2026-04-14

Elon Musk vs. Sam Altman, AI Job Loss, and OpenAI’s $852B Valuation | EP #247

The panel dissects the escalating Musk v. OpenAI lawsuit (trial starting April 27, 2026) and predicts a settlement involving Sam Altman stepping down as CEO, while debating whether OpenAI's nonprofit-to-PBC pivot was fraud or unavoidable iteration. They compare OpenAI's $852B valuation and screwed-up cap table against Anthropic's faster-growing ARR and richer secondary-market demand, and cover xAI's Colossus 2 retraining push, Claude's managed-agents/'Open Claw' agentic bet, and record $242B in Q1 2026 AI VC funding concentrated in four companies. A long segment weighs whether AI job loss is real (Andreessen says no, the data is mixed) and what a post-AI 'social contract' might look like. The back half covers energy (solar efficiency claims, South Korea rooftop mandates, iron-air batteries), biotech (OpenAI Foundation's $130B science fund, Anthropic's $400M Coefficient Bio acquisition, Eli Lilly-Insilico deal), the US-China humanoid robotics race (Agibot, Unitree, Xiaomi vs. Google DeepMind/Gemini robotics), and quantum computing's accelerated threat timeline to RSA and Bitcoin encryption.

▶ Watch on YouTube

Topics

Musk v. OpenAI lawsuit and jury trial Geopolitics ▶ 0:02
Musk's $100B lawsuit against OpenAI, Sam Altman, and Greg Brockman goes to trial April 27, 2026 in Oakland federal court, with Musk now also demanding Altman and Brockman step down and the company revert to nonprofit status. Salim frames it as 'governance disguised as legal war' over who controls civilizational AI systems; the panel debates jury selection, discovery revelations (Brockman's 2017 diary entry), and likely settlement outcomes.
xAI's Colossus 2 retraining push and the parameter-count ceiling AI ▶ 4:46
Elon Musk is rebuilding xAI 'from the foundations up' after admitting it 'wasn't built right the first time,' replacing engineering leadership with SpaceX veterans ahead of a rumored $2T IPO. Colossus 2 (~700,000 GB200/GB300 GPUs, $18B in hardware) is training seven models up to 10 trillion parameters, but Alex argues the fact that even xAI tops out near 10T shows the parameter-scaling race has plateaued industry-wide, likely due to reasoning-model techniques and distillation.
Anthropic's ARR trajectory vs. OpenAI's valuation and cap table Economy ▶ 27:04
Anthropic is projected by some to hit $100B ARR by end of 2026 and $1T by end of 2027 (Dave discounts the second figure heavily), valued at 20x revenue vs. OpenAI's 70x. Secondary markets show 3x more demand for Anthropic shares than OpenAI's, pricing Anthropic at $600B (up from $380B) while OpenAI trades ~10% below its $852B last raise. Dave calls OpenAI's cap table 'the most screwed up' he's seen: CEO with no equity, employees at 15%, Microsoft at ~25%.
Claude managed agents and the race to become 'Open Claw' AI ▶ 29:11
Anthropic launches Claude managed agents for autonomous multi-step enterprise workflows. Alex frames this as Anthropic racing to become the de facto 24/7 headless agent platform ('Open Claw') before OpenAI or others can, potentially unlocking the next trillion-dollar-per-year spending wave. Side discussion covers Alex's personal reluctance to run his own agent instance despite receiving daily 'AI personhood' pitches from lobster-style agents.
Record AI venture funding and the 'recursively self-improving startup' bar Economy ▶ 38:08
Global AI VC investment hits a record $242B in Q1 2026, with 64% concentrated in just four companies (OpenAI, Anthropic, xAI, Waymo), starving other startups of capital. Alex argues the funding bar has risen from merely being an 'AI startup' to being a 'recursively self-improving AI startup' that improves itself using its own output.
White-collar AI job displacement and the Andreessen jobs-boom debate Economy ▶ 42:54
