NVIDIA Paid $20B for Groq. The Inference Chip Gap Is Real

On December 24, 2025, NVIDIA paid $20 billion for a non-exclusive license to Groq’s Language Processing Unit (LPU) architecture and hired Groq’s CEO Jonathan Ross plus core engineering talent. The deal signals that even NVIDIA, with ~80% of AI training silicon, recognizes GPUs are architecturally wrong for inference decode — …