"Bernie Demands the Labs Stop, Wall Street Turns GPUs Into Bonds, Grok 4.7 Takes #1 ft. Emad Mostaque" — Peter Diamandis (Moonshots)

Why this is in the vault

Weekly AI-news roundup with recurring cast (Diamandis, Alex Wissner-Gross, Dave Blundin, Salim Ismail) plus guest Emad Mostaque, covering two threads directly relevant to RDCO's AI-infra watch and AI-policy tracking: Nvidia's move to securitize GPU compute as a Wall Street-backed financial asset class, and Senator Bernie Sanders' formal pause-AI letter to Anthropic/Meta/OpenAI following a Stanford bacteriophage-design result. Also tracks frontier-model cadence (Grok 4.6/4.7 vs Opus 5) and Hollywood's AI-generated-film economics, both recurring RDCO watch items.

Episode summary

The Moonshots quintet covers nine stories: the $101M Healthspan XPRIZE finals, AI-generated feature films undercutting Hollywood budgets by ~98%, Grok 4.6's release and the rumored 4.7 (trained on SpaceX engineering data), Nvidia's $500B+ GPU-financing partnership with Wall Street asset managers, a "post-transformer" architecture claim (dragon hatchling) that panelist Alex Wissner-Gross debunks as unprincipled, Bernie Sanders' AI-pause letter paired with a Stanford AI-designed-bacteriophage story, Anthropic/EU AI-content watermarking and labeling moves, Zuckerberg's "personal superintelligence" essay, and an eVTOL ("flying car") industry consolidation (Archer acquiring three Boeing units). Emad Mostaque, returning guest and founder of Intelligent Internet, contributes UK/EU regulatory perspective and technical color on video-gen and bio-design models he has hands-on experience with (having built media-generation and biology models previously).

Key arguments / segments

• [00:01:03–00:19:00] Healthspan XPRIZE finals recap: $101M prize, 800 teams competed, 10 teams awarded $1M each; framed as the same catalytic-prize model that built the commercial space industry after the 1996 Ansari X Prize.
• [00:19:02–00:37:00] Hollywood economics collapse: Higgsfield's AI-generated feature film "Cully Hill Boys" cost $2M / 4 weeks / 28 people vs. $20-100M / 12-18 months traditionally — roughly 2% of cost, 6% of time. 9 of top 10 text-to-video leaderboard models are now Chinese-made.
• [00:40:00–00:50:00] Grok 4.6 released, matching GPT-5.6 on Artificial Analysis Intelligence Index (score 61); Wissner-Gross frames it as xAI using the Cursor acquisition's reasoning-trace data to "catch up" to frontier the way Chinese labs allegedly distill from Claude, while xAI's compute advantage lets it try to leapfrog. Grok 4.7 (rumored, ~2T parameters) claimed by Elon to beat Opus.
• [00:51:47–01:04:02] Nvidia announced a $500B+ financing partnership with Apollo, BlackRock, Blackstone, Brookfield, and KKR to fund AI-infrastructure buildout via third-party capital rather than Nvidia's own balance sheet — panelists debate whether this echoes mortgage-backed-securities risk (Larry Fink comparison) if GPU depreciation assumptions break on an architecture or physics breakthrough.
• [01:04:36–01:11:01] "Dragon hatchling" post-transformer architecture claim on ARC-AGI: Wissner-Gross calls it "a hot mess" — a grab-bag of biomimetic motifs that overfits to one benchmark rather than a principled, Bitter-Lesson-compliant simplification.
• [01:11:31–01:31:52] Bernie Sanders sent a formal letter to Anthropic/Meta/OpenAI CEOs demanding an AI development pause, citing a Stanford team's use of the Evo2 generative model to design 16 viable, previously-nonexistent bacteriophages. Panel consensus (Mostaque, Wissner-Gross, Blundin) rejects a pause as unenforceable and counterproductive, arguing for "defensive co-scaling" (better detection/monitoring, e.g. ubiquitous DNA sequencers, prompt-level monitoring) instead of capability throttling.
• [01:32:44–01:41:00] Anthropic began embedding invisible watermarks in all Claude-generated text; the EU is launching mandatory AI-content labeling. Panel is broadly critical, comparing it to GDPR cookie banners and predicting rapid circumvention (watermark-removal tools already surfaced on X within 24 hours).
• [01:41:35–01:56:00] Zuckerberg published a "personal superintelligence" essay + video announcing open-weight Muse models (30B dense, plus upcoming Muse-Park 1.2) and a "Future Is for Everyone" community-benefit fund for data-center host towns; panel is split between praising the open-source push and skeptical of Meta's underlying ad-distribution incentives.
• [01:56:00–02:06:00] eVTOL consolidation: Archer Aviation acquired three Boeing units (Wisk Aero, Insitu, SkyGrid AI) with Boeing taking an equity stake; Joby and Archer lead FAA certification, targeting ~$3/seat-mile (Joby) with Uber-X-level pricing targeted longer term.

