Jobs at Etched

Etched is building AI chips that are hard-coded for individual model architectures. Our first product (Sohu) only supports transformers, but has an order of magnitude more throughput and lower latency than a B200. With Etched ASICs, you can build products that would be impossible with GPUs, like real-time video generation models and extremely deep & parallel chain-of-thought reasoning agents.

Etched is building what it hopes will be the defining inference chip of the transformer era. Unlike general-purpose GPUs, its Sohu chip is an application-specific integrated circuit (ASIC) designed exclusively to run transformer-based models. By stripping out everything unrelated to that architecture, Etched claims Sohu can deliver 140x better performance per dollar than traditional GPUs. The company has raised $125 million in under two years and is aiming to sell its hardware to hyperscalers and major AI infra buyers.

The early signal is promising. Etched has customers with tens of millions of dollars in committed preorders, despite the chip still being pre-launch. With energy and latency constraints becoming existential for large-scale AI deployments, Sohu’s efficiency pitch is resonating. Founders Gavin Uberti and Chris Zhu, both Thiel Fellows and former ML infra engineers, are betting that the AI landscape will consolidate around transformers—and that customers will trade flexibility for speed and cost at scale.

Etched is early and high-risk. If the architecture shifts or hyperscalers design competing silicon in-house, the window could close quickly. But in a segment where speed and energy efficiency are becoming non-negotiable, Etched is one of the few startups with a focused, credible plan to build purpose-built hardware for the models that matter most today.

Written as of July 2025