Groq

Groq designs what it calls a language processing unit, a chip built around one job: producing tokens fast. Where a general-purpose GPU has to be good at many things, Groq’s architecture gives up that flexibility in exchange for very low latency, which is why its demos tend to show text appearing faster than you can read it.

Not to be confused with Grok, xAI’s chatbot, which shares an unfortunate amount of the same spelling. In 2026 Nvidia struck a licensing and hiring arrangement with Groq that lawmakers criticised as a way of buying a company without triggering merger review, a structure Nvidia later reused elsewhere.