity out
[52:13] there, and if you're coming out third
[52:16] and, you know, SpaceX is skyrocketing
[52:18] and and Anthropic is, you know, growing
[52:21] at 640% per year, uh, it's dangerous to
[52:24] come out third.
[52:26] >> It is, but if you if you said, "Okay,
[52:28] the counterargument is then, let's get
[52:30] let those other two IPOs happen. Let's
[52:32] let them digest their hundreds of
[52:34] billions of dollars and let the market
[52:36] recharge, and then we'll go out later
[52:38] when it's recharged." You're going to be
[52:40] waiting years.
[52:40] >> Yeah. you know, you're not talking about
[52:42] 3 months later. So then then you're
[52:44] like, okay, you can't you can't wait
[52:46] years. Anything could happen in between.
[52:47] Meanwhile, these guys are fully tanked
[52:49] up and moving 100,000 miles an hour. No,
[52:53] you don't want you don't want to be that
[52:54] company. So, you got to go. You got to
[52:56] go.
[52:56] >> When you've got these trillion dollar
[52:57] IPOs, the whole kind of corporate rapper
[52:59] seems way too small for what's going on.
[53:03] >> This is so unprecedented compared to
[53:05] anything in human history. It's just
[53:06] absolutely wild.
[53:08] >> Yeah. Does it break?
[53:09] >> Yeah. Go on, Alex.
[53:11] >> Remember also these concerns? April was
[53:13] a whole two months ago. That's several
[53:15] lifetimes in singularity time scaling
[53:18] ago. Open AAI had recently come out of
[53:21] its code red. They were shutting down
[53:24] Sora and some of their other divisions
[53:26] trying to become anthropic faster than
[53:28] anthropic could become Open AI. pivoting
[53:30] to codeex and then arguably most
[53:33] importantly taking Stargate which was
[53:35] originally conceived as being OpenAI
[53:38] owned and operated data centers and
[53:41] rebranding basically a leasing scheme as
[53:44] the new Stargate. So I I my sense as an
[53:48] outsider without access to internal
[53:50] accounting is that through a combination
[53:53] of new revenue sources like basically
[53:56] becoming the codeex company uh removing
[53:59] other expenses like Sora and like their
[54:02] AI for science division and then
[54:05] effectively lobomizing Stargate and
[54:08] switching it over to a leasing model
[54:09] rather than an ownership model which if
[54:11] I were CFO of OpenAI I would have been
[54:14] scared to death of the ownership. model
[54:16] given all of the price girrations in the
[54:18] GPU market. I I would expect that given
[54:21] all of those measures that open AI again
[54:24] not investment advice but comes out of
[54:26] all of those with a much more stable and
[54:29] predictable prospectus than the open AI
[54:32] that came into this year that looked
[54:34] much more like a consumer play volatile
[54:38] that was going to own its data centers.
[54:40] Scary.
[54:40] >> They they pulled it off, Alex. They
[54:42] really pulled it off. They turned it
[54:44] around.
[54:45] They did, but that's probably what
[54:46] freaked out Sarah. If you looked at the
[54:48] forecast from a year ago, it it had 20
[54:50] billion of ad revenue and I forget, but
[54:52] many, many billions of consumer
[54:54] subscription revenue and then a little
[54:55] bit of enterprise.
[54:57] >> And now, if you look at the forecast,
[54:58] it's a complete inversion of the colors
[55:00] on the bar chart. That always scares a
[55:02] CFO, right?
[55:03] >> Brian, you know, I have to imagine that
[55:05] being a founder tech company is the big
[55:08] advantage that you have over traditional
[55:11] large corporations. I mean, being able
[55:13] to say we're going this way uh versus
[55:15] that way, but for all my friends who are
[55:18] public company CF CEOs, uh it it's not
[55:22] fun. How do you think about about, you
[55:25] know, remembering when you were a couple
[55:27] months before your IPO? What's going on
[55:29] inside of a company theoretically like
[55:31] this?
[55:34] >> Well, a couple things come to mind. One
[55:36] is um there's always there should be
[55:38] some healthy tension, I think, between
[55:40] like a founder, CEO, and a CFO. I mean,
[55:42] that's founders are great at many
[55:44] things. Um, but if you just have too
[55:46] much founder energy, sometimes they blow
[55:47] the place up, right? We can probably
[55:48] think of our favorite companies in that
[55:50] example. Um, but simultaneously, you
[55:53] don't just want to have risk-minded
[55:54] operators uh running the show. And that
[55:57] that's just a recipe for incrementalism
[55:59] or even slow decline, right? So, I think
[56:01] it's sometimes the press likes to report
[56:03] on these things of like, oh my gosh,
[56:05] there's drama happening between, you
[56:06] know, I'm like, that's called a healthy
[56:08] conversation. So I imagine that they're
[56:10] they're actually working like relatively
[56:12] well together on this and it's a high
[56:13] stakes moment for both of them.
[56:15] >> Um the other thing is like sometimes
[56:17] people criticize OpenAI like oh my gosh
[56:19] they missed their their target uh their
[56:21] end ofear target to get a billion users
[56:23] or something like that and you know
[56:24] internally we all we often use like the
[56:26] OKR system and we we tell we tell
[56:29] everybody set uncomfortably ambitious
[56:30] goals and if you hit at least 70% of it
[56:32] that's considered a good outcome. So
[56:34] when whenever companies switch into the
[56:36] public mindset, they're often setting
[56:38] their goals start to become more like
[56:40] appropriate for external investors where
[56:42] wow, we just hit our goals every quarter
[56:43] around here. Who you know? Um and so
[56:46] anyway, there's a lot of t stuff like
[56:47] that that happens behind the scenes. I
[56:49] think the highle bit is right is what
[56:51] you guys already covered is they've got
[56:53] to get out just because that's what
[56:55] their competitors are doing and that's
[56:57] where the capital is going to all go.
[56:59] So, it's going to be uncomfortable and
[57:00] messy and I'm a little worried about um
[57:03] these valuations where retail is going
[57:06] to, you know, be the the buyer of this
[57:08] and take the hit.
[57:10] >> Yeah, it could be pretty choppy here.
[57:12] Um, you know, just self because we
[57:14] talked about how AI is sort of taking
[57:16] all the oxygen out of the room. Like
[57:18] selfishly, I think other techos are kind
[57:20] of like, all right, finally get these
[57:21] things public so there can be a real
[57:22] mark on it and maybe um some of the the
[57:25] hype dies down a little bit. I I also
[57:27] think there's going to be fierce
[57:28] competition amongst these these models
[57:30] because they keep like the open source
[57:32] models are basically just as good and 3
[57:34] to six months behind for like 99%
[57:37] >> that much if that much.
[57:38] >> Yeah. And they they're like literally
[57:40] 99% cheaper for inference. So I think
[57:43] that the demand for intelligence is
[57:45] almost infinite. It probably is infinite
[57:47] but the costs are going to fall like way
[57:49] more than Mo's law. So um you know are
[57:52] is the harness that these guys are
[57:54] making whether it's codec or claude uh
[57:56] claude code are those valuable? I don't
[57:58] think the harness is where the value is.
[58:00] I think that um like the the the
[58:03] depreciation on these models like how
[58:05] how fastly how fast the price of them
[58:07] drop for the same level of intelligence
[58:09] is going to be dramatic. And I think
[58:10] within 12 18 months 80% of our workloads
[58:14] are going to be going to models that are
[58:15] 99% cheaper.
[58:18] >> Yeah. I Brian I had a meeting with a
[58:20] Morgan Stanley guy right after the
[58:22] Facebook IPO and he said oh my god this
[58:25] was the worst disaster in in US
[58:27] financial history with the stock was
[58:29] down I think about half after the IPO
[58:32] and nothing wrong with the company
[58:33] obviously and it was a great investment
[58:35] in hindsight obviously but there just
[58:37] wasn't enough liquidity so they they
[58:39] dumped it in all their private client
[58:41] accounts then watched it go down by half
[58:43] and now we're doing this three
[58:46] backtoback IPOs
[58:48] each one of which is a factor of 10
[58:51] bigger than anything we've ever seen
[58:53] before. And so I I'm just curious, the
[58:55] logistics of this could be a nightmare.
[58:58] Even if the companies are great and even
[59:00] if five years from now they're way up
[59:01] from their IPO valuations, is there
[59:03] really that much money in the world?
[59:06] Like does it actually exist?
[59:09] >> And it'll be we'll find out, I guess.
[59:11] >> Yeah. And and right now, you know, I
[59:12] have a lot of friends who like if I'm
[59:14] not, you know, pitching my AI story for
[59:16] my company, I can't get money. AI is
[59:19] sucking the oxygen out of out of the
[59:21] room everywhere. All right, this next
[59:24] story generally stopped me in my tracks.
[59:26] Uh, you know, Google, one of the biggest
[59:28] AI infrastructure players on the planet
[59:30] with its own TPUs and data centers, is
[59:32] paying SpaceX 11 billion per year
[59:35] through 2029 to access 110,000 NVIDIA
[59:39] GPUs inside of XAI's data center. Uh,
[59:42] you know, SpaceX is now clearly a
[59:44] hyperscaler. And it's ironic, Google is
[59:47] renting compute from Elon's rocket
[59:49] company, which by the way, Google is an
[59:50] investor in Elon's rocket company. I
[59:52] love this quote from Austin Reef who put
[59:54] it. It's crazy that with one deal,
[59:58] SpaceX IPO went from 100x revenues to
[01:00:01] 50x revenues. Uh, one contract have the
[01:00:05] company's valuation multiple, you know,
[01:00:07] and and this tells us, you know, clearly
[01:00:09] the story. There's a massive AI compute
[01:00:12] shortage.
[01:00:14] >> Yeah. So
[01:00:15] >> love to get your take on this Alex
[01:00:16] because uh you know there was a view of
[01:00:19] the world where Elon was building a
[01:00:21] frontier lab with a you know
[01:00:23] first-in-class model. Now if he rents
[01:00:26] all the compute to Anthropic and to
[01:00:28] Google then clearly he's backed out of
[01:00:31] that game completely. I'm not quite sure
[01:00:33] you know when he did the first rental to
[01:00:35] Anthropic he said that's Colossus one.
[01:00:38] Colossus 2 is still grinding away on the
[01:00:40] you know the the mother of all frontier
[01:00:42] models. You're going to actually when we
[01:00:44] interviewed him in Austin, remember
[01:00:46] Peter? He said that you're going to be
[01:00:48] absolutely blown away by it was Grock 5.