Notable claims

• Higgsfield's AI-generated feature film reportedly hit a $700M annualized revenue run rate roughly 18 months after founding (per Wissner-Gross, unverified third-party figure cited on the pod).
• Nvidia is organizing $500B+ in third-party institutional financing (Apollo, BlackRock, Blackstone, Brookfield, KKR) for AI-infrastructure buildout — a shift from Nvidia financing customer purchases directly to structuring compute as an investable asset class.
• Stanford researchers used the Evo2 generative model (open-source, ~40B parameters) to design ~300 candidate bacteriophage genomes, producing 16 viable phages effective against E. coli strains with no natural resistance — cited as the scientific trigger for Sanders' pause letter.
• Global cost of age-related chronic disease is cited at ~$20T/year against a ~$120-130T global economy.
• Joby Aviation targets ~$3/seat-mile in early commercial eVTOL service (Dubai launch this year), with a longer-term cost target around $15-25/trip.

Guests

• Emad Mostaque — Founder of Intelligent Internet (formerly founder/CEO of Stability AI); recurring Moonshots guest with direct technical background in generative media and biological-sequence models; based in the UK, offers European regulatory perspective on the AI-pause and EU labeling stories.

Recurring panel (not "guests" in the single-episode sense but central to the discussion): Alex Wissner-Gross (AI researcher, "in-house superintelligence"), Dave Blundin (investor), Salim Ismail (Exponential Organizations author/founder of OpenExO).

Sponsorship

Three sponsor reads in this episode: Google for Startups (generative-media deployment guide/blueprint for Google DeepMind models), Blitzy (autonomous AI software-development platform, claims 5x engineering velocity), and Fountain Life (longevity/preventive-health membership service — Diamandis is affiliated as a member of its medical team advisory presence, a standing host-adjacent relationship disclosed in prior Moonshots episodes).

Mapping against Ray Data Co

Direct relevance to two active RDCO watch threads: (1) the GPU-financialization story (Nvidia + Wall Street asset managers structuring compute as a bond-like investable asset) extends the compute-economics thread already tracked via the dwarkesh Jensen Huang and Dylan Patel episodes — useful corroborating data point for any investing thesis on the AI-infrastructure capital cycle, particularly the depreciation-risk objection raised on this pod (an architecture or physics breakthrough stranding financed GPU assets). (2) The Bernie Sanders pause-letter story is a policy-signal data point for RDCO's low-priority AI-policy watch — worth a light touch, not an active thesis input. Weak-to-medium mapping otherwise: the Hollywood AI-film economics and eVTOL stories are interesting frontier-tracking color but don't connect to an active RDCO bet.

Related

• [[2026-07-19-moonshots-ep-272-emad-mostaque]] — prior Moonshots episode with the same guest, useful for continuity on Mostaque's positions.
• [[2026-08-03-dwarkesh-why-smarter-ai-models-could-drive-up-compute-prices-10x]] — compute-pricing/economics thread this episode's Nvidia-financing story directly extends.
• [[2026-04-15-dwarkesh-jensen-huang-nvidia-moat]] — Nvidia strategy background relevant to the GPU-as-financial-asset framing.