transformer-inference
(2)
Reiner Pope
Reiner Pope is the co-founder and CEO of MatX, an AI chip startup based in Mountain View, California, designing purpose-built silicon for large language models. A former Google Senior Staff Software Engineer who led AI software development for Google's TPUs and served as Efficiency Lead for PaLM, Pope left Google one week before ChatGPT launched to build what he calls 'the best chips for LLMs that physics allows.' MatX has raised approximately $604M to date, including a $500M Series B led by Jane Street and Situational Awareness LP, and is building a hybrid SRAM-HBM chip architecture targeting the world's leading AI labs.
Positron AI
Positron AI designs purpose-built inference hardware for transformer models, aiming to make Nvidia GPUs optional for running large language models at production scale. Its first product, Atlas, ships from US fabs and claims roughly 3x lower latency and 4x better performance-per-watt versus an H100 system.