
Lightning fast chips with hardcoded AI models
Lamb Labs is a YC S26 startup building custom inference chips that hardcode LLM weights directly into silicon. We're developing architectures around binary and ternary model weights, with the goal of making AI inference dramatically faster and more energy-efficient.
We believe there is an opportunity to rethink how LLMs are implemented in hardware, rather than simply adapting existing processor architectures to run them.
We're pre-product, and our team currently consists of our two founders, Thomas and Niki. We're looking for an exceptional engineer or researcher to help develop the architecture behind our first chips, exploring new approaches to model-specific computation, memory, and automated hardware generation.