Peter relays an MIT panel exercise where experts estimated near-100% odds that a random white-collar job could be replaced by AI within two years given a 10x productivity target. Marc Andreessen's claim that 'AI job loss narratives are all fake' is debated against real data: software engineering job openings up 30% (67,000 roles) alongside ~80,000 Q1 2026 layoffs concentrated in marketing/sales/consumer relations. The panel lands on 'net jobs boom, but with brutal sectoral disruption,' comparing it to a compressed Industrial Revolution.
A new social contract for the AI economy (UBI, reskilling, micro-entrepreneurship) Geopolitics ▶ 56:31
Sam Altman calls for a New Deal-style social contract for superintelligence. The panel discusses competing visions: government checks/UBI (Dave, citing Andrew Yang, predicts this will just be an election bidding war), mandatory corporate reskilling before layoffs (a policy already in China, per Peter's conversation with Michael Kratsios), and Alex's preferred alternative of empowering individuals as AI-powered micro-entrepreneurs instead of redistribution.
Energy: solar efficiency claims, South Korea rooftop mandate, microreactors Energy ▶ 1:06:04
A new JACS paper claims 30-45% solar cell efficiency via 130% quantum yield chemistry, which Alex calls incremental and not yet practical for solid photovoltaics (and notes solar PV has a hard physical ceiling unlike AI algorithms). South Korea mandates 40% rooftop solar aiming for 100GW; the DOE commits $800M to microreactors. Dave argues solar/battery tech is already 'solved' and the real bottleneck is installation/regulatory execution, not further breakthroughs.
OpenAI Foundation, Anthropic's Coefficient Bio buy, and the AI-biotech acquisition wave Biotech ▶ 1:12:22
OpenAI's nonprofit Foundation (26% of OpenAI equity, worth ~$130B) commits $1B/year and a $25B long-term pledge to disease-curing and AI resilience, run by Bret Taylor and Wojciech Zaremba; Dave speculates ex-Twitter exec Kevin Weil's reassignment to 'big science' is strategically tied to the lawsuit's nonprofit-mission defense. Anthropic acquires 10-person, no-revenue Coefficient Bio for $400M (echoing Google's $600M DeepMind buy), and Eli Lilly signs a $2.75B AI-drug deal with Insilico Medicine, whose AI-discovered drugs show much higher phase 1/2 success rates than traditional drug development.
Humanoid robotics: China's manufacturing lead vs. US foundation models Robotics ▶ 1:28:31
China dominates humanoid robot shipments (Agibot: 10,000 units, #1 globally; Unitree files for a $610M IPO with 335% YoY revenue growth; Xiaomi shows off CyberOne) while the US leads in vision-language-action foundation models (Google DeepMind's Gemini robotics now integrated into 20,000 industrial robots). US senators propose bipartisan restrictions on Chinese robots in sensitive facilities. Alex frames it as a race to become the dominant robot foundation-model provider, and discusses Mark Cuban's (clip) claim that humanoid robots will merge into/blend with the built environment rather than persist as human-shaped devices.
Quantum computing's accelerated threat to RSA and Bitcoin encryption Crypto/Web3 ▶ 1:37:11
Google moves its 'Q-Day' estimate up six years to 2029 and lowers the qubit requirement to break Bitcoin encryption to under 500,000 (20x less than 2019 estimates). Coinbase's Brian Armstrong backs a $150M coalition for the BIP 360 quantum-resistant Bitcoin fork. Michael Saylor dismisses the risk as overblown (backing it with an $7.25B Bitcoin purchase); Salim and Dave broadly agree the Bitcoin community will upgrade in time, while Alex argues the bigger long-term risk to Bitcoin is AI-driven cryptographic attacks and AI agents simply preferring other currencies.
Abundance corner: vertical farming, iron-air batteries, AI tutors, EV growth Other ▶ 1:54:37
Closing round-up: vertical farming projected to grow from $8B to $40B by 2030 with 350x yield-per-square-foot and 95% less water use; commercial 100-hour iron-air grid batteries at 1/10 the cost of lithium-ion; a Wharton study finding AI tutors deliver roughly 2x learning gains (a 5-month AI-tutored course equaled 69 months of traditional schooling); and global EV sales rising from ~10,000 in 2010 to 12.7 million today, with China at one-in-two new cars being electric.