[01:00:50] >> Yeah,
[01:00:51] >> it's it's it's brilliant. And you know,
[01:00:53] here we are. That was due out in March
[01:00:55] and here we are. It's June.
[01:00:57] >> Well, he's clearly refocused his
[01:00:58] attention.
[01:01:00] >> Well, all of Colossus 2, do you know,
[01:01:02] Alex?
[01:01:03] >> I I think a lot of it uh is my
[01:01:05] understanding. So, I I took a bit of
[01:01:07] heat for saying a few episodes ago that
[01:01:10] I I thought Grock in the wake of the
[01:01:12] anthropic SpaceX deal was on life
[01:01:14] support and Peter and I had a back and
[01:01:17] forth with Elon on X talking about it. I
[01:01:20] I do think Elon and SpaceX AI are going
[01:01:23] to come back to the frontier at some
[01:01:25] point. It's not obvious to me that the
[01:01:28] acquisition basically the the deacto
[01:01:31] acquisition of Curser is going to be
[01:01:33] that moment. Uh or if that'll put them
[01:01:36] maybe just behind the the frontier in
[01:01:39] what effectively has become a second
[01:01:41] tier of the frontier. Call it the Google
[01:01:43] tier uh until Google hopefully leaprogs.
[01:01:46] It would certainly be great to get more
[01:01:48] competition among frontier models. But I
[01:01:51] if if I'm Elon, I'm I'm thinking the
[01:01:54] best shot that SpaceX AI has to get back
[01:01:58] to being the frontier is by being a
[01:02:01] hyperscaler first. That ultimately
[01:02:03] algorithmic
[01:02:05] wars, which is what we're seeing right
[01:02:07] now, will burn themselves out. We'll
[01:02:08] discover through recursive
[01:02:10] self-improvement over the next few years
[01:02:12] what a perfect AI model looks like. And
[01:02:15] then the war, the competition moves to
[01:02:18] who is the biggest, baddest hyperscaler,
[01:02:21] who has the most hardware and the most
[01:02:23] compute.
[01:02:24] >> And Alex, don't forget the anthropic
[01:02:26] deal can be cancelled with, you know,
[01:02:28] some like a quarter.
[01:02:29] >> It's month by month. I think it's month
[01:02:31] by month.
[01:02:31] >> And then this is only a three-year
[01:02:33] contract. So when when Elon Elon never
[01:02:37] uh never plays second fiddle, right? He
[01:02:39] will be working on
[01:02:42] >> He wrote back to us, he'll never give
[01:02:44] up. never underline. But what I would
[01:02:47] say is that the question is how do you
[01:02:49] get back to the frontier? And I I
[01:02:51] suspect the thinking may be become a
[01:02:53] hyperscaler for the moment. Put Grock on
[01:02:56] de facto life support for a year or two.
[01:02:59] Use revenue from the Google deal and the
[01:03:01] anthropic deal to accumulate enough of a
[01:03:04] head of steam on the SpaceX side that
[01:03:07] you're able to go and build the Dyson
[01:03:08] swarm in sun-synchronous orbit. And then
[01:03:11] once you have 10x 100x more compute than
[01:03:14] everyone else, then you go and try to
[01:03:17] reach a frontier status by training on
[01:03:19] the
[01:03:19] >> world and and buy every model that you
[01:03:22] want to. He's going to have a currency
[01:03:23] that he can go on shopping sprees with.
[01:03:27] >> Well, he just did that with cursor, but
[01:03:29] that may or may not be enough to get him
[01:03:31] to the frontier. It may require
[01:03:32] >> one of many.
[01:03:33] >> There may be some things that money
[01:03:35] can't buy, but that compute can. and his
[01:03:37] thinking maybe speculatively he can buy
[01:03:40] enough compute to buy his way back to
[01:03:41] the frontier.
[01:03:43] >> You know if Google needs this much
[01:03:44] external compute like at this scale that
[01:03:48] tells you the bottleneck has completely
[01:03:49] shifted from model design to
[01:03:51] infrastructure. I'm finding this hard to
[01:03:53] comprehend that Google needs this level
[01:03:56] of compute. That's a crazy crazy number
[01:03:58] they're paying for this.
[01:04:00] >> I mean I thought they had their own
[01:04:01] infrastructure. Is that is the compute
[01:04:03] demand so crazy?
[01:04:06] >> Yeah. No, it's everything is sold out,
[01:04:08] Seem. I mean, they'll take everything.
[01:04:09] >> Yeah,
[01:04:11] >> it's completely So, so just numerically,
[01:04:14] this is a this is 110,000 GPUs here. I
[01:04:17] thought he bought half a million. No, he
[01:04:19] did the other deal with Anthropic
[01:04:20] already.
[01:04:21] >> Um, so does that mean he's still got a
[01:04:23] couple hundred thousand GPUs for
[01:04:25] internal XAI training or not?
[01:04:27] >> I I think these are different
[01:04:29] generations. So, the Colossus one was
[01:04:32] sort of an unholy ad mixture of several
[01:04:34] different I think it it was a mixture of
[01:04:37] H100's and several other generations. Uh
[01:04:40] so uh Colossus 2 was supposed to be a
[01:04:44] pure mixture. Remember also Google most
[01:04:46] of their compute takes the TPU format.
[01:04:48] You have customers that don't
[01:04:50] necessarily want to run their workloads
[01:04:52] on TPUs. They may want the Nvidia CUDA
[01:04:55] stack. And so there are variety of
[01:04:57] reasons why if if you're Google Cloud
[01:04:59] Platform, you want to be able to to rope
[01:05:02] in additional GPUs if you can find them.
[01:05:05] All right, I'm going to move us along
[01:05:06] unless someone else has another point
[01:05:08] you want to make here. Uh, this next
[01:05:10] story is sort of science fiction
[01:05:11] becoming real. I love the scope and
[01:05:14] audacity of uh of Elon's plans. This is
[01:05:17] their AI1 satellite they just unveiled.
[01:05:20] The next generation AI satellite
[01:05:22] designed, Alex, to launch humanity's
[01:05:24] first Dyson Swarm here. The spec the
[01:05:26] specs are wild. 150 kW of peak compute,
[01:05:30] 70 kW per ton. This is a two-tonon
[01:05:32] satellite, a 70 m wingspan, basically
[01:05:36] the wingspan of a an Airbus 380 with 110
[01:05:40] square meters of deployable radiative
[01:05:42] cooling and an integrative
[01:05:44] micrometeorite shield. Right, there's
[01:05:46] lots of micromedorites hitting the
[01:05:48] planet all the time. Uh Elon's
[01:05:50] description is worth me reading. He
[01:05:52] says, quote, "The AI satellite is much
[01:05:54] simpler than a Starling satellite. The
[01:05:56] AI satellite is essentially a lot of
[01:05:58] solar cells. You still need some laser
[01:06:00] links, but you don't have all the super
[01:06:02] complex antennas that you have on a
[01:06:04] Starlink satellite. It's easier to
[01:06:06] design than uh than the AI satellite.
[01:06:09] It's a lot bigger. A lot of this
[01:06:11] technology already exists with Starlink
[01:06:14] V3. So, here he is. Uh he's making it
[01:06:17] real. That's what he does all the time.
[01:06:19] Build the tech to make it real. Alex,
[01:06:21] what do you make of it?
[01:06:22] >> A few thoughts. First, it's beautiful.
[01:06:25] >> Yeah, it is beautiful. Second, it's so
[01:06:27] beautiful. In in in my daily newsletter,
[01:06:29] I featured an artistic rendering of this
[01:06:32] design. It's beautiful. Second thought,
[01:06:35] look at how much of its surface area has
[01:06:37] nothing to do at all with compute. Look
[01:06:39] at the solar arrays for power. Look at
[01:06:41] the radiators for heat dissipation. Look
[01:06:44] how tiny, at least by surface area. All
[01:06:46] of this is for compute. What that says
[01:06:49] to me is, think a few generations out.
[01:06:51] Imagine now we're in the early 2030s and
[01:06:56] maybe compact fusion is finally working.
[01:06:59] Wouldn't that radically change the form
[01:07:01] factor?
[01:07:02] >> Power beaming.
[01:07:04] >> Power beaming. Sure. But even with power
[01:07:06] beaming, you still need to uh you still
[01:07:08] need to radiate the the power from uh
[01:07:11] from heat dissipation uh from the the
[01:07:13] GPUs and you still need to receive the
[01:07:15] power. uh yes, it'll enable them to be
[01:07:17] more focused, but you're still you're
[01:07:20] still depending on some sort of
[01:07:21] radiative transfer in both directions.
[01:07:23] But imagine like a few generations out,
[01:07:26] you have some some radically innovative
[01:07:29] radiator system to get rid of your waste
[01:07:31] heat. And maybe you're able to locally
[01:07:33] power yourself without needing the sun.
[01:07:36] These can be incredibly compact. This is
[01:07:38] the most disjointed, the largest,
[01:07:41] bulkiest, most mainframe era node in the
[01:07:45] Dyson swarm we're ever likely to see.
[01:07:47] Imagine now in the future something
[01:07:49] that's far more compact, something that
[01:07:52] is looks maybe a little bit more
[01:07:54] spherical and a little bit less like a
[01:07:56] leaf blowing in the solar wind. This is
[01:07:59] the first generation of I think the node
[01:08:01] in our Dyson swarm Elon does over and
[01:08:04] over again, right? his Merlin engines,
[01:08:06] you know, V1, V2, V3, they get smaller,
[01:08:09] more compact, get it working and then
[01:08:11] simplify.
[01:08:12] >> Yes,
[01:08:14] >> Brian. Uh, any appreciation that you
[01:08:17] want to share on this one?
[01:08:20] >> I mean, Elon's the best in the world at
[01:08:21] hardware. No doubt about it. I, you were
[01:08:23] talking about micromedorites. I was just
[01:08:25] imagining they must have some redundancy
[01:08:26] here. Like, if if there's some bullet,
[01:08:28] you know, some bullet fragments can fly
[01:08:30] through this thing either on the solar
[01:08:31] panel or on the liquid radiators. Maybe
[01:08:34] there's different compartments like, you
[01:08:35] know, it could it could sustain uh five
[01:08:38] bullet hits into the radiators and still
[01:08:40] get sufficient cooling or something like
[01:08:42] that. And then the other thing I'm just
[01:08:43] wondering is how does it all fold up
[01:08:45] inside Starship? But I'm sure I'm sure
[01:08:46] it'll be a beautiful origami. And by the
[01:08:48] way, the point you just made about
[01:08:50] Starship, the reason, you know, SpaceX
[01:08:52] can do this is because of Starship and
[01:08:55] its volume. you know, no one else I
[01:08:58] don't think any other vehicle is going
[01:08:59] to have the level of capacity to build
[01:09:02] or deploy satellites like this. So,
[01:09:06] >> but also keep in mind again so 150 kow
[01:09:11] we were speaking previously uh with
[01:09:14] Andrew and uh talking about cerebrus.
