ATRIUMsearch → argument graph
Audio · 2026-05-22 · 1h 42m · 18 moments

SpaceX's $2T Case, Nvidia's Shock Selloff, America Turns on AI, Trump Pulls AI Order, Bond Crisis?

(0:00) Gavin Baker joins the show! (0:30) Andrej Karpathy joins Anthropic; hypergrowth and profitability (12:42) Why Americans have turned on AI, anti-human perception (27:22) Trump pulls AI EO, US-China AI relationship, dystopian AI layoffs (45:19) SpaceX S-1 tear down! Breaking down the three major businesses and the case for a $2T valuation (1:11:22) Nvidia smashes earnings but stock falls, why people are shorting chips (1:22:25) Market update: Flashing red signals, oil, inflation ✦ AI generated

timeline · colored by role

01
Prediction

Karpathy joining Anthropic to lead recursive self-improvement could unlock an order-of-magnitude improvement in AI capability on a yearly basis — a new form of Moore's law where model quality goes parabolic.

Chamath argues that Karpathy's recursive self-learning approach, combined with Anthropic's existing lead, could produce a 10x yearly improvement in model quality, creating a new Moore's law trajectory.

transcript

Chamath: I think this idea of recursive self-learning puts these models on a combination of overdrive and autopilot. And so if you put those two things together, I think that you start to, you could potentially live out this idea that there's an order of magnitude improvement on a yearly basis. So like this new form of Moore's law. So then the model quality just goes absolutely parabolically, just like this, straight up.

extends · 1

02
Prediction

Recursive self-improvement and continual learning, combined with the work of singular talents like Andrej Karpathy, could pull the future of AI forward in a very real way, potentially making annual order-of-magnitude improvements seem conservative.

Gavin Baker argues that Anthropic's hiring of Andrej Karpathy for recursive self-improvement is a major deal, and that if combined with continual learning, performance gains could be far faster than linear projections.

transcript

Gavin Baker: I do think what Karpathy is working on, recursive self-improvement is really important, and unlocking that in continual learning, maybe the two final frontiers for AI. And just the idea of recursive self-improvement, that the model, while it is training, during a forward pass, has input into its training, or another model has input into the training. I think that could be really powerful. And I think Chamath's statistics of 10Xing every year might seem conservative if that comes to pass. And then, of course, continual learning is the holy grail, where the model learns from experiences the way humans do. And that's something we haven't unlocked yet. And those two combined, I think, would They might pull the future forward in a very real way.

extends · 1gives example · 1supports · 1

03
Data

LLM companies (OpenAI, Anthropic, Google, xAI, plus Cursor and open-source) are on track for $200-400 billion in aggregate ARR by end of this year at high margins, proving strong ROI on AI infrastructure — currently being achieved on the strictest possible definition excluding the most profitable use cases.

Gavin Baker presents data that the private LLM industry (including Google) is heading toward $200-400B ARR with ~80% inference margins, demonstrating real returns even excluding GPU-driven improvements in ads and recommendations at Facebook, Amazon, and Google.

transcript

Gavin Baker: the success is extraordinary. It's undeniable. I think the fact that they are now, they were EBIT positive per the Wall Street Journal in the most recent quarter is a really important fact for kind of the whole AI narrative. Because now there's you could talk about circular funding, you could talk about ROI, and we could go look at the ROIC of the hyperscalers. But if OpenAI and Anthropic are at, call it $100 billion of ARR now with 80 percent-ish gross margins on inference, like the returns are there. And then if we add in, and they're growing really fast, if we add in Gemini, we add in Cursor, we add in XAI, we add in open source, you know, it's not hard to see 200, 300, $400 billion of ARR at the end of this year at high margins.

supports · 1

04
Prediction

Recursive self-improvement, where an AI model has input into its own training, could put model improvement on a parabolic trajectory resembling a new form of Moore's law.

