Research on the supply mode and cost structure of railway Artificial Intelligence computing power

To address the issues of selecting the supply mode and controlling the computing power cost in railway artificial intelligence scenarios, this paper, based on an analysis of the comprehensive computing power demand characteristics of railway AI, conducts a comparative study of three supply modes: self-built intelligent computing centers, on-demand rental of computing power, and hybrid computing power configuration. It also uses the Total Cost of Ownership (TCO) method to analyze the cost structure of the three modes. The research shows that the self-built mode is suitable for scenarios with high security requirements, stable training tasks, and long - term reuse value; the rental mode is suitable for short-term, fluctuating, and rapid deployment needs; and the hybrid mode can achieve a better balance between security control, cost constraints, and elastic supply. Further, considering the business characteristics of railway AI, which are “concentrated training and distributed inference”, this paper takes the intelligent recognition model for vehicle fault images as an example to estimate the cost of the hybrid configuration mode: self-built training computing power and rental inference computing power, providing a reference for the comprehensive computing power resource planning, supply mode selection, and configuration optimization of railway AI.