[01:09:16] Cerebrus like a wafer scale engine is
[01:09:19] what like 10 20 30 kow 20 kow
[01:09:23] >> 20 kow. So 70 kow is a ton for this
[01:09:28] >> versus like 20 kow for a single wafer
[01:09:32] that's what fraction of a kilogram. So
[01:09:34] imagine how far we are from the physical
[01:09:38] physically efficient frontier here. When
[01:09:40] we talk all the time on on the pod royal
[01:09:43] we about the Dyson swarm. Some of us
[01:09:45] talk about disassembling the moon or or
[01:09:47] other planets and what what what a waste
[01:09:49] by atom count most of our solar system
[01:09:51] is. Drink. Yeah, drink.
[01:09:54] >> What a waste by mass most of our solar
[01:09:56] system is. If it takes an entire ton to
[01:09:59] generate 150 kilowatts of compute in low
[01:10:03] Earth orbit or sun-synchronous orbit,
[01:10:05] we're orders of magnitude, but that says
[01:10:07] to me, we're orders of magnitude away
[01:10:08] from efficiently turning most of the
[01:10:11] matter of our solar system, most of it
[01:10:13] by mass anyway, into compute. So, plenty
[01:10:15] of scaling to go as well.
[01:10:17] >> Dave,
[01:10:17] >> well, Peter, you're the you're the
[01:10:18] master of launching. Uh if this is about
[01:10:21] a $6 million GPU compute thing, you
[01:10:24] know, 150 kilowatts is about translates
[01:10:26] to about 6 million of GPU. But what is
[01:10:29] the launch cost of this one unit? So
[01:10:31] this would be equivalent to like an
[01:10:32] NV72, you know, if it were in Nvidia
[01:10:35] terms. That's up there in space.
[01:10:37] >> You want to get down you want to get it
[01:10:39] down to 100 dollars per kilogram, right?
[01:10:42] That's the target price uh at the end of
[01:10:45] the day. Uh and remember his original
[01:10:47] filing was for 500,000 of these
[01:10:50] satellites. Uh you know the calculation
[01:10:52] is like a launch per hour of Starship 24
[01:10:55] hours a day, 7 days a week to deploy
[01:10:57] that and then he upped it to a million
[01:10:58] satellites. Uh
[01:11:01] I I I think again the audacity of his
[01:11:04] scale is extraordinary but he backs into
[01:11:07] the numbers as a first principal thinker
[01:11:09] and he makes it work.
[01:11:10] >> Yeah. A couple of thoughts here. you
[01:11:11] know the AI infrastructure now stops
[01:11:15] being a real estate problem and starts
[01:11:17] becoming a launch problem right and this
[01:11:19] is one of the moments where we're seeing
[01:11:20] the category change in real time where
[01:11:23] it's not a rocket company it's not a
[01:11:24] satellite internet company it's like
[01:11:26] literally civilizational infrastructure
[01:11:29] company I such an unbelievable demand
[01:11:32] for power that we have to do this and
[01:11:34] let's remember that this was you know
[01:11:36] data orbital compute was not in
[01:11:39] anybody's bingo card a year ago and here
[01:11:40] we designed it and like planning on
[01:11:43] getting it out there. Incredible.
[01:11:44] >> And and even more, every large AI
[01:11:46] company is talking about putting up
[01:11:48] their orbital data centers. Uh and now
[01:11:50] that new Glenn is not functioning,
[01:11:52] right? It's going to be down for at
[01:11:54] least a year. Um maybe more. Uh SpaceX
[01:11:57] is the only story in town. We'll see if
[01:11:59] Relativity Space makes it happen or
[01:12:02] Rocket Rocket Labs gets their larger
[01:12:04] vehicle operating. Uh this next story,
[01:12:06] you know, Elon loves building the
[01:12:07] machines that build the machines. And
[01:12:09] and so this week, SpaceX announced a
[01:12:12] Gigasat factory. Love the term giga. Uh
[01:12:15] once again, of course, in Texas, and
[01:12:17] they're going to be producing the A1 the
[01:12:19] AI1 satellites uh in late 2027. The
[01:12:22] scale is enormous, 1,000 acres capacity
[01:12:25] for 11 million square ft of facilities.
[01:12:28] And Dave, you remember when we were with
[01:12:29] Elon um at the Gigafactory and this is
[01:12:32] the Gigafactory playbook once again, he
[01:12:35] was building out again 11 million square
[01:12:37] ft for Optimus production. Uh so the
[01:12:41] important thing here is it's fully
[01:12:43] integrated. You know, they're going to
[01:12:44] produce the solar ingots, the wafers,
[01:12:46] the solar cells, and the entire AI uh
[01:12:50] one satellite all on one campus. Uh and
[01:12:53] vertical integration matters. You know,
[01:12:54] if you control the entire supply chain
[01:12:56] from raw materials to finished products,
[01:12:58] you control your cost and your timeline.
[01:13:00] And that's probably the number one thing
[01:13:03] that Elon does extremely well is
[01:13:06] manufacturing.
[01:13:07] >> Uh Dave,
[01:13:08] >> is and and look at look at Texas just
[01:13:11] running away with every one of these
[01:13:12] projects. It's crazy.
[01:13:14] >> Yeah. You know, and then in all the
[01:13:16] posts online, if you look on X,
[01:13:17] everyone's like, "Oh my gosh, he's
[01:13:18] taking over all of Texas." Look, a
[01:13:20] thousand acres is about two golf
[01:13:22] courses. So next time you're flying,
[01:13:24] look down. It's it's just two golf
[01:13:26] courses, guys. He's not taking over all
[01:13:29] the land in Texas. But the the impact of
[01:13:31] this on the economy in that state is
[01:13:33] unbelievably healthy. Unbelievably good.
[01:13:36] So I don't know why more states aren't
[01:13:37] competing for this.
[01:13:38] >> I think there's an important point that
[01:13:40] we're sleeping on, which is vertical
[01:13:42] integration. Yeah, sure. He's he's doing
[01:13:44] that, but it's not necessarily just for
[01:13:46] his own sake or because he wants to be
[01:13:49] the the master of the supply chain for
[01:13:51] this. It's because if you follow
[01:13:53] carefully SpaceX's announcements in
[01:13:55] connection with its IPO, they've been
[01:13:58] releasing demo videos, concept videos of
[01:14:00] doing this on the moon. And if you're
[01:14:03] not in a position to vertically
[01:14:05] integrate production of these AI
[01:14:08] orbiting data centers, then you're not
[01:14:10] in a position to start manufacturing
[01:14:12] them off the Earth, namely on the lunar
[01:14:15] surface. So he and SpaceX released this
[01:14:18] video of an electromagnetic slingshot
[01:14:21] launching bam bam bam AI orbital data
[01:14:25] centers with a rail gun from the surface
[01:14:28] of the moon presumably manufactured on
[01:14:30] the surface of the moon.
[01:14:31] >> Yes. So I I think that that's the
[01:14:33] essential thing that that we arguably
[01:14:36] should be talking about here. If you
[01:14:37] have vertical integration of the solar
[01:14:40] and the compute, then you can do that on
[01:14:42] the moon and then uh your your your
[01:14:46] delta V is far more favorable and you
[01:14:48] can just start slinging these things out
[01:14:50] and building the actual Dyson swarm, not
[01:14:52] from the Earth's surface where the
[01:14:53] economics for heavy launch are less
[01:14:56] favorable. Yeah. I'll never forget I was
[01:14:57] with Elon um at SpaceX headquarters and
[01:15:01] we were having a conversation about, you
[01:15:04] know, the fact he was bemoning the fact
[01:15:05] he has to make everything, right? We
[01:15:07] we're hearing that same thing from Brett
[01:15:09] Adcock at Figure and a lot of other
[01:15:11] high-tech companies and there was a
[01:15:14] product called Pika, which is the heat
[01:15:16] shielding uh for the Dragon capsule and
[01:15:19] he was getting screwed by the
[01:15:20] manufacturer and so he just turned to
[01:15:23] the manufacturer and said, "Forget it.
[01:15:24] I'm going to make it myself and put you
[01:15:26] out of business. And just just the
[01:15:28] ferocious mindset he has is
[01:15:31] extraordinary. Um
[01:15:32] >> well, do you have a prediction, Peter?
[01:15:34] It'd be great to get everyone's
[01:15:35] prediction, but this partnership right
[01:15:37] now between Daario and Elon is insanely
[01:15:39] powerful because because you know, Alex
[01:15:41] is pointing out continually that new
[01:15:42] physics is going to be invented by AI
[01:15:44] imminently. And that new physics, if it
[01:15:46] goes right into these gigafs, uh it
[01:15:49] could be just mind-blowing what comes
[01:15:51] out the other side. And and also if
[01:15:53] Daario wins the race to
[01:15:54] self-improvement, then he's going to
[01:15:56] have the smartest AI, you know, and
[01:15:58] rapidly accelerating, but Elon will have
[01:16:01] the space-based data centers and the
[01:16:03] manufacturing capability to turn that
[01:16:05] into a closed loop, you know, self
[01:16:07] self-manufacturing closed loop, which
[01:16:09] Daario hasn't even started on. You know,
[01:16:11] there's no there's no physical stuff
[01:16:12] going on at Anthropic at all yet. Uh,
[01:16:15] and you know, you know, where the rubber
[01:16:16] really hits the road is the AI that
[01:16:18] designs chips. So here's Elon building
[01:16:21] the Terraab, but the chip design itself
[01:16:24] comes from the smartest AI, which is,
[01:16:26] you know, probably going to be Daario.
[01:16:28] May maybe, you know, OpenAI has a
[01:16:29] chance, too. But that that duopoly
[01:16:32] becomes incredibly powerful if they
[01:16:35] stick together. But do you think they'll
[01:16:37] still be friends five years?
