"ByteDance is creating a new chip for AI inference loads. The chip is similar to those produced by Nvidia partner Groq. The move represents an expansion into language processing units (LPUs)."