NVIDIA Vera Rubin: 30x Throughput/Watt, 35x Cost Reduction

NVIDIA's Vera Rubin platform shows a significant leap in Agentic AI performance. It delivers a massive 30 times increase in throughput per watt compared to Blackwell. What's more, Vera Rubin offers a 35 times token cost reduction for Agentic AI workloads. These results were measured using real-world agentic coding trajectories and the SemiAnalysis AgentX benchmark. This benchmark evaluates AI infrastructure in agentic-coding inference across various models like Kimi K3 and DeepSeek V4 Pro. The platform's efficiency is critical for AI operators. It means they can support more AI capacity within existing power and infrastructure budgets or significantly lower operating costs for the same capacity. This advancement signals a major step forward for Agentic AI development.