Groq Banks $650M to Compete in AI Inference, Not Training

Groq has secured $650 million in funding to double down on inference—running trained models in production—rather than training them from scratch. Its LPU chip excels at speed and cost-per-token metrics that enterprises now prioritize as AI moves from experimentation to deployment. The move comes as frontier model weights stabilize, making hardware competition increasingly about inference efficiency rather than raw training power.