[01:16:38] >> Yeah, these guys, I mean, all of the
[01:16:40] players here have formed partnerships
[01:16:42] and broken partnerships and reshuffled
[01:16:45] the deck, you know, probably multiple
[01:16:47] times over the last few years. So what's
[01:16:48] your question? Is is the partnership
[01:16:50] going to stick together?
[01:16:51] >> Well, five years from today, are Daario
[01:16:54] and Elon um friends?
[01:16:56] >> Yeah. Like like shaking hands and
[01:16:58] >> No, I haven't seen honestly I love I
[01:17:01] love Elon, but I haven't seen him
[01:17:03] partner uh for the long term ever. Uh
[01:17:08] every every partnership comes together,
[01:17:11] he takes the lead and runs with it. I'll
[01:17:15] register a prediction which is I I think
[01:17:16] there are going to be such crazy things
[01:17:18] happening in the next few years that we
[01:17:20] we'll look back and laugh at ourselves
[01:17:22] for even asking the question of whether
[01:17:24] Elon will be shaking hands with
[01:17:25] >> Dar answer to the question
[01:17:28] >> interesting
[01:17:29] >> questioning the question itself
[01:17:32] thoughts
[01:17:35] >> uh I totally agree with you Peter I
[01:17:37] think this is like uh just I'm I'm just
[01:17:40] boggled by the level of of somebody
[01:17:43] thinking Looking at civilization scale
[01:17:45] on a non-stop basis, and I love it.
[01:17:47] >> This episode is brought to you by
[01:17:49] Blitzy, autonomous software development
[01:17:51] with infinite code context. Blitzy uses
[01:17:55] thousands of specialized AI agents that
[01:17:57] think for hours to understand
[01:18:00] enterprisecale code bases with millions
[01:18:02] of lines of code. Engineers start every
[01:18:05] development sprint with the Blitzy
[01:18:07] platform, bringing in their development
[01:18:09] requirements. The Blitzy platform
[01:18:11] provides a plan, then generates and
[01:18:14] pre-ompiles code for each task. Blitzy
[01:18:16] delivers 80% or more of the development
[01:18:19] work autonomously while providing a
[01:18:21] guide for the final 20% of human
[01:18:24] development work required to complete
[01:18:26] the sprint. Enterprises are achieving a
[01:18:29] 5x engineering velocity increase when
[01:18:31] incorporating Blitzy as their preIDE
[01:18:34] development tool, pairing it with their
[01:18:36] coding co-pilot of choice to bring an AI
[01:18:39] native SDLC into their org. Ready to 5x
[01:18:42] your engineering velocity? Visit
[01:18:44] blitzy.com to schedule a demo and start
[01:18:47] building with Blitzy today.
[01:18:51] >> All right. Uh, amazing. All right, let's
[01:18:53] move to our next story here. Let's put
[01:18:55] some financial figures on this. Um
[01:18:58] SpaceX valuation is uh right now in poly
[01:19:02] market expected to hit 2.13 trillion at
[01:19:05] the close of the first day of market. Uh
[01:19:08] and here's the news. Morgan Stanley
[01:19:10] projects SpaceX revenue could grow from
[01:19:12] get this 18.7 billion in 2025. That's
[01:19:16] the last number we have on the books to
[01:19:18] 3.4 trillion by 2040. Uh at those
[01:19:22] revenues at 3.4 4 trillion by 2040. That
[01:19:25] puts SpaceX's value at some place
[01:19:28] between 50 trillion to 100 trillion.
[01:19:30] Right? We're seeing the birth
[01:19:32] potentially the first hundred trillion
[01:19:33] dollar company. Uh and and these charts
[01:19:37] down below tell the story, right? So
[01:19:39] here's the revenue numbers today. You
[01:19:41] know, uh Starlink is 11 billion of
[01:19:44] revenue. Launch revenue is a measly 4
[01:19:47] billion. Uh AI revenue from XAI at 3.2
[01:19:52] billion. And here comes Google at 11
[01:19:54] billion and then SpaceX uh you know
[01:19:58] total of $18 billion. So uh crazy
[01:20:02] numbers. Any thoughts on this? I mean I
[01:20:05] don't think people again no no company
[01:20:07] in history has ever hit a trillion
[01:20:09] dollars of revenue. So just to put this
[01:20:12] in context, you're talking about the
[01:20:13] really really big guys like Walmart and
[01:20:15] Amazon. You know they're getting close
[01:20:16] but they're not at a trillion yet. Uh so
[01:20:19] this four trillion.
[01:20:21] >> Yeah. On the on the other hand, a
[01:20:22] trillion is what, like 3% of US GDP. So,
[01:20:26] and that's today's GDP. If if we undergo
[01:20:29] his predicted uh 3xing year-over-year or
[01:20:33] 10xing year-over-year of GDP, that still
[01:20:35] makes this a relatively small drop in
[01:20:37] the bucket. I I think it not investment
[01:20:39] advice obviously but I I think it is
[01:20:41] very much a conceivable scenario and one
[01:20:44] in which maybe the Dyson swarm gets
[01:20:47] built and most of or a large chunk of
[01:20:49] civilization runs on the compute that
[01:20:52] the Dyson swarm provides. Yeah, sure. A
[01:20:54] few trillion dollars makes total sense.
[01:20:56] >> Seem, you want to close us out?
[01:20:58] >> I got nothing.
[01:21:00] >> Okay. All right. Fair.
[01:21:02] >> I I have a question for Brian though,
[01:21:04] just on that narrow point. So let's say
[01:21:06] we do build the Dyson swarm. Uh let's
[01:21:08] say uh most of the compute in our solar
[01:21:11] system is no longer earth-based and
[01:21:14] presumably most of our economy consists
[01:21:16] of AI agents trading. What do you think
[01:21:19] will be the role if any of
[01:21:21] cryptocurrency in that future?
[01:21:25] >> Well, I think Bitcoin will be the new
[01:21:27] gold standard and then the payments will
[01:21:29] be happening on chain. So um capital
[01:21:32] formation will probably happen on chain.
[01:21:34] borrowing and lending, you know,
[01:21:37] that's that's the financial system that
[01:21:39] the AI agents would end up using. So
[01:21:41] that's the most likely outcome in my
[01:21:43] point of view.
[01:21:44] >> So So you think basically uh there's
[01:21:47] almost no impact on margin regarding
[01:21:49] whether our compute is terrestrial or
[01:21:52] orbital on how the crypto rails or or
[01:21:55] broader financial rails of our
[01:21:57] civilization operate. no latency
[01:21:59] considerations, no trust based
[01:22:01] considerations, no decentralization
[01:22:03] considerations like doesn't make a
[01:22:04] difference.
[01:22:06] >> Well, I think a lot of those things will
[01:22:08] get better on a crypto-based system like
[01:22:10] the decentralization characteristics,
[01:22:11] this um soundness of money, right? Um if
[01:22:15] you're talking about syncing up a
[01:22:17] blockchain between multiple planets like
[01:22:19] Mars and Earth, there would be some
[01:22:21] delay in that. But I don't see any
[01:22:22] reason why, you know, if you need to
[01:22:24] transfer money between Mars and Earth,
[01:22:26] um it couldn't happen on chain and have
[01:22:28] a laser link between them or something
[01:22:29] like that. And you know, um so they're
[01:22:32] not going to be real-time payments
[01:22:34] between those planets, but there could
[01:22:35] be uh appropriately delayed payments. Uh
[01:22:39] they won't need to settle as often. So
[01:22:40] >> are these conversations you're having in
[01:22:42] your executive committee every day?
[01:22:45] >> My my conversations are sometimes a lot
[01:22:47] more mundane about like how do I get
[01:22:49] these two people to work together, you
[01:22:50] know? But um we still got to solve all
[01:22:52] the human problems uh before we start
[01:22:54] building interplanetary financial
[01:22:56] systems.
[01:22:56] >> By the way, I have to point out, you
[01:22:58] know, this projection by Morgan Stanley
[01:23:00] is definitely a fundraising tool ahead
[01:23:03] of the IPO. You know, Morgan Stanley is
[01:23:05] one of the banks that's going to profit
[01:23:06] from this deal. Uh there's, you know,
[01:23:09] I'm reading every day on X that players
[01:23:12] are opting out of SpaceX. It's
[01:23:13] overvalued. They're valuing it at 700
[01:23:15] million, not, you know, 1.7 trillion.
[01:23:18] So, we're going to see. I you know I I
[01:23:21] fundamentally believe in the long-term
[01:23:22] value of SpaceX. Uh I just hope retail
[01:23:25] investors don't get hit uh in the in the
[01:23:29] process. So while we have Brian uh Brian
[01:23:31] Longevity and Biotech, one of my
[01:23:33] favorite stories. So first off,
[01:23:35] congratulations on your recent raise at
[01:23:38] New Limit, $435
[01:23:41] million to push us towards uh age
[01:23:44] reversal. if you don't mind, uh, tell
[01:23:46] us, you know, origin story of New Limit
[01:23:49] and, uh, why you and Blake started it
[01:23:51] and what are you guys doing there?
[01:23:54] >> Yeah. Well, actually, Jacob Kimmel and
[01:23:55] Blake and I all started it together as
[01:23:57] co-founders and Jacob's the CEO now, so
[01:23:59] he's really he's brilliant.
[01:24:02] >> He is. Um, yeah. Yeah. So, the origin
[01:24:05] story goes actually back to the IPO of
[01:24:06] Coinbase, believe it or not, and 2021.
[01:24:09] Having gotten some liquidity from
[01:24:11] Coinbase and thinking about the next 10
[01:24:13] years and what were going to be the big
[01:24:15] technology trends that could drive
[01:24:16] civilizational progress, I kind of felt
[01:24:18] like a lot of the big ones had good
[01:24:20] teams working on it, right? AI and
[01:24:22] fusion energy and brain machine
[01:24:23] interfaces and all the rest space. Um,
[01:24:26] one I one that I didn't see great teams
[01:24:28] working on as much and that maybe seemed
[01:24:30] underfunded was longevity. And so I I
[01:24:33] reached out to a few folks, started
[01:24:34] hosting some dinners um with just top
[01:24:37] scientists and biotech CEOs and kind of
[01:24:38] went around the table and said, "All
[01:24:39] right, what's on the horizon that's most
[01:24:41] exciting to you, but it's underfunded?"
