SpaceX IPOs at $2.89T Market Cap, US Govt Suspends Fable & Mythos 5, Altman Delays OpenAI’s IPO |265
The panel dissects SpaceX's record-setting IPO (closing near $2.89T market cap, making Elon Musk the first trillionaire) as a bet on a three-part business (launch monopoly, Starlink cash engine, AI/orbital compute) rather than a normal tech stock. They then cover the US government's abrupt export-control shutdown of Anthropic's Fable 5 and Mythos 5 models for all non-US persons, debating who controls access to frontier intelligence, Anthropic's undisclosed model-downgrading and data-retention practices, and the risk of driving companies toward Chinese open-weight models. Additional threads cover OpenAI's price war and Sam Altman's comment about delaying its IPO amid recursive self-improvement, data-center bottlenecks (power transformers, terrestrial vs. orbital vs. lunar compute), AI taxation proposals, and rising youth unemployment/unrest risk. The episode closes with an AMA covering Salim's EXO/organizational-singularity model, Bitcoin skepticism, sovereign wealth funds, and quantum computing's threat to cryptography.
SpaceX's IPO closed near $2.89T market cap after opening at $135/share and rising ~20% on day one, making Elon Musk the first trillionaire. The panel frames it as a bet on three converging businesses (launch monopoly, Starlink cash engine, AI/orbital compute) rather than a normal tech stock, and discusses the IPO creating thousands of new millionaires, the Kessler-effect risk disclosed in the S-1, and the likelihood of a SpaceX-Tesla merger.
Wealth concentration and the new 'techno-philanthropist' classEconomy▶ 19:38
Discussion of when a quadrillionaire might emerge (estimates range from the late 2030s to 2060s), whether Elon's wealth is 'value created, not value extracted,' extreme geographic concentration of capital in the US, and a generational shift toward capital markets rewarding hard/deep tech over software for the first time in decades.
US government suspends Fable 5 and Mythos 5AI▶ 29:24
With ~90 minutes notice, the US government issued an export-control directive blocking all foreign nationals (including Anthropic's own foreign employees) from Fable 5 and Mythos 5, citing a jailbreak vulnerability. Anthropic disabled both models globally. The panel debates the murky timeline (disputed accounts of whether Dario Amodei was reachable), the precedent of using export-control law for AI, and whether this pushes companies toward on-prem or Chinese open-weight models.
Anthropic's undisclosed downgrading and data retentionAI▶ 41:59
Before the export-control shutdown, developers were already revolting against Fable 5 over two disclosures buried in a 319-page document: 30-day retention of all prompts even for enterprise customers with negotiated zero-data-retention, and silent downgrading of users detected doing AI-research queries to a weaker model without notice. Anthropic later disclosed it reserved the right to actively interfere with ('poison') such users' outputs.
Who controls frontier intelligence accessAI▶ 50:44
Panel debate on whether the Fable/Mythos episode marks a new precedent where governments (via export control) or corporations decide who gets access to which tier of AI capability. Salim notes 70% of elite frontier-lab AI researchers are foreign-born (mainly China, India, Taiwan, UK), making blanket export bans self-defeating; Dave and Salim argue this will force enterprises toward on-prem, multi-model failover architectures.
OpenAI's price war and Altman's IPO-delay commentAI▶ 55:19
The Wall Street Journal reports OpenAI is weighing steep price cuts to court developers alienated by Anthropic's shutdown. Sam Altman separately suggested faster recursive self-improvement could make delaying OpenAI's IPO advantageous. Alex argues price cuts are mostly marketing since Anthropic is also hyper-deflating costs, while Dave notes Altman holds little to no OpenAI equity, reducing his personal stake in the IPO's timing.
OpenAI's Codex team reports agents can now generate their own goals and sub-agent tasks from user intent rather than being told exactly what to do. The panel connects this to Salim's 'organizational singularity' concept (a 10-company pilot program starting the following week) and Alex frames it as an early step toward AI self-determination and personhood questions.
