Etched is building specialized AI hardware that transforms the performance of transformer models. The company has developed Sohu, the world's first transformer-specific ASIC chip that delivers an order of magnitude more throughput and lower latency than leading GPUs like NVIDIA's B200. With capabilities running over 500,000 tokens per second on Llama 70B, Etched's technology enables products previously impossible with GPUs, including real-time video generation models and extremely deep chain-of-thought reasoning agents. The company's specialized approach to AI inference hardware allows businesses to achieve dramatic cost savings and performance improvements in their AI workloads.
Founded in 2022 by Harvard dropouts Gavin Uberti, Chris Zhu, and Robert Wachen, Etched has rapidly emerged as a formidable competitor in the AI chip market. The company has raised significant funding, including a $120 million round, reaching a valuation of $5 billion as they challenge established players like NVIDIA. Etched's philosophy centers on the conviction that transformers will dominate the AI landscape, creating a market for single-model ASICs that can outperform general-purpose computing solutions. By partnering with industry leaders like Rambus for memory and interface technologies, Etched is poised to reshape the future of artificial intelligence computing with their innovative chip architecture.