[01:24:43] And one of the folks mentioned
[01:24:45] epigenetic reprogramming, which is this
[01:24:48] area that talks about how you can
[01:24:50] reprogram cells. Um Shina Yamanaka
[01:24:53] famously won the Nobel Prize for this in
[01:24:55] 2012, showing you could reprogram an old
[01:24:57] skin cell into a young uh embryionic
[01:25:00] stem cell. And he by doing so with just
[01:25:02] these four proteins added to the cell he
[01:25:04] had this kind of remarkable discovery
[01:25:06] that he changed both the age of the cell
[01:25:07] and the type of the cell. So at new
[01:25:09] limit we're trying to do half of what
[01:25:11] Shiny Yamanaka did. We don't want to
[01:25:13] change the type of the cell we just want
[01:25:14] to change the age. And what we mean by
[01:25:16] that is really just restoring the
[01:25:18] function that the cell had when it was
[01:25:19] younger. And um we built a uh high
[01:25:23] throughput screening system. It starts
[01:25:25] with AI to talk explore the you know 10
[01:25:29] quadrillion different possibilities and
[01:25:31] combinations of different proteins you
[01:25:33] could use to reprogram a cell and it
[01:25:35] starts to it it originally it ingested
[01:25:37] all the existing literature. Now it's
[01:25:39] been ingesting the the wet lab data
[01:25:40] that's been coming out of New Limit
[01:25:42] which we have the largest data set out
[01:25:44] there by far. And so it recommends the
[01:25:46] next set of experiments that we run in
[01:25:48] our wet lab. We do these poolled screens
[01:25:49] and see if phenotypically we can we can
[01:25:52] make cells look younger. Then we take
[01:25:54] the best hits out of that in the funnel
[01:25:55] and we put them through functional
[01:25:57] assays which are a little slower and
[01:25:58] more expensive to see if they actually
[01:26:00] act younger. Um, for instance, we might
[01:26:03] take a reprogrammed liver cell and see
[01:26:05] if it can process caffeine and
[01:26:07] acetaminophen and alcohol or whatever
[01:26:09] kind of like a young a young liver cell.
[01:26:11] And then the the best functional assay
[01:26:14] candidates we then are starting to run
[01:26:15] through um non-human primate and and
[01:26:18] human trials. So yeah, the process has
[01:26:21] gone faster than I would have expected.
[01:26:23] I thought this was going to be like a 5
[01:26:24] or 10 year kind of basic research
[01:26:26] endeavor, but um we've actually been
[01:26:29] able to demonstrate successful
[01:26:30] reprogramming of human cells now in our
[01:26:32] first drug candidates going into the
[01:26:33] clinic next year, hopefully followed by
[01:26:34] a bunch more.
[01:26:35] >> Amazing. Human trial next year and you
[01:26:37] and Blake put in the first 100 million.
[01:26:39] I remember that. I remember coming to
[01:26:41] your lab and and seeing it. Uh full
[01:26:43] disclosure, I'm also a shareholder, an
[01:26:46] investor in UI. Very proud of that. So,
[01:26:49] congrats on the progress. And uh other
[01:26:51] news related in epigenetic reprogramming
[01:26:54] uh Life Bioscience is another one of my
[01:26:55] portfolio companies announced they've uh
[01:26:58] they've dosed their first patient with
[01:27:00] their OSK treatment today. So yes, we're
[01:27:05] on the verge of uh longevity escape
[01:27:07] velocity. Brian, what do you what do you
[01:27:10] hold as a target for, you know, when do
[01:27:13] we hit LE? Do you have a number in your
[01:27:16] head? Do you have a a goal?
[01:27:19] You know, man, we need Kurszswall to
[01:27:22] come plot another curve for us, don't
[01:27:23] we?
[01:27:24] >> He's got his prediction. His prediction
[01:27:26] is 2033.
[01:27:27] >> Huh.
[01:27:29] >> So,
[01:27:29] >> I'd have to go look at his data on that.
[01:27:30] I actually I haven't looked at the exact
[01:27:32] year that it would be projected to
[01:27:33] happen. I'm just been again trying to
[01:27:35] solve more mundane problems like how do
[01:27:37] we hire how do we hire the next uh you
[01:27:39] know bioinformaticist or whatever. But
[01:27:41] um yeah, I hope he's right. I mean, he
[01:27:44] has been right on so many things. It
[01:27:46] wouldn't surprise me if he's right
[01:27:46] again. I have an idiosyncratic position.
[01:27:49] Uh perhaps unsurprisingly, I I will go
[01:27:52] out on a limb and say that I think LEV
[01:27:57] is going to be spiky in the same sense
[01:27:59] that AGI is spiky and that on certain
[01:28:02] spikes it seems possible it might be
[01:28:05] achieved later this year. And the reason
[01:28:09] why I say that is and I I wrote about
[01:28:11] this in my newsletter and I know Peter
[01:28:13] you immediately said, "Oh, but but but
[01:28:16] uh Uh in the past few weeks there were
[01:28:19] uh there was a really interesting paper
[01:28:22] published on the first uh placebo uh
[01:28:26] rather double blinded study on HIV
[01:28:29] patients receiving GLP1s looking at
[01:28:32] epigenetic clocks and and I know all the
[01:28:34] caveats. Epigenetic clocks orth style
[01:28:38] clocks aren't uh as as good assays at
[01:28:41] determining biological age as a variety
[01:28:43] of functional measures. I know all of
[01:28:45] that.
[01:28:45] >> Okay. Nonetheless, with all of those
[01:28:47] caveats out the way, out of the way, for
[01:28:50] the first time, to my knowledge, in the
[01:28:52] literature, I'm starting to see evidence
[01:28:54] of some sort of measurable at least
[01:28:57] double-digit age reversal as a result of
[01:29:00] GLP1s. And that makes me ask whether as
[01:29:04] we're we're now with Redat True Tide and
[01:29:06] and other uh Chinese have their own I
[01:29:10] think second and third generation GLP1s
[01:29:13] whether we're going to start to see
[01:29:16] hidden under our noses in the same sense
[01:29:17] that we flew past the touring test
[01:29:20] without remarking that AGI basically
[01:29:22] passed the touring test. Did we with a
[01:29:24] whimper or not with a bang whether this
[01:29:27] year maybe next year but as soon as this
[01:29:29] year we're going to in a spiky way in a
[01:29:32] subpopul
[01:29:34] fly bath fly past lev without broader
[01:29:38] notice that it just happened.
[01:29:40] >> Well, we're going to have some level of
[01:29:42] proof, right? We have this $101 million
[01:29:44] healthpan prize going on where we're
[01:29:46] measuring functional reversal of age.
[01:29:49] Right? Again, the biomarkers right now,
[01:29:52] you know, you can test your biomarkers
[01:29:54] on a whole clock and many different
[01:29:55] clocks. The problem is if you test
[01:29:57] yourself in the morning and test
[01:29:59] yourself at night, you'll get different
[01:30:00] answers. And so, I don't think there's a
[01:30:03] reliable biioarker for aging. Brian, I
[01:30:05] don't know if you agree with that or
[01:30:06] not, but what I care about is less a
[01:30:09] number on a page, but more functional.
[01:30:12] Do you have the ability to process um
[01:30:16] you know liver function that you did
[01:30:18] earlier in life? Do you have the
[01:30:20] cognition, the immune, the muscle that
[01:30:21] you had 20 years younger? So I think
[01:30:24] we're at the beginning of that curve and
[01:30:25] just like in the singularity we're we're
[01:30:28] you know living through it. I think
[01:30:30] we're in the process and I agree GLP-1
[01:30:32] drugs are probably one of the very first
[01:30:34] longevity drugs out there. Brian, any
[01:30:36] thoughts or maybe a question to Peter
[01:30:38] and Brian on this? You think we're going
[01:30:40] to be arguing n years from now? Let's
[01:30:42] say it's not this year. Let's say it's
[01:30:44] the early 2030s. We're going to be
[01:30:46] arguing over whether we've passed LEV or
[01:30:49] not. And we're going to bring folks onto
[01:30:51] the pod years from now saying, "Oh, no,
[01:30:53] it definitely hasn't been hit because my
[01:30:55] favorite vanity benchmark hasn't been
[01:30:57] passed yet." While others will say, "No,
[01:30:59] it was actually passed 5 years ago."
[01:31:02] >> Brian, I like your theory. I like your
[01:31:04] theory on that, Alex. I I would agree
[01:31:06] with the touring test comparison where
[01:31:08] we probably will go past it and people
[01:31:10] won't nobody will react. Um I I think
[01:31:14] that's probably likely. I don't think
[01:31:15] it'll happen this year or next year, but
[01:31:18] um 2033 could could potentially happen.
[01:31:21] I mean, it's like the FDA approval part
[01:31:23] is the is the slow part, right? The AI
[01:31:25] part is going really fast.
[01:31:26] >> I will fly anywhere for the treatment.
[01:31:29] >> I want to make a couple of com I want to
[01:31:31] make a couple of comments here. Um, you
[01:31:33] know, if you look, if we talk about LEV,
[01:31:36] we won't know for a long time because
[01:31:38] people won't be dying for quite a long
[01:31:40] time. We won't really know when it's hit
[01:31:42] except for medical certain medical
[01:31:43] tests, etc. So, I think this is going to
[01:31:45] drag out by definition. My favorite
[01:31:48] comment from the life extension world is
[01:31:51] that the baby that's going to live to a
[01:31:52] thousand years old is already alive and
[01:31:54] that just blows your mind.
[01:31:55] >> Yeah. And and I agree with you, right?
[01:31:57] The the whole basis of the la of our
[01:31:59] health span prize was not to be a
[01:32:01] longevity prize where you have to wait
[01:32:03] 30 years to award a winner. It's an age
[01:32:06] reversal, right? Do you have the
[01:32:08] function that you had earlier? You know,
[01:32:10] one of my favorite things right now that
[01:32:12] we haven't reported on and I'm working
[01:32:15] to get Martin Rothblat on the pod. Um
[01:32:17] Martine's the CEO of United
[01:32:19] Therapeutics. Uh is work she has been
[01:32:22] working on uh restoring your thymus. So
[01:32:26] going from your own uh IPSC stem cells,
[01:32:29] your induced pleot potent stem cells and
[01:32:31] regrowing a thymus and of course you
[01:32:33] lose your thymus in your 20s completely
[01:32:37] and it's one of the most important
[01:32:38] organs for regrowing. Have you been
[01:32:39] thinking about uh about thymus uh
[01:32:42] capabilities at new limit?
