Extropic's real bet: thermodynamic chips replacing matrix multiplications, not cheap RNGs
rosemaryoannah · reddit · 2026-09-03
A detailed Reddit breakdown of Extropic's thermodynamic computing thesis walks through the seven steps of LLM token generation and argues the expensive part is producing scores via massive matrix multiplications, not final sampling. So if Extropic were only using thermal noise as a cheap RNG, it would be a marginal optimization. The author contends the real bet is hardware-native stochastic computation: pbits physically evolving toward useful probability distributions, replacing deterministic matrix work — meaning the chips aren't drop-in GPU replacements but a wager on new low-energy AI architectures built for this hardware.
More from Infra
- IETF drafts OAuth 2.1 Authorization Framework, set to obsolete OAuth 2.0 (RFC 6749) — jedisct1 · 2026-09-03
- Figure commits $3.5B to deploy up to 100,000 NVIDIA Vera Rubin GPUs with Nscale — adcock_brett · 2026-09-03
- Nvidia's vertical integration signals next-gen app layer push beyond CUDA — RachelVT42 · 2026-09-03
- ARK mid-year AI review: compute buildout and adoption keep beating expectations — downingARK · 2026-09-03
- Buying a refurbished 8xA100 server to colocate and rent out: one user's GPU math — Alarming-Ad8154 · 2026-09-03
- Hidden consumable under AI servers: PCB drill-bit demand is growing far faster than board volumes — tengyanAI · 2026-09-03