Groq

Groq builds custom tensor-streaming processors designed specifically for low-latency machine learning inference. Their platform integrates deterministic chip designs with a compiler-controlled execution model to run AI workloads in real time. Sells inference access directly via GroqCloud, its own API/serving platform -- the exact dual silicon/serving straddle the taxonomy anticipated.

Ecosystem functionAn inference chip pioneer acquired by Nvidia to expand its real-time processing capabilities.

Business modelInference hardware and cloud services