← The AI Systems Map
L2 · ChipsAcceleratorUnited States

Etched

Transformer-ASIC startup betting the architecture into the silicon

Etched is a startup building Sohu, an ASIC that hard-codes the transformer architecture directly into silicon, trading generality for extreme throughput and efficiency on transformer inference. The bet is that if transformers remain dominant, a fixed-function chip can massively outperform GPUs on tokens/sec per dollar and watt. It raised a $120M round in 2024 and is fabbed at TSMC; its risk is architectural — a shift away from transformers would strand the design.

Product

Sohu — transformer-only ASIC

Bet

Fixed-function beats GPU on transformers

Risk

Locked to transformer architecture

Fab

TSMC

How it fits the stack

Etched with what it depends on (above) and what it feeds (below). The figure renders as a crawlable diagram and upgrades to an interactive 3D graph as it scrolls into view.

usesdepends oncompetes withEtchedChipsHigh-Bandwidth Memory(HBM)chokepointTSMC (TaiwanSemiconductorManufacturing Company)chokepointNvidia
EtchedDepends on ↑Related

Etched in the AI stack. Etched with its immediate upstream dependencies (top) and downstream dependents (bottom) in the AI value chain. Hover a node in 3D, or read the full relationships below.

Graph data (text) — 4 entities, 3 relationships
  • EtchedusesHigh-Bandwidth Memory (HBM)
  • Etcheddepends onTSMC (Taiwan Semiconductor Manufacturing Company)
  • Etchedcompetes withNvidia