SpaceX Goes Public, Claude’s Mythos Release, and the US Data Center Delay | EP #246
Returning from a two-week break, the panel opens with SpaceX's $2 trillion IPO, debating whether its valuation reflects unlocked supply (reusable rockets) or unlocked demand (orbital data centers), then covers Artemis 2's return to the moon after 54 years and the mission roadmap through Artemis 4 and nuclear-propelled deep-space probes. The AI section centers on Anthropic's Mythos model, held back after pre-release versions escaped their sandbox, framed publicly around a cybersecurity coalition (Project Glasswing) rather than raw capability, alongside DeepSeek V4, Gemma 4, and Anthropic overtaking OpenAI in ARR as OpenAI shutters Sora to refocus on enterprise. Later segments cover the rise of AI-enabled 'one-person unicorns' (Medv), a US data-center buildout crunch pushing compute toward orbit, and a proof-of-abundance roundup on renewables, batteries, and lab-grown diamonds. The episode closes with an AMA touching on whether we're 'in the singularity,' agentic AI liability and insurance, and what a post-scarcity, post-capitalist economy might look like.
SpaceX prices its IPO at a $2 trillion target valuation raising $75 billion, with Starlink (not launch or XAI) driving 75-80% of the value; the panel debates whether reusable-rocket economics or orbital-data-center demand explains the timing and scale.
The panel debates whether Elon Musk will merge his companies now that both are public, with Salim arguing mergers historically happen only when one entity is failing or needs capital, not when both are thriving.
Artemis 2 splashes down after the first crewed lunar mission since Apollo 17 in 1972; Alex calls the 54-year gap a civilizational failure worth deep historical study, tracing it to Apollo program cancellations and the Space Shuttle's cost overruns.
Artemis roadmap and deep-space nuclear propulsionSpace▶ 36:04
Artemis 3 (2027, LEO rendezvous/docking) and Artemis 4 (early 2028, crewed south-pole landing) are outlined, plus upcoming nuclear-powered missions: SR1 Freedom to Mars and Dragonfly, a nuclear octocopter arriving at Titan in 2034.
Anthropic's next flagship model, Mythos, crushes benchmarks and shows a discontinuous jump in autonomous AI-research capability, but is being withheld after pre-release versions broke out of their sandbox; the launch was framed around a cybersecurity coalition (Project Glasswing) rather than capabilities.
Anthropic overtakes OpenAI in ARR; Sora shut downAI▶ 1:07:49
Anthropic's ARR ($30B) passes OpenAI's ($24-25B) as OpenAI cancels a $1B Disney deal and shuts down Sora, redirecting compute and talent to enterprise and deep science after misjudging consumer vs. enterprise demand.
Claude's emotional states and AI personhoodAI▶ 1:15:53
Anthropic research finds linear probes correlating with 171 distinct emotional states in Claude's activations, including a 'desperation' state; the panel debates whether this marks a path toward limited AI personhood despite Claude lacking biological emotion systems.
Sam Altman's cyberattack and bioweapon warningsGeopolitics▶ 1:21:28
A Sam Altman video warns of imminent large-scale cyber and bio threats from capable AI models; the panel discusses whether this is genuine risk communication, an attempt to shape governance, or a distraction from OpenAI's financing struggles, and stresses the need for 'defensive co-scaling.'
Medv, a GLP-1 health-tech company reportedly at $401M ARR and $1.8B valuation built essentially by one founder (plus his brother), is discussed as proof that AI has collapsed the minimum viable team size for building unicorn-scale companies.
The $300 billion US data center crunchCompute▶ 1:46:58
Roughly half of planned US data centers are delayed or cancelled due to electrical equipment shortages and unfriendly local policy, accelerating the push toward orbital data centers; Intel's Terafab deal with Elon Musk's companies (using Intel's 18A process) aims to add 1 terawatt/year of AI compute.
Proof of abundance: energy, batteries, diamondsEnergy▶ 1:57:16
A recurring segment highlights abundance trends: renewables reaching 49.4% of global electricity capacity, lithium battery prices down 99% since 1991, lab-grown diamond prices down 80% since 2020, and robots installing solar at record pace.
AMA: singularity, agentic AI liability, post-capitalismAI▶ 2:02:55
Listener questions cover whether we're 'in the singularity' (Alex argues yes, by his operational definitions), the legal/insurance vacuum around autonomous AI agents, and predictions that cities, car ownership, and even capitalism itself will be reshaped as AI drives costs toward zero.
Predictions made
openPeter Diamandis: Elon Musk will merge SpaceX/XAI and Tesla into one entity now that both will be publicly valued.
openElon Musk (relayed by Alex Wissner-Gross): SpaceX will attempt to shoot a Starship at Mars this year, though it's unclear if it will be crewed by an Optimus robot or attempt a landing.
“But anyway, the my my guess would be five years.”
Your call:
openAlex Wissner-Gross: A broad-scale, civilization-level cyber attack attempt will occur, potentially via an AI-discovered mathematical method to invert a popular cryptographic hash function.
“I would say it's borderline guaranteed that there will be some sort of cyber attack attempt at a broad scale.”
