Vera Rubin’s NVL72 architecture reportedly delivers 67 times better performance per dollar for agentic inference workloads, according to analysis by SemiAnalysis. The advancement suggests significant improvements in the cost-efficiency of AI inference systems designed for agentic applications.