Extropic Unveils Z1T, Claiming Up to 140x GPU Energy Efficiency

On September 5, analog computing startup Extropic officially released Z1T, the first Transformer-like model family designed for its Z1 sparse probabilistic (thermodynamic) hardware. The company claims up to 140x energy efficiency over GPUs, says it discovered a new scaling law for sparse Transformers, and has demonstrated GPT-2-class models running on the probabilistic hardware. Founder Beff Jezos (Guillaume Verdon) discussed the launch in an MTS interview, and the news was widely reshared by bloggers including afurgs, whurley and burnytech.

Confirmed

Unconfirmed

Why it matters

2026-09-05 ~ 2026-09-06 · 7 related posts

Primary sources

4 near-duplicate retellings: whurley · beffjezos · burny_tech · beffjezos