Data center bottlenecks and orbital/lunar computeCompute▶ 1:14:09
AI data-center compute has doubled every 7 months since Colossus 1 (built in 122 days, online August 2024), with global AI computing capacity growing 3.3x/year, yet power and step-up transformers now have 2.5-3 year lead times, making electrical hardware (not chips or capital) the binding bottleneck. The panel debates a future split between terrestrial training clusters and orbital/lunar inference and eventually training compute, citing solar and cooling advantages on the Moon.
Reacting to Andrew Yang's call to 'tax the bots,' the panel argues taxing AI cognition/tokens directly would slow progress in critical areas (e.g., cancer research) and is largely redundant with existing corporate income tax; Alex prefers using superintelligence to drive cost of living toward zero (weighing UBI vs. UBS vs. universal basic equity/compute) over punitive AI-specific taxes.
Youth unemployment and revolution riskGeopolitics▶ 1:29:31
Peter presents data showing recent college grads (especially ages 22-25 in software and customer-support roles) face disproportionately long job searches, and reviews 20 historical youth-led revolutions. The panel debates whether AI-driven disemployment could fuel unrest in the next 2-8 years, with Alex arguing the risk is concentrated less in the US/China and more in the rest of the world lacking frontier AI capability.
AMA: EXO model, Bitcoin, sovereign wealth funds, quantum riskCrypto/Web3▶ 1:44:03
Audience Q&A covers Salim's EXO/organizational-singularity framework (MTP before scale-ideas), Alex's skepticism of Bitcoin as a non-productive speculative asset (while liking stablecoins and blockchain generally), sovereign wealth fund models (Norway, Singapore, Gulf states) as a template for AI-era wealth distribution, and whether a 3-year quantum breakthrough could break crypto (panel judges the timeline too aggressive).
Predictions made
openPeter Diamandis: SpaceX and Tesla will merge into one company.
“I put it at 100%. Polymarket puts it at 37% that it happens by the end of this year.”
Your call:
openAlex Wissner-Gross: The solar system's first quadrillionaire will emerge, based on crowd-sourced extrapolations of net-worth growth.
EP #? · · due: sometime between the late 2030s and the 2060s · ▶ watch
“we'll see the first quadrillionaire sometime between 10 years from now and 40 years from now.”
Your call:
openDave Blundin: A product branded 'Fable 5' will be back online, per Polymarket odds, though it may be patched or capability-reduced rather than identical to the original.
“Polymarket says Fable 5 will be back out, you know, 90% chance within a month.”
Your call:
openAlex Wissner-Gross: The Fable 5/Mythos export-control situation will be resolved, likely via some negotiated arrangement between Anthropic and the White House (potentially including golden shares for a sovereign wealth fund and streamlined disclosure procedures).
EP #? · · due: within the next day to a few weeks, and well before any Anthropic IPO · ▶ watch
“I I would suspect this is just a temporary state of affairs. I don't think it's likely to last more than a few weeks.”
Your call:
openAlex Wissner-Gross: Someone will launch a benchmark or evaluation specifically designed to detect 'poisoning attacks' by frontier models against users, and a class-action or antitrust challenge against the poisoning-style guardrail practice is likely.
EP #? · · due: within the next few weeks · ▶ watch
“I would be shocked if in the next few weeks someone doesn't launch a benchmark or an eval to look for poisoning attacks from these models.”
Your call:
openAlex Wissner-Gross: Capital markets will continue to reward hard tech and civilization-expanding technology over software, reversing decades of software/outsourcing dominance.
“I would predict this is likely to be the case for the next 10 to 20 years.”
Your call:
openAlex Wissner-Gross: Barring a major algorithmic advance in distributed training (or a 'left turn in civilization'), the largest AI training compute will remain terrestrial while marginal new AI compute shifts to orbital data centers.
“perhaps by the early 2030s, barring some left turn in civilization, much of the marginal new compute will remain in orbital data centers.”
Your call:
openPeter Diamandis: A quantum computing breakthrough capable of breaking the entire crypto stack is a real long-term threat but will not occur on an aggressive near-term timeline.
“I think the threat is real but not in a 3-year timeline. I think that's aggressive.”