Chamath argues that Karpathy's work on recursive self-learning, combined with massive compute, could enable an order-of-magnitude improvement in model quality every year, creating a new Moore's law for AI.

transcript

Chamath: I think this idea of recursive self-learning puts these models on a combination of overdrive and autopilot. And so if you put those two things together, I think that you start to, You could potentially live out this idea that there's an order of magnitude improvement on a yearly basis. So like this new form of Moore's law. So then the model quality just goes absolutely parabolically, just like this, straight up. I think a bunch of compute at the problem and these things learn really quick, I think is the I ordered a bit there.

extends · 1gives example · 1supports · 1

05
Claim

LLM companies including OpenAI, Anthropic, and Google are on track for $200-400 billion of aggregate ARR by end of this year at high margins, proving strong ROI on AI infrastructure.

Gavin Baker breaks down the economics of the AI industry, estimating that LLM companies will generate $200-400 billion in ARR this year with ~80% gross margins on inference, making the returns on GPU spend undeniable.

transcript

Gavin Baker: The success is extraordinary. It's undeniable. I think the fact that they are now, they were EBIT positive per the Wall Street Journal in the most recent quarter is a really important fact for kind of the whole AI narrative. Because now there's you could talk about circular funding, you could talk about ROI, and we could go look at the ROIC of the hyperscalers. But if OpenAI and Anthropic are at, call it $100 billion of ARR now with 80 percent-ish gross margins on inference, like the returns are there. And then if we add in, and they're growing really fast, if we add in Gemini, we add in Cursor, we add in XAI, we add in open source, you know, it's not hard to see 200, 300, $400 billion of ARR at the end of this year at high margins.

supports · 1

06
Mechanism

The anti-AI sentiment in America is driven by three layers: a power-imbalance narrative where AI creates outsized returns for the few, foreign state actors funding anti-technology sentiment as a competitive strategy, and a deep psychological anti-humanist framing that offends human ego — rooted in the same disruption as the Copernican revolution.

Friedberg lays out a three-layer explanation for why AI is hated: economic asymmetry where few benefit while many wait for diffusion; foreign state actors (tracing back to KGB Cold War tactics) fueling anti-progress sentiment; and an anti-humanist existential discomfort analogous to heliocentricity displacing humanity from the center of the universe.

transcript

Friedberg: I think that there's like an underlying view that technology creates leverage for a small group of people, which creates power imbalances. And nothing represents that more than AI, that a small number of people that control and profit from and benefit from AI are going to end up getting outsized returns relative to the broader population, that the time to diffusion of the technology, because ultimately all technologies commoditize and diffuse, but the time to diffusion here is such that it's going to be extremely asymmetric for society. ... Secondly, I think that there's a deep amount of external energy that's fueling this anti-technology sentiment in the United States and has been for decades. I think to Gavin's point, I don't think it's just China with NGOs today. I think that there is a long history of state actors intervening in media activities in foreign nations to try and create the sentiment and fuel a sentiment that reduces progress in that competitive state. I think this goes all the way back to KGB design during the Cold War and it's been refined and honed and improved over time. ... And then the third piece is like when the Copernican revolution happened, it was a mind, like heliocentricity was a totally new way of thinking for humans. ... There's something about AI that's very like not human-centric and it kind of shifts and fucks with the ego of the human. It's almost anti-humanist. And I think that that's like a deep psychological current for a lot of people and their disdain for this technology.

provides context · 1

07
Claim

Americans have turned on AI because it creates a perception of power imbalance where a small number of people capture outsized returns while the broader population sees no immediate benefit, and this is compounded by foreign state actors fueling anti-tech sentiment.

