Etched @ Princeton

​Etched is building Sohu, an AI inference chip that's >10x faster than NVIDIA GPUs. Sohu is hyper-specialized for running transformers: the architecture that powers all frontier AI models today (from OpenAI to xAI, Meta, Deepseek, etc.). By specializing, Sohu offers disruptively better throughput, latency, and cost per token. Sohu makes new AI products like real-time video and swarms of deep research agents deployable at humanity-scale.

​Now 150-people strong, the team has attracted some of the top engineers, AI researchers, PhDs, quants, and industry legends, like Brian Loiler (who built HGX/DGX, >80% of NVIDIA’s revenue) and David Munday (who built Google’s TPU v1-v5 software and firmware stack). Etched has raised hundreds of millions from leading investors, including Peter Thiel, Jane Street, and Two Sigma, all betting that Etched will unlock the AI inference singularity.

​​Undergrads, grad/PhD students, and researchers are all welcome to join. Please RSVP as space is very limited.