Groq raised $650 million to expand its AI inference cloud, with 13 data centers, more than five million developers, NVIDIA LPX integration, and a 200 MW capacity target by the end of 2027. The deal shows why serving AI models is becoming its own infrastructure market, separate from the training race.