Etched

Etched (Etched.ai, Inc.) is a semiconductor startup founded in 2022 by Harvard dropouts Gavin Uberti, Chris Zhu, and Robert Wachen. Based in Cupertino, California, the company specializes in application-specific integrated circuits (ASICs) designed for transformer-based AI models. Their technology aims to outperform general-purpose GPUs in inference tasks, enabling faster and more cost-effective processing for large language models (LLMs). The flagship product, Sohu, is the first chip specifically built for transformer AI architecture. It is fabricated on TSMC's 4nm process and features 144 GB of HBM3E memory per chip. Sohu optimizes matrix multiplication patterns for transformer inference, achieving significantly higher performance compared to traditional GPU systems. Etched also offers complete server systems that integrate Sohu chips and software, facilitating easy model porting and scalability for advanced AI workloads. The company targets hyperscalers, AI research firms, and enterprises that require robust solutions for real-time applications and large-scale data processing.