In this highly technical and wide-ranging podcast, Guom Verdon, founder of Extropic and a pioneer in thermodynamic computing, shares his journey from aspiring theoretical physicist to innovator in a new computing paradigm. Growing up fascinated by quantum mechanics and inspired by figures like Feynman and Hawking, Verdon initially pursued quantum computing and machine learning at Alphabet. However, he became disillusioned with the reductionist approach in physics and recognized the need to embrace complexity and probabilistic models to better understand and simulate the universe. This realization led him to develop thermodynamic computing, which leverages the stochastic physics of electrons to perform probabilistic inference more efficiently than classical deterministic or quantum computers.
Verdon explains that traditional computing strives to minimize noise and maintain determinism, which becomes increasingly energy-intensive and impractical as devices shrink and approach physical limits. In contrast, thermodynamic computing embraces noise and stochasticity, operating closer to equilibrium states to achieve massive gains in energy efficiency. His company Extropic has developed silicon-based chips that harness natural stochastic electron dynamics to accelerate Markov Chain Monte Carlo algorithms, foundational for probabilistic machine learning models like energy-based models (EBMs). These chips represent a new class of physics-based computers that integrate analog stochastic processes with digital control, promising 1,000 to 100,000 times energy efficiency improvements at the chip level.
The discussion highlights the broader implications of this technology for AI and computing. Current AI workloads increasingly rely on probabilistic sampling methods, yet run inefficiently on deterministic hardware. Thermodynamic computing could unlock the full potential of EBMs and other probabilistic models by providing native hardware acceleration, enabling more scalable, energy-efficient AI systems. Verdon envisions a heterogeneous computing future where quantum, thermodynamic, and classical deterministic processors coexist, each specialized for different computational tasks. This shift is seen as inevitable due to fundamental physical limits—what he terms the “thermal danger zone”—that constrain further scaling of traditional digital computers.
Beyond technology, Verdon introduces his philosophical movement, Effective Accelerationism (EAC), which advocates embracing growth and technological acceleration as a cultural imperative to avoid stagnation or decline. Rooted in thermodynamic principles and complex systems thinking, EAC promotes maximizing civilization’s free energy consumption as a measure of progress. He argues that resisting acceleration is counterproductive and that societies must balance exploration (high variance) and exploitation (optimization) to thrive. This worldview also informs his stance on AI governance, decentralization, and democratization of intelligence, emphasizing the importance of distributed control over cognitive extensions to prevent monopolistic or authoritarian dominance.
In closing, Verdon invites listeners to engage with Extropic’s emerging technology through upcoming private alpha access and encourages the machine learning community to deepen their understanding of probabilistic models. He stresses that thermodynamic computing is not just a niche innovation but a foundational shift with broad applications in AI, optimization, and scientific simulation. The conversation blends deep technical insights with visionary perspectives on the future of computation, intelligence, and civilization, making it a compelling exploration of how pushing compute to the limits of physics can reshape technology and society.
