A Californian AI-chip startup designing a processor custom-built to run transformer models, the algorithms that underpin most large language models. By specialising entirely in transformers, Etched can strip away hardware needed for other uses and simplify the software running on the chip, yielding greater efficiency. The approach carries risk: designing a new chip typically takes 12-18 months, whereas AI algorithms evolve far faster, and a chip built around today's dominant architecture could become obsolete if the field shifts. As of early 2026 the chip remains in development.
Those who don't know, talk. Those who don't talk, know.