Etched

Etched's August 18, 2026 release states that its first rack was shipped to Jane Street in July and is being deployed in Jane Street's workloads; a quoted Jane Street statement says the rack is running in its data center. This completed first delivery is distinct from the earlier A0 silicon validation and customer-contract observation.

Etched states its A0 silicon returned from TSMC N4P and its first rack-scale product is undergoing customer validation, with early customer tests and production initiated against customer contracts.

Etched is developing transformer-specific AI accelerators, but public evidence of broad production deployment remains early.

Etched's July 2026 update describes co-design across chips, packages, PCBs, cold plates and interconnects for frontier inference clusters. This adds specific system-architecture disciplines beyond the earlier Sohu transformer-ASIC description.

Etched develops Sohu, an ASIC architecture specialized for transformer inference with strong technical differentiation.

Etched develops Sohu, an ASIC architecture specialized for transformer inference with strong technical differentiation.

Etched identifies platform lead Brian Loiler with prior NVIDIA HGX/DGX engineering experience and software lead David Munday with prior TPU software leadership, adding named systems-organization expertise beyond physical facilities.

Etched's official site reports that it has opened a Taiwan factory and built a data center, test house and NPI prototyping lab in its San Jose office to support 24/7 engineering cycles.

Etched has assembled chip-design and systems talent and financing to pursue advanced AI semiconductor production.

Etched describes inference-cluster co-design across chips, packages, PCBs, cold plates and interconnects in its July 23, 2026 update. This adds concrete packaging, thermal and board-level disciplines supporting inferred engineering portability beyond the existing ASIC/compiler description.

ASIC, compiler and inference systems skills transfer across semiconductor and AI compute employers, though architecture specialization remains.

ASIC, compiler and inference systems skills transfer across semiconductor and AI compute employers, though architecture specialization remains.