Gavin Uberti, Robert Wachen, and Chris Zhu
AI inference is becoming one of the largest computing markets in history and almost none of today’s hardware was built for it. Every modern AI product runs on a constant stream of token generation, context processing, and agentic execution loops, most of it bottlenecked by chips designed for a different era of compute.
Etched’s silicon is purpose-built to change that. The company’s transformer-native architecture handles the full demands of modern inference, from prefill-heavy throughput to ultra-low-latency decode, all without the thermal and memory compromises that plague general-purpose processors. Where most hardware drops more GPUs on top of a broken memory architecture, Etched is rebuilding the compute layer of AI from the ground up.
