This is the value prop of Groq and Cerebras. They don't have the best models, but they have the fastest inference, and Groq has both the lowest cost and fastest speed.