[01:32:45] >> It's not one of the ones we've targeted
[01:32:47] yet, but uh yeah, I think it's it's on
[01:32:50] the list somewhere. Um we're going to
[01:32:51] have to get to all the major tissues at
[01:32:53] some point. I
[01:32:53] >> I was arguing with Jacob. I want muscle
[01:32:55] rejuvenation, please. Yeah.
[01:32:57] >> Yeah. I mean, that would pair nicely
[01:32:59] with the GOP ones, right?
[01:33:00] >> For sure.
[01:33:01] >> I want to say something I've been
[01:33:02] wanting to say for months and months,
[01:33:04] and I can only say on this episode,
[01:33:07] >> which is, you know, as we talk about
[01:33:08] life extension, there's been two things
[01:33:10] true in the world for the history of
[01:33:12] humanity, death and taxes. And we may,
[01:33:16] this will crack death and Bitcoin is
[01:33:18] going to crack taxes. So, there you go.
[01:33:22] >> I love this tweet. I saw
[01:33:23] >> I love it. I love it when a when a
[01:33:24] brilliant successful entrepreneur from
[01:33:26] outside of biotech gets into biotech
[01:33:28] because you get that fresh view kind of
[01:33:30] you know what is possible you know
[01:33:33] you're not jaded you're not calloused
[01:33:34] you're just like what is possible really
[01:33:36] curious to know what you buy for 435
[01:33:38] million and does it surprise you like
[01:33:41] where that money goes
[01:33:43] >> yeah well on the on the idea of going
[01:33:44] into biotech I mean there's definitely
[01:33:48] some Elon inspiration there which is
[01:33:49] like we're we're living through a golden
[01:33:50] age of software where fortunes are being
[01:33:52] made And so I think the these hard tech
[01:33:54] problems deserve more capital and you
[01:33:58] know it's almost to like SpaceX probably
[01:34:00] would not have existed unless some a
[01:34:02] wealthy individual like Elon who had had
[01:34:04] an exit had at least some of his own
[01:34:05] capital to go try and do it. Same thing
[01:34:07] with Tesla, right? So venture capital
[01:34:10] sometimes doesn't get you all the way
[01:34:11] there. It's almost like to do these real
[01:34:13] moonshots, you have to like find someone
[01:34:15] who made a billion dollars in software
[01:34:17] and then and then go found step two go
[01:34:19] found the moonshots. But I I hope that
[01:34:21] that's not the case forever. I think a
[01:34:22] lot of the money that's been made in
[01:34:24] crypto and AI will actually go into the
[01:34:25] next moonshots and and all of that
[01:34:28] stuff. Um but sorry your question was
[01:34:30] about um which part again? Sorry.
[01:34:32] >> Well, 435 million you know that automate
[01:34:36] a bunch of wet lab tests or does that
[01:34:37] give you like concurrent many many lines
[01:34:40] of research or what does it
[01:34:41] >> run a human trial?
[01:34:43] >> It's trials a lot of trials. I mean in a
[01:34:47] software company most of your costs are
[01:34:49] people and then some AWS and stuff like
[01:34:51] that. Um in a biotech company typically
[01:34:54] it's like it's people and then materials
[01:34:56] like re like reagents and things. Um but
[01:34:59] in this case now that we're getting
[01:35:01] ready to go into the clinic yeah it's
[01:35:03] you know your average phase one trial
[01:35:05] might be I don't know 10 or 20 million
[01:35:06] and you can kind of go up from there on
[01:35:08] phase two and phase three. So you need
[01:35:09] to get enough shots on goal to get some
[01:35:11] of these candidates to come out the
[01:35:13] other side and have a big impact. So,
[01:35:14] we've got enough capital now to run a
[01:35:16] number of trials, which is good.
[01:35:18] >> Alex, I love this tweet I saw uh from
[01:35:22] Sam Suare starting to mourn all the
[01:35:24] people who died before the singularity.
[01:35:27] >> Yeah,
[01:35:27] >> I'm mourning all the the people who've
[01:35:29] died period. And I I think you know when
[01:35:32] to Brian's point about all of the grand
[01:35:35] challenges that there aren't yet
[01:35:37] billionaires solving, I think I'll throw
[01:35:39] this out to the the universe, to the
[01:35:42] economy. I would love to back or
[01:35:44] otherwise support someone building a
[01:35:47] company to use advanced compute to
[01:35:50] digitally resurrect everyone who's ever
[01:35:52] died. I think it's a tragedy that we've
[01:35:53] had so many billions of deaths over the
[01:35:55] millennia. Let's fix it.
[01:35:57] >> I love that. I love that. All right. Our
[01:35:59] next story here is a fun one. Our last
[01:36:02] three stories in the field. Uh Columbia
[01:36:04] University researchers successfully
[01:36:06] edited a PCSK9 gene. This is, you know,
[01:36:08] your LDL, your bad cholesterol and an
[01:36:11] HBG gene for hemoglobin in embryos in
[01:36:15] vitro. So, uh, you know, this is opening
[01:36:19] up the conversation. I remember when I
[01:36:20] was, uh, uh, at the Whitehead Institute
[01:36:23] doing my medical degree, uh, and the
[01:36:27] first gene editing capabilities of
[01:36:30] restriction enzymes came out. a lot of
[01:36:33] hand ringing around uh embryo editing,
[01:36:36] zygote editing. You know, 2018, a
[01:36:39] scientist in China uh Hi Janu uh
[01:36:42] successfully edited the CCR5 gene in two
[01:36:46] young girls that were brought to term
[01:36:48] healthy uh to make HIV resistance. So
[01:36:51] the question is, are we going to start,
[01:36:53] you know, when this gets reliable
[01:36:55] enough, are we going to start editing
[01:36:57] our kids? Um, you know, my son Dax just
[01:37:00] did a project on this at school and his
[01:37:03] point was we give our kids the best
[01:37:05] food, the best education, the best
[01:37:07] friends, the best clothing. Why not
[01:37:09] start with the best jeans? So, I am
[01:37:12] curious uh what your thoughts are guys.
[01:37:15] >> Yes, of course. Like this is this is
[01:37:18] much more advanced than even Gatka. The
[01:37:20] whole plot of Gatka was it's it's still
[01:37:23] your genes, your parents' genes, just
[01:37:25] the best combination with extreme in
[01:37:27] vitro selection of embryos. This goes
[01:37:30] beyond this to to do base editing of
[01:37:33] embryos. And you know, base editing
[01:37:36] itself is is looking tremendously
[01:37:39] promising, invented by David Louu and a
[01:37:41] collaborator at Harvard. Better than
[01:37:43] first generation crisper. you're able to
[01:37:46] to do uh single nucleotide swaps uh in
[01:37:50] in DNA with minimal error rates. This is
[01:37:54] tremendous and I I can't for the life of
[01:37:56] me understand or otherwise why this
[01:38:00] wouldn't become a broadspread practice.
[01:38:03] Yeah, obviously uh you know, shadows of
[01:38:06] eugenics come to mind, but at the end of
[01:38:09] the day, if you're getting rid of
[01:38:10] hereditary disease or if you've got
[01:38:13] some, you know, if you're a family below
[01:38:17] 5 foot and you want your kids to be
[01:38:18] taller, uh you know, the ethics of this
[01:38:22] are going to evolve and morality around
[01:38:23] this is going to evolve and societal
[01:38:25] norms are going to evolve. Seem, where
[01:38:26] do you come out on this?
[01:38:28] >> Well, two things. One is this is going
[01:38:30] to have a a lot of people freak out. But
[01:38:32] let's note that we have an old word for
[01:38:33] all this is called breeding. For
[01:38:35] thousands of years, we've been crossing
[01:38:37] dogs and cats and horses and humans to
[01:38:38] select the traits that we want. What's
[01:38:41] happening now is biology is becoming a
[01:38:43] digital programmable domain. Disease
[01:38:45] prevention is the most compelling case
[01:38:47] here. But the governance things that
[01:38:49] this will bring up are going to be
[01:38:50] unbelievable because we're going to be
[01:38:52] editing diseases like right before birth
[01:38:55] at this point. I said this is incredible
[01:38:58] between uh sequencing crisper IBF AI
[01:39:00] modeling like anything is now possible
[01:39:03] just think of yourself your 50 trillion
[01:39:05] cells for every human being in the world
[01:39:07] is now a software engineering problem.
[01:39:09] How do we think about that?
[01:39:11] >> Brian.
[01:39:11] >> Yeah. Well, I think one of the one of
[01:39:13] the bigger
[01:39:14] >> please
[01:39:14] >> problems with this particular one of the
[01:39:16] bigger complexities is jurisdiction. So
[01:39:18] that if you're if you live in a country,
[01:39:20] so suppose your parents are in the US
[01:39:22] and you go to a a Caribbean island and
[01:39:24] that Caribbean island is say, "Yeah, do
[01:39:26] whatever you want." Um, so then you've
[01:39:28] got a baby, you then you bring that baby
[01:39:30] back in the US. That's inevitably going
[01:39:32] to happen. So regardless of what your
[01:39:34] local laws decide they would like it to
[01:39:36] be, it's not going to actually work
[01:39:38] locally. So that that's a really big
[01:39:41] complexity in this whole this whole
[01:39:43] area.
[01:39:44] >> Brian, you want to weigh in?
[01:39:46] >> Yeah, it's a great point. I mean, I I
[01:39:48] think this is going to happen, too. I
[01:39:49] think it'll be good for humanity um on
[01:39:51] net. I think a lot of people will freak
[01:39:53] out about it for a while, but it's I
[01:39:54] think there was a Pew Research study I
[01:39:57] saw that something like 80% of Americans
[01:40:00] would support um embryo editing for
[01:40:03] disease prevention. So, that was more
[01:40:05] than I expected. I thought 80% is pretty
[01:40:07] high, but you know, only about 20% I
[01:40:09] think supported it for quote
[01:40:11] enhancement. Um the part that I've never
[01:40:14] been able to figure out is what's the
[01:40:16] line between disease prevention and
[01:40:17] enhancement, right? Um not having a
[01:40:19] disease sounds like a pretty good
[01:40:20] enhancement to me. I mean there's
[01:40:22] there's cases where like let's say
[01:40:24] osteoporosis, right? So if you have
[01:40:26] that, it's a disease. But if you had a
[01:40:29] gen a gene edited to prevent it, you
[01:40:31] have stronger bones. Stronger bones
[01:40:33] sounds like an enhancement. I I
[01:40:34] basically think the line is so blurry
[01:40:36] that
[01:40:36] >> people will start with this for disease
[01:40:39] prevention and then it'll uh upgrade
[01:40:40] over time. And I think anything's, you
[01:40:43] know, on the table, even IQ or whatever.