Friedberg argues that the AI backlash is driven by a fundamental human discomfort with technologies that create asymmetric power, combined with a long history of foreign state actors using media to slow technological progress in competitor nations.

transcript

Friedberg: I think that there's like an underlying view that technology creates leverage for a small group of people, which creates power imbalances. And nothing represents that more than AI, that a small number of people that control and profit from and benefit from AI are going to end up getting outsized returns relative to the broader population, that the time to diffusion of the technology, because ultimately all technologies commoditize and diffuse, but the time to diffusion here is such that it's going to be extremely asymmetric for society. ... Secondly, I think that there's a deep amount of external energy that's fueling this anti-technology sentiment in the United States and has been for decades. I think to Gavin's point, I don't think it's just China with NGOs today. I think that there is a long history of state actors intervening in media activities in foreign nations to try and create the sentiment and fuel a sentiment that reduces progress in that competitive state. I think this goes all the way back to KGB design during the Cold War

gives example · 1provides context · 1

08
Claim

The United States cannot slow down AI development because it is in an existential race with China, and the only path to peace is a mutually assured capability balance, similar to the nuclear arms race after the Manhattan Project.

Friedberg argues that the US cannot slow AI progress due to the geopolitical race with China, drawing a parallel to the Cold War nuclear arms race where balance prevented unilateral dominance.

transcript

Friedberg: I think after the Manhattan Project, the research labs were stood up to maintain our scientists that worked on the Manhattan Project from effectively leaching back or leaking back to Russia and Germany and other places that were adversary to the United States. ... When the proliferation began, there was no stopping it. You had to have this balance in the world. Otherwise, you have effectively an asymmetric power that can do whatever it wants globally. I think there's that moment in the world right now where if the United States does not advance its AI technology, the availability of it, TBD, industry, taxation, all these things that we're talking about doing, there will be someone else that will. And if someone else does, we can go through what would happen.

explains mechanism · 1supports · 1

09
Context

Slowing down AI is not possible — the US cannot afford to fall behind because the game theory of the US-China AI race mirrors the nuclear arms race, where asymmetric power is unacceptable and a balance must be reached through competition, not restraint.

Friedberg argues using the Manhattan Project and nuclear arms race as analogies that once a transformative technology exists, you cannot uninvent it; if America slows down, a competitor (China) will advance, leading to dangerous asymmetry. The goal is a mutual detente, not unilateral restraint.

transcript

Friedberg: after the Manhattan Project, the research labs were stood up to maintain our scientists that worked on the Manhattan Project from effectively leaching back or leaking back to Russia and Germany and other places that were adversary to the United States. And they all were against the nuclear bomb. They worked on it because it was necessary for the United States security. But then when Russia got a hold of the secrets, they were leaked because people were worried that if the US had all the power, there would be no counterbalance to the power. ... When the proliferation began, there was no stopping it. And you had to have this balance in the world. Otherwise, you have effectively an asymmetric power that can do whatever it wants globally. I think there's that moment in the world right now where if the United States does not advance its AI technology, the availability of it, TBD, industry, taxation, all these things that we're talking about doing, there will be someone else that will. ... As you do that analysis, you realize, wait a second, that's probably not a healthy place for the world to be. It's also probably not a healthy place for the United States to be the only one with AI. And so I think what we end up seeing is if we do try and slow down AI, we kind of lose this moment of balance that's necessary when you have a technology proliferation, like we saw with the arms race after World War II.

provides context · 1rebuts · 2supports · 4

10
Claim

Rather than assuming jobs like truck driving or warehouse sorting must be preserved, we should ask the people doing those jobs whether they actually want them — the premise that these are desirable jobs that need protection is unexamined and may be false, given Amazon's 35-40% warehouse churn.

Chamath argues that before creating policy to protect jobs from automation, we should survey the actual workers — many of whom may not want these high-churn, low-satisfaction jobs. The moral high ground of job protection may not reflect the preferences of the workers themselves.

transcript

Chamath: I think it's interesting that in all of those discussions, I've yet to see an actual survey of only the truck drivers and only the package sorters. The question that I would have is, do the people that do these jobs want these jobs? And if they do, then there's a reasonable claim to make to keep those jobs the way that they are. If you're saying, this is the job that I do, I love it, I'm able to provide for my family, great. That's a very different argument than, well, you know what, Amazon has 35 or 40% churn inside of their warehouses. And we should probably ask the question, why is that? Because if it was such a great job, I suspect the churn would be 3 or 4%. So what exactly is it that we want to protect? And have you asked them? And I think that this is just, again, a bunch of people in the peanut gallery who want to take a moral high ground and try to make some other group of people feel guilty or feel bad.

