中信证券：推理+训练驱动算力需求井喷，云厂商资本开支或加码

Core Viewpoint - The report from CITIC Securities indicates that as AI commercialization accelerates by 2026, the demand for computing power is shifting from a single "training-driven" model to a dual "training + inference" model, which is expected to maintain rapid growth [1] Group 1: Computing Power Demand - The demand for computing power is transitioning to a dual-driven model, combining both training and inference [1] - This shift is anticipated to sustain high growth rates in the future [1] Group 2: Domestic Chip Development - Continuous breakthroughs in domestic computing power chips in terms of process and architecture are expected [1] - The production capacity and yield of advanced domestic processes are improving [1] - The introduction of super nodes is likely to enhance the capabilities of domestic cloud vendors [1] Group 3: Market Opportunities - The increased investment in computing power by domestic cloud vendors is expected to lead to the continuous expansion of domestic intelligent computing centers [1] - This expansion will facilitate the formation of a closed loop among "chips-models-applications" [1] - There are significant opportunities for value reassessment in IDC and network sectors due to the accelerated investment in domestic computing power [1]