[01:40:44] It's like, should we have more smart
[01:40:46] people in the world? Probably. I mean,
[01:40:47] that's seems like a good thing to me.
[01:40:49] >> And ethics change, right? Because I
[01:40:50] mean, if you remember a while ago, IVF
[01:40:52] was considered immoral in the early
[01:40:54] days.
[01:40:55] >> And now it's it's enabled millions of
[01:40:58] families to be formed.
[01:41:00] >> Well, but not to state the obvious, but
[01:41:02] uh you know, if you're 7 foot tall or
[01:41:05] higher, you have a 50% chance of making
[01:41:07] the NBA. And if you make the NBA, that's
[01:41:09] a $20 million a year paycheck. So if
[01:41:11] you're a couple, you know, just a
[01:41:13] regular everyday couple in the US and
[01:41:15] you have the option to have a 7 foot
[01:41:16] long tall child by design, you're going
[01:41:19] to go to that Caribbean island and make
[01:41:21] your seven foot tall for economic
[01:41:23] reasons. How, you know, think of what
[01:41:25] could go wrong then.
[01:41:27] >> So there there's going to be some
[01:41:29] serious guardrails. There has to be
[01:41:31] guardrails around this.
[01:41:32] >> You know, there's a company I love uh
[01:41:34] called Nucleus. Uh Ken Sedagi, the CEO,
[01:41:38] is going to be at my my abundance
[01:41:40] longevity trip in October. Uh Brian, I'm
[01:41:43] I'm hoping to have uh someone from
[01:41:46] Uliimmit there as well. But what nucleus
[01:41:48] does is you fertilize, you know, 20 uh
[01:41:52] 20 eggs, you know, 20 zygot and then you
[01:41:57] sequence them and you can choose the one
[01:41:59] out of the 20 that have the attributes
[01:42:01] instead of playing dice and wait, you
[01:42:04] know, and hoping for the lucky sperm. Uh
[01:42:06] you can actually pick the one that has,
[01:42:07] you know, the right attributes from
[01:42:09] within the gene population that you
[01:42:10] have. Separate from base editing, a step
[01:42:12] before that, but still something you can
[01:42:14] do today. Remember the line from Gatka,
[01:42:16] "Your child is still you, simply the
[01:42:19] best of you,
[01:42:20] >> except that's not even true with base
[01:42:22] editing. It's better than the best of
[01:42:24] the parents."
[01:42:25] >> Well, Brian, listen, thank you for
[01:42:26] joining us today. Uh, grateful. I know
[01:42:29] you've got, you know, the world's
[01:42:30] largest, uh, crypto company in the world
[01:42:32] to to run. So, uh, appreciate your time,
[01:42:36] my friend.
[01:42:37] >> Yeah, thank you for having me. This was
[01:42:38] a great conversation. I love talking
[01:42:40] with people who want to build a future.
[01:42:42] >> Yeah, for sure. And that is most
[01:42:44] definitely this this amazing group. All
[01:42:46] right. Uh our next story here is
[01:42:47] Anthropic launches Fable 5 and Mythos 5.
[01:42:50] Uh here we go. Uh Andre Kaparthy uh who
[01:42:54] has joined uh our friends at Anthropic
[01:42:57] is now shilling for the newest models.
[01:43:00] Here's his quote. Super exciting
[01:43:01] release. Fable 5 is the same underlying
[01:43:04] model as Mythos, but with added
[01:43:06] safeguards. You can 10x your test suite,
[01:43:09] autooptimize code, run giant research
[01:43:12] projects with custom HTML for the
[01:43:14] results, anything. Alex, over to you.
[01:43:18] >> Very impressive release. Anthropic is
[01:43:21] back in the lead now. Uh, until today,
[01:43:24] GPT 5.5 was the lead, the
[01:43:27] state-of-the-art across most of these
[01:43:30] benchmarks. today for probably about 5
[01:43:33] minutes. Anthropic takes back the crown
[01:43:35] across most measures of super
[01:43:38] intelligence. A few notes. The
[01:43:40] distinction between Fable 5 versus
[01:43:42] Mythos 5. Mythos is a uh as reported a
[01:43:46] less inhibited version of Fable. Or
[01:43:50] conversely, Fable is a more inhibited
[01:43:52] version of mythos that has certain
[01:43:55] scaffolding and other restrictions that
[01:43:57] prevent it from having more ambitious
[01:43:59] conversations that might relate to
[01:44:01] biology, chemistry, cyber security,
[01:44:04] perhaps other areas. But I've used it.
[01:44:06] It's incredible. I gave it my typical
[01:44:10] benchmark, which is favorite eval,
[01:44:12] asking it to oneshot a cyberpunk
[01:44:16] FT first person shooter that's visually
[01:44:18] stunning. Did an amazing job, no errors,
[01:44:21] ran immediately, had a nice soundtrack
[01:44:23] that came with it. What's more
[01:44:25] interesting if you look at the
[01:44:26] benchmarks and then also the
[01:44:28] demonstrations far be it for me to
[01:44:30] suggest that Pokemon games are actually
[01:44:32] the secret to super intelligence but if
[01:44:35] you look at the demos that have been
[01:44:37] coming out from anthropic and otherwise
[01:44:39] it is able to play just from pure visual
[01:44:42] reasoning now this is Fable 5 or mythos
[01:44:44] 5 doesn't really seem to matter it's
[01:44:46] able to based on just watching the
[01:44:49] screens of games like various Pokémon
[01:44:51] games it's able to win them and that
[01:44:54] requires long-term long range reasoning
[01:44:57] capabilities that we really haven't seen
[01:44:59] until now. So my guess is Anthropic has
[01:45:03] probably been very aggressively doing
[01:45:05] RLVR reinforcement learning with
[01:45:07] verifiable rewards on long range
[01:45:10] reasoning challenges. maybe game
[01:45:12] playing, maybe challenges that involve
[01:45:14] spatial reasoning, maybe very large code
[01:45:16] bases, maybe very large capture the flag
[01:45:19] CTF hunts, very large sprawling both in
[01:45:24] probably, my guess would be in space and
[01:45:26] in time RLVR challenges to get some of
[01:45:29] these results that it's just incredible
[01:45:32] what you can oneshot now with Fable 5 uh
[01:45:36] with reasoning set to high. And again,
[01:45:39] difficult to predict how long Anthropic
[01:45:41] will retain the crown before, say, GPT
[01:45:44] 5.6 comes out, but it is exciting to see
[01:45:47] the Frontier move once again.
[01:45:49] >> Dave, you've been playing?
[01:45:51] >> Yeah. Well, it's only been out for about
[01:45:52] 4 hours, but I've been playing for 4
[01:45:54] hours, so uh it's I mean, what we're
[01:45:57] living through, Alex is constantly
[01:45:59] posting on our internal feed. It's
[01:46:00] happening. It's happening. I mean, it's
[01:46:02] really happening now. This is Oh my god.
[01:46:05] You know what's interesting to me? a
[01:46:07] bunch of things. First of all, they
[01:46:08] doubled the price, so it's expensive as
[01:46:10] all hell. Uh this this idea of commodity
[01:46:13] intelligence is not happening. It's it's
[01:46:15] damn expensive if you want the most
[01:46:17] brilliant thing. But what it produces is
[01:46:21] so incredibly intelligent that it takes
[01:46:24] you quite a while to read and understand
[01:46:26] what it did. And it's like, well, I'll
[01:46:28] just move on while you're reading and
[01:46:30] trying to catch up.
[01:46:32] >> And so that's an interesting new thing
[01:46:33] for me. Um, and also I think this is
[01:46:36] interesting that Anthropic had it in the
[01:46:38] bag. You know, Alex just mentioned that,
[01:46:41] you know, the leap frog is going on, but
[01:46:43] actually OpenAI pushed out their best
[01:46:46] and Anthropic then said, "Okay, well, we
[01:46:48] had this ready to go." So, you know,
[01:46:50] we're just trying to make it safer, but
[01:46:52] now we'll come out and go over the top.
[01:46:53] >> And we've been saying they're they're
[01:46:54] pulling their punches. They're they are
[01:46:57] they're they're gaming this.
[01:46:58] >> Uh, as you're heading towards your
[01:47:01] flight, do you want to add anything
[01:47:02] here?
[01:47:03] Uh, no. What occurred to me here is
[01:47:06] that, you know, when you have you have
[01:47:08] the um split now between the safety
[01:47:12] guard rails and the performance guard
[01:47:15] rails. And I thought that was really
[01:47:16] interesting
[01:47:19] because you've got the same underlying
[01:47:21] model with added safeguards, right?
[01:47:23] That's the whole safety debate in one
[01:47:24] sentence right there.
[01:47:26] >> Well, actually, they even changed the
[01:47:27] name, too. You know, you've got the the
[01:47:29] Mythos, which we've been talking about
[01:47:30] for a while, and then Fable is the
[01:47:32] heavily guardrailed, won't do a cyber
[01:47:34] threat, won't do a nuclear weapon, you
[01:47:37] know, version of the exact same model.
[01:47:39] Uh so they even even differentiated it
[01:47:41] by name to support exactly what you're
[01:47:43] saying. lame
[01:47:44] >> with with pretty broad buffers like you
[01:47:46] can friends of mine have tried having
[01:47:48] conversations perfectly innocent
[01:47:50] conversations with fable five high about
[01:47:53] biology and from what I've heard it has
[01:47:57] pretty broad margins for what it will
[01:47:59] simply revert back to opus for if if it
[01:48:02] thinks you're asking anything in biology
[01:48:05] or chemistry that could remotely be a
[01:48:08] dangerous subject it'll just downgrade
[01:48:10] you immediately to opus
[01:48:11] >> you know funny you say that Alex cuz uh
[01:48:13] you know I'm often trying to work from a
[01:48:14] plane and SSH doesn't work on the plane
[01:48:18] Wi-Fi and and so I, you know, use a
[01:48:20] workaround to try and get access to all
[01:48:21] my remote agents. And that also vomited
[01:48:23] saying looks a little bit like maybe
[01:48:25] you're doing something sketchy and it
[01:48:28] gave me the uh the anthropic, you know,
[01:48:30] we're we're not going to answer that
[01:48:32] query thing, you know, repeatedly. So, I
[01:48:34] had to actually find a workaround for
[01:48:36] that, too. So yeah, they've really
[01:48:38] really locked it down to try and avoid
[01:48:40] the
[01:48:40] >> I predict as we head towards IPO season
[01:48:43] for these two companies, these Frontier
[01:48:45] Labs, we're going to start to see the
[01:48:46] rate of model release increase. They're
[01:48:48] going to be leaprogging who's going to
[01:48:50] be out front at the day of the IPO.
