Nvidia said its Groq 3 LPX inference accelerator entered full production, positioning the chip for agentic AI workloads as competition for inference grows.