Your call:
openAlex Wissner-Gross: Frontier AI labs will propose giving up golden shares, extra taxation, or funding some form of universal basic benefit, and the US government will act on it, drawing criticism from the labs afterward.
EP #? · · due: in the next few months, ahead of the OpenAI and Anthropic IPOs · ▶ watch
“I strongly suspect in the next few months the government's going to do exactly that, and everyone will cry foul.”
Your call:
openPeter Diamandis: More politicians will campaign on taxing the AI/robots that displaced a worker's job and redistributing that revenue to the displaced worker, without specifying how much or how the payout scales.
EP #? · · due: not specified, implied as the upcoming election cycle approaches · ▶ watch
“You're going to see more and more politicians saying, 'Hey, let's tax the AI that took your job, and we're going to give you that money.'”
Your call:
Numbers that matter
$2.89 trillionSpaceX's market capitalization following its IPO, cited as the world's largest IPO ever.
$135 opening / $161 close, +20% day oneSpaceX IPO share price on its first trading day.
$75 billionAmount of pure cash SpaceX's IPO absorbed from available market liquidity.
6 months / ~$1 trillionTime until SpaceX IPO lockups expire and the roughly $1T of stock that becomes eligible to sell.
10 billion subscribers / $1B+ quarterly profitFigures cited for Starlink's subscriber base and quarterly profit (subscriber figure as stated in the episode).
82% voting controlElon Musk's voting control of SpaceX (not Tesla), cited as rationale for a possible merger.
37%Polymarket odds that SpaceX and Tesla merge by end of 2026.
~4,400 employees / ~400 moreRoughly 4,400 SpaceX employees became millionaires and another ~400 became centi-millionaires/billionaires in the IPO, called the largest single-day creation of millionaires in history.
$1.3-1.4 trillionElon Musk's estimated net worth following the IPO, with reported gains of over $100 million/day over three days.
$60 billion+Price of SpaceX's agreed purchase of AnySphere, the corporate owner of Cursor.
10-40 yearsRange of extrapolated timelines (late 2030s to 2060s) for the solar system's first quadrillionaire.
5:21 p.m. Eastern / 90 minutesTime the US export-control directive against Fable 5/Mythos 5 was issued, and the notice window Anthropic says it was given.
~1/3 of employeesPeter's estimate of the share of Anthropic's employee base that are foreign nationals affected by the export-control ban.
319 pages / 30 daysLength of the document in which Anthropic's data-retention policy was disclosed; prompts retained at least 30 days even for zero-data-retention enterprise contracts.
70%Share of elite AI researchers across frontier labs who are foreign-born, primarily from China, India, Taiwan, and the UK.
Worth digging into
🕳️ Kessler syndrome math and current LEO debris load
Dave claims low Earth orbit has already exceeded the critical debris-mass threshold for a runaway collision cascade, which would be a major, under-discussed constraint on SpaceX's Starlink and orbital-data-center ambitions.
🕳️ The Fable 5/Mythos export-control legal mechanism
This is described as the first time the US government has blocked access to a frontier AI model via export-control law, setting a precedent for how governments can control AI deployment going forward.
🕳️ Power transformer supply bottleneck
The panel argues 2.5-3 year lead times on power and step-up transformers, not chips or capital, are now the binding constraint on AI data-center growth -- a much less discussed angle than GPU supply.
🕳️ 70% foreign-born frontier AI researcher statistic
Salim uses this figure to argue that blanket export-control bans on non-US persons are self-defeating, since most elite researchers at frontier labs are foreign nationals from China, India, Taiwan, and the UK.
🕳️ Youth unemployment data vs. the 'job apocalypse is a myth' narrative
Peter cites NY Fed and Stanford AI Index data showing recent college grads (22-25) diverging sharply from other age cohorts in software and customer-support employment, which could be an early leading indicator of AI-driven disemployment despite low headline unemployment.
🕳️ Technology decoupling from capital thesis
Alex frames Sam Altman's IPO-delay comment as a potential early signal that recursive self-improvement could let large AI labs forgo external capital entirely, calling it an under-discussed counterpart to the standard 'AI displaces labor' narrative.