“Groq's LPU-based inference is more than 5 times lower in cost compared to GPU-based solutions.”