NVIDIA's official blog explains that AI factories, costing roughly $60 million per megawatt, maximize return on investment through three key factors: productivity, durability, and fungibility. The company highlights that its Vera Rubin NVL72 systems deliver over 30 times higher throughput per megawatt and up to 45 times lower cost per million tokens compared to GB300 NVL72, according to SemiAnalysis AgentX data. NVIDIA GPUs, such as the A100, continue to earn years after shipment, with CoreWeave extending bookings for units introduced in 2020 through 2029. Additionally, NVIDIA AI factories are designed to run every type of AI model and non-AI workloads, enhancing their long-term utility and demand.