11
Mechanism

The US and China should jointly establish KYC (Know Your Customer) rules for frontier AI models, similar to biological safety testing, to prevent bad actors from creating weapons like bioweapons, while otherwise letting the race continue.

Chamath argues that the US and China should agree on a simple battery of safety tests for frontier models, akin to FDA testing, to prevent misuse for bioweapons or terrorism, while otherwise allowing competition to proceed.

transcript

Chamath: I think it's actually good that China is less than nine months behind us. I think it allows us to find a detente where we have a certain magnitude of capability that they also have, and that allows all of us to then seek peace and abundance. ... So I think what we need to do We probably need KYC. I think that should be something that us in China get together and say, you don't want it to get into the hands of people you can't control. You probably already KYC those models anyway inside of China. You already review those training runs before you allow these models to get released. We already know that that's happening. So we should probably do some sort of KYC so some crazy person doesn't create some biological weapon.

12
Claim

The US should not impose unilateral AI model testing requirements — the liability system (courts, self-regulation) already incentivizes responsible behavior, and government power granted through regulation is a one-way ratchet that never gets taken back.

Gavin Baker opposes requiring frontier AI models to pass government safety testing before release, arguing that the US already has self-regulation and a court system that holds companies accountable after harm occurs, and that regulatory power granted to government tends to grow permanently.

transcript

Gavin Baker: the one thing that's great about America is there is... We have other forms of regulation. Self-regulation, sure. We have self, one, we have self-regulation. Also, we have the courts. And if an AI model company behaves irresponsibly, they know that there are ways that people who have been harmed can seek recourse. And so we already have a system that encourages responsible behavior on the part of the model makers. ... once you give something, give a power to the government, It's almost never taken back and it begins to grow and it's kind of a one-way, a one-way path. One-way ratchet.

extends · 1rebuts · 2supports · 1

13
Mechanism

The AI PR crisis stems from a psychological anti-humanist perception — AI displaces humans from the center of the world, triggering deep discomfort similar to the Copernican revolution.

Friedberg argues that beyond economic fears and foreign influence, AI triggers a deep psychological reaction because it is inherently anti-humanist — it shifts humans from the center of the intellectual universe, much like heliocentrism did.

transcript

Friedberg: When the Copernican revolution happened, it was a mind, like heliocentricity was a totally new way of thinking for humans. And it was deeply disruptive to the church. And it was deeply disruptive to the power centers, which were the centers that could tell people Earth is at the center of the universe, we're in control, we're the direct channel to God. And the idea that the sun is at the center of the solar system and we spin around it and we're a tiny speck in the universe was very hard for people to grasp. There's something about AI that's very like not human-centric and it kind of shifts and fucks with the ego of the human. It's almost anti-humanist. And I think that that's like a deep psychological current for a lot of people and their disdain for this technology. It fuels it.

provides context · 1

14
Claim

Frontier model regulation through government executive orders is premature and dangerous because government power is a one-way ratchet that never gets taken back.

Gavin Baker argues against government-mandated testing of frontier AI models, pointing to the existing self-regulation and legal liability system in America, and warns that once government power is granted it never gets taken back.

transcript

Gavin Baker: I just, you know, to me, once you give something, give a power to the government, It's almost never taken back and it begins to grow and it's kind of a one-way, a one-way path.

supports · 1

15
Mechanism

The US cannot slow down AI development because if it does, China will advance ahead, creating a dangerous asymmetric power dynamic — just like the nuclear arms race after the Manhattan Project.

