Groq Pivots to Inference, Betting Training Becomes Commodity

Groq confirmed a $650 million funding round and a deliberate shift toward AI inference workloads, six months after its $20 billion non-exclusive licensing agreement with Nvidia. The nine-year-old chipmaker's LPU architecture is optimized for low-latency token generation, not training. As frontier model weights stabilize and enterprises move from experimentation to production, inference throughput and cost-per-token metrics are driving procurement. Existing investors backing the round signals internal confidence in the inference-focused strategy.