Extropic is building thermodynamic computing hardware

Extropic says it’s ditching power-hungry graphics cards for a new kind of "thermodynamic" AI chip—think dice-rolling hardware that leans into randomness to save energy. They teased a prototype board (XTR-0) and an open-source Python toolkit (THRML), but the comments didn’t come to clap; they came to brawl. One camp is hyped for anything that could slash data center power bills, the other says this feels more like sizzle than steak. A rival got name-dropped fast—Normal Computing—as users pointed out that Extropic has already “pivoted” from superconductors to standard chip tech, which set off a “what changed?” pile-on.

The hottest take? A blunt reduction: it’s basically a hardware random-number machine aimed at the de-noising step in image-making AI. Cue the memes: “so… a very fancy dice shaker?” Others blasted the launch for being impossible to parse without a PhD, calling it shiny marketing with zero plain-English proof. Still, there’s cautious optimism—one commenter is “75% believing” but waiting for real benchmarks, and a few are rooting for anything that makes AI less of a power-guzzling monster. Verdict from the thread: bold idea, messy messaging. Until someone posts numbers, the vibe is equal parts curiosity and side-eye, with jokes about “thermodynamics” sounding a lot like “hot air.”