Trooper argues using the Manhattan Project analogy that AI is in a proliferation race with China where slowing down would create dangerous power asymmetry, and that a balance where both sides have comparable capability is necessary for global stability.

transcript

David Sacks: If the United States does not advance its AI technology, the availability of it, TBD, industry, taxation, all these things that we're talking about doing, there will be someone else that will. And if someone else does, We can go through what would happen. There's a complicated game theory on this, but what would happen if China had sufficiently advanced models and sufficiently advanced scaled deployment of those models relative to the United States? As you do that analysis, you realize, wait a second, that's probably not a healthy place for the world to be. It's also probably not a healthy place for the United States to be the only one with AI. And so I think what we end up seeing is if we do try and slow down AI, we kind of lose this moment of balance that's necessary when you have a technology proliferation, like we saw with the arms race after World War II.

explains mechanism · 1supports · 2

16
Prediction

Cities that ban self-driving vehicles will feel barbaric and unsafe, and the technology will inevitably prevail because autonomous driving will have near-zero death rates, making wrongful death lawsuits against human-driven alternatives inevitable.

Gavin Baker argues that cities banning self-driving cars will feel barbaric compared to places with autonomous vehicles, and that the technology will win out due to vastly superior safety, leading to wrongful death lawsuits against municipalities that block it.

transcript

Gavin Baker: I think going to a city where you can't get in a Waymo or a cyber cab is going to feel barbaric and unsafe. ... I don't know if you remember, but the early days of Uber, sometimes you go to a city where there was no Uber. Yeah, it'd be incredibly frustrating. Well, I'm not going to come back until they have Uber. It's so inconvenient. And I think, so whatever individual municipalities decide, I do think, one, Chamath's point is really powerful. There's 50,000 automotive deaths per year in the United States, if I recall correctly, and a million globally, that's not tolerable. And there will for sure be wrongful death lawsuits.

provides context · 1supports · 1

17
Claim

Crime is now a choice that municipalities can opt out of using AI-powered tools like gunshot detection, drones, and license plate cameras at minimal cost — cities that choose not to deploy them are effectively choosing to be pro-crime.

Gavin Baker argues that AI surveillance tools make crime a solvable problem at de minimis cost, and that municipalities that refuse them — like Cambridge turning off gunshot detectors — are making a deliberate choice to allow crime, citing Las Vegas's successful deployment as proof.

transcript

Gavin Baker: Crime is now a choice. You know, I think that the Cambridge state has voted to turn off gunshot detectors 2 days ago. So the geniuses coming out of Harvard in that town decided gunshot detection shouldn't occur. It's wild because, you know, there's a theory that it disadvantages, you know, that it might lead to an illegal migrant who's shooting a gun, being apprehended, and we don't want that. And A16Z had a great essay on flock. We can really, really solve crime, and it's just a choice. And different states and municipalities will make different choices to be pro-crime or anti-crime. I'm sure they don't cast it as pro-crime. There's an, you know, some sort of moral or ethical reason. They're making that choice. But people will vote with their feet over time, and then voters will vote with their votes.

18
Example

Major tech companies are now conducting layoffs that are explicitly driven by AI replacing middle-management and measurement roles, not just post-pandemic bloat reduction, which is creating a dystopian perception of the technology.

Jason argues that the public's fear of AI is justified by real-world examples where Cloudflare and Meta explicitly cited AI as the reason for laying off thousands of workers, including managers and H-1B visa holders, even while posting record profits.

transcript

Jason: I think we have to recognize that the layoffs that are occurring in big tech and in a lot of these places are not just the bloating issue anymore. ... The first one I want to give you an example of is Matthew Prince who's the CEO of Cloudflare, incredible company, public company. Two weeks ago, I laid off more than 20% of my workforce. I didn't do it because Cloudflare is struggling. We posted record revenue growth, have strong free cash flow, and are adding an unprecedented number of customers, yada, yada, yada. And he says basically, he's getting rid of measurers. Measurers are the people who manage people and who measure data. And he just says, we're getting rid of all those people. They're unnecessary because of AI, and we'll be adding people in other positions. At the same time, Zuckerberg did another round of layoffs, and they were done in a way that people felt was not considered and a bit, what's the word? Dystopian? Dystopian, thank you, sir. He did them in a pretty dystopian way.