[01:48:52] Who's getting the news?
[01:48:53] >> At least at least, Peter, we have
[01:48:55] competition. The situation could be
[01:48:56] worse. We could just have a singleton, a
[01:48:59] single lab that is that is not getting
[01:49:00] leaprogged at all. At least we have two
[01:49:02] competitors at the moment in the lead.
[01:49:04] >> Yeah, for sure. All right, our final
[01:49:07] story for today. Uh, after more than a
[01:49:09] decade of sucking, Apple finally made
[01:49:12] its move to improve Siri. Uh, they've
[01:49:15] announced a multi-year partnership with
[01:49:16] Google to power Siri with Gemini AI,
[01:49:19] rebuilding Siri from the ground up. Uh,
[01:49:22] the big shift, Siri is now an agent, not
[01:49:25] just a voice interface. The key
[01:49:27] breakthrough is personal context. Yes,
[01:49:29] finally. Siri can now reason across your
[01:49:32] messages, your emails, your notes, your
[01:49:34] photos, and actually get things done.
[01:49:36] And it can actually, you know, when I'm
[01:49:38] sending a text message to Kristen, it
[01:49:42] can spell it correctly or it can spell
[01:49:44] my name, you know, correctly. It doesn't
[01:49:47] most of the time. Uh they're calling it
[01:49:49] persistent AI workspace coming to beta
[01:49:52] later this year. Uh focused in English.
[01:49:54] Uh two huge implications here. First,
[01:49:57] Apple is essentially admitting it lost
[01:49:59] the Frontier model or the foundation
[01:50:00] model race and chose Google to rent
[01:50:03] brains rather than build its own. And
[01:50:05] and it's a stunning, you know,
[01:50:06] concession from the company that has
[01:50:09] always priding itself on its own stack.
[01:50:12] Second, the real emote was never the
[01:50:14] model. It's the personal context. So,
[01:50:18] uh, Alex, you wanna you want to share?
[01:50:20] >> I'll different perspectives here. So one
[01:50:23] the glass half empty perspective is
[01:50:26] Apple for arguably the first time in
[01:50:29] modern history has has outsourced an a
[01:50:32] key element of its tech stack which
[01:50:35] Apple it it's just not as with Elon and
[01:50:38] vertical integration. It's not in
[01:50:40] Apple's cultural DNA to to want to rent
[01:50:44] out someone else's technology that's
[01:50:46] going to be so core to the system.
[01:50:48] That's the glass half empty. second
[01:50:50] perspective and that perspective says
[01:50:53] the foundation model is not analogous to
[01:50:56] say an operating system. It's more
[01:50:58] analogous to the role of a search engine
[01:51:00] which has to be localized. So you'll
[01:51:01] note Apple didn't announce this in China
[01:51:04] yet. It didn't announce it in the EU yet
[01:51:06] for different reasons. But focusing just
[01:51:08] on China, I would expect Apple to strike
[01:51:11] some partnership that's roughly
[01:51:12] analogous to the partnership with Google
[01:51:15] with one of the three big Chinese AI
[01:51:18] labs or tech companies. And that will
[01:51:21] represent a localization roughly
[01:51:23] analogous to the way Apple localizes its
[01:51:26] other web search capabilities in China
[01:51:29] with one of the Chinese companies
[01:51:31] analogously to the way it default uh to
[01:51:34] Google for web search. to the glass half
[01:51:36] full argument is well foundation models
[01:51:40] however foundational they are are
[01:51:42] intrinsically localizable and so Apple
[01:51:45] has merely chosen for its American
[01:51:48] franchise Google and maybe it'll choose
[01:51:51] another franchisee for Europe and
[01:51:53] another one for China
[01:51:57] Google's been doing this with Android
[01:51:59] and its suite of products for a while
[01:52:01] now right so this is Apple sort of
[01:52:03] catching up I mean I've almost you moved
[01:52:05] to an Android phone so many times just
[01:52:08] because of its ability to, you know,
[01:52:10] contextualize all of my my personal
[01:52:12] context.
[01:52:14] >> A third perspective. Yes. And a third
[01:52:17] perspective is if you buy that
[01:52:19] foundation model economics are
[01:52:21] hyperdelationary and are the cost is
[01:52:23] going down by 40x year-over-year or
[01:52:26] anything remotely like that, then you
[01:52:28] should view this entire bit of Apple
[01:52:30] losing its tech sovereignty as a nothing
[01:52:33] burger. And yes, Apple maybe is is
[01:52:35] struggling to distill Gemini down to its
[01:52:39] private cloud compute combined with
[01:52:41] iPhones this year, but in a year or two,
[01:52:44] there will have been so much
[01:52:45] hyperdelation in the cost of compute
[01:52:47] that for the present capabilities of new
[01:52:50] Siri that they will easily fit into the
[01:52:53] new iPhones. and that any loss of
[01:52:55] sovereignty or any use, we haven't even
[01:52:58] touched on this semi- embarrassingly.
[01:53:01] It's not even the case that Apple is
[01:53:03] just able to distill Gemini, Google's
[01:53:05] Gemini or its version thereof into the
[01:53:08] iPhones. And it's not even the case that
[01:53:10] they're able to operate it with a
[01:53:11] combination of Apple's own cloud
[01:53:14] combined with the the iPhone edge
[01:53:16] compute. They actually need Google's
[01:53:17] cloud to do this because it's so comput
[01:53:19] intensive. But all of that over the next
[01:53:21] year or two can shrink and hyperdelate
[01:53:23] down to to just working on an edge
[01:53:25] device as these algorithms get more
[01:53:27] advanced.
[01:53:28] >> Yeah. Dave, any thoughts on Apple?
[01:53:30] >> Yeah, lots. Well, anytime in tech that
[01:53:32] you don't do anything for a long period
[01:53:34] of time, you eventually get crushed. The
[01:53:36] question is how does it emerge? The way
[01:53:39] it would play out if Apple doesn't do
[01:53:40] something aggressive uh is you lose the
[01:53:43] interface. You know, Siri is now Google,
[01:53:46] but Google is Android and Android is the
[01:53:49] one competitor to the iPhone. Remember,
[01:53:51] all the revenue still comes from the
[01:53:52] iPhone. So, you lose the interface. At
[01:53:55] the same time, TSMC
[01:53:58] moves the chip manufacturing away from
[01:54:00] the M5 and M3 and M4 to a much higher
[01:54:04] paying customer in AI data center use.
[01:54:06] So Nvidia or Elon take the manufacturing
[01:54:10] capacity away for the underlying chip
[01:54:12] because they just pay more to TSMC. So
[01:54:15] then you've lost your hardware
[01:54:17] manufacturing under the covers and
[01:54:19] you've lost the interface on the other
[01:54:21] side and then everybody just moves to
[01:54:22] Android because they're talking to
[01:54:23] Google or Gemini all day long anyway.
[01:54:26] And so also the machines get the the
[01:54:29] actual physical phones get manufactured
[01:54:31] by Samsung using Samsung's uh chip
[01:54:33] manufacturing capability which is you
[01:54:35] know immune to TSMC.
[01:54:37] >> Selma I want to acknowledge your
[01:54:38] persistence and your ability to stay in
[01:54:41] the conversation.
[01:54:42] >> This is a new low. I'm in my airplane
[01:54:43] seat. So let's just let it just go past
[01:54:47] us.
[01:54:47] >> Okay. Closing closing thoughts on Apple
[01:54:49] and uh and Siri.
[01:54:53] >> Your closing thought that it's good.
[01:54:55] I'll believe that it's good when I see
[01:54:56] it.
[01:54:57] >> Yeah, we've been so disappointed over
[01:54:59] and over again. I you know, you and I
[01:55:01] met the founders of Siri before they
[01:55:02] sold to Apple. Uh and it was it prom it
[01:55:06] had such promise back then and it just
[01:55:08] uh it stuck around just way too long. Uh
[01:55:12] gentlemen, uh I think that's a wrap.
[01:55:14] Enjoyed having Brian here going from
[01:55:16] Bitcoin to longevity. Um anyway, uh
[01:55:20] crazy week ahead. I I'll see you guys in
[01:55:22] a few day for our next episode. Uh,
[01:55:24] thank you as always for your commitment
[01:55:26] to this podcast. Everybody watching,
[01:55:29] thank you for your, you know, support of
[01:55:31] Moonshots. We're here to give you the
[01:55:33] news, help you understand it, help you
[01:55:35] be optimistic. Uh, it is the most
[01:55:37] extraordinary time ever to be alive. I
[01:55:39] like to say, you know, during the
[01:55:40] singularity, don't sleep, don't blink.
[01:55:43] Uh,
[01:55:45] >> the words live long, the words live long
[01:55:47] and prosper.
[01:55:49] >> Yeah, they do.
[01:55:52] Take care, gentlemen. Safe travels
[01:55:54] everybody.
[01:55:54] >> Peace and long life.
[01:55:55] >> Take care folks.
[01:55:59] >> If you made it to the end of this
[01:56:00] episode, which you obviously did, I
[01:56:03] consider you a moonshot mate. Every
[01:56:04] week, my moonshot mates and I spend a
[01:56:06] lot of energy and time to really deliver
[01:56:08] you the news that matters. If you're a
[01:56:10] subscriber, thank you. If you're not a
[01:56:12] subscriber yet, please consider
[01:56:14] subscribing so you get the news as it
[01:56:16] comes out. I also want to invite you to
[01:56:18] join me on my weekly newsletter called
[01:56:20] Metatrens. I have a research team. You
[01:56:22] may not know this, but we spend the
[01:56:24] entire week looking at the meta trends
[01:56:26] that are impacting your family, your
[01:56:28] company, your industry, your nation. And
[01:56:31] I put this into a two-minute read every
[01:56:33] week. If you'd like to get access to the
[01:56:35] Metatrends newsletter every week, go to
[01:56:37] diamandis.com/tatrends.
[01:56:39] That's diamandis.com/metatrends.
[01:56:42] Thank you again for joining us today.
[01:56:44] It's a blast for us to put this together
[01:56:46] every week.