“Forget Everything You Believed About Computing” — Peter H. Diamandis Moonshots EP #102
Episode summary
Diamandis interviews Guillaume “Gill” Verdon, founder of the effective accelerationism (e/acc) movement and CEO of Extropic AI. Verdon is a quantum physicist who spent three years at Google working with Sergey Brin on quantum technologies before founding Extropic to build “thermodynamic computers” — a third branch of computing alongside classical digital and quantum. The conversation covers the e/acc philosophy (embracing technological acceleration rather than fearing it), the Kardashev scale for measuring civilizational energy consumption, and Extropic’s technical vision of embedding AI algorithms directly into the physics of electrons to achieve energy efficiency orders of magnitude beyond current GPUs and potentially even the human brain.
Key arguments / segments
- [00:00:00] e/acc origins: Verdon wrote the accelerationist manifesto during a lull while founding Extropic; went viral as a counter-narrative to AI doom/overregulation
- [00:03:00] Moonshot vision: brain-scale AI processor that is more energy efficient than the human brain (14-20W for 100B neurons); current GPU equivalents are tens to hundreds of millions of times less efficient
- [00:10:00] Variance as survival strategy: biological analogy of genetic diversity; argues centralized AI control reduces variance and makes civilization fragile
- [00:28:00] Universe 25 / mouse utopia: warns that eliminating struggle leads to civilizational collapse; humans need to “uplevel” ambitions
- [00:29:00] Kardashev scale: humanity is still below Type 1; 8,000x more solar energy hits Earth than we consume annually
- [00:42:00] Entrepreneurial advice: stay away from typical white-collar software plays; deep tech and the “world of atoms” is where durable moats exist
- [00:47:00] Thermodynamic computing explained: a third branch (neither deterministic digital nor quantum) where probabilistic states of electrons are the compute medium, potentially thousands of times more energy efficient
Notable claims
- GPUs are tens to hundreds of millions of times less energy efficient than the brain for equivalent AI workloads
- Extropic aims to surpass even brain-level efficiency by embedding ML algorithms into the physics of electrons
- Current LLMs will saturate near typical human intelligence until AI systems are embodied and can query the physical environment directly
Bias / sponsor flags
- Fountain Life sponsorship: extended mid-roll by Diamandis (he co-founded it with Tony Robbins) — health diagnostics pitch
- Verdon is promoting his own company (Extropic AI) and the e/acc movement he founded; no independent benchmarks or peer review cited for efficiency claims
- Efficiency claims for thermodynamic computing are aspirational, not demonstrated at scale
Relevance to Ray Data Co
Moderate. The deep-tech moat argument (atoms > bits) is a useful lens for evaluating startup defensibility. The e/acc framing — that centralized AI control reduces variance and fragility — aligns with the decentralization thesis. The thermodynamic computing concept is worth tracking as a potential paradigm shift in AI infrastructure costs, but is pre-commercial and speculative.