gives example · 1

Highlight slides
Karpathy's Recursive Self-Improvement at Anthropic✦ from: Karpathy joining Anthropic to lead recursive self-improvement could unlock an order-of-magnitude improvement in AI capability on a yearly basis — a new form of Moore's law where model quality goes parabolic.A New Moore's Law for AI✦ from: Karpathy joining Anthropic to lead recursive self-improvement could unlock an order-of-magnitude improvement in AI capability on a yearly basis — a new form of Moore's law where model quality goes parabolic.Recursive Self-Improvement: A New Frontier✦ from: Recursive self-improvement and continual learning, combined with the work of singular talents like Andrej Karpathy, could pull the future of AI forward in a very real way, potentially making annual order-of-magnitude improvements seem conservative.Continual Learning: The Holy Grail✦ from: Recursive self-improvement and continual learning, combined with the work of singular talents like Andrej Karpathy, could pull the future of AI forward in a very real way, potentially making annual order-of-magnitude improvements seem conservative.Order-of-Magnitude Gains Become Conservative✦ from: Recursive self-improvement and continual learning, combined with the work of singular talents like Andrej Karpathy, could pull the future of AI forward in a very real way, potentially making annual order-of-magnitude improvements seem conservative.Recursive Self-Improvement as AI's New Moore's Law✦ from: Recursive self-improvement, where an AI model has input into its own training, could put model improvement on a parabolic trajectory resembling a new form of Moore's law.The Trajectory: Order-of-Magnitude Every Year✦ from: Recursive self-improvement, where an AI model has input into its own training, could put model improvement on a parabolic trajectory resembling a new form of Moore's law.LLM Companies on Track for $200–400B ARR✦ from: LLM companies including OpenAI, Anthropic, and Google are on track for $200-400 billion of aggregate ARR by end of this year at high margins, proving strong ROI on AI infrastructure.Aggregate Revenue Build-Up✦ from: LLM companies including OpenAI, Anthropic, and Google are on track for $200-400 billion of aggregate ARR by end of this year at high margins, proving strong ROI on AI infrastructure.The three layers of anti-AI sentiment✦ from: The anti-AI sentiment in America is driven by three layers: a power-imbalance narrative where AI creates outsized returns for the few, foreign state actors funding anti-technology sentiment as a competitive strategy, and a deep psychological anti-humanist framing that offends human ego — rooted in the same disruption as the Copernican revolution.Layer 1: Power imbalance via slow diffusion✦ from: The anti-AI sentiment in America is driven by three layers: a power-imbalance narrative where AI creates outsized returns for the few, foreign state actors funding anti-technology sentiment as a competitive strategy, and a deep psychological anti-humanist framing that offends human ego — rooted in the same disruption as the Copernican revolution.Layers 2 & 3: External funding + psychological ego shock✦ from: The anti-AI sentiment in America is driven by three layers: a power-imbalance narrative where AI creates outsized returns for the few, foreign state actors funding anti-technology sentiment as a competitive strategy, and a deep psychological anti-humanist framing that offends human ego — rooted in the same disruption as the Copernican revolution.The AI Power Imbalance✦ from: Americans have turned on AI because it creates a perception of power imbalance where a small number of people capture outsized returns while the broader population sees no immediate benefit, and this is compounded by foreign state actors fueling anti-tech sentiment.Foreign State Actors Fuel the Backlash✦ from: Americans have turned on AI because it creates a perception of power imbalance where a small number of people capture outsized returns while the broader population sees no immediate benefit, and this is compounded by foreign state actors fueling anti-tech sentiment.
Related episodes