Models & Architectures
LPU (Language Processing Unit)
A chip built specifically for fast AI inference, pioneered by Groq as an alternative to general-purpose GPUs.
Also known as: language processing unit,fast inference chip
An LPU, or Language Processing Unit, is a chip designed from scratch to run trained language models as fast as possible, rather than being adapted from graphics hardware like a GPU. Groq popularised the term with chips built around deterministic, sequential token generation, which trades general-purpose flexibility for very high, predictable inference speed on language models specifically. LPUs cannot train models — only run them once trained elsewhere — so they compete with GPUs on inference workloads rather than replacing them outright. The speed advantage matters most for latency-sensitive products: voice agents, live coding assistants, and any AI workflow chaining many model calls together.