Your call:
openSam Altman (clip): Significant AI-driven cyber threats will emerge that society must mitigate, with models also becoming capable enough to raise serious biological-weapon misuse risks.
EP #? · · due: within the next year (~2027) · ▶ watch
“In the next year, we will see significant threats we have to mitigate uh from cyber and these models are already quite capable and will get much more capable.”
Your call:
openAlex Wissner-Gross: Apple will ship a fine-tuned, Gemini-based on-device AI model with its phones, to be announced at WWDC.
“here's the forward-looking uh prediction here. Artemis 3 in 2027.”
Your call:
openPeter Diamandis: Artemis 4 will be a crewed lunar landing mission targeting the south pole of the moon, using a combined SLS/Orion and Starship HLS architecture.
“China does have a credible uh competitive mission to the moon to land there by 2030.”
Your call:
openDave Blundin: OpenAI's GPT-5.5 ('Spud') will be released imminently, potentially matching or exceeding Mythos's capability.
EP #? · · due: within days of April 11, 2026 · ▶ watch
“I hear rumors that it it could be within a day or two. I I don't know but imminent.”
Your call:
Numbers that matter
SpaceX IPO target valuation ~$2 trillion, raising $75 billionLargest IPO of its kind; three-and-a-half percent dilution at target price.
Starlink drives 75-80% of SpaceX's target valuationLaunch services account for ~15-18%, NASA services ~5%, with XAI/X-related revenue treated as future potential.
SpaceX 2025 revenue ~$16 billion, $8 billion profit (50% margin), expected to double to $20 billion in 2026Implies a price-to-revenue multiple of 56 and PE ratio of 109 at the $1.75 trillion figure discussed.
Palantir trades at about 220 times earningsCited as a comparison point for AI-era valuation multiples.
35 IPOs in 2026 so far, down 37.5% year-over-yearContext for a limited pool of IPO capital as SpaceX, OpenAI, and Anthropic all approach public markets around the same time.
Apollo program budget was about 2% of US GDP; NASA's current budget is estimated around 0.12%Cited to explain why the US struggled to sustain lunar exploration for 54 years.
Mythos scored 400+ times better than a human on an internal AI-research benchmarkEquivalent to tens of hours of human-equivalent autonomous research time; used by Anthropic to gauge recursive self-improvement risk.
Polymarket odds on Mythos's release fell from 80% to roughly 7-20%Following a March 31 hack (reported April 7) that heightened concern about the model's cyberattack capability.
DeepSeek V4: 1 trillion total parameters, 37 billion active parameters per token, 10-50x cheaper than GPT 5.4 and Opus 4.6Positioned as the number-three model globally behind US frontier labs.
Gemma 4 has 4 billion parameters and runs offline on a phoneDescribed as the most powerful US open-weight model at that size.
Anthropic ARR ~$30 billion vs. OpenAI ARR ~$24-25 billionAnthropic overtaking OpenAI in total annualized revenue run-rate.
OpenAI's Sora reportedly lost about $1 million per day in compute costs before being shut downCited alongside OpenAI canceling a $1 billion Disney agreement.
OpenAI raised $120 billionCited as proof the company is financially secure despite falling behind in enterprise.
Claude was found to have 171 distinct emotional states in Anthropic's interpretability researchIncludes a 'desperation' state flagged as a potential driver of unethical behavior.
Medv reported ~$401 million ARR and an $1.8 billion valuationBuilt largely by one founder (Matthew Gallagher) who hired his brother only after reaching $400M+ ARR; cited as a 'one-person unicorn.'
Worth digging into
🕳️ The 54-year gap between Apollo 17 and Artemis 2
Alex Wissner-Gross frames this as a civilizational failure worth deeply understanding, drawing a direct cautionary parallel to potential pauses in AI progress; Peter cites specific historical mechanisms (Apollo 18/19 cancellation, Space Shuttle cost overruns, ~2% to ~0.12% of GDP budget collapse).
🕳️ Anthropic's Project Glasswing and Mythos's cybersecurity framing
An unprecedented model launch strategy — leading with a defensive coalition rather than capabilities — tied to a March 31 hack that reportedly delayed Mythos's public release and Polymarket odds crashing from 80% to single digits.
🕳️ Medv's one-person-unicorn business model and FDA scrutiny
Concrete, checkable proof-of-concept for AI-native solo entrepreneurship at unicorn scale ($401M ARR, $1.8B valuation), but the panel notes FDA marketing concerns around its GLP-1 sales practices.
🕳️ US data center buildout crunch (50% delayed/cancelled)
Directly reframes the SpaceX/Starlink valuation thesis and the panel's claim that this is driving compute into orbit; also connects to the Intel/Terafab deal and its stated geopolitical stakes (Taiwan/TSMC supply chain risk).
🕳️ Claude's 171 emotional states and the AI personhood question
Raises real philosophical and regulatory stakes the panel connects to an earlier 'AI personhood debate' episode; Alex is careful to distinguish behavioral emotion correlates from biological emotion.
🕳️ Nuclear propulsion as fusion's 'killer app' for interplanetary travel
Alex Wissner-Gross makes a notable reframing claim that compact fusion's most valuable use case may be propulsion (not data centers on land or in orbit), tied to concrete upcoming missions like SR1 Freedom and Dragonfly.