At the AI Infrastructure Summit, NVIDIA showcased its DSX MaxLPS power scheduling system, specifically designed for AI factories. This system significantly enhances energy efficiency, increasing token throughput per megawatt by up to 40%. Currently, this technology has been integrated by Emerald AI into its Conductor platform. Among its features, the DSX Flex function dynamically adjusts the data center's power consumption based on grid conditions, successfully receiving 200 grid status signals without requiring user intervention. Through Lambda testing, this function achieved a 24% increase in token throughput under a fixed power budget. Additionally, NVIDIA revealed that the Vera Rubin and Groq 3 LPX racks can accommodate up to 40% more GPUs within the same site power consumption range, with token throughput increasing by up to 35%. Notably, the Groq 3 LPX achieved an impressive performance of 2,529 output tokens per user per second under specific workloads. Multiple enterprises have tested the Vera CPU, reporting substantial improvements in performance, throughput, and query task throughput. Furthermore, tests conducted by different companies also found significant optimizations in aspects such as security sandbox startup speed, orchestration step latency, overall latency, and analytical query performance.