As the “digital brain” and core foundational support for the development of intelligent transportation and connected vehicles, the performance of data centers directly determines the operational capability of intelligent transportation systems. In the process of advancing the vehicle-road-cloud collaborative architecture, the demand for high-performance computing power in data centers has experienced explosive growth. The substantial increase in computing tasks has posed severe challenges to thermal management, making efficient and reliable cooling systems an indispensable core component. Centrifugal compressor water-cooling units are the mainstream cooling solution for large-capacity scenarios, and their design optimization is crucial for improving the energy efficiency and performance of the entire cooling system. This paper proposes a one-dimensional performance prediction method for centrifugal compressors based on an empirical loss model, and realizes the iterative calculation of parameters in the entire flow path from the impeller inlet to the diffuser outlet through Python programming. A systematic impact assessment was carried out for major loss mechanisms such as surface friction, tip clearance, and wake mixing under standard operating conditions and critical operating conditions. The results show that the original model has high prediction accuracy under standard operating conditions, with isentropic efficiency error not exceeding 5%; however, under critical operating conditions, the efficiency prediction deviation reaches 7.54% due to the neglect of coupling effects between various losses. To address this issue, this paper introduces deviation correction factors related to flow rate, rotational speed, and density, which significantly improve the model’s prediction capability under extreme operating conditions: the efficiency error under critical operating conditions is reduced to 1.54%, and only 0.3% under rated operating conditions. This model provides a reliable tool for compressor performance prediction and extreme operating boundary identification, and has high application value in engineering practice.