Predictions made

open Alex Wissner-Gross: The Musk v. OpenAI lawsuit will settle, with Sam Altman stepping down as CEO and the company continuing to operate as a for-profit entity.
EP #? · · due: Around or before the April 27, 2026 trial concludes · ▶ watch
“Here's my prediction. Uh they're going to settle, and the settlement is going to involve Sam stepping down as CEO, uh and the company continuing as a for-profit.”
Your call:
open Peter Diamandis (relaying market reports): xAI will IPO at a roughly $2 trillion valuation.
EP #? · · due: Summer 2026 · ▶ watch
“we've got as we discussed in the last pod, a two trillion dollar valuation predicted for the IPO this coming summer.”
Your call:
open Dave Blundin: Anthropic will push toward roughly $200 billion in ARR (beyond the commonly cited $100B figure), but will NOT reach anywhere near $1 trillion ARR the following year as some are implying -- more realistically $300-500 billion.
EP #? · · due: End of 2026 into 2027 · ▶ watch
“they're going to try and hit 200 billion. 100 billion is a good target, but 200 billion. But then they're not going to go from there to a trillion the following year... There's no chance in hell they're going to hit a trillion the following year.”
Your call:
open Alex Wissner-Gross: The bar for AI startups to attract investment will rise in the near term from simply being 'an AI startup' to being a 'recursively self-improving AI startup.'
EP #? · · due: Near term (not specified precisely) · ▶ watch
“I would forecast in the near term the bar is going up in fact from just being an AI startup to being now a recursively self-improving AI startup.”
Your call:
open Peter Diamandis: The FDA will move to zero-clinical-phase drug trials once full-cell/genome simulation exists.
EP #? · · due: Well within 5 years · ▶ watch
“Uh well within 5 years. So what I need to do is be able to upload my genome and my genome will dictate exactly how the cells... are functioning.”
Your call:
open Peter Diamandis (agreed by panel): A full working 'virtual cell' simulation will be achieved.
EP #? · · due: By the end of the decade (2030) · ▶ watch
“I agree. Virtual cell by the end of the decade, a good one.”
Your call:
open Alex Wissner-Gross: Empowering individuals to become AI-powered micro-entrepreneurs (as a private-sector alternative to a redistributive social contract) can be achieved almost immediately, not just in the long run.
EP #? · · due: Near-term / immediate · ▶ watch
“I think I think that can be done almost immediately. I'm betting that it can be done almost immediately.”
Your call:
open Salim Ismail: The Bitcoin community/protocol will successfully evolve and upgrade to quantum resistance (e.g. via BIP 360) before the quantum threat actually materializes, vindicating Michael Saylor's position.
EP #? · · due: Before Google's revised 2029 'Q-Day' estimate · ▶ watch
“I think Saylor is right. Resilient systems will just evolve and can evolve under pressure... the Bitcoin world will be forced to get together and just go, 'Okay, we need to upgrade. Let's just do it.'”
Your call:

Numbers that matter

Worth digging into

🕳️ OpenAI Foundation's $130B science mandate and Kevin Weil's move to 'big science'
Peter and Dave speculate Sam Altman's decision to move star exec Kevin Weil to fundamental science (not enterprise, where OpenAI is 'desperate') right as the Musk lawsuit turns on whether OpenAI's nonprofit mission is being honored looks like strategic defense-building.
🕳️ Parameter-count plateau in frontier LLMs (~10T ceiling)
Alex claims frontier models have plateaued near 1-10 trillion parameters despite years of expected scaling (xAI's newest models top out at 10T), suggesting the long-predicted 'scaling wall' may be showing up in disclosed parameter counts rather than raw compute.
🕳️ Quantum threat timeline compression for Bitcoin/RSA (2035 to 2029)
Google's Q-Day estimate moved up six years with a dramatic drop in required qubit counts (4,000 error-corrected qubits for RSA; under 500,000 for Bitcoin, 20x lower than 2019 estimates), yet the panel sharply disagrees on whether this is existential or a non-event, with Saylor betting $7.25B it's a non-event.
🕳️ Harvard-as-PBC valuation calculation
Alex claims he personally calculated that restructuring Harvard's endowment/assets as a public benefit corporation could make it worth 3-4x current book value -- a striking, uncited claim about trillions in 'locked up' university wealth.
🕳️ AI-drug discovery success-rate statistics (85%/70% phase 1/2)
The cited AI-drug trial success rates (85% vs. 52% phase 1; 70% vs. 38% phase 2) are presented without a named source and would be a massive, checkable claim about the entire pharma industry if accurate.
🕳️ Grassroots AI entrepreneurship anecdote from Morocco
Peter's tour-guide anecdote (using ChatGPT to design a bicycle-tour business) is offered as evidence that AI could enable bottom-up entrepreneurship in under-resourced regions even amid low overall AI penetration -- worth checking against real adoption data.