“Groq's LPU architecture can run Mixture of Experts (MoE) models within its static scheduling framework by using scatter and gather operations to fetch